1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (C) 1991, 1992 Linus Torvalds
4 * Copyright (C) 1994, Karl Keyte: Added support for disk statistics
5 * Elevator latency, (C) 2000 Andrea Arcangeli <andrea@suse.de> SuSE
6 * Queue request tables / lock, selectable elevator, Jens Axboe <axboe@suse.de>
7 * kernel-doc documentation started by NeilBrown <neilb@cse.unsw.edu.au>
9 * bio rewrite, highmem i/o, etc, Jens Axboe <axboe@suse.de> - may 2001
13 * This handles all read/write requests to block devices
15 #include <linux/kernel.h>
16 #include <linux/module.h>
17 #include <linux/backing-dev.h>
18 #include <linux/bio.h>
19 #include <linux/blkdev.h>
20 #include <linux/blk-mq.h>
21 #include <linux/highmem.h>
23 #include <linux/pagemap.h>
24 #include <linux/kernel_stat.h>
25 #include <linux/string.h>
26 #include <linux/init.h>
27 #include <linux/completion.h>
28 #include <linux/slab.h>
29 #include <linux/swap.h>
30 #include <linux/writeback.h>
31 #include <linux/task_io_accounting_ops.h>
32 #include <linux/fault-inject.h>
33 #include <linux/list_sort.h>
34 #include <linux/delay.h>
35 #include <linux/ratelimit.h>
36 #include <linux/pm_runtime.h>
37 #include <linux/blk-cgroup.h>
38 #include <linux/t10-pi.h>
39 #include <linux/debugfs.h>
40 #include <linux/bpf.h>
41 #include <linux/psi.h>
42 #include <linux/sched/sysctl.h>
43 #include <linux/blk-crypto.h>
45 #define CREATE_TRACE_POINTS
46 #include <trace/events/block.h>
50 #include "blk-mq-sched.h"
52 #include "blk-rq-qos.h"
54 struct dentry *blk_debugfs_root;
56 EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_remap);
57 EXPORT_TRACEPOINT_SYMBOL_GPL(block_rq_remap);
58 EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_complete);
59 EXPORT_TRACEPOINT_SYMBOL_GPL(block_split);
60 EXPORT_TRACEPOINT_SYMBOL_GPL(block_unplug);
62 DEFINE_IDA(blk_queue_ida);
65 * For queue allocation
67 struct kmem_cache *blk_requestq_cachep;
70 * Controlling structure to kblockd
72 static struct workqueue_struct *kblockd_workqueue;
75 * blk_queue_flag_set - atomically set a queue flag
76 * @flag: flag to be set
79 void blk_queue_flag_set(unsigned int flag, struct request_queue *q)
81 set_bit(flag, &q->queue_flags);
83 EXPORT_SYMBOL(blk_queue_flag_set);
86 * blk_queue_flag_clear - atomically clear a queue flag
87 * @flag: flag to be cleared
90 void blk_queue_flag_clear(unsigned int flag, struct request_queue *q)
92 clear_bit(flag, &q->queue_flags);
94 EXPORT_SYMBOL(blk_queue_flag_clear);
97 * blk_queue_flag_test_and_set - atomically test and set a queue flag
98 * @flag: flag to be set
101 * Returns the previous value of @flag - 0 if the flag was not set and 1 if
102 * the flag was already set.
104 bool blk_queue_flag_test_and_set(unsigned int flag, struct request_queue *q)
106 return test_and_set_bit(flag, &q->queue_flags);
108 EXPORT_SYMBOL_GPL(blk_queue_flag_test_and_set);
110 void blk_rq_init(struct request_queue *q, struct request *rq)
112 memset(rq, 0, sizeof(*rq));
114 INIT_LIST_HEAD(&rq->queuelist);
116 rq->__sector = (sector_t) -1;
117 INIT_HLIST_NODE(&rq->hash);
118 RB_CLEAR_NODE(&rq->rb_node);
119 rq->tag = BLK_MQ_NO_TAG;
120 rq->internal_tag = BLK_MQ_NO_TAG;
121 rq->start_time_ns = ktime_get_ns();
123 refcount_set(&rq->ref, 1);
124 blk_crypto_rq_set_defaults(rq);
126 EXPORT_SYMBOL(blk_rq_init);
128 #define REQ_OP_NAME(name) [REQ_OP_##name] = #name
129 static const char *const blk_op_name[] = {
133 REQ_OP_NAME(DISCARD),
134 REQ_OP_NAME(SECURE_ERASE),
135 REQ_OP_NAME(ZONE_RESET),
136 REQ_OP_NAME(ZONE_RESET_ALL),
137 REQ_OP_NAME(ZONE_OPEN),
138 REQ_OP_NAME(ZONE_CLOSE),
139 REQ_OP_NAME(ZONE_FINISH),
140 REQ_OP_NAME(ZONE_APPEND),
141 REQ_OP_NAME(WRITE_SAME),
142 REQ_OP_NAME(WRITE_ZEROES),
143 REQ_OP_NAME(SCSI_IN),
144 REQ_OP_NAME(SCSI_OUT),
146 REQ_OP_NAME(DRV_OUT),
151 * blk_op_str - Return string XXX in the REQ_OP_XXX.
154 * Description: Centralize block layer function to convert REQ_OP_XXX into
155 * string format. Useful in the debugging and tracing bio or request. For
156 * invalid REQ_OP_XXX it returns string "UNKNOWN".
