1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (C) 1991, 1992 Linus Torvalds
4 * Copyright (C) 1994, Karl Keyte: Added support for disk statistics
5 * Elevator latency, (C) 2000 Andrea Arcangeli <andrea@suse.de> SuSE
6 * Queue request tables / lock, selectable elevator, Jens Axboe <axboe@suse.de>
7 * kernel-doc documentation started by NeilBrown <neilb@cse.unsw.edu.au>
9 * bio rewrite, highmem i/o, etc, Jens Axboe <axboe@suse.de> - may 2001
13 * This handles all read/write requests to block devices
15 #include <linux/kernel.h>
16 #include <linux/module.h>
17 #include <linux/backing-dev.h>
18 #include <linux/bio.h>
19 #include <linux/blkdev.h>
20 #include <linux/blk-mq.h>
21 #include <linux/highmem.h>
23 #include <linux/pagemap.h>
24 #include <linux/kernel_stat.h>
25 #include <linux/string.h>
26 #include <linux/init.h>
27 #include <linux/completion.h>
28 #include <linux/slab.h>
29 #include <linux/swap.h>
30 #include <linux/writeback.h>
31 #include <linux/task_io_accounting_ops.h>
32 #include <linux/fault-inject.h>
33 #include <linux/list_sort.h>
34 #include <linux/delay.h>
35 #include <linux/ratelimit.h>
36 #include <linux/pm_runtime.h>
37 #include <linux/blk-cgroup.h>
38 #include <linux/t10-pi.h>
39 #include <linux/debugfs.h>
40 #include <linux/bpf.h>
41 #include <linux/psi.h>
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);
62 DEFINE_IDA(blk_queue_ida);
65 * For queue allocation
67 struct kmem_cache *blk_requestq_cachep;
70 * Controlling structure to kblockd
72 static struct workqueue_struct *kblockd_workqueue;
75 * blk_queue_flag_set - atomically set a queue flag
76 * @flag: flag to be set
79 void blk_queue_flag_set(unsigned int flag, struct request_queue *q)
81 set_bit(flag, &q->queue_flags);
83 EXPORT_SYMBOL(blk_queue_flag_set);
86 * blk_queue_flag_clear - atomically clear a queue flag
87 * @flag: flag to be cleared
90 void blk_queue_flag_clear(unsigned int flag, struct request_queue *q)
92 clear_bit(flag, &q->queue_flags);
94 EXPORT_SYMBOL(blk_queue_flag_clear);
97 * blk_queue_flag_test_and_set - atomically test and set a queue flag
98 * @flag: flag to be set
101 * Returns the previous value of @flag - 0 if the flag was not set and 1 if
102 * the flag was already set.
104 bool blk_queue_flag_test_and_set(unsigned int flag, struct request_queue *q)
106 return test_and_set_bit(flag, &q->queue_flags);
108 EXPORT_SYMBOL_GPL(blk_queue_flag_test_and_set);
110 void blk_rq_init(struct request_queue *q, struct request *rq)
112 memset(rq, 0, sizeof(*rq));
114 INIT_LIST_HEAD(&rq->queuelist);
116 rq->__sector = (sector_t) -1;
117 INIT_HLIST_NODE(&rq->hash);
118 RB_CLEAR_NODE(&rq->rb_node);
119 rq->tag = BLK_MQ_NO_TAG;
120 rq->internal_tag = BLK_MQ_NO_TAG;
121 rq->start_time_ns = ktime_get_ns();
123 refcount_set(&rq->ref, 1);
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),
143 REQ_OP_NAME(SCSI_IN),
144 REQ_OP_NAME(SCSI_OUT),
146 REQ_OP_NAME(DRV_OUT),
151 * blk_op_str - Return string XXX in the REQ_OP_XXX.
154 * Description: Centralize block layer function to convert REQ_OP_XXX into
155 * string format. Useful in the debugging and tracing bio or request. For
156 * invalid REQ_OP_XXX it returns string "UNKNOWN".
158 inline const char *blk_op_str(unsigned int op)
160 const char *op_str = "UNKNOWN";
162 if (op < ARRAY_SIZE(blk_op_name) && blk_op_name[op])
163 op_str = blk_op_name[op];
167 EXPORT_SYMBOL_GPL(blk_op_str);
169 static const struct {
173 [BLK_STS_OK] = { 0, "" },
174 [BLK_STS_NOTSUPP] = { -EOPNOTSUPP, "operation not supported" },
175 [BLK_STS_TIMEOUT] = { -ETIMEDOUT, "timeout" },
176 [BLK_STS_NOSPC] = { -ENOSPC, "critical space allocation" },
177 [BLK_STS_TRANSPORT] = { -ENOLINK, "recoverable transport" },
178 [BLK_STS_TARGET] = { -EREMOTEIO, "critical target" },
179 [BLK_STS_NEXUS] = { -EBADE, "critical nexus" },
180 [BLK_STS_MEDIUM] = { -ENODATA, "critical medium" },
181 [BLK_STS_PROTECTION] = { -EILSEQ, "protection" },
182 [BLK_STS_RESOURCE] = { -ENOMEM, "kernel resource" },
183 [BLK_STS_DEV_RESOURCE] = { -EBUSY, "device resource" },
184 [BLK_STS_AGAIN] = { -EAGAIN, "nonblocking retry" },
186 /* device mapper special case, should not leak out: */
187 [BLK_STS_DM_REQUEUE] = { -EREMCHG, "dm internal retry" },
189 /* zone device specific errors */
190 [BLK_STS_ZONE_OPEN_RESOURCE] = { -ETOOMANYREFS, "open zones exceeded" },
191 [BLK_STS_ZONE_ACTIVE_RESOURCE] = { -EOVERFLOW, "active zones exceeded" },
193 /* everything else not covered above: */
194 [BLK_STS_IOERR] = { -EIO, "I/O" },
197 blk_status_t errno_to_blk_status(int errno)
201 for (i = 0; i < ARRAY_SIZE(blk_errors); i++) {
202 if (blk_errors[i].errno == errno)
203 return (__force blk_status_t)i;
206 return BLK_STS_IOERR;
208 EXPORT_SYMBOL_GPL(errno_to_blk_status);
210 int blk_status_to_errno(blk_status_t status)
212 int idx = (__force int)status;
214 if (WARN_ON_ONCE(idx >= ARRAY_SIZE(blk_errors)))
216 return blk_errors[idx].errno;
218 EXPORT_SYMBOL_GPL(blk_status_to_errno);
220 static void print_req_error(struct request *req, blk_status_t status,
223 int idx = (__force int)status;
225 if (WARN_ON_ONCE(idx >= ARRAY_SIZE(blk_errors)))
228 printk_ratelimited(KERN_ERR
229 "%s: %s error, dev %s, sector %llu op 0x%x:(%s) flags 0x%x "
230 "phys_seg %u prio class %u\n",
231 caller, blk_errors[idx].name,
232 req->rq_disk ? req->rq_disk->disk_name : "?",
233 blk_rq_pos(req), req_op(req), blk_op_str(req_op(req)),
234 req->cmd_flags & ~REQ_OP_MASK,
235 req->nr_phys_segments,
236 IOPRIO_PRIO_CLASS(req->ioprio));
239 static void req_bio_endio(struct request *rq, struct bio *bio,
240 unsigned int nbytes, blk_status_t error)
243 bio->bi_status = error;
245 if (unlikely(rq->rq_flags & RQF_QUIET))
246 bio_set_flag(bio, BIO_QUIET);
248 bio_advance(bio, nbytes);
250 if (req_op(rq) == REQ_OP_ZONE_APPEND && error == BLK_STS_OK) {
252 * Partial zone append completions cannot be supported as the
253 * BIO fragments may end up not being written sequentially.
