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 refcount_set(&rq->ref, 1);
125 blk_crypto_rq_set_defaults(rq);
127 EXPORT_SYMBOL(blk_rq_init);
129 #define REQ_OP_NAME(name) [REQ_OP_##name] = #name
130 static const char *const blk_op_name[] = {
134 REQ_OP_NAME(DISCARD),
135 REQ_OP_NAME(SECURE_ERASE),
136 REQ_OP_NAME(ZONE_RESET),
137 REQ_OP_NAME(ZONE_RESET_ALL),
138 REQ_OP_NAME(ZONE_OPEN),
139 REQ_OP_NAME(ZONE_CLOSE),
140 REQ_OP_NAME(ZONE_FINISH),
141 REQ_OP_NAME(ZONE_APPEND),
142 REQ_OP_NAME(WRITE_SAME),
143 REQ_OP_NAME(WRITE_ZEROES),
144 REQ_OP_NAME(SCSI_IN),
145 REQ_OP_NAME(SCSI_OUT),
147 REQ_OP_NAME(DRV_OUT),
152 * blk_op_str - Return string XXX in the REQ_OP_XXX.
155 * Description: Centralize block layer function to convert REQ_OP_XXX into
156 * string format. Useful in the debugging and tracing bio or request. For
157 * invalid REQ_OP_XXX it returns string "UNKNOWN".
159 inline const char *blk_op_str(unsigned int op)
161 const char *op_str = "UNKNOWN";
163 if (op < ARRAY_SIZE(blk_op_name) && blk_op_name[op])
164 op_str = blk_op_name[op];
168 EXPORT_SYMBOL_GPL(blk_op_str);
170 static const struct {
174 [BLK_STS_OK] = { 0, "" },
175 [BLK_STS_NOTSUPP] = { -EOPNOTSUPP, "operation not supported" },
176 [BLK_STS_TIMEOUT] = { -ETIMEDOUT, "timeout" },
177 [BLK_STS_NOSPC] = { -ENOSPC, "critical space allocation" },
178 [BLK_STS_TRANSPORT] = { -ENOLINK, "recoverable transport" },
179 [BLK_STS_TARGET] = { -EREMOTEIO, "critical target" },
180 [BLK_STS_NEXUS] = { -EBADE, "critical nexus" },
181 [BLK_STS_MEDIUM] = { -ENODATA, "critical medium" },
182 [BLK_STS_PROTECTION] = { -EILSEQ, "protection" },
183 [BLK_STS_RESOURCE] = { -ENOMEM, "kernel resource" },
184 [BLK_STS_DEV_RESOURCE] = { -EBUSY, "device resource" },
185 [BLK_STS_AGAIN] = { -EAGAIN, "nonblocking retry" },
187 /* device mapper special case, should not leak out: */
188 [BLK_STS_DM_REQUEUE] = { -EREMCHG, "dm internal retry" },
190 /* zone device specific errors */
191 [BLK_STS_ZONE_OPEN_RESOURCE] = { -ETOOMANYREFS, "open zones exceeded" },
192 [BLK_STS_ZONE_ACTIVE_RESOURCE] = { -EOVERFLOW, "active zones exceeded" },
194 /* everything else not covered above: */
195 [BLK_STS_IOERR] = { -EIO, "I/O" },
198 blk_status_t errno_to_blk_status(int errno)
202 for (i = 0; i < ARRAY_SIZE(blk_errors); i++) {
203 if (blk_errors[i].errno == errno)
204 return (__force blk_status_t)i;
207 return BLK_STS_IOERR;
209 EXPORT_SYMBOL_GPL(errno_to_blk_status);
211 int blk_status_to_errno(blk_status_t status)
213 int idx = (__force int)status;
215 if (WARN_ON_ONCE(idx >= ARRAY_SIZE(blk_errors)))
217 return blk_errors[idx].errno;
219 EXPORT_SYMBOL_GPL(blk_status_to_errno);
221 static void print_req_error(struct request *req, blk_status_t status,
224 int idx = (__force int)status;
226 if (WARN_ON_ONCE(idx >= ARRAY_SIZE(blk_errors)))
229 printk_ratelimited(KERN_ERR
230 "%s: %s error, dev %s, sector %llu op 0x%x:(%s) flags 0x%x "
231 "phys_seg %u prio class %u\n",
232 caller, blk_errors[idx].name,
233 req->rq_disk ? req->rq_disk->disk_name : "?",
234 blk_rq_pos(req), req_op(req), blk_op_str(req_op(req)),
235 req->cmd_flags & ~REQ_OP_MASK,
236 req->nr_phys_segments,
237 IOPRIO_PRIO_CLASS(req->ioprio));
240 static void req_bio_endio(struct request *rq, struct bio *bio,
241 unsigned int nbytes, blk_status_t error)
244 bio->bi_status = error;
246 if (unlikely(rq->rq_flags & RQF_QUIET))
247 bio_set_flag(bio, BIO_QUIET);
249 bio_advance(bio, nbytes);
251 if (req_op(rq) == REQ_OP_ZONE_APPEND && error == BLK_STS_OK) {
253 * Partial zone append completions cannot be supported as the
254 * BIO fragments may end up not being written sequentially.
256 if (bio->bi_iter.bi_size)
257 bio->bi_status = BLK_STS_IOERR;
259 bio->bi_iter.bi_sector = rq->__sector;
262 /* don't actually finish bio if it's part of flush sequence */
263 if (bio->bi_iter.bi_size == 0 && !(rq->rq_flags & RQF_FLUSH_SEQ))
267 void blk_dump_rq_flags(struct request *rq, char *msg)
269 printk(KERN_INFO "%s: dev %s: flags=%llx\n", msg,
270 rq->rq_disk ? rq->rq_disk->disk_name : "?",
271 (unsigned long long) rq->cmd_flags);
273 printk(KERN_INFO " sector %llu, nr/cnr %u/%u\n",
274 (unsigned long long)blk_rq_pos(rq),
275 blk_rq_sectors(rq), blk_rq_cur_sectors(rq));
276 printk(KERN_INFO " bio %p, biotail %p, len %u\n",
277 rq->bio, rq->biotail, blk_rq_bytes(rq));
279 EXPORT_SYMBOL(blk_dump_rq_flags);
282 * blk_sync_queue - cancel any pending callbacks on a queue
286 * The block layer may perform asynchronous callback activity
287 * on a queue, such as calling the unplug function after a timeout.
