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/bio.h>
18 #include <linux/blkdev.h>
19 #include <linux/blk-mq.h>
20 #include <linux/blk-pm.h>
21 #include <linux/highmem.h>
23 #include <linux/pagemap.h>
24 #include <linux/kernel_stat.h>
25 #include <linux/string.h>
26 #include <linux/init.h>
27 #include <linux/completion.h>
28 #include <linux/slab.h>
29 #include <linux/swap.h>
30 #include <linux/writeback.h>
31 #include <linux/task_io_accounting_ops.h>
32 #include <linux/fault-inject.h>
33 #include <linux/list_sort.h>
34 #include <linux/delay.h>
35 #include <linux/ratelimit.h>
36 #include <linux/pm_runtime.h>
37 #include <linux/blk-cgroup.h>
38 #include <linux/t10-pi.h>
39 #include <linux/debugfs.h>
40 #include <linux/bpf.h>
41 #include <linux/psi.h>
42 #include <linux/sched/sysctl.h>
43 #include <linux/blk-crypto.h>
45 #define CREATE_TRACE_POINTS
46 #include <trace/events/block.h>
50 #include "blk-mq-sched.h"
52 #include "blk-rq-qos.h"
54 struct dentry *blk_debugfs_root;
56 EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_remap);
57 EXPORT_TRACEPOINT_SYMBOL_GPL(block_rq_remap);
58 EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_complete);
59 EXPORT_TRACEPOINT_SYMBOL_GPL(block_split);
60 EXPORT_TRACEPOINT_SYMBOL_GPL(block_unplug);
61 EXPORT_TRACEPOINT_SYMBOL_GPL(block_rq_insert);
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);
120 rq->tag = BLK_MQ_NO_TAG;
121 rq->internal_tag = BLK_MQ_NO_TAG;
122 rq->start_time_ns = ktime_get_ns();
124 blk_crypto_rq_set_defaults(rq);
126 EXPORT_SYMBOL(blk_rq_init);
128 #define REQ_OP_NAME(name) [REQ_OP_##name] = #name
129 static const char *const blk_op_name[] = {
133 REQ_OP_NAME(DISCARD),
134 REQ_OP_NAME(SECURE_ERASE),
135 REQ_OP_NAME(ZONE_RESET),
136 REQ_OP_NAME(ZONE_RESET_ALL),
137 REQ_OP_NAME(ZONE_OPEN),
138 REQ_OP_NAME(ZONE_CLOSE),
139 REQ_OP_NAME(ZONE_FINISH),
140 REQ_OP_NAME(ZONE_APPEND),
141 REQ_OP_NAME(WRITE_SAME),
142 REQ_OP_NAME(WRITE_ZEROES),
144 REQ_OP_NAME(DRV_OUT),
149 * blk_op_str - Return string XXX in the REQ_OP_XXX.
152 * Description: Centralize block layer function to convert REQ_OP_XXX into
153 * string format. Useful in the debugging and tracing bio or request. For
154 * invalid REQ_OP_XXX it returns string "UNKNOWN".
156 inline const char *blk_op_str(unsigned int op)
158 const char *op_str = "UNKNOWN";
160 if (op < ARRAY_SIZE(blk_op_name) && blk_op_name[op])
161 op_str = blk_op_name[op];
165 EXPORT_SYMBOL_GPL(blk_op_str);
167 static const struct {
171 [BLK_STS_OK] = { 0, "" },
172 [BLK_STS_NOTSUPP] = { -EOPNOTSUPP, "operation not supported" },
173 [BLK_STS_TIMEOUT] = { -ETIMEDOUT, "timeout" },
174 [BLK_STS_NOSPC] = { -ENOSPC, "critical space allocation" },
175 [BLK_STS_TRANSPORT] = { -ENOLINK, "recoverable transport" },
176 [BLK_STS_TARGET] = { -EREMOTEIO, "critical target" },
177 [BLK_STS_NEXUS] = { -EBADE, "critical nexus" },
178 [BLK_STS_MEDIUM] = { -ENODATA, "critical medium" },
179 [BLK_STS_PROTECTION] = { -EILSEQ, "protection" },
180 [BLK_STS_RESOURCE] = { -ENOMEM, "kernel resource" },
181 [BLK_STS_DEV_RESOURCE] = { -EBUSY, "device resource" },
182 [BLK_STS_AGAIN] = { -EAGAIN, "nonblocking retry" },
184 /* device mapper special case, should not leak out: */
185 [BLK_STS_DM_REQUEUE] = { -EREMCHG, "dm internal retry" },
187 /* zone device specific errors */
188 [BLK_STS_ZONE_OPEN_RESOURCE] = { -ETOOMANYREFS, "open zones exceeded" },
189 [BLK_STS_ZONE_ACTIVE_RESOURCE] = { -EOVERFLOW, "active zones exceeded" },
191 /* everything else not covered above: */
192 [BLK_STS_IOERR] = { -EIO, "I/O" },
195 blk_status_t errno_to_blk_status(int errno)
199 for (i = 0; i < ARRAY_SIZE(blk_errors); i++) {
200 if (blk_errors[i].errno == errno)
201 return (__force blk_status_t)i;
204 return BLK_STS_IOERR;
206 EXPORT_SYMBOL_GPL(errno_to_blk_status);
208 int blk_status_to_errno(blk_status_t status)
210 int idx = (__force int)status;
212 if (WARN_ON_ONCE(idx >= ARRAY_SIZE(blk_errors)))
214 return blk_errors[idx].errno;
216 EXPORT_SYMBOL_GPL(blk_status_to_errno);
218 static void print_req_error(struct request *req, blk_status_t status,
221 int idx = (__force int)status;
223 if (WARN_ON_ONCE(idx >= ARRAY_SIZE(blk_errors)))
226 printk_ratelimited(KERN_ERR
227 "%s: %s error, dev %s, sector %llu op 0x%x:(%s) flags 0x%x "
228 "phys_seg %u prio class %u\n",
229 caller, blk_errors[idx].name,
230 req->rq_disk ? req->rq_disk->disk_name : "?",
231 blk_rq_pos(req), req_op(req), blk_op_str(req_op(req)),
232 req->cmd_flags & ~REQ_OP_MASK,
233 req->nr_phys_segments,
234 IOPRIO_PRIO_CLASS(req->ioprio));
237 static void req_bio_endio(struct request *rq, struct bio *bio,
238 unsigned int nbytes, blk_status_t error)
241 bio->bi_status = error;
243 if (unlikely(rq->rq_flags & RQF_QUIET))
244 bio_set_flag(bio, BIO_QUIET);
246 bio_advance(bio, nbytes);
248 if (req_op(rq) == REQ_OP_ZONE_APPEND && error == BLK_STS_OK) {
250 * Partial zone append completions cannot be supported as the
251 * BIO fragments may end up not being written sequentially.
