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/blk-pm.h>
22 #include <linux/highmem.h>
24 #include <linux/pagemap.h>
25 #include <linux/kernel_stat.h>
26 #include <linux/string.h>
27 #include <linux/init.h>
28 #include <linux/completion.h>
29 #include <linux/slab.h>
30 #include <linux/swap.h>
31 #include <linux/writeback.h>
32 #include <linux/task_io_accounting_ops.h>
33 #include <linux/fault-inject.h>
34 #include <linux/list_sort.h>
35 #include <linux/delay.h>
36 #include <linux/ratelimit.h>
37 #include <linux/pm_runtime.h>
38 #include <linux/blk-cgroup.h>
39 #include <linux/t10-pi.h>
40 #include <linux/debugfs.h>
41 #include <linux/bpf.h>
42 #include <linux/psi.h>
43 #include <linux/sched/sysctl.h>
44 #include <linux/blk-crypto.h>
46 #define CREATE_TRACE_POINTS
47 #include <trace/events/block.h>
51 #include "blk-mq-sched.h"
53 #include "blk-rq-qos.h"
55 struct dentry *blk_debugfs_root;
57 EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_remap);
58 EXPORT_TRACEPOINT_SYMBOL_GPL(block_rq_remap);
59 EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_complete);
60 EXPORT_TRACEPOINT_SYMBOL_GPL(block_split);
61 EXPORT_TRACEPOINT_SYMBOL_GPL(block_unplug);
63 DEFINE_IDA(blk_queue_ida);
66 * For queue allocation
68 struct kmem_cache *blk_requestq_cachep;
71 * Controlling structure to kblockd
73 static struct workqueue_struct *kblockd_workqueue;
76 * blk_queue_flag_set - atomically set a queue flag
77 * @flag: flag to be set
80 void blk_queue_flag_set(unsigned int flag, struct request_queue *q)
82 set_bit(flag, &q->queue_flags);
84 EXPORT_SYMBOL(blk_queue_flag_set);
87 * blk_queue_flag_clear - atomically clear a queue flag
88 * @flag: flag to be cleared
91 void blk_queue_flag_clear(unsigned int flag, struct request_queue *q)
93 clear_bit(flag, &q->queue_flags);
95 EXPORT_SYMBOL(blk_queue_flag_clear);
98 * blk_queue_flag_test_and_set - atomically test and set a queue flag
99 * @flag: flag to be set
102 * Returns the previous value of @flag - 0 if the flag was not set and 1 if
103 * the flag was already set.
105 bool blk_queue_flag_test_and_set(unsigned int flag, struct request_queue *q)
107 return test_and_set_bit(flag, &q->queue_flags);
109 EXPORT_SYMBOL_GPL(blk_queue_flag_test_and_set);
111 void blk_rq_init(struct request_queue *q, struct request *rq)
113 memset(rq, 0, sizeof(*rq));
115 INIT_LIST_HEAD(&rq->queuelist);
117 rq->__sector = (sector_t) -1;
118 INIT_HLIST_NODE(&rq->hash);
119 RB_CLEAR_NODE(&rq->rb_node);
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),
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 /* @q is and will stay empty, shutdown and put */
420 EXPORT_SYMBOL(blk_cleanup_queue);
423 * blk_queue_enter() - try to increase q->q_usage_counter
424 * @q: request queue pointer
425 * @flags: BLK_MQ_REQ_NOWAIT and/or BLK_MQ_REQ_PM
427 int blk_queue_enter(struct request_queue *q, blk_mq_req_flags_t flags)
429 const bool pm = flags & BLK_MQ_REQ_PM;
432 bool success = false;
435 if (percpu_ref_tryget_live(&q->q_usage_counter)) {
437 * The code that increments the pm_only counter is
438 * responsible for ensuring that that counter is
439 * globally visible before the queue is unfrozen.
441 if ((pm && queue_rpm_status(q) != RPM_SUSPENDED) ||
442 !blk_queue_pm_only(q)) {
445 percpu_ref_put(&q->q_usage_counter);
453 if (flags & BLK_MQ_REQ_NOWAIT)
457 * read pair of barrier in blk_freeze_queue_start(),
458 * we need to order reading __PERCPU_REF_DEAD flag of
459 * .q_usage_counter and reading .mq_freeze_depth or
460 * queue dying flag, otherwise the following wait may
461 * never return if the two reads are reordered.