255 if (bio->bi_iter.bi_size)
256 bio->bi_status = BLK_STS_IOERR;
258 bio->bi_iter.bi_sector = rq->__sector;
261 /* don't actually finish bio if it's part of flush sequence */
262 if (bio->bi_iter.bi_size == 0 && !(rq->rq_flags & RQF_FLUSH_SEQ))
266 void blk_dump_rq_flags(struct request *rq, char *msg)
268 printk(KERN_INFO "%s: dev %s: flags=%llx\n", msg,
269 rq->rq_disk ? rq->rq_disk->disk_name : "?",
270 (unsigned long long) rq->cmd_flags);
272 printk(KERN_INFO " sector %llu, nr/cnr %u/%u\n",
273 (unsigned long long)blk_rq_pos(rq),
274 blk_rq_sectors(rq), blk_rq_cur_sectors(rq));
275 printk(KERN_INFO " bio %p, biotail %p, len %u\n",
276 rq->bio, rq->biotail, blk_rq_bytes(rq));
278 EXPORT_SYMBOL(blk_dump_rq_flags);
281 * blk_sync_queue - cancel any pending callbacks on a queue
285 * The block layer may perform asynchronous callback activity
286 * on a queue, such as calling the unplug function after a timeout.
287 * A block device may call blk_sync_queue to ensure that any
288 * such activity is cancelled, thus allowing it to release resources
289 * that the callbacks might use. The caller must already have made sure
290 * that its ->submit_bio will not re-add plugging prior to calling
293 * This function does not cancel any asynchronous activity arising
294 * out of elevator or throttling code. That would require elevator_exit()
295 * and blkcg_exit_queue() to be called with queue lock initialized.
298 void blk_sync_queue(struct request_queue *q)
300 del_timer_sync(&q->timeout);
301 cancel_work_sync(&q->timeout_work);
303 EXPORT_SYMBOL(blk_sync_queue);
306 * blk_set_pm_only - increment pm_only counter
307 * @q: request queue pointer
309 void blk_set_pm_only(struct request_queue *q)
311 atomic_inc(&q->pm_only);
313 EXPORT_SYMBOL_GPL(blk_set_pm_only);
315 void blk_clear_pm_only(struct request_queue *q)
319 pm_only = atomic_dec_return(&q->pm_only);
320 WARN_ON_ONCE(pm_only < 0);
322 wake_up_all(&q->mq_freeze_wq);
324 EXPORT_SYMBOL_GPL(blk_clear_pm_only);
327 * blk_put_queue - decrement the request_queue refcount
328 * @q: the request_queue structure to decrement the refcount for
330 * Decrements the refcount of the request_queue kobject. When this reaches 0
331 * we'll have blk_release_queue() called.
333 * Context: Any context, but the last reference must not be dropped from
336 void blk_put_queue(struct request_queue *q)
338 kobject_put(&q->kobj);
340 EXPORT_SYMBOL(blk_put_queue);
342 void blk_set_queue_dying(struct request_queue *q)
344 blk_queue_flag_set(QUEUE_FLAG_DYING, q);
347 * When queue DYING flag is set, we need to block new req
348 * entering queue, so we call blk_freeze_queue_start() to
349 * prevent I/O from crossing blk_queue_enter().
351 blk_freeze_queue_start(q);
354 blk_mq_wake_waiters(q);
356 /* Make blk_queue_enter() reexamine the DYING flag. */
357 wake_up_all(&q->mq_freeze_wq);
359 EXPORT_SYMBOL_GPL(blk_set_queue_dying);
362 * blk_cleanup_queue - shutdown a request queue
363 * @q: request queue to shutdown
365 * Mark @q DYING, drain all pending requests, mark @q DEAD, destroy and
366 * put it. All future requests will be failed immediately with -ENODEV.
370 void blk_cleanup_queue(struct request_queue *q)
372 /* cannot be called from atomic context */
375 WARN_ON_ONCE(blk_queue_registered(q));
377 /* mark @q DYING, no new request or merges will be allowed afterwards */
378 blk_set_queue_dying(q);
380 blk_queue_flag_set(QUEUE_FLAG_NOMERGES, q);
381 blk_queue_flag_set(QUEUE_FLAG_NOXMERGES, q);
384 * Drain all requests queued before DYING marking. Set DEAD flag to
385 * prevent that blk_mq_run_hw_queues() accesses the hardware queues
386 * after draining finished.
392 blk_queue_flag_set(QUEUE_FLAG_DEAD, q);
394 /* for synchronous bio-based driver finish in-flight integrity i/o */
395 blk_flush_integrity();
397 /* @q won't process any more request, flush async actions */
398 del_timer_sync(&q->backing_dev_info->laptop_mode_wb_timer);
402 blk_mq_exit_queue(q);
405 * In theory, request pool of sched_tags belongs to request queue.
406 * However, the current implementation requires tag_set for freeing
407 * requests, so free the pool now.
409 * Queue has become frozen, there can't be any in-queue requests, so
410 * it is safe to free requests now.
412 mutex_lock(&q->sysfs_lock);
414 blk_mq_sched_free_requests(q);
415 mutex_unlock(&q->sysfs_lock);
417 percpu_ref_exit(&q->q_usage_counter);
419 /* @q is and will stay empty, shutdown and put */
422 EXPORT_SYMBOL(blk_cleanup_queue);
425 * blk_queue_enter() - try to increase q->q_usage_counter
426 * @q: request queue pointer
427 * @flags: BLK_MQ_REQ_NOWAIT and/or BLK_MQ_REQ_PREEMPT
429 int blk_queue_enter(struct request_queue *q, blk_mq_req_flags_t flags)
431 const bool pm = flags & BLK_MQ_REQ_PREEMPT;
434 bool success = false;
437 if (percpu_ref_tryget_live(&q->q_usage_counter)) {
439 * The code that increments the pm_only counter is
440 * responsible for ensuring that that counter is
441 * globally visible before the queue is unfrozen.
