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/kernel_stat.h>
24 #include <linux/string.h>
25 #include <linux/init.h>
26 #include <linux/completion.h>
27 #include <linux/slab.h>
28 #include <linux/swap.h>
29 #include <linux/writeback.h>
30 #include <linux/task_io_accounting_ops.h>
31 #include <linux/fault-inject.h>
32 #include <linux/list_sort.h>
33 #include <linux/delay.h>
34 #include <linux/ratelimit.h>
35 #include <linux/pm_runtime.h>
36 #include <linux/blk-cgroup.h>
37 #include <linux/debugfs.h>
38 #include <linux/bpf.h>
39 #include <linux/psi.h>
41 #define CREATE_TRACE_POINTS
42 #include <trace/events/block.h>
46 #include "blk-mq-sched.h"
48 #include "blk-rq-qos.h"
50 #ifdef CONFIG_DEBUG_FS
51 struct dentry *blk_debugfs_root;
54 EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_remap);
55 EXPORT_TRACEPOINT_SYMBOL_GPL(block_rq_remap);
56 EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_complete);
57 EXPORT_TRACEPOINT_SYMBOL_GPL(block_split);
58 EXPORT_TRACEPOINT_SYMBOL_GPL(block_unplug);
60 DEFINE_IDA(blk_queue_ida);
63 * For queue allocation
65 struct kmem_cache *blk_requestq_cachep;
68 * Controlling structure to kblockd
70 static struct workqueue_struct *kblockd_workqueue;
73 * blk_queue_flag_set - atomically set a queue flag
74 * @flag: flag to be set
77 void blk_queue_flag_set(unsigned int flag, struct request_queue *q)
79 set_bit(flag, &q->queue_flags);
81 EXPORT_SYMBOL(blk_queue_flag_set);
84 * blk_queue_flag_clear - atomically clear a queue flag
85 * @flag: flag to be cleared
88 void blk_queue_flag_clear(unsigned int flag, struct request_queue *q)
90 clear_bit(flag, &q->queue_flags);
92 EXPORT_SYMBOL(blk_queue_flag_clear);
95 * blk_queue_flag_test_and_set - atomically test and set a queue flag
96 * @flag: flag to be set
99 * Returns the previous value of @flag - 0 if the flag was not set and 1 if
100 * the flag was already set.
102 bool blk_queue_flag_test_and_set(unsigned int flag, struct request_queue *q)
104 return test_and_set_bit(flag, &q->queue_flags);
106 EXPORT_SYMBOL_GPL(blk_queue_flag_test_and_set);
108 void blk_rq_init(struct request_queue *q, struct request *rq)
110 memset(rq, 0, sizeof(*rq));
112 INIT_LIST_HEAD(&rq->queuelist);
114 rq->__sector = (sector_t) -1;
115 INIT_HLIST_NODE(&rq->hash);
116 RB_CLEAR_NODE(&rq->rb_node);
118 rq->internal_tag = -1;
119 rq->start_time_ns = ktime_get_ns();
121 refcount_set(&rq->ref, 1);
123 EXPORT_SYMBOL(blk_rq_init);
125 #define REQ_OP_NAME(name) [REQ_OP_##name] = #name
126 static const char *const blk_op_name[] = {
130 REQ_OP_NAME(DISCARD),
131 REQ_OP_NAME(SECURE_ERASE),
132 REQ_OP_NAME(ZONE_RESET),
133 REQ_OP_NAME(ZONE_RESET_ALL),
134 REQ_OP_NAME(WRITE_SAME),
135 REQ_OP_NAME(WRITE_ZEROES),
136 REQ_OP_NAME(SCSI_IN),
137 REQ_OP_NAME(SCSI_OUT),
139 REQ_OP_NAME(DRV_OUT),
144 * blk_op_str - Return string XXX in the REQ_OP_XXX.
147 * Description: Centralize block layer function to convert REQ_OP_XXX into
148 * string format. Useful in the debugging and tracing bio or request. For
149 * invalid REQ_OP_XXX it returns string "UNKNOWN".
151 inline const char *blk_op_str(unsigned int op)
153 const char *op_str = "UNKNOWN";
155 if (op < ARRAY_SIZE(blk_op_name) && blk_op_name[op])
156 op_str = blk_op_name[op];
160 EXPORT_SYMBOL_GPL(blk_op_str);
162 static const struct {
166 [BLK_STS_OK] = { 0, "" },
167 [BLK_STS_NOTSUPP] = { -EOPNOTSUPP, "operation not supported" },
168 [BLK_STS_TIMEOUT] = { -ETIMEDOUT, "timeout" },
169 [BLK_STS_NOSPC] = { -ENOSPC, "critical space allocation" },
170 [BLK_STS_TRANSPORT] = { -ENOLINK, "recoverable transport" },
171 [BLK_STS_TARGET] = { -EREMOTEIO, "critical target" },
172 [BLK_STS_NEXUS] = { -EBADE, "critical nexus" },
173 [BLK_STS_MEDIUM] = { -ENODATA, "critical medium" },
174 [BLK_STS_PROTECTION] = { -EILSEQ, "protection" },
175 [BLK_STS_RESOURCE] = { -ENOMEM, "kernel resource" },
176 [BLK_STS_DEV_RESOURCE] = { -EBUSY, "device resource" },
177 [BLK_STS_AGAIN] = { -EAGAIN, "nonblocking retry" },
179 /* device mapper special case, should not leak out: */
180 [BLK_STS_DM_REQUEUE] = { -EREMCHG, "dm internal retry" },
182 /* everything else not covered above: */
183 [BLK_STS_IOERR] = { -EIO, "I/O" },
186 blk_status_t errno_to_blk_status(int errno)
190 for (i = 0; i < ARRAY_SIZE(blk_errors); i++) {
191 if (blk_errors[i].errno == errno)
192 return (__force blk_status_t)i;
195 return BLK_STS_IOERR;
197 EXPORT_SYMBOL_GPL(errno_to_blk_status);
199 int blk_status_to_errno(blk_status_t status)
201 int idx = (__force int)status;
203 if (WARN_ON_ONCE(idx >= ARRAY_SIZE(blk_errors)))
205 return blk_errors[idx].errno;
207 EXPORT_SYMBOL_GPL(blk_status_to_errno);
209 static void print_req_error(struct request *req, blk_status_t status,
212 int idx = (__force int)status;
214 if (WARN_ON_ONCE(idx >= ARRAY_SIZE(blk_errors)))
217 printk_ratelimited(KERN_ERR
218 "%s: %s error, dev %s, sector %llu op 0x%x:(%s) flags 0x%x "
219 "phys_seg %u prio class %u\n",
220 caller, blk_errors[idx].name,
221 req->rq_disk ? req->rq_disk->disk_name : "?",
222 blk_rq_pos(req), req_op(req), blk_op_str(req_op(req)),
223 req->cmd_flags & ~REQ_OP_MASK,
224 req->nr_phys_segments,
225 IOPRIO_PRIO_CLASS(req->ioprio));
228 static void req_bio_endio(struct request *rq, struct bio *bio,
229 unsigned int nbytes, blk_status_t error)
232 bio->bi_status = error;
234 if (unlikely(rq->rq_flags & RQF_QUIET))
235 bio_set_flag(bio, BIO_QUIET);
237 bio_advance(bio, nbytes);
239 /* don't actually finish bio if it's part of flush sequence */
240 if (bio->bi_iter.bi_size == 0 && !(rq->rq_flags & RQF_FLUSH_SEQ))
244 void blk_dump_rq_flags(struct request *rq, char *msg)
246 printk(KERN_INFO "%s: dev %s: flags=%llx\n", msg,
247 rq->rq_disk ? rq->rq_disk->disk_name : "?",
248 (unsigned long long) rq->cmd_flags);
250 printk(KERN_INFO " sector %llu, nr/cnr %u/%u\n",
251 (unsigned long long)blk_rq_pos(rq),
252 blk_rq_sectors(rq), blk_rq_cur_sectors(rq));
253 printk(KERN_INFO " bio %p, biotail %p, len %u\n",
254 rq->bio, rq->biotail, blk_rq_bytes(rq));
256 EXPORT_SYMBOL(blk_dump_rq_flags);
259 * blk_sync_queue - cancel any pending callbacks on a queue
263 * The block layer may perform asynchronous callback activity
264 * on a queue, such as calling the unplug function after a timeout.
