Loading...
1// SPDX-License-Identifier: GPL-2.0
2/*
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>
8 * - July2000
9 * bio rewrite, highmem i/o, etc, Jens Axboe <axboe@suse.de> - may 2001
10 */
11
12/*
13 * This handles all read/write requests to block devices
14 */
15#include <linux/kernel.h>
16#include <linux/module.h>
17#include <linux/bio.h>
18#include <linux/blkdev.h>
19#include <linux/blk-pm.h>
20#include <linux/blk-integrity.h>
21#include <linux/highmem.h>
22#include <linux/mm.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/t10-pi.h>
38#include <linux/debugfs.h>
39#include <linux/bpf.h>
40#include <linux/part_stat.h>
41#include <linux/sched/sysctl.h>
42#include <linux/blk-crypto.h>
43
44#define CREATE_TRACE_POINTS
45#include <trace/events/block.h>
46
47#include "blk.h"
48#include "blk-mq-sched.h"
49#include "blk-pm.h"
50#include "blk-cgroup.h"
51#include "blk-throttle.h"
52
53struct dentry *blk_debugfs_root;
54
55EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_remap);
56EXPORT_TRACEPOINT_SYMBOL_GPL(block_rq_remap);
57EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_complete);
58EXPORT_TRACEPOINT_SYMBOL_GPL(block_split);
59EXPORT_TRACEPOINT_SYMBOL_GPL(block_unplug);
60EXPORT_TRACEPOINT_SYMBOL_GPL(block_rq_insert);
61
62static DEFINE_IDA(blk_queue_ida);
63
64/*
65 * For queue allocation
66 */
67static struct kmem_cache *blk_requestq_cachep;
68
69/*
70 * Controlling structure to kblockd
71 */
72static struct workqueue_struct *kblockd_workqueue;
73
74/**
75 * blk_queue_flag_set - atomically set a queue flag
76 * @flag: flag to be set
77 * @q: request queue
78 */
79void blk_queue_flag_set(unsigned int flag, struct request_queue *q)
80{
81 set_bit(flag, &q->queue_flags);
82}
83EXPORT_SYMBOL(blk_queue_flag_set);
84
85/**
86 * blk_queue_flag_clear - atomically clear a queue flag
87 * @flag: flag to be cleared
88 * @q: request queue
89 */
90void blk_queue_flag_clear(unsigned int flag, struct request_queue *q)
91{
92 clear_bit(flag, &q->queue_flags);
93}
94EXPORT_SYMBOL(blk_queue_flag_clear);
95
96/**
97 * blk_queue_flag_test_and_set - atomically test and set a queue flag
98 * @flag: flag to be set
99 * @q: request queue
100 *
101 * Returns the previous value of @flag - 0 if the flag was not set and 1 if
102 * the flag was already set.
103 */
104bool blk_queue_flag_test_and_set(unsigned int flag, struct request_queue *q)
105{
106 return test_and_set_bit(flag, &q->queue_flags);
107}
108EXPORT_SYMBOL_GPL(blk_queue_flag_test_and_set);
109
110#define REQ_OP_NAME(name) [REQ_OP_##name] = #name
111static const char *const blk_op_name[] = {
112 REQ_OP_NAME(READ),
113 REQ_OP_NAME(WRITE),
114 REQ_OP_NAME(FLUSH),
115 REQ_OP_NAME(DISCARD),
116 REQ_OP_NAME(SECURE_ERASE),
117 REQ_OP_NAME(ZONE_RESET),
118 REQ_OP_NAME(ZONE_RESET_ALL),
119 REQ_OP_NAME(ZONE_OPEN),
120 REQ_OP_NAME(ZONE_CLOSE),
121 REQ_OP_NAME(ZONE_FINISH),
122 REQ_OP_NAME(ZONE_APPEND),
123 REQ_OP_NAME(WRITE_ZEROES),
124 REQ_OP_NAME(DRV_IN),
125 REQ_OP_NAME(DRV_OUT),
126};
127#undef REQ_OP_NAME
128
129/**
130 * blk_op_str - Return string XXX in the REQ_OP_XXX.
131 * @op: REQ_OP_XXX.
132 *
133 * Description: Centralize block layer function to convert REQ_OP_XXX into
134 * string format. Useful in the debugging and tracing bio or request. For
135 * invalid REQ_OP_XXX it returns string "UNKNOWN".
136 */
137inline const char *blk_op_str(enum req_op op)
138{
139 const char *op_str = "UNKNOWN";
140
141 if (op < ARRAY_SIZE(blk_op_name) && blk_op_name[op])
142 op_str = blk_op_name[op];
143
144 return op_str;
145}
146EXPORT_SYMBOL_GPL(blk_op_str);
147
148static const struct {
149 int errno;
150 const char *name;
151} blk_errors[] = {
152 [BLK_STS_OK] = { 0, "" },
153 [BLK_STS_NOTSUPP] = { -EOPNOTSUPP, "operation not supported" },
154 [BLK_STS_TIMEOUT] = { -ETIMEDOUT, "timeout" },
155 [BLK_STS_NOSPC] = { -ENOSPC, "critical space allocation" },
156 [BLK_STS_TRANSPORT] = { -ENOLINK, "recoverable transport" },
157 [BLK_STS_TARGET] = { -EREMOTEIO, "critical target" },
158 [BLK_STS_NEXUS] = { -EBADE, "critical nexus" },
159 [BLK_STS_MEDIUM] = { -ENODATA, "critical medium" },
160 [BLK_STS_PROTECTION] = { -EILSEQ, "protection" },
161 [BLK_STS_RESOURCE] = { -ENOMEM, "kernel resource" },
162 [BLK_STS_DEV_RESOURCE] = { -EBUSY, "device resource" },
163 [BLK_STS_AGAIN] = { -EAGAIN, "nonblocking retry" },
164 [BLK_STS_OFFLINE] = { -ENODEV, "device offline" },
165
166 /* device mapper special case, should not leak out: */
167 [BLK_STS_DM_REQUEUE] = { -EREMCHG, "dm internal retry" },
168
169 /* zone device specific errors */
170 [BLK_STS_ZONE_OPEN_RESOURCE] = { -ETOOMANYREFS, "open zones exceeded" },
171 [BLK_STS_ZONE_ACTIVE_RESOURCE] = { -EOVERFLOW, "active zones exceeded" },
172
173 /* everything else not covered above: */
174 [BLK_STS_IOERR] = { -EIO, "I/O" },
175};
176
177blk_status_t errno_to_blk_status(int errno)
178{
179 int i;
180
181 for (i = 0; i < ARRAY_SIZE(blk_errors); i++) {
182 if (blk_errors[i].errno == errno)
183 return (__force blk_status_t)i;
184 }
185
186 return BLK_STS_IOERR;
187}
188EXPORT_SYMBOL_GPL(errno_to_blk_status);
189
190int blk_status_to_errno(blk_status_t status)
191{
192 int idx = (__force int)status;
193
194 if (WARN_ON_ONCE(idx >= ARRAY_SIZE(blk_errors)))
195 return -EIO;
196 return blk_errors[idx].errno;
197}
198EXPORT_SYMBOL_GPL(blk_status_to_errno);
199
200const char *blk_status_to_str(blk_status_t status)
201{
202 int idx = (__force int)status;
203
204 if (WARN_ON_ONCE(idx >= ARRAY_SIZE(blk_errors)))
205 return "<null>";
206 return blk_errors[idx].name;
207}
208
209/**
210 * blk_sync_queue - cancel any pending callbacks on a queue
211 * @q: the queue
212 *
213 * Description:
214 * The block layer may perform asynchronous callback activity
215 * on a queue, such as calling the unplug function after a timeout.
216 * A block device may call blk_sync_queue to ensure that any
217 * such activity is cancelled, thus allowing it to release resources
218 * that the callbacks might use. The caller must already have made sure
219 * that its ->submit_bio will not re-add plugging prior to calling
220 * this function.
221 *
222 * This function does not cancel any asynchronous activity arising
223 * out of elevator or throttling code. That would require elevator_exit()
224 * and blkcg_exit_queue() to be called with queue lock initialized.
225 *
226 */
227void blk_sync_queue(struct request_queue *q)
228{
229 del_timer_sync(&q->timeout);
230 cancel_work_sync(&q->timeout_work);
231}
232EXPORT_SYMBOL(blk_sync_queue);
233
234/**
235 * blk_set_pm_only - increment pm_only counter
236 * @q: request queue pointer
237 */
238void blk_set_pm_only(struct request_queue *q)
239{
240 atomic_inc(&q->pm_only);
241}
242EXPORT_SYMBOL_GPL(blk_set_pm_only);
243
244void blk_clear_pm_only(struct request_queue *q)
245{
246 int pm_only;
247
248 pm_only = atomic_dec_return(&q->pm_only);
249 WARN_ON_ONCE(pm_only < 0);
250 if (pm_only == 0)
251 wake_up_all(&q->mq_freeze_wq);
252}
253EXPORT_SYMBOL_GPL(blk_clear_pm_only);
254
255static void blk_free_queue_rcu(struct rcu_head *rcu_head)
256{
257 struct request_queue *q = container_of(rcu_head,
258 struct request_queue, rcu_head);
259
260 percpu_ref_exit(&q->q_usage_counter);
261 kmem_cache_free(blk_requestq_cachep, q);
262}
263
264static void blk_free_queue(struct request_queue *q)
265{
266 if (q->poll_stat)
267 blk_stat_remove_callback(q, q->poll_cb);
268 blk_stat_free_callback(q->poll_cb);
269
270 blk_free_queue_stats(q->stats);
271 kfree(q->poll_stat);
272
273 if (queue_is_mq(q))
274 blk_mq_release(q);
275
276 ida_free(&blk_queue_ida, q->id);
277 call_rcu(&q->rcu_head, blk_free_queue_rcu);
278}
279
280/**
281 * blk_put_queue - decrement the request_queue refcount
282 * @q: the request_queue structure to decrement the refcount for
283 *
284 * Decrements the refcount of the request_queue and free it when the refcount
285 * reaches 0.
286 */
287void blk_put_queue(struct request_queue *q)
288{
289 if (refcount_dec_and_test(&q->refs))
290 blk_free_queue(q);
291}
292EXPORT_SYMBOL(blk_put_queue);
293
294void blk_queue_start_drain(struct request_queue *q)
295{
296 /*
297 * When queue DYING flag is set, we need to block new req
298 * entering queue, so we call blk_freeze_queue_start() to
299 * prevent I/O from crossing blk_queue_enter().
300 */
301 blk_freeze_queue_start(q);
302 if (queue_is_mq(q))
303 blk_mq_wake_waiters(q);
304 /* Make blk_queue_enter() reexamine the DYING flag. */
305 wake_up_all(&q->mq_freeze_wq);
306}
307
308/**
309 * blk_queue_enter() - try to increase q->q_usage_counter
310 * @q: request queue pointer
311 * @flags: BLK_MQ_REQ_NOWAIT and/or BLK_MQ_REQ_PM
312 */
313int blk_queue_enter(struct request_queue *q, blk_mq_req_flags_t flags)
314{
315 const bool pm = flags & BLK_MQ_REQ_PM;
316
317 while (!blk_try_enter_queue(q, pm)) {
318 if (flags & BLK_MQ_REQ_NOWAIT)
319 return -EAGAIN;
320
321 /*
322 * read pair of barrier in blk_freeze_queue_start(), we need to
323 * order reading __PERCPU_REF_DEAD flag of .q_usage_counter and
324 * reading .mq_freeze_depth or queue dying flag, otherwise the
325 * following wait may never return if the two reads are
326 * reordered.
327 */
328 smp_rmb();
329 wait_event(q->mq_freeze_wq,
330 (!q->mq_freeze_depth &&
331 blk_pm_resume_queue(pm, q)) ||
332 blk_queue_dying(q));
333 if (blk_queue_dying(q))
334 return -ENODEV;
335 }
336
337 return 0;
338}
339
340int __bio_queue_enter(struct request_queue *q, struct bio *bio)
341{
342 while (!blk_try_enter_queue(q, false)) {
343 struct gendisk *disk = bio->bi_bdev->bd_disk;
344
345 if (bio->bi_opf & REQ_NOWAIT) {
346 if (test_bit(GD_DEAD, &disk->state))
347 goto dead;
348 bio_wouldblock_error(bio);
349 return -EAGAIN;
350 }
351
352 /*
353 * read pair of barrier in blk_freeze_queue_start(), we need to
354 * order reading __PERCPU_REF_DEAD flag of .q_usage_counter and
355 * reading .mq_freeze_depth or queue dying flag, otherwise the
356 * following wait may never return if the two reads are
357 * reordered.