253 if (bio->bi_iter.bi_size)
254 bio->bi_status = BLK_STS_IOERR;
256 bio->bi_iter.bi_sector = rq->__sector;
259 /* don't actually finish bio if it's part of flush sequence */
260 if (bio->bi_iter.bi_size == 0 && !(rq->rq_flags & RQF_FLUSH_SEQ))
264 void blk_dump_rq_flags(struct request *rq, char *msg)
266 printk(KERN_INFO "%s: dev %s: flags=%llx\n", msg,
267 rq->rq_disk ? rq->rq_disk->disk_name : "?",
268 (unsigned long long) rq->cmd_flags);
270 printk(KERN_INFO " sector %llu, nr/cnr %u/%u\n",
271 (unsigned long long)blk_rq_pos(rq),
272 blk_rq_sectors(rq), blk_rq_cur_sectors(rq));
273 printk(KERN_INFO " bio %p, biotail %p, len %u\n",
274 rq->bio, rq->biotail, blk_rq_bytes(rq));
276 EXPORT_SYMBOL(blk_dump_rq_flags);
279 * blk_sync_queue - cancel any pending callbacks on a queue
283 * The block layer may perform asynchronous callback activity
284 * on a queue, such as calling the unplug function after a timeout.
285 * A block device may call blk_sync_queue to ensure that any
286 * such activity is cancelled, thus allowing it to release resources
287 * that the callbacks might use. The caller must already have made sure
288 * that its ->submit_bio will not re-add plugging prior to calling
291 * This function does not cancel any asynchronous activity arising
292 * out of elevator or throttling code. That would require elevator_exit()
293 * and blkcg_exit_queue() to be called with queue lock initialized.
296 void blk_sync_queue(struct request_queue *q)
298 del_timer_sync(&q->timeout);
299 cancel_work_sync(&q->timeout_work);
301 EXPORT_SYMBOL(blk_sync_queue);
304 * blk_set_pm_only - increment pm_only counter
305 * @q: request queue pointer
307 void blk_set_pm_only(struct request_queue *q)
309 atomic_inc(&q->pm_only);
311 EXPORT_SYMBOL_GPL(blk_set_pm_only);
313 void blk_clear_pm_only(struct request_queue *q)
317 pm_only = atomic_dec_return(&q->pm_only);
318 WARN_ON_ONCE(pm_only < 0);
320 wake_up_all(&q->mq_freeze_wq);
322 EXPORT_SYMBOL_GPL(blk_clear_pm_only);
325 * blk_put_queue - decrement the request_queue refcount
326 * @q: the request_queue structure to decrement the refcount for
328 * Decrements the refcount of the request_queue kobject. When this reaches 0
329 * we'll have blk_release_queue() called.
331 * Context: Any context, but the last reference must not be dropped from
334 void blk_put_queue(struct request_queue *q)
336 kobject_put(&q->kobj);
338 EXPORT_SYMBOL(blk_put_queue);
340 void blk_set_queue_dying(struct request_queue *q)
342 blk_queue_flag_set(QUEUE_FLAG_DYING, q);
345 * When queue DYING flag is set, we need to block new req
346 * entering queue, so we call blk_freeze_queue_start() to
347 * prevent I/O from crossing blk_queue_enter().
349 blk_freeze_queue_start(q);
352 blk_mq_wake_waiters(q);
354 /* Make blk_queue_enter() reexamine the DYING flag. */
355 wake_up_all(&q->mq_freeze_wq);
357 EXPORT_SYMBOL_GPL(blk_set_queue_dying);
360 * blk_cleanup_queue - shutdown a request queue
361 * @q: request queue to shutdown
363 * Mark @q DYING, drain all pending requests, mark @q DEAD, destroy and
364 * put it. All future requests will be failed immediately with -ENODEV.
368 void blk_cleanup_queue(struct request_queue *q)
370 /* cannot be called from atomic context */
373 WARN_ON_ONCE(blk_queue_registered(q));
375 /* mark @q DYING, no new request or merges will be allowed afterwards */
376 blk_set_queue_dying(q);
378 blk_queue_flag_set(QUEUE_FLAG_NOMERGES, q);
379 blk_queue_flag_set(QUEUE_FLAG_NOXMERGES, q);
382 * Drain all requests queued before DYING marking. Set DEAD flag to
383 * prevent that blk_mq_run_hw_queues() accesses the hardware queues
384 * after draining finished.
390 blk_queue_flag_set(QUEUE_FLAG_DEAD, q);
392 /* for synchronous bio-based driver finish in-flight integrity i/o */
393 blk_flush_integrity();
397 blk_mq_exit_queue(q);
400 * In theory, request pool of sched_tags belongs to request queue.
401 * However, the current implementation requires tag_set for freeing
402 * requests, so free the pool now.
404 * Queue has become frozen, there can't be any in-queue requests, so
405 * it is safe to free requests now.
407 mutex_lock(&q->sysfs_lock);
409 blk_mq_sched_free_requests(q);
410 mutex_unlock(&q->sysfs_lock);
412 percpu_ref_exit(&q->q_usage_counter);
414 /* @q is and will stay empty, shutdown and put */
417 EXPORT_SYMBOL(blk_cleanup_queue);
420 * blk_queue_enter() - try to increase q->q_usage_counter
421 * @q: request queue pointer
422 * @flags: BLK_MQ_REQ_NOWAIT and/or BLK_MQ_REQ_PM
424 int blk_queue_enter(struct request_queue *q, blk_mq_req_flags_t flags)
426 const bool pm = flags & BLK_MQ_REQ_PM;
429 bool success = false;
432 if (percpu_ref_tryget_live(&q->q_usage_counter)) {
434 * The code that increments the pm_only counter is
435 * responsible for ensuring that that counter is
436 * globally visible before the queue is unfrozen.
438 if ((pm && queue_rpm_status(q) != RPM_SUSPENDED) ||
439 !blk_queue_pm_only(q)) {
442 percpu_ref_put(&q->q_usage_counter);
450 if (flags & BLK_MQ_REQ_NOWAIT)
454 * read pair of barrier in blk_freeze_queue_start(),
455 * we need to order reading __PERCPU_REF_DEAD flag of
456 * .q_usage_counter and reading .mq_freeze_depth or
457 * queue dying flag, otherwise the following wait may
458 * never return if the two reads are reordered.