265 * A block device may call blk_sync_queue to ensure that any
266 * such activity is cancelled, thus allowing it to release resources
267 * that the callbacks might use. The caller must already have made sure
268 * that its ->make_request_fn will not re-add plugging prior to calling
271 * This function does not cancel any asynchronous activity arising
272 * out of elevator or throttling code. That would require elevator_exit()
273 * and blkcg_exit_queue() to be called with queue lock initialized.
276 void blk_sync_queue(struct request_queue *q)
278 del_timer_sync(&q->timeout);
279 cancel_work_sync(&q->timeout_work);
281 EXPORT_SYMBOL(blk_sync_queue);
284 * blk_set_pm_only - increment pm_only counter
285 * @q: request queue pointer
287 void blk_set_pm_only(struct request_queue *q)
289 atomic_inc(&q->pm_only);
291 EXPORT_SYMBOL_GPL(blk_set_pm_only);
293 void blk_clear_pm_only(struct request_queue *q)
297 pm_only = atomic_dec_return(&q->pm_only);
298 WARN_ON_ONCE(pm_only < 0);
300 wake_up_all(&q->mq_freeze_wq);
302 EXPORT_SYMBOL_GPL(blk_clear_pm_only);
304 void blk_put_queue(struct request_queue *q)
306 kobject_put(&q->kobj);
308 EXPORT_SYMBOL(blk_put_queue);
310 void blk_set_queue_dying(struct request_queue *q)
312 blk_queue_flag_set(QUEUE_FLAG_DYING, q);
315 * When queue DYING flag is set, we need to block new req
316 * entering queue, so we call blk_freeze_queue_start() to
317 * prevent I/O from crossing blk_queue_enter().
319 blk_freeze_queue_start(q);
322 blk_mq_wake_waiters(q);
324 /* Make blk_queue_enter() reexamine the DYING flag. */
325 wake_up_all(&q->mq_freeze_wq);
327 EXPORT_SYMBOL_GPL(blk_set_queue_dying);
330 * blk_cleanup_queue - shutdown a request queue
331 * @q: request queue to shutdown
333 * Mark @q DYING, drain all pending requests, mark @q DEAD, destroy and
334 * put it. All future requests will be failed immediately with -ENODEV.
336 void blk_cleanup_queue(struct request_queue *q)
338 /* mark @q DYING, no new request or merges will be allowed afterwards */
339 mutex_lock(&q->sysfs_lock);
340 blk_set_queue_dying(q);
342 blk_queue_flag_set(QUEUE_FLAG_NOMERGES, q);
343 blk_queue_flag_set(QUEUE_FLAG_NOXMERGES, q);
344 blk_queue_flag_set(QUEUE_FLAG_DYING, q);
345 mutex_unlock(&q->sysfs_lock);
348 * Drain all requests queued before DYING marking. Set DEAD flag to
349 * prevent that blk_mq_run_hw_queues() accesses the hardware queues
350 * after draining finished.
356 blk_queue_flag_set(QUEUE_FLAG_DEAD, q);
358 /* for synchronous bio-based driver finish in-flight integrity i/o */
359 blk_flush_integrity();
361 /* @q won't process any more request, flush async actions */
362 del_timer_sync(&q->backing_dev_info->laptop_mode_wb_timer);
366 blk_mq_exit_queue(q);
369 * In theory, request pool of sched_tags belongs to request queue.
370 * However, the current implementation requires tag_set for freeing
371 * requests, so free the pool now.
373 * Queue has become frozen, there can't be any in-queue requests, so
374 * it is safe to free requests now.
376 mutex_lock(&q->sysfs_lock);
378 blk_mq_sched_free_requests(q);
379 mutex_unlock(&q->sysfs_lock);
381 percpu_ref_exit(&q->q_usage_counter);
383 /* @q is and will stay empty, shutdown and put */
386 EXPORT_SYMBOL(blk_cleanup_queue);
388 struct request_queue *blk_alloc_queue(gfp_t gfp_mask)
390 return blk_alloc_queue_node(gfp_mask, NUMA_NO_NODE);
392 EXPORT_SYMBOL(blk_alloc_queue);
395 * blk_queue_enter() - try to increase q->q_usage_counter
396 * @q: request queue pointer
397 * @flags: BLK_MQ_REQ_NOWAIT and/or BLK_MQ_REQ_PREEMPT
399 int blk_queue_enter(struct request_queue *q, blk_mq_req_flags_t flags)
401 const bool pm = flags & BLK_MQ_REQ_PREEMPT;
404 bool success = false;
407 if (percpu_ref_tryget_live(&q->q_usage_counter)) {
409 * The code that increments the pm_only counter is
410 * responsible for ensuring that that counter is
411 * globally visible before the queue is unfrozen.