443 if (pm || !blk_queue_pm_only(q)) {
446 percpu_ref_put(&q->q_usage_counter);
454 if (flags & BLK_MQ_REQ_NOWAIT)
458 * read pair of barrier in blk_freeze_queue_start(),
459 * we need to order reading __PERCPU_REF_DEAD flag of
460 * .q_usage_counter and reading .mq_freeze_depth or
461 * queue dying flag, otherwise the following wait may
462 * never return if the two reads are reordered.
466 wait_event(q->mq_freeze_wq,
467 (!q->mq_freeze_depth &&
468 (pm || (blk_pm_request_resume(q),
469 !blk_queue_pm_only(q)))) ||
471 if (blk_queue_dying(q))
476 static inline int bio_queue_enter(struct bio *bio)
478 struct request_queue *q = bio->bi_disk->queue;
479 bool nowait = bio->bi_opf & REQ_NOWAIT;
482 ret = blk_queue_enter(q, nowait ? BLK_MQ_REQ_NOWAIT : 0);
484 if (nowait && !blk_queue_dying(q))
485 bio_wouldblock_error(bio);
493 void blk_queue_exit(struct request_queue *q)
495 percpu_ref_put(&q->q_usage_counter);
498 static void blk_queue_usage_counter_release(struct percpu_ref *ref)
500 struct request_queue *q =
501 container_of(ref, struct request_queue, q_usage_counter);
503 wake_up_all(&q->mq_freeze_wq);
506 static void blk_rq_timed_out_timer(struct timer_list *t)
508 struct request_queue *q = from_timer(q, t, timeout);
510 kblockd_schedule_work(&q->timeout_work);
513 static void blk_timeout_work(struct work_struct *work)
517 struct request_queue *blk_alloc_queue(int node_id)
519 struct request_queue *q;
522 q = kmem_cache_alloc_node(blk_requestq_cachep,
523 GFP_KERNEL | __GFP_ZERO, node_id);
527 q->last_merge = NULL;
529 q->id = ida_simple_get(&blk_queue_ida, 0, 0, GFP_KERNEL);
533 ret = bioset_init(&q->bio_split, BIO_POOL_SIZE, 0, BIOSET_NEED_BVECS);
537 q->backing_dev_info = bdi_alloc(node_id);
538 if (!q->backing_dev_info)
541 q->stats = blk_alloc_queue_stats();
547 atomic_set(&q->nr_active_requests_shared_sbitmap, 0);
549 timer_setup(&q->backing_dev_info->laptop_mode_wb_timer,
550 laptop_mode_timer_fn, 0);
551 timer_setup(&q->timeout, blk_rq_timed_out_timer, 0);
552 INIT_WORK(&q->timeout_work, blk_timeout_work);
553 INIT_LIST_HEAD(&q->icq_list);
554 #ifdef CONFIG_BLK_CGROUP
555 INIT_LIST_HEAD(&q->blkg_list);
558 kobject_init(&q->kobj, &blk_queue_ktype);
560 mutex_init(&q->debugfs_mutex);
561 mutex_init(&q->sysfs_lock);
562 mutex_init(&q->sysfs_dir_lock);
563 spin_lock_init(&q->queue_lock);
565 init_waitqueue_head(&q->mq_freeze_wq);
566 mutex_init(&q->mq_freeze_lock);
569 * Init percpu_ref in atomic mode so that it's faster to shutdown.
570 * See blk_register_queue() for details.
572 if (percpu_ref_init(&q->q_usage_counter,
573 blk_queue_usage_counter_release,
574 PERCPU_REF_INIT_ATOMIC, GFP_KERNEL))
577 if (blkcg_init_queue(q))
580 blk_queue_dma_alignment(q, 511);
581 blk_set_default_limits(&q->limits);
582 q->nr_requests = BLKDEV_MAX_RQ;
587 percpu_ref_exit(&q->q_usage_counter);
589 blk_free_queue_stats(q->stats);
591 bdi_put(q->backing_dev_info);
593 bioset_exit(&q->bio_split);
595 ida_simple_remove(&blk_queue_ida, q->id);
597 kmem_cache_free(blk_requestq_cachep, q);
600 EXPORT_SYMBOL(blk_alloc_queue);
603 * blk_get_queue - increment the request_queue refcount
604 * @q: the request_queue structure to increment the refcount for
606 * Increment the refcount of the request_queue kobject.
608 * Context: Any context.
610 bool blk_get_queue(struct request_queue *q)
612 if (likely(!blk_queue_dying(q))) {
619 EXPORT_SYMBOL(blk_get_queue);
622 * blk_get_request - allocate a request
623 * @q: request queue to allocate a request for
624 * @op: operation (REQ_OP_*) and REQ_* flags, e.g. REQ_SYNC.
625 * @flags: BLK_MQ_REQ_* flags, e.g. BLK_MQ_REQ_NOWAIT.
627 struct request *blk_get_request(struct request_queue *q, unsigned int op,
628 blk_mq_req_flags_t flags)
632 WARN_ON_ONCE(op & REQ_NOWAIT);
633 WARN_ON_ONCE(flags & ~(BLK_MQ_REQ_NOWAIT | BLK_MQ_REQ_PREEMPT));
635 req = blk_mq_alloc_request(q, op, flags);
636 if (!IS_ERR(req) && q->mq_ops->initialize_rq_fn)
637 q->mq_ops->initialize_rq_fn(req);
641 EXPORT_SYMBOL(blk_get_request);
643 void blk_put_request(struct request *req)
645 blk_mq_free_request(req);
647 EXPORT_SYMBOL(blk_put_request);
649 static void handle_bad_sector(struct bio *bio, sector_t maxsector)
651 char b[BDEVNAME_SIZE];
653 pr_info_ratelimited("attempt to access beyond end of device\n"
654 "%s: rw=%d, want=%llu, limit=%llu\n",
655 bio_devname(bio, b), bio->bi_opf,
656 bio_end_sector(bio), maxsector);
659 #ifdef CONFIG_FAIL_MAKE_REQUEST
661 static DECLARE_FAULT_ATTR(fail_make_request);
663 static int __init setup_fail_make_request(char *str)
665 return setup_fault_attr(&fail_make_request, str);
667 __setup("fail_make_request=", setup_fail_make_request);
669 static bool should_fail_request(struct block_device *part, unsigned int bytes)
671 return part->bd_make_it_fail && should_fail(&fail_make_request, bytes);
674 static int __init fail_make_request_debugfs(void)
676 struct dentry *dir = fault_create_debugfs_attr("fail_make_request",
677 NULL, &fail_make_request);
679 return PTR_ERR_OR_ZERO(dir);
682 late_initcall(fail_make_request_debugfs);
684 #else /* CONFIG_FAIL_MAKE_REQUEST */
686 static inline bool should_fail_request(struct block_device *part,
692 #endif /* CONFIG_FAIL_MAKE_REQUEST */
694 static inline bool bio_check_ro(struct bio *bio, struct block_device *part)
696 const int op = bio_op(bio);
698 if (part->bd_read_only && op_is_write(op)) {
699 char b[BDEVNAME_SIZE];
701 if (op_is_flush(bio->bi_opf) && !bio_sectors(bio))
705 "Trying to write to read-only block-device %s (partno %d)\n",
706 bio_devname(bio, b), part->bd_partno);
707 /* Older lvm-tools actually trigger this */
714 static noinline int should_fail_bio(struct bio *bio)
716 if (should_fail_request(bio->bi_disk->part0, bio->bi_iter.bi_size))
720 ALLOW_ERROR_INJECTION(should_fail_bio, ERRNO);
723 * Check whether this bio extends beyond the end of the device or partition.