358 */
359 smp_rmb();
360 wait_event(q->mq_freeze_wq,
361 (!q->mq_freeze_depth &&
362 blk_pm_resume_queue(false, q)) ||
363 test_bit(GD_DEAD, &disk->state));
364 if (test_bit(GD_DEAD, &disk->state))
365 goto dead;
366 }
367
368 return 0;
369dead:
370 bio_io_error(bio);
371 return -ENODEV;
372}
373
374void blk_queue_exit(struct request_queue *q)
375{
376 percpu_ref_put(&q->q_usage_counter);
377}
378
379static void blk_queue_usage_counter_release(struct percpu_ref *ref)
380{
381 struct request_queue *q =
382 container_of(ref, struct request_queue, q_usage_counter);
383
384 wake_up_all(&q->mq_freeze_wq);
385}
386
387static void blk_rq_timed_out_timer(struct timer_list *t)
388{
389 struct request_queue *q = from_timer(q, t, timeout);
390
391 kblockd_schedule_work(&q->timeout_work);
392}
393
394static void blk_timeout_work(struct work_struct *work)
395{
396}
397
398struct request_queue *blk_alloc_queue(int node_id)
399{
400 struct request_queue *q;
401
402 q = kmem_cache_alloc_node(blk_requestq_cachep, GFP_KERNEL | __GFP_ZERO,
403 node_id);
404 if (!q)
405 return NULL;
406
407 q->last_merge = NULL;
408
409 q->id = ida_alloc(&blk_queue_ida, GFP_KERNEL);
410 if (q->id < 0)
411 goto fail_q;
412
413 q->stats = blk_alloc_queue_stats();
414 if (!q->stats)
415 goto fail_id;
416
417 q->node = node_id;
418
419 atomic_set(&q->nr_active_requests_shared_tags, 0);
420
421 timer_setup(&q->timeout, blk_rq_timed_out_timer, 0);
422 INIT_WORK(&q->timeout_work, blk_timeout_work);
423 INIT_LIST_HEAD(&q->icq_list);
424
425 refcount_set(&q->refs, 1);
426 mutex_init(&q->debugfs_mutex);
427 mutex_init(&q->sysfs_lock);
428 mutex_init(&q->sysfs_dir_lock);
429 spin_lock_init(&q->queue_lock);
430
431 init_waitqueue_head(&q->mq_freeze_wq);
432 mutex_init(&q->mq_freeze_lock);
433
434 /*
435 * Init percpu_ref in atomic mode so that it's faster to shutdown.
436 * See blk_register_queue() for details.
437 */
438 if (percpu_ref_init(&q->q_usage_counter,
439 blk_queue_usage_counter_release,
440 PERCPU_REF_INIT_ATOMIC, GFP_KERNEL))
441 goto fail_stats;
442
443 blk_set_default_limits(&q->limits);
444 q->nr_requests = BLKDEV_DEFAULT_RQ;
445
446 return q;
447
448fail_stats:
449 blk_free_queue_stats(q->stats);
450fail_id:
451 ida_free(&blk_queue_ida, q->id);
452fail_q:
453 kmem_cache_free(blk_requestq_cachep, q);
454 return NULL;
455}
456
457/**
458 * blk_get_queue - increment the request_queue refcount
459 * @q: the request_queue structure to increment the refcount for
460 *
461 * Increment the refcount of the request_queue kobject.
462 *
463 * Context: Any context.
464 */
465bool blk_get_queue(struct request_queue *q)
466{
467 if (unlikely(blk_queue_dying(q)))
468 return false;
469 refcount_inc(&q->refs);
470 return true;
471}
472EXPORT_SYMBOL(blk_get_queue);
473
474#ifdef CONFIG_FAIL_MAKE_REQUEST
475
476static DECLARE_FAULT_ATTR(fail_make_request);
477
478static int __init setup_fail_make_request(char *str)
479{
480 return setup_fault_attr(&fail_make_request, str);
481}
482__setup("fail_make_request=", setup_fail_make_request);
483
484bool should_fail_request(struct block_device *part, unsigned int bytes)
485{
486 return part->bd_make_it_fail && should_fail(&fail_make_request, bytes);
487}
488
489static int __init fail_make_request_debugfs(void)
490{
491 struct dentry *dir = fault_create_debugfs_attr("fail_make_request",
492 NULL, &fail_make_request);
493
494 return PTR_ERR_OR_ZERO(dir);
495}
496
497late_initcall(fail_make_request_debugfs);
498#endif /* CONFIG_FAIL_MAKE_REQUEST */
499
500static inline void bio_check_ro(struct bio *bio)
501{
502 if (op_is_write(bio_op(bio)) && bdev_read_only(bio->bi_bdev)) {
503 if (op_is_flush(bio->bi_opf) && !bio_sectors(bio))
504 return;
505 pr_warn("Trying to write to read-only block-device %pg\n",
506 bio->bi_bdev);
507 /* Older lvm-tools actually trigger this */
508 }
509}
510
511static noinline int should_fail_bio(struct bio *bio)
512{
513 if (should_fail_request(bdev_whole(bio->bi_bdev), bio->bi_iter.bi_size))
514 return -EIO;
515 return 0;
516}
517ALLOW_ERROR_INJECTION(should_fail_bio, ERRNO);
518
519/*
520 * Check whether this bio extends beyond the end of the device or partition.
521 * This may well happen - the kernel calls bread() without checking the size of
522 * the device, e.g., when mounting a file system.
523 */
524static inline int bio_check_eod(struct bio *bio)
525{
526 sector_t maxsector = bdev_nr_sectors(bio->bi_bdev);
527 unsigned int nr_sectors = bio_sectors(bio);
528
529 if (nr_sectors && maxsector &&
530 (nr_sectors > maxsector ||
531 bio->bi_iter.bi_sector > maxsector - nr_sectors)) {
532 pr_info_ratelimited("%s: attempt to access beyond end of device\n"
533 "%pg: rw=%d, sector=%llu, nr_sectors = %u limit=%llu\n",
534 current->comm, bio->bi_bdev, bio->bi_opf,
535 bio->bi_iter.bi_sector, nr_sectors, maxsector);
536 return -EIO;
537 }
538 return 0;
539}
540
541/*
542 * Remap block n of partition p to block n+start(p) of the disk.
543 */
544static int blk_partition_remap(struct bio *bio)
545{
546 struct block_device *p = bio->bi_bdev;
547
548 if (unlikely(should_fail_request(p, bio->bi_iter.bi_size)))
549 return -EIO;
550 if (bio_sectors(bio)) {
551 bio->bi_iter.bi_sector += p->bd_start_sect;
552 trace_block_bio_remap(bio, p->bd_dev,
553 bio->bi_iter.bi_sector -
554 p->bd_start_sect);
555 }
556 bio_set_flag(bio, BIO_REMAPPED);
557 return 0;
558}
559
560/*
561 * Check write append to a zoned block device.
562 */
563static inline blk_status_t blk_check_zone_append(struct request_queue *q,
564 struct bio *bio)
565{
566 int nr_sectors = bio_sectors(bio);
567
568 /* Only applicable to zoned block devices */
569 if (!bdev_is_zoned(bio->bi_bdev))
570 return BLK_STS_NOTSUPP;
571
572 /* The bio sector must point to the start of a sequential zone */
573 if (bio->bi_iter.bi_sector & (bdev_zone_sectors(bio->bi_bdev) - 1) ||
574 !bio_zone_is_seq(bio))
575 return BLK_STS_IOERR;
576
577 /*
578 * Not allowed to cross zone boundaries. Otherwise, the BIO will be
579 * split and could result in non-contiguous sectors being written in
580 * different zones.
581 */
582 if (nr_sectors > q->limits.chunk_sectors)
583 return BLK_STS_IOERR;
584
585 /* Make sure the BIO is small enough and will not get split */
586 if (nr_sectors > q->limits.max_zone_append_sectors)
587 return BLK_STS_IOERR;
588
589 bio->bi_opf |= REQ_NOMERGE;
590
591 return BLK_STS_OK;
592}
593
594static void __submit_bio(struct bio *bio)
595{
596 struct gendisk *disk = bio->bi_bdev->bd_disk;
597
598 if (unlikely(!blk_crypto_bio_prep(&bio)))
599 return;
600
601 if (!disk->fops->submit_bio) {
602 blk_mq_submit_bio(bio);
603 } else if (likely(bio_queue_enter(bio) == 0)) {
604 disk->fops->submit_bio(bio);
605 blk_queue_exit(disk->queue);
606 }
607}
608
609/*
610 * The loop in this function may be a bit non-obvious, and so deserves some
611 * explanation:
612 *
613 * - Before entering the loop, bio->bi_next is NULL (as all callers ensure
614 * that), so we have a list with a single bio.
615 * - We pretend that we have just taken it off a longer list, so we assign
616 * bio_list to a pointer to the bio_list_on_stack, thus initialising the
617 * bio_list of new bios to be added. ->submit_bio() may indeed add some more
618 * bios through a recursive call to submit_bio_noacct. If it did, we find a
619 * non-NULL value in bio_list and re-enter the loop from the top.
620 * - In this case we really did just take the bio of the top of the list (no
621 * pretending) and so remove it from bio_list, and call into ->submit_bio()
622 * again.
623 *
624 * bio_list_on_stack[0] contains bios submitted by the current ->submit_bio.
625 * bio_list_on_stack[1] contains bios that were submitted before the current
626 * ->submit_bio, but that haven't been processed yet.
627 */
628static void __submit_bio_noacct(struct bio *bio)
629{
630 struct bio_list bio_list_on_stack[2];
631
632 BUG_ON(bio->bi_next);
633
634 bio_list_init(&bio_list_on_stack[0]);
635 current->bio_list = bio_list_on_stack;
636
637 do {
638 struct request_queue *q = bdev_get_queue(bio->bi_bdev);
639 struct bio_list lower, same;
640
641 /*
642 * Create a fresh bio_list for all subordinate requests.
643 */
644 bio_list_on_stack[1] = bio_list_on_stack[0];
645 bio_list_init(&bio_list_on_stack[0]);
646
647 __submit_bio(bio);
648
649 /*
650 * Sort new bios into those for a lower level and those for the
651 * same level.
652 */
653 bio_list_init(&lower);
654 bio_list_init(&same);
655 while ((bio = bio_list_pop(&bio_list_on_stack[0])) != NULL)
656 if (q == bdev_get_queue(bio->bi_bdev))
657 bio_list_add(&same, bio);
658 else
659 bio_list_add(&lower, bio);
660
661 /*
662 * Now assemble so we handle the lowest level first.
663 */
664 bio_list_merge(&bio_list_on_stack[0], &lower);
665 bio_list_merge(&bio_list_on_stack[0], &same);
666 bio_list_merge(&bio_list_on_stack[0], &bio_list_on_stack[1]);
667 } while ((bio = bio_list_pop(&bio_list_on_stack[0])));
668
669 current->bio_list = NULL;
670}
671
672static void __submit_bio_noacct_mq(struct bio *bio)
673{
674 struct bio_list bio_list[2] = { };
675
676 current->bio_list = bio_list;
677
678 do {
679 __submit_bio(bio);
680 } while ((bio = bio_list_pop(&bio_list[0])));
681
682 current->bio_list = NULL;
683}
684
685void submit_bio_noacct_nocheck(struct bio *bio)
686{
687 /*
688 * We only want one ->submit_bio to be active at a time, else stack
689 * usage with stacked devices could be a problem. Use current->bio_list
690 * to collect a list of requests submited by a ->submit_bio method while
691 * it is active, and then process them after it returned.
692 */
693 if (current->bio_list)
694 bio_list_add(¤t->bio_list[0], bio);
695 else if (!bio->bi_bdev->bd_disk->fops->submit_bio)
696 __submit_bio_noacct_mq(bio);
697 else
698 __submit_bio_noacct(bio);
699}
700
701/**
702 * submit_bio_noacct - re-submit a bio to the block device layer for I/O
703 * @bio: The bio describing the location in memory and on the device.
704 *
705 * This is a version of submit_bio() that shall only be used for I/O that is
706 * resubmitted to lower level drivers by stacking block drivers. All file
707 * systems and other upper level users of the block layer should use
708 * submit_bio() instead.
709 */
710void submit_bio_noacct(struct bio *bio)
711{
712 struct block_device *bdev = bio->bi_bdev;
713 struct request_queue *q = bdev_get_queue(bdev);
714 blk_status_t status = BLK_STS_IOERR;
715 struct blk_plug *plug;
716
717 might_sleep();
718
719 plug = blk_mq_plug(bio);
720 if (plug && plug->nowait)
721 bio->bi_opf |= REQ_NOWAIT;
722
723 /*
724 * For a REQ_NOWAIT based request, return -EOPNOTSUPP
725 * if queue does not support NOWAIT.
726 */
727 if ((bio->bi_opf & REQ_NOWAIT) && !bdev_nowait(bdev))
728 goto not_supported;
729
730 if (should_fail_bio(bio))
731 goto end_io;
732 bio_check_ro(bio);
733 if (!bio_flagged(bio, BIO_REMAPPED)) {
734 if (unlikely(bio_check_eod(bio)))
735 goto end_io;
736 if (bdev->bd_partno && unlikely(blk_partition_remap(bio)))
737 goto end_io;
738 }
739
740 /*
741 * Filter flush bio's early so that bio based drivers without flush
742 * support don't have to worry about them.