465 wait_event(q->mq_freeze_wq,
466 (!q->mq_freeze_depth &&
467 blk_pm_resume_queue(pm, q)) ||
469 if (blk_queue_dying(q))
474 static inline int bio_queue_enter(struct bio *bio)
476 struct request_queue *q = bio->bi_disk->queue;
477 bool nowait = bio->bi_opf & REQ_NOWAIT;
480 ret = blk_queue_enter(q, nowait ? BLK_MQ_REQ_NOWAIT : 0);
482 if (nowait && !blk_queue_dying(q))
483 bio_wouldblock_error(bio);
491 void blk_queue_exit(struct request_queue *q)
493 percpu_ref_put(&q->q_usage_counter);
496 static void blk_queue_usage_counter_release(struct percpu_ref *ref)
498 struct request_queue *q =
499 container_of(ref, struct request_queue, q_usage_counter);
501 wake_up_all(&q->mq_freeze_wq);
504 static void blk_rq_timed_out_timer(struct timer_list *t)
506 struct request_queue *q = from_timer(q, t, timeout);
508 kblockd_schedule_work(&q->timeout_work);
511 static void blk_timeout_work(struct work_struct *work)
515 struct request_queue *blk_alloc_queue(int node_id)
517 struct request_queue *q;
520 q = kmem_cache_alloc_node(blk_requestq_cachep,
521 GFP_KERNEL | __GFP_ZERO, node_id);
525 q->last_merge = NULL;
527 q->id = ida_simple_get(&blk_queue_ida, 0, 0, GFP_KERNEL);
531 ret = bioset_init(&q->bio_split, BIO_POOL_SIZE, 0, BIOSET_NEED_BVECS);
535 q->backing_dev_info = bdi_alloc(node_id);
536 if (!q->backing_dev_info)
539 q->stats = blk_alloc_queue_stats();
545 atomic_set(&q->nr_active_requests_shared_sbitmap, 0);
547 timer_setup(&q->backing_dev_info->laptop_mode_wb_timer,
548 laptop_mode_timer_fn, 0);
549 timer_setup(&q->timeout, blk_rq_timed_out_timer, 0);
550 INIT_WORK(&q->timeout_work, blk_timeout_work);
551 INIT_LIST_HEAD(&q->icq_list);
552 #ifdef CONFIG_BLK_CGROUP
553 INIT_LIST_HEAD(&q->blkg_list);
556 kobject_init(&q->kobj, &blk_queue_ktype);
558 mutex_init(&q->debugfs_mutex);
559 mutex_init(&q->sysfs_lock);
560 mutex_init(&q->sysfs_dir_lock);
561 spin_lock_init(&q->queue_lock);
563 init_waitqueue_head(&q->mq_freeze_wq);
564 mutex_init(&q->mq_freeze_lock);
567 * Init percpu_ref in atomic mode so that it's faster to shutdown.
568 * See blk_register_queue() for details.
570 if (percpu_ref_init(&q->q_usage_counter,
571 blk_queue_usage_counter_release,
572 PERCPU_REF_INIT_ATOMIC, GFP_KERNEL))
575 if (blkcg_init_queue(q))
578 blk_queue_dma_alignment(q, 511);
579 blk_set_default_limits(&q->limits);
580 q->nr_requests = BLKDEV_MAX_RQ;
585 percpu_ref_exit(&q->q_usage_counter);
587 blk_free_queue_stats(q->stats);
589 bdi_put(q->backing_dev_info);
591 bioset_exit(&q->bio_split);
593 ida_simple_remove(&blk_queue_ida, q->id);
595 kmem_cache_free(blk_requestq_cachep, q);
598 EXPORT_SYMBOL(blk_alloc_queue);
601 * blk_get_queue - increment the request_queue refcount
602 * @q: the request_queue structure to increment the refcount for
604 * Increment the refcount of the request_queue kobject.
606 * Context: Any context.
608 bool blk_get_queue(struct request_queue *q)
610 if (likely(!blk_queue_dying(q))) {
617 EXPORT_SYMBOL(blk_get_queue);
620 * blk_get_request - allocate a request
621 * @q: request queue to allocate a request for
622 * @op: operation (REQ_OP_*) and REQ_* flags, e.g. REQ_SYNC.
623 * @flags: BLK_MQ_REQ_* flags, e.g. BLK_MQ_REQ_NOWAIT.
625 struct request *blk_get_request(struct request_queue *q, unsigned int op,
626 blk_mq_req_flags_t flags)
630 WARN_ON_ONCE(op & REQ_NOWAIT);
631 WARN_ON_ONCE(flags & ~(BLK_MQ_REQ_NOWAIT | BLK_MQ_REQ_PM));
633 req = blk_mq_alloc_request(q, op, flags);
634 if (!IS_ERR(req) && q->mq_ops->initialize_rq_fn)
635 q->mq_ops->initialize_rq_fn(req);
639 EXPORT_SYMBOL(blk_get_request);
641 void blk_put_request(struct request *req)
643 blk_mq_free_request(req);
645 EXPORT_SYMBOL(blk_put_request);
647 static void handle_bad_sector(struct bio *bio, sector_t maxsector)
649 char b[BDEVNAME_SIZE];
651 pr_info_ratelimited("attempt to access beyond end of device\n"
652 "%s: rw=%d, want=%llu, limit=%llu\n",
653 bio_devname(bio, b), bio->bi_opf,
654 bio_end_sector(bio), maxsector);
657 #ifdef CONFIG_FAIL_MAKE_REQUEST
659 static DECLARE_FAULT_ATTR(fail_make_request);
661 static int __init setup_fail_make_request(char *str)
663 return setup_fault_attr(&fail_make_request, str);
665 __setup("fail_make_request=", setup_fail_make_request);
667 static bool should_fail_request(struct hd_struct *part, unsigned int bytes)
669 return part->make_it_fail && should_fail(&fail_make_request, bytes);
672 static int __init fail_make_request_debugfs(void)
674 struct dentry *dir = fault_create_debugfs_attr("fail_make_request",
675 NULL, &fail_make_request);
677 return PTR_ERR_OR_ZERO(dir);
680 late_initcall(fail_make_request_debugfs);
682 #else /* CONFIG_FAIL_MAKE_REQUEST */
684 static inline bool should_fail_request(struct hd_struct *part,
690 #endif /* CONFIG_FAIL_MAKE_REQUEST */
692 static inline bool bio_check_ro(struct bio *bio, struct hd_struct *part)
694 const int op = bio_op(bio);
696 if (part->policy && op_is_write(op)) {
697 char b[BDEVNAME_SIZE];
699 if (op_is_flush(bio->bi_opf) && !bio_sectors(bio))
701 pr_warn("Trying to write to read-only block-device %s (partno %d)\n",
702 bio_devname(bio, b), part->partno);
703 /* Older lvm-tools actually trigger this */
710 static noinline int should_fail_bio(struct bio *bio)
712 if (should_fail_request(&bio->bi_disk->part0, bio->bi_iter.bi_size))
716 ALLOW_ERROR_INJECTION(should_fail_bio, ERRNO);
719 * Check whether this bio extends beyond the end of the device or partition.