724 * This may well happen - the kernel calls bread() without checking the size of
725 * the device, e.g., when mounting a file system.
727 static inline int bio_check_eod(struct bio *bio, sector_t maxsector)
729 unsigned int nr_sectors = bio_sectors(bio);
731 if (nr_sectors && maxsector &&
732 (nr_sectors > maxsector ||
733 bio->bi_iter.bi_sector > maxsector - nr_sectors)) {
734 handle_bad_sector(bio, maxsector);
741 * Remap block n of partition p to block n+start(p) of the disk.
743 static inline int blk_partition_remap(struct bio *bio)
745 struct block_device *p;
749 p = __disk_get_part(bio->bi_disk, bio->bi_partno);
752 if (unlikely(should_fail_request(p, bio->bi_iter.bi_size)))
754 if (unlikely(bio_check_ro(bio, p)))
757 if (bio_sectors(bio)) {
758 if (bio_check_eod(bio, bdev_nr_sectors(p)))
760 bio->bi_iter.bi_sector += p->bd_start_sect;
761 trace_block_bio_remap(bio, p->bd_dev,
762 bio->bi_iter.bi_sector -
773 * Check write append to a zoned block device.
775 static inline blk_status_t blk_check_zone_append(struct request_queue *q,
778 sector_t pos = bio->bi_iter.bi_sector;
779 int nr_sectors = bio_sectors(bio);
781 /* Only applicable to zoned block devices */
782 if (!blk_queue_is_zoned(q))
783 return BLK_STS_NOTSUPP;
785 /* The bio sector must point to the start of a sequential zone */
786 if (pos & (blk_queue_zone_sectors(q) - 1) ||
787 !blk_queue_zone_is_seq(q, pos))
788 return BLK_STS_IOERR;
791 * Not allowed to cross zone boundaries. Otherwise, the BIO will be
792 * split and could result in non-contiguous sectors being written in
795 if (nr_sectors > q->limits.chunk_sectors)
796 return BLK_STS_IOERR;
798 /* Make sure the BIO is small enough and will not get split */
799 if (nr_sectors > q->limits.max_zone_append_sectors)
800 return BLK_STS_IOERR;
802 bio->bi_opf |= REQ_NOMERGE;
807 static noinline_for_stack bool submit_bio_checks(struct bio *bio)
809 struct request_queue *q = bio->bi_disk->queue;
810 blk_status_t status = BLK_STS_IOERR;
811 struct blk_plug *plug;
815 plug = blk_mq_plug(q, bio);
816 if (plug && plug->nowait)
817 bio->bi_opf |= REQ_NOWAIT;
820 * For a REQ_NOWAIT based request, return -EOPNOTSUPP
821 * if queue does not support NOWAIT.
823 if ((bio->bi_opf & REQ_NOWAIT) && !blk_queue_nowait(q))
826 if (should_fail_bio(bio))
829 if (bio->bi_partno) {
830 if (unlikely(blk_partition_remap(bio)))
833 if (unlikely(bio_check_ro(bio, bio->bi_disk->part0)))
835 if (unlikely(bio_check_eod(bio, get_capacity(bio->bi_disk))))
840 * Filter flush bio's early so that bio based drivers without flush
841 * support don't have to worry about them.
843 if (op_is_flush(bio->bi_opf) &&
844 !test_bit(QUEUE_FLAG_WC, &q->queue_flags)) {
845 bio->bi_opf &= ~(REQ_PREFLUSH | REQ_FUA);
846 if (!bio_sectors(bio)) {
852 if (!test_bit(QUEUE_FLAG_POLL, &q->queue_flags))
853 bio->bi_opf &= ~REQ_HIPRI;
855 switch (bio_op(bio)) {
857 if (!blk_queue_discard(q))
860 case REQ_OP_SECURE_ERASE:
861 if (!blk_queue_secure_erase(q))
864 case REQ_OP_WRITE_SAME:
865 if (!q->limits.max_write_same_sectors)
868 case REQ_OP_ZONE_APPEND:
869 status = blk_check_zone_append(q, bio);
870 if (status != BLK_STS_OK)
873 case REQ_OP_ZONE_RESET:
874 case REQ_OP_ZONE_OPEN:
875 case REQ_OP_ZONE_CLOSE:
876 case REQ_OP_ZONE_FINISH:
877 if (!blk_queue_is_zoned(q))
880 case REQ_OP_ZONE_RESET_ALL:
881 if (!blk_queue_is_zoned(q) || !blk_queue_zone_resetall(q))
884 case REQ_OP_WRITE_ZEROES:
885 if (!q->limits.max_write_zeroes_sectors)
893 * Various block parts want %current->io_context, so allocate it up
894 * front rather than dealing with lots of pain to allocate it only
895 * where needed. This may fail and the block layer knows how to live
898 if (unlikely(!current->io_context))
899 create_task_io_context(current, GFP_ATOMIC, q->node);
901 if (blk_throtl_bio(bio)) {
902 blkcg_bio_issue_init(bio);
906 blk_cgroup_bio_start(bio);
907 blkcg_bio_issue_init(bio);
909 if (!bio_flagged(bio, BIO_TRACE_COMPLETION)) {
910 trace_block_bio_queue(bio);
911 /* Now that enqueuing has been traced, we need to trace
912 * completion as well.
914 bio_set_flag(bio, BIO_TRACE_COMPLETION);
919 status = BLK_STS_NOTSUPP;
921 bio->bi_status = status;
926 static blk_qc_t __submit_bio(struct bio *bio)
928 struct gendisk *disk = bio->bi_disk;
929 blk_qc_t ret = BLK_QC_T_NONE;
931 if (blk_crypto_bio_prep(&bio)) {
932 if (!disk->fops->submit_bio)
933 return blk_mq_submit_bio(bio);
934 ret = disk->fops->submit_bio(bio);
936 blk_queue_exit(disk->queue);
941 * The loop in this function may be a bit non-obvious, and so deserves some
944 * - Before entering the loop, bio->bi_next is NULL (as all callers ensure
945 * that), so we have a list with a single bio.