743 */
744 if (op_is_flush(bio->bi_opf) &&
745 !test_bit(QUEUE_FLAG_WC, &q->queue_flags)) {
746 bio->bi_opf &= ~(REQ_PREFLUSH | REQ_FUA);
747 if (!bio_sectors(bio)) {
748 status = BLK_STS_OK;
749 goto end_io;
750 }
751 }
752
753 if (!test_bit(QUEUE_FLAG_POLL, &q->queue_flags))
754 bio_clear_polled(bio);
755
756 switch (bio_op(bio)) {
757 case REQ_OP_DISCARD:
758 if (!bdev_max_discard_sectors(bdev))
759 goto not_supported;
760 break;
761 case REQ_OP_SECURE_ERASE:
762 if (!bdev_max_secure_erase_sectors(bdev))
763 goto not_supported;
764 break;
765 case REQ_OP_ZONE_APPEND:
766 status = blk_check_zone_append(q, bio);
767 if (status != BLK_STS_OK)
768 goto end_io;
769 break;
770 case REQ_OP_ZONE_RESET:
771 case REQ_OP_ZONE_OPEN:
772 case REQ_OP_ZONE_CLOSE:
773 case REQ_OP_ZONE_FINISH:
774 if (!bdev_is_zoned(bio->bi_bdev))
775 goto not_supported;
776 break;
777 case REQ_OP_ZONE_RESET_ALL:
778 if (!bdev_is_zoned(bio->bi_bdev) || !blk_queue_zone_resetall(q))
779 goto not_supported;
780 break;
781 case REQ_OP_WRITE_ZEROES:
782 if (!q->limits.max_write_zeroes_sectors)
783 goto not_supported;
784 break;
785 default:
786 break;
787 }
788
789 if (blk_throtl_bio(bio))
790 return;
791
792 blk_cgroup_bio_start(bio);
793 blkcg_bio_issue_init(bio);
794
795 if (!bio_flagged(bio, BIO_TRACE_COMPLETION)) {
796 trace_block_bio_queue(bio);
797 /* Now that enqueuing has been traced, we need to trace
798 * completion as well.
799 */
800 bio_set_flag(bio, BIO_TRACE_COMPLETION);
801 }
802 submit_bio_noacct_nocheck(bio);
803 return;
804
805not_supported:
806 status = BLK_STS_NOTSUPP;
807end_io:
808 bio->bi_status = status;
809 bio_endio(bio);
810}
811EXPORT_SYMBOL(submit_bio_noacct);
812
813/**
814 * submit_bio - submit a bio to the block device layer for I/O
815 * @bio: The &struct bio which describes the I/O
816 *
817 * submit_bio() is used to submit I/O requests to block devices. It is passed a
818 * fully set up &struct bio that describes the I/O that needs to be done. The
819 * bio will be send to the device described by the bi_bdev field.
820 *
821 * The success/failure status of the request, along with notification of
822 * completion, is delivered asynchronously through the ->bi_end_io() callback
823 * in @bio. The bio must NOT be touched by the caller until ->bi_end_io() has
824 * been called.
825 */
826void submit_bio(struct bio *bio)
827{
828 if (blkcg_punt_bio_submit(bio))
829 return;
830
831 if (bio_op(bio) == REQ_OP_READ) {
832 task_io_account_read(bio->bi_iter.bi_size);
833 count_vm_events(PGPGIN, bio_sectors(bio));
834 } else if (bio_op(bio) == REQ_OP_WRITE) {
835 count_vm_events(PGPGOUT, bio_sectors(bio));
836 }
837
838 submit_bio_noacct(bio);
839}
840EXPORT_SYMBOL(submit_bio);
841
842/**
843 * bio_poll - poll for BIO completions
844 * @bio: bio to poll for
845 * @iob: batches of IO
846 * @flags: BLK_POLL_* flags that control the behavior
847 *
848 * Poll for completions on queue associated with the bio. Returns number of
849 * completed entries found.
850 *
851 * Note: the caller must either be the context that submitted @bio, or
852 * be in a RCU critical section to prevent freeing of @bio.
853 */
854int bio_poll(struct bio *bio, struct io_comp_batch *iob, unsigned int flags)
855{
856 struct request_queue *q = bdev_get_queue(bio->bi_bdev);
857 blk_qc_t cookie = READ_ONCE(bio->bi_cookie);
858 int ret = 0;
859
860 if (cookie == BLK_QC_T_NONE ||
861 !test_bit(QUEUE_FLAG_POLL, &q->queue_flags))
862 return 0;
863
864 /*
865 * As the requests that require a zone lock are not plugged in the
866 * first place, directly accessing the plug instead of using
867 * blk_mq_plug() should not have any consequences during flushing for
868 * zoned devices.
869 */
870 blk_flush_plug(current->plug, false);
871
872 if (bio_queue_enter(bio))
873 return 0;
874 if (queue_is_mq(q)) {
875 ret = blk_mq_poll(q, cookie, iob, flags);
876 } else {
877 struct gendisk *disk = q->disk;
878
879 if (disk && disk->fops->poll_bio)
880 ret = disk->fops->poll_bio(bio, iob, flags);
881 }
882 blk_queue_exit(q);
883 return ret;
884}
885EXPORT_SYMBOL_GPL(bio_poll);
886
887/*
888 * Helper to implement file_operations.iopoll. Requires the bio to be stored
889 * in iocb->private, and cleared before freeing the bio.
890 */
891int iocb_bio_iopoll(struct kiocb *kiocb, struct io_comp_batch *iob,
892 unsigned int flags)
893{
894 struct bio *bio;
895 int ret = 0;
896
897 /*
898 * Note: the bio cache only uses SLAB_TYPESAFE_BY_RCU, so bio can
899 * point to a freshly allocated bio at this point. If that happens
900 * we have a few cases to consider:
901 *
902 * 1) the bio is beeing initialized and bi_bdev is NULL. We can just
903 * simply nothing in this case
904 * 2) the bio points to a not poll enabled device. bio_poll will catch
905 * this and return 0
906 * 3) the bio points to a poll capable device, including but not
907 * limited to the one that the original bio pointed to. In this
908 * case we will call into the actual poll method and poll for I/O,
909 * even if we don't need to, but it won't cause harm either.
910 *
911 * For cases 2) and 3) above the RCU grace period ensures that bi_bdev
912 * is still allocated. Because partitions hold a reference to the whole
913 * device bdev and thus disk, the disk is also still valid. Grabbing
914 * a reference to the queue in bio_poll() ensures the hctxs and requests
915 * are still valid as well.
916 */
917 rcu_read_lock();
918 bio = READ_ONCE(kiocb->private);
919 if (bio && bio->bi_bdev)
920 ret = bio_poll(bio, iob, flags);
921 rcu_read_unlock();
922
923 return ret;
924}
925EXPORT_SYMBOL_GPL(iocb_bio_iopoll);
926
927void update_io_ticks(struct block_device *part, unsigned long now, bool end)
928{
929 unsigned long stamp;
930again:
931 stamp = READ_ONCE(part->bd_stamp);
932 if (unlikely(time_after(now, stamp))) {
933 if (likely(try_cmpxchg(&part->bd_stamp, &stamp, now)))
934 __part_stat_add(part, io_ticks, end ? now - stamp : 1);
935 }
936 if (part->bd_partno) {
937 part = bdev_whole(part);
938 goto again;
939 }
940}
941
942unsigned long bdev_start_io_acct(struct block_device *bdev,
943 unsigned int sectors, enum req_op op,
944 unsigned long start_time)
945{
946 const int sgrp = op_stat_group(op);
947
948 part_stat_lock();
949 update_io_ticks(bdev, start_time, false);
950 part_stat_inc(bdev, ios[sgrp]);
951 part_stat_add(bdev, sectors[sgrp], sectors);
952 part_stat_local_inc(bdev, in_flight[op_is_write(op)]);
953 part_stat_unlock();
954
955 return start_time;
956}
957EXPORT_SYMBOL(bdev_start_io_acct);
958
959/**
960 * bio_start_io_acct - start I/O accounting for bio based drivers
961 * @bio: bio to start account for
962 *
963 * Returns the start time that should be passed back to bio_end_io_acct().
964 */
965unsigned long bio_start_io_acct(struct bio *bio)
966{
967 return bdev_start_io_acct(bio->bi_bdev, bio_sectors(bio),
968 bio_op(bio), jiffies);
969}
970EXPORT_SYMBOL_GPL(bio_start_io_acct);
971
972void bdev_end_io_acct(struct block_device *bdev, enum req_op op,
973 unsigned long start_time)
974{
975 const int sgrp = op_stat_group(op);
976 unsigned long now = READ_ONCE(jiffies);
977 unsigned long duration = now - start_time;
978
979 part_stat_lock();
980 update_io_ticks(bdev, now, true);
981 part_stat_add(bdev, nsecs[sgrp], jiffies_to_nsecs(duration));
982 part_stat_local_dec(bdev, in_flight[op_is_write(op)]);
983 part_stat_unlock();
984}
985EXPORT_SYMBOL(bdev_end_io_acct);
986
987void bio_end_io_acct_remapped(struct bio *bio, unsigned long start_time,
988 struct block_device *orig_bdev)
989{
990 bdev_end_io_acct(orig_bdev, bio_op(bio), start_time);
991}
992EXPORT_SYMBOL_GPL(bio_end_io_acct_remapped);
993
994/**
995 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
996 * @q : the queue of the device being checked
997 *
998 * Description:
999 * Check if underlying low-level drivers of a device are busy.
1000 * If the drivers want to export their busy state, they must set own
1001 * exporting function using blk_queue_lld_busy() first.
1002 *
1003 * Basically, this function is used only by request stacking drivers
1004 * to stop dispatching requests to underlying devices when underlying
1005 * devices are busy. This behavior helps more I/O merging on the queue
1006 * of the request stacking driver and prevents I/O throughput regression
1007 * on burst I/O load.
1008 *
1009 * Return:
1010 * 0 - Not busy (The request stacking driver should dispatch request)
1011 * 1 - Busy (The request stacking driver should stop dispatching request)
1012 */
1013int blk_lld_busy(struct request_queue *q)
1014{
1015 if (queue_is_mq(q) && q->mq_ops->busy)
1016 return q->mq_ops->busy(q);
1017
1018 return 0;
1019}
1020EXPORT_SYMBOL_GPL(blk_lld_busy);
1021
1022int kblockd_schedule_work(struct work_struct *work)
1023{
1024 return queue_work(kblockd_workqueue, work);
1025}
1026EXPORT_SYMBOL(kblockd_schedule_work);
1027
1028int kblockd_mod_delayed_work_on(int cpu, struct delayed_work *dwork,
1029 unsigned long delay)
1030{
1031 return mod_delayed_work_on(cpu, kblockd_workqueue, dwork, delay);
1032}
1033EXPORT_SYMBOL(kblockd_mod_delayed_work_on);
1034
1035void blk_start_plug_nr_ios(struct blk_plug *plug, unsigned short nr_ios)
1036{
1037 struct task_struct *tsk = current;
1038
1039 /*
1040 * If this is a nested plug, don't actually assign it.
1041 */
1042 if (tsk->plug)
1043 return;
1044
1045 plug->mq_list = NULL;
1046 plug->cached_rq = NULL;
1047 plug->nr_ios = min_t(unsigned short, nr_ios, BLK_MAX_REQUEST_COUNT);
1048 plug->rq_count = 0;
1049 plug->multiple_queues = false;
1050 plug->has_elevator = false;
1051 plug->nowait = false;
1052 INIT_LIST_HEAD(&plug->cb_list);
1053
1054 /*
1055 * Store ordering should not be needed here, since a potential
1056 * preempt will imply a full memory barrier
1057 */
1058 tsk->plug = plug;
1059}
1060
1061/**
1062 * blk_start_plug - initialize blk_plug and track it inside the task_struct
1063 * @plug: The &struct blk_plug that needs to be initialized
1064 *
1065 * Description:
1066 * blk_start_plug() indicates to the block layer an intent by the caller
1067 * to submit multiple I/O requests in a batch. The block layer may use
1068 * this hint to defer submitting I/Os from the caller until blk_finish_plug()
1069 * is called. However, the block layer may choose to submit requests
1070 * before a call to blk_finish_plug() if the number of queued I/Os
1071 * exceeds %BLK_MAX_REQUEST_COUNT, or if the size of the I/O is larger than
1072 * %BLK_PLUG_FLUSH_SIZE. The queued I/Os may also be submitted early if
1073 * the task schedules (see below).
1074 *
1075 * Tracking blk_plug inside the task_struct will help with auto-flushing the
1076 * pending I/O should the task end up blocking between blk_start_plug() and
1077 * blk_finish_plug(). This is important from a performance perspective, but
1078 * also ensures that we don't deadlock. For instance, if the task is blocking
1079 * for a memory allocation, memory reclaim could end up wanting to free a
1080 * page belonging to that request that is currently residing in our private
1081 * plug. By flushing the pending I/O when the process goes to sleep, we avoid
1082 * this kind of deadlock.
