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();
541 q->backing_dev_info->ra_pages = VM_READAHEAD_PAGES;
542 q->backing_dev_info->io_pages = VM_READAHEAD_PAGES;
543 q->backing_dev_info->capabilities = BDI_CAP_CGROUP_WRITEBACK;
546 timer_setup(&q->backing_dev_info->laptop_mode_wb_timer,
547 laptop_mode_timer_fn, 0);
548 timer_setup(&q->timeout, blk_rq_timed_out_timer, 0);
549 INIT_WORK(&q->timeout_work, blk_timeout_work);
550 INIT_LIST_HEAD(&q->icq_list);
551 #ifdef CONFIG_BLK_CGROUP
552 INIT_LIST_HEAD(&q->blkg_list);
555 kobject_init(&q->kobj, &blk_queue_ktype);
557 mutex_init(&q->debugfs_mutex);
558 mutex_init(&q->sysfs_lock);
559 mutex_init(&q->sysfs_dir_lock);
560 spin_lock_init(&q->queue_lock);
562 init_waitqueue_head(&q->mq_freeze_wq);
563 mutex_init(&q->mq_freeze_lock);
566 * Init percpu_ref in atomic mode so that it's faster to shutdown.
567 * See blk_register_queue() for details.
569 if (percpu_ref_init(&q->q_usage_counter,
570 blk_queue_usage_counter_release,
571 PERCPU_REF_INIT_ATOMIC, GFP_KERNEL))
574 if (blkcg_init_queue(q))
577 blk_queue_dma_alignment(q, 511);
578 blk_set_default_limits(&q->limits);
579 q->nr_requests = BLKDEV_MAX_RQ;
584 percpu_ref_exit(&q->q_usage_counter);
586 blk_free_queue_stats(q->stats);
588 bdi_put(q->backing_dev_info);
590 bioset_exit(&q->bio_split);
592 ida_simple_remove(&blk_queue_ida, q->id);
594 kmem_cache_free(blk_requestq_cachep, q);
597 EXPORT_SYMBOL(blk_alloc_queue);
600 * blk_get_queue - increment the request_queue refcount
601 * @q: the request_queue structure to increment the refcount for
603 * Increment the refcount of the request_queue kobject.
605 * Context: Any context.
607 bool blk_get_queue(struct request_queue *q)
609 if (likely(!blk_queue_dying(q))) {
616 EXPORT_SYMBOL(blk_get_queue);
619 * blk_get_request - allocate a request
620 * @q: request queue to allocate a request for
621 * @op: operation (REQ_OP_*) and REQ_* flags, e.g. REQ_SYNC.
622 * @flags: BLK_MQ_REQ_* flags, e.g. BLK_MQ_REQ_NOWAIT.
624 struct request *blk_get_request(struct request_queue *q, unsigned int op,
625 blk_mq_req_flags_t flags)
629 WARN_ON_ONCE(op & REQ_NOWAIT);
630 WARN_ON_ONCE(flags & ~(BLK_MQ_REQ_NOWAIT | BLK_MQ_REQ_PREEMPT));
632 req = blk_mq_alloc_request(q, op, flags);
633 if (!IS_ERR(req) && q->mq_ops->initialize_rq_fn)
634 q->mq_ops->initialize_rq_fn(req);
638 EXPORT_SYMBOL(blk_get_request);
640 void blk_put_request(struct request *req)
642 blk_mq_free_request(req);
644 EXPORT_SYMBOL(blk_put_request);
646 static void handle_bad_sector(struct bio *bio, sector_t maxsector)
648 char b[BDEVNAME_SIZE];
650 printk(KERN_INFO "attempt to access beyond end of device\n");
651 printk(KERN_INFO "%s: rw=%d, want=%Lu, limit=%Lu\n",
652 bio_devname(bio, b), bio->bi_opf,
653 (unsigned long long)bio_end_sector(bio),
654 (long long)maxsector);
657 #ifdef CONFIG_FAIL_MAKE_REQUEST
659 static DECLARE_FAULT_ATTR(fail_make_request);
661 static int __init setup_fail_make_request(char *str)
663 return setup_fault_attr(&fail_make_request, str);
665 __setup("fail_make_request=", setup_fail_make_request);
667 static bool should_fail_request(struct hd_struct *part, unsigned int bytes)
669 return part->make_it_fail && should_fail(&fail_make_request, bytes);
672 static int __init fail_make_request_debugfs(void)
674 struct dentry *dir = fault_create_debugfs_attr("fail_make_request",
675 NULL, &fail_make_request);
677 return PTR_ERR_OR_ZERO(dir);
680 late_initcall(fail_make_request_debugfs);
682 #else /* CONFIG_FAIL_MAKE_REQUEST */
684 static inline bool should_fail_request(struct hd_struct *part,
690 #endif /* CONFIG_FAIL_MAKE_REQUEST */
692 static inline bool bio_check_ro(struct bio *bio, struct hd_struct *part)
694 const int op = bio_op(bio);
696 if (part->policy && op_is_write(op)) {
697 char b[BDEVNAME_SIZE];
699 if (op_is_flush(bio->bi_opf) && !bio_sectors(bio))
703 "Trying to write to read-only block-device %s (partno %d)\n",
704 bio_devname(bio, b), part->partno);
705 /* Older lvm-tools actually trigger this */
712 static noinline int should_fail_bio(struct bio *bio)
714 if (should_fail_request(&bio->bi_disk->part0, bio->bi_iter.bi_size))
718 ALLOW_ERROR_INJECTION(should_fail_bio, ERRNO);
721 * Check whether this bio extends beyond the end of the device or partition.