158 inline const char *blk_op_str(unsigned int op)
160 const char *op_str = "UNKNOWN";
162 if (op < ARRAY_SIZE(blk_op_name) && blk_op_name[op])
163 op_str = blk_op_name[op];
167 EXPORT_SYMBOL_GPL(blk_op_str);
169 static const struct {
173 [BLK_STS_OK] = { 0, "" },
174 [BLK_STS_NOTSUPP] = { -EOPNOTSUPP, "operation not supported" },
175 [BLK_STS_TIMEOUT] = { -ETIMEDOUT, "timeout" },
176 [BLK_STS_NOSPC] = { -ENOSPC, "critical space allocation" },
177 [BLK_STS_TRANSPORT] = { -ENOLINK, "recoverable transport" },
178 [BLK_STS_TARGET] = { -EREMOTEIO, "critical target" },
179 [BLK_STS_NEXUS] = { -EBADE, "critical nexus" },
180 [BLK_STS_MEDIUM] = { -ENODATA, "critical medium" },
181 [BLK_STS_PROTECTION] = { -EILSEQ, "protection" },
182 [BLK_STS_RESOURCE] = { -ENOMEM, "kernel resource" },
183 [BLK_STS_DEV_RESOURCE] = { -EBUSY, "device resource" },
184 [BLK_STS_AGAIN] = { -EAGAIN, "nonblocking retry" },
186 /* device mapper special case, should not leak out: */
187 [BLK_STS_DM_REQUEUE] = { -EREMCHG, "dm internal retry" },
189 /* everything else not covered above: */
190 [BLK_STS_IOERR] = { -EIO, "I/O" },
193 blk_status_t errno_to_blk_status(int errno)
197 for (i = 0; i < ARRAY_SIZE(blk_errors); i++) {
198 if (blk_errors[i].errno == errno)
199 return (__force blk_status_t)i;
202 return BLK_STS_IOERR;
204 EXPORT_SYMBOL_GPL(errno_to_blk_status);
206 int blk_status_to_errno(blk_status_t status)
208 int idx = (__force int)status;
210 if (WARN_ON_ONCE(idx >= ARRAY_SIZE(blk_errors)))
212 return blk_errors[idx].errno;
214 EXPORT_SYMBOL_GPL(blk_status_to_errno);
216 static void print_req_error(struct request *req, blk_status_t status,
219 int idx = (__force int)status;
221 if (WARN_ON_ONCE(idx >= ARRAY_SIZE(blk_errors)))
224 printk_ratelimited(KERN_ERR
225 "%s: %s error, dev %s, sector %llu op 0x%x:(%s) flags 0x%x "
226 "phys_seg %u prio class %u\n",
227 caller, blk_errors[idx].name,
228 req->rq_disk ? req->rq_disk->disk_name : "?",
229 blk_rq_pos(req), req_op(req), blk_op_str(req_op(req)),
230 req->cmd_flags & ~REQ_OP_MASK,
231 req->nr_phys_segments,
232 IOPRIO_PRIO_CLASS(req->ioprio));
235 static void req_bio_endio(struct request *rq, struct bio *bio,
236 unsigned int nbytes, blk_status_t error)
239 bio->bi_status = error;
241 if (unlikely(rq->rq_flags & RQF_QUIET))
242 bio_set_flag(bio, BIO_QUIET);
244 bio_advance(bio, nbytes);
246 if (req_op(rq) == REQ_OP_ZONE_APPEND && error == BLK_STS_OK) {
248 * Partial zone append completions cannot be supported as the
249 * BIO fragments may end up not being written sequentially.
251 if (bio->bi_iter.bi_size)
252 bio->bi_status = BLK_STS_IOERR;
254 bio->bi_iter.bi_sector = rq->__sector;
257 /* don't actually finish bio if it's part of flush sequence */
258 if (bio->bi_iter.bi_size == 0 && !(rq->rq_flags & RQF_FLUSH_SEQ))
262 void blk_dump_rq_flags(struct request *rq, char *msg)
264 printk(KERN_INFO "%s: dev %s: flags=%llx\n", msg,
265 rq->rq_disk ? rq->rq_disk->disk_name : "?",
266 (unsigned long long) rq->cmd_flags);
268 printk(KERN_INFO " sector %llu, nr/cnr %u/%u\n",
269 (unsigned long long)blk_rq_pos(rq),
270 blk_rq_sectors(rq), blk_rq_cur_sectors(rq));
271 printk(KERN_INFO " bio %p, biotail %p, len %u\n",
272 rq->bio, rq->biotail, blk_rq_bytes(rq));
274 EXPORT_SYMBOL(blk_dump_rq_flags);
277 * blk_sync_queue - cancel any pending callbacks on a queue
281 * The block layer may perform asynchronous callback activity
282 * on a queue, such as calling the unplug function after a timeout.
283 * A block device may call blk_sync_queue to ensure that any
284 * such activity is cancelled, thus allowing it to release resources
285 * that the callbacks might use. The caller must already have made sure
286 * that its ->submit_bio will not re-add plugging prior to calling
289 * This function does not cancel any asynchronous activity arising
290 * out of elevator or throttling code. That would require elevator_exit()
291 * and blkcg_exit_queue() to be called with queue lock initialized.
294 void blk_sync_queue(struct request_queue *q)
296 del_timer_sync(&q->timeout);
297 cancel_work_sync(&q->timeout_work);
299 EXPORT_SYMBOL(blk_sync_queue);
302 * blk_set_pm_only - increment pm_only counter
303 * @q: request queue pointer
305 void blk_set_pm_only(struct request_queue *q)
307 atomic_inc(&q->pm_only);
309 EXPORT_SYMBOL_GPL(blk_set_pm_only);
311 void blk_clear_pm_only(struct request_queue *q)
315 pm_only = atomic_dec_return(&q->pm_only);
316 WARN_ON_ONCE(pm_only < 0);
318 wake_up_all(&q->mq_freeze_wq);
320 EXPORT_SYMBOL_GPL(blk_clear_pm_only);
323 * blk_put_queue - decrement the request_queue refcount
324 * @q: the request_queue structure to decrement the refcount for
326 * Decrements the refcount of the request_queue kobject. When this reaches 0
327 * we'll have blk_release_queue() called.
329 * Context: Any context, but the last reference must not be dropped from
332 void blk_put_queue(struct request_queue *q)
334 kobject_put(&q->kobj);
336 EXPORT_SYMBOL(blk_put_queue);
338 void blk_set_queue_dying(struct request_queue *q)
340 blk_queue_flag_set(QUEUE_FLAG_DYING, q);
343 * When queue DYING flag is set, we need to block new req
344 * entering queue, so we call blk_freeze_queue_start() to
345 * prevent I/O from crossing blk_queue_enter().
347 blk_freeze_queue_start(q);
350 blk_mq_wake_waiters(q);
352 /* Make blk_queue_enter() reexamine the DYING flag. */
353 wake_up_all(&q->mq_freeze_wq);
355 EXPORT_SYMBOL_GPL(blk_set_queue_dying);
358 * blk_cleanup_queue - shutdown a request queue
359 * @q: request queue to shutdown
361 * Mark @q DYING, drain all pending requests, mark @q DEAD, destroy and
362 * put it. All future requests will be failed immediately with -ENODEV.
366 void blk_cleanup_queue(struct request_queue *q)
368 /* cannot be called from atomic context */
371 WARN_ON_ONCE(blk_queue_registered(q));
373 /* mark @q DYING, no new request or merges will be allowed afterwards */
374 blk_set_queue_dying(q);
376 blk_queue_flag_set(QUEUE_FLAG_NOMERGES, q);
377 blk_queue_flag_set(QUEUE_FLAG_NOXMERGES, q);
380 * Drain all requests queued before DYING marking. Set DEAD flag to
381 * prevent that blk_mq_run_hw_queues() accesses the hardware queues
382 * after draining finished.
388 blk_queue_flag_set(QUEUE_FLAG_DEAD, q);
390 /* for synchronous bio-based driver finish in-flight integrity i/o */
391 blk_flush_integrity();
393 /* @q won't process any more request, flush async actions */
394 del_timer_sync(&q->backing_dev_info->laptop_mode_wb_timer);
398 blk_mq_exit_queue(q);
401 * In theory, request pool of sched_tags belongs to request queue.
402 * However, the current implementation requires tag_set for freeing
403 * requests, so free the pool now.
405 * Queue has become frozen, there can't be any in-queue requests, so
406 * it is safe to free requests now.