462 wait_event(q->mq_freeze_wq,
463 (!q->mq_freeze_depth &&
464 blk_pm_resume_queue(pm, q)) ||
466 if (blk_queue_dying(q))
471 static inline int bio_queue_enter(struct bio *bio)
473 struct request_queue *q = bio->bi_bdev->bd_disk->queue;
474 bool nowait = bio->bi_opf & REQ_NOWAIT;
477 ret = blk_queue_enter(q, nowait ? BLK_MQ_REQ_NOWAIT : 0);
479 if (nowait && !blk_queue_dying(q))
480 bio_wouldblock_error(bio);
488 void blk_queue_exit(struct request_queue *q)
490 percpu_ref_put(&q->q_usage_counter);
493 static void blk_queue_usage_counter_release(struct percpu_ref *ref)
495 struct request_queue *q =
496 container_of(ref, struct request_queue, q_usage_counter);
498 wake_up_all(&q->mq_freeze_wq);
501 static void blk_rq_timed_out_timer(struct timer_list *t)
503 struct request_queue *q = from_timer(q, t, timeout);
505 kblockd_schedule_work(&q->timeout_work);
508 static void blk_timeout_work(struct work_struct *work)
512 struct request_queue *blk_alloc_queue(int node_id)
514 struct request_queue *q;
517 q = kmem_cache_alloc_node(blk_requestq_cachep,
518 GFP_KERNEL | __GFP_ZERO, node_id);
522 q->last_merge = NULL;
524 q->id = ida_simple_get(&blk_queue_ida, 0, 0, GFP_KERNEL);
528 ret = bioset_init(&q->bio_split, BIO_POOL_SIZE, 0, 0);
532 q->stats = blk_alloc_queue_stats();
538 atomic_set(&q->nr_active_requests_shared_sbitmap, 0);
540 timer_setup(&q->timeout, blk_rq_timed_out_timer, 0);
541 INIT_WORK(&q->timeout_work, blk_timeout_work);
542 INIT_LIST_HEAD(&q->icq_list);
543 #ifdef CONFIG_BLK_CGROUP
544 INIT_LIST_HEAD(&q->blkg_list);
547 kobject_init(&q->kobj, &blk_queue_ktype);
549 mutex_init(&q->debugfs_mutex);
550 mutex_init(&q->sysfs_lock);
551 mutex_init(&q->sysfs_dir_lock);
552 spin_lock_init(&q->queue_lock);
554 init_waitqueue_head(&q->mq_freeze_wq);
555 mutex_init(&q->mq_freeze_lock);
558 * Init percpu_ref in atomic mode so that it's faster to shutdown.
559 * See blk_register_queue() for details.
561 if (percpu_ref_init(&q->q_usage_counter,
562 blk_queue_usage_counter_release,
563 PERCPU_REF_INIT_ATOMIC, GFP_KERNEL))
566 if (blkcg_init_queue(q))
569 blk_queue_dma_alignment(q, 511);
570 blk_set_default_limits(&q->limits);
571 q->nr_requests = BLKDEV_MAX_RQ;
576 percpu_ref_exit(&q->q_usage_counter);
578 blk_free_queue_stats(q->stats);
580 bioset_exit(&q->bio_split);
582 ida_simple_remove(&blk_queue_ida, q->id);
584 kmem_cache_free(blk_requestq_cachep, q);
589 * blk_get_queue - increment the request_queue refcount
590 * @q: the request_queue structure to increment the refcount for
592 * Increment the refcount of the request_queue kobject.
594 * Context: Any context.
596 bool blk_get_queue(struct request_queue *q)
598 if (likely(!blk_queue_dying(q))) {
605 EXPORT_SYMBOL(blk_get_queue);
608 * blk_get_request - allocate a request
609 * @q: request queue to allocate a request for
610 * @op: operation (REQ_OP_*) and REQ_* flags, e.g. REQ_SYNC.
611 * @flags: BLK_MQ_REQ_* flags, e.g. BLK_MQ_REQ_NOWAIT.
613 struct request *blk_get_request(struct request_queue *q, unsigned int op,
614 blk_mq_req_flags_t flags)
618 WARN_ON_ONCE(op & REQ_NOWAIT);
619 WARN_ON_ONCE(flags & ~(BLK_MQ_REQ_NOWAIT | BLK_MQ_REQ_PM));
621 req = blk_mq_alloc_request(q, op, flags);
622 if (!IS_ERR(req) && q->mq_ops->initialize_rq_fn)
623 q->mq_ops->initialize_rq_fn(req);
627 EXPORT_SYMBOL(blk_get_request);
629 void blk_put_request(struct request *req)
631 blk_mq_free_request(req);
633 EXPORT_SYMBOL(blk_put_request);
635 static void handle_bad_sector(struct bio *bio, sector_t maxsector)
637 char b[BDEVNAME_SIZE];
639 pr_info_ratelimited("attempt to access beyond end of device\n"
640 "%s: rw=%d, want=%llu, limit=%llu\n",
641 bio_devname(bio, b), bio->bi_opf,
642 bio_end_sector(bio), maxsector);
645 #ifdef CONFIG_FAIL_MAKE_REQUEST
647 static DECLARE_FAULT_ATTR(fail_make_request);
649 static int __init setup_fail_make_request(char *str)
651 return setup_fault_attr(&fail_make_request, str);
653 __setup("fail_make_request=", setup_fail_make_request);
655 static bool should_fail_request(struct block_device *part, unsigned int bytes)
657 return part->bd_make_it_fail && should_fail(&fail_make_request, bytes);
660 static int __init fail_make_request_debugfs(void)
662 struct dentry *dir = fault_create_debugfs_attr("fail_make_request",
663 NULL, &fail_make_request);
665 return PTR_ERR_OR_ZERO(dir);
668 late_initcall(fail_make_request_debugfs);
670 #else /* CONFIG_FAIL_MAKE_REQUEST */
672 static inline bool should_fail_request(struct block_device *part,
678 #endif /* CONFIG_FAIL_MAKE_REQUEST */
680 static inline bool bio_check_ro(struct bio *bio)
682 if (op_is_write(bio_op(bio)) && bdev_read_only(bio->bi_bdev)) {
683 char b[BDEVNAME_SIZE];
685 if (op_is_flush(bio->bi_opf) && !bio_sectors(bio))
689 "Trying to write to read-only block-device %s (partno %d)\n",
690 bio_devname(bio, b), bio->bi_bdev->bd_partno);
691 /* Older lvm-tools actually trigger this */
698 static noinline int should_fail_bio(struct bio *bio)
700 if (should_fail_request(bdev_whole(bio->bi_bdev), bio->bi_iter.bi_size))
704 ALLOW_ERROR_INJECTION(should_fail_bio, ERRNO);
707 * Check whether this bio extends beyond the end of the device or partition.