946 * - We pretend that we have just taken it off a longer list, so we assign
947 * bio_list to a pointer to the bio_list_on_stack, thus initialising the
948 * bio_list of new bios to be added. ->submit_bio() may indeed add some more
949 * bios through a recursive call to submit_bio_noacct. If it did, we find a
950 * non-NULL value in bio_list and re-enter the loop from the top.
951 * - In this case we really did just take the bio of the top of the list (no
952 * pretending) and so remove it from bio_list, and call into ->submit_bio()
955 * bio_list_on_stack[0] contains bios submitted by the current ->submit_bio.
956 * bio_list_on_stack[1] contains bios that were submitted before the current
957 * ->submit_bio_bio, but that haven't been processed yet.
959 static blk_qc_t __submit_bio_noacct(struct bio *bio)
961 struct bio_list bio_list_on_stack[2];
962 blk_qc_t ret = BLK_QC_T_NONE;
964 BUG_ON(bio->bi_next);
966 bio_list_init(&bio_list_on_stack[0]);
967 current->bio_list = bio_list_on_stack;
970 struct request_queue *q = bio->bi_disk->queue;
971 struct bio_list lower, same;
973 if (unlikely(bio_queue_enter(bio) != 0))
977 * Create a fresh bio_list for all subordinate requests.
979 bio_list_on_stack[1] = bio_list_on_stack[0];
980 bio_list_init(&bio_list_on_stack[0]);
982 ret = __submit_bio(bio);
985 * Sort new bios into those for a lower level and those for the
988 bio_list_init(&lower);
989 bio_list_init(&same);
990 while ((bio = bio_list_pop(&bio_list_on_stack[0])) != NULL)
991 if (q == bio->bi_disk->queue)
992 bio_list_add(&same, bio);
994 bio_list_add(&lower, bio);
997 * Now assemble so we handle the lowest level first.
999 bio_list_merge(&bio_list_on_stack[0], &lower);
1000 bio_list_merge(&bio_list_on_stack[0], &same);
1001 bio_list_merge(&bio_list_on_stack[0], &bio_list_on_stack[1]);
1002 } while ((bio = bio_list_pop(&bio_list_on_stack[0])));
1004 current->bio_list = NULL;
1008 static blk_qc_t __submit_bio_noacct_mq(struct bio *bio)
1010 struct bio_list bio_list[2] = { };
1011 blk_qc_t ret = BLK_QC_T_NONE;
1013 current->bio_list = bio_list;
1016 struct gendisk *disk = bio->bi_disk;
1018 if (unlikely(bio_queue_enter(bio) != 0))
1021 if (!blk_crypto_bio_prep(&bio)) {
1022 blk_queue_exit(disk->queue);
1023 ret = BLK_QC_T_NONE;
1027 ret = blk_mq_submit_bio(bio);
1028 } while ((bio = bio_list_pop(&bio_list[0])));
1030 current->bio_list = NULL;
1035 * submit_bio_noacct - re-submit a bio to the block device layer for I/O
1036 * @bio: The bio describing the location in memory and on the device.
1038 * This is a version of submit_bio() that shall only be used for I/O that is
1039 * resubmitted to lower level drivers by stacking block drivers. All file
1040 * systems and other upper level users of the block layer should use
1041 * submit_bio() instead.
1043 blk_qc_t submit_bio_noacct(struct bio *bio)
1045 if (!submit_bio_checks(bio))
1046 return BLK_QC_T_NONE;
1049 * We only want one ->submit_bio to be active at a time, else stack
1050 * usage with stacked devices could be a problem. Use current->bio_list
1051 * to collect a list of requests submited by a ->submit_bio method while
1052 * it is active, and then process them after it returned.
1054 if (current->bio_list) {
1055 bio_list_add(¤t->bio_list[0], bio);
1056 return BLK_QC_T_NONE;
1059 if (!bio->bi_disk->fops->submit_bio)
1060 return __submit_bio_noacct_mq(bio);
1061 return __submit_bio_noacct(bio);
1063 EXPORT_SYMBOL(submit_bio_noacct);
1066 * submit_bio - submit a bio to the block device layer for I/O
1067 * @bio: The &struct bio which describes the I/O
1069 * submit_bio() is used to submit I/O requests to block devices. It is passed a
1070 * fully set up &struct bio that describes the I/O that needs to be done. The
1071 * bio will be send to the device described by the bi_disk and bi_partno fields.
1073 * The success/failure status of the request, along with notification of
1074 * completion, is delivered asynchronously through the ->bi_end_io() callback
1075 * in @bio. The bio must NOT be touched by thecaller until ->bi_end_io() has
1078 blk_qc_t submit_bio(struct bio *bio)
1080 if (blkcg_punt_bio_submit(bio))
1081 return BLK_QC_T_NONE;
1084 * If it's a regular read/write or a barrier with data attached,
1085 * go through the normal accounting stuff before submission.
1087 if (bio_has_data(bio)) {
1090 if (unlikely(bio_op(bio) == REQ_OP_WRITE_SAME))
1091 count = queue_logical_block_size(bio->bi_disk->queue) >> 9;
1093 count = bio_sectors(bio);
1095 if (op_is_write(bio_op(bio))) {
1096 count_vm_events(PGPGOUT, count);
1098 task_io_account_read(bio->bi_iter.bi_size);
1099 count_vm_events(PGPGIN, count);
1102 if (unlikely(block_dump)) {
1103 char b[BDEVNAME_SIZE];
1104 printk(KERN_DEBUG "%s(%d): %s block %Lu on %s (%u sectors)\n",
1105 current->comm, task_pid_nr(current),
1106 op_is_write(bio_op(bio)) ? "WRITE" : "READ",
1107 (unsigned long long)bio->bi_iter.bi_sector,
1108 bio_devname(bio, b), count);
1113 * If we're reading data that is part of the userspace workingset, count
1114 * submission time as memory stall. When the device is congested, or
1115 * the submitting cgroup IO-throttled, submission can be a significant
1116 * part of overall IO time.
1118 if (unlikely(bio_op(bio) == REQ_OP_READ &&
1119 bio_flagged(bio, BIO_WORKINGSET))) {
1120 unsigned long pflags;
1123 psi_memstall_enter(&pflags);
1124 ret = submit_bio_noacct(bio);
1125 psi_memstall_leave(&pflags);
1130 return submit_bio_noacct(bio);
1132 EXPORT_SYMBOL(submit_bio);
1135 * blk_cloned_rq_check_limits - Helper function to check a cloned request
1136 * for the new queue limits
1138 * @rq: the request being checked
1141 * @rq may have been made based on weaker limitations of upper-level queues
1142 * in request stacking drivers, and it may violate the limitation of @q.
1143 * Since the block layer and the underlying device driver trust @rq
1144 * after it is inserted to @q, it should be checked against @q before
1145 * the insertion using this generic function.
