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 #include "blk-ioprio.h"
53
54 struct dentry *blk_debugfs_root;
55
56 EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_remap);
57 EXPORT_TRACEPOINT_SYMBOL_GPL(block_rq_remap);
58 EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_complete);
59 EXPORT_TRACEPOINT_SYMBOL_GPL(block_split);
60 EXPORT_TRACEPOINT_SYMBOL_GPL(block_unplug);
61 EXPORT_TRACEPOINT_SYMBOL_GPL(block_rq_insert);
62
63 static DEFINE_IDA(blk_queue_ida);
64
65 /*
66 * For queue allocation
67 */
68 static struct kmem_cache *blk_requestq_cachep;
69
70 /*
71 * Controlling structure to kblockd
72 */
73 static struct workqueue_struct *kblockd_workqueue;
74
75 /**
76 * blk_queue_flag_set - atomically set a queue flag
77 * @flag: flag to be set
78 * @q: request queue
79 */
80 void blk_queue_flag_set(unsigned int flag, struct request_queue *q)
81 {
82 set_bit(flag, &q->queue_flags);
83 }
84 EXPORT_SYMBOL(blk_queue_flag_set);
85
86 /**
87 * blk_queue_flag_clear - atomically clear a queue flag
88 * @flag: flag to be cleared
89 * @q: request queue
90 */
91 void blk_queue_flag_clear(unsigned int flag, struct request_queue *q)
92 {
93 clear_bit(flag, &q->queue_flags);
94 }
95 EXPORT_SYMBOL(blk_queue_flag_clear);
96
97 #define REQ_OP_NAME(name) [REQ_OP_##name] = #name
98 static const char *const blk_op_name[] = {
99 REQ_OP_NAME(READ),
100 REQ_OP_NAME(WRITE),
101 REQ_OP_NAME(FLUSH),
102 REQ_OP_NAME(DISCARD),
103 REQ_OP_NAME(SECURE_ERASE),
104 REQ_OP_NAME(ZONE_RESET),
105 REQ_OP_NAME(ZONE_RESET_ALL),
106 REQ_OP_NAME(ZONE_OPEN),
107 REQ_OP_NAME(ZONE_CLOSE),
108 REQ_OP_NAME(ZONE_FINISH),
109 REQ_OP_NAME(ZONE_APPEND),
110 REQ_OP_NAME(WRITE_ZEROES),
111 REQ_OP_NAME(DRV_IN),
112 REQ_OP_NAME(DRV_OUT),
113 };
114 #undef REQ_OP_NAME
115
116 /**
117 * blk_op_str - Return string XXX in the REQ_OP_XXX.
118 * @op: REQ_OP_XXX.
119 *
120 * Description: Centralize block layer function to convert REQ_OP_XXX into
121 * string format. Useful in the debugging and tracing bio or request. For
122 * invalid REQ_OP_XXX it returns string "UNKNOWN".
123 */
124 inline const char *blk_op_str(enum req_op op)
125 {
126 const char *op_str = "UNKNOWN";
127
128 if (op < ARRAY_SIZE(blk_op_name) && blk_op_name[op])
129 op_str = blk_op_name[op];
130
131 return op_str;
132 }
133 EXPORT_SYMBOL_GPL(blk_op_str);
134
135 static const struct {
136 int errno;
137 const char *name;
138 } blk_errors[] = {
139 [BLK_STS_OK] = { 0, "" },
140 [BLK_STS_NOTSUPP] = { -EOPNOTSUPP, "operation not supported" },
141 [BLK_STS_TIMEOUT] = { -ETIMEDOUT, "timeout" },
142 [BLK_STS_NOSPC] = { -ENOSPC, "critical space allocation" },
143 [BLK_STS_TRANSPORT] = { -ENOLINK, "recoverable transport" },
144 [BLK_STS_TARGET] = { -EREMOTEIO, "critical target" },
145 [BLK_STS_RESV_CONFLICT] = { -EBADE, "reservation conflict" },
146 [BLK_STS_MEDIUM] = { -ENODATA, "critical medium" },
147 [BLK_STS_PROTECTION] = { -EILSEQ, "protection" },
148 [BLK_STS_RESOURCE] = { -ENOMEM, "kernel resource" },
149 [BLK_STS_DEV_RESOURCE] = { -EBUSY, "device resource" },
150 [BLK_STS_AGAIN] = { -EAGAIN, "nonblocking retry" },
151 [BLK_STS_OFFLINE] = { -ENODEV, "device offline" },
152
153 /* device mapper special case, should not leak out: */
154 [BLK_STS_DM_REQUEUE] = { -EREMCHG, "dm internal retry" },
155
156 /* zone device specific errors */
157 [BLK_STS_ZONE_OPEN_RESOURCE] = { -ETOOMANYREFS, "open zones exceeded" },