720 * This may well happen - the kernel calls bread() without checking the size of
721 * the device, e.g., when mounting a file system.
723 static inline int bio_check_eod(struct bio *bio, sector_t maxsector)
725 unsigned int nr_sectors = bio_sectors(bio);
727 if (nr_sectors && maxsector &&
728 (nr_sectors > maxsector ||
729 bio->bi_iter.bi_sector > maxsector - nr_sectors)) {
730 handle_bad_sector(bio, maxsector);
737 * Remap block n of partition p to block n+start(p) of the disk.
739 static inline int blk_partition_remap(struct bio *bio)
745 p = __disk_get_part(bio->bi_disk, bio->bi_partno);
748 if (unlikely(should_fail_request(p, bio->bi_iter.bi_size)))
750 if (unlikely(bio_check_ro(bio, p)))
753 if (bio_sectors(bio)) {
754 if (bio_check_eod(bio, part_nr_sects_read(p)))
756 bio->bi_iter.bi_sector += p->start_sect;
757 trace_block_bio_remap(bio->bi_disk->queue, bio, part_devt(p),
758 bio->bi_iter.bi_sector - p->start_sect);
768 * Check write append to a zoned block device.
770 static inline blk_status_t blk_check_zone_append(struct request_queue *q,
773 sector_t pos = bio->bi_iter.bi_sector;
774 int nr_sectors = bio_sectors(bio);
776 /* Only applicable to zoned block devices */
777 if (!blk_queue_is_zoned(q))
778 return BLK_STS_NOTSUPP;
780 /* The bio sector must point to the start of a sequential zone */
781 if (pos & (blk_queue_zone_sectors(q) - 1) ||
782 !blk_queue_zone_is_seq(q, pos))
783 return BLK_STS_IOERR;
786 * Not allowed to cross zone boundaries. Otherwise, the BIO will be
787 * split and could result in non-contiguous sectors being written in
790 if (nr_sectors > q->limits.chunk_sectors)
791 return BLK_STS_IOERR;
793 /* Make sure the BIO is small enough and will not get split */
794 if (nr_sectors > q->limits.max_zone_append_sectors)
795 return BLK_STS_IOERR;
797 bio->bi_opf |= REQ_NOMERGE;
802 static noinline_for_stack bool submit_bio_checks(struct bio *bio)
804 struct request_queue *q = bio->bi_disk->queue;
805 blk_status_t status = BLK_STS_IOERR;
806 struct blk_plug *plug;
810 plug = blk_mq_plug(q, bio);
811 if (plug && plug->nowait)
812 bio->bi_opf |= REQ_NOWAIT;
815 * For a REQ_NOWAIT based request, return -EOPNOTSUPP
816 * if queue does not support NOWAIT.
818 if ((bio->bi_opf & REQ_NOWAIT) && !blk_queue_nowait(q))
821 if (should_fail_bio(bio))
824 if (bio->bi_partno) {
825 if (unlikely(blk_partition_remap(bio)))
828 if (unlikely(bio_check_ro(bio, &bio->bi_disk->part0)))
830 if (unlikely(bio_check_eod(bio, get_capacity(bio->bi_disk))))
835 * Filter flush bio's early so that bio based drivers without flush
836 * support don't have to worry about them.
838 if (op_is_flush(bio->bi_opf) &&
839 !test_bit(QUEUE_FLAG_WC, &q->queue_flags)) {
840 bio->bi_opf &= ~(REQ_PREFLUSH | REQ_FUA);
841 if (!bio_sectors(bio)) {
847 if (!test_bit(QUEUE_FLAG_POLL, &q->queue_flags))
848 bio->bi_opf &= ~REQ_HIPRI;
850 switch (bio_op(bio)) {
852 if (!blk_queue_discard(q))
855 case REQ_OP_SECURE_ERASE:
856 if (!blk_queue_secure_erase(q))
859 case REQ_OP_WRITE_SAME:
860 if (!q->limits.max_write_same_sectors)
863 case REQ_OP_ZONE_APPEND:
864 status = blk_check_zone_append(q, bio);
865 if (status != BLK_STS_OK)
868 case REQ_OP_ZONE_RESET:
869 case REQ_OP_ZONE_OPEN:
870 case REQ_OP_ZONE_CLOSE:
871 case REQ_OP_ZONE_FINISH:
872 if (!blk_queue_is_zoned(q))
875 case REQ_OP_ZONE_RESET_ALL:
876 if (!blk_queue_is_zoned(q) || !blk_queue_zone_resetall(q))
879 case REQ_OP_WRITE_ZEROES:
880 if (!q->limits.max_write_zeroes_sectors)
888 * Various block parts want %current->io_context, so allocate it up
889 * front rather than dealing with lots of pain to allocate it only
890 * where needed. This may fail and the block layer knows how to live
893 if (unlikely(!current->io_context))
894 create_task_io_context(current, GFP_ATOMIC, q->node);
896 if (blk_throtl_bio(bio))
899 blk_cgroup_bio_start(bio);
900 blkcg_bio_issue_init(bio);
902 if (!bio_flagged(bio, BIO_TRACE_COMPLETION)) {
903 trace_block_bio_queue(q, bio);
904 /* Now that enqueuing has been traced, we need to trace
905 * completion as well.