408 mutex_lock(&q->sysfs_lock);
410 blk_mq_sched_free_requests(q);
411 mutex_unlock(&q->sysfs_lock);
413 percpu_ref_exit(&q->q_usage_counter);
415 /* @q is and will stay empty, shutdown and put */
418 EXPORT_SYMBOL(blk_cleanup_queue);
421 * blk_queue_enter() - try to increase q->q_usage_counter
422 * @q: request queue pointer
423 * @flags: BLK_MQ_REQ_NOWAIT and/or BLK_MQ_REQ_PREEMPT
425 int blk_queue_enter(struct request_queue *q, blk_mq_req_flags_t flags)
427 const bool pm = flags & BLK_MQ_REQ_PREEMPT;
430 bool success = false;
433 if (percpu_ref_tryget_live(&q->q_usage_counter)) {
435 * The code that increments the pm_only counter is
436 * responsible for ensuring that that counter is
437 * globally visible before the queue is unfrozen.
439 if (pm || !blk_queue_pm_only(q)) {
442 percpu_ref_put(&q->q_usage_counter);
450 if (flags & BLK_MQ_REQ_NOWAIT)
454 * read pair of barrier in blk_freeze_queue_start(),
455 * we need to order reading __PERCPU_REF_DEAD flag of
456 * .q_usage_counter and reading .mq_freeze_depth or
457 * queue dying flag, otherwise the following wait may
458 * never return if the two reads are reordered.
462 wait_event(q->mq_freeze_wq,
463 (!q->mq_freeze_depth &&
464 (pm || (blk_pm_request_resume(q),
465 !blk_queue_pm_only(q)))) ||
467 if (blk_queue_dying(q))
472 static inline int bio_queue_enter(struct bio *bio)
474 struct request_queue *q = bio->bi_disk->queue;
475 bool nowait = bio->bi_opf & REQ_NOWAIT;
478 ret = blk_queue_enter(q, nowait ? BLK_MQ_REQ_NOWAIT : 0);
480 if (nowait && !blk_queue_dying(q))
481 bio_wouldblock_error(bio);
489 void blk_queue_exit(struct request_queue *q)
491 percpu_ref_put(&q->q_usage_counter);
494 static void blk_queue_usage_counter_release(struct percpu_ref *ref)
496 struct request_queue *q =
497 container_of(ref, struct request_queue, q_usage_counter);
499 wake_up_all(&q->mq_freeze_wq);
502 static void blk_rq_timed_out_timer(struct timer_list *t)
504 struct request_queue *q = from_timer(q, t, timeout);
506 kblockd_schedule_work(&q->timeout_work);
509 static void blk_timeout_work(struct work_struct *work)
513 struct request_queue *blk_alloc_queue(int node_id)
515 struct request_queue *q;
518 q = kmem_cache_alloc_node(blk_requestq_cachep,
519 GFP_KERNEL | __GFP_ZERO, node_id);
523 q->last_merge = NULL;
525 q->id = ida_simple_get(&blk_queue_ida, 0, 0, GFP_KERNEL);
529 ret = bioset_init(&q->bio_split, BIO_POOL_SIZE, 0, BIOSET_NEED_BVECS);
533 q->backing_dev_info = bdi_alloc(node_id);
534 if (!q->backing_dev_info)
537 q->stats = blk_alloc_queue_stats();
543 atomic_set(&q->nr_active_requests_shared_sbitmap, 0);
545 timer_setup(&q->backing_dev_info->laptop_mode_wb_timer,
546 laptop_mode_timer_fn, 0);
547 timer_setup(&q->timeout, blk_rq_timed_out_timer, 0);
548 INIT_WORK(&q->timeout_work, blk_timeout_work);
549 INIT_LIST_HEAD(&q->icq_list);
550 #ifdef CONFIG_BLK_CGROUP
551 INIT_LIST_HEAD(&q->blkg_list);
554 kobject_init(&q->kobj, &blk_queue_ktype);
556 mutex_init(&q->debugfs_mutex);
557 mutex_init(&q->sysfs_lock);
558 mutex_init(&q->sysfs_dir_lock);
559 spin_lock_init(&q->queue_lock);
561 init_waitqueue_head(&q->mq_freeze_wq);
562 mutex_init(&q->mq_freeze_lock);
565 * Init percpu_ref in atomic mode so that it's faster to shutdown.
566 * See blk_register_queue() for details.
568 if (percpu_ref_init(&q->q_usage_counter,
569 blk_queue_usage_counter_release,
570 PERCPU_REF_INIT_ATOMIC, GFP_KERNEL))
573 if (blkcg_init_queue(q))
576 blk_queue_dma_alignment(q, 511);
577 blk_set_default_limits(&q->limits);
578 q->nr_requests = BLKDEV_MAX_RQ;
583 percpu_ref_exit(&q->q_usage_counter);
585 blk_free_queue_stats(q->stats);
587 bdi_put(q->backing_dev_info);
589 bioset_exit(&q->bio_split);
591 ida_simple_remove(&blk_queue_ida, q->id);
593 kmem_cache_free(blk_requestq_cachep, q);
596 EXPORT_SYMBOL(blk_alloc_queue);
599 * blk_get_queue - increment the request_queue refcount
600 * @q: the request_queue structure to increment the refcount for
602 * Increment the refcount of the request_queue kobject.
604 * Context: Any context.
606 bool blk_get_queue(struct request_queue *q)
608 if (likely(!blk_queue_dying(q))) {
615 EXPORT_SYMBOL(blk_get_queue);
618 * blk_get_request - allocate a request
619 * @q: request queue to allocate a request for
620 * @op: operation (REQ_OP_*) and REQ_* flags, e.g. REQ_SYNC.
621 * @flags: BLK_MQ_REQ_* flags, e.g. BLK_MQ_REQ_NOWAIT.