288 * A block device may call blk_sync_queue to ensure that any
289 * such activity is cancelled, thus allowing it to release resources
290 * that the callbacks might use. The caller must already have made sure
291 * that its ->submit_bio will not re-add plugging prior to calling
294 * This function does not cancel any asynchronous activity arising
295 * out of elevator or throttling code. That would require elevator_exit()
296 * and blkcg_exit_queue() to be called with queue lock initialized.
299 void blk_sync_queue(struct request_queue *q)
301 del_timer_sync(&q->timeout);
302 cancel_work_sync(&q->timeout_work);
304 EXPORT_SYMBOL(blk_sync_queue);
307 * blk_set_pm_only - increment pm_only counter
308 * @q: request queue pointer
310 void blk_set_pm_only(struct request_queue *q)
312 atomic_inc(&q->pm_only);
314 EXPORT_SYMBOL_GPL(blk_set_pm_only);
316 void blk_clear_pm_only(struct request_queue *q)
320 pm_only = atomic_dec_return(&q->pm_only);
321 WARN_ON_ONCE(pm_only < 0);
323 wake_up_all(&q->mq_freeze_wq);
325 EXPORT_SYMBOL_GPL(blk_clear_pm_only);
328 * blk_put_queue - decrement the request_queue refcount
329 * @q: the request_queue structure to decrement the refcount for
331 * Decrements the refcount of the request_queue kobject. When this reaches 0
332 * we'll have blk_release_queue() called.
334 * Context: Any context, but the last reference must not be dropped from
337 void blk_put_queue(struct request_queue *q)
339 kobject_put(&q->kobj);
341 EXPORT_SYMBOL(blk_put_queue);
343 void blk_set_queue_dying(struct request_queue *q)
345 blk_queue_flag_set(QUEUE_FLAG_DYING, q);
348 * When queue DYING flag is set, we need to block new req
349 * entering queue, so we call blk_freeze_queue_start() to
350 * prevent I/O from crossing blk_queue_enter().
352 blk_freeze_queue_start(q);
355 blk_mq_wake_waiters(q);
357 /* Make blk_queue_enter() reexamine the DYING flag. */
358 wake_up_all(&q->mq_freeze_wq);
360 EXPORT_SYMBOL_GPL(blk_set_queue_dying);
363 * blk_cleanup_queue - shutdown a request queue
364 * @q: request queue to shutdown
366 * Mark @q DYING, drain all pending requests, mark @q DEAD, destroy and
367 * put it. All future requests will be failed immediately with -ENODEV.
371 void blk_cleanup_queue(struct request_queue *q)
373 /* cannot be called from atomic context */
376 WARN_ON_ONCE(blk_queue_registered(q));
378 /* mark @q DYING, no new request or merges will be allowed afterwards */
379 blk_set_queue_dying(q);
381 blk_queue_flag_set(QUEUE_FLAG_NOMERGES, q);
382 blk_queue_flag_set(QUEUE_FLAG_NOXMERGES, q);
385 * Drain all requests queued before DYING marking. Set DEAD flag to
386 * prevent that blk_mq_run_hw_queues() accesses the hardware queues
387 * after draining finished.
393 blk_queue_flag_set(QUEUE_FLAG_DEAD, q);
395 /* for synchronous bio-based driver finish in-flight integrity i/o */
396 blk_flush_integrity();
398 /* @q won't process any more request, flush async actions */
399 del_timer_sync(&q->backing_dev_info->laptop_mode_wb_timer);
403 blk_mq_exit_queue(q);
406 * In theory, request pool of sched_tags belongs to request queue.
407 * However, the current implementation requires tag_set for freeing
408 * requests, so free the pool now.
410 * Queue has become frozen, there can't be any in-queue requests, so
411 * it is safe to free requests now.
413 mutex_lock(&q->sysfs_lock);
415 blk_mq_sched_free_requests(q);
416 mutex_unlock(&q->sysfs_lock);
418 percpu_ref_exit(&q->q_usage_counter);
420 /* @q is and will stay empty, shutdown and put */
423 EXPORT_SYMBOL(blk_cleanup_queue);
426 * blk_queue_enter() - try to increase q->q_usage_counter
427 * @q: request queue pointer
428 * @flags: BLK_MQ_REQ_NOWAIT and/or BLK_MQ_REQ_PM
430 int blk_queue_enter(struct request_queue *q, blk_mq_req_flags_t flags)
432 const bool pm = flags & BLK_MQ_REQ_PM;
435 bool success = false;
438 if (percpu_ref_tryget_live(&q->q_usage_counter)) {
440 * The code that increments the pm_only counter is
441 * responsible for ensuring that that counter is
442 * globally visible before the queue is unfrozen.
444 if ((pm && queue_rpm_status(q) != RPM_SUSPENDED) ||
445 !blk_queue_pm_only(q)) {
448 percpu_ref_put(&q->q_usage_counter);
456 if (flags & BLK_MQ_REQ_NOWAIT)
460 * read pair of barrier in blk_freeze_queue_start(),
461 * we need to order reading __PERCPU_REF_DEAD flag of
462 * .q_usage_counter and reading .mq_freeze_depth or
463 * queue dying flag, otherwise the following wait may
464 * never return if the two reads are reordered.