413 if (pm || !blk_queue_pm_only(q)) {
416 percpu_ref_put(&q->q_usage_counter);
424 if (flags & BLK_MQ_REQ_NOWAIT)
428 * read pair of barrier in blk_freeze_queue_start(),
429 * we need to order reading __PERCPU_REF_DEAD flag of
430 * .q_usage_counter and reading .mq_freeze_depth or
431 * queue dying flag, otherwise the following wait may
432 * never return if the two reads are reordered.
436 wait_event(q->mq_freeze_wq,
437 (!q->mq_freeze_depth &&
438 (pm || (blk_pm_request_resume(q),
439 !blk_queue_pm_only(q)))) ||
441 if (blk_queue_dying(q))
446 void blk_queue_exit(struct request_queue *q)
448 percpu_ref_put(&q->q_usage_counter);
451 static void blk_queue_usage_counter_release(struct percpu_ref *ref)
453 struct request_queue *q =
454 container_of(ref, struct request_queue, q_usage_counter);
456 wake_up_all(&q->mq_freeze_wq);
459 static void blk_rq_timed_out_timer(struct timer_list *t)
461 struct request_queue *q = from_timer(q, t, timeout);
463 kblockd_schedule_work(&q->timeout_work);
466 static void blk_timeout_work(struct work_struct *work)
471 * blk_alloc_queue_node - allocate a request queue
472 * @gfp_mask: memory allocation flags
473 * @node_id: NUMA node to allocate memory from
475 struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id)
477 struct request_queue *q;
480 q = kmem_cache_alloc_node(blk_requestq_cachep,
481 gfp_mask | __GFP_ZERO, node_id);
485 q->last_merge = NULL;
487 q->id = ida_simple_get(&blk_queue_ida, 0, 0, gfp_mask);
491 ret = bioset_init(&q->bio_split, BIO_POOL_SIZE, 0, BIOSET_NEED_BVECS);
495 q->backing_dev_info = bdi_alloc_node(gfp_mask, node_id);
496 if (!q->backing_dev_info)
499 q->stats = blk_alloc_queue_stats();
503 q->backing_dev_info->ra_pages = VM_READAHEAD_PAGES;
504 q->backing_dev_info->capabilities = BDI_CAP_CGROUP_WRITEBACK;
505 q->backing_dev_info->name = "block";
508 timer_setup(&q->backing_dev_info->laptop_mode_wb_timer,
509 laptop_mode_timer_fn, 0);
510 timer_setup(&q->timeout, blk_rq_timed_out_timer, 0);
511 INIT_WORK(&q->timeout_work, blk_timeout_work);
512 INIT_LIST_HEAD(&q->icq_list);
513 #ifdef CONFIG_BLK_CGROUP
514 INIT_LIST_HEAD(&q->blkg_list);
517 kobject_init(&q->kobj, &blk_queue_ktype);
519 #ifdef CONFIG_BLK_DEV_IO_TRACE
520 mutex_init(&q->blk_trace_mutex);
522 mutex_init(&q->sysfs_lock);
523 mutex_init(&q->sysfs_dir_lock);
524 spin_lock_init(&q->queue_lock);
526 init_waitqueue_head(&q->mq_freeze_wq);
527 mutex_init(&q->mq_freeze_lock);
530 * Init percpu_ref in atomic mode so that it's faster to shutdown.
531 * See blk_register_queue() for details.
533 if (percpu_ref_init(&q->q_usage_counter,
534 blk_queue_usage_counter_release,
535 PERCPU_REF_INIT_ATOMIC, GFP_KERNEL))
538 if (blkcg_init_queue(q))
544 percpu_ref_exit(&q->q_usage_counter);
546 blk_free_queue_stats(q->stats);
548 bdi_put(q->backing_dev_info);
550 bioset_exit(&q->bio_split);
552 ida_simple_remove(&blk_queue_ida, q->id);
554 kmem_cache_free(blk_requestq_cachep, q);
557 EXPORT_SYMBOL(blk_alloc_queue_node);
559 bool blk_get_queue(struct request_queue *q)
561 if (likely(!blk_queue_dying(q))) {
568 EXPORT_SYMBOL(blk_get_queue);
571 * blk_get_request - allocate a request
572 * @q: request queue to allocate a request for
573 * @op: operation (REQ_OP_*) and REQ_* flags, e.g. REQ_SYNC.
574 * @flags: BLK_MQ_REQ_* flags, e.g. BLK_MQ_REQ_NOWAIT.
576 struct request *blk_get_request(struct request_queue *q, unsigned int op,
577 blk_mq_req_flags_t flags)
581 WARN_ON_ONCE(op & REQ_NOWAIT);
582 WARN_ON_ONCE(flags & ~(BLK_MQ_REQ_NOWAIT | BLK_MQ_REQ_PREEMPT));
584 req = blk_mq_alloc_request(q, op, flags);
585 if (!IS_ERR(req) && q->mq_ops->initialize_rq_fn)
586 q->mq_ops->initialize_rq_fn(req);
590 EXPORT_SYMBOL(blk_get_request);
592 void blk_put_request(struct request *req)
594 blk_mq_free_request(req);
596 EXPORT_SYMBOL(blk_put_request);
598 bool bio_attempt_back_merge(struct request *req, struct bio *bio,
599 unsigned int nr_segs)
601 const int ff = bio->bi_opf & REQ_FAILFAST_MASK;
603 if (!ll_back_merge_fn(req, bio, nr_segs))
606 trace_block_bio_backmerge(req->q, req, bio);
608 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
609 blk_rq_set_mixed_merge(req);
611 req->biotail->bi_next = bio;
613 req->__data_len += bio->bi_iter.bi_size;
615 blk_account_io_start(req, false);
619 bool bio_attempt_front_merge(struct request *req, struct bio *bio,
620 unsigned int nr_segs)
622 const int ff = bio->bi_opf & REQ_FAILFAST_MASK;
624 if (!ll_front_merge_fn(req, bio, nr_segs))
627 trace_block_bio_frontmerge(req->q, req, bio);
629 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
630 blk_rq_set_mixed_merge(req);
632 bio->bi_next = req->bio;
635 req->__sector = bio->bi_iter.bi_sector;
636 req->__data_len += bio->bi_iter.bi_size;
638 blk_account_io_start(req, false);
642 bool bio_attempt_discard_merge(struct request_queue *q, struct request *req,
645 unsigned short segments = blk_rq_nr_discard_segments(req);
647 if (segments >= queue_max_discard_segments(q))
649 if (blk_rq_sectors(req) + bio_sectors(bio) >
650 blk_rq_get_max_sectors(req, blk_rq_pos(req)))
653 req->biotail->bi_next = bio;
655 req->__data_len += bio->bi_iter.bi_size;
656 req->nr_phys_segments = segments + 1;
658 blk_account_io_start(req, false);
661 req_set_nomerge(q, req);
666 * blk_attempt_plug_merge - try to merge with %current's plugged list
667 * @q: request_queue new bio is being queued at
668 * @bio: new bio being queued
669 * @nr_segs: number of segments in @bio
670 * @same_queue_rq: pointer to &struct request that gets filled in when
671 * another request associated with @q is found on the plug list
672 * (optional, may be %NULL)
674 * Determine whether @bio being queued on @q can be merged with a request
675 * on %current's plugged list. Returns %true if merge was successful,
678 * Plugging coalesces IOs from the same issuer for the same purpose without
679 * going through @q->queue_lock. As such it's more of an issuing mechanism
680 * than scheduling, and the request, while may have elvpriv data, is not
681 * added on the elevator at this point. In addition, we don't have
682 * reliable access to the elevator outside queue lock. Only check basic
683 * merging parameters without querying the elevator.