623 struct request *blk_get_request(struct request_queue *q, unsigned int op,
624 blk_mq_req_flags_t flags)
628 WARN_ON_ONCE(op & REQ_NOWAIT);
629 WARN_ON_ONCE(flags & ~(BLK_MQ_REQ_NOWAIT | BLK_MQ_REQ_PREEMPT));
631 req = blk_mq_alloc_request(q, op, flags);
632 if (!IS_ERR(req) && q->mq_ops->initialize_rq_fn)
633 q->mq_ops->initialize_rq_fn(req);
637 EXPORT_SYMBOL(blk_get_request);
639 void blk_put_request(struct request *req)
641 blk_mq_free_request(req);
643 EXPORT_SYMBOL(blk_put_request);
645 static void handle_bad_sector(struct bio *bio, sector_t maxsector)
647 char b[BDEVNAME_SIZE];
649 printk(KERN_INFO "attempt to access beyond end of device\n");
650 printk(KERN_INFO "%s: rw=%d, want=%Lu, limit=%Lu\n",
651 bio_devname(bio, b), bio->bi_opf,
652 (unsigned long long)bio_end_sector(bio),
653 (long long)maxsector);
656 #ifdef CONFIG_FAIL_MAKE_REQUEST
658 static DECLARE_FAULT_ATTR(fail_make_request);
660 static int __init setup_fail_make_request(char *str)
662 return setup_fault_attr(&fail_make_request, str);
664 __setup("fail_make_request=", setup_fail_make_request);
666 static bool should_fail_request(struct hd_struct *part, unsigned int bytes)
668 return part->make_it_fail && should_fail(&fail_make_request, bytes);
671 static int __init fail_make_request_debugfs(void)
673 struct dentry *dir = fault_create_debugfs_attr("fail_make_request",
674 NULL, &fail_make_request);
676 return PTR_ERR_OR_ZERO(dir);
679 late_initcall(fail_make_request_debugfs);
681 #else /* CONFIG_FAIL_MAKE_REQUEST */
683 static inline bool should_fail_request(struct hd_struct *part,
689 #endif /* CONFIG_FAIL_MAKE_REQUEST */
691 static inline bool bio_check_ro(struct bio *bio, struct hd_struct *part)
693 const int op = bio_op(bio);
695 if (part->policy && op_is_write(op)) {
696 char b[BDEVNAME_SIZE];
698 if (op_is_flush(bio->bi_opf) && !bio_sectors(bio))
702 "Trying to write to read-only block-device %s (partno %d)\n",
703 bio_devname(bio, b), part->partno);
704 /* Older lvm-tools actually trigger this */
711 static noinline int should_fail_bio(struct bio *bio)
713 if (should_fail_request(&bio->bi_disk->part0, bio->bi_iter.bi_size))
717 ALLOW_ERROR_INJECTION(should_fail_bio, ERRNO);
720 * Check whether this bio extends beyond the end of the device or partition.
721 * This may well happen - the kernel calls bread() without checking the size of
722 * the device, e.g., when mounting a file system.
724 static inline int bio_check_eod(struct bio *bio, sector_t maxsector)
726 unsigned int nr_sectors = bio_sectors(bio);
728 if (nr_sectors && maxsector &&
729 (nr_sectors > maxsector ||
730 bio->bi_iter.bi_sector > maxsector - nr_sectors)) {
731 handle_bad_sector(bio, maxsector);
738 * Remap block n of partition p to block n+start(p) of the disk.
740 static inline int blk_partition_remap(struct bio *bio)
746 p = __disk_get_part(bio->bi_disk, bio->bi_partno);
749 if (unlikely(should_fail_request(p, bio->bi_iter.bi_size)))
751 if (unlikely(bio_check_ro(bio, p)))
754 if (bio_sectors(bio)) {
755 if (bio_check_eod(bio, part_nr_sects_read(p)))
757 bio->bi_iter.bi_sector += p->start_sect;
758 trace_block_bio_remap(bio->bi_disk->queue, bio, part_devt(p),
759 bio->bi_iter.bi_sector - p->start_sect);
769 * Check write append to a zoned block device.
771 static inline blk_status_t blk_check_zone_append(struct request_queue *q,
774 sector_t pos = bio->bi_iter.bi_sector;
775 int nr_sectors = bio_sectors(bio);
777 /* Only applicable to zoned block devices */
778 if (!blk_queue_is_zoned(q))
779 return BLK_STS_NOTSUPP;
781 /* The bio sector must point to the start of a sequential zone */
782 if (pos & (blk_queue_zone_sectors(q) - 1) ||
783 !blk_queue_zone_is_seq(q, pos))
784 return BLK_STS_IOERR;
787 * Not allowed to cross zone boundaries. Otherwise, the BIO will be
788 * split and could result in non-contiguous sectors being written in
791 if (nr_sectors > q->limits.chunk_sectors)
792 return BLK_STS_IOERR;
794 /* Make sure the BIO is small enough and will not get split */
795 if (nr_sectors > q->limits.max_zone_append_sectors)
796 return BLK_STS_IOERR;
798 bio->bi_opf |= REQ_NOMERGE;
803 static noinline_for_stack bool submit_bio_checks(struct bio *bio)
805 struct request_queue *q = bio->bi_disk->queue;
806 blk_status_t status = BLK_STS_IOERR;
807 struct blk_plug *plug;
811 plug = blk_mq_plug(q, bio);
812 if (plug && plug->nowait)
813 bio->bi_opf |= REQ_NOWAIT;
816 * For a REQ_NOWAIT based request, return -EOPNOTSUPP
817 * if queue does not support NOWAIT.
819 if ((bio->bi_opf & REQ_NOWAIT) && !blk_queue_nowait(q))
822 if (should_fail_bio(bio))
825 if (bio->bi_partno) {
826 if (unlikely(blk_partition_remap(bio)))
829 if (unlikely(bio_check_ro(bio, &bio->bi_disk->part0)))
831 if (unlikely(bio_check_eod(bio, get_capacity(bio->bi_disk))))
836 * Filter flush bio's early so that bio based drivers without flush
837 * support don't have to worry about them.
839 if (op_is_flush(bio->bi_opf) &&
840 !test_bit(QUEUE_FLAG_WC, &q->queue_flags)) {
841 bio->bi_opf &= ~(REQ_PREFLUSH | REQ_FUA);
842 if (!bio_sectors(bio)) {
848 if (!test_bit(QUEUE_FLAG_POLL, &q->queue_flags))
849 bio->bi_opf &= ~REQ_HIPRI;
851 switch (bio_op(bio)) {
853 if (!blk_queue_discard(q))
856 case REQ_OP_SECURE_ERASE:
857 if (!blk_queue_secure_erase(q))
860 case REQ_OP_WRITE_SAME:
861 if (!q->limits.max_write_same_sectors)
864 case REQ_OP_ZONE_APPEND:
865 status = blk_check_zone_append(q, bio);
866 if (status != BLK_STS_OK)
869 case REQ_OP_ZONE_RESET:
870 case REQ_OP_ZONE_OPEN:
871 case REQ_OP_ZONE_CLOSE:
872 case REQ_OP_ZONE_FINISH:
873 if (!blk_queue_is_zoned(q))
876 case REQ_OP_ZONE_RESET_ALL:
877 if (!blk_queue_is_zoned(q) || !blk_queue_zone_resetall(q))
880 case REQ_OP_WRITE_ZEROES:
881 if (!q->limits.max_write_zeroes_sectors)
889 * Various block parts want %current->io_context, so allocate it up
890 * front rather than dealing with lots of pain to allocate it only
891 * where needed. This may fail and the block layer knows how to live
894 if (unlikely(!current->io_context))
895 create_task_io_context(current, GFP_ATOMIC, q->node);
897 if (blk_throtl_bio(bio)) {
898 blkcg_bio_issue_init(bio);
902 blk_cgroup_bio_start(bio);
903 blkcg_bio_issue_init(bio);
905 if (!bio_flagged(bio, BIO_TRACE_COMPLETION)) {
906 trace_block_bio_queue(q, bio);
907 /* Now that enqueuing has been traced, we need to trace
908 * completion as well.