708 * This may well happen - the kernel calls bread() without checking the size of
709 * the device, e.g., when mounting a file system.
711 static inline int bio_check_eod(struct bio *bio)
713 sector_t maxsector = bdev_nr_sectors(bio->bi_bdev);
714 unsigned int nr_sectors = bio_sectors(bio);
716 if (nr_sectors && maxsector &&
717 (nr_sectors > maxsector ||
718 bio->bi_iter.bi_sector > maxsector - nr_sectors)) {
719 handle_bad_sector(bio, maxsector);
726 * Remap block n of partition p to block n+start(p) of the disk.
728 static int blk_partition_remap(struct bio *bio)
730 struct block_device *p = bio->bi_bdev;
732 if (unlikely(should_fail_request(p, bio->bi_iter.bi_size)))
734 if (bio_sectors(bio)) {
735 bio->bi_iter.bi_sector += p->bd_start_sect;
736 trace_block_bio_remap(bio, p->bd_dev,
737 bio->bi_iter.bi_sector -
740 bio_set_flag(bio, BIO_REMAPPED);
745 * Check write append to a zoned block device.
747 static inline blk_status_t blk_check_zone_append(struct request_queue *q,
750 sector_t pos = bio->bi_iter.bi_sector;
751 int nr_sectors = bio_sectors(bio);
753 /* Only applicable to zoned block devices */
754 if (!blk_queue_is_zoned(q))
755 return BLK_STS_NOTSUPP;
757 /* The bio sector must point to the start of a sequential zone */
758 if (pos & (blk_queue_zone_sectors(q) - 1) ||
759 !blk_queue_zone_is_seq(q, pos))
760 return BLK_STS_IOERR;
763 * Not allowed to cross zone boundaries. Otherwise, the BIO will be
764 * split and could result in non-contiguous sectors being written in
767 if (nr_sectors > q->limits.chunk_sectors)
768 return BLK_STS_IOERR;
770 /* Make sure the BIO is small enough and will not get split */
771 if (nr_sectors > q->limits.max_zone_append_sectors)
772 return BLK_STS_IOERR;
774 bio->bi_opf |= REQ_NOMERGE;
779 static noinline_for_stack bool submit_bio_checks(struct bio *bio)
781 struct block_device *bdev = bio->bi_bdev;
782 struct request_queue *q = bdev->bd_disk->queue;
783 blk_status_t status = BLK_STS_IOERR;
784 struct blk_plug *plug;
788 plug = blk_mq_plug(q, bio);
789 if (plug && plug->nowait)
790 bio->bi_opf |= REQ_NOWAIT;
793 * For a REQ_NOWAIT based request, return -EOPNOTSUPP
794 * if queue does not support NOWAIT.
796 if ((bio->bi_opf & REQ_NOWAIT) && !blk_queue_nowait(q))
799 if (should_fail_bio(bio))
801 if (unlikely(bio_check_ro(bio)))
803 if (!bio_flagged(bio, BIO_REMAPPED)) {
804 if (unlikely(bio_check_eod(bio)))
806 if (bdev->bd_partno && unlikely(blk_partition_remap(bio)))
811 * Filter flush bio's early so that bio based drivers without flush
812 * support don't have to worry about them.
814 if (op_is_flush(bio->bi_opf) &&
815 !test_bit(QUEUE_FLAG_WC, &q->queue_flags)) {
816 bio->bi_opf &= ~(REQ_PREFLUSH | REQ_FUA);
817 if (!bio_sectors(bio)) {
823 if (!test_bit(QUEUE_FLAG_POLL, &q->queue_flags))
824 bio_clear_hipri(bio);
826 switch (bio_op(bio)) {
828 if (!blk_queue_discard(q))
831 case REQ_OP_SECURE_ERASE:
832 if (!blk_queue_secure_erase(q))
835 case REQ_OP_WRITE_SAME:
836 if (!q->limits.max_write_same_sectors)
839 case REQ_OP_ZONE_APPEND:
840 status = blk_check_zone_append(q, bio);
841 if (status != BLK_STS_OK)
844 case REQ_OP_ZONE_RESET:
845 case REQ_OP_ZONE_OPEN:
846 case REQ_OP_ZONE_CLOSE:
847 case REQ_OP_ZONE_FINISH:
848 if (!blk_queue_is_zoned(q))
851 case REQ_OP_ZONE_RESET_ALL:
852 if (!blk_queue_is_zoned(q) || !blk_queue_zone_resetall(q))
855 case REQ_OP_WRITE_ZEROES:
856 if (!q->limits.max_write_zeroes_sectors)
864 * Various block parts want %current->io_context, so allocate it up
865 * front rather than dealing with lots of pain to allocate it only
866 * where needed. This may fail and the block layer knows how to live
869 if (unlikely(!current->io_context))
870 create_task_io_context(current, GFP_ATOMIC, q->node);
872 if (blk_throtl_bio(bio)) {
873 blkcg_bio_issue_init(bio);
877 blk_cgroup_bio_start(bio);
878 blkcg_bio_issue_init(bio);
880 if (!bio_flagged(bio, BIO_TRACE_COMPLETION)) {
881 trace_block_bio_queue(bio);
882 /* Now that enqueuing has been traced, we need to trace
883 * completion as well.
885 bio_set_flag(bio, BIO_TRACE_COMPLETION);
890 status = BLK_STS_NOTSUPP;
892 bio->bi_status = status;
897 static blk_qc_t __submit_bio(struct bio *bio)
899 struct gendisk *disk = bio->bi_bdev->bd_disk;
900 blk_qc_t ret = BLK_QC_T_NONE;
902 if (unlikely(bio_queue_enter(bio) != 0))
903 return BLK_QC_T_NONE;
905 if (!submit_bio_checks(bio) || !blk_crypto_bio_prep(&bio))
907 if (disk->fops->submit_bio) {
908 ret = disk->fops->submit_bio(bio);
911 return blk_mq_submit_bio(bio);
914 blk_queue_exit(disk->queue);
919 * The loop in this function may be a bit non-obvious, and so deserves some
922 * - Before entering the loop, bio->bi_next is NULL (as all callers ensure
923 * that), so we have a list with a single bio.