722 * This may well happen - the kernel calls bread() without checking the size of
723 * the device, e.g., when mounting a file system.
725 static inline int bio_check_eod(struct bio *bio, sector_t maxsector)
727 unsigned int nr_sectors = bio_sectors(bio);
729 if (nr_sectors && maxsector &&
730 (nr_sectors > maxsector ||
731 bio->bi_iter.bi_sector > maxsector - nr_sectors)) {
732 handle_bad_sector(bio, maxsector);
739 * Remap block n of partition p to block n+start(p) of the disk.
741 static inline int blk_partition_remap(struct bio *bio)
747 p = __disk_get_part(bio->bi_disk, bio->bi_partno);
750 if (unlikely(should_fail_request(p, bio->bi_iter.bi_size)))
752 if (unlikely(bio_check_ro(bio, p)))
755 if (bio_sectors(bio)) {
756 if (bio_check_eod(bio, part_nr_sects_read(p)))
758 bio->bi_iter.bi_sector += p->start_sect;
759 trace_block_bio_remap(bio->bi_disk->queue, bio, part_devt(p),
760 bio->bi_iter.bi_sector - p->start_sect);
770 * Check write append to a zoned block device.
772 static inline blk_status_t blk_check_zone_append(struct request_queue *q,
775 sector_t pos = bio->bi_iter.bi_sector;
776 int nr_sectors = bio_sectors(bio);
778 /* Only applicable to zoned block devices */
779 if (!blk_queue_is_zoned(q))
780 return BLK_STS_NOTSUPP;
782 /* The bio sector must point to the start of a sequential zone */
783 if (pos & (blk_queue_zone_sectors(q) - 1) ||
784 !blk_queue_zone_is_seq(q, pos))
785 return BLK_STS_IOERR;
788 * Not allowed to cross zone boundaries. Otherwise, the BIO will be
789 * split and could result in non-contiguous sectors being written in
792 if (nr_sectors > q->limits.chunk_sectors)
793 return BLK_STS_IOERR;
795 /* Make sure the BIO is small enough and will not get split */
796 if (nr_sectors > q->limits.max_zone_append_sectors)
797 return BLK_STS_IOERR;
799 bio->bi_opf |= REQ_NOMERGE;
804 static noinline_for_stack bool submit_bio_checks(struct bio *bio)
806 struct request_queue *q = bio->bi_disk->queue;
807 blk_status_t status = BLK_STS_IOERR;
808 struct blk_plug *plug;
812 plug = blk_mq_plug(q, bio);
813 if (plug && plug->nowait)
814 bio->bi_opf |= REQ_NOWAIT;
817 * For a REQ_NOWAIT based request, return -EOPNOTSUPP
818 * if queue is not a request based queue.
820 if ((bio->bi_opf & REQ_NOWAIT) && !queue_is_mq(q))
823 if (should_fail_bio(bio))
826 if (bio->bi_partno) {
827 if (unlikely(blk_partition_remap(bio)))
830 if (unlikely(bio_check_ro(bio, &bio->bi_disk->part0)))
832 if (unlikely(bio_check_eod(bio, get_capacity(bio->bi_disk))))
837 * Filter flush bio's early so that bio based drivers without flush
838 * support don't have to worry about them.