1083 */
1084void blk_start_plug(struct blk_plug *plug)
1085{
1086 blk_start_plug_nr_ios(plug, 1);
1087}
1088EXPORT_SYMBOL(blk_start_plug);
1089
1090static void flush_plug_callbacks(struct blk_plug *plug, bool from_schedule)
1091{
1092 LIST_HEAD(callbacks);
1093
1094 while (!list_empty(&plug->cb_list)) {
1095 list_splice_init(&plug->cb_list, &callbacks);
1096
1097 while (!list_empty(&callbacks)) {
1098 struct blk_plug_cb *cb = list_first_entry(&callbacks,
1099 struct blk_plug_cb,
1100 list);
1101 list_del(&cb->list);
1102 cb->callback(cb, from_schedule);
1103 }
1104 }
1105}
1106
1107struct blk_plug_cb *blk_check_plugged(blk_plug_cb_fn unplug, void *data,
1108 int size)
1109{
1110 struct blk_plug *plug = current->plug;
1111 struct blk_plug_cb *cb;
1112
1113 if (!plug)
1114 return NULL;
1115
1116 list_for_each_entry(cb, &plug->cb_list, list)
1117 if (cb->callback == unplug && cb->data == data)
1118 return cb;
1119
1120 /* Not currently on the callback list */
1121 BUG_ON(size < sizeof(*cb));
1122 cb = kzalloc(size, GFP_ATOMIC);
1123 if (cb) {
1124 cb->data = data;
1125 cb->callback = unplug;
1126 list_add(&cb->list, &plug->cb_list);
1127 }
1128 return cb;
1129}
1130EXPORT_SYMBOL(blk_check_plugged);
1131
1132void __blk_flush_plug(struct blk_plug *plug, bool from_schedule)
1133{
1134 if (!list_empty(&plug->cb_list))
1135 flush_plug_callbacks(plug, from_schedule);
1136 if (!rq_list_empty(plug->mq_list))
1137 blk_mq_flush_plug_list(plug, from_schedule);
1138 /*
1139 * Unconditionally flush out cached requests, even if the unplug
1140 * event came from schedule. Since we know hold references to the
1141 * queue for cached requests, we don't want a blocked task holding
1142 * up a queue freeze/quiesce event.
1143 */
1144 if (unlikely(!rq_list_empty(plug->cached_rq)))
1145 blk_mq_free_plug_rqs(plug);
1146}
1147
1148/**
1149 * blk_finish_plug - mark the end of a batch of submitted I/O
1150 * @plug: The &struct blk_plug passed to blk_start_plug()
1151 *
1152 * Description:
1153 * Indicate that a batch of I/O submissions is complete. This function
1154 * must be paired with an initial call to blk_start_plug(). The intent
1155 * is to allow the block layer to optimize I/O submission. See the
1156 * documentation for blk_start_plug() for more information.
1157 */
1158void blk_finish_plug(struct blk_plug *plug)
1159{
1160 if (plug == current->plug) {
1161 __blk_flush_plug(plug, false);
1162 current->plug = NULL;
1163 }
1164}
1165EXPORT_SYMBOL(blk_finish_plug);
1166
1167void blk_io_schedule(void)
1168{
1169 /* Prevent hang_check timer from firing at us during very long I/O */
1170 unsigned long timeout = sysctl_hung_task_timeout_secs * HZ / 2;
1171
1172 if (timeout)
1173 io_schedule_timeout(timeout);
1174 else
1175 io_schedule();
1176}
1177EXPORT_SYMBOL_GPL(blk_io_schedule);
1178
1179int __init blk_dev_init(void)
1180{
1181 BUILD_BUG_ON((__force u32)REQ_OP_LAST >= (1 << REQ_OP_BITS));
1182 BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 *
1183 sizeof_field(struct request, cmd_flags));
1184 BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 *
1185 sizeof_field(struct bio, bi_opf));
1186
1187 /* used for unplugging and affects IO latency/throughput - HIGHPRI */
1188 kblockd_workqueue = alloc_workqueue("kblockd",
1189 WQ_MEM_RECLAIM | WQ_HIGHPRI, 0);
1190 if (!kblockd_workqueue)
1191 panic("Failed to create kblockd\n");
1192
1193 blk_requestq_cachep = kmem_cache_create("request_queue",
1194 sizeof(struct request_queue), 0, SLAB_PANIC, NULL);
1195
1196 blk_debugfs_root = debugfs_create_dir("block", NULL);
1197
1198 return 0;
1199}
1// SPDX-License-Identifier: GPL-2.0
2/*
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>
8 * - July2000
9 * bio rewrite, highmem i/o, etc, Jens Axboe <axboe@suse.de> - may 2001
10 */
11
12/*
13 * This handles all read/write requests to block devices
14 */
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>
22#include <linux/mm.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>
44
45#define CREATE_TRACE_POINTS
46#include <trace/events/block.h>
47
48#include "blk.h"
49#include "blk-mq.h"
50#include "blk-mq-sched.h"
51#include "blk-pm.h"
52#include "blk-rq-qos.h"
53
54struct dentry *blk_debugfs_root;
55
56EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_remap);
57EXPORT_TRACEPOINT_SYMBOL_GPL(block_rq_remap);
58EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_complete);
59EXPORT_TRACEPOINT_SYMBOL_GPL(block_split);
60EXPORT_TRACEPOINT_SYMBOL_GPL(block_unplug);
61
62DEFINE_IDA(blk_queue_ida);
63
64/*
65 * For queue allocation
66 */
67struct kmem_cache *blk_requestq_cachep;
68
69/*
70 * Controlling structure to kblockd
71 */
72static struct workqueue_struct *kblockd_workqueue;
73
74/**
75 * blk_queue_flag_set - atomically set a queue flag
76 * @flag: flag to be set
77 * @q: request queue
78 */
79void blk_queue_flag_set(unsigned int flag, struct request_queue *q)
80{
81 set_bit(flag, &q->queue_flags);
82}
83EXPORT_SYMBOL(blk_queue_flag_set);
84
85/**
86 * blk_queue_flag_clear - atomically clear a queue flag
87 * @flag: flag to be cleared
88 * @q: request queue
89 */
90void blk_queue_flag_clear(unsigned int flag, struct request_queue *q)
91{
92 clear_bit(flag, &q->queue_flags);
93}
94EXPORT_SYMBOL(blk_queue_flag_clear);
95
96/**
97 * blk_queue_flag_test_and_set - atomically test and set a queue flag
98 * @flag: flag to be set
99 * @q: request queue
100 *
101 * Returns the previous value of @flag - 0 if the flag was not set and 1 if
102 * the flag was already set.
103 */
104bool blk_queue_flag_test_and_set(unsigned int flag, struct request_queue *q)
105{
106 return test_and_set_bit(flag, &q->queue_flags);
107}
108EXPORT_SYMBOL_GPL(blk_queue_flag_test_and_set);
109
110void blk_rq_init(struct request_queue *q, struct request *rq)
111{
112 memset(rq, 0, sizeof(*rq));
113
114 INIT_LIST_HEAD(&rq->queuelist);
115 rq->q = q;
116 rq->__sector = (sector_t) -1;
117 INIT_HLIST_NODE(&rq->hash);
118 RB_CLEAR_NODE(&rq->rb_node);
119 rq->tag = -1;
120 rq->internal_tag = -1;
121 rq->start_time_ns = ktime_get_ns();
122 rq->part = NULL;
123 refcount_set(&rq->ref, 1);
124 blk_crypto_rq_set_defaults(rq);
125}
126EXPORT_SYMBOL(blk_rq_init);
127
128#define REQ_OP_NAME(name) [REQ_OP_##name] = #name
129static const char *const blk_op_name[] = {
130 REQ_OP_NAME(READ),
131 REQ_OP_NAME(WRITE),
132 REQ_OP_NAME(FLUSH),
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),
145 REQ_OP_NAME(DRV_IN),
146 REQ_OP_NAME(DRV_OUT),
147};
148#undef REQ_OP_NAME
149
150/**
151 * blk_op_str - Return string XXX in the REQ_OP_XXX.
152 * @op: REQ_OP_XXX.
153 *
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".
157 */
158inline const char *blk_op_str(unsigned int op)
159{
160 const char *op_str = "UNKNOWN";
161
162 if (op < ARRAY_SIZE(blk_op_name) && blk_op_name[op])
163 op_str = blk_op_name[op];
164
165 return op_str;
166}
167EXPORT_SYMBOL_GPL(blk_op_str);
168
169static const struct {
170 int errno;
171 const char *name;
172} blk_errors[] = {
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" },
185
186 /* device mapper special case, should not leak out: */
187 [BLK_STS_DM_REQUEUE] = { -EREMCHG, "dm internal retry" },
188
189 /* everything else not covered above: */
190 [BLK_STS_IOERR] = { -EIO, "I/O" },
191};
192
193blk_status_t errno_to_blk_status(int errno)
194{
195 int i;
196
197 for (i = 0; i < ARRAY_SIZE(blk_errors); i++) {
198 if (blk_errors[i].errno == errno)
199 return (__force blk_status_t)i;
200 }
201
202 return BLK_STS_IOERR;
203}
204EXPORT_SYMBOL_GPL(errno_to_blk_status);
205
206int blk_status_to_errno(blk_status_t status)
207{
208 int idx = (__force int)status;
209
210 if (WARN_ON_ONCE(idx >= ARRAY_SIZE(blk_errors)))
211 return -EIO;
212 return blk_errors[idx].errno;
213}
214EXPORT_SYMBOL_GPL(blk_status_to_errno);
215
216static void print_req_error(struct request *req, blk_status_t status,
217 const char *caller)
218{
219 int idx = (__force int)status;
220
221 if (WARN_ON_ONCE(idx >= ARRAY_SIZE(blk_errors)))
222 return;
223
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));
233}
234
235static void req_bio_endio(struct request *rq, struct bio *bio,
236 unsigned int nbytes, blk_status_t error)
237{
238 if (error)
239 bio->bi_status = error;
240
241 if (unlikely(rq->rq_flags & RQF_QUIET))
242 bio_set_flag(bio, BIO_QUIET);
243
244 bio_advance(bio, nbytes);
245
246 if (req_op(rq) == REQ_OP_ZONE_APPEND && error == BLK_STS_OK) {
247 /*
248 * Partial zone append completions cannot be supported as the
249 * BIO fragments may end up not being written sequentially.
250 */
251 if (bio->bi_iter.bi_size)
252 bio->bi_status = BLK_STS_IOERR;
253 else
254 bio->bi_iter.bi_sector = rq->__sector;
255 }
256
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))
259 bio_endio(bio);
260}
261
262void blk_dump_rq_flags(struct request *rq, char *msg)
263{
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);
267
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));
273}
274EXPORT_SYMBOL(blk_dump_rq_flags);
275
276/**
277 * blk_sync_queue - cancel any pending callbacks on a queue
278 * @q: the queue
279 *
280 * Description:
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
287 * this function.
288 *
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.
292 *
293 */
294void blk_sync_queue(struct request_queue *q)
295{
296 del_timer_sync(&q->timeout);
297 cancel_work_sync(&q->timeout_work);
298}
299EXPORT_SYMBOL(blk_sync_queue);
300
301/**
302 * blk_set_pm_only - increment pm_only counter
303 * @q: request queue pointer
304 */
305void blk_set_pm_only(struct request_queue *q)
306{
307 atomic_inc(&q->pm_only);
308}
309EXPORT_SYMBOL_GPL(blk_set_pm_only);
310
311void blk_clear_pm_only(struct request_queue *q)
312{
313 int pm_only;
314
315 pm_only = atomic_dec_return(&q->pm_only);
316 WARN_ON_ONCE(pm_only < 0);
317 if (pm_only == 0)
318 wake_up_all(&q->mq_freeze_wq);
319}
320EXPORT_SYMBOL_GPL(blk_clear_pm_only);
321
322/**
323 * blk_put_queue - decrement the request_queue refcount
324 * @q: the request_queue structure to decrement the refcount for
325 *
326 * Decrements the refcount of the request_queue kobject. When this reaches 0
327 * we'll have blk_release_queue() called.
328 *
329 * Context: Any context, but the last reference must not be dropped from
330 * atomic context.
331 */
332void blk_put_queue(struct request_queue *q)
333{
334 kobject_put(&q->kobj);
335}
336EXPORT_SYMBOL(blk_put_queue);
337
338void blk_set_queue_dying(struct request_queue *q)
339{
340 blk_queue_flag_set(QUEUE_FLAG_DYING, q);
341
342 /*
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().
346 */
347 blk_freeze_queue_start(q);
348
349 if (queue_is_mq(q))
350 blk_mq_wake_waiters(q);
351
352 /* Make blk_queue_enter() reexamine the DYING flag. */
353 wake_up_all(&q->mq_freeze_wq);
354}
355EXPORT_SYMBOL_GPL(blk_set_queue_dying);
356
357/**
358 * blk_cleanup_queue - shutdown a request queue
359 * @q: request queue to shutdown
360 *
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.
363 *
364 * Context: can sleep
365 */
366void blk_cleanup_queue(struct request_queue *q)
367{
368 /* cannot be called from atomic context */
369 might_sleep();
370
371 WARN_ON_ONCE(blk_queue_registered(q));
372
373 /* mark @q DYING, no new request or merges will be allowed afterwards */
374 blk_set_queue_dying(q);
375
376 blk_queue_flag_set(QUEUE_FLAG_NOMERGES, q);
377 blk_queue_flag_set(QUEUE_FLAG_NOXMERGES, q);
378
379 /*
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.