685 * Caller must ensure !blk_queue_nomerges(q) beforehand.
687 bool blk_attempt_plug_merge(struct request_queue *q, struct bio *bio,
688 unsigned int nr_segs, struct request **same_queue_rq)
690 struct blk_plug *plug;
692 struct list_head *plug_list;
694 plug = blk_mq_plug(q, bio);
698 plug_list = &plug->mq_list;
700 list_for_each_entry_reverse(rq, plug_list, queuelist) {
703 if (rq->q == q && same_queue_rq) {
705 * Only blk-mq multiple hardware queues case checks the
706 * rq in the same queue, there should be only one such
712 if (rq->q != q || !blk_rq_merge_ok(rq, bio))
715 switch (blk_try_merge(rq, bio)) {
716 case ELEVATOR_BACK_MERGE:
717 merged = bio_attempt_back_merge(rq, bio, nr_segs);
719 case ELEVATOR_FRONT_MERGE:
720 merged = bio_attempt_front_merge(rq, bio, nr_segs);
722 case ELEVATOR_DISCARD_MERGE:
723 merged = bio_attempt_discard_merge(q, rq, bio);
736 static void handle_bad_sector(struct bio *bio, sector_t maxsector)
738 char b[BDEVNAME_SIZE];
740 printk(KERN_INFO "attempt to access beyond end of device\n");
741 printk(KERN_INFO "%s: rw=%d, want=%Lu, limit=%Lu\n",
742 bio_devname(bio, b), bio->bi_opf,
743 (unsigned long long)bio_end_sector(bio),
744 (long long)maxsector);
747 #ifdef CONFIG_FAIL_MAKE_REQUEST
749 static DECLARE_FAULT_ATTR(fail_make_request);
751 static int __init setup_fail_make_request(char *str)
753 return setup_fault_attr(&fail_make_request, str);
755 __setup("fail_make_request=", setup_fail_make_request);
757 static bool should_fail_request(struct hd_struct *part, unsigned int bytes)
759 return part->make_it_fail && should_fail(&fail_make_request, bytes);
762 static int __init fail_make_request_debugfs(void)
764 struct dentry *dir = fault_create_debugfs_attr("fail_make_request",
765 NULL, &fail_make_request);
767 return PTR_ERR_OR_ZERO(dir);
770 late_initcall(fail_make_request_debugfs);
772 #else /* CONFIG_FAIL_MAKE_REQUEST */
774 static inline bool should_fail_request(struct hd_struct *part,
780 #endif /* CONFIG_FAIL_MAKE_REQUEST */
782 static inline bool bio_check_ro(struct bio *bio, struct hd_struct *part)
784 const int op = bio_op(bio);
786 if (part->policy && op_is_write(op)) {
787 char b[BDEVNAME_SIZE];
789 if (op_is_flush(bio->bi_opf) && !bio_sectors(bio))
793 "generic_make_request: Trying to write "
794 "to read-only block-device %s (partno %d)\n",
795 bio_devname(bio, b), part->partno);
796 /* Older lvm-tools actually trigger this */
803 static noinline int should_fail_bio(struct bio *bio)
805 if (should_fail_request(&bio->bi_disk->part0, bio->bi_iter.bi_size))
809 ALLOW_ERROR_INJECTION(should_fail_bio, ERRNO);
812 * Check whether this bio extends beyond the end of the device or partition.
813 * This may well happen - the kernel calls bread() without checking the size of
814 * the device, e.g., when mounting a file system.
816 static inline int bio_check_eod(struct bio *bio, sector_t maxsector)
818 unsigned int nr_sectors = bio_sectors(bio);
820 if (nr_sectors && maxsector &&
821 (nr_sectors > maxsector ||
822 bio->bi_iter.bi_sector > maxsector - nr_sectors)) {
823 handle_bad_sector(bio, maxsector);
830 * Remap block n of partition p to block n+start(p) of the disk.
832 static inline int blk_partition_remap(struct bio *bio)
838 p = __disk_get_part(bio->bi_disk, bio->bi_partno);
841 if (unlikely(should_fail_request(p, bio->bi_iter.bi_size)))
843 if (unlikely(bio_check_ro(bio, p)))
847 * Zone reset does not include bi_size so bio_sectors() is always 0.
848 * Include a test for the reset op code and perform the remap if needed.
850 if (bio_sectors(bio) || bio_op(bio) == REQ_OP_ZONE_RESET) {
851 if (bio_check_eod(bio, part_nr_sects_read(p)))
853 bio->bi_iter.bi_sector += p->start_sect;
854 trace_block_bio_remap(bio->bi_disk->queue, bio, part_devt(p),
855 bio->bi_iter.bi_sector - p->start_sect);
864 static noinline_for_stack bool
865 generic_make_request_checks(struct bio *bio)
867 struct request_queue *q;
868 int nr_sectors = bio_sectors(bio);
869 blk_status_t status = BLK_STS_IOERR;
870 char b[BDEVNAME_SIZE];
874 q = bio->bi_disk->queue;
877 "generic_make_request: Trying to access "
878 "nonexistent block-device %s (%Lu)\n",
879 bio_devname(bio, b), (long long)bio->bi_iter.bi_sector);
884 * For a REQ_NOWAIT based request, return -EOPNOTSUPP
885 * if queue is not a request based queue.