158 [BLK_STS_ZONE_ACTIVE_RESOURCE] = { -EOVERFLOW, "active zones exceeded" },
159
160 /* Command duration limit device-side timeout */
161 [BLK_STS_DURATION_LIMIT] = { -ETIME, "duration limit exceeded" },
162
163 [BLK_STS_INVAL] = { -EINVAL, "invalid" },
164
165 /* everything else not covered above: */
166 [BLK_STS_IOERR] = { -EIO, "I/O" },
167 };
168
169 blk_status_t errno_to_blk_status(int errno)
170 {
171 int i;
172
173 for (i = 0; i < ARRAY_SIZE(blk_errors); i++) {
174 if (blk_errors[i].errno == errno)
175 return (__force blk_status_t)i;
176 }
177
178 return BLK_STS_IOERR;
179 }
180 EXPORT_SYMBOL_GPL(errno_to_blk_status);
181
182 int blk_status_to_errno(blk_status_t status)
183 {
184 int idx = (__force int)status;
185
186 if (WARN_ON_ONCE(idx >= ARRAY_SIZE(blk_errors)))
187 return -EIO;
188 return blk_errors[idx].errno;
189 }
190 EXPORT_SYMBOL_GPL(blk_status_to_errno);
191
192 const char *blk_status_to_str(blk_status_t status)
193 {
194 int idx = (__force int)status;
195
196 if (WARN_ON_ONCE(idx >= ARRAY_SIZE(blk_errors)))
197 return "<null>";
198 return blk_errors[idx].name;
199 }
200 EXPORT_SYMBOL_GPL(blk_status_to_str);
201
202 /**
203 * blk_sync_queue - cancel any pending callbacks on a queue
204 * @q: the queue
205 *
206 * Description:
207 * The block layer may perform asynchronous callback activity
208 * on a queue, such as calling the unplug function after a timeout.
209 * A block device may call blk_sync_queue to ensure that any
210 * such activity is cancelled, thus allowing it to release resources
211 * that the callbacks might use. The caller must already have made sure
212 * that its ->submit_bio will not re-add plugging prior to calling
213 * this function.
214 *
215 * This function does not cancel any asynchronous activity arising
216 * out of elevator or throttling code. That would require elevator_exit()
217 * and blkcg_exit_queue() to be called with queue lock initialized.
218 *
219 */
220 void blk_sync_queue(struct request_queue *q)
221 {
222 del_timer_sync(&q->timeout);
223 cancel_work_sync(&q->timeout_work);
224 }
225 EXPORT_SYMBOL(blk_sync_queue);
226
227 /**
228 * blk_set_pm_only - increment pm_only counter
229 * @q: request queue pointer
230 */
231 void blk_set_pm_only(struct request_queue *q)
232 {
233 atomic_inc(&q->pm_only);
234 }
235 EXPORT_SYMBOL_GPL(blk_set_pm_only);
236
237 void blk_clear_pm_only(struct request_queue *q)
238 {
239 int pm_only;
240
241 pm_only = atomic_dec_return(&q->pm_only);
242 WARN_ON_ONCE(pm_only < 0);
243 if (pm_only == 0)
244 wake_up_all(&q->mq_freeze_wq);
245 }
246 EXPORT_SYMBOL_GPL(blk_clear_pm_only);
247
248 static void blk_free_queue_rcu(struct rcu_head *rcu_head)
249 {
250 struct request_queue *q = container_of(rcu_head,
251 struct request_queue, rcu_head);
252
253 percpu_ref_exit(&q->q_usage_counter);
254 kmem_cache_free(blk_requestq_cachep, q);
255 }
256
257 static void blk_free_queue(struct request_queue *q)
258 {
259 blk_free_queue_stats(q->stats);
260 if (queue_is_mq(q))
261 blk_mq_release(q);
262
263 ida_free(&blk_queue_ida, q->id);
264 call_rcu(&q->rcu_head, blk_free_queue_rcu);
265 }
266
267 /**
268 * blk_put_queue - decrement the request_queue refcount
269 * @q: the request_queue structure to decrement the refcount for
270 *
271 * Decrements the refcount of the request_queue and free it when the refcount
272 * reaches 0.