383 */
384 blk_freeze_queue(q);
385
386 rq_qos_exit(q);
387
388 blk_queue_flag_set(QUEUE_FLAG_DEAD, q);
389
390 /* for synchronous bio-based driver finish in-flight integrity i/o */
391 blk_flush_integrity();
392
393 /* @q won't process any more request, flush async actions */
394 del_timer_sync(&q->backing_dev_info->laptop_mode_wb_timer);
395 blk_sync_queue(q);
396
397 if (queue_is_mq(q))
398 blk_mq_exit_queue(q);
399
400 /*
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.
404 *
405 * Queue has become frozen, there can't be any in-queue requests, so
406 * it is safe to free requests now.
407 */
408 mutex_lock(&q->sysfs_lock);
409 if (q->elevator)
410 blk_mq_sched_free_requests(q);
411 mutex_unlock(&q->sysfs_lock);
412
413 percpu_ref_exit(&q->q_usage_counter);
414
415 /* @q is and will stay empty, shutdown and put */
416 blk_put_queue(q);
417}
418EXPORT_SYMBOL(blk_cleanup_queue);
419
420/**
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
424 */
425int blk_queue_enter(struct request_queue *q, blk_mq_req_flags_t flags)
426{
427 const bool pm = flags & BLK_MQ_REQ_PREEMPT;
428
429 while (true) {
430 bool success = false;
431
432 rcu_read_lock();
433 if (percpu_ref_tryget_live(&q->q_usage_counter)) {
434 /*
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.
438 */
439 if (pm || !blk_queue_pm_only(q)) {
440 success = true;
441 } else {
442 percpu_ref_put(&q->q_usage_counter);
443 }
444 }
445 rcu_read_unlock();
446
447 if (success)
448 return 0;
449
450 if (flags & BLK_MQ_REQ_NOWAIT)
451 return -EBUSY;
452
453 /*
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.
459 */
460 smp_rmb();
461
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)))) ||
466 blk_queue_dying(q));
467 if (blk_queue_dying(q))
468 return -ENODEV;
469 }
470}
471
472static inline int bio_queue_enter(struct bio *bio)
473{
474 struct request_queue *q = bio->bi_disk->queue;
475 bool nowait = bio->bi_opf & REQ_NOWAIT;
476 int ret;
477
478 ret = blk_queue_enter(q, nowait ? BLK_MQ_REQ_NOWAIT : 0);
479 if (unlikely(ret)) {
480 if (nowait && !blk_queue_dying(q))
481 bio_wouldblock_error(bio);
482 else
483 bio_io_error(bio);
484 }
485
486 return ret;
487}
488
489void blk_queue_exit(struct request_queue *q)
490{
491 percpu_ref_put(&q->q_usage_counter);
492}
493
494static void blk_queue_usage_counter_release(struct percpu_ref *ref)
495{
496 struct request_queue *q =
497 container_of(ref, struct request_queue, q_usage_counter);
498
499 wake_up_all(&q->mq_freeze_wq);
500}
501
502static void blk_rq_timed_out_timer(struct timer_list *t)
503{
504 struct request_queue *q = from_timer(q, t, timeout);
505
506 kblockd_schedule_work(&q->timeout_work);
507}
508
509static void blk_timeout_work(struct work_struct *work)
510{
511}
512
513struct request_queue *blk_alloc_queue(int node_id)
514{
515 struct request_queue *q;
516 int ret;
517
518 q = kmem_cache_alloc_node(blk_requestq_cachep,
519 GFP_KERNEL | __GFP_ZERO, node_id);
520 if (!q)
521 return NULL;
522
523 q->last_merge = NULL;
524
525 q->id = ida_simple_get(&blk_queue_ida, 0, 0, GFP_KERNEL);
526 if (q->id < 0)
527 goto fail_q;
528
529 ret = bioset_init(&q->bio_split, BIO_POOL_SIZE, 0, BIOSET_NEED_BVECS);
530 if (ret)
531 goto fail_id;
532
533 q->backing_dev_info = bdi_alloc(node_id);
534 if (!q->backing_dev_info)
535 goto fail_split;
536
537 q->stats = blk_alloc_queue_stats();
538 if (!q->stats)
539 goto fail_stats;
540
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;
544 q->node = node_id;
545
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);
553#endif
554
555 kobject_init(&q->kobj, &blk_queue_ktype);
556
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);
561
562 init_waitqueue_head(&q->mq_freeze_wq);
563 mutex_init(&q->mq_freeze_lock);
564
565 /*
566 * Init percpu_ref in atomic mode so that it's faster to shutdown.
567 * See blk_register_queue() for details.
568 */
569 if (percpu_ref_init(&q->q_usage_counter,
570 blk_queue_usage_counter_release,
571 PERCPU_REF_INIT_ATOMIC, GFP_KERNEL))
572 goto fail_bdi;
573
574 if (blkcg_init_queue(q))
575 goto fail_ref;
576
577 blk_queue_dma_alignment(q, 511);
578 blk_set_default_limits(&q->limits);
579 q->nr_requests = BLKDEV_MAX_RQ;
580
581 return q;
582
583fail_ref:
584 percpu_ref_exit(&q->q_usage_counter);
585fail_bdi:
586 blk_free_queue_stats(q->stats);
587fail_stats:
588 bdi_put(q->backing_dev_info);
589fail_split:
590 bioset_exit(&q->bio_split);
591fail_id:
592 ida_simple_remove(&blk_queue_ida, q->id);
593fail_q:
594 kmem_cache_free(blk_requestq_cachep, q);
595 return NULL;
596}
597EXPORT_SYMBOL(blk_alloc_queue);
598
599/**
600 * blk_get_queue - increment the request_queue refcount
601 * @q: the request_queue structure to increment the refcount for
602 *
603 * Increment the refcount of the request_queue kobject.
604 *
605 * Context: Any context.
606 */
607bool blk_get_queue(struct request_queue *q)
608{
609 if (likely(!blk_queue_dying(q))) {
610 __blk_get_queue(q);
611 return true;
612 }
613
614 return false;
615}
616EXPORT_SYMBOL(blk_get_queue);
617
618/**
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.
623 */
624struct request *blk_get_request(struct request_queue *q, unsigned int op,
625 blk_mq_req_flags_t flags)
626{
627 struct request *req;
628
629 WARN_ON_ONCE(op & REQ_NOWAIT);
630 WARN_ON_ONCE(flags & ~(BLK_MQ_REQ_NOWAIT | BLK_MQ_REQ_PREEMPT));
631
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);
635
636 return req;
637}
638EXPORT_SYMBOL(blk_get_request);
639
640void blk_put_request(struct request *req)
641{
642 blk_mq_free_request(req);
643}
644EXPORT_SYMBOL(blk_put_request);
645
646static void blk_account_io_merge_bio(struct request *req)
647{
648 if (!blk_do_io_stat(req))
649 return;
650
651 part_stat_lock();
652 part_stat_inc(req->part, merges[op_stat_group(req_op(req))]);
653 part_stat_unlock();
654}
655
656bool bio_attempt_back_merge(struct request *req, struct bio *bio,
657 unsigned int nr_segs)
658{
659 const int ff = bio->bi_opf & REQ_FAILFAST_MASK;
660
661 if (!ll_back_merge_fn(req, bio, nr_segs))
662 return false;
663
664 trace_block_bio_backmerge(req->q, req, bio);
665 rq_qos_merge(req->q, req, bio);
666
667 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
668 blk_rq_set_mixed_merge(req);
669
670 req->biotail->bi_next = bio;
671 req->biotail = bio;
672 req->__data_len += bio->bi_iter.bi_size;
673
674 bio_crypt_free_ctx(bio);
675
676 blk_account_io_merge_bio(req);
677 return true;
678}
679
680bool bio_attempt_front_merge(struct request *req, struct bio *bio,
681 unsigned int nr_segs)
682{
683 const int ff = bio->bi_opf & REQ_FAILFAST_MASK;
684
685 if (!ll_front_merge_fn(req, bio, nr_segs))
686 return false;
687
688 trace_block_bio_frontmerge(req->q, req, bio);
689 rq_qos_merge(req->q, req, bio);
690
691 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
692 blk_rq_set_mixed_merge(req);
693
694 bio->bi_next = req->bio;
695 req->bio = bio;
696
697 req->__sector = bio->bi_iter.bi_sector;
698 req->__data_len += bio->bi_iter.bi_size;
699
700 bio_crypt_do_front_merge(req, bio);
701
702 blk_account_io_merge_bio(req);
703 return true;
704}
705
706bool bio_attempt_discard_merge(struct request_queue *q, struct request *req,
707 struct bio *bio)
708{
709 unsigned short segments = blk_rq_nr_discard_segments(req);
710
711 if (segments >= queue_max_discard_segments(q))
712 goto no_merge;
713 if (blk_rq_sectors(req) + bio_sectors(bio) >
714 blk_rq_get_max_sectors(req, blk_rq_pos(req)))
715 goto no_merge;
716
717 rq_qos_merge(q, req, bio);
718
719 req->biotail->bi_next = bio;
720 req->biotail = bio;
721 req->__data_len += bio->bi_iter.bi_size;
722 req->nr_phys_segments = segments + 1;
723
724 blk_account_io_merge_bio(req);
725 return true;
726no_merge:
727 req_set_nomerge(q, req);
728 return false;
729}
730
731/**
732 * blk_attempt_plug_merge - try to merge with %current's plugged list
733 * @q: request_queue new bio is being queued at
734 * @bio: new bio being queued
735 * @nr_segs: number of segments in @bio
736 * @same_queue_rq: pointer to &struct request that gets filled in when
737 * another request associated with @q is found on the plug list
738 * (optional, may be %NULL)
739 *
740 * Determine whether @bio being queued on @q can be merged with a request
741 * on %current's plugged list. Returns %true if merge was successful,
742 * otherwise %false.
743 *
744 * Plugging coalesces IOs from the same issuer for the same purpose without
745 * going through @q->queue_lock. As such it's more of an issuing mechanism
746 * than scheduling, and the request, while may have elvpriv data, is not
747 * added on the elevator at this point. In addition, we don't have
748 * reliable access to the elevator outside queue lock. Only check basic
749 * merging parameters without querying the elevator.
750 *
751 * Caller must ensure !blk_queue_nomerges(q) beforehand.
752 */
753bool blk_attempt_plug_merge(struct request_queue *q, struct bio *bio,
754 unsigned int nr_segs, struct request **same_queue_rq)
755{
756 struct blk_plug *plug;
757 struct request *rq;
758 struct list_head *plug_list;
759
760 plug = blk_mq_plug(q, bio);
761 if (!plug)
762 return false;
763
764 plug_list = &plug->mq_list;
765
766 list_for_each_entry_reverse(rq, plug_list, queuelist) {
767 bool merged = false;
768
769 if (rq->q == q && same_queue_rq) {
770 /*
771 * Only blk-mq multiple hardware queues case checks the
772 * rq in the same queue, there should be only one such
773 * rq in a queue
774 **/
775 *same_queue_rq = rq;
776 }
777
778 if (rq->q != q || !blk_rq_merge_ok(rq, bio))
779 continue;
780
781 switch (blk_try_merge(rq, bio)) {
782 case ELEVATOR_BACK_MERGE:
783 merged = bio_attempt_back_merge(rq, bio, nr_segs);
784 break;
785 case ELEVATOR_FRONT_MERGE:
786 merged = bio_attempt_front_merge(rq, bio, nr_segs);
787 break;
788 case ELEVATOR_DISCARD_MERGE:
789 merged = bio_attempt_discard_merge(q, rq, bio);
790 break;
791 default:
792 break;
793 }
794
795 if (merged)
796 return true;
797 }
798
799 return false;
800}
801
802static void handle_bad_sector(struct bio *bio, sector_t maxsector)
803{
804 char b[BDEVNAME_SIZE];
805
806 printk(KERN_INFO "attempt to access beyond end of device\n");
807 printk(KERN_INFO "%s: rw=%d, want=%Lu, limit=%Lu\n",
808 bio_devname(bio, b), bio->bi_opf,
809 (unsigned long long)bio_end_sector(bio),
810 (long long)maxsector);
811}
812
813#ifdef CONFIG_FAIL_MAKE_REQUEST
814
815static DECLARE_FAULT_ATTR(fail_make_request);
816
817static int __init setup_fail_make_request(char *str)
818{
819 return setup_fault_attr(&fail_make_request, str);
820}
821__setup("fail_make_request=", setup_fail_make_request);
822
823static bool should_fail_request(struct hd_struct *part, unsigned int bytes)
824{
825 return part->make_it_fail && should_fail(&fail_make_request, bytes);
826}
827
828static int __init fail_make_request_debugfs(void)
829{
830 struct dentry *dir = fault_create_debugfs_attr("fail_make_request",
831 NULL, &fail_make_request);
832
833 return PTR_ERR_OR_ZERO(dir);
834}
835
836late_initcall(fail_make_request_debugfs);
837
838#else /* CONFIG_FAIL_MAKE_REQUEST */
839
840static inline bool should_fail_request(struct hd_struct *part,
841 unsigned int bytes)
842{
843 return false;
844}
845
846#endif /* CONFIG_FAIL_MAKE_REQUEST */
847
848static inline bool bio_check_ro(struct bio *bio, struct hd_struct *part)
849{
850 const int op = bio_op(bio);
851
852 if (part->policy && op_is_write(op)) {
853 char b[BDEVNAME_SIZE];
854
855 if (op_is_flush(bio->bi_opf) && !bio_sectors(bio))
856 return false;
857
858 WARN_ONCE(1,
859 "Trying to write to read-only block-device %s (partno %d)\n",
860 bio_devname(bio, b), part->partno);
861 /* Older lvm-tools actually trigger this */
862 return false;
863 }
864
865 return false;
866}
867
868static noinline int should_fail_bio(struct bio *bio)
869{
870 if (should_fail_request(&bio->bi_disk->part0, bio->bi_iter.bi_size))
871 return -EIO;
872 return 0;
873}
874ALLOW_ERROR_INJECTION(should_fail_bio, ERRNO);
875
876/*
877 * Check whether this bio extends beyond the end of the device or partition.