907 bio_set_flag(bio, BIO_TRACE_COMPLETION);
912 status = BLK_STS_NOTSUPP;
914 bio->bi_status = status;
919 static blk_qc_t __submit_bio(struct bio *bio)
921 struct gendisk *disk = bio->bi_disk;
922 blk_qc_t ret = BLK_QC_T_NONE;
924 if (blk_crypto_bio_prep(&bio)) {
925 if (!disk->fops->submit_bio)
926 return blk_mq_submit_bio(bio);
927 ret = disk->fops->submit_bio(bio);
929 blk_queue_exit(disk->queue);
934 * The loop in this function may be a bit non-obvious, and so deserves some
937 * - Before entering the loop, bio->bi_next is NULL (as all callers ensure
938 * that), so we have a list with a single bio.
939 * - We pretend that we have just taken it off a longer list, so we assign
940 * bio_list to a pointer to the bio_list_on_stack, thus initialising the
941 * bio_list of new bios to be added. ->submit_bio() may indeed add some more
942 * bios through a recursive call to submit_bio_noacct. If it did, we find a
943 * non-NULL value in bio_list and re-enter the loop from the top.
944 * - In this case we really did just take the bio of the top of the list (no
945 * pretending) and so remove it from bio_list, and call into ->submit_bio()
948 * bio_list_on_stack[0] contains bios submitted by the current ->submit_bio.
949 * bio_list_on_stack[1] contains bios that were submitted before the current
950 * ->submit_bio_bio, but that haven't been processed yet.
952 static blk_qc_t __submit_bio_noacct(struct bio *bio)
954 struct bio_list bio_list_on_stack[2];
955 blk_qc_t ret = BLK_QC_T_NONE;
957 BUG_ON(bio->bi_next);
959 bio_list_init(&bio_list_on_stack[0]);
960 current->bio_list = bio_list_on_stack;
963 struct request_queue *q = bio->bi_disk->queue;
964 struct bio_list lower, same;
966 if (unlikely(bio_queue_enter(bio) != 0))
970 * Create a fresh bio_list for all subordinate requests.
972 bio_list_on_stack[1] = bio_list_on_stack[0];
973 bio_list_init(&bio_list_on_stack[0]);
975 ret = __submit_bio(bio);
978 * Sort new bios into those for a lower level and those for the
981 bio_list_init(&lower);
982 bio_list_init(&same);
983 while ((bio = bio_list_pop(&bio_list_on_stack[0])) != NULL)
984 if (q == bio->bi_disk->queue)
985 bio_list_add(&same, bio);
987 bio_list_add(&lower, bio);
990 * Now assemble so we handle the lowest level first.
992 bio_list_merge(&bio_list_on_stack[0], &lower);
993 bio_list_merge(&bio_list_on_stack[0], &same);
994 bio_list_merge(&bio_list_on_stack[0], &bio_list_on_stack[1]);
995 } while ((bio = bio_list_pop(&bio_list_on_stack[0])));
997 current->bio_list = NULL;
1001 static blk_qc_t __submit_bio_noacct_mq(struct bio *bio)
1003 struct bio_list bio_list[2] = { };
1004 blk_qc_t ret = BLK_QC_T_NONE;
1006 current->bio_list = bio_list;
1009 struct gendisk *disk = bio->bi_disk;
1011 if (unlikely(bio_queue_enter(bio) != 0))
1014 if (!blk_crypto_bio_prep(&bio)) {
1015 blk_queue_exit(disk->queue);
1016 ret = BLK_QC_T_NONE;
1020 ret = blk_mq_submit_bio(bio);
1021 } while ((bio = bio_list_pop(&bio_list[0])));
1023 current->bio_list = NULL;
1028 * submit_bio_noacct - re-submit a bio to the block device layer for I/O
1029 * @bio: The bio describing the location in memory and on the device.
1031 * This is a version of submit_bio() that shall only be used for I/O that is
1032 * resubmitted to lower level drivers by stacking block drivers. All file
1033 * systems and other upper level users of the block layer should use
1034 * submit_bio() instead.
1036 blk_qc_t submit_bio_noacct(struct bio *bio)
1038 if (!submit_bio_checks(bio))
1039 return BLK_QC_T_NONE;
1042 * We only want one ->submit_bio to be active at a time, else stack
1043 * usage with stacked devices could be a problem. Use current->bio_list
1044 * to collect a list of requests submited by a ->submit_bio method while
1045 * it is active, and then process them after it returned.
1047 if (current->bio_list) {
1048 bio_list_add(¤t->bio_list[0], bio);
1049 return BLK_QC_T_NONE;
1052 if (!bio->bi_disk->fops->submit_bio)
1053 return __submit_bio_noacct_mq(bio);
1054 return __submit_bio_noacct(bio);
1056 EXPORT_SYMBOL(submit_bio_noacct);
1059 * submit_bio - submit a bio to the block device layer for I/O
1060 * @bio: The &struct bio which describes the I/O
1062 * submit_bio() is used to submit I/O requests to block devices. It is passed a
1063 * fully set up &struct bio that describes the I/O that needs to be done. The
1064 * bio will be send to the device described by the bi_disk and bi_partno fields.