840 if (op_is_flush(bio->bi_opf) &&
841 !test_bit(QUEUE_FLAG_WC, &q->queue_flags)) {
842 bio->bi_opf &= ~(REQ_PREFLUSH | REQ_FUA);
843 if (!bio_sectors(bio)) {
849 if (!test_bit(QUEUE_FLAG_POLL, &q->queue_flags))
850 bio->bi_opf &= ~REQ_HIPRI;
852 switch (bio_op(bio)) {
854 if (!blk_queue_discard(q))
857 case REQ_OP_SECURE_ERASE:
858 if (!blk_queue_secure_erase(q))
861 case REQ_OP_WRITE_SAME:
862 if (!q->limits.max_write_same_sectors)
865 case REQ_OP_ZONE_APPEND:
866 status = blk_check_zone_append(q, bio);
867 if (status != BLK_STS_OK)
870 case REQ_OP_ZONE_RESET:
871 case REQ_OP_ZONE_OPEN:
872 case REQ_OP_ZONE_CLOSE:
873 case REQ_OP_ZONE_FINISH:
874 if (!blk_queue_is_zoned(q))
877 case REQ_OP_ZONE_RESET_ALL:
878 if (!blk_queue_is_zoned(q) || !blk_queue_zone_resetall(q))
881 case REQ_OP_WRITE_ZEROES:
882 if (!q->limits.max_write_zeroes_sectors)
890 * Various block parts want %current->io_context, so allocate it up
891 * front rather than dealing with lots of pain to allocate it only
892 * where needed. This may fail and the block layer knows how to live
895 if (unlikely(!current->io_context))
896 create_task_io_context(current, GFP_ATOMIC, q->node);
898 if (blk_throtl_bio(bio)) {
899 blkcg_bio_issue_init(bio);
903 blk_cgroup_bio_start(bio);
904 blkcg_bio_issue_init(bio);
906 if (!bio_flagged(bio, BIO_TRACE_COMPLETION)) {
907 trace_block_bio_queue(q, bio);
908 /* Now that enqueuing has been traced, we need to trace
909 * completion as well.
911 bio_set_flag(bio, BIO_TRACE_COMPLETION);
916 status = BLK_STS_NOTSUPP;
918 bio->bi_status = status;
923 static blk_qc_t __submit_bio(struct bio *bio)
925 struct gendisk *disk = bio->bi_disk;
926 blk_qc_t ret = BLK_QC_T_NONE;
928 if (blk_crypto_bio_prep(&bio)) {
929 if (!disk->fops->submit_bio)
930 return blk_mq_submit_bio(bio);
931 ret = disk->fops->submit_bio(bio);
933 blk_queue_exit(disk->queue);
938 * The loop in this function may be a bit non-obvious, and so deserves some
941 * - Before entering the loop, bio->bi_next is NULL (as all callers ensure
942 * that), so we have a list with a single bio.
943 * - We pretend that we have just taken it off a longer list, so we assign
944 * bio_list to a pointer to the bio_list_on_stack, thus initialising the
945 * bio_list of new bios to be added. ->submit_bio() may indeed add some more
946 * bios through a recursive call to submit_bio_noacct. If it did, we find a
947 * non-NULL value in bio_list and re-enter the loop from the top.
948 * - In this case we really did just take the bio of the top of the list (no
949 * pretending) and so remove it from bio_list, and call into ->submit_bio()
952 * bio_list_on_stack[0] contains bios submitted by the current ->submit_bio.
953 * bio_list_on_stack[1] contains bios that were submitted before the current
954 * ->submit_bio_bio, but that haven't been processed yet.
956 static blk_qc_t __submit_bio_noacct(struct bio *bio)
958 struct bio_list bio_list_on_stack[2];
959 blk_qc_t ret = BLK_QC_T_NONE;
961 BUG_ON(bio->bi_next);
963 bio_list_init(&bio_list_on_stack[0]);
964 current->bio_list = bio_list_on_stack;
967 struct request_queue *q = bio->bi_disk->queue;
968 struct bio_list lower, same;
970 if (unlikely(bio_queue_enter(bio) != 0))
974 * Create a fresh bio_list for all subordinate requests.
976 bio_list_on_stack[1] = bio_list_on_stack[0];
977 bio_list_init(&bio_list_on_stack[0]);
979 ret = __submit_bio(bio);
982 * Sort new bios into those for a lower level and those for the
985 bio_list_init(&lower);
986 bio_list_init(&same);
987 while ((bio = bio_list_pop(&bio_list_on_stack[0])) != NULL)
988 if (q == bio->bi_disk->queue)
989 bio_list_add(&same, bio);
991 bio_list_add(&lower, bio);
994 * Now assemble so we handle the lowest level first.
996 bio_list_merge(&bio_list_on_stack[0], &lower);
997 bio_list_merge(&bio_list_on_stack[0], &same);
998 bio_list_merge(&bio_list_on_stack[0], &bio_list_on_stack[1]);
999 } while ((bio = bio_list_pop(&bio_list_on_stack[0])));
1001 current->bio_list = NULL;
1005 static blk_qc_t __submit_bio_noacct_mq(struct bio *bio)
1007 struct bio_list bio_list[2] = { };
1008 blk_qc_t ret = BLK_QC_T_NONE;
1010 current->bio_list = bio_list;
1013 struct gendisk *disk = bio->bi_disk;
1015 if (unlikely(bio_queue_enter(bio) != 0))
1018 if (!blk_crypto_bio_prep(&bio)) {
1019 blk_queue_exit(disk->queue);
1020 ret = BLK_QC_T_NONE;
1024 ret = blk_mq_submit_bio(bio);
1025 } while ((bio = bio_list_pop(&bio_list[0])));
1027 current->bio_list = NULL;
1032 * submit_bio_noacct - re-submit a bio to the block device layer for I/O
1033 * @bio: The bio describing the location in memory and on the device.