878 * This may well happen - the kernel calls bread() without checking the size of
879 * the device, e.g., when mounting a file system.
880 */
881static inline int bio_check_eod(struct bio *bio, sector_t maxsector)
882{
883 unsigned int nr_sectors = bio_sectors(bio);
884
885 if (nr_sectors && maxsector &&
886 (nr_sectors > maxsector ||
887 bio->bi_iter.bi_sector > maxsector - nr_sectors)) {
888 handle_bad_sector(bio, maxsector);
889 return -EIO;
890 }
891 return 0;
892}
893
894/*
895 * Remap block n of partition p to block n+start(p) of the disk.
896 */
897static inline int blk_partition_remap(struct bio *bio)
898{
899 struct hd_struct *p;
900 int ret = -EIO;
901
902 rcu_read_lock();
903 p = __disk_get_part(bio->bi_disk, bio->bi_partno);
904 if (unlikely(!p))
905 goto out;
906 if (unlikely(should_fail_request(p, bio->bi_iter.bi_size)))
907 goto out;
908 if (unlikely(bio_check_ro(bio, p)))
909 goto out;
910
911 if (bio_sectors(bio)) {
912 if (bio_check_eod(bio, part_nr_sects_read(p)))
913 goto out;
914 bio->bi_iter.bi_sector += p->start_sect;
915 trace_block_bio_remap(bio->bi_disk->queue, bio, part_devt(p),
916 bio->bi_iter.bi_sector - p->start_sect);
917 }
918 bio->bi_partno = 0;
919 ret = 0;
920out:
921 rcu_read_unlock();
922 return ret;
923}
924
925/*
926 * Check write append to a zoned block device.
927 */
928static inline blk_status_t blk_check_zone_append(struct request_queue *q,
929 struct bio *bio)
930{
931 sector_t pos = bio->bi_iter.bi_sector;
932 int nr_sectors = bio_sectors(bio);
933
934 /* Only applicable to zoned block devices */
935 if (!blk_queue_is_zoned(q))
936 return BLK_STS_NOTSUPP;
937
938 /* The bio sector must point to the start of a sequential zone */
939 if (pos & (blk_queue_zone_sectors(q) - 1) ||
940 !blk_queue_zone_is_seq(q, pos))
941 return BLK_STS_IOERR;
942
943 /*
944 * Not allowed to cross zone boundaries. Otherwise, the BIO will be
945 * split and could result in non-contiguous sectors being written in
946 * different zones.
947 */
948 if (nr_sectors > q->limits.chunk_sectors)
949 return BLK_STS_IOERR;
950
951 /* Make sure the BIO is small enough and will not get split */
952 if (nr_sectors > q->limits.max_zone_append_sectors)
953 return BLK_STS_IOERR;
954
955 bio->bi_opf |= REQ_NOMERGE;
956
957 return BLK_STS_OK;
958}
959
960static noinline_for_stack bool submit_bio_checks(struct bio *bio)
961{
962 struct request_queue *q = bio->bi_disk->queue;
963 blk_status_t status = BLK_STS_IOERR;
964 struct blk_plug *plug;
965
966 might_sleep();
967
968 plug = blk_mq_plug(q, bio);
969 if (plug && plug->nowait)
970 bio->bi_opf |= REQ_NOWAIT;
971
972 /*
973 * For a REQ_NOWAIT based request, return -EOPNOTSUPP
974 * if queue is not a request based queue.
975 */
976 if ((bio->bi_opf & REQ_NOWAIT) && !queue_is_mq(q))
977 goto not_supported;
978
979 if (should_fail_bio(bio))
980 goto end_io;
981
982 if (bio->bi_partno) {
983 if (unlikely(blk_partition_remap(bio)))
984 goto end_io;
985 } else {
986 if (unlikely(bio_check_ro(bio, &bio->bi_disk->part0)))
987 goto end_io;
988 if (unlikely(bio_check_eod(bio, get_capacity(bio->bi_disk))))
989 goto end_io;
990 }
991
992 /*
993 * Filter flush bio's early so that bio based drivers without flush
994 * support don't have to worry about them.
995 */
996 if (op_is_flush(bio->bi_opf) &&
997 !test_bit(QUEUE_FLAG_WC, &q->queue_flags)) {
998 bio->bi_opf &= ~(REQ_PREFLUSH | REQ_FUA);
999 if (!bio_sectors(bio)) {
1000 status = BLK_STS_OK;
1001 goto end_io;
1002 }
1003 }
1004
1005 if (!test_bit(QUEUE_FLAG_POLL, &q->queue_flags))
1006 bio->bi_opf &= ~REQ_HIPRI;
1007
1008 switch (bio_op(bio)) {
1009 case REQ_OP_DISCARD:
1010 if (!blk_queue_discard(q))
1011 goto not_supported;
1012 break;
1013 case REQ_OP_SECURE_ERASE:
1014 if (!blk_queue_secure_erase(q))
1015 goto not_supported;
1016 break;
1017 case REQ_OP_WRITE_SAME:
1018 if (!q->limits.max_write_same_sectors)
1019 goto not_supported;
1020 break;
1021 case REQ_OP_ZONE_APPEND:
1022 status = blk_check_zone_append(q, bio);
1023 if (status != BLK_STS_OK)
1024 goto end_io;
1025 break;
1026 case REQ_OP_ZONE_RESET:
1027 case REQ_OP_ZONE_OPEN:
1028 case REQ_OP_ZONE_CLOSE:
1029 case REQ_OP_ZONE_FINISH:
1030 if (!blk_queue_is_zoned(q))
1031 goto not_supported;
1032 break;
1033 case REQ_OP_ZONE_RESET_ALL:
1034 if (!blk_queue_is_zoned(q) || !blk_queue_zone_resetall(q))
1035 goto not_supported;
1036 break;
1037 case REQ_OP_WRITE_ZEROES:
1038 if (!q->limits.max_write_zeroes_sectors)
1039 goto not_supported;
1040 break;
1041 default:
1042 break;
1043 }
1044
1045 /*
1046 * Various block parts want %current->io_context, so allocate it up
1047 * front rather than dealing with lots of pain to allocate it only
1048 * where needed. This may fail and the block layer knows how to live
1049 * with it.
1050 */
1051 if (unlikely(!current->io_context))
1052 create_task_io_context(current, GFP_ATOMIC, q->node);
1053
1054 if (blk_throtl_bio(bio)) {
1055 blkcg_bio_issue_init(bio);
1056 return false;
1057 }
1058
1059 blk_cgroup_bio_start(bio);
1060 blkcg_bio_issue_init(bio);
1061
1062 if (!bio_flagged(bio, BIO_TRACE_COMPLETION)) {
1063 trace_block_bio_queue(q, bio);
1064 /* Now that enqueuing has been traced, we need to trace
1065 * completion as well.
1066 */
1067 bio_set_flag(bio, BIO_TRACE_COMPLETION);
1068 }
1069 return true;
1070
1071not_supported:
1072 status = BLK_STS_NOTSUPP;
1073end_io:
1074 bio->bi_status = status;
1075 bio_endio(bio);
1076 return false;
1077}
1078
1079static blk_qc_t __submit_bio(struct bio *bio)
1080{
1081 struct gendisk *disk = bio->bi_disk;
1082 blk_qc_t ret = BLK_QC_T_NONE;
1083
1084 if (blk_crypto_bio_prep(&bio)) {
1085 if (!disk->fops->submit_bio)
1086 return blk_mq_submit_bio(bio);
1087 ret = disk->fops->submit_bio(bio);
1088 }
1089 blk_queue_exit(disk->queue);
1090 return ret;
1091}
1092
1093/*
1094 * The loop in this function may be a bit non-obvious, and so deserves some
1095 * explanation:
1096 *
1097 * - Before entering the loop, bio->bi_next is NULL (as all callers ensure
1098 * that), so we have a list with a single bio.
1099 * - We pretend that we have just taken it off a longer list, so we assign
1100 * bio_list to a pointer to the bio_list_on_stack, thus initialising the
1101 * bio_list of new bios to be added. ->submit_bio() may indeed add some more
1102 * bios through a recursive call to submit_bio_noacct. If it did, we find a
1103 * non-NULL value in bio_list and re-enter the loop from the top.
1104 * - In this case we really did just take the bio of the top of the list (no
1105 * pretending) and so remove it from bio_list, and call into ->submit_bio()
1106 * again.
1107 *
1108 * bio_list_on_stack[0] contains bios submitted by the current ->submit_bio.
1109 * bio_list_on_stack[1] contains bios that were submitted before the current
1110 * ->submit_bio_bio, but that haven't been processed yet.
1111 */
1112static blk_qc_t __submit_bio_noacct(struct bio *bio)
1113{
1114 struct bio_list bio_list_on_stack[2];
1115 blk_qc_t ret = BLK_QC_T_NONE;
1116
1117 BUG_ON(bio->bi_next);
1118
1119 bio_list_init(&bio_list_on_stack[0]);
1120 current->bio_list = bio_list_on_stack;
1121
1122 do {
1123 struct request_queue *q = bio->bi_disk->queue;
1124 struct bio_list lower, same;
1125
1126 if (unlikely(bio_queue_enter(bio) != 0))
1127 continue;
1128
1129 /*
1130 * Create a fresh bio_list for all subordinate requests.
1131 */
1132 bio_list_on_stack[1] = bio_list_on_stack[0];
1133 bio_list_init(&bio_list_on_stack[0]);
1134
1135 ret = __submit_bio(bio);
1136
1137 /*
1138 * Sort new bios into those for a lower level and those for the
1139 * same level.
1140 */
1141 bio_list_init(&lower);
1142 bio_list_init(&same);
1143 while ((bio = bio_list_pop(&bio_list_on_stack[0])) != NULL)
1144 if (q == bio->bi_disk->queue)
1145 bio_list_add(&same, bio);
1146 else
1147 bio_list_add(&lower, bio);
1148
1149 /*
1150 * Now assemble so we handle the lowest level first.
1151 */
1152 bio_list_merge(&bio_list_on_stack[0], &lower);
1153 bio_list_merge(&bio_list_on_stack[0], &same);
1154 bio_list_merge(&bio_list_on_stack[0], &bio_list_on_stack[1]);
1155 } while ((bio = bio_list_pop(&bio_list_on_stack[0])));
1156
1157 current->bio_list = NULL;
1158 return ret;
1159}
1160
1161static blk_qc_t __submit_bio_noacct_mq(struct bio *bio)
1162{
1163 struct bio_list bio_list[2] = { };
1164 blk_qc_t ret = BLK_QC_T_NONE;
1165
1166 current->bio_list = bio_list;
1167
1168 do {
1169 struct gendisk *disk = bio->bi_disk;
1170
1171 if (unlikely(bio_queue_enter(bio) != 0))
1172 continue;
1173
1174 if (!blk_crypto_bio_prep(&bio)) {
1175 blk_queue_exit(disk->queue);
1176 ret = BLK_QC_T_NONE;
1177 continue;
1178 }
1179
1180 ret = blk_mq_submit_bio(bio);
1181 } while ((bio = bio_list_pop(&bio_list[0])));
1182
1183 current->bio_list = NULL;
1184 return ret;
1185}
1186
1187/**
1188 * submit_bio_noacct - re-submit a bio to the block device layer for I/O
1189 * @bio: The bio describing the location in memory and on the device.
1190 *
1191 * This is a version of submit_bio() that shall only be used for I/O that is
1192 * resubmitted to lower level drivers by stacking block drivers. All file
1193 * systems and other upper level users of the block layer should use
1194 * submit_bio() instead.
1195 */
1196blk_qc_t submit_bio_noacct(struct bio *bio)
1197{
1198 if (!submit_bio_checks(bio))
1199 return BLK_QC_T_NONE;
1200
1201 /*
1202 * We only want one ->submit_bio to be active at a time, else stack
1203 * usage with stacked devices could be a problem. Use current->bio_list
1204 * to collect a list of requests submited by a ->submit_bio method while
1205 * it is active, and then process them after it returned.