1147 * Request stacking drivers like request-based dm may change the queue
1148 * limits when retrying requests on other queues. Those requests need
1149 * to be checked against the new queue limits again during dispatch.
1151 static blk_status_t blk_cloned_rq_check_limits(struct request_queue *q,
1154 unsigned int max_sectors = blk_queue_get_max_sectors(q, req_op(rq));
1156 if (blk_rq_sectors(rq) > max_sectors) {
1158 * SCSI device does not have a good way to return if
1159 * Write Same/Zero is actually supported. If a device rejects
1160 * a non-read/write command (discard, write same,etc.) the
1161 * low-level device driver will set the relevant queue limit to
1162 * 0 to prevent blk-lib from issuing more of the offending
1163 * operations. Commands queued prior to the queue limit being
1164 * reset need to be completed with BLK_STS_NOTSUPP to avoid I/O
1165 * errors being propagated to upper layers.
1167 if (max_sectors == 0)
1168 return BLK_STS_NOTSUPP;
1170 printk(KERN_ERR "%s: over max size limit. (%u > %u)\n",
1171 __func__, blk_rq_sectors(rq), max_sectors);
1172 return BLK_STS_IOERR;
1176 * queue's settings related to segment counting like q->bounce_pfn
1177 * may differ from that of other stacking queues.
1178 * Recalculate it to check the request correctly on this queue's
1181 rq->nr_phys_segments = blk_recalc_rq_segments(rq);
1182 if (rq->nr_phys_segments > queue_max_segments(q)) {
1183 printk(KERN_ERR "%s: over max segments limit. (%hu > %hu)\n",
1184 __func__, rq->nr_phys_segments, queue_max_segments(q));
1185 return BLK_STS_IOERR;
1192 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
1193 * @q: the queue to submit the request
1194 * @rq: the request being queued
1196 blk_status_t blk_insert_cloned_request(struct request_queue *q, struct request *rq)
1200 ret = blk_cloned_rq_check_limits(q, rq);
1201 if (ret != BLK_STS_OK)
1205 should_fail_request(rq->rq_disk->part0, blk_rq_bytes(rq)))
1206 return BLK_STS_IOERR;
1208 if (blk_crypto_insert_cloned_request(rq))
1209 return BLK_STS_IOERR;
1211 if (blk_queue_io_stat(q))
1212 blk_account_io_start(rq);
1215 * Since we have a scheduler attached on the top device,
1216 * bypass a potential scheduler on the bottom device for
1219 return blk_mq_request_issue_directly(rq, true);
1221 EXPORT_SYMBOL_GPL(blk_insert_cloned_request);
1224 * blk_rq_err_bytes - determine number of bytes till the next failure boundary
1225 * @rq: request to examine
1228 * A request could be merge of IOs which require different failure
1229 * handling. This function determines the number of bytes which
1230 * can be failed from the beginning of the request without
1231 * crossing into area which need to be retried further.
1234 * The number of bytes to fail.
1236 unsigned int blk_rq_err_bytes(const struct request *rq)
1238 unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK;
1239 unsigned int bytes = 0;
1242 if (!(rq->rq_flags & RQF_MIXED_MERGE))
1243 return blk_rq_bytes(rq);
1246 * Currently the only 'mixing' which can happen is between
1247 * different fastfail types. We can safely fail portions
1248 * which have all the failfast bits that the first one has -
1249 * the ones which are at least as eager to fail as the first
1252 for (bio = rq->bio; bio; bio = bio->bi_next) {
1253 if ((bio->bi_opf & ff) != ff)
1255 bytes += bio->bi_iter.bi_size;
1258 /* this could lead to infinite loop */
1259 BUG_ON(blk_rq_bytes(rq) && !bytes);
1262 EXPORT_SYMBOL_GPL(blk_rq_err_bytes);
1264 static void update_io_ticks(struct block_device *part, unsigned long now,
1267 unsigned long stamp;
1269 stamp = READ_ONCE(part->bd_stamp);
1270 if (unlikely(stamp != now)) {
1271 if (likely(cmpxchg(&part->bd_stamp, stamp, now) == stamp))
1272 __part_stat_add(part, io_ticks, end ? now - stamp : 1);
1274 if (part->bd_partno) {
1275 part = bdev_whole(part);
1280 static void blk_account_io_completion(struct request *req, unsigned int bytes)
1282 if (req->part && blk_do_io_stat(req)) {
1283 const int sgrp = op_stat_group(req_op(req));
1286 part_stat_add(req->part, sectors[sgrp], bytes >> 9);
1291 void blk_account_io_done(struct request *req, u64 now)
1294 * Account IO completion. flush_rq isn't accounted as a
1295 * normal IO on queueing nor completion. Accounting the
1296 * containing request is enough.
1298 if (req->part && blk_do_io_stat(req) &&
1299 !(req->rq_flags & RQF_FLUSH_SEQ)) {
1300 const int sgrp = op_stat_group(req_op(req));
1303 update_io_ticks(req->part, jiffies, true);
1304 part_stat_inc(req->part, ios[sgrp]);
1305 part_stat_add(req->part, nsecs[sgrp], now - req->start_time_ns);
1310 void blk_account_io_start(struct request *rq)
1312 if (!blk_do_io_stat(rq))
1315 rq->part = disk_map_sector_rcu(rq->rq_disk, blk_rq_pos(rq));
1318 update_io_ticks(rq->part, jiffies, false);
1322 static unsigned long __part_start_io_acct(struct block_device *part,
1323 unsigned int sectors, unsigned int op)
1325 const int sgrp = op_stat_group(op);
1326 unsigned long now = READ_ONCE(jiffies);
1329 update_io_ticks(part, now, false);
1330 part_stat_inc(part, ios[sgrp]);
1331 part_stat_add(part, sectors[sgrp], sectors);
1332 part_stat_local_inc(part, in_flight[op_is_write(op)]);
1338 unsigned long part_start_io_acct(struct gendisk *disk, struct block_device **part,
1341 *part = disk_map_sector_rcu(disk, bio->bi_iter.bi_sector);
1343 return __part_start_io_acct(*part, bio_sectors(bio), bio_op(bio));
1345 EXPORT_SYMBOL_GPL(part_start_io_acct);
1347 unsigned long disk_start_io_acct(struct gendisk *disk, unsigned int sectors,
1350 return __part_start_io_acct(disk->part0, sectors, op);
1352 EXPORT_SYMBOL(disk_start_io_acct);
1354 static void __part_end_io_acct(struct block_device *part, unsigned int op,
1355 unsigned long start_time)
1357 const int sgrp = op_stat_group(op);
1358 unsigned long now = READ_ONCE(jiffies);
1359 unsigned long duration = now - start_time;
1362 update_io_ticks(part, now, true);
1363 part_stat_add(part, nsecs[sgrp], jiffies_to_nsecs(duration));
1364 part_stat_local_dec(part, in_flight[op_is_write(op)]);
1368 void part_end_io_acct(struct block_device *part, struct bio *bio,
1369 unsigned long start_time)
1371 __part_end_io_acct(part, bio_op(bio), start_time);
1373 EXPORT_SYMBOL_GPL(part_end_io_acct);
1375 void disk_end_io_acct(struct gendisk *disk, unsigned int op,
1376 unsigned long start_time)
1378 __part_end_io_acct(disk->part0, op, start_time);
1380 EXPORT_SYMBOL(disk_end_io_acct);
1383 * Steal bios from a request and add them to a bio list.