924 * - We pretend that we have just taken it off a longer list, so we assign
925 * bio_list to a pointer to the bio_list_on_stack, thus initialising the
926 * bio_list of new bios to be added. ->submit_bio() may indeed add some more
927 * bios through a recursive call to submit_bio_noacct. If it did, we find a
928 * non-NULL value in bio_list and re-enter the loop from the top.
929 * - In this case we really did just take the bio of the top of the list (no
930 * pretending) and so remove it from bio_list, and call into ->submit_bio()
933 * bio_list_on_stack[0] contains bios submitted by the current ->submit_bio.
934 * bio_list_on_stack[1] contains bios that were submitted before the current
935 * ->submit_bio_bio, but that haven't been processed yet.
937 static blk_qc_t __submit_bio_noacct(struct bio *bio)
939 struct bio_list bio_list_on_stack[2];
940 blk_qc_t ret = BLK_QC_T_NONE;
942 BUG_ON(bio->bi_next);
944 bio_list_init(&bio_list_on_stack[0]);
945 current->bio_list = bio_list_on_stack;
948 struct request_queue *q = bio->bi_bdev->bd_disk->queue;
949 struct bio_list lower, same;
952 * Create a fresh bio_list for all subordinate requests.
954 bio_list_on_stack[1] = bio_list_on_stack[0];
955 bio_list_init(&bio_list_on_stack[0]);
957 ret = __submit_bio(bio);
960 * Sort new bios into those for a lower level and those for the
963 bio_list_init(&lower);
964 bio_list_init(&same);
965 while ((bio = bio_list_pop(&bio_list_on_stack[0])) != NULL)
966 if (q == bio->bi_bdev->bd_disk->queue)
967 bio_list_add(&same, bio);
969 bio_list_add(&lower, bio);
972 * Now assemble so we handle the lowest level first.
974 bio_list_merge(&bio_list_on_stack[0], &lower);
975 bio_list_merge(&bio_list_on_stack[0], &same);
976 bio_list_merge(&bio_list_on_stack[0], &bio_list_on_stack[1]);
977 } while ((bio = bio_list_pop(&bio_list_on_stack[0])));
979 current->bio_list = NULL;
983 static blk_qc_t __submit_bio_noacct_mq(struct bio *bio)
985 struct bio_list bio_list[2] = { };
988 current->bio_list = bio_list;
991 ret = __submit_bio(bio);
992 } while ((bio = bio_list_pop(&bio_list[0])));
994 current->bio_list = NULL;
999 * submit_bio_noacct - re-submit a bio to the block device layer for I/O
1000 * @bio: The bio describing the location in memory and on the device.
1002 * This is a version of submit_bio() that shall only be used for I/O that is
1003 * resubmitted to lower level drivers by stacking block drivers. All file
1004 * systems and other upper level users of the block layer should use
1005 * submit_bio() instead.
1007 blk_qc_t submit_bio_noacct(struct bio *bio)
1010 * We only want one ->submit_bio to be active at a time, else stack
1011 * usage with stacked devices could be a problem. Use current->bio_list
1012 * to collect a list of requests submited by a ->submit_bio method while
1013 * it is active, and then process them after it returned.
1015 if (current->bio_list) {
1016 bio_list_add(¤t->bio_list[0], bio);
1017 return BLK_QC_T_NONE;
1020 if (!bio->bi_bdev->bd_disk->fops->submit_bio)
1021 return __submit_bio_noacct_mq(bio);
1022 return __submit_bio_noacct(bio);
1024 EXPORT_SYMBOL(submit_bio_noacct);
1027 * submit_bio - submit a bio to the block device layer for I/O
1028 * @bio: The &struct bio which describes the I/O
1030 * submit_bio() is used to submit I/O requests to block devices. It is passed a
1031 * fully set up &struct bio that describes the I/O that needs to be done. The
1032 * bio will be send to the device described by the bi_bdev field.
1034 * The success/failure status of the request, along with notification of
1035 * completion, is delivered asynchronously through the ->bi_end_io() callback
1036 * in @bio. The bio must NOT be touched by thecaller until ->bi_end_io() has
1039 blk_qc_t submit_bio(struct bio *bio)
1041 if (blkcg_punt_bio_submit(bio))
1042 return BLK_QC_T_NONE;
1045 * If it's a regular read/write or a barrier with data attached,
1046 * go through the normal accounting stuff before submission.
1048 if (bio_has_data(bio)) {
1051 if (unlikely(bio_op(bio) == REQ_OP_WRITE_SAME))
1052 count = queue_logical_block_size(
1053 bio->bi_bdev->bd_disk->queue) >> 9;
1055 count = bio_sectors(bio);
1057 if (op_is_write(bio_op(bio))) {
1058 count_vm_events(PGPGOUT, count);
1060 task_io_account_read(bio->bi_iter.bi_size);
1061 count_vm_events(PGPGIN, count);
1066 * If we're reading data that is part of the userspace workingset, count
1067 * submission time as memory stall. When the device is congested, or
1068 * the submitting cgroup IO-throttled, submission can be a significant
1069 * part of overall IO time.
1071 if (unlikely(bio_op(bio) == REQ_OP_READ &&
1072 bio_flagged(bio, BIO_WORKINGSET))) {
1073 unsigned long pflags;
1076 psi_memstall_enter(&pflags);
1077 ret = submit_bio_noacct(bio);
1078 psi_memstall_leave(&pflags);
1083 return submit_bio_noacct(bio);
1085 EXPORT_SYMBOL(submit_bio);
1088 * blk_cloned_rq_check_limits - Helper function to check a cloned request
1089 * for the new queue limits
1091 * @rq: the request being checked
1094 * @rq may have been made based on weaker limitations of upper-level queues
1095 * in request stacking drivers, and it may violate the limitation of @q.
1096 * Since the block layer and the underlying device driver trust @rq
1097 * after it is inserted to @q, it should be checked against @q before
1098 * the insertion using this generic function.
1100 * Request stacking drivers like request-based dm may change the queue
1101 * limits when retrying requests on other queues. Those requests need
1102 * to be checked against the new queue limits again during dispatch.
1104 static blk_status_t blk_cloned_rq_check_limits(struct request_queue *q,
1107 unsigned int max_sectors = blk_queue_get_max_sectors(q, req_op(rq));
1109 if (blk_rq_sectors(rq) > max_sectors) {
1111 * SCSI device does not have a good way to return if
1112 * Write Same/Zero is actually supported. If a device rejects
1113 * a non-read/write command (discard, write same,etc.) the
1114 * low-level device driver will set the relevant queue limit to
1115 * 0 to prevent blk-lib from issuing more of the offending
1116 * operations. Commands queued prior to the queue limit being
1117 * reset need to be completed with BLK_STS_NOTSUPP to avoid I/O
1118 * errors being propagated to upper layers.