468 wait_event(q->mq_freeze_wq,
469 (!q->mq_freeze_depth &&
470 blk_pm_resume_queue(pm, q)) ||
472 if (blk_queue_dying(q))
477 static inline int bio_queue_enter(struct bio *bio)
479 struct request_queue *q = bio->bi_bdev->bd_disk->queue;
480 bool nowait = bio->bi_opf & REQ_NOWAIT;
483 ret = blk_queue_enter(q, nowait ? BLK_MQ_REQ_NOWAIT : 0);
485 if (nowait && !blk_queue_dying(q))
486 bio_wouldblock_error(bio);
494 void blk_queue_exit(struct request_queue *q)
496 percpu_ref_put(&q->q_usage_counter);
499 static void blk_queue_usage_counter_release(struct percpu_ref *ref)
501 struct request_queue *q =
502 container_of(ref, struct request_queue, q_usage_counter);
504 wake_up_all(&q->mq_freeze_wq);
507 static void blk_rq_timed_out_timer(struct timer_list *t)
509 struct request_queue *q = from_timer(q, t, timeout);
511 kblockd_schedule_work(&q->timeout_work);
514 static void blk_timeout_work(struct work_struct *work)
518 struct request_queue *blk_alloc_queue(int node_id)
520 struct request_queue *q;
523 q = kmem_cache_alloc_node(blk_requestq_cachep,
524 GFP_KERNEL | __GFP_ZERO, node_id);
528 q->last_merge = NULL;
530 q->id = ida_simple_get(&blk_queue_ida, 0, 0, GFP_KERNEL);
534 ret = bioset_init(&q->bio_split, BIO_POOL_SIZE, 0, 0);
538 q->backing_dev_info = bdi_alloc(node_id);
539 if (!q->backing_dev_info)
542 q->stats = blk_alloc_queue_stats();
548 atomic_set(&q->nr_active_requests_shared_sbitmap, 0);
550 timer_setup(&q->backing_dev_info->laptop_mode_wb_timer,
551 laptop_mode_timer_fn, 0);
552 timer_setup(&q->timeout, blk_rq_timed_out_timer, 0);
553 INIT_WORK(&q->timeout_work, blk_timeout_work);
554 INIT_LIST_HEAD(&q->icq_list);
555 #ifdef CONFIG_BLK_CGROUP
556 INIT_LIST_HEAD(&q->blkg_list);
559 kobject_init(&q->kobj, &blk_queue_ktype);
561 mutex_init(&q->debugfs_mutex);
562 mutex_init(&q->sysfs_lock);
563 mutex_init(&q->sysfs_dir_lock);
564 spin_lock_init(&q->queue_lock);
566 init_waitqueue_head(&q->mq_freeze_wq);
567 mutex_init(&q->mq_freeze_lock);
570 * Init percpu_ref in atomic mode so that it's faster to shutdown.
571 * See blk_register_queue() for details.
573 if (percpu_ref_init(&q->q_usage_counter,
574 blk_queue_usage_counter_release,
575 PERCPU_REF_INIT_ATOMIC, GFP_KERNEL))
578 if (blkcg_init_queue(q))
581 blk_queue_dma_alignment(q, 511);
582 blk_set_default_limits(&q->limits);
583 q->nr_requests = BLKDEV_MAX_RQ;
588 percpu_ref_exit(&q->q_usage_counter);
590 blk_free_queue_stats(q->stats);
592 bdi_put(q->backing_dev_info);
594 bioset_exit(&q->bio_split);
596 ida_simple_remove(&blk_queue_ida, q->id);
598 kmem_cache_free(blk_requestq_cachep, q);
601 EXPORT_SYMBOL(blk_alloc_queue);
604 * blk_get_queue - increment the request_queue refcount
605 * @q: the request_queue structure to increment the refcount for
607 * Increment the refcount of the request_queue kobject.
609 * Context: Any context.
611 bool blk_get_queue(struct request_queue *q)
613 if (likely(!blk_queue_dying(q))) {
620 EXPORT_SYMBOL(blk_get_queue);
623 * blk_get_request - allocate a request
624 * @q: request queue to allocate a request for
625 * @op: operation (REQ_OP_*) and REQ_* flags, e.g. REQ_SYNC.
626 * @flags: BLK_MQ_REQ_* flags, e.g. BLK_MQ_REQ_NOWAIT.
628 struct request *blk_get_request(struct request_queue *q, unsigned int op,
629 blk_mq_req_flags_t flags)
633 WARN_ON_ONCE(op & REQ_NOWAIT);
634 WARN_ON_ONCE(flags & ~(BLK_MQ_REQ_NOWAIT | BLK_MQ_REQ_PM));
636 req = blk_mq_alloc_request(q, op, flags);
637 if (!IS_ERR(req) && q->mq_ops->initialize_rq_fn)
638 q->mq_ops->initialize_rq_fn(req);
642 EXPORT_SYMBOL(blk_get_request);
644 void blk_put_request(struct request *req)
646 blk_mq_free_request(req);
648 EXPORT_SYMBOL(blk_put_request);
650 static void handle_bad_sector(struct bio *bio, sector_t maxsector)
652 char b[BDEVNAME_SIZE];
654 pr_info_ratelimited("attempt to access beyond end of device\n"
655 "%s: rw=%d, want=%llu, limit=%llu\n",
656 bio_devname(bio, b), bio->bi_opf,
657 bio_end_sector(bio), maxsector);
660 #ifdef CONFIG_FAIL_MAKE_REQUEST
662 static DECLARE_FAULT_ATTR(fail_make_request);
664 static int __init setup_fail_make_request(char *str)
666 return setup_fault_attr(&fail_make_request, str);
668 __setup("fail_make_request=", setup_fail_make_request);
670 static bool should_fail_request(struct block_device *part, unsigned int bytes)
672 return part->bd_make_it_fail && should_fail(&fail_make_request, bytes);
675 static int __init fail_make_request_debugfs(void)
677 struct dentry *dir = fault_create_debugfs_attr("fail_make_request",
678 NULL, &fail_make_request);
680 return PTR_ERR_OR_ZERO(dir);
683 late_initcall(fail_make_request_debugfs);
685 #else /* CONFIG_FAIL_MAKE_REQUEST */
687 static inline bool should_fail_request(struct block_device *part,
693 #endif /* CONFIG_FAIL_MAKE_REQUEST */
695 static inline bool bio_check_ro(struct bio *bio)
697 if (op_is_write(bio_op(bio)) && bdev_read_only(bio->bi_bdev)) {
698 char b[BDEVNAME_SIZE];
700 if (op_is_flush(bio->bi_opf) && !bio_sectors(bio))
704 "Trying to write to read-only block-device %s (partno %d)\n",
705 bio_devname(bio, b), bio->bi_bdev->bd_partno);
706 /* Older lvm-tools actually trigger this */
713 static noinline int should_fail_bio(struct bio *bio)
715 if (should_fail_request(bdev_whole(bio->bi_bdev), bio->bi_iter.bi_size))
719 ALLOW_ERROR_INJECTION(should_fail_bio, ERRNO);
722 * Check whether this bio extends beyond the end of the device or partition.
723 * This may well happen - the kernel calls bread() without checking the size of
724 * the device, e.g., when mounting a file system.
726 static inline int bio_check_eod(struct bio *bio)
728 sector_t maxsector = bdev_nr_sectors(bio->bi_bdev);
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 int blk_partition_remap(struct bio *bio)
745 struct block_device *p = bio->bi_bdev;
747 if (unlikely(should_fail_request(p, bio->bi_iter.bi_size)))
749 if (bio_sectors(bio)) {
750 bio->bi_iter.bi_sector += p->bd_start_sect;
751 trace_block_bio_remap(bio, p->bd_dev,
752 bio->bi_iter.bi_sector -
755 bio_set_flag(bio, BIO_REMAPPED);
760 * Check write append to a zoned block device.