1035 * This is a version of submit_bio() that shall only be used for I/O that is
1036 * resubmitted to lower level drivers by stacking block drivers. All file
1037 * systems and other upper level users of the block layer should use
1038 * submit_bio() instead.
1040 blk_qc_t submit_bio_noacct(struct bio *bio)
1042 if (!submit_bio_checks(bio))
1043 return BLK_QC_T_NONE;
1046 * We only want one ->submit_bio to be active at a time, else stack
1047 * usage with stacked devices could be a problem. Use current->bio_list
1048 * to collect a list of requests submited by a ->submit_bio method while
1049 * it is active, and then process them after it returned.
1051 if (current->bio_list) {
1052 bio_list_add(¤t->bio_list[0], bio);
1053 return BLK_QC_T_NONE;
1056 if (!bio->bi_disk->fops->submit_bio)
1057 return __submit_bio_noacct_mq(bio);
1058 return __submit_bio_noacct(bio);
1060 EXPORT_SYMBOL(submit_bio_noacct);
1063 * submit_bio - submit a bio to the block device layer for I/O
1064 * @bio: The &struct bio which describes the I/O
1066 * submit_bio() is used to submit I/O requests to block devices. It is passed a
1067 * fully set up &struct bio that describes the I/O that needs to be done. The
1068 * bio will be send to the device described by the bi_disk and bi_partno fields.
1070 * The success/failure status of the request, along with notification of
1071 * completion, is delivered asynchronously through the ->bi_end_io() callback
1072 * in @bio. The bio must NOT be touched by thecaller until ->bi_end_io() has
1075 blk_qc_t submit_bio(struct bio *bio)
1077 if (blkcg_punt_bio_submit(bio))
1078 return BLK_QC_T_NONE;
1081 * If it's a regular read/write or a barrier with data attached,
1082 * go through the normal accounting stuff before submission.
1084 if (bio_has_data(bio)) {
1087 if (unlikely(bio_op(bio) == REQ_OP_WRITE_SAME))
1088 count = queue_logical_block_size(bio->bi_disk->queue) >> 9;
1090 count = bio_sectors(bio);
1092 if (op_is_write(bio_op(bio))) {
1093 count_vm_events(PGPGOUT, count);
1095 task_io_account_read(bio->bi_iter.bi_size);
1096 count_vm_events(PGPGIN, count);
1099 if (unlikely(block_dump)) {
1100 char b[BDEVNAME_SIZE];
1101 printk(KERN_DEBUG "%s(%d): %s block %Lu on %s (%u sectors)\n",
1102 current->comm, task_pid_nr(current),
1103 op_is_write(bio_op(bio)) ? "WRITE" : "READ",
1104 (unsigned long long)bio->bi_iter.bi_sector,
1105 bio_devname(bio, b), count);
1110 * If we're reading data that is part of the userspace workingset, count
1111 * submission time as memory stall. When the device is congested, or
1112 * the submitting cgroup IO-throttled, submission can be a significant
1113 * part of overall IO time.
1115 if (unlikely(bio_op(bio) == REQ_OP_READ &&
1116 bio_flagged(bio, BIO_WORKINGSET))) {
1117 unsigned long pflags;
1120 psi_memstall_enter(&pflags);
1121 ret = submit_bio_noacct(bio);
1122 psi_memstall_leave(&pflags);
1127 return submit_bio_noacct(bio);
1129 EXPORT_SYMBOL(submit_bio);
1132 * blk_cloned_rq_check_limits - Helper function to check a cloned request
1133 * for the new queue limits
1135 * @rq: the request being checked
1138 * @rq may have been made based on weaker limitations of upper-level queues
1139 * in request stacking drivers, and it may violate the limitation of @q.
1140 * Since the block layer and the underlying device driver trust @rq
1141 * after it is inserted to @q, it should be checked against @q before
1142 * the insertion using this generic function.
1144 * Request stacking drivers like request-based dm may change the queue
1145 * limits when retrying requests on other queues. Those requests need
1146 * to be checked against the new queue limits again during dispatch.