910 bio_set_flag(bio, BIO_TRACE_COMPLETION);
915 status = BLK_STS_NOTSUPP;
917 bio->bi_status = status;
922 static blk_qc_t __submit_bio(struct bio *bio)
924 struct gendisk *disk = bio->bi_disk;
925 blk_qc_t ret = BLK_QC_T_NONE;
927 if (blk_crypto_bio_prep(&bio)) {
928 if (!disk->fops->submit_bio)
929 return blk_mq_submit_bio(bio);
930 ret = disk->fops->submit_bio(bio);
932 blk_queue_exit(disk->queue);
937 * The loop in this function may be a bit non-obvious, and so deserves some
940 * - Before entering the loop, bio->bi_next is NULL (as all callers ensure
941 * that), so we have a list with a single bio.
942 * - We pretend that we have just taken it off a longer list, so we assign
943 * bio_list to a pointer to the bio_list_on_stack, thus initialising the
944 * bio_list of new bios to be added. ->submit_bio() may indeed add some more
945 * bios through a recursive call to submit_bio_noacct. If it did, we find a
946 * non-NULL value in bio_list and re-enter the loop from the top.
947 * - In this case we really did just take the bio of the top of the list (no
948 * pretending) and so remove it from bio_list, and call into ->submit_bio()
951 * bio_list_on_stack[0] contains bios submitted by the current ->submit_bio.
952 * bio_list_on_stack[1] contains bios that were submitted before the current
953 * ->submit_bio_bio, but that haven't been processed yet.
955 static blk_qc_t __submit_bio_noacct(struct bio *bio)
957 struct bio_list bio_list_on_stack[2];
958 blk_qc_t ret = BLK_QC_T_NONE;
960 BUG_ON(bio->bi_next);
962 bio_list_init(&bio_list_on_stack[0]);
963 current->bio_list = bio_list_on_stack;
966 struct request_queue *q = bio->bi_disk->queue;
967 struct bio_list lower, same;
969 if (unlikely(bio_queue_enter(bio) != 0))
973 * Create a fresh bio_list for all subordinate requests.
975 bio_list_on_stack[1] = bio_list_on_stack[0];
976 bio_list_init(&bio_list_on_stack[0]);
978 ret = __submit_bio(bio);
981 * Sort new bios into those for a lower level and those for the
984 bio_list_init(&lower);
985 bio_list_init(&same);
986 while ((bio = bio_list_pop(&bio_list_on_stack[0])) != NULL)
987 if (q == bio->bi_disk->queue)
988 bio_list_add(&same, bio);
990 bio_list_add(&lower, bio);
993 * Now assemble so we handle the lowest level first.
995 bio_list_merge(&bio_list_on_stack[0], &lower);
996 bio_list_merge(&bio_list_on_stack[0], &same);
997 bio_list_merge(&bio_list_on_stack[0], &bio_list_on_stack[1]);
998 } while ((bio = bio_list_pop(&bio_list_on_stack[0])));
1000 current->bio_list = NULL;
1004 static blk_qc_t __submit_bio_noacct_mq(struct bio *bio)
1006 struct bio_list bio_list[2] = { };
1007 blk_qc_t ret = BLK_QC_T_NONE;
1009 current->bio_list = bio_list;
1012 struct gendisk *disk = bio->bi_disk;
1014 if (unlikely(bio_queue_enter(bio) != 0))
1017 if (!blk_crypto_bio_prep(&bio)) {
1018 blk_queue_exit(disk->queue);
1019 ret = BLK_QC_T_NONE;
1023 ret = blk_mq_submit_bio(bio);
1024 } while ((bio = bio_list_pop(&bio_list[0])));
1026 current->bio_list = NULL;
1031 * submit_bio_noacct - re-submit a bio to the block device layer for I/O
1032 * @bio: The bio describing the location in memory and on the device.
1034 * This is a version of submit_bio() that shall only be used for I/O that is
1035 * resubmitted to lower level drivers by stacking block drivers. All file
1036 * systems and other upper level users of the block layer should use
1037 * submit_bio() instead.
1039 blk_qc_t submit_bio_noacct(struct bio *bio)
1041 if (!submit_bio_checks(bio))
1042 return BLK_QC_T_NONE;
1045 * We only want one ->submit_bio to be active at a time, else stack
1046 * usage with stacked devices could be a problem. Use current->bio_list
1047 * to collect a list of requests submited by a ->submit_bio method while
1048 * it is active, and then process them after it returned.
1050 if (current->bio_list) {
1051 bio_list_add(¤t->bio_list[0], bio);
1052 return BLK_QC_T_NONE;
1055 if (!bio->bi_disk->fops->submit_bio)
1056 return __submit_bio_noacct_mq(bio);
1057 return __submit_bio_noacct(bio);
1059 EXPORT_SYMBOL(submit_bio_noacct);
1062 * submit_bio - submit a bio to the block device layer for I/O
1063 * @bio: The &struct bio which describes the I/O
1065 * submit_bio() is used to submit I/O requests to block devices. It is passed a
1066 * fully set up &struct bio that describes the I/O that needs to be done. The
1067 * bio will be send to the device described by the bi_disk and bi_partno fields.
1069 * The success/failure status of the request, along with notification of
1070 * completion, is delivered asynchronously through the ->bi_end_io() callback
1071 * in @bio. The bio must NOT be touched by thecaller until ->bi_end_io() has
1074 blk_qc_t submit_bio(struct bio *bio)
1076 if (blkcg_punt_bio_submit(bio))
1077 return BLK_QC_T_NONE;
1080 * If it's a regular read/write or a barrier with data attached,
1081 * go through the normal accounting stuff before submission.
1083 if (bio_has_data(bio)) {
1086 if (unlikely(bio_op(bio) == REQ_OP_WRITE_SAME))
1087 count = queue_logical_block_size(bio->bi_disk->queue) >> 9;
1089 count = bio_sectors(bio);
1091 if (op_is_write(bio_op(bio))) {
1092 count_vm_events(PGPGOUT, count);
1094 task_io_account_read(bio->bi_iter.bi_size);
1095 count_vm_events(PGPGIN, count);
1098 if (unlikely(block_dump)) {
1099 char b[BDEVNAME_SIZE];
1100 printk(KERN_DEBUG "%s(%d): %s block %Lu on %s (%u sectors)\n",
1101 current->comm, task_pid_nr(current),
1102 op_is_write(bio_op(bio)) ? "WRITE" : "READ",
1103 (unsigned long long)bio->bi_iter.bi_sector,
1104 bio_devname(bio, b), count);
1109 * If we're reading data that is part of the userspace workingset, count
1110 * submission time as memory stall. When the device is congested, or
1111 * the submitting cgroup IO-throttled, submission can be a significant
1112 * part of overall IO time.
1114 if (unlikely(bio_op(bio) == REQ_OP_READ &&
1115 bio_flagged(bio, BIO_WORKINGSET))) {
1116 unsigned long pflags;
1119 psi_memstall_enter(&pflags);
1120 ret = submit_bio_noacct(bio);
1121 psi_memstall_leave(&pflags);
1126 return submit_bio_noacct(bio);
1128 EXPORT_SYMBOL(submit_bio);
1131 * blk_cloned_rq_check_limits - Helper function to check a cloned request
1132 * for the new queue limits
1134 * @rq: the request being checked
1137 * @rq may have been made based on weaker limitations of upper-level queues
1138 * in request stacking drivers, and it may violate the limitation of @q.