273 */
274 void blk_put_queue(struct request_queue *q)
275 {
276 if (refcount_dec_and_test(&q->refs))
277 blk_free_queue(q);
278 }
279 EXPORT_SYMBOL(blk_put_queue);
280
281 void blk_queue_start_drain(struct request_queue *q)
282 {
283 /*
284 * When queue DYING flag is set, we need to block new req
285 * entering queue, so we call blk_freeze_queue_start() to
286 * prevent I/O from crossing blk_queue_enter().
287 */
288 blk_freeze_queue_start(q);
289 if (queue_is_mq(q))
290 blk_mq_wake_waiters(q);
291 /* Make blk_queue_enter() reexamine the DYING flag. */
292 wake_up_all(&q->mq_freeze_wq);
293 }
294
295 /**
296 * blk_queue_enter() - try to increase q->q_usage_counter
297 * @q: request queue pointer
298 * @flags: BLK_MQ_REQ_NOWAIT and/or BLK_MQ_REQ_PM
299 */
300 int blk_queue_enter(struct request_queue *q, blk_mq_req_flags_t flags)
301 {
302 const bool pm = flags & BLK_MQ_REQ_PM;
303
304 while (!blk_try_enter_queue(q, pm)) {
305 if (flags & BLK_MQ_REQ_NOWAIT)
306 return -EAGAIN;
307
308 /*
309 * read pair of barrier in blk_freeze_queue_start(), we need to
310 * order reading __PERCPU_REF_DEAD flag of .q_usage_counter and
311 * reading .mq_freeze_depth or queue dying flag, otherwise the
312 * following wait may never return if the two reads are
313 * reordered.
314 */
315 smp_rmb();
316 wait_event(q->mq_freeze_wq,
317 (!q->mq_freeze_depth &&
318 blk_pm_resume_queue(pm, q)) ||
319 blk_queue_dying(q));
320 if (blk_queue_dying(q))
321 return -ENODEV;
322 }
323
324 return 0;
325 }
326
327 int __bio_queue_enter(struct request_queue *q, struct bio *bio)
328 {
329 while (!blk_try_enter_queue(q, false)) {
330 struct gendisk *disk = bio->bi_bdev->bd_disk;
331
332 if (bio->bi_opf & REQ_NOWAIT) {
333 if (test_bit(GD_DEAD, &disk->state))
334 goto dead;
335 bio_wouldblock_error(bio);
336 return -EAGAIN;
337 }
338
339 /*
340 * read pair of barrier in blk_freeze_queue_start(), we need to
341 * order reading __PERCPU_REF_DEAD flag of .q_usage_counter and
342 * reading .mq_freeze_depth or queue dying flag, otherwise the
343 * following wait may never return if the two reads are
344 * reordered.
345 */
346 smp_rmb();
347 wait_event(q->mq_freeze_wq,
348 (!q->mq_freeze_depth &&
349 blk_pm_resume_queue(false, q)) ||
350 test_bit(GD_DEAD, &disk->state));
351 if (test_bit(GD_DEAD, &disk->state))
352 goto dead;
353 }
354
355 return 0;
356 dead:
357 bio_io_error(bio);
358 return -ENODEV;
359 }
360
361 void blk_queue_exit(struct request_queue *q)
362 {
363 percpu_ref_put(&q->q_usage_counter);
364 }
365
366 static void blk_queue_usage_counter_release(struct percpu_ref *ref)
367 {
368 struct request_queue *q =
369 container_of(ref, struct request_queue, q_usage_counter);
370
371 wake_up_all(&q->mq_freeze_wq);
372 }
373
374 static void blk_rq_timed_out_timer(struct timer_list *t)
375 {
376 struct request_queue *q = from_timer(q, t, timeout);
377
378 kblockd_schedule_work(&q->timeout_work);
379 }
380
381 static void blk_timeout_work(struct work_struct *work)
382 {
383 }
384
385 struct request_queue *blk_alloc_queue(struct queue_limits *lim, int node_id)
386 {
387 struct request_queue *q;
388 int error;
389
390 q = kmem_cache_alloc_node(blk_requestq_cachep, GFP_KERNEL | __GFP_ZERO,