1148 static blk_status_t blk_cloned_rq_check_limits(struct request_queue *q,
1151 if (blk_rq_sectors(rq) > blk_queue_get_max_sectors(q, req_op(rq))) {
1152 printk(KERN_ERR "%s: over max size limit. (%u > %u)\n",
1153 __func__, blk_rq_sectors(rq),
1154 blk_queue_get_max_sectors(q, req_op(rq)));
1155 return BLK_STS_IOERR;
1159 * queue's settings related to segment counting like q->bounce_pfn
1160 * may differ from that of other stacking queues.
1161 * Recalculate it to check the request correctly on this queue's
1164 rq->nr_phys_segments = blk_recalc_rq_segments(rq);
1165 if (rq->nr_phys_segments > queue_max_segments(q)) {
1166 printk(KERN_ERR "%s: over max segments limit. (%hu > %hu)\n",
1167 __func__, rq->nr_phys_segments, queue_max_segments(q));
1168 return BLK_STS_IOERR;
1175 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
1176 * @q: the queue to submit the request
1177 * @rq: the request being queued
1179 blk_status_t blk_insert_cloned_request(struct request_queue *q, struct request *rq)
1181 if (blk_cloned_rq_check_limits(q, rq))
1182 return BLK_STS_IOERR;
1185 should_fail_request(&rq->rq_disk->part0, blk_rq_bytes(rq)))
1186 return BLK_STS_IOERR;
1188 if (blk_crypto_insert_cloned_request(rq))
1189 return BLK_STS_IOERR;
1191 if (blk_queue_io_stat(q))
1192 blk_account_io_start(rq);
1195 * Since we have a scheduler attached on the top device,
1196 * bypass a potential scheduler on the bottom device for
1199 return blk_mq_request_issue_directly(rq, true);
1201 EXPORT_SYMBOL_GPL(blk_insert_cloned_request);
1204 * blk_rq_err_bytes - determine number of bytes till the next failure boundary
1205 * @rq: request to examine
1208 * A request could be merge of IOs which require different failure
1209 * handling. This function determines the number of bytes which
1210 * can be failed from the beginning of the request without
1211 * crossing into area which need to be retried further.
1214 * The number of bytes to fail.
1216 unsigned int blk_rq_err_bytes(const struct request *rq)
1218 unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK;
1219 unsigned int bytes = 0;
1222 if (!(rq->rq_flags & RQF_MIXED_MERGE))
1223 return blk_rq_bytes(rq);
1226 * Currently the only 'mixing' which can happen is between
1227 * different fastfail types. We can safely fail portions
1228 * which have all the failfast bits that the first one has -
1229 * the ones which are at least as eager to fail as the first
1232 for (bio = rq->bio; bio; bio = bio->bi_next) {
1233 if ((bio->bi_opf & ff) != ff)
1235 bytes += bio->bi_iter.bi_size;
1238 /* this could lead to infinite loop */
1239 BUG_ON(blk_rq_bytes(rq) && !bytes);
1242 EXPORT_SYMBOL_GPL(blk_rq_err_bytes);
1244 static void update_io_ticks(struct hd_struct *part, unsigned long now, bool end)
1246 unsigned long stamp;
1248 stamp = READ_ONCE(part->stamp);
1249 if (unlikely(stamp != now)) {
1250 if (likely(cmpxchg(&part->stamp, stamp, now) == stamp))
1251 __part_stat_add(part, io_ticks, end ? now - stamp : 1);
1254 part = &part_to_disk(part)->part0;
1259 static void blk_account_io_completion(struct request *req, unsigned int bytes)
1261 if (req->part && blk_do_io_stat(req)) {
1262 const int sgrp = op_stat_group(req_op(req));
1263 struct hd_struct *part;
1267 part_stat_add(part, sectors[sgrp], bytes >> 9);
1272 void blk_account_io_done(struct request *req, u64 now)
1275 * Account IO completion. flush_rq isn't accounted as a
1276 * normal IO on queueing nor completion. Accounting the
1277 * containing request is enough.