887 if ((bio->bi_opf & REQ_NOWAIT) && !queue_is_mq(q))
890 if (should_fail_bio(bio))
893 if (bio->bi_partno) {
894 if (unlikely(blk_partition_remap(bio)))
897 if (unlikely(bio_check_ro(bio, &bio->bi_disk->part0)))
899 if (unlikely(bio_check_eod(bio, get_capacity(bio->bi_disk))))
904 * Filter flush bio's early so that make_request based
905 * drivers without flush support don't have to worry
908 if (op_is_flush(bio->bi_opf) &&
909 !test_bit(QUEUE_FLAG_WC, &q->queue_flags)) {
910 bio->bi_opf &= ~(REQ_PREFLUSH | REQ_FUA);
917 if (!test_bit(QUEUE_FLAG_POLL, &q->queue_flags))
918 bio->bi_opf &= ~REQ_HIPRI;
920 switch (bio_op(bio)) {
922 if (!blk_queue_discard(q))
925 case REQ_OP_SECURE_ERASE:
926 if (!blk_queue_secure_erase(q))
929 case REQ_OP_WRITE_SAME:
930 if (!q->limits.max_write_same_sectors)
933 case REQ_OP_ZONE_RESET:
934 if (!blk_queue_is_zoned(q))
937 case REQ_OP_ZONE_RESET_ALL:
938 if (!blk_queue_is_zoned(q) || !blk_queue_zone_resetall(q))
941 case REQ_OP_WRITE_ZEROES:
942 if (!q->limits.max_write_zeroes_sectors)
950 * Various block parts want %current->io_context and lazy ioc
951 * allocation ends up trading a lot of pain for a small amount of
952 * memory. Just allocate it upfront. This may fail and block
953 * layer knows how to live with it.
955 create_io_context(GFP_ATOMIC, q->node);
957 if (!blkcg_bio_issue_check(q, bio))
960 if (!bio_flagged(bio, BIO_TRACE_COMPLETION)) {
961 trace_block_bio_queue(q, bio);
962 /* Now that enqueuing has been traced, we need to trace
963 * completion as well.
965 bio_set_flag(bio, BIO_TRACE_COMPLETION);
970 status = BLK_STS_NOTSUPP;
972 bio->bi_status = status;
978 * generic_make_request - hand a buffer to its device driver for I/O
979 * @bio: The bio describing the location in memory and on the device.
981 * generic_make_request() is used to make I/O requests of block
982 * devices. It is passed a &struct bio, which describes the I/O that needs
985 * generic_make_request() does not return any status. The
986 * success/failure status of the request, along with notification of
987 * completion, is delivered asynchronously through the bio->bi_end_io
988 * function described (one day) else where.
990 * The caller of generic_make_request must make sure that bi_io_vec
991 * are set to describe the memory buffer, and that bi_dev and bi_sector are
992 * set to describe the device address, and the
993 * bi_end_io and optionally bi_private are set to describe how
994 * completion notification should be signaled.
996 * generic_make_request and the drivers it calls may use bi_next if this
997 * bio happens to be merged with someone else, and may resubmit the bio to
998 * a lower device by calling into generic_make_request recursively, which
999 * means the bio should NOT be touched after the call to ->make_request_fn.
1001 blk_qc_t generic_make_request(struct bio *bio)
1004 * bio_list_on_stack[0] contains bios submitted by the current
1006 * bio_list_on_stack[1] contains bios that were submitted before
1007 * the current make_request_fn, but that haven't been processed
1010 struct bio_list bio_list_on_stack[2];
1011 blk_qc_t ret = BLK_QC_T_NONE;
1013 if (!generic_make_request_checks(bio))
1017 * We only want one ->make_request_fn to be active at a time, else
1018 * stack usage with stacked devices could be a problem. So use
1019 * current->bio_list to keep a list of requests submited by a
1020 * make_request_fn function. current->bio_list is also used as a
1021 * flag to say if generic_make_request is currently active in this
1022 * task or not. If it is NULL, then no make_request is active. If
1023 * it is non-NULL, then a make_request is active, and new requests
1024 * should be added at the tail
1026 if (current->bio_list) {
1027 bio_list_add(¤t->bio_list[0], bio);
1031 /* following loop may be a bit non-obvious, and so deserves some
1033 * Before entering the loop, bio->bi_next is NULL (as all callers
1034 * ensure that) so we have a list with a single bio.
1035 * We pretend that we have just taken it off a longer list, so
1036 * we assign bio_list to a pointer to the bio_list_on_stack,
1037 * thus initialising the bio_list of new bios to be
1038 * added. ->make_request() may indeed add some more bios
1039 * through a recursive call to generic_make_request. If it
1040 * did, we find a non-NULL value in bio_list and re-enter the loop
1041 * from the top. In this case we really did just take the bio
1042 * of the top of the list (no pretending) and so remove it from
1043 * bio_list, and call into ->make_request() again.
1045 BUG_ON(bio->bi_next);
1046 bio_list_init(&bio_list_on_stack[0]);
1047 current->bio_list = bio_list_on_stack;
1049 struct request_queue *q = bio->bi_disk->queue;
1050 blk_mq_req_flags_t flags = bio->bi_opf & REQ_NOWAIT ?
1051 BLK_MQ_REQ_NOWAIT : 0;
1053 if (likely(blk_queue_enter(q, flags) == 0)) {
1054 struct bio_list lower, same;
1056 /* Create a fresh bio_list for all subordinate requests */
1057 bio_list_on_stack[1] = bio_list_on_stack[0];
1058 bio_list_init(&bio_list_on_stack[0]);
1059 ret = q->make_request_fn(q, bio);
1063 /* sort new bios into those for a lower level
1064 * and those for the same level
1066 bio_list_init(&lower);
1067 bio_list_init(&same);
1068 while ((bio = bio_list_pop(&bio_list_on_stack[0])) != NULL)
1069 if (q == bio->bi_disk->queue)
1070 bio_list_add(&same, bio);
1072 bio_list_add(&lower, bio);
1073 /* now assemble so we handle the lowest level first */
1074 bio_list_merge(&bio_list_on_stack[0], &lower);
1075 bio_list_merge(&bio_list_on_stack[0], &same);
1076 bio_list_merge(&bio_list_on_stack[0], &bio_list_on_stack[1]);
1078 if (unlikely(!blk_queue_dying(q) &&
1079 (bio->bi_opf & REQ_NOWAIT)))
1080 bio_wouldblock_error(bio);
1084 bio = bio_list_pop(&bio_list_on_stack[0]);
1086 current->bio_list = NULL; /* deactivate */
1091 EXPORT_SYMBOL(generic_make_request);
1094 * direct_make_request - hand a buffer directly to its device driver for I/O
1095 * @bio: The bio describing the location in memory and on the device.