1066 * The success/failure status of the request, along with notification of
1067 * completion, is delivered asynchronously through the ->bi_end_io() callback
1068 * in @bio. The bio must NOT be touched by thecaller until ->bi_end_io() has
1071 blk_qc_t submit_bio(struct bio *bio)
1073 if (blkcg_punt_bio_submit(bio))
1074 return BLK_QC_T_NONE;
1077 * If it's a regular read/write or a barrier with data attached,
1078 * go through the normal accounting stuff before submission.
1080 if (bio_has_data(bio)) {
1083 if (unlikely(bio_op(bio) == REQ_OP_WRITE_SAME))
1084 count = queue_logical_block_size(bio->bi_disk->queue) >> 9;
1086 count = bio_sectors(bio);
1088 if (op_is_write(bio_op(bio))) {
1089 count_vm_events(PGPGOUT, count);
1091 task_io_account_read(bio->bi_iter.bi_size);
1092 count_vm_events(PGPGIN, count);
1095 if (unlikely(block_dump)) {
1096 char b[BDEVNAME_SIZE];
1097 printk(KERN_DEBUG "%s(%d): %s block %Lu on %s (%u sectors)\n",
1098 current->comm, task_pid_nr(current),
1099 op_is_write(bio_op(bio)) ? "WRITE" : "READ",
1100 (unsigned long long)bio->bi_iter.bi_sector,
1101 bio_devname(bio, b), count);
1106 * If we're reading data that is part of the userspace workingset, count
1107 * submission time as memory stall. When the device is congested, or
1108 * the submitting cgroup IO-throttled, submission can be a significant
1109 * part of overall IO time.
1111 if (unlikely(bio_op(bio) == REQ_OP_READ &&
1112 bio_flagged(bio, BIO_WORKINGSET))) {
1113 unsigned long pflags;
1116 psi_memstall_enter(&pflags);
1117 ret = submit_bio_noacct(bio);
1118 psi_memstall_leave(&pflags);
1123 return submit_bio_noacct(bio);
1125 EXPORT_SYMBOL(submit_bio);
1128 * blk_cloned_rq_check_limits - Helper function to check a cloned request
1129 * for the new queue limits
1131 * @rq: the request being checked
1134 * @rq may have been made based on weaker limitations of upper-level queues
1135 * in request stacking drivers, and it may violate the limitation of @q.
1136 * Since the block layer and the underlying device driver trust @rq
1137 * after it is inserted to @q, it should be checked against @q before
1138 * the insertion using this generic function.
1140 * Request stacking drivers like request-based dm may change the queue
1141 * limits when retrying requests on other queues. Those requests need
1142 * to be checked against the new queue limits again during dispatch.
1144 static blk_status_t blk_cloned_rq_check_limits(struct request_queue *q,
1147 unsigned int max_sectors = blk_queue_get_max_sectors(q, req_op(rq));
1149 if (blk_rq_sectors(rq) > max_sectors) {
1151 * SCSI device does not have a good way to return if
1152 * Write Same/Zero is actually supported. If a device rejects
1153 * a non-read/write command (discard, write same,etc.) the
1154 * low-level device driver will set the relevant queue limit to
1155 * 0 to prevent blk-lib from issuing more of the offending
1156 * operations. Commands queued prior to the queue limit being
1157 * reset need to be completed with BLK_STS_NOTSUPP to avoid I/O
1158 * errors being propagated to upper layers.
1160 if (max_sectors == 0)
1161 return BLK_STS_NOTSUPP;
1163 printk(KERN_ERR "%s: over max size limit. (%u > %u)\n",
1164 __func__, blk_rq_sectors(rq), max_sectors);
1165 return BLK_STS_IOERR;
1169 * queue's settings related to segment counting like q->bounce_pfn
1170 * may differ from that of other stacking queues.
1171 * Recalculate it to check the request correctly on this queue's
1174 rq->nr_phys_segments = blk_recalc_rq_segments(rq);
1175 if (rq->nr_phys_segments > queue_max_segments(q)) {
1176 printk(KERN_ERR "%s: over max segments limit. (%hu > %hu)\n",
1177 __func__, rq->nr_phys_segments, queue_max_segments(q));
1178 return BLK_STS_IOERR;
1185 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
1186 * @q: the queue to submit the request
1187 * @rq: the request being queued
1189 blk_status_t blk_insert_cloned_request(struct request_queue *q, struct request *rq)
1193 ret = blk_cloned_rq_check_limits(q, rq);
1194 if (ret != BLK_STS_OK)
1198 should_fail_request(&rq->rq_disk->part0, blk_rq_bytes(rq)))
1199 return BLK_STS_IOERR;
1201 if (blk_crypto_insert_cloned_request(rq))
1202 return BLK_STS_IOERR;
1204 if (blk_queue_io_stat(q))
1205 blk_account_io_start(rq);
1208 * Since we have a scheduler attached on the top device,
1209 * bypass a potential scheduler on the bottom device for
1212 return blk_mq_request_issue_directly(rq, true);
1214 EXPORT_SYMBOL_GPL(blk_insert_cloned_request);
1217 * blk_rq_err_bytes - determine number of bytes till the next failure boundary
1218 * @rq: request to examine
1221 * A request could be merge of IOs which require different failure
1222 * handling. This function determines the number of bytes which
1223 * can be failed from the beginning of the request without
1224 * crossing into area which need to be retried further.