1279 if (req->part && blk_do_io_stat(req) &&
1280 !(req->rq_flags & RQF_FLUSH_SEQ)) {
1281 const int sgrp = op_stat_group(req_op(req));
1282 struct hd_struct *part;
1287 update_io_ticks(part, jiffies, true);
1288 part_stat_inc(part, ios[sgrp]);
1289 part_stat_add(part, nsecs[sgrp], now - req->start_time_ns);
1292 hd_struct_put(part);
1296 void blk_account_io_start(struct request *rq)
1298 if (!blk_do_io_stat(rq))
1301 rq->part = disk_map_sector_rcu(rq->rq_disk, blk_rq_pos(rq));
1304 update_io_ticks(rq->part, jiffies, false);
1308 unsigned long disk_start_io_acct(struct gendisk *disk, unsigned int sectors,
1311 struct hd_struct *part = &disk->part0;
1312 const int sgrp = op_stat_group(op);
1313 unsigned long now = READ_ONCE(jiffies);
1316 update_io_ticks(part, now, false);
1317 part_stat_inc(part, ios[sgrp]);
1318 part_stat_add(part, sectors[sgrp], sectors);
1319 part_stat_local_inc(part, in_flight[op_is_write(op)]);
1324 EXPORT_SYMBOL(disk_start_io_acct);
1326 void disk_end_io_acct(struct gendisk *disk, unsigned int op,
1327 unsigned long start_time)
1329 struct hd_struct *part = &disk->part0;
1330 const int sgrp = op_stat_group(op);
1331 unsigned long now = READ_ONCE(jiffies);
1332 unsigned long duration = now - start_time;
1335 update_io_ticks(part, now, true);
1336 part_stat_add(part, nsecs[sgrp], jiffies_to_nsecs(duration));
1337 part_stat_local_dec(part, in_flight[op_is_write(op)]);
1340 EXPORT_SYMBOL(disk_end_io_acct);
1343 * Steal bios from a request and add them to a bio list.
1344 * The request must not have been partially completed before.
1346 void blk_steal_bios(struct bio_list *list, struct request *rq)
1350 list->tail->bi_next = rq->bio;
1352 list->head = rq->bio;
1353 list->tail = rq->biotail;
1361 EXPORT_SYMBOL_GPL(blk_steal_bios);
1364 * blk_update_request - Special helper function for request stacking drivers
1365 * @req: the request being processed
1366 * @error: block status code
1367 * @nr_bytes: number of bytes to complete @req
1370 * Ends I/O on a number of bytes attached to @req, but doesn't complete
1371 * the request structure even if @req doesn't have leftover.
1372 * If @req has leftover, sets it up for the next range of segments.
1374 * This special helper function is only for request stacking drivers
1375 * (e.g. request-based dm) so that they can handle partial completion.
1376 * Actual device drivers should use blk_mq_end_request instead.
1378 * Passing the result of blk_rq_bytes() as @nr_bytes guarantees
1379 * %false return from this function.
1382 * The RQF_SPECIAL_PAYLOAD flag is ignored on purpose in both
1383 * blk_rq_bytes() and in blk_update_request().
1386 * %false - this request doesn't have any more data
1387 * %true - this request has more data
1389 bool blk_update_request(struct request *req, blk_status_t error,
1390 unsigned int nr_bytes)
1394 trace_block_rq_complete(req, blk_status_to_errno(error), nr_bytes);
1399 #ifdef CONFIG_BLK_DEV_INTEGRITY
1400 if (blk_integrity_rq(req) && req_op(req) == REQ_OP_READ &&
1401 error == BLK_STS_OK)
1402 req->q->integrity.profile->complete_fn(req, nr_bytes);
1405 if (unlikely(error && !blk_rq_is_passthrough(req) &&
1406 !(req->rq_flags & RQF_QUIET)))
1407 print_req_error(req, error, __func__);
1409 blk_account_io_completion(req, nr_bytes);
1413 struct bio *bio = req->bio;
1414 unsigned bio_bytes = min(bio->bi_iter.bi_size, nr_bytes);
1416 if (bio_bytes == bio->bi_iter.bi_size)
1417 req->bio = bio->bi_next;
1419 /* Completion has already been traced */
1420 bio_clear_flag(bio, BIO_TRACE_COMPLETION);
1421 req_bio_endio(req, bio, bio_bytes, error);
1423 total_bytes += bio_bytes;
1424 nr_bytes -= bio_bytes;
1435 * Reset counters so that the request stacking driver
1436 * can find how many bytes remain in the request
1439 req->__data_len = 0;
1443 req->__data_len -= total_bytes;
1445 /* update sector only for requests with clear definition of sector */
1446 if (!blk_rq_is_passthrough(req))
1447 req->__sector += total_bytes >> 9;
1449 /* mixed attributes always follow the first bio */
1450 if (req->rq_flags & RQF_MIXED_MERGE) {
1451 req->cmd_flags &= ~REQ_FAILFAST_MASK;
1452 req->cmd_flags |= req->bio->bi_opf & REQ_FAILFAST_MASK;
1455 if (!(req->rq_flags & RQF_SPECIAL_PAYLOAD)) {
1457 * If total number of sectors is less than the first segment
1458 * size, something has gone terribly wrong.
1460 if (blk_rq_bytes(req) < blk_rq_cur_bytes(req)) {
1461 blk_dump_rq_flags(req, "request botched");
1462 req->__data_len = blk_rq_cur_bytes(req);
1465 /* recalculate the number of segments */
1466 req->nr_phys_segments = blk_recalc_rq_segments(req);
1471 EXPORT_SYMBOL_GPL(blk_update_request);
1473 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
1475 * rq_flush_dcache_pages - Helper function to flush all pages in a request
1476 * @rq: the request to be flushed
1479 * Flush all pages in @rq.