1384 * The request must not have been partially completed before.
1386 void blk_steal_bios(struct bio_list *list, struct request *rq)
1390 list->tail->bi_next = rq->bio;
1392 list->head = rq->bio;
1393 list->tail = rq->biotail;
1401 EXPORT_SYMBOL_GPL(blk_steal_bios);
1404 * blk_update_request - Special helper function for request stacking drivers
1405 * @req: the request being processed
1406 * @error: block status code
1407 * @nr_bytes: number of bytes to complete @req
1410 * Ends I/O on a number of bytes attached to @req, but doesn't complete
1411 * the request structure even if @req doesn't have leftover.
1412 * If @req has leftover, sets it up for the next range of segments.
1414 * This special helper function is only for request stacking drivers
1415 * (e.g. request-based dm) so that they can handle partial completion.
1416 * Actual device drivers should use blk_mq_end_request instead.
1418 * Passing the result of blk_rq_bytes() as @nr_bytes guarantees
1419 * %false return from this function.
1422 * The RQF_SPECIAL_PAYLOAD flag is ignored on purpose in both
1423 * blk_rq_bytes() and in blk_update_request().
1426 * %false - this request doesn't have any more data
1427 * %true - this request has more data
1429 bool blk_update_request(struct request *req, blk_status_t error,
1430 unsigned int nr_bytes)
1434 trace_block_rq_complete(req, blk_status_to_errno(error), nr_bytes);
1439 #ifdef CONFIG_BLK_DEV_INTEGRITY
1440 if (blk_integrity_rq(req) && req_op(req) == REQ_OP_READ &&
1441 error == BLK_STS_OK)
1442 req->q->integrity.profile->complete_fn(req, nr_bytes);
1445 if (unlikely(error && !blk_rq_is_passthrough(req) &&
1446 !(req->rq_flags & RQF_QUIET)))
1447 print_req_error(req, error, __func__);
1449 blk_account_io_completion(req, nr_bytes);
1453 struct bio *bio = req->bio;
1454 unsigned bio_bytes = min(bio->bi_iter.bi_size, nr_bytes);
1456 if (bio_bytes == bio->bi_iter.bi_size)
1457 req->bio = bio->bi_next;
1459 /* Completion has already been traced */
1460 bio_clear_flag(bio, BIO_TRACE_COMPLETION);
1461 req_bio_endio(req, bio, bio_bytes, error);
1463 total_bytes += bio_bytes;
1464 nr_bytes -= bio_bytes;
1475 * Reset counters so that the request stacking driver
1476 * can find how many bytes remain in the request
1479 req->__data_len = 0;
1483 req->__data_len -= total_bytes;
1485 /* update sector only for requests with clear definition of sector */
1486 if (!blk_rq_is_passthrough(req))
1487 req->__sector += total_bytes >> 9;
1489 /* mixed attributes always follow the first bio */
1490 if (req->rq_flags & RQF_MIXED_MERGE) {
1491 req->cmd_flags &= ~REQ_FAILFAST_MASK;
1492 req->cmd_flags |= req->bio->bi_opf & REQ_FAILFAST_MASK;
1495 if (!(req->rq_flags & RQF_SPECIAL_PAYLOAD)) {
1497 * If total number of sectors is less than the first segment
1498 * size, something has gone terribly wrong.
1500 if (blk_rq_bytes(req) < blk_rq_cur_bytes(req)) {
1501 blk_dump_rq_flags(req, "request botched");
1502 req->__data_len = blk_rq_cur_bytes(req);
1505 /* recalculate the number of segments */
1506 req->nr_phys_segments = blk_recalc_rq_segments(req);
1511 EXPORT_SYMBOL_GPL(blk_update_request);
1513 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
1515 * rq_flush_dcache_pages - Helper function to flush all pages in a request
1516 * @rq: the request to be flushed
1519 * Flush all pages in @rq.
1521 void rq_flush_dcache_pages(struct request *rq)
1523 struct req_iterator iter;
1524 struct bio_vec bvec;
1526 rq_for_each_segment(bvec, rq, iter)
1527 flush_dcache_page(bvec.bv_page);
1529 EXPORT_SYMBOL_GPL(rq_flush_dcache_pages);
1533 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
1534 * @q : the queue of the device being checked
1537 * Check if underlying low-level drivers of a device are busy.
1538 * If the drivers want to export their busy state, they must set own
1539 * exporting function using blk_queue_lld_busy() first.
1541 * Basically, this function is used only by request stacking drivers
1542 * to stop dispatching requests to underlying devices when underlying
1543 * devices are busy. This behavior helps more I/O merging on the queue
1544 * of the request stacking driver and prevents I/O throughput regression
1545 * on burst I/O load.
1548 * 0 - Not busy (The request stacking driver should dispatch request)
1549 * 1 - Busy (The request stacking driver should stop dispatching request)
1551 int blk_lld_busy(struct request_queue *q)
1553 if (queue_is_mq(q) && q->mq_ops->busy)
1554 return q->mq_ops->busy(q);
1558 EXPORT_SYMBOL_GPL(blk_lld_busy);
1561 * blk_rq_unprep_clone - Helper function to free all bios in a cloned request
1562 * @rq: the clone request to be cleaned up
1565 * Free all bios in @rq for a cloned request.
1567 void blk_rq_unprep_clone(struct request *rq)
1571 while ((bio = rq->bio) != NULL) {
1572 rq->bio = bio->bi_next;
1577 EXPORT_SYMBOL_GPL(blk_rq_unprep_clone);
1580 * blk_rq_prep_clone - Helper function to setup clone request
1581 * @rq: the request to be setup
1582 * @rq_src: original request to be cloned
1583 * @bs: bio_set that bios for clone are allocated from
1584 * @gfp_mask: memory allocation mask for bio
1585 * @bio_ctr: setup function to be called for each clone bio.
1586 * Returns %0 for success, non %0 for failure.
1587 * @data: private data to be passed to @bio_ctr
1590 * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq.
1591 * Also, pages which the original bios are pointing to are not copied
1592 * and the cloned bios just point same pages.
1593 * So cloned bios must be completed before original bios, which means
1594 * the caller must complete @rq before @rq_src.