1097 * This function behaves like generic_make_request(), but does not protect
1098 * against recursion. Must only be used if the called driver is known
1099 * to not call generic_make_request (or direct_make_request) again from
1100 * its make_request function. (Calling direct_make_request again from
1101 * a workqueue is perfectly fine as that doesn't recurse).
1103 blk_qc_t direct_make_request(struct bio *bio)
1105 struct request_queue *q = bio->bi_disk->queue;
1106 bool nowait = bio->bi_opf & REQ_NOWAIT;
1109 if (!generic_make_request_checks(bio))
1110 return BLK_QC_T_NONE;
1112 if (unlikely(blk_queue_enter(q, nowait ? BLK_MQ_REQ_NOWAIT : 0))) {
1113 if (nowait && !blk_queue_dying(q))
1114 bio->bi_status = BLK_STS_AGAIN;
1116 bio->bi_status = BLK_STS_IOERR;
1118 return BLK_QC_T_NONE;
1121 ret = q->make_request_fn(q, bio);
1125 EXPORT_SYMBOL_GPL(direct_make_request);
1128 * submit_bio - submit a bio to the block device layer for I/O
1129 * @bio: The &struct bio which describes the I/O
1131 * submit_bio() is very similar in purpose to generic_make_request(), and
1132 * uses that function to do most of the work. Both are fairly rough
1133 * interfaces; @bio must be presetup and ready for I/O.
1136 blk_qc_t submit_bio(struct bio *bio)
1138 bool workingset_read = false;
1139 unsigned long pflags;
1142 if (blkcg_punt_bio_submit(bio))
1143 return BLK_QC_T_NONE;
1146 * If it's a regular read/write or a barrier with data attached,
1147 * go through the normal accounting stuff before submission.
1149 if (bio_has_data(bio)) {
1152 if (unlikely(bio_op(bio) == REQ_OP_WRITE_SAME))
1153 count = queue_logical_block_size(bio->bi_disk->queue) >> 9;
1155 count = bio_sectors(bio);
1157 if (op_is_write(bio_op(bio))) {
1158 count_vm_events(PGPGOUT, count);
1160 if (bio_flagged(bio, BIO_WORKINGSET))
1161 workingset_read = true;
1162 task_io_account_read(bio->bi_iter.bi_size);
1163 count_vm_events(PGPGIN, count);
1166 if (unlikely(block_dump)) {
1167 char b[BDEVNAME_SIZE];
1168 printk(KERN_DEBUG "%s(%d): %s block %Lu on %s (%u sectors)\n",
1169 current->comm, task_pid_nr(current),
1170 op_is_write(bio_op(bio)) ? "WRITE" : "READ",
1171 (unsigned long long)bio->bi_iter.bi_sector,
1172 bio_devname(bio, b), count);
1177 * If we're reading data that is part of the userspace
1178 * workingset, count submission time as memory stall. When the
1179 * device is congested, or the submitting cgroup IO-throttled,
1180 * submission can be a significant part of overall IO time.
1182 if (workingset_read)
1183 psi_memstall_enter(&pflags);
1185 ret = generic_make_request(bio);
1187 if (workingset_read)
1188 psi_memstall_leave(&pflags);
1192 EXPORT_SYMBOL(submit_bio);
1195 * blk_cloned_rq_check_limits - Helper function to check a cloned request
1196 * for new the queue limits
1198 * @rq: the request being checked
1201 * @rq may have been made based on weaker limitations of upper-level queues
1202 * in request stacking drivers, and it may violate the limitation of @q.
1203 * Since the block layer and the underlying device driver trust @rq
1204 * after it is inserted to @q, it should be checked against @q before
1205 * the insertion using this generic function.
1207 * Request stacking drivers like request-based dm may change the queue
1208 * limits when retrying requests on other queues. Those requests need
1209 * to be checked against the new queue limits again during dispatch.
1211 static int blk_cloned_rq_check_limits(struct request_queue *q,
1214 if (blk_rq_sectors(rq) > blk_queue_get_max_sectors(q, req_op(rq))) {
1215 printk(KERN_ERR "%s: over max size limit. (%u > %u)\n",
1216 __func__, blk_rq_sectors(rq),
1217 blk_queue_get_max_sectors(q, req_op(rq)));
1222 * queue's settings related to segment counting like q->bounce_pfn
1223 * may differ from that of other stacking queues.
1224 * Recalculate it to check the request correctly on this queue's
1227 rq->nr_phys_segments = blk_recalc_rq_segments(rq);
1228 if (rq->nr_phys_segments > queue_max_segments(q)) {
1229 printk(KERN_ERR "%s: over max segments limit. (%hu > %hu)\n",
1230 __func__, rq->nr_phys_segments, queue_max_segments(q));
1238 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
1239 * @q: the queue to submit the request
1240 * @rq: the request being queued
1242 blk_status_t blk_insert_cloned_request(struct request_queue *q, struct request *rq)
1244 if (blk_cloned_rq_check_limits(q, rq))
1245 return BLK_STS_IOERR;
1248 should_fail_request(&rq->rq_disk->part0, blk_rq_bytes(rq)))
1249 return BLK_STS_IOERR;
1251 if (blk_queue_io_stat(q))
1252 blk_account_io_start(rq, true);
1255 * Since we have a scheduler attached on the top device,
1256 * bypass a potential scheduler on the bottom device for
1259 return blk_mq_request_issue_directly(rq, true);
1261 EXPORT_SYMBOL_GPL(blk_insert_cloned_request);
1264 * blk_rq_err_bytes - determine number of bytes till the next failure boundary
1265 * @rq: request to examine
1268 * A request could be merge of IOs which require different failure
1269 * handling. This function determines the number of bytes which
1270 * can be failed from the beginning of the request without
1271 * crossing into area which need to be retried further.
1274 * The number of bytes to fail.