1120 if (max_sectors == 0)
1121 return BLK_STS_NOTSUPP;
1123 printk(KERN_ERR "%s: over max size limit. (%u > %u)\n",
1124 __func__, blk_rq_sectors(rq), max_sectors);
1125 return BLK_STS_IOERR;
1129 * The queue settings related to segment counting may differ from the
1132 rq->nr_phys_segments = blk_recalc_rq_segments(rq);
1133 if (rq->nr_phys_segments > queue_max_segments(q)) {
1134 printk(KERN_ERR "%s: over max segments limit. (%hu > %hu)\n",
1135 __func__, rq->nr_phys_segments, queue_max_segments(q));
1136 return BLK_STS_IOERR;
1143 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
1144 * @q: the queue to submit the request
1145 * @rq: the request being queued
1147 blk_status_t blk_insert_cloned_request(struct request_queue *q, struct request *rq)
1151 ret = blk_cloned_rq_check_limits(q, rq);
1152 if (ret != BLK_STS_OK)
1156 should_fail_request(rq->rq_disk->part0, blk_rq_bytes(rq)))
1157 return BLK_STS_IOERR;
1159 if (blk_crypto_insert_cloned_request(rq))
1160 return BLK_STS_IOERR;
1162 if (blk_queue_io_stat(q))
1163 blk_account_io_start(rq);
1166 * Since we have a scheduler attached on the top device,
1167 * bypass a potential scheduler on the bottom device for
1170 return blk_mq_request_issue_directly(rq, true);
1172 EXPORT_SYMBOL_GPL(blk_insert_cloned_request);
1175 * blk_rq_err_bytes - determine number of bytes till the next failure boundary
1176 * @rq: request to examine
1179 * A request could be merge of IOs which require different failure
1180 * handling. This function determines the number of bytes which
1181 * can be failed from the beginning of the request without
1182 * crossing into area which need to be retried further.
1185 * The number of bytes to fail.
1187 unsigned int blk_rq_err_bytes(const struct request *rq)
1189 unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK;
1190 unsigned int bytes = 0;
1193 if (!(rq->rq_flags & RQF_MIXED_MERGE))
1194 return blk_rq_bytes(rq);
1197 * Currently the only 'mixing' which can happen is between
1198 * different fastfail types. We can safely fail portions
1199 * which have all the failfast bits that the first one has -
1200 * the ones which are at least as eager to fail as the first
1203 for (bio = rq->bio; bio; bio = bio->bi_next) {
1204 if ((bio->bi_opf & ff) != ff)
1206 bytes += bio->bi_iter.bi_size;
1209 /* this could lead to infinite loop */
1210 BUG_ON(blk_rq_bytes(rq) && !bytes);
1213 EXPORT_SYMBOL_GPL(blk_rq_err_bytes);
1215 static void update_io_ticks(struct block_device *part, unsigned long now,
1218 unsigned long stamp;
1220 stamp = READ_ONCE(part->bd_stamp);
1221 if (unlikely(time_after(now, stamp))) {
1222 if (likely(cmpxchg(&part->bd_stamp, stamp, now) == stamp))
1223 __part_stat_add(part, io_ticks, end ? now - stamp : 1);
1225 if (part->bd_partno) {
1226 part = bdev_whole(part);
1231 static void blk_account_io_completion(struct request *req, unsigned int bytes)
1233 if (req->part && blk_do_io_stat(req)) {
1234 const int sgrp = op_stat_group(req_op(req));
1237 part_stat_add(req->part, sectors[sgrp], bytes >> 9);
1242 void blk_account_io_done(struct request *req, u64 now)
1245 * Account IO completion. flush_rq isn't accounted as a
1246 * normal IO on queueing nor completion. Accounting the
1247 * containing request is enough.
1249 if (req->part && blk_do_io_stat(req) &&
1250 !(req->rq_flags & RQF_FLUSH_SEQ)) {
1251 const int sgrp = op_stat_group(req_op(req));
1254 update_io_ticks(req->part, jiffies, true);
1255 part_stat_inc(req->part, ios[sgrp]);
1256 part_stat_add(req->part, nsecs[sgrp], now - req->start_time_ns);
1261 void blk_account_io_start(struct request *rq)
1263 if (!blk_do_io_stat(rq))
1266 /* passthrough requests can hold bios that do not have ->bi_bdev set */
1267 if (rq->bio && rq->bio->bi_bdev)
1268 rq->part = rq->bio->bi_bdev;
1270 rq->part = rq->rq_disk->part0;
1273 update_io_ticks(rq->part, jiffies, false);
1277 static unsigned long __part_start_io_acct(struct block_device *part,
1278 unsigned int sectors, unsigned int op)
1280 const int sgrp = op_stat_group(op);
1281 unsigned long now = READ_ONCE(jiffies);
1284 update_io_ticks(part, now, false);
1285 part_stat_inc(part, ios[sgrp]);
1286 part_stat_add(part, sectors[sgrp], sectors);
1287 part_stat_local_inc(part, in_flight[op_is_write(op)]);
1294 * bio_start_io_acct - start I/O accounting for bio based drivers
1295 * @bio: bio to start account for
1297 * Returns the start time that should be passed back to bio_end_io_acct().
1299 unsigned long bio_start_io_acct(struct bio *bio)
1301 return __part_start_io_acct(bio->bi_bdev, bio_sectors(bio), bio_op(bio));
1303 EXPORT_SYMBOL_GPL(bio_start_io_acct);
1305 unsigned long disk_start_io_acct(struct gendisk *disk, unsigned int sectors,
1308 return __part_start_io_acct(disk->part0, sectors, op);
1310 EXPORT_SYMBOL(disk_start_io_acct);
1312 static void __part_end_io_acct(struct block_device *part, unsigned int op,
1313 unsigned long start_time)
1315 const int sgrp = op_stat_group(op);
1316 unsigned long now = READ_ONCE(jiffies);
1317 unsigned long duration = now - start_time;
1320 update_io_ticks(part, now, true);
1321 part_stat_add(part, nsecs[sgrp], jiffies_to_nsecs(duration));
1322 part_stat_local_dec(part, in_flight[op_is_write(op)]);
1326 void bio_end_io_acct_remapped(struct bio *bio, unsigned long start_time,
1327 struct block_device *orig_bdev)
1329 __part_end_io_acct(orig_bdev, bio_op(bio), start_time);
1331 EXPORT_SYMBOL_GPL(bio_end_io_acct_remapped);
1333 void disk_end_io_acct(struct gendisk *disk, unsigned int op,
1334 unsigned long start_time)
1336 __part_end_io_acct(disk->part0, op, start_time);
1338 EXPORT_SYMBOL(disk_end_io_acct);
1341 * Steal bios from a request and add them to a bio list.