1481 void rq_flush_dcache_pages(struct request *rq)
1483 struct req_iterator iter;
1484 struct bio_vec bvec;
1486 rq_for_each_segment(bvec, rq, iter)
1487 flush_dcache_page(bvec.bv_page);
1489 EXPORT_SYMBOL_GPL(rq_flush_dcache_pages);
1493 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
1494 * @q : the queue of the device being checked
1497 * Check if underlying low-level drivers of a device are busy.
1498 * If the drivers want to export their busy state, they must set own
1499 * exporting function using blk_queue_lld_busy() first.
1501 * Basically, this function is used only by request stacking drivers
1502 * to stop dispatching requests to underlying devices when underlying
1503 * devices are busy. This behavior helps more I/O merging on the queue
1504 * of the request stacking driver and prevents I/O throughput regression
1505 * on burst I/O load.
1508 * 0 - Not busy (The request stacking driver should dispatch request)
1509 * 1 - Busy (The request stacking driver should stop dispatching request)
1511 int blk_lld_busy(struct request_queue *q)
1513 if (queue_is_mq(q) && q->mq_ops->busy)
1514 return q->mq_ops->busy(q);
1518 EXPORT_SYMBOL_GPL(blk_lld_busy);
1521 * blk_rq_unprep_clone - Helper function to free all bios in a cloned request
1522 * @rq: the clone request to be cleaned up
1525 * Free all bios in @rq for a cloned request.
1527 void blk_rq_unprep_clone(struct request *rq)
1531 while ((bio = rq->bio) != NULL) {
1532 rq->bio = bio->bi_next;
1537 EXPORT_SYMBOL_GPL(blk_rq_unprep_clone);
1540 * blk_rq_prep_clone - Helper function to setup clone request
1541 * @rq: the request to be setup
1542 * @rq_src: original request to be cloned
1543 * @bs: bio_set that bios for clone are allocated from
1544 * @gfp_mask: memory allocation mask for bio
1545 * @bio_ctr: setup function to be called for each clone bio.
1546 * Returns %0 for success, non %0 for failure.
1547 * @data: private data to be passed to @bio_ctr
1550 * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq.
1551 * Also, pages which the original bios are pointing to are not copied
1552 * and the cloned bios just point same pages.
1553 * So cloned bios must be completed before original bios, which means
1554 * the caller must complete @rq before @rq_src.
1556 int blk_rq_prep_clone(struct request *rq, struct request *rq_src,
1557 struct bio_set *bs, gfp_t gfp_mask,
1558 int (*bio_ctr)(struct bio *, struct bio *, void *),
1561 struct bio *bio, *bio_src;
1566 __rq_for_each_bio(bio_src, rq_src) {
1567 bio = bio_clone_fast(bio_src, gfp_mask, bs);
1571 if (bio_ctr && bio_ctr(bio, bio_src, data))
1575 rq->biotail->bi_next = bio;
1578 rq->bio = rq->biotail = bio;
1581 /* Copy attributes of the original request to the clone request. */
1582 rq->__sector = blk_rq_pos(rq_src);
1583 rq->__data_len = blk_rq_bytes(rq_src);
1584 if (rq_src->rq_flags & RQF_SPECIAL_PAYLOAD) {
1585 rq->rq_flags |= RQF_SPECIAL_PAYLOAD;
1586 rq->special_vec = rq_src->special_vec;
1588 rq->nr_phys_segments = rq_src->nr_phys_segments;
1589 rq->ioprio = rq_src->ioprio;
1592 blk_crypto_rq_bio_prep(rq, rq->bio, gfp_mask);
1599 blk_rq_unprep_clone(rq);
1603 EXPORT_SYMBOL_GPL(blk_rq_prep_clone);
1605 int kblockd_schedule_work(struct work_struct *work)
1607 return queue_work(kblockd_workqueue, work);
1609 EXPORT_SYMBOL(kblockd_schedule_work);
1611 int kblockd_mod_delayed_work_on(int cpu, struct delayed_work *dwork,
1612 unsigned long delay)
1614 return mod_delayed_work_on(cpu, kblockd_workqueue, dwork, delay);
1616 EXPORT_SYMBOL(kblockd_mod_delayed_work_on);
1619 * blk_start_plug - initialize blk_plug and track it inside the task_struct
1620 * @plug: The &struct blk_plug that needs to be initialized
1623 * blk_start_plug() indicates to the block layer an intent by the caller
1624 * to submit multiple I/O requests in a batch. The block layer may use
1625 * this hint to defer submitting I/Os from the caller until blk_finish_plug()
1626 * is called. However, the block layer may choose to submit requests
1627 * before a call to blk_finish_plug() if the number of queued I/Os
1628 * exceeds %BLK_MAX_REQUEST_COUNT, or if the size of the I/O is larger than
1629 * %BLK_PLUG_FLUSH_SIZE. The queued I/Os may also be submitted early if
1630 * the task schedules (see below).