1206 */
1207 if (current->bio_list) {
1208 bio_list_add(¤t->bio_list[0], bio);
1209 return BLK_QC_T_NONE;
1210 }
1211
1212 if (!bio->bi_disk->fops->submit_bio)
1213 return __submit_bio_noacct_mq(bio);
1214 return __submit_bio_noacct(bio);
1215}
1216EXPORT_SYMBOL(submit_bio_noacct);
1217
1218/**
1219 * submit_bio - submit a bio to the block device layer for I/O
1220 * @bio: The &struct bio which describes the I/O
1221 *
1222 * submit_bio() is used to submit I/O requests to block devices. It is passed a
1223 * fully set up &struct bio that describes the I/O that needs to be done. The
1224 * bio will be send to the device described by the bi_disk and bi_partno fields.
1225 *
1226 * The success/failure status of the request, along with notification of
1227 * completion, is delivered asynchronously through the ->bi_end_io() callback
1228 * in @bio. The bio must NOT be touched by thecaller until ->bi_end_io() has
1229 * been called.
1230 */
1231blk_qc_t submit_bio(struct bio *bio)
1232{
1233 if (blkcg_punt_bio_submit(bio))
1234 return BLK_QC_T_NONE;
1235
1236 /*
1237 * If it's a regular read/write or a barrier with data attached,
1238 * go through the normal accounting stuff before submission.
1239 */
1240 if (bio_has_data(bio)) {
1241 unsigned int count;
1242
1243 if (unlikely(bio_op(bio) == REQ_OP_WRITE_SAME))
1244 count = queue_logical_block_size(bio->bi_disk->queue) >> 9;
1245 else
1246 count = bio_sectors(bio);
1247
1248 if (op_is_write(bio_op(bio))) {
1249 count_vm_events(PGPGOUT, count);
1250 } else {
1251 task_io_account_read(bio->bi_iter.bi_size);
1252 count_vm_events(PGPGIN, count);
1253 }
1254
1255 if (unlikely(block_dump)) {
1256 char b[BDEVNAME_SIZE];
1257 printk(KERN_DEBUG "%s(%d): %s block %Lu on %s (%u sectors)\n",
1258 current->comm, task_pid_nr(current),
1259 op_is_write(bio_op(bio)) ? "WRITE" : "READ",
1260 (unsigned long long)bio->bi_iter.bi_sector,
1261 bio_devname(bio, b), count);
1262 }
1263 }
1264
1265 /*
1266 * If we're reading data that is part of the userspace workingset, count
1267 * submission time as memory stall. When the device is congested, or
1268 * the submitting cgroup IO-throttled, submission can be a significant
1269 * part of overall IO time.
1270 */
1271 if (unlikely(bio_op(bio) == REQ_OP_READ &&
1272 bio_flagged(bio, BIO_WORKINGSET))) {
1273 unsigned long pflags;
1274 blk_qc_t ret;
1275
1276 psi_memstall_enter(&pflags);
1277 ret = submit_bio_noacct(bio);
1278 psi_memstall_leave(&pflags);
1279
1280 return ret;
1281 }
1282
1283 return submit_bio_noacct(bio);
1284}
1285EXPORT_SYMBOL(submit_bio);
1286
1287/**
1288 * blk_cloned_rq_check_limits - Helper function to check a cloned request
1289 * for the new queue limits
1290 * @q: the queue
1291 * @rq: the request being checked
1292 *
1293 * Description:
1294 * @rq may have been made based on weaker limitations of upper-level queues
1295 * in request stacking drivers, and it may violate the limitation of @q.
1296 * Since the block layer and the underlying device driver trust @rq
1297 * after it is inserted to @q, it should be checked against @q before
1298 * the insertion using this generic function.
1299 *
1300 * Request stacking drivers like request-based dm may change the queue
1301 * limits when retrying requests on other queues. Those requests need
1302 * to be checked against the new queue limits again during dispatch.
1303 */
1304static int blk_cloned_rq_check_limits(struct request_queue *q,
1305 struct request *rq)
1306{
1307 if (blk_rq_sectors(rq) > blk_queue_get_max_sectors(q, req_op(rq))) {
1308 printk(KERN_ERR "%s: over max size limit. (%u > %u)\n",
1309 __func__, blk_rq_sectors(rq),
1310 blk_queue_get_max_sectors(q, req_op(rq)));
1311 return -EIO;
1312 }
1313
1314 /*
1315 * queue's settings related to segment counting like q->bounce_pfn
1316 * may differ from that of other stacking queues.
1317 * Recalculate it to check the request correctly on this queue's
1318 * limitation.
1319 */
1320 rq->nr_phys_segments = blk_recalc_rq_segments(rq);
1321 if (rq->nr_phys_segments > queue_max_segments(q)) {
1322 printk(KERN_ERR "%s: over max segments limit. (%hu > %hu)\n",
1323 __func__, rq->nr_phys_segments, queue_max_segments(q));
1324 return -EIO;
1325 }
1326
1327 return 0;
1328}
1329
1330/**
1331 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
1332 * @q: the queue to submit the request
1333 * @rq: the request being queued
1334 */
1335blk_status_t blk_insert_cloned_request(struct request_queue *q, struct request *rq)
1336{
1337 if (blk_cloned_rq_check_limits(q, rq))
1338 return BLK_STS_IOERR;
1339
1340 if (rq->rq_disk &&
1341 should_fail_request(&rq->rq_disk->part0, blk_rq_bytes(rq)))
1342 return BLK_STS_IOERR;
1343
1344 if (blk_crypto_insert_cloned_request(rq))
1345 return BLK_STS_IOERR;
1346
1347 if (blk_queue_io_stat(q))
1348 blk_account_io_start(rq);
1349
1350 /*
1351 * Since we have a scheduler attached on the top device,
1352 * bypass a potential scheduler on the bottom device for
1353 * insert.
1354 */
1355 return blk_mq_request_issue_directly(rq, true);
1356}
1357EXPORT_SYMBOL_GPL(blk_insert_cloned_request);
1358
1359/**
1360 * blk_rq_err_bytes - determine number of bytes till the next failure boundary
1361 * @rq: request to examine
1362 *
1363 * Description:
1364 * A request could be merge of IOs which require different failure
1365 * handling. This function determines the number of bytes which
1366 * can be failed from the beginning of the request without
1367 * crossing into area which need to be retried further.
1368 *
1369 * Return:
1370 * The number of bytes to fail.
1371 */
1372unsigned int blk_rq_err_bytes(const struct request *rq)
1373{
1374 unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK;
1375 unsigned int bytes = 0;
1376 struct bio *bio;
1377
1378 if (!(rq->rq_flags & RQF_MIXED_MERGE))
1379 return blk_rq_bytes(rq);
1380
1381 /*
1382 * Currently the only 'mixing' which can happen is between
1383 * different fastfail types. We can safely fail portions
1384 * which have all the failfast bits that the first one has -
1385 * the ones which are at least as eager to fail as the first
1386 * one.
1387 */
1388 for (bio = rq->bio; bio; bio = bio->bi_next) {
1389 if ((bio->bi_opf & ff) != ff)
1390 break;
1391 bytes += bio->bi_iter.bi_size;
1392 }
1393
1394 /* this could lead to infinite loop */
1395 BUG_ON(blk_rq_bytes(rq) && !bytes);
1396 return bytes;
1397}
1398EXPORT_SYMBOL_GPL(blk_rq_err_bytes);
1399
1400static void update_io_ticks(struct hd_struct *part, unsigned long now, bool end)
1401{
1402 unsigned long stamp;
1403again:
1404 stamp = READ_ONCE(part->stamp);
1405 if (unlikely(stamp != now)) {
1406 if (likely(cmpxchg(&part->stamp, stamp, now) == stamp))
1407 __part_stat_add(part, io_ticks, end ? now - stamp : 1);
1408 }
1409 if (part->partno) {
1410 part = &part_to_disk(part)->part0;
1411 goto again;
1412 }
1413}
1414
1415static void blk_account_io_completion(struct request *req, unsigned int bytes)
1416{
1417 if (req->part && blk_do_io_stat(req)) {
1418 const int sgrp = op_stat_group(req_op(req));
1419 struct hd_struct *part;
1420
1421 part_stat_lock();
1422 part = req->part;
1423 part_stat_add(part, sectors[sgrp], bytes >> 9);
1424 part_stat_unlock();
1425 }
1426}
1427
1428void blk_account_io_done(struct request *req, u64 now)
1429{
1430 /*
1431 * Account IO completion. flush_rq isn't accounted as a
1432 * normal IO on queueing nor completion. Accounting the
1433 * containing request is enough.
1434 */
1435 if (req->part && blk_do_io_stat(req) &&
1436 !(req->rq_flags & RQF_FLUSH_SEQ)) {
1437 const int sgrp = op_stat_group(req_op(req));
1438 struct hd_struct *part;
1439
1440 part_stat_lock();
1441 part = req->part;
1442
1443 update_io_ticks(part, jiffies, true);
1444 part_stat_inc(part, ios[sgrp]);
1445 part_stat_add(part, nsecs[sgrp], now - req->start_time_ns);
1446 part_stat_unlock();
1447
1448 hd_struct_put(part);
1449 }
1450}
1451
1452void blk_account_io_start(struct request *rq)
1453{
1454 if (!blk_do_io_stat(rq))
1455 return;
1456
1457 rq->part = disk_map_sector_rcu(rq->rq_disk, blk_rq_pos(rq));
1458
1459 part_stat_lock();
1460 update_io_ticks(rq->part, jiffies, false);
1461 part_stat_unlock();
1462}
1463
1464unsigned long disk_start_io_acct(struct gendisk *disk, unsigned int sectors,
1465 unsigned int op)
1466{
1467 struct hd_struct *part = &disk->part0;
1468 const int sgrp = op_stat_group(op);
1469 unsigned long now = READ_ONCE(jiffies);
1470
1471 part_stat_lock();
1472 update_io_ticks(part, now, false);
1473 part_stat_inc(part, ios[sgrp]);
1474 part_stat_add(part, sectors[sgrp], sectors);
1475 part_stat_local_inc(part, in_flight[op_is_write(op)]);
1476 part_stat_unlock();
1477
1478 return now;
1479}
1480EXPORT_SYMBOL(disk_start_io_acct);
1481
1482void disk_end_io_acct(struct gendisk *disk, unsigned int op,
1483 unsigned long start_time)
1484{
1485 struct hd_struct *part = &disk->part0;
1486 const int sgrp = op_stat_group(op);
1487 unsigned long now = READ_ONCE(jiffies);
1488 unsigned long duration = now - start_time;
1489
1490 part_stat_lock();
1491 update_io_ticks(part, now, true);
1492 part_stat_add(part, nsecs[sgrp], jiffies_to_nsecs(duration));
1493 part_stat_local_dec(part, in_flight[op_is_write(op)]);
1494 part_stat_unlock();
1495}
1496EXPORT_SYMBOL(disk_end_io_acct);
1497
1498/*
1499 * Steal bios from a request and add them to a bio list.
1500 * The request must not have been partially completed before.
1501 */
1502void blk_steal_bios(struct bio_list *list, struct request *rq)
1503{
1504 if (rq->bio) {
1505 if (list->tail)
1506 list->tail->bi_next = rq->bio;
1507 else
1508 list->head = rq->bio;
1509 list->tail = rq->biotail;
1510
1511 rq->bio = NULL;
1512 rq->biotail = NULL;
1513 }
1514
1515 rq->__data_len = 0;
1516}
1517EXPORT_SYMBOL_GPL(blk_steal_bios);
1518
1519/**
1520 * blk_update_request - Special helper function for request stacking drivers
1521 * @req: the request being processed
1522 * @error: block status code
1523 * @nr_bytes: number of bytes to complete @req
1524 *
1525 * Description:
1526 * Ends I/O on a number of bytes attached to @req, but doesn't complete
1527 * the request structure even if @req doesn't have leftover.
1528 * If @req has leftover, sets it up for the next range of segments.
1529 *
1530 * This special helper function is only for request stacking drivers
1531 * (e.g. request-based dm) so that they can handle partial completion.
1532 * Actual device drivers should use blk_mq_end_request instead.
1533 *
1534 * Passing the result of blk_rq_bytes() as @nr_bytes guarantees
1535 * %false return from this function.
1536 *
1537 * Note:
1538 * The RQF_SPECIAL_PAYLOAD flag is ignored on purpose in both
1539 * blk_rq_bytes() and in blk_update_request().
1540 *
1541 * Return:
1542 * %false - this request doesn't have any more data
1543 * %true - this request has more data
1544 **/
1545bool blk_update_request(struct request *req, blk_status_t error,
1546 unsigned int nr_bytes)
1547{
1548 int total_bytes;
1549
1550 trace_block_rq_complete(req, blk_status_to_errno(error), nr_bytes);
1551
1552 if (!req->bio)
1553 return false;
1554
1555#ifdef CONFIG_BLK_DEV_INTEGRITY
1556 if (blk_integrity_rq(req) && req_op(req) == REQ_OP_READ &&
1557 error == BLK_STS_OK)
1558 req->q->integrity.profile->complete_fn(req, nr_bytes);
1559#endif
1560
1561 if (unlikely(error && !blk_rq_is_passthrough(req) &&
1562 !(req->rq_flags & RQF_QUIET)))
1563 print_req_error(req, error, __func__);
1564
1565 blk_account_io_completion(req, nr_bytes);
1566
1567 total_bytes = 0;
1568 while (req->bio) {
1569 struct bio *bio = req->bio;
1570 unsigned bio_bytes = min(bio->bi_iter.bi_size, nr_bytes);
1571
1572 if (bio_bytes == bio->bi_iter.bi_size)
1573 req->bio = bio->bi_next;
1574
1575 /* Completion has already been traced */
1576 bio_clear_flag(bio, BIO_TRACE_COMPLETION);
1577 req_bio_endio(req, bio, bio_bytes, error);
1578
1579 total_bytes += bio_bytes;
1580 nr_bytes -= bio_bytes;
1581
1582 if (!nr_bytes)
1583 break;
1584 }
1585
1586 /*
1587 * completely done
1588 */
1589 if (!req->bio) {
1590 /*
1591 * Reset counters so that the request stacking driver
1592 * can find how many bytes remain in the request
1593 * later.