1227 * The number of bytes to fail.
1229 unsigned int blk_rq_err_bytes(const struct request *rq)
1231 unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK;
1232 unsigned int bytes = 0;
1235 if (!(rq->rq_flags & RQF_MIXED_MERGE))
1236 return blk_rq_bytes(rq);
1239 * Currently the only 'mixing' which can happen is between
1240 * different fastfail types. We can safely fail portions
1241 * which have all the failfast bits that the first one has -
1242 * the ones which are at least as eager to fail as the first
1245 for (bio = rq->bio; bio; bio = bio->bi_next) {
1246 if ((bio->bi_opf & ff) != ff)
1248 bytes += bio->bi_iter.bi_size;
1251 /* this could lead to infinite loop */
1252 BUG_ON(blk_rq_bytes(rq) && !bytes);
1255 EXPORT_SYMBOL_GPL(blk_rq_err_bytes);
1257 static void update_io_ticks(struct hd_struct *part, unsigned long now, bool end)
1259 unsigned long stamp;
1261 stamp = READ_ONCE(part->stamp);
1262 if (unlikely(stamp != now)) {
1263 if (likely(cmpxchg(&part->stamp, stamp, now) == stamp))
1264 __part_stat_add(part, io_ticks, end ? now - stamp : 1);
1267 part = &part_to_disk(part)->part0;
1272 static void blk_account_io_completion(struct request *req, unsigned int bytes)
1274 if (req->part && blk_do_io_stat(req)) {
1275 const int sgrp = op_stat_group(req_op(req));
1276 struct hd_struct *part;
1280 part_stat_add(part, sectors[sgrp], bytes >> 9);
1285 void blk_account_io_done(struct request *req, u64 now)
1288 * Account IO completion. flush_rq isn't accounted as a
1289 * normal IO on queueing nor completion. Accounting the
1290 * containing request is enough.
1292 if (req->part && blk_do_io_stat(req) &&
1293 !(req->rq_flags & RQF_FLUSH_SEQ)) {
1294 const int sgrp = op_stat_group(req_op(req));
1295 struct hd_struct *part;
1300 update_io_ticks(part, jiffies, true);
1301 part_stat_inc(part, ios[sgrp]);
1302 part_stat_add(part, nsecs[sgrp], now - req->start_time_ns);
1305 hd_struct_put(part);
1309 void blk_account_io_start(struct request *rq)
1311 if (!blk_do_io_stat(rq))
1314 rq->part = disk_map_sector_rcu(rq->rq_disk, blk_rq_pos(rq));
1317 update_io_ticks(rq->part, jiffies, false);
1321 static unsigned long __part_start_io_acct(struct hd_struct *part,
1322 unsigned int sectors, unsigned int op)
1324 const int sgrp = op_stat_group(op);
1325 unsigned long now = READ_ONCE(jiffies);
1328 update_io_ticks(part, now, false);
1329 part_stat_inc(part, ios[sgrp]);
1330 part_stat_add(part, sectors[sgrp], sectors);
1331 part_stat_local_inc(part, in_flight[op_is_write(op)]);
1337 unsigned long part_start_io_acct(struct gendisk *disk, struct hd_struct **part,
1340 *part = disk_map_sector_rcu(disk, bio->bi_iter.bi_sector);
1342 return __part_start_io_acct(*part, bio_sectors(bio), bio_op(bio));
1344 EXPORT_SYMBOL_GPL(part_start_io_acct);
1346 unsigned long disk_start_io_acct(struct gendisk *disk, unsigned int sectors,
1349 return __part_start_io_acct(&disk->part0, sectors, op);
1351 EXPORT_SYMBOL(disk_start_io_acct);
1353 static void __part_end_io_acct(struct hd_struct *part, unsigned int op,
1354 unsigned long start_time)
1356 const int sgrp = op_stat_group(op);
1357 unsigned long now = READ_ONCE(jiffies);
1358 unsigned long duration = now - start_time;
1361 update_io_ticks(part, now, true);
1362 part_stat_add(part, nsecs[sgrp], jiffies_to_nsecs(duration));
1363 part_stat_local_dec(part, in_flight[op_is_write(op)]);
1367 void part_end_io_acct(struct hd_struct *part, struct bio *bio,
1368 unsigned long start_time)
1370 __part_end_io_acct(part, bio_op(bio), start_time);
1371 hd_struct_put(part);
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);
1446 * Upper layers may call blk_crypto_evict_key() anytime after the last
1447 * bio_endio(). Therefore, the keyslot must be released before that.