1342 * The request must not have been partially completed before.
1344 void blk_steal_bios(struct bio_list *list, struct request *rq)
1348 list->tail->bi_next = rq->bio;
1350 list->head = rq->bio;
1351 list->tail = rq->biotail;
1359 EXPORT_SYMBOL_GPL(blk_steal_bios);
1362 * blk_update_request - Complete multiple bytes without completing the request
1363 * @req: the request being processed
1364 * @error: block status code
1365 * @nr_bytes: number of bytes to complete for @req
1368 * Ends I/O on a number of bytes attached to @req, but doesn't complete
1369 * the request structure even if @req doesn't have leftover.
1370 * If @req has leftover, sets it up for the next range of segments.
1372 * Passing the result of blk_rq_bytes() as @nr_bytes guarantees
1373 * %false return from this function.
1376 * The RQF_SPECIAL_PAYLOAD flag is ignored on purpose in this function
1377 * except in the consistency check at the end of this function.
1380 * %false - this request doesn't have any more data
1381 * %true - this request has more data
1383 bool blk_update_request(struct request *req, blk_status_t error,
1384 unsigned int nr_bytes)
1388 trace_block_rq_complete(req, blk_status_to_errno(error), nr_bytes);
1393 #ifdef CONFIG_BLK_DEV_INTEGRITY
1394 if (blk_integrity_rq(req) && req_op(req) == REQ_OP_READ &&
1395 error == BLK_STS_OK)
1396 req->q->integrity.profile->complete_fn(req, nr_bytes);
1399 if (unlikely(error && !blk_rq_is_passthrough(req) &&
1400 !(req->rq_flags & RQF_QUIET)))
1401 print_req_error(req, error, __func__);
1403 blk_account_io_completion(req, nr_bytes);
1407 struct bio *bio = req->bio;
1408 unsigned bio_bytes = min(bio->bi_iter.bi_size, nr_bytes);
1410 if (bio_bytes == bio->bi_iter.bi_size)
1411 req->bio = bio->bi_next;
1413 /* Completion has already been traced */
1414 bio_clear_flag(bio, BIO_TRACE_COMPLETION);
1415 req_bio_endio(req, bio, bio_bytes, error);
1417 total_bytes += bio_bytes;
1418 nr_bytes -= bio_bytes;
1429 * Reset counters so that the request stacking driver
1430 * can find how many bytes remain in the request
1433 req->__data_len = 0;
1437 req->__data_len -= total_bytes;
1439 /* update sector only for requests with clear definition of sector */
1440 if (!blk_rq_is_passthrough(req))
1441 req->__sector += total_bytes >> 9;
1443 /* mixed attributes always follow the first bio */
1444 if (req->rq_flags & RQF_MIXED_MERGE) {
1445 req->cmd_flags &= ~REQ_FAILFAST_MASK;
1446 req->cmd_flags |= req->bio->bi_opf & REQ_FAILFAST_MASK;
1449 if (!(req->rq_flags & RQF_SPECIAL_PAYLOAD)) {
1451 * If total number of sectors is less than the first segment
1452 * size, something has gone terribly wrong.
1454 if (blk_rq_bytes(req) < blk_rq_cur_bytes(req)) {
1455 blk_dump_rq_flags(req, "request botched");
1456 req->__data_len = blk_rq_cur_bytes(req);
1459 /* recalculate the number of segments */
1460 req->nr_phys_segments = blk_recalc_rq_segments(req);
1465 EXPORT_SYMBOL_GPL(blk_update_request);
1467 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
1469 * rq_flush_dcache_pages - Helper function to flush all pages in a request
1470 * @rq: the request to be flushed
1473 * Flush all pages in @rq.
1475 void rq_flush_dcache_pages(struct request *rq)
1477 struct req_iterator iter;
1478 struct bio_vec bvec;
1480 rq_for_each_segment(bvec, rq, iter)
1481 flush_dcache_page(bvec.bv_page);
1483 EXPORT_SYMBOL_GPL(rq_flush_dcache_pages);
1487 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
1488 * @q : the queue of the device being checked
1491 * Check if underlying low-level drivers of a device are busy.
1492 * If the drivers want to export their busy state, they must set own
1493 * exporting function using blk_queue_lld_busy() first.
1495 * Basically, this function is used only by request stacking drivers
1496 * to stop dispatching requests to underlying devices when underlying
1497 * devices are busy. This behavior helps more I/O merging on the queue
1498 * of the request stacking driver and prevents I/O throughput regression
1499 * on burst I/O load.
1502 * 0 - Not busy (The request stacking driver should dispatch request)
1503 * 1 - Busy (The request stacking driver should stop dispatching request)
1505 int blk_lld_busy(struct request_queue *q)
1507 if (queue_is_mq(q) && q->mq_ops->busy)
1508 return q->mq_ops->busy(q);
1512 EXPORT_SYMBOL_GPL(blk_lld_busy);
1515 * blk_rq_unprep_clone - Helper function to free all bios in a cloned request
1516 * @rq: the clone request to be cleaned up
1519 * Free all bios in @rq for a cloned request.
1521 void blk_rq_unprep_clone(struct request *rq)
1525 while ((bio = rq->bio) != NULL) {
1526 rq->bio = bio->bi_next;
1531 EXPORT_SYMBOL_GPL(blk_rq_unprep_clone);
1534 * blk_rq_prep_clone - Helper function to setup clone request
1535 * @rq: the request to be setup
1536 * @rq_src: original request to be cloned
1537 * @bs: bio_set that bios for clone are allocated from
1538 * @gfp_mask: memory allocation mask for bio
1539 * @bio_ctr: setup function to be called for each clone bio.
1540 * Returns %0 for success, non %0 for failure.
1541 * @data: private data to be passed to @bio_ctr
1544 * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq.
1545 * Also, pages which the original bios are pointing to are not copied
1546 * and the cloned bios just point same pages.
1547 * So cloned bios must be completed before original bios, which means
1548 * the caller must complete @rq before @rq_src.