1594 */
1595 req->__data_len = 0;
1596 return false;
1597 }
1598
1599 req->__data_len -= total_bytes;
1600
1601 /* update sector only for requests with clear definition of sector */
1602 if (!blk_rq_is_passthrough(req))
1603 req->__sector += total_bytes >> 9;
1604
1605 /* mixed attributes always follow the first bio */
1606 if (req->rq_flags & RQF_MIXED_MERGE) {
1607 req->cmd_flags &= ~REQ_FAILFAST_MASK;
1608 req->cmd_flags |= req->bio->bi_opf & REQ_FAILFAST_MASK;
1609 }
1610
1611 if (!(req->rq_flags & RQF_SPECIAL_PAYLOAD)) {
1612 /*
1613 * If total number of sectors is less than the first segment
1614 * size, something has gone terribly wrong.
1615 */
1616 if (blk_rq_bytes(req) < blk_rq_cur_bytes(req)) {
1617 blk_dump_rq_flags(req, "request botched");
1618 req->__data_len = blk_rq_cur_bytes(req);
1619 }
1620
1621 /* recalculate the number of segments */
1622 req->nr_phys_segments = blk_recalc_rq_segments(req);
1623 }
1624
1625 return true;
1626}
1627EXPORT_SYMBOL_GPL(blk_update_request);
1628
1629#if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
1630/**
1631 * rq_flush_dcache_pages - Helper function to flush all pages in a request
1632 * @rq: the request to be flushed
1633 *
1634 * Description:
1635 * Flush all pages in @rq.
1636 */
1637void rq_flush_dcache_pages(struct request *rq)
1638{
1639 struct req_iterator iter;
1640 struct bio_vec bvec;
1641
1642 rq_for_each_segment(bvec, rq, iter)
1643 flush_dcache_page(bvec.bv_page);
1644}
1645EXPORT_SYMBOL_GPL(rq_flush_dcache_pages);
1646#endif
1647
1648/**
1649 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
1650 * @q : the queue of the device being checked
1651 *
1652 * Description:
1653 * Check if underlying low-level drivers of a device are busy.
1654 * If the drivers want to export their busy state, they must set own
1655 * exporting function using blk_queue_lld_busy() first.
1656 *
1657 * Basically, this function is used only by request stacking drivers
1658 * to stop dispatching requests to underlying devices when underlying
1659 * devices are busy. This behavior helps more I/O merging on the queue
1660 * of the request stacking driver and prevents I/O throughput regression
1661 * on burst I/O load.
1662 *
1663 * Return:
1664 * 0 - Not busy (The request stacking driver should dispatch request)
1665 * 1 - Busy (The request stacking driver should stop dispatching request)
1666 */
1667int blk_lld_busy(struct request_queue *q)
1668{
1669 if (queue_is_mq(q) && q->mq_ops->busy)
1670 return q->mq_ops->busy(q);
1671
1672 return 0;
1673}
1674EXPORT_SYMBOL_GPL(blk_lld_busy);
1675
1676/**
1677 * blk_rq_unprep_clone - Helper function to free all bios in a cloned request
1678 * @rq: the clone request to be cleaned up
1679 *
1680 * Description:
1681 * Free all bios in @rq for a cloned request.
1682 */
1683void blk_rq_unprep_clone(struct request *rq)
1684{
1685 struct bio *bio;
1686
1687 while ((bio = rq->bio) != NULL) {
1688 rq->bio = bio->bi_next;
1689
1690 bio_put(bio);
1691 }
1692}
1693EXPORT_SYMBOL_GPL(blk_rq_unprep_clone);
1694
1695/**
1696 * blk_rq_prep_clone - Helper function to setup clone request
1697 * @rq: the request to be setup
1698 * @rq_src: original request to be cloned
1699 * @bs: bio_set that bios for clone are allocated from
1700 * @gfp_mask: memory allocation mask for bio
1701 * @bio_ctr: setup function to be called for each clone bio.
1702 * Returns %0 for success, non %0 for failure.
1703 * @data: private data to be passed to @bio_ctr
1704 *
1705 * Description:
1706 * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq.
1707 * Also, pages which the original bios are pointing to are not copied
1708 * and the cloned bios just point same pages.
1709 * So cloned bios must be completed before original bios, which means
1710 * the caller must complete @rq before @rq_src.
1711 */
1712int blk_rq_prep_clone(struct request *rq, struct request *rq_src,
1713 struct bio_set *bs, gfp_t gfp_mask,
1714 int (*bio_ctr)(struct bio *, struct bio *, void *),
1715 void *data)
1716{
1717 struct bio *bio, *bio_src;
1718
1719 if (!bs)
1720 bs = &fs_bio_set;
1721
1722 __rq_for_each_bio(bio_src, rq_src) {
1723 bio = bio_clone_fast(bio_src, gfp_mask, bs);
1724 if (!bio)
1725 goto free_and_out;
1726
1727 if (bio_ctr && bio_ctr(bio, bio_src, data))
1728 goto free_and_out;
1729
1730 if (rq->bio) {
1731 rq->biotail->bi_next = bio;
1732 rq->biotail = bio;
1733 } else
1734 rq->bio = rq->biotail = bio;
1735 }
1736
1737 /* Copy attributes of the original request to the clone request. */
1738 rq->__sector = blk_rq_pos(rq_src);
1739 rq->__data_len = blk_rq_bytes(rq_src);
1740 if (rq_src->rq_flags & RQF_SPECIAL_PAYLOAD) {
1741 rq->rq_flags |= RQF_SPECIAL_PAYLOAD;
1742 rq->special_vec = rq_src->special_vec;
1743 }
1744 rq->nr_phys_segments = rq_src->nr_phys_segments;
1745 rq->ioprio = rq_src->ioprio;
1746
1747 if (rq->bio)
1748 blk_crypto_rq_bio_prep(rq, rq->bio, gfp_mask);
1749
1750 return 0;
1751
1752free_and_out:
1753 if (bio)
1754 bio_put(bio);
1755 blk_rq_unprep_clone(rq);
1756
1757 return -ENOMEM;
1758}
1759EXPORT_SYMBOL_GPL(blk_rq_prep_clone);
1760
1761int kblockd_schedule_work(struct work_struct *work)
1762{
1763 return queue_work(kblockd_workqueue, work);
1764}
1765EXPORT_SYMBOL(kblockd_schedule_work);
1766
1767int kblockd_mod_delayed_work_on(int cpu, struct delayed_work *dwork,
1768 unsigned long delay)
1769{
1770 return mod_delayed_work_on(cpu, kblockd_workqueue, dwork, delay);
1771}
1772EXPORT_SYMBOL(kblockd_mod_delayed_work_on);
1773
1774/**
1775 * blk_start_plug - initialize blk_plug and track it inside the task_struct
1776 * @plug: The &struct blk_plug that needs to be initialized
1777 *
1778 * Description:
1779 * blk_start_plug() indicates to the block layer an intent by the caller
1780 * to submit multiple I/O requests in a batch. The block layer may use
1781 * this hint to defer submitting I/Os from the caller until blk_finish_plug()
1782 * is called. However, the block layer may choose to submit requests
1783 * before a call to blk_finish_plug() if the number of queued I/Os
1784 * exceeds %BLK_MAX_REQUEST_COUNT, or if the size of the I/O is larger than
1785 * %BLK_PLUG_FLUSH_SIZE. The queued I/Os may also be submitted early if
1786 * the task schedules (see below).
1787 *
1788 * Tracking blk_plug inside the task_struct will help with auto-flushing the
1789 * pending I/O should the task end up blocking between blk_start_plug() and
1790 * blk_finish_plug(). This is important from a performance perspective, but
1791 * also ensures that we don't deadlock. For instance, if the task is blocking
1792 * for a memory allocation, memory reclaim could end up wanting to free a
1793 * page belonging to that request that is currently residing in our private
1794 * plug. By flushing the pending I/O when the process goes to sleep, we avoid
1795 * this kind of deadlock.
1796 */
1797void blk_start_plug(struct blk_plug *plug)
1798{
1799 struct task_struct *tsk = current;
1800
1801 /*
1802 * If this is a nested plug, don't actually assign it.
1803 */
1804 if (tsk->plug)
1805 return;
1806
1807 INIT_LIST_HEAD(&plug->mq_list);
1808 INIT_LIST_HEAD(&plug->cb_list);
1809 plug->rq_count = 0;
1810 plug->multiple_queues = false;
1811 plug->nowait = false;
1812
1813 /*
1814 * Store ordering should not be needed here, since a potential
1815 * preempt will imply a full memory barrier
1816 */
1817 tsk->plug = plug;
1818}
1819EXPORT_SYMBOL(blk_start_plug);
1820
1821static void flush_plug_callbacks(struct blk_plug *plug, bool from_schedule)
1822{
1823 LIST_HEAD(callbacks);
1824
1825 while (!list_empty(&plug->cb_list)) {
1826 list_splice_init(&plug->cb_list, &callbacks);
1827
1828 while (!list_empty(&callbacks)) {
1829 struct blk_plug_cb *cb = list_first_entry(&callbacks,
1830 struct blk_plug_cb,
1831 list);
1832 list_del(&cb->list);
1833 cb->callback(cb, from_schedule);
1834 }
1835 }
1836}
1837
1838struct blk_plug_cb *blk_check_plugged(blk_plug_cb_fn unplug, void *data,
1839 int size)
1840{
1841 struct blk_plug *plug = current->plug;
1842 struct blk_plug_cb *cb;
1843
1844 if (!plug)
1845 return NULL;
1846
1847 list_for_each_entry(cb, &plug->cb_list, list)
1848 if (cb->callback == unplug && cb->data == data)
1849 return cb;
1850
1851 /* Not currently on the callback list */
1852 BUG_ON(size < sizeof(*cb));
1853 cb = kzalloc(size, GFP_ATOMIC);
1854 if (cb) {
1855 cb->data = data;
1856 cb->callback = unplug;
1857 list_add(&cb->list, &plug->cb_list);
1858 }
1859 return cb;
1860}
1861EXPORT_SYMBOL(blk_check_plugged);
1862
1863void blk_flush_plug_list(struct blk_plug *plug, bool from_schedule)
1864{
1865 flush_plug_callbacks(plug, from_schedule);
1866
1867 if (!list_empty(&plug->mq_list))
1868 blk_mq_flush_plug_list(plug, from_schedule);
1869}
1870
1871/**
1872 * blk_finish_plug - mark the end of a batch of submitted I/O
1873 * @plug: The &struct blk_plug passed to blk_start_plug()
1874 *
1875 * Description:
1876 * Indicate that a batch of I/O submissions is complete. This function
1877 * must be paired with an initial call to blk_start_plug(). The intent
1878 * is to allow the block layer to optimize I/O submission. See the
1879 * documentation for blk_start_plug() for more information.
1880 */
1881void blk_finish_plug(struct blk_plug *plug)
1882{
1883 if (plug != current->plug)
1884 return;
1885 blk_flush_plug_list(plug, false);
1886
1887 current->plug = NULL;
1888}
1889EXPORT_SYMBOL(blk_finish_plug);
1890
1891void blk_io_schedule(void)
1892{
1893 /* Prevent hang_check timer from firing at us during very long I/O */
1894 unsigned long timeout = sysctl_hung_task_timeout_secs * HZ / 2;
1895
1896 if (timeout)
1897 io_schedule_timeout(timeout);
1898 else
1899 io_schedule();
1900}
1901EXPORT_SYMBOL_GPL(blk_io_schedule);
1902
1903int __init blk_dev_init(void)
1904{
1905 BUILD_BUG_ON(REQ_OP_LAST >= (1 << REQ_OP_BITS));
1906 BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 *
1907 sizeof_field(struct request, cmd_flags));
1908 BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 *
1909 sizeof_field(struct bio, bi_opf));
1910
1911 /* used for unplugging and affects IO latency/throughput - HIGHPRI */
1912 kblockd_workqueue = alloc_workqueue("kblockd",
1913 WQ_MEM_RECLAIM | WQ_HIGHPRI, 0);
1914 if (!kblockd_workqueue)
1915 panic("Failed to create kblockd\n");
1916
1917 blk_requestq_cachep = kmem_cache_create("request_queue",
1918 sizeof(struct request_queue), 0, SLAB_PANIC, NULL);
1919
1920 blk_debugfs_root = debugfs_create_dir("block", NULL);
1921
1922 return 0;
1923}