1139 * Since the block layer and the underlying device driver trust @rq
1140 * after it is inserted to @q, it should be checked against @q before
1141 * the insertion using this generic function.
1143 * Request stacking drivers like request-based dm may change the queue
1144 * limits when retrying requests on other queues. Those requests need
1145 * to be checked against the new queue limits again during dispatch.
1147 static blk_status_t blk_cloned_rq_check_limits(struct request_queue *q,
1150 unsigned int max_sectors = blk_queue_get_max_sectors(q, req_op(rq));
1152 if (blk_rq_sectors(rq) > max_sectors) {
1154 * SCSI device does not have a good way to return if
1155 * Write Same/Zero is actually supported. If a device rejects
1156 * a non-read/write command (discard, write same,etc.) the
1157 * low-level device driver will set the relevant queue limit to
1158 * 0 to prevent blk-lib from issuing more of the offending
1159 * operations. Commands queued prior to the queue limit being
1160 * reset need to be completed with BLK_STS_NOTSUPP to avoid I/O
1161 * errors being propagated to upper layers.
1163 if (max_sectors == 0)
1164 return BLK_STS_NOTSUPP;
1166 printk(KERN_ERR "%s: over max size limit. (%u > %u)\n",
1167 __func__, blk_rq_sectors(rq), max_sectors);
1168 return BLK_STS_IOERR;
1172 * queue's settings related to segment counting like q->bounce_pfn
1173 * may differ from that of other stacking queues.
1174 * Recalculate it to check the request correctly on this queue's
1177 rq->nr_phys_segments = blk_recalc_rq_segments(rq);
1178 if (rq->nr_phys_segments > queue_max_segments(q)) {
1179 printk(KERN_ERR "%s: over max segments limit. (%hu > %hu)\n",
1180 __func__, rq->nr_phys_segments, queue_max_segments(q));
1181 return BLK_STS_IOERR;
1188 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
1189 * @q: the queue to submit the request
1190 * @rq: the request being queued
1192 blk_status_t blk_insert_cloned_request(struct request_queue *q, struct request *rq)
1196 ret = blk_cloned_rq_check_limits(q, rq);
1197 if (ret != BLK_STS_OK)
1201 should_fail_request(&rq->rq_disk->part0, blk_rq_bytes(rq)))
1202 return BLK_STS_IOERR;
1204 if (blk_crypto_insert_cloned_request(rq))
1205 return BLK_STS_IOERR;
1207 if (blk_queue_io_stat(q))
1208 blk_account_io_start(rq);
1211 * Since we have a scheduler attached on the top device,
1212 * bypass a potential scheduler on the bottom device for
1215 return blk_mq_request_issue_directly(rq, true);
1217 EXPORT_SYMBOL_GPL(blk_insert_cloned_request);
1220 * blk_rq_err_bytes - determine number of bytes till the next failure boundary
1221 * @rq: request to examine
1224 * A request could be merge of IOs which require different failure
1225 * handling. This function determines the number of bytes which
1226 * can be failed from the beginning of the request without
1227 * crossing into area which need to be retried further.
1230 * The number of bytes to fail.
1232 unsigned int blk_rq_err_bytes(const struct request *rq)
1234 unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK;
1235 unsigned int bytes = 0;
1238 if (!(rq->rq_flags & RQF_MIXED_MERGE))
1239 return blk_rq_bytes(rq);
1242 * Currently the only 'mixing' which can happen is between
1243 * different fastfail types. We can safely fail portions
1244 * which have all the failfast bits that the first one has -
1245 * the ones which are at least as eager to fail as the first
1248 for (bio = rq->bio; bio; bio = bio->bi_next) {
1249 if ((bio->bi_opf & ff) != ff)
1251 bytes += bio->bi_iter.bi_size;
1254 /* this could lead to infinite loop */
1255 BUG_ON(blk_rq_bytes(rq) && !bytes);
1258 EXPORT_SYMBOL_GPL(blk_rq_err_bytes);
1260 static void update_io_ticks(struct hd_struct *part, unsigned long now, bool end)
1262 unsigned long stamp;
1264 stamp = READ_ONCE(part->stamp);
1265 if (unlikely(stamp != now)) {
1266 if (likely(cmpxchg(&part->stamp, stamp, now) == stamp))
1267 __part_stat_add(part, io_ticks, end ? now - stamp : 1);
1270 part = &part_to_disk(part)->part0;
1275 static void blk_account_io_completion(struct request *req, unsigned int bytes)
1277 if (req->part && blk_do_io_stat(req)) {
1278 const int sgrp = op_stat_group(req_op(req));
1279 struct hd_struct *part;
1283 part_stat_add(part, sectors[sgrp], bytes >> 9);
1288 void blk_account_io_done(struct request *req, u64 now)
1291 * Account IO completion. flush_rq isn't accounted as a
1292 * normal IO on queueing nor completion. Accounting the
1293 * containing request is enough.
1295 if (req->part && blk_do_io_stat(req) &&
1296 !(req->rq_flags & RQF_FLUSH_SEQ)) {
1297 const int sgrp = op_stat_group(req_op(req));
1298 struct hd_struct *part;
1303 update_io_ticks(part, jiffies, true);
1304 part_stat_inc(part, ios[sgrp]);
1305 part_stat_add(part, nsecs[sgrp], now - req->start_time_ns);
1308 hd_struct_put(part);
1312 void blk_account_io_start(struct request *rq)
1314 if (!blk_do_io_stat(rq))
1317 rq->part = disk_map_sector_rcu(rq->rq_disk, blk_rq_pos(rq));
1320 update_io_ticks(rq->part, jiffies, false);
1324 static unsigned long __part_start_io_acct(struct hd_struct *part,
1325 unsigned int sectors, unsigned int op)
1327 const int sgrp = op_stat_group(op);
1328 unsigned long now = READ_ONCE(jiffies);
1331 update_io_ticks(part, now, false);
1332 part_stat_inc(part, ios[sgrp]);
1333 part_stat_add(part, sectors[sgrp], sectors);
1334 part_stat_local_inc(part, in_flight[op_is_write(op)]);
1340 unsigned long part_start_io_acct(struct gendisk *disk, struct hd_struct **part,
1343 *part = disk_map_sector_rcu(disk, bio->bi_iter.bi_sector);
1345 return __part_start_io_acct(*part, bio_sectors(bio), bio_op(bio));
1347 EXPORT_SYMBOL_GPL(part_start_io_acct);
1349 unsigned long disk_start_io_acct(struct gendisk *disk, unsigned int sectors,
1352 return __part_start_io_acct(&disk->part0, sectors, op);
1354 EXPORT_SYMBOL(disk_start_io_acct);
1356 static void __part_end_io_acct(struct hd_struct *part, unsigned int op,
1357 unsigned long start_time)
1359 const int sgrp = op_stat_group(op);
1360 unsigned long now = READ_ONCE(jiffies);
1361 unsigned long duration = now - start_time;
1364 update_io_ticks(part, now, true);
1365 part_stat_add(part, nsecs[sgrp], jiffies_to_nsecs(duration));
1366 part_stat_local_dec(part, in_flight[op_is_write(op)]);
1370 void part_end_io_acct(struct hd_struct *part, struct bio *bio,
1371 unsigned long start_time)
1373 __part_end_io_acct(part, bio_op(bio), start_time);
1374 hd_struct_put(part);
1376 EXPORT_SYMBOL_GPL(part_end_io_acct);
1378 void disk_end_io_acct(struct gendisk *disk, unsigned int op,
1379 unsigned long start_time)
1381 __part_end_io_acct(&disk->part0, op, start_time);
1383 EXPORT_SYMBOL(disk_end_io_acct);
1386 * Steal bios from a request and add them to a bio list.