1632 * Tracking blk_plug inside the task_struct will help with auto-flushing the
1633 * pending I/O should the task end up blocking between blk_start_plug() and
1634 * blk_finish_plug(). This is important from a performance perspective, but
1635 * also ensures that we don't deadlock. For instance, if the task is blocking
1636 * for a memory allocation, memory reclaim could end up wanting to free a
1637 * page belonging to that request that is currently residing in our private
1638 * plug. By flushing the pending I/O when the process goes to sleep, we avoid
1639 * this kind of deadlock.
1641 void blk_start_plug(struct blk_plug *plug)
1643 struct task_struct *tsk = current;
1646 * If this is a nested plug, don't actually assign it.
1651 INIT_LIST_HEAD(&plug->mq_list);
1652 INIT_LIST_HEAD(&plug->cb_list);
1654 plug->multiple_queues = false;
1655 plug->nowait = false;
1658 * Store ordering should not be needed here, since a potential
1659 * preempt will imply a full memory barrier
1663 EXPORT_SYMBOL(blk_start_plug);
1665 static void flush_plug_callbacks(struct blk_plug *plug, bool from_schedule)
1667 LIST_HEAD(callbacks);
1669 while (!list_empty(&plug->cb_list)) {
1670 list_splice_init(&plug->cb_list, &callbacks);
1672 while (!list_empty(&callbacks)) {
1673 struct blk_plug_cb *cb = list_first_entry(&callbacks,
1676 list_del(&cb->list);
1677 cb->callback(cb, from_schedule);
1682 struct blk_plug_cb *blk_check_plugged(blk_plug_cb_fn unplug, void *data,
1685 struct blk_plug *plug = current->plug;
1686 struct blk_plug_cb *cb;
1691 list_for_each_entry(cb, &plug->cb_list, list)
1692 if (cb->callback == unplug && cb->data == data)
1695 /* Not currently on the callback list */
1696 BUG_ON(size < sizeof(*cb));
1697 cb = kzalloc(size, GFP_ATOMIC);
1700 cb->callback = unplug;
1701 list_add(&cb->list, &plug->cb_list);
1705 EXPORT_SYMBOL(blk_check_plugged);
1707 void blk_flush_plug_list(struct blk_plug *plug, bool from_schedule)
1709 flush_plug_callbacks(plug, from_schedule);
1711 if (!list_empty(&plug->mq_list))
1712 blk_mq_flush_plug_list(plug, from_schedule);
1716 * blk_finish_plug - mark the end of a batch of submitted I/O
1717 * @plug: The &struct blk_plug passed to blk_start_plug()
1720 * Indicate that a batch of I/O submissions is complete. This function
1721 * must be paired with an initial call to blk_start_plug(). The intent
1722 * is to allow the block layer to optimize I/O submission. See the
1723 * documentation for blk_start_plug() for more information.
1725 void blk_finish_plug(struct blk_plug *plug)
1727 if (plug != current->plug)
1729 blk_flush_plug_list(plug, false);
1731 current->plug = NULL;
1733 EXPORT_SYMBOL(blk_finish_plug);
1735 void blk_io_schedule(void)
1737 /* Prevent hang_check timer from firing at us during very long I/O */
1738 unsigned long timeout = sysctl_hung_task_timeout_secs * HZ / 2;
1741 io_schedule_timeout(timeout);
1745 EXPORT_SYMBOL_GPL(blk_io_schedule);
1747 int __init blk_dev_init(void)
1749 BUILD_BUG_ON(REQ_OP_LAST >= (1 << REQ_OP_BITS));
1750 BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 *
1751 sizeof_field(struct request, cmd_flags));
1752 BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 *
1753 sizeof_field(struct bio, bi_opf));
1755 /* used for unplugging and affects IO latency/throughput - HIGHPRI */
1756 kblockd_workqueue = alloc_workqueue("kblockd",
1757 WQ_MEM_RECLAIM | WQ_HIGHPRI, 0);
1758 if (!kblockd_workqueue)
1759 panic("Failed to create kblockd\n");
1761 blk_requestq_cachep = kmem_cache_create("request_queue",
1762 sizeof(struct request_queue), 0, SLAB_PANIC, NULL);
1764 blk_debugfs_root = debugfs_create_dir("block", NULL);