762 static inline blk_status_t blk_check_zone_append(struct request_queue *q,
765 sector_t pos = bio->bi_iter.bi_sector;
766 int nr_sectors = bio_sectors(bio);
768 /* Only applicable to zoned block devices */
769 if (!blk_queue_is_zoned(q))
770 return BLK_STS_NOTSUPP;
772 /* The bio sector must point to the start of a sequential zone */
773 if (pos & (blk_queue_zone_sectors(q) - 1) ||
774 !blk_queue_zone_is_seq(q, pos))
775 return BLK_STS_IOERR;
778 * Not allowed to cross zone boundaries. Otherwise, the BIO will be
779 * split and could result in non-contiguous sectors being written in
782 if (nr_sectors > q->limits.chunk_sectors)
783 return BLK_STS_IOERR;
785 /* Make sure the BIO is small enough and will not get split */
786 if (nr_sectors > q->limits.max_zone_append_sectors)
787 return BLK_STS_IOERR;
789 bio->bi_opf |= REQ_NOMERGE;
794 static noinline_for_stack bool submit_bio_checks(struct bio *bio)
796 struct block_device *bdev = bio->bi_bdev;
797 struct request_queue *q = bdev->bd_disk->queue;
798 blk_status_t status = BLK_STS_IOERR;
799 struct blk_plug *plug;
803 plug = blk_mq_plug(q, bio);
804 if (plug && plug->nowait)
805 bio->bi_opf |= REQ_NOWAIT;
808 * For a REQ_NOWAIT based request, return -EOPNOTSUPP
809 * if queue does not support NOWAIT.
811 if ((bio->bi_opf & REQ_NOWAIT) && !blk_queue_nowait(q))
814 if (should_fail_bio(bio))
816 if (unlikely(bio_check_ro(bio)))
818 if (!bio_flagged(bio, BIO_REMAPPED)) {
819 if (unlikely(bio_check_eod(bio)))
821 if (bdev->bd_partno && unlikely(blk_partition_remap(bio)))
826 * Filter flush bio's early so that bio based drivers without flush
827 * support don't have to worry about them.
829 if (op_is_flush(bio->bi_opf) &&
830 !test_bit(QUEUE_FLAG_WC, &q->queue_flags)) {
831 bio->bi_opf &= ~(REQ_PREFLUSH | REQ_FUA);
832 if (!bio_sectors(bio)) {
838 if (!test_bit(QUEUE_FLAG_POLL, &q->queue_flags))
839 bio->bi_opf &= ~REQ_HIPRI;
841 switch (bio_op(bio)) {
843 if (!blk_queue_discard(q))
846 case REQ_OP_SECURE_ERASE:
847 if (!blk_queue_secure_erase(q))
850 case REQ_OP_WRITE_SAME:
851 if (!q->limits.max_write_same_sectors)
854 case REQ_OP_ZONE_APPEND:
855 status = blk_check_zone_append(q, bio);
856 if (status != BLK_STS_OK)
859 case REQ_OP_ZONE_RESET:
860 case REQ_OP_ZONE_OPEN:
861 case REQ_OP_ZONE_CLOSE:
862 case REQ_OP_ZONE_FINISH:
863 if (!blk_queue_is_zoned(q))
866 case REQ_OP_ZONE_RESET_ALL:
867 if (!blk_queue_is_zoned(q) || !blk_queue_zone_resetall(q))
870 case REQ_OP_WRITE_ZEROES:
871 if (!q->limits.max_write_zeroes_sectors)
879 * Various block parts want %current->io_context, so allocate it up
880 * front rather than dealing with lots of pain to allocate it only
881 * where needed. This may fail and the block layer knows how to live
884 if (unlikely(!current->io_context))
885 create_task_io_context(current, GFP_ATOMIC, q->node);
887 if (blk_throtl_bio(bio)) {
888 blkcg_bio_issue_init(bio);
892 blk_cgroup_bio_start(bio);
893 blkcg_bio_issue_init(bio);
895 if (!bio_flagged(bio, BIO_TRACE_COMPLETION)) {
896 trace_block_bio_queue(bio);
897 /* Now that enqueuing has been traced, we need to trace
898 * completion as well.
900 bio_set_flag(bio, BIO_TRACE_COMPLETION);
905 status = BLK_STS_NOTSUPP;
907 bio->bi_status = status;
912 static blk_qc_t __submit_bio(struct bio *bio)
914 struct gendisk *disk = bio->bi_bdev->bd_disk;
915 blk_qc_t ret = BLK_QC_T_NONE;
917 if (blk_crypto_bio_prep(&bio)) {
918 if (!disk->fops->submit_bio)
919 return blk_mq_submit_bio(bio);
920 ret = disk->fops->submit_bio(bio);
922 blk_queue_exit(disk->queue);
927 * The loop in this function may be a bit non-obvious, and so deserves some
930 * - Before entering the loop, bio->bi_next is NULL (as all callers ensure
931 * that), so we have a list with a single bio.
932 * - We pretend that we have just taken it off a longer list, so we assign
933 * bio_list to a pointer to the bio_list_on_stack, thus initialising the
934 * bio_list of new bios to be added. ->submit_bio() may indeed add some more
935 * bios through a recursive call to submit_bio_noacct. If it did, we find a
936 * non-NULL value in bio_list and re-enter the loop from the top.
937 * - In this case we really did just take the bio of the top of the list (no
938 * pretending) and so remove it from bio_list, and call into ->submit_bio()
941 * bio_list_on_stack[0] contains bios submitted by the current ->submit_bio.
942 * bio_list_on_stack[1] contains bios that were submitted before the current
943 * ->submit_bio_bio, but that haven't been processed yet.
945 static blk_qc_t __submit_bio_noacct(struct bio *bio)
947 struct bio_list bio_list_on_stack[2];
948 blk_qc_t ret = BLK_QC_T_NONE;
950 BUG_ON(bio->bi_next);
952 bio_list_init(&bio_list_on_stack[0]);
953 current->bio_list = bio_list_on_stack;
956 struct request_queue *q = bio->bi_bdev->bd_disk->queue;
957 struct bio_list lower, same;
959 if (unlikely(bio_queue_enter(bio) != 0))
963 * Create a fresh bio_list for all subordinate requests.