1387 * The request must not have been partially completed before.
1389 void blk_steal_bios(struct bio_list *list, struct request *rq)
1393 list->tail->bi_next = rq->bio;
1395 list->head = rq->bio;
1396 list->tail = rq->biotail;
1404 EXPORT_SYMBOL_GPL(blk_steal_bios);
1407 * blk_update_request - Special helper function for request stacking drivers
1408 * @req: the request being processed
1409 * @error: block status code
1410 * @nr_bytes: number of bytes to complete @req
1413 * Ends I/O on a number of bytes attached to @req, but doesn't complete
1414 * the request structure even if @req doesn't have leftover.
1415 * If @req has leftover, sets it up for the next range of segments.
1417 * This special helper function is only for request stacking drivers
1418 * (e.g. request-based dm) so that they can handle partial completion.
1419 * Actual device drivers should use blk_mq_end_request instead.
1421 * Passing the result of blk_rq_bytes() as @nr_bytes guarantees
1422 * %false return from this function.
1425 * The RQF_SPECIAL_PAYLOAD flag is ignored on purpose in both
1426 * blk_rq_bytes() and in blk_update_request().
1429 * %false - this request doesn't have any more data
1430 * %true - this request has more data
1432 bool blk_update_request(struct request *req, blk_status_t error,
1433 unsigned int nr_bytes)
1437 trace_block_rq_complete(req, blk_status_to_errno(error), nr_bytes);
1442 #ifdef CONFIG_BLK_DEV_INTEGRITY
1443 if (blk_integrity_rq(req) && req_op(req) == REQ_OP_READ &&
1444 error == BLK_STS_OK)
1445 req->q->integrity.profile->complete_fn(req, nr_bytes);
1448 if (unlikely(error && !blk_rq_is_passthrough(req) &&
1449 !(req->rq_flags & RQF_QUIET)))
1450 print_req_error(req, error, __func__);
1452 blk_account_io_completion(req, nr_bytes);
1456 struct bio *bio = req->bio;
1457 unsigned bio_bytes = min(bio->bi_iter.bi_size, nr_bytes);
1459 if (bio_bytes == bio->bi_iter.bi_size)
1460 req->bio = bio->bi_next;
1462 /* Completion has already been traced */
1463 bio_clear_flag(bio, BIO_TRACE_COMPLETION);
1464 req_bio_endio(req, bio, bio_bytes, error);
1466 total_bytes += bio_bytes;
1467 nr_bytes -= bio_bytes;
1478 * Reset counters so that the request stacking driver
1479 * can find how many bytes remain in the request
1482 req->__data_len = 0;
1486 req->__data_len -= total_bytes;
1488 /* update sector only for requests with clear definition of sector */
1489 if (!blk_rq_is_passthrough(req))
1490 req->__sector += total_bytes >> 9;
1492 /* mixed attributes always follow the first bio */
1493 if (req->rq_flags & RQF_MIXED_MERGE) {
1494 req->cmd_flags &= ~REQ_FAILFAST_MASK;
1495 req->cmd_flags |= req->bio->bi_opf & REQ_FAILFAST_MASK;
1498 if (!(req->rq_flags & RQF_SPECIAL_PAYLOAD)) {
1500 * If total number of sectors is less than the first segment
1501 * size, something has gone terribly wrong.
1503 if (blk_rq_bytes(req) < blk_rq_cur_bytes(req)) {
1504 blk_dump_rq_flags(req, "request botched");
1505 req->__data_len = blk_rq_cur_bytes(req);
1508 /* recalculate the number of segments */
1509 req->nr_phys_segments = blk_recalc_rq_segments(req);
1514 EXPORT_SYMBOL_GPL(blk_update_request);
1516 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
1518 * rq_flush_dcache_pages - Helper function to flush all pages in a request
1519 * @rq: the request to be flushed
1522 * Flush all pages in @rq.
1524 void rq_flush_dcache_pages(struct request *rq)
1526 struct req_iterator iter;
1527 struct bio_vec bvec;
1529 rq_for_each_segment(bvec, rq, iter)
1530 flush_dcache_page(bvec.bv_page);
1532 EXPORT_SYMBOL_GPL(rq_flush_dcache_pages);
1536 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
1537 * @q : the queue of the device being checked
1540 * Check if underlying low-level drivers of a device are busy.
1541 * If the drivers want to export their busy state, they must set own
1542 * exporting function using blk_queue_lld_busy() first.
1544 * Basically, this function is used only by request stacking drivers
1545 * to stop dispatching requests to underlying devices when underlying
1546 * devices are busy. This behavior helps more I/O merging on the queue
1547 * of the request stacking driver and prevents I/O throughput regression
1548 * on burst I/O load.
1551 * 0 - Not busy (The request stacking driver should dispatch request)
1552 * 1 - Busy (The request stacking driver should stop dispatching request)
1554 int blk_lld_busy(struct request_queue *q)
1556 if (queue_is_mq(q) && q->mq_ops->busy)
1557 return q->mq_ops->busy(q);
1561 EXPORT_SYMBOL_GPL(blk_lld_busy);
1564 * blk_rq_unprep_clone - Helper function to free all bios in a cloned request
1565 * @rq: the clone request to be cleaned up
1568 * Free all bios in @rq for a cloned request.
1570 void blk_rq_unprep_clone(struct request *rq)
1574 while ((bio = rq->bio) != NULL) {
1575 rq->bio = bio->bi_next;
1580 EXPORT_SYMBOL_GPL(blk_rq_unprep_clone);
1583 * blk_rq_prep_clone - Helper function to setup clone request
1584 * @rq: the request to be setup
1585 * @rq_src: original request to be cloned
1586 * @bs: bio_set that bios for clone are allocated from
1587 * @gfp_mask: memory allocation mask for bio
1588 * @bio_ctr: setup function to be called for each clone bio.
1589 * Returns %0 for success, non %0 for failure.
1590 * @data: private data to be passed to @bio_ctr
1593 * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq.