1550 int blk_rq_prep_clone(struct request *rq, struct request *rq_src,
1551 struct bio_set *bs, gfp_t gfp_mask,
1552 int (*bio_ctr)(struct bio *, struct bio *, void *),
1555 struct bio *bio, *bio_src;
1560 __rq_for_each_bio(bio_src, rq_src) {
1561 bio = bio_clone_fast(bio_src, gfp_mask, bs);
1565 if (bio_ctr && bio_ctr(bio, bio_src, data))
1569 rq->biotail->bi_next = bio;
1572 rq->bio = rq->biotail = bio;
1577 /* Copy attributes of the original request to the clone request. */
1578 rq->__sector = blk_rq_pos(rq_src);
1579 rq->__data_len = blk_rq_bytes(rq_src);
1580 if (rq_src->rq_flags & RQF_SPECIAL_PAYLOAD) {
1581 rq->rq_flags |= RQF_SPECIAL_PAYLOAD;
1582 rq->special_vec = rq_src->special_vec;
1584 rq->nr_phys_segments = rq_src->nr_phys_segments;
1585 rq->ioprio = rq_src->ioprio;
1587 if (rq->bio && blk_crypto_rq_bio_prep(rq, rq->bio, gfp_mask) < 0)
1595 blk_rq_unprep_clone(rq);
1599 EXPORT_SYMBOL_GPL(blk_rq_prep_clone);
1601 int kblockd_schedule_work(struct work_struct *work)
1603 return queue_work(kblockd_workqueue, work);
1605 EXPORT_SYMBOL(kblockd_schedule_work);
1607 int kblockd_mod_delayed_work_on(int cpu, struct delayed_work *dwork,
1608 unsigned long delay)
1610 return mod_delayed_work_on(cpu, kblockd_workqueue, dwork, delay);
1612 EXPORT_SYMBOL(kblockd_mod_delayed_work_on);
1615 * blk_start_plug - initialize blk_plug and track it inside the task_struct
1616 * @plug: The &struct blk_plug that needs to be initialized
1619 * blk_start_plug() indicates to the block layer an intent by the caller
1620 * to submit multiple I/O requests in a batch. The block layer may use
1621 * this hint to defer submitting I/Os from the caller until blk_finish_plug()
1622 * is called. However, the block layer may choose to submit requests
1623 * before a call to blk_finish_plug() if the number of queued I/Os
1624 * exceeds %BLK_MAX_REQUEST_COUNT, or if the size of the I/O is larger than
1625 * %BLK_PLUG_FLUSH_SIZE. The queued I/Os may also be submitted early if
1626 * the task schedules (see below).
1628 * Tracking blk_plug inside the task_struct will help with auto-flushing the
1629 * pending I/O should the task end up blocking between blk_start_plug() and
1630 * blk_finish_plug(). This is important from a performance perspective, but
1631 * also ensures that we don't deadlock. For instance, if the task is blocking
1632 * for a memory allocation, memory reclaim could end up wanting to free a
1633 * page belonging to that request that is currently residing in our private
1634 * plug. By flushing the pending I/O when the process goes to sleep, we avoid
1635 * this kind of deadlock.
1637 void blk_start_plug(struct blk_plug *plug)
1639 struct task_struct *tsk = current;
1642 * If this is a nested plug, don't actually assign it.
1647 INIT_LIST_HEAD(&plug->mq_list);
1648 INIT_LIST_HEAD(&plug->cb_list);
1650 plug->multiple_queues = false;
1651 plug->nowait = false;
1654 * Store ordering should not be needed here, since a potential
1655 * preempt will imply a full memory barrier
1659 EXPORT_SYMBOL(blk_start_plug);
1661 static void flush_plug_callbacks(struct blk_plug *plug, bool from_schedule)
1663 LIST_HEAD(callbacks);
1665 while (!list_empty(&plug->cb_list)) {
1666 list_splice_init(&plug->cb_list, &callbacks);
1668 while (!list_empty(&callbacks)) {
1669 struct blk_plug_cb *cb = list_first_entry(&callbacks,
1672 list_del(&cb->list);
1673 cb->callback(cb, from_schedule);
1678 struct blk_plug_cb *blk_check_plugged(blk_plug_cb_fn unplug, void *data,
1681 struct blk_plug *plug = current->plug;
1682 struct blk_plug_cb *cb;
1687 list_for_each_entry(cb, &plug->cb_list, list)
1688 if (cb->callback == unplug && cb->data == data)
1691 /* Not currently on the callback list */
1692 BUG_ON(size < sizeof(*cb));
1693 cb = kzalloc(size, GFP_ATOMIC);
1696 cb->callback = unplug;
1697 list_add(&cb->list, &plug->cb_list);
1701 EXPORT_SYMBOL(blk_check_plugged);
1703 void blk_flush_plug_list(struct blk_plug *plug, bool from_schedule)
1705 flush_plug_callbacks(plug, from_schedule);
1707 if (!list_empty(&plug->mq_list))
1708 blk_mq_flush_plug_list(plug, from_schedule);
1712 * blk_finish_plug - mark the end of a batch of submitted I/O
1713 * @plug: The &struct blk_plug passed to blk_start_plug()
1716 * Indicate that a batch of I/O submissions is complete. This function
1717 * must be paired with an initial call to blk_start_plug(). The intent
1718 * is to allow the block layer to optimize I/O submission. See the
1719 * documentation for blk_start_plug() for more information.
1721 void blk_finish_plug(struct blk_plug *plug)
1723 if (plug != current->plug)
1725 blk_flush_plug_list(plug, false);
1727 current->plug = NULL;
1729 EXPORT_SYMBOL(blk_finish_plug);
1731 void blk_io_schedule(void)
1733 /* Prevent hang_check timer from firing at us during very long I/O */
1734 unsigned long timeout = sysctl_hung_task_timeout_secs * HZ / 2;
1737 io_schedule_timeout(timeout);
1741 EXPORT_SYMBOL_GPL(blk_io_schedule);
1743 int __init blk_dev_init(void)
1745 BUILD_BUG_ON(REQ_OP_LAST >= (1 << REQ_OP_BITS));
1746 BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 *
1747 sizeof_field(struct request, cmd_flags));
1748 BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 *
1749 sizeof_field(struct bio, bi_opf));
1751 /* used for unplugging and affects IO latency/throughput - HIGHPRI */
1752 kblockd_workqueue = alloc_workqueue("kblockd",
1753 WQ_MEM_RECLAIM | WQ_HIGHPRI, 0);
1754 if (!kblockd_workqueue)
1755 panic("Failed to create kblockd\n");
1757 blk_requestq_cachep = kmem_cache_create("request_queue",
1758 sizeof(struct request_queue), 0, SLAB_PANIC, NULL);
1760 blk_debugfs_root = debugfs_create_dir("block", NULL);