965 bio_list_on_stack[1] = bio_list_on_stack[0];
966 bio_list_init(&bio_list_on_stack[0]);
968 ret = __submit_bio(bio);
971 * Sort new bios into those for a lower level and those for the
974 bio_list_init(&lower);
975 bio_list_init(&same);
976 while ((bio = bio_list_pop(&bio_list_on_stack[0])) != NULL)
977 if (q == bio->bi_bdev->bd_disk->queue)
978 bio_list_add(&same, bio);
980 bio_list_add(&lower, bio);
983 * Now assemble so we handle the lowest level first.
985 bio_list_merge(&bio_list_on_stack[0], &lower);
986 bio_list_merge(&bio_list_on_stack[0], &same);
987 bio_list_merge(&bio_list_on_stack[0], &bio_list_on_stack[1]);
988 } while ((bio = bio_list_pop(&bio_list_on_stack[0])));
990 current->bio_list = NULL;
994 static blk_qc_t __submit_bio_noacct_mq(struct bio *bio)
996 struct bio_list bio_list[2] = { };
997 blk_qc_t ret = BLK_QC_T_NONE;
999 current->bio_list = bio_list;
1002 struct gendisk *disk = bio->bi_bdev->bd_disk;
1004 if (unlikely(bio_queue_enter(bio) != 0))
1007 if (!blk_crypto_bio_prep(&bio)) {
1008 blk_queue_exit(disk->queue);
1009 ret = BLK_QC_T_NONE;
1013 ret = blk_mq_submit_bio(bio);
1014 } while ((bio = bio_list_pop(&bio_list[0])));
1016 current->bio_list = NULL;
1021 * submit_bio_noacct - re-submit a bio to the block device layer for I/O
1022 * @bio: The bio describing the location in memory and on the device.
1024 * This is a version of submit_bio() that shall only be used for I/O that is
1025 * resubmitted to lower level drivers by stacking block drivers. All file
1026 * systems and other upper level users of the block layer should use
1027 * submit_bio() instead.
1029 blk_qc_t submit_bio_noacct(struct bio *bio)
1031 if (!submit_bio_checks(bio))
1032 return BLK_QC_T_NONE;
1035 * We only want one ->submit_bio to be active at a time, else stack
1036 * usage with stacked devices could be a problem. Use current->bio_list
1037 * to collect a list of requests submited by a ->submit_bio method while
1038 * it is active, and then process them after it returned.
1040 if (current->bio_list) {
1041 bio_list_add(¤t->bio_list[0], bio);
1042 return BLK_QC_T_NONE;
1045 if (!bio->bi_bdev->bd_disk->fops->submit_bio)
1046 return __submit_bio_noacct_mq(bio);
1047 return __submit_bio_noacct(bio);
1049 EXPORT_SYMBOL(submit_bio_noacct);
1052 * submit_bio - submit a bio to the block device layer for I/O
1053 * @bio: The &struct bio which describes the I/O
1055 * submit_bio() is used to submit I/O requests to block devices. It is passed a
1056 * fully set up &struct bio that describes the I/O that needs to be done. The
1057 * bio will be send to the device described by the bi_bdev field.
1059 * The success/failure status of the request, along with notification of
1060 * completion, is delivered asynchronously through the ->bi_end_io() callback
1061 * in @bio. The bio must NOT be touched by thecaller until ->bi_end_io() has
1064 blk_qc_t submit_bio(struct bio *bio)
1066 if (blkcg_punt_bio_submit(bio))
1067 return BLK_QC_T_NONE;
1070 * If it's a regular read/write or a barrier with data attached,
1071 * go through the normal accounting stuff before submission.
1073 if (bio_has_data(bio)) {
1076 if (unlikely(bio_op(bio) == REQ_OP_WRITE_SAME))
1077 count = queue_logical_block_size(
1078 bio->bi_bdev->bd_disk->queue) >> 9;
1080 count = bio_sectors(bio);
1082 if (op_is_write(bio_op(bio))) {
1083 count_vm_events(PGPGOUT, count);
1085 task_io_account_read(bio->bi_iter.bi_size);
1086 count_vm_events(PGPGIN, count);
1089 if (unlikely(block_dump)) {
1090 char b[BDEVNAME_SIZE];
1091 printk(KERN_DEBUG "%s(%d): %s block %Lu on %s (%u sectors)\n",
1092 current->comm, task_pid_nr(current),
1093 op_is_write(bio_op(bio)) ? "WRITE" : "READ",
1094 (unsigned long long)bio->bi_iter.bi_sector,
1095 bio_devname(bio, b), count);
1100 * If we're reading data that is part of the userspace workingset, count
1101 * submission time as memory stall. When the device is congested, or
1102 * the submitting cgroup IO-throttled, submission can be a significant
1103 * part of overall IO time.
1105 if (unlikely(bio_op(bio) == REQ_OP_READ &&
1106 bio_flagged(bio, BIO_WORKINGSET))) {
1107 unsigned long pflags;
1110 psi_memstall_enter(&pflags);
1111 ret = submit_bio_noacct(bio);
1112 psi_memstall_leave(&pflags);
1117 return submit_bio_noacct(bio);
1119 EXPORT_SYMBOL(submit_bio);
1122 * blk_cloned_rq_check_limits - Helper function to check a cloned request
1123 * for the new queue limits
1125 * @rq: the request being checked
1128 * @rq may have been made based on weaker limitations of upper-level queues
1129 * in request stacking drivers, and it may violate the limitation of @q.
1130 * Since the block layer and the underlying device driver trust @rq
1131 * after it is inserted to @q, it should be checked against @q before
1132 * the insertion using this generic function.
1134 * Request stacking drivers like request-based dm may change the queue
1135 * limits when retrying requests on other queues. Those requests need
1136 * to be checked against the new queue limits again during dispatch.
1138 static blk_status_t blk_cloned_rq_check_limits(struct request_queue *q,
1141 unsigned int max_sectors = blk_queue_get_max_sectors(q, req_op(rq));
1143 if (blk_rq_sectors(rq) > max_sectors) {
1145 * SCSI device does not have a good way to return if
1146 * Write Same/Zero is actually supported. If a device rejects
1147 * a non-read/write command (discard, write same,etc.) the
1148 * low-level device driver will set the relevant queue limit to
1149 * 0 to prevent blk-lib from issuing more of the offending
1150 * operations. Commands queued prior to the queue limit being
1151 * reset need to be completed with BLK_STS_NOTSUPP to avoid I/O
1152 * errors being propagated to upper layers.
1154 if (max_sectors == 0)
1155 return BLK_STS_NOTSUPP;
1157 printk(KERN_ERR "%s: over max size limit. (%u > %u)\n",
1158 __func__, blk_rq_sectors(rq), max_sectors);
1159 return BLK_STS_IOERR;
1163 * queue's settings related to segment counting like q->bounce_pfn
1164 * may differ from that of other stacking queues.