1276 unsigned int blk_rq_err_bytes(const struct request *rq)
1278 unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK;
1279 unsigned int bytes = 0;
1282 if (!(rq->rq_flags & RQF_MIXED_MERGE))
1283 return blk_rq_bytes(rq);
1286 * Currently the only 'mixing' which can happen is between
1287 * different fastfail types. We can safely fail portions
1288 * which have all the failfast bits that the first one has -
1289 * the ones which are at least as eager to fail as the first
1292 for (bio = rq->bio; bio; bio = bio->bi_next) {
1293 if ((bio->bi_opf & ff) != ff)
1295 bytes += bio->bi_iter.bi_size;
1298 /* this could lead to infinite loop */
1299 BUG_ON(blk_rq_bytes(rq) && !bytes);
1302 EXPORT_SYMBOL_GPL(blk_rq_err_bytes);
1304 void blk_account_io_completion(struct request *req, unsigned int bytes)
1306 if (blk_do_io_stat(req)) {
1307 const int sgrp = op_stat_group(req_op(req));
1308 struct hd_struct *part;
1312 part_stat_add(part, sectors[sgrp], bytes >> 9);
1317 void blk_account_io_done(struct request *req, u64 now)
1320 * Account IO completion. flush_rq isn't accounted as a
1321 * normal IO on queueing nor completion. Accounting the
1322 * containing request is enough.
1324 if (blk_do_io_stat(req) && !(req->rq_flags & RQF_FLUSH_SEQ)) {
1325 const int sgrp = op_stat_group(req_op(req));
1326 struct hd_struct *part;
1331 update_io_ticks(part, jiffies);
1332 part_stat_inc(part, ios[sgrp]);
1333 part_stat_add(part, nsecs[sgrp], now - req->start_time_ns);
1334 part_stat_add(part, time_in_queue, nsecs_to_jiffies64(now - req->start_time_ns));
1335 part_dec_in_flight(req->q, part, rq_data_dir(req));
1337 hd_struct_put(part);
1342 void blk_account_io_start(struct request *rq, bool new_io)
1344 struct hd_struct *part;
1345 int rw = rq_data_dir(rq);
1347 if (!blk_do_io_stat(rq))
1354 part_stat_inc(part, merges[rw]);
1356 part = disk_map_sector_rcu(rq->rq_disk, blk_rq_pos(rq));
1357 if (!hd_struct_try_get(part)) {
1359 * The partition is already being removed,
1360 * the request will be accounted on the disk only
1362 * We take a reference on disk->part0 although that
1363 * partition will never be deleted, so we can treat
1364 * it as any other partition.
1366 part = &rq->rq_disk->part0;
1367 hd_struct_get(part);
1369 part_inc_in_flight(rq->q, part, rw);
1373 update_io_ticks(part, jiffies);
1379 * Steal bios from a request and add them to a bio list.
1380 * The request must not have been partially completed before.
1382 void blk_steal_bios(struct bio_list *list, struct request *rq)
1386 list->tail->bi_next = rq->bio;
1388 list->head = rq->bio;
1389 list->tail = rq->biotail;
1397 EXPORT_SYMBOL_GPL(blk_steal_bios);
1400 * blk_update_request - Special helper function for request stacking drivers
1401 * @req: the request being processed
1402 * @error: block status code
1403 * @nr_bytes: number of bytes to complete @req
1406 * Ends I/O on a number of bytes attached to @req, but doesn't complete
1407 * the request structure even if @req doesn't have leftover.
1408 * If @req has leftover, sets it up for the next range of segments.
1410 * This special helper function is only for request stacking drivers
1411 * (e.g. request-based dm) so that they can handle partial completion.
1412 * Actual device drivers should use blk_mq_end_request instead.
1414 * Passing the result of blk_rq_bytes() as @nr_bytes guarantees
1415 * %false return from this function.
1418 * The RQF_SPECIAL_PAYLOAD flag is ignored on purpose in both
1419 * blk_rq_bytes() and in blk_update_request().
1422 * %false - this request doesn't have any more data
1423 * %true - this request has more data
1425 bool blk_update_request(struct request *req, blk_status_t error,
1426 unsigned int nr_bytes)
1430 trace_block_rq_complete(req, blk_status_to_errno(error), nr_bytes);
1435 if (unlikely(error && !blk_rq_is_passthrough(req) &&
1436 !(req->rq_flags & RQF_QUIET)))
1437 print_req_error(req, error, __func__);
1439 blk_account_io_completion(req, nr_bytes);
1443 struct bio *bio = req->bio;
1444 unsigned bio_bytes = min(bio->bi_iter.bi_size, nr_bytes);
1446 if (bio_bytes == bio->bi_iter.bi_size)
1447 req->bio = bio->bi_next;
1449 /* Completion has already been traced */
1450 bio_clear_flag(bio, BIO_TRACE_COMPLETION);
1451 req_bio_endio(req, bio, bio_bytes, error);
1453 total_bytes += bio_bytes;
1454 nr_bytes -= bio_bytes;
1465 * Reset counters so that the request stacking driver
1466 * can find how many bytes remain in the request
1469 req->__data_len = 0;
1473 req->__data_len -= total_bytes;
1475 /* update sector only for requests with clear definition of sector */
1476 if (!blk_rq_is_passthrough(req))
1477 req->__sector += total_bytes >> 9;
1479 /* mixed attributes always follow the first bio */
1480 if (req->rq_flags & RQF_MIXED_MERGE) {
1481 req->cmd_flags &= ~REQ_FAILFAST_MASK;
1482 req->cmd_flags |= req->bio->bi_opf & REQ_FAILFAST_MASK;
1485 if (!(req->rq_flags & RQF_SPECIAL_PAYLOAD)) {
1487 * If total number of sectors is less than the first segment
1488 * size, something has gone terribly wrong.
1490 if (blk_rq_bytes(req) < blk_rq_cur_bytes(req)) {
1491 blk_dump_rq_flags(req, "request botched");
1492 req->__data_len = blk_rq_cur_bytes(req);
1495 /* recalculate the number of segments */
1496 req->nr_phys_segments = blk_recalc_rq_segments(req);
1501 EXPORT_SYMBOL_GPL(blk_update_request);
1503 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
1505 * rq_flush_dcache_pages - Helper function to flush all pages in a request
1506 * @rq: the request to be flushed
1509 * Flush all pages in @rq.
1511 void rq_flush_dcache_pages(struct request *rq)
1513 struct req_iterator iter;
1514 struct bio_vec bvec;
1516 rq_for_each_segment(bvec, rq, iter)
1517 flush_dcache_page(bvec.bv_page);
1519 EXPORT_SYMBOL_GPL(rq_flush_dcache_pages);
1523 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
1524 * @q : the queue of the device being checked
1527 * Check if underlying low-level drivers of a device are busy.
1528 * If the drivers want to export their busy state, they must set own
1529 * exporting function using blk_queue_lld_busy() first.