1594 * Also, pages which the original bios are pointing to are not copied
1595 * and the cloned bios just point same pages.
1596 * So cloned bios must be completed before original bios, which means
1597 * the caller must complete @rq before @rq_src.
1599 int blk_rq_prep_clone(struct request *rq, struct request *rq_src,
1600 struct bio_set *bs, gfp_t gfp_mask,
1601 int (*bio_ctr)(struct bio *, struct bio *, void *),
1604 struct bio *bio, *bio_src;
1609 __rq_for_each_bio(bio_src, rq_src) {
1610 bio = bio_clone_fast(bio_src, gfp_mask, bs);
1614 if (bio_ctr && bio_ctr(bio, bio_src, data))
1618 rq->biotail->bi_next = bio;
1621 rq->bio = rq->biotail = bio;
1624 /* Copy attributes of the original request to the clone request. */
1625 rq->__sector = blk_rq_pos(rq_src);
1626 rq->__data_len = blk_rq_bytes(rq_src);
1627 if (rq_src->rq_flags & RQF_SPECIAL_PAYLOAD) {
1628 rq->rq_flags |= RQF_SPECIAL_PAYLOAD;
1629 rq->special_vec = rq_src->special_vec;
1631 rq->nr_phys_segments = rq_src->nr_phys_segments;
1632 rq->ioprio = rq_src->ioprio;
1635 blk_crypto_rq_bio_prep(rq, rq->bio, gfp_mask);
1642 blk_rq_unprep_clone(rq);
1646 EXPORT_SYMBOL_GPL(blk_rq_prep_clone);
1648 int kblockd_schedule_work(struct work_struct *work)
1650 return queue_work(kblockd_workqueue, work);
1652 EXPORT_SYMBOL(kblockd_schedule_work);
1654 int kblockd_mod_delayed_work_on(int cpu, struct delayed_work *dwork,
1655 unsigned long delay)
1657 return mod_delayed_work_on(cpu, kblockd_workqueue, dwork, delay);
1659 EXPORT_SYMBOL(kblockd_mod_delayed_work_on);
1662 * blk_start_plug - initialize blk_plug and track it inside the task_struct
1663 * @plug: The &struct blk_plug that needs to be initialized
1666 * blk_start_plug() indicates to the block layer an intent by the caller
1667 * to submit multiple I/O requests in a batch. The block layer may use
1668 * this hint to defer submitting I/Os from the caller until blk_finish_plug()
1669 * is called. However, the block layer may choose to submit requests
1670 * before a call to blk_finish_plug() if the number of queued I/Os
1671 * exceeds %BLK_MAX_REQUEST_COUNT, or if the size of the I/O is larger than
1672 * %BLK_PLUG_FLUSH_SIZE. The queued I/Os may also be submitted early if
1673 * the task schedules (see below).
1675 * Tracking blk_plug inside the task_struct will help with auto-flushing the
1676 * pending I/O should the task end up blocking between blk_start_plug() and
1677 * blk_finish_plug(). This is important from a performance perspective, but
1678 * also ensures that we don't deadlock. For instance, if the task is blocking
1679 * for a memory allocation, memory reclaim could end up wanting to free a
1680 * page belonging to that request that is currently residing in our private
1681 * plug. By flushing the pending I/O when the process goes to sleep, we avoid
1682 * this kind of deadlock.
1684 void blk_start_plug(struct blk_plug *plug)
1686 struct task_struct *tsk = current;
1689 * If this is a nested plug, don't actually assign it.
1694 INIT_LIST_HEAD(&plug->mq_list);
1695 INIT_LIST_HEAD(&plug->cb_list);
1697 plug->multiple_queues = false;
1698 plug->nowait = false;
1701 * Store ordering should not be needed here, since a potential
1702 * preempt will imply a full memory barrier
1706 EXPORT_SYMBOL(blk_start_plug);
1708 static void flush_plug_callbacks(struct blk_plug *plug, bool from_schedule)
1710 LIST_HEAD(callbacks);
1712 while (!list_empty(&plug->cb_list)) {
1713 list_splice_init(&plug->cb_list, &callbacks);
1715 while (!list_empty(&callbacks)) {
1716 struct blk_plug_cb *cb = list_first_entry(&callbacks,
1719 list_del(&cb->list);
1720 cb->callback(cb, from_schedule);
1725 struct blk_plug_cb *blk_check_plugged(blk_plug_cb_fn unplug, void *data,
1728 struct blk_plug *plug = current->plug;
1729 struct blk_plug_cb *cb;
1734 list_for_each_entry(cb, &plug->cb_list, list)
1735 if (cb->callback == unplug && cb->data == data)
1738 /* Not currently on the callback list */
1739 BUG_ON(size < sizeof(*cb));
1740 cb = kzalloc(size, GFP_ATOMIC);
1743 cb->callback = unplug;
1744 list_add(&cb->list, &plug->cb_list);
1748 EXPORT_SYMBOL(blk_check_plugged);
1750 void blk_flush_plug_list(struct blk_plug *plug, bool from_schedule)
1752 flush_plug_callbacks(plug, from_schedule);
1754 if (!list_empty(&plug->mq_list))
1755 blk_mq_flush_plug_list(plug, from_schedule);
1759 * blk_finish_plug - mark the end of a batch of submitted I/O
1760 * @plug: The &struct blk_plug passed to blk_start_plug()
1763 * Indicate that a batch of I/O submissions is complete. This function
1764 * must be paired with an initial call to blk_start_plug(). The intent
1765 * is to allow the block layer to optimize I/O submission. See the
1766 * documentation for blk_start_plug() for more information.
1768 void blk_finish_plug(struct blk_plug *plug)
1770 if (plug != current->plug)
1772 blk_flush_plug_list(plug, false);
1774 current->plug = NULL;
1776 EXPORT_SYMBOL(blk_finish_plug);
1778 void blk_io_schedule(void)
1780 /* Prevent hang_check timer from firing at us during very long I/O */
1781 unsigned long timeout = sysctl_hung_task_timeout_secs * HZ / 2;
1784 io_schedule_timeout(timeout);
1788 EXPORT_SYMBOL_GPL(blk_io_schedule);
1790 int __init blk_dev_init(void)
1792 BUILD_BUG_ON(REQ_OP_LAST >= (1 << REQ_OP_BITS));
1793 BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 *
1794 sizeof_field(struct request, cmd_flags));
1795 BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 *
1796 sizeof_field(struct bio, bi_opf));
1798 /* used for unplugging and affects IO latency/throughput - HIGHPRI */
1799 kblockd_workqueue = alloc_workqueue("kblockd",
1800 WQ_MEM_RECLAIM | WQ_HIGHPRI, 0);
1801 if (!kblockd_workqueue)
1802 panic("Failed to create kblockd\n");
1804 blk_requestq_cachep = kmem_cache_create("request_queue",
1805 sizeof(struct request_queue), 0, SLAB_PANIC, NULL);
1807 blk_debugfs_root = debugfs_create_dir("block", NULL);