1165 * Recalculate it to check the request correctly on this queue's
1168 rq->nr_phys_segments = blk_recalc_rq_segments(rq);
1169 if (rq->nr_phys_segments > queue_max_segments(q)) {
1170 printk(KERN_ERR "%s: over max segments limit. (%hu > %hu)\n",
1171 __func__, rq->nr_phys_segments, queue_max_segments(q));
1172 return BLK_STS_IOERR;
1179 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
1180 * @q: the queue to submit the request
1181 * @rq: the request being queued
1183 blk_status_t blk_insert_cloned_request(struct request_queue *q, struct request *rq)
1187 ret = blk_cloned_rq_check_limits(q, rq);
1188 if (ret != BLK_STS_OK)
1192 should_fail_request(rq->rq_disk->part0, blk_rq_bytes(rq)))
1193 return BLK_STS_IOERR;
1195 if (blk_crypto_insert_cloned_request(rq))
1196 return BLK_STS_IOERR;
1198 if (blk_queue_io_stat(q))
1199 blk_account_io_start(rq);
1202 * Since we have a scheduler attached on the top device,
1203 * bypass a potential scheduler on the bottom device for
1206 return blk_mq_request_issue_directly(rq, true);
1208 EXPORT_SYMBOL_GPL(blk_insert_cloned_request);
1211 * blk_rq_err_bytes - determine number of bytes till the next failure boundary
1212 * @rq: request to examine
1215 * A request could be merge of IOs which require different failure
1216 * handling. This function determines the number of bytes which
1217 * can be failed from the beginning of the request without
1218 * crossing into area which need to be retried further.
1221 * The number of bytes to fail.
1223 unsigned int blk_rq_err_bytes(const struct request *rq)
1225 unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK;
1226 unsigned int bytes = 0;
1229 if (!(rq->rq_flags & RQF_MIXED_MERGE))
1230 return blk_rq_bytes(rq);
1233 * Currently the only 'mixing' which can happen is between
1234 * different fastfail types. We can safely fail portions
1235 * which have all the failfast bits that the first one has -
1236 * the ones which are at least as eager to fail as the first
1239 for (bio = rq->bio; bio; bio = bio->bi_next) {
1240 if ((bio->bi_opf & ff) != ff)
1242 bytes += bio->bi_iter.bi_size;
1245 /* this could lead to infinite loop */
1246 BUG_ON(blk_rq_bytes(rq) && !bytes);
1249 EXPORT_SYMBOL_GPL(blk_rq_err_bytes);
1251 static void update_io_ticks(struct block_device *part, unsigned long now,
1254 unsigned long stamp;
1256 stamp = READ_ONCE(part->bd_stamp);
1257 if (unlikely(stamp != now)) {
1258 if (likely(cmpxchg(&part->bd_stamp, stamp, now) == stamp))
1259 __part_stat_add(part, io_ticks, end ? now - stamp : 1);
1261 if (part->bd_partno) {
1262 part = bdev_whole(part);
1267 static void blk_account_io_completion(struct request *req, unsigned int bytes)
1269 if (req->part && blk_do_io_stat(req)) {
1270 const int sgrp = op_stat_group(req_op(req));
1273 part_stat_add(req->part, sectors[sgrp], bytes >> 9);
1278 void blk_account_io_done(struct request *req, u64 now)
1281 * Account IO completion. flush_rq isn't accounted as a
1282 * normal IO on queueing nor completion. Accounting the
1283 * containing request is enough.
1285 if (req->part && blk_do_io_stat(req) &&
1286 !(req->rq_flags & RQF_FLUSH_SEQ)) {
1287 const int sgrp = op_stat_group(req_op(req));
1290 update_io_ticks(req->part, jiffies, true);
1291 part_stat_inc(req->part, ios[sgrp]);
1292 part_stat_add(req->part, nsecs[sgrp], now - req->start_time_ns);
1297 void blk_account_io_start(struct request *rq)
1299 if (!blk_do_io_stat(rq))
1302 /* passthrough requests can hold bios that do not have ->bi_bdev set */
1303 if (rq->bio && rq->bio->bi_bdev)
1304 rq->part = rq->bio->bi_bdev;
1306 rq->part = rq->rq_disk->part0;
1309 update_io_ticks(rq->part, jiffies, false);
1313 static unsigned long __part_start_io_acct(struct block_device *part,
1314 unsigned int sectors, unsigned int op)
1316 const int sgrp = op_stat_group(op);
1317 unsigned long now = READ_ONCE(jiffies);
1320 update_io_ticks(part, now, false);
1321 part_stat_inc(part, ios[sgrp]);
1322 part_stat_add(part, sectors[sgrp], sectors);
1323 part_stat_local_inc(part, in_flight[op_is_write(op)]);
1330 * bio_start_io_acct - start I/O accounting for bio based drivers
1331 * @bio: bio to start account for
1333 * Returns the start time that should be passed back to bio_end_io_acct().
1335 unsigned long bio_start_io_acct(struct bio *bio)
1337 return __part_start_io_acct(bio->bi_bdev, bio_sectors(bio), bio_op(bio));
1339 EXPORT_SYMBOL_GPL(bio_start_io_acct);
1341 unsigned long disk_start_io_acct(struct gendisk *disk, unsigned int sectors,
1344 return __part_start_io_acct(disk->part0, sectors, op);
1346 EXPORT_SYMBOL(disk_start_io_acct);
1348 static void __part_end_io_acct(struct block_device *part, unsigned int op,
1349 unsigned long start_time)
1351 const int sgrp = op_stat_group(op);
1352 unsigned long now = READ_ONCE(jiffies);
1353 unsigned long duration = now - start_time;
1356 update_io_ticks(part, now, true);
1357 part_stat_add(part, nsecs[sgrp], jiffies_to_nsecs(duration));
1358 part_stat_local_dec(part, in_flight[op_is_write(op)]);
1362 void bio_end_io_acct_remapped(struct bio *bio, unsigned long start_time,
1363 struct block_device *orig_bdev)
1365 __part_end_io_acct(orig_bdev, bio_op(bio), start_time);
1367 EXPORT_SYMBOL_GPL(bio_end_io_acct_remapped);
1369 void disk_end_io_acct(struct gendisk *disk, unsigned int op,
1370 unsigned long start_time)
1372 __part_end_io_acct(disk->part0, op, start_time);
1374 EXPORT_SYMBOL(disk_end_io_acct);
1377 * Steal bios from a request and add them to a bio list.