1449 if (blk_crypto_rq_has_keyslot(req) && nr_bytes >= blk_rq_bytes(req))
1450 __blk_crypto_rq_put_keyslot(req);
1452 if (unlikely(error && !blk_rq_is_passthrough(req) &&
1453 !(req->rq_flags & RQF_QUIET)))
1454 print_req_error(req, error, __func__);
1456 blk_account_io_completion(req, nr_bytes);
1460 struct bio *bio = req->bio;
1461 unsigned bio_bytes = min(bio->bi_iter.bi_size, nr_bytes);
1463 if (bio_bytes == bio->bi_iter.bi_size)
1464 req->bio = bio->bi_next;
1466 /* Completion has already been traced */
1467 bio_clear_flag(bio, BIO_TRACE_COMPLETION);
1468 req_bio_endio(req, bio, bio_bytes, error);
1470 total_bytes += bio_bytes;
1471 nr_bytes -= bio_bytes;
1482 * Reset counters so that the request stacking driver
1483 * can find how many bytes remain in the request
1486 req->__data_len = 0;
1490 req->__data_len -= total_bytes;
1492 /* update sector only for requests with clear definition of sector */
1493 if (!blk_rq_is_passthrough(req))
1494 req->__sector += total_bytes >> 9;
1496 /* mixed attributes always follow the first bio */
1497 if (req->rq_flags & RQF_MIXED_MERGE) {
1498 req->cmd_flags &= ~REQ_FAILFAST_MASK;
1499 req->cmd_flags |= req->bio->bi_opf & REQ_FAILFAST_MASK;
1502 if (!(req->rq_flags & RQF_SPECIAL_PAYLOAD)) {
1504 * If total number of sectors is less than the first segment
1505 * size, something has gone terribly wrong.
1507 if (blk_rq_bytes(req) < blk_rq_cur_bytes(req)) {
1508 blk_dump_rq_flags(req, "request botched");
1509 req->__data_len = blk_rq_cur_bytes(req);
1512 /* recalculate the number of segments */
1513 req->nr_phys_segments = blk_recalc_rq_segments(req);
1518 EXPORT_SYMBOL_GPL(blk_update_request);
1520 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
1522 * rq_flush_dcache_pages - Helper function to flush all pages in a request
1523 * @rq: the request to be flushed
1526 * Flush all pages in @rq.
1528 void rq_flush_dcache_pages(struct request *rq)
1530 struct req_iterator iter;
1531 struct bio_vec bvec;
1533 rq_for_each_segment(bvec, rq, iter)
1534 flush_dcache_page(bvec.bv_page);
1536 EXPORT_SYMBOL_GPL(rq_flush_dcache_pages);
1540 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
1541 * @q : the queue of the device being checked
1544 * Check if underlying low-level drivers of a device are busy.
1545 * If the drivers want to export their busy state, they must set own
1546 * exporting function using blk_queue_lld_busy() first.
1548 * Basically, this function is used only by request stacking drivers
1549 * to stop dispatching requests to underlying devices when underlying
1550 * devices are busy. This behavior helps more I/O merging on the queue
1551 * of the request stacking driver and prevents I/O throughput regression
1552 * on burst I/O load.
1555 * 0 - Not busy (The request stacking driver should dispatch request)
1556 * 1 - Busy (The request stacking driver should stop dispatching request)
1558 int blk_lld_busy(struct request_queue *q)
1560 if (queue_is_mq(q) && q->mq_ops->busy)
1561 return q->mq_ops->busy(q);
1565 EXPORT_SYMBOL_GPL(blk_lld_busy);
1568 * blk_rq_unprep_clone - Helper function to free all bios in a cloned request
1569 * @rq: the clone request to be cleaned up
1572 * Free all bios in @rq for a cloned request.
1574 void blk_rq_unprep_clone(struct request *rq)
1578 while ((bio = rq->bio) != NULL) {
1579 rq->bio = bio->bi_next;
1584 EXPORT_SYMBOL_GPL(blk_rq_unprep_clone);
1587 * blk_rq_prep_clone - Helper function to setup clone request
1588 * @rq: the request to be setup
1589 * @rq_src: original request to be cloned
1590 * @bs: bio_set that bios for clone are allocated from
1591 * @gfp_mask: memory allocation mask for bio
1592 * @bio_ctr: setup function to be called for each clone bio.
1593 * Returns %0 for success, non %0 for failure.
1594 * @data: private data to be passed to @bio_ctr
1597 * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq.
1598 * Also, pages which the original bios are pointing to are not copied
1599 * and the cloned bios just point same pages.
1600 * So cloned bios must be completed before original bios, which means
1601 * the caller must complete @rq before @rq_src.
1603 int blk_rq_prep_clone(struct request *rq, struct request *rq_src,
1604 struct bio_set *bs, gfp_t gfp_mask,
1605 int (*bio_ctr)(struct bio *, struct bio *, void *),
1608 struct bio *bio, *bio_src;
1613 __rq_for_each_bio(bio_src, rq_src) {
1614 bio = bio_clone_fast(bio_src, gfp_mask, bs);
1618 if (bio_ctr && bio_ctr(bio, bio_src, data))
1622 rq->biotail->bi_next = bio;
1625 rq->bio = rq->biotail = bio;
1630 /* Copy attributes of the original request to the clone request. */
1631 rq->__sector = blk_rq_pos(rq_src);
1632 rq->__data_len = blk_rq_bytes(rq_src);
1633 if (rq_src->rq_flags & RQF_SPECIAL_PAYLOAD) {
1634 rq->rq_flags |= RQF_SPECIAL_PAYLOAD;
1635 rq->special_vec = rq_src->special_vec;
1637 rq->nr_phys_segments = rq_src->nr_phys_segments;
1638 rq->ioprio = rq_src->ioprio;
1640 if (rq->bio && blk_crypto_rq_bio_prep(rq, rq->bio, gfp_mask) < 0)
1648 blk_rq_unprep_clone(rq);
1652 EXPORT_SYMBOL_GPL(blk_rq_prep_clone);
1654 int kblockd_schedule_work(struct work_struct *work)
1656 return queue_work(kblockd_workqueue, work);
1658 EXPORT_SYMBOL(kblockd_schedule_work);
1660 int kblockd_mod_delayed_work_on(int cpu, struct delayed_work *dwork,
1661 unsigned long delay)
1663 return mod_delayed_work_on(cpu, kblockd_workqueue, dwork, delay);
1665 EXPORT_SYMBOL(kblockd_mod_delayed_work_on);
1668 * blk_start_plug - initialize blk_plug and track it inside the task_struct
1669 * @plug: The &struct blk_plug that needs to be initialized
1672 * blk_start_plug() indicates to the block layer an intent by the caller
1673 * to submit multiple I/O requests in a batch. The block layer may use
1674 * this hint to defer submitting I/Os from the caller until blk_finish_plug()
1675 * is called. However, the block layer may choose to submit requests
1676 * before a call to blk_finish_plug() if the number of queued I/Os
1677 * exceeds %BLK_MAX_REQUEST_COUNT, or if the size of the I/O is larger than
1678 * %BLK_PLUG_FLUSH_SIZE. The queued I/Os may also be submitted early if
1679 * the task schedules (see below).