1531 * Basically, this function is used only by request stacking drivers
1532 * to stop dispatching requests to underlying devices when underlying
1533 * devices are busy. This behavior helps more I/O merging on the queue
1534 * of the request stacking driver and prevents I/O throughput regression
1535 * on burst I/O load.
1538 * 0 - Not busy (The request stacking driver should dispatch request)
1539 * 1 - Busy (The request stacking driver should stop dispatching request)
1541 int blk_lld_busy(struct request_queue *q)
1543 if (queue_is_mq(q) && q->mq_ops->busy)
1544 return q->mq_ops->busy(q);
1548 EXPORT_SYMBOL_GPL(blk_lld_busy);
1551 * blk_rq_unprep_clone - Helper function to free all bios in a cloned request
1552 * @rq: the clone request to be cleaned up
1555 * Free all bios in @rq for a cloned request.
1557 void blk_rq_unprep_clone(struct request *rq)
1561 while ((bio = rq->bio) != NULL) {
1562 rq->bio = bio->bi_next;
1567 EXPORT_SYMBOL_GPL(blk_rq_unprep_clone);
1570 * Copy attributes of the original request to the clone request.
1571 * The actual data parts (e.g. ->cmd, ->sense) are not copied.
1573 static void __blk_rq_prep_clone(struct request *dst, struct request *src)
1575 dst->__sector = blk_rq_pos(src);
1576 dst->__data_len = blk_rq_bytes(src);
1577 if (src->rq_flags & RQF_SPECIAL_PAYLOAD) {
1578 dst->rq_flags |= RQF_SPECIAL_PAYLOAD;
1579 dst->special_vec = src->special_vec;
1581 dst->nr_phys_segments = src->nr_phys_segments;
1582 dst->ioprio = src->ioprio;
1583 dst->extra_len = src->extra_len;
1587 * blk_rq_prep_clone - Helper function to setup clone request
1588 * @rq: the request to be setup
1589 * @rq_src: original request to be cloned
1590 * @bs: bio_set that bios for clone are allocated from
1591 * @gfp_mask: memory allocation mask for bio
1592 * @bio_ctr: setup function to be called for each clone bio.
1593 * Returns %0 for success, non %0 for failure.
1594 * @data: private data to be passed to @bio_ctr
1597 * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq.
1598 * The actual data parts of @rq_src (e.g. ->cmd, ->sense)
1599 * are not copied, and copying such parts is the caller's responsibility.
1600 * Also, pages which the original bios are pointing to are not copied
1601 * and the cloned bios just point same pages.
1602 * So cloned bios must be completed before original bios, which means
1603 * the caller must complete @rq before @rq_src.
1605 int blk_rq_prep_clone(struct request *rq, struct request *rq_src,
1606 struct bio_set *bs, gfp_t gfp_mask,
1607 int (*bio_ctr)(struct bio *, struct bio *, void *),
1610 struct bio *bio, *bio_src;
1615 __rq_for_each_bio(bio_src, rq_src) {
1616 bio = bio_clone_fast(bio_src, gfp_mask, bs);
1620 if (bio_ctr && bio_ctr(bio, bio_src, data))
1624 rq->biotail->bi_next = bio;
1627 rq->bio = rq->biotail = bio;
1630 __blk_rq_prep_clone(rq, rq_src);
1637 blk_rq_unprep_clone(rq);
1641 EXPORT_SYMBOL_GPL(blk_rq_prep_clone);
1643 int kblockd_schedule_work(struct work_struct *work)
1645 return queue_work(kblockd_workqueue, work);
1647 EXPORT_SYMBOL(kblockd_schedule_work);
1649 int kblockd_schedule_work_on(int cpu, struct work_struct *work)
1651 return queue_work_on(cpu, kblockd_workqueue, work);
1653 EXPORT_SYMBOL(kblockd_schedule_work_on);
1655 int kblockd_mod_delayed_work_on(int cpu, struct delayed_work *dwork,
1656 unsigned long delay)
1658 return mod_delayed_work_on(cpu, kblockd_workqueue, dwork, delay);
1660 EXPORT_SYMBOL(kblockd_mod_delayed_work_on);
1663 * blk_start_plug - initialize blk_plug and track it inside the task_struct
1664 * @plug: The &struct blk_plug that needs to be initialized
1667 * blk_start_plug() indicates to the block layer an intent by the caller
1668 * to submit multiple I/O requests in a batch. The block layer may use
1669 * this hint to defer submitting I/Os from the caller until blk_finish_plug()
1670 * is called. However, the block layer may choose to submit requests
1671 * before a call to blk_finish_plug() if the number of queued I/Os
1672 * exceeds %BLK_MAX_REQUEST_COUNT, or if the size of the I/O is larger than
1673 * %BLK_PLUG_FLUSH_SIZE. The queued I/Os may also be submitted early if
1674 * the task schedules (see below).
1676 * Tracking blk_plug inside the task_struct will help with auto-flushing the
1677 * pending I/O should the task end up blocking between blk_start_plug() and
1678 * blk_finish_plug(). This is important from a performance perspective, but
1679 * also ensures that we don't deadlock. For instance, if the task is blocking
1680 * for a memory allocation, memory reclaim could end up wanting to free a
1681 * page belonging to that request that is currently residing in our private
1682 * plug. By flushing the pending I/O when the process goes to sleep, we avoid
1683 * this kind of deadlock.
1685 void blk_start_plug(struct blk_plug *plug)
1687 struct task_struct *tsk = current;
1690 * If this is a nested plug, don't actually assign it.
1695 INIT_LIST_HEAD(&plug->mq_list);
1696 INIT_LIST_HEAD(&plug->cb_list);
1698 plug->multiple_queues = 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 int __init blk_dev_init(void)
1780 BUILD_BUG_ON(REQ_OP_LAST >= (1 << REQ_OP_BITS));
1781 BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 *
1782 FIELD_SIZEOF(struct request, cmd_flags));
1783 BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 *
1784 FIELD_SIZEOF(struct bio, bi_opf));
1786 /* used for unplugging and affects IO latency/throughput - HIGHPRI */
1787 kblockd_workqueue = alloc_workqueue("kblockd",
1788 WQ_MEM_RECLAIM | WQ_HIGHPRI, 0);
1789 if (!kblockd_workqueue)
1790 panic("Failed to create kblockd\n");
1792 blk_requestq_cachep = kmem_cache_create("request_queue",
1793 sizeof(struct request_queue), 0, SLAB_PANIC, NULL);
1795 #ifdef CONFIG_DEBUG_FS
1796 blk_debugfs_root = debugfs_create_dir("block", NULL);