1378 * The request must not have been partially completed before.
1380 void blk_steal_bios(struct bio_list *list, struct request *rq)
1384 list->tail->bi_next = rq->bio;
1386 list->head = rq->bio;
1387 list->tail = rq->biotail;
1395 EXPORT_SYMBOL_GPL(blk_steal_bios);
1398 * blk_update_request - Special helper function for request stacking drivers
1399 * @req: the request being processed
1400 * @error: block status code
1401 * @nr_bytes: number of bytes to complete @req
1404 * Ends I/O on a number of bytes attached to @req, but doesn't complete
1405 * the request structure even if @req doesn't have leftover.
1406 * If @req has leftover, sets it up for the next range of segments.
1408 * This special helper function is only for request stacking drivers
1409 * (e.g. request-based dm) so that they can handle partial completion.
1410 * Actual device drivers should use blk_mq_end_request instead.
1412 * Passing the result of blk_rq_bytes() as @nr_bytes guarantees
1413 * %false return from this function.
1416 * The RQF_SPECIAL_PAYLOAD flag is ignored on purpose in both
1417 * blk_rq_bytes() and in blk_update_request().
1420 * %false - this request doesn't have any more data
1421 * %true - this request has more data
1423 bool blk_update_request(struct request *req, blk_status_t error,
1424 unsigned int nr_bytes)
1428 trace_block_rq_complete(req, blk_status_to_errno(error), nr_bytes);
1433 #ifdef CONFIG_BLK_DEV_INTEGRITY
1434 if (blk_integrity_rq(req) && req_op(req) == REQ_OP_READ &&
1435 error == BLK_STS_OK)
1436 req->q->integrity.profile->complete_fn(req, nr_bytes);
1439 if (unlikely(error && !blk_rq_is_passthrough(req) &&
1440 !(req->rq_flags & RQF_QUIET)))
1441 print_req_error(req, error, __func__);
1443 blk_account_io_completion(req, nr_bytes);
1447 struct bio *bio = req->bio;
1448 unsigned bio_bytes = min(bio->bi_iter.bi_size, nr_bytes);
1450 if (bio_bytes == bio->bi_iter.bi_size)
1451 req->bio = bio->bi_next;
1453 /* Completion has already been traced */
1454 bio_clear_flag(bio, BIO_TRACE_COMPLETION);
1455 req_bio_endio(req, bio, bio_bytes, error);
1457 total_bytes += bio_bytes;
1458 nr_bytes -= bio_bytes;
1469 * Reset counters so that the request stacking driver
1470 * can find how many bytes remain in the request
1473 req->__data_len = 0;
1477 req->__data_len -= total_bytes;
1479 /* update sector only for requests with clear definition of sector */
1480 if (!blk_rq_is_passthrough(req))
1481 req->__sector += total_bytes >> 9;
1483 /* mixed attributes always follow the first bio */
1484 if (req->rq_flags & RQF_MIXED_MERGE) {
1485 req->cmd_flags &= ~REQ_FAILFAST_MASK;
1486 req->cmd_flags |= req->bio->bi_opf & REQ_FAILFAST_MASK;
1489 if (!(req->rq_flags & RQF_SPECIAL_PAYLOAD)) {
1491 * If total number of sectors is less than the first segment
1492 * size, something has gone terribly wrong.
1494 if (blk_rq_bytes(req) < blk_rq_cur_bytes(req)) {
1495 blk_dump_rq_flags(req, "request botched");
1496 req->__data_len = blk_rq_cur_bytes(req);
1499 /* recalculate the number of segments */
1500 req->nr_phys_segments = blk_recalc_rq_segments(req);
1505 EXPORT_SYMBOL_GPL(blk_update_request);
1507 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
1509 * rq_flush_dcache_pages - Helper function to flush all pages in a request
1510 * @rq: the request to be flushed
1513 * Flush all pages in @rq.
1515 void rq_flush_dcache_pages(struct request *rq)
1517 struct req_iterator iter;
1518 struct bio_vec bvec;
1520 rq_for_each_segment(bvec, rq, iter)
1521 flush_dcache_page(bvec.bv_page);
1523 EXPORT_SYMBOL_GPL(rq_flush_dcache_pages);
1527 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
1528 * @q : the queue of the device being checked
1531 * Check if underlying low-level drivers of a device are busy.
1532 * If the drivers want to export their busy state, they must set own
1533 * exporting function using blk_queue_lld_busy() first.
1535 * Basically, this function is used only by request stacking drivers
1536 * to stop dispatching requests to underlying devices when underlying
1537 * devices are busy. This behavior helps more I/O merging on the queue
1538 * of the request stacking driver and prevents I/O throughput regression
1539 * on burst I/O load.
1542 * 0 - Not busy (The request stacking driver should dispatch request)
1543 * 1 - Busy (The request stacking driver should stop dispatching request)
1545 int blk_lld_busy(struct request_queue *q)
1547 if (queue_is_mq(q) && q->mq_ops->busy)
1548 return q->mq_ops->busy(q);
1552 EXPORT_SYMBOL_GPL(blk_lld_busy);
1555 * blk_rq_unprep_clone - Helper function to free all bios in a cloned request
1556 * @rq: the clone request to be cleaned up
1559 * Free all bios in @rq for a cloned request.
1561 void blk_rq_unprep_clone(struct request *rq)
1565 while ((bio = rq->bio) != NULL) {
1566 rq->bio = bio->bi_next;
1571 EXPORT_SYMBOL_GPL(blk_rq_unprep_clone);
1574 * blk_rq_prep_clone - Helper function to setup clone request
1575 * @rq: the request to be setup
1576 * @rq_src: original request to be cloned
1577 * @bs: bio_set that bios for clone are allocated from
1578 * @gfp_mask: memory allocation mask for bio
1579 * @bio_ctr: setup function to be called for each clone bio.
1580 * Returns %0 for success, non %0 for failure.
1581 * @data: private data to be passed to @bio_ctr
1584 * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq.
1585 * Also, pages which the original bios are pointing to are not copied
1586 * and the cloned bios just point same pages.
1587 * So cloned bios must be completed before original bios, which means
1588 * the caller must complete @rq before @rq_src.