1596 int blk_rq_prep_clone(struct request *rq, struct request *rq_src,
1597 struct bio_set *bs, gfp_t gfp_mask,
1598 int (*bio_ctr)(struct bio *, struct bio *, void *),
1601 struct bio *bio, *bio_src;
1606 __rq_for_each_bio(bio_src, rq_src) {
1607 bio = bio_clone_fast(bio_src, gfp_mask, bs);
1611 if (bio_ctr && bio_ctr(bio, bio_src, data))
1615 rq->biotail->bi_next = bio;
1618 rq->bio = rq->biotail = bio;
1623 /* Copy attributes of the original request to the clone request. */
1624 rq->__sector = blk_rq_pos(rq_src);
1625 rq->__data_len = blk_rq_bytes(rq_src);
1626 if (rq_src->rq_flags & RQF_SPECIAL_PAYLOAD) {
1627 rq->rq_flags |= RQF_SPECIAL_PAYLOAD;
1628 rq->special_vec = rq_src->special_vec;
1630 rq->nr_phys_segments = rq_src->nr_phys_segments;
1631 rq->ioprio = rq_src->ioprio;
1633 if (rq->bio && blk_crypto_rq_bio_prep(rq, rq->bio, gfp_mask) < 0)
1641 blk_rq_unprep_clone(rq);
1645 EXPORT_SYMBOL_GPL(blk_rq_prep_clone);
1647 int kblockd_schedule_work(struct work_struct *work)
1649 return queue_work(kblockd_workqueue, work);
1651 EXPORT_SYMBOL(kblockd_schedule_work);
1653 int kblockd_mod_delayed_work_on(int cpu, struct delayed_work *dwork,
1654 unsigned long delay)
1656 return mod_delayed_work_on(cpu, kblockd_workqueue, dwork, delay);
1658 EXPORT_SYMBOL(kblockd_mod_delayed_work_on);
1661 * blk_start_plug - initialize blk_plug and track it inside the task_struct
1662 * @plug: The &struct blk_plug that needs to be initialized
1665 * blk_start_plug() indicates to the block layer an intent by the caller
1666 * to submit multiple I/O requests in a batch. The block layer may use
1667 * this hint to defer submitting I/Os from the caller until blk_finish_plug()
1668 * is called. However, the block layer may choose to submit requests
1669 * before a call to blk_finish_plug() if the number of queued I/Os
1670 * exceeds %BLK_MAX_REQUEST_COUNT, or if the size of the I/O is larger than
1671 * %BLK_PLUG_FLUSH_SIZE. The queued I/Os may also be submitted early if
1672 * the task schedules (see below).
1674 * Tracking blk_plug inside the task_struct will help with auto-flushing the
1675 * pending I/O should the task end up blocking between blk_start_plug() and
1676 * blk_finish_plug(). This is important from a performance perspective, but
1677 * also ensures that we don't deadlock. For instance, if the task is blocking
1678 * for a memory allocation, memory reclaim could end up wanting to free a
1679 * page belonging to that request that is currently residing in our private
1680 * plug. By flushing the pending I/O when the process goes to sleep, we avoid
1681 * this kind of deadlock.
1683 void blk_start_plug(struct blk_plug *plug)
1685 struct task_struct *tsk = current;
1688 * If this is a nested plug, don't actually assign it.
1693 INIT_LIST_HEAD(&plug->mq_list);
1694 INIT_LIST_HEAD(&plug->cb_list);
1696 plug->multiple_queues = false;
1697 plug->nowait = false;
1700 * Store ordering should not be needed here, since a potential
1701 * preempt will imply a full memory barrier
1705 EXPORT_SYMBOL(blk_start_plug);
1707 static void flush_plug_callbacks(struct blk_plug *plug, bool from_schedule)
1709 LIST_HEAD(callbacks);
1711 while (!list_empty(&plug->cb_list)) {
1712 list_splice_init(&plug->cb_list, &callbacks);
1714 while (!list_empty(&callbacks)) {
1715 struct blk_plug_cb *cb = list_first_entry(&callbacks,
1718 list_del(&cb->list);
1719 cb->callback(cb, from_schedule);
1724 struct blk_plug_cb *blk_check_plugged(blk_plug_cb_fn unplug, void *data,
1727 struct blk_plug *plug = current->plug;
1728 struct blk_plug_cb *cb;
1733 list_for_each_entry(cb, &plug->cb_list, list)
1734 if (cb->callback == unplug && cb->data == data)
1737 /* Not currently on the callback list */
1738 BUG_ON(size < sizeof(*cb));
1739 cb = kzalloc(size, GFP_ATOMIC);
1742 cb->callback = unplug;
1743 list_add(&cb->list, &plug->cb_list);
1747 EXPORT_SYMBOL(blk_check_plugged);
1749 void blk_flush_plug_list(struct blk_plug *plug, bool from_schedule)
1751 flush_plug_callbacks(plug, from_schedule);
1753 if (!list_empty(&plug->mq_list))
1754 blk_mq_flush_plug_list(plug, from_schedule);
1758 * blk_finish_plug - mark the end of a batch of submitted I/O
1759 * @plug: The &struct blk_plug passed to blk_start_plug()
1762 * Indicate that a batch of I/O submissions is complete. This function
1763 * must be paired with an initial call to blk_start_plug(). The intent
1764 * is to allow the block layer to optimize I/O submission. See the
1765 * documentation for blk_start_plug() for more information.
1767 void blk_finish_plug(struct blk_plug *plug)
1769 if (plug != current->plug)
1771 blk_flush_plug_list(plug, false);
1773 current->plug = NULL;
1775 EXPORT_SYMBOL(blk_finish_plug);
1777 void blk_io_schedule(void)
1779 /* Prevent hang_check timer from firing at us during very long I/O */
1780 unsigned long timeout = sysctl_hung_task_timeout_secs * HZ / 2;
1783 io_schedule_timeout(timeout);
1787 EXPORT_SYMBOL_GPL(blk_io_schedule);
1789 int __init blk_dev_init(void)
1791 BUILD_BUG_ON(REQ_OP_LAST >= (1 << REQ_OP_BITS));
1792 BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 *
1793 sizeof_field(struct request, cmd_flags));
1794 BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 *
1795 sizeof_field(struct bio, bi_opf));
1797 /* used for unplugging and affects IO latency/throughput - HIGHPRI */
1798 kblockd_workqueue = alloc_workqueue("kblockd",
1799 WQ_MEM_RECLAIM | WQ_HIGHPRI, 0);
1800 if (!kblockd_workqueue)
1801 panic("Failed to create kblockd\n");
1803 blk_requestq_cachep = kmem_cache_create("request_queue",
1804 sizeof(struct request_queue), 0, SLAB_PANIC, NULL);
1806 blk_debugfs_root = debugfs_create_dir("block", NULL);