1681 * Tracking blk_plug inside the task_struct will help with auto-flushing the
1682 * pending I/O should the task end up blocking between blk_start_plug() and
1683 * blk_finish_plug(). This is important from a performance perspective, but
1684 * also ensures that we don't deadlock. For instance, if the task is blocking
1685 * for a memory allocation, memory reclaim could end up wanting to free a
1686 * page belonging to that request that is currently residing in our private
1687 * plug. By flushing the pending I/O when the process goes to sleep, we avoid
1688 * this kind of deadlock.
1690 void blk_start_plug(struct blk_plug *plug)
1692 struct task_struct *tsk = current;
1695 * If this is a nested plug, don't actually assign it.
1700 INIT_LIST_HEAD(&plug->mq_list);
1701 INIT_LIST_HEAD(&plug->cb_list);
1703 plug->multiple_queues = false;
1704 plug->nowait = false;
1707 * Store ordering should not be needed here, since a potential
1708 * preempt will imply a full memory barrier
1712 EXPORT_SYMBOL(blk_start_plug);
1714 static void flush_plug_callbacks(struct blk_plug *plug, bool from_schedule)
1716 LIST_HEAD(callbacks);
1718 while (!list_empty(&plug->cb_list)) {
1719 list_splice_init(&plug->cb_list, &callbacks);
1721 while (!list_empty(&callbacks)) {
1722 struct blk_plug_cb *cb = list_first_entry(&callbacks,
1725 list_del(&cb->list);
1726 cb->callback(cb, from_schedule);
1731 struct blk_plug_cb *blk_check_plugged(blk_plug_cb_fn unplug, void *data,
1734 struct blk_plug *plug = current->plug;
1735 struct blk_plug_cb *cb;
1740 list_for_each_entry(cb, &plug->cb_list, list)
1741 if (cb->callback == unplug && cb->data == data)
1744 /* Not currently on the callback list */
1745 BUG_ON(size < sizeof(*cb));
1746 cb = kzalloc(size, GFP_ATOMIC);
1749 cb->callback = unplug;
1750 list_add(&cb->list, &plug->cb_list);
1754 EXPORT_SYMBOL(blk_check_plugged);
1756 void blk_flush_plug_list(struct blk_plug *plug, bool from_schedule)
1758 flush_plug_callbacks(plug, from_schedule);
1760 if (!list_empty(&plug->mq_list))
1761 blk_mq_flush_plug_list(plug, from_schedule);
1765 * blk_finish_plug - mark the end of a batch of submitted I/O
1766 * @plug: The &struct blk_plug passed to blk_start_plug()
1769 * Indicate that a batch of I/O submissions is complete. This function
1770 * must be paired with an initial call to blk_start_plug(). The intent
1771 * is to allow the block layer to optimize I/O submission. See the
1772 * documentation for blk_start_plug() for more information.
1774 void blk_finish_plug(struct blk_plug *plug)
1776 if (plug != current->plug)
1778 blk_flush_plug_list(plug, false);
1780 current->plug = NULL;
1782 EXPORT_SYMBOL(blk_finish_plug);
1784 void blk_io_schedule(void)
1786 /* Prevent hang_check timer from firing at us during very long I/O */
1787 unsigned long timeout = sysctl_hung_task_timeout_secs * HZ / 2;
1790 io_schedule_timeout(timeout);
1794 EXPORT_SYMBOL_GPL(blk_io_schedule);
1796 int __init blk_dev_init(void)
1798 BUILD_BUG_ON(REQ_OP_LAST >= (1 << REQ_OP_BITS));
1799 BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 *
1800 sizeof_field(struct request, cmd_flags));
1801 BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 *
1802 sizeof_field(struct bio, bi_opf));
1804 /* used for unplugging and affects IO latency/throughput - HIGHPRI */
1805 kblockd_workqueue = alloc_workqueue("kblockd",
1806 WQ_MEM_RECLAIM | WQ_HIGHPRI, 0);
1807 if (!kblockd_workqueue)
1808 panic("Failed to create kblockd\n");
1810 blk_requestq_cachep = kmem_cache_create("request_queue",
1811 sizeof(struct request_queue), 0, SLAB_PANIC, NULL);
1813 blk_debugfs_root = debugfs_create_dir("block", NULL);