1590 int blk_rq_prep_clone(struct request *rq, struct request *rq_src,
1591 struct bio_set *bs, gfp_t gfp_mask,
1592 int (*bio_ctr)(struct bio *, struct bio *, void *),
1595 struct bio *bio, *bio_src;
1600 __rq_for_each_bio(bio_src, rq_src) {
1601 bio = bio_clone_fast(bio_src, gfp_mask, bs);
1605 if (bio_ctr && bio_ctr(bio, bio_src, data))
1609 rq->biotail->bi_next = bio;
1612 rq->bio = rq->biotail = bio;
1617 /* Copy attributes of the original request to the clone request. */
1618 rq->__sector = blk_rq_pos(rq_src);
1619 rq->__data_len = blk_rq_bytes(rq_src);
1620 if (rq_src->rq_flags & RQF_SPECIAL_PAYLOAD) {
1621 rq->rq_flags |= RQF_SPECIAL_PAYLOAD;
1622 rq->special_vec = rq_src->special_vec;
1624 rq->nr_phys_segments = rq_src->nr_phys_segments;
1625 rq->ioprio = rq_src->ioprio;
1627 if (rq->bio && blk_crypto_rq_bio_prep(rq, rq->bio, gfp_mask) < 0)
1635 blk_rq_unprep_clone(rq);
1639 EXPORT_SYMBOL_GPL(blk_rq_prep_clone);
1641 int kblockd_schedule_work(struct work_struct *work)
1643 return queue_work(kblockd_workqueue, work);
1645 EXPORT_SYMBOL(kblockd_schedule_work);
1647 int kblockd_mod_delayed_work_on(int cpu, struct delayed_work *dwork,
1648 unsigned long delay)
1650 return mod_delayed_work_on(cpu, kblockd_workqueue, dwork, delay);
1652 EXPORT_SYMBOL(kblockd_mod_delayed_work_on);
1655 * blk_start_plug - initialize blk_plug and track it inside the task_struct
1656 * @plug: The &struct blk_plug that needs to be initialized
1659 * blk_start_plug() indicates to the block layer an intent by the caller
1660 * to submit multiple I/O requests in a batch. The block layer may use
1661 * this hint to defer submitting I/Os from the caller until blk_finish_plug()
1662 * is called. However, the block layer may choose to submit requests
1663 * before a call to blk_finish_plug() if the number of queued I/Os
1664 * exceeds %BLK_MAX_REQUEST_COUNT, or if the size of the I/O is larger than
1665 * %BLK_PLUG_FLUSH_SIZE. The queued I/Os may also be submitted early if
1666 * the task schedules (see below).
1668 * Tracking blk_plug inside the task_struct will help with auto-flushing the
1669 * pending I/O should the task end up blocking between blk_start_plug() and
1670 * blk_finish_plug(). This is important from a performance perspective, but
1671 * also ensures that we don't deadlock. For instance, if the task is blocking
1672 * for a memory allocation, memory reclaim could end up wanting to free a
1673 * page belonging to that request that is currently residing in our private
1674 * plug. By flushing the pending I/O when the process goes to sleep, we avoid
1675 * this kind of deadlock.
1677 void blk_start_plug(struct blk_plug *plug)
1679 struct task_struct *tsk = current;
1682 * If this is a nested plug, don't actually assign it.
1687 INIT_LIST_HEAD(&plug->mq_list);
1688 INIT_LIST_HEAD(&plug->cb_list);
1690 plug->multiple_queues = false;
1691 plug->nowait = false;
1694 * Store ordering should not be needed here, since a potential
1695 * preempt will imply a full memory barrier
1699 EXPORT_SYMBOL(blk_start_plug);
1701 static void flush_plug_callbacks(struct blk_plug *plug, bool from_schedule)
1703 LIST_HEAD(callbacks);
1705 while (!list_empty(&plug->cb_list)) {
1706 list_splice_init(&plug->cb_list, &callbacks);
1708 while (!list_empty(&callbacks)) {
1709 struct blk_plug_cb *cb = list_first_entry(&callbacks,
1712 list_del(&cb->list);
1713 cb->callback(cb, from_schedule);
1718 struct blk_plug_cb *blk_check_plugged(blk_plug_cb_fn unplug, void *data,
1721 struct blk_plug *plug = current->plug;
1722 struct blk_plug_cb *cb;
1727 list_for_each_entry(cb, &plug->cb_list, list)
1728 if (cb->callback == unplug && cb->data == data)
1731 /* Not currently on the callback list */
1732 BUG_ON(size < sizeof(*cb));
1733 cb = kzalloc(size, GFP_ATOMIC);
1736 cb->callback = unplug;
1737 list_add(&cb->list, &plug->cb_list);
1741 EXPORT_SYMBOL(blk_check_plugged);
1743 void blk_flush_plug_list(struct blk_plug *plug, bool from_schedule)
1745 flush_plug_callbacks(plug, from_schedule);
1747 if (!list_empty(&plug->mq_list))
1748 blk_mq_flush_plug_list(plug, from_schedule);
1752 * blk_finish_plug - mark the end of a batch of submitted I/O
1753 * @plug: The &struct blk_plug passed to blk_start_plug()
1756 * Indicate that a batch of I/O submissions is complete. This function
1757 * must be paired with an initial call to blk_start_plug(). The intent
1758 * is to allow the block layer to optimize I/O submission. See the
1759 * documentation for blk_start_plug() for more information.
1761 void blk_finish_plug(struct blk_plug *plug)
1763 if (plug != current->plug)
1765 blk_flush_plug_list(plug, false);
1767 current->plug = NULL;
1769 EXPORT_SYMBOL(blk_finish_plug);
1771 void blk_io_schedule(void)
1773 /* Prevent hang_check timer from firing at us during very long I/O */
1774 unsigned long timeout = sysctl_hung_task_timeout_secs * HZ / 2;
1777 io_schedule_timeout(timeout);
1781 EXPORT_SYMBOL_GPL(blk_io_schedule);
1783 int __init blk_dev_init(void)
1785 BUILD_BUG_ON(REQ_OP_LAST >= (1 << REQ_OP_BITS));
1786 BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 *
1787 sizeof_field(struct request, cmd_flags));
1788 BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 *
1789 sizeof_field(struct bio, bi_opf));
1791 /* used for unplugging and affects IO latency/throughput - HIGHPRI */
1792 kblockd_workqueue = alloc_workqueue("kblockd",
1793 WQ_MEM_RECLAIM | WQ_HIGHPRI, 0);
1794 if (!kblockd_workqueue)
1795 panic("Failed to create kblockd\n");
1797 blk_requestq_cachep = kmem_cache_create("request_queue",
1798 sizeof(struct request_queue), 0, SLAB_PANIC, NULL);
1800 blk_debugfs_root = debugfs_create_dir("block", NULL);