391 node_id);
392 if (!q)
393 return ERR_PTR(-ENOMEM);
394
395 q->last_merge = NULL;
396
397 q->id = ida_alloc(&blk_queue_ida, GFP_KERNEL);
398 if (q->id < 0) {
399 error = q->id;
400 goto fail_q;
401 }
402
403 q->stats = blk_alloc_queue_stats();
404 if (!q->stats) {
405 error = -ENOMEM;
406 goto fail_id;
407 }
408
409 error = blk_set_default_limits(lim);
410 if (error)
411 goto fail_stats;
412 q->limits = *lim;
413
414 q->node = node_id;
415
416 atomic_set(&q->nr_active_requests_shared_tags, 0);
417
418 timer_setup(&q->timeout, blk_rq_timed_out_timer, 0);
419 INIT_WORK(&q->timeout_work, blk_timeout_work);
420 INIT_LIST_HEAD(&q->icq_list);
421
422 refcount_set(&q->refs, 1);
423 mutex_init(&q->debugfs_mutex);
424 mutex_init(&q->sysfs_lock);
425 mutex_init(&q->sysfs_dir_lock);
426 mutex_init(&q->limits_lock);
427 mutex_init(&q->rq_qos_mutex);
428 spin_lock_init(&q->queue_lock);
429
430 init_waitqueue_head(&q->mq_freeze_wq);
431 mutex_init(&q->mq_freeze_lock);
432
433 blkg_init_queue(q);
434
435 /*
436 * Init percpu_ref in atomic mode so that it's faster to shutdown.
437 * See blk_register_queue() for details.
438 */
439 error = percpu_ref_init(&q->q_usage_counter,
440 blk_queue_usage_counter_release,
441 PERCPU_REF_INIT_ATOMIC, GFP_KERNEL);
442 if (error)
443 goto fail_stats;
444
445 q->nr_requests = BLKDEV_DEFAULT_RQ;
446
447 return q;
448
449 fail_stats:
450 blk_free_queue_stats(q->stats);
451 fail_id:
452 ida_free(&blk_queue_ida, q->id);
453 fail_q:
454 kmem_cache_free(blk_requestq_cachep, q);
455 return ERR_PTR(error);
456 }
457
458 /**
459 * blk_get_queue - increment the request_queue refcount
460 * @q: the request_queue structure to increment the refcount for
461 *
462 * Increment the refcount of the request_queue kobject.
463 *
464 * Context: Any context.
465 */
466 bool blk_get_queue(struct request_queue *q)
467 {
468 if (unlikely(blk_queue_dying(q)))
469 return false;
470 refcount_inc(&q->refs);
471 return true;
472 }
473 EXPORT_SYMBOL(blk_get_queue);
474
475 #ifdef CONFIG_FAIL_MAKE_REQUEST
476
477 static DECLARE_FAULT_ATTR(fail_make_request);
478
479 static int __init setup_fail_make_request(char *str)
480 {
481 return setup_fault_attr(&fail_make_request, str);
482 }
483 __setup("fail_make_request=", setup_fail_make_request);
484
485 bool should_fail_request(struct block_device *part, unsigned int bytes)
486 {
487 return bdev_test_flag(part, BD_MAKE_IT_FAIL) &&
488 should_fail(&fail_make_request, bytes);
489 }
490
491 static int __init fail_make_request_debugfs(void)
492 {
493 struct dentry *dir = fault_create_debugfs_attr("fail_make_request",
494 NULL, &fail_make_request);
495
496 return PTR_ERR_OR_ZERO(dir);
497 }
498
499 late_initcall(fail_make_request_debugfs);
500 #endif /* CONFIG_FAIL_MAKE_REQUEST */
501
502 static inline void bio_check_ro(struct bio *bio)
503 {
504 if (op_is_write(bio_op(bio)) && bdev_read_only(bio->bi_bdev)) {
505 if (op_is_flush(bio->bi_opf) && !bio_sectors(bio))
506 return;
507
508 if (bdev_test_flag(bio->bi_bdev, BD_RO_WARNED))
509 return;
510
511 bdev_set_flag(bio->bi_bdev, BD_RO_WARNED);
512
513 /*
514 * Use ioctl to set underlying disk of raid/dm to read-only
515 * will trigger this.
516 */
517 pr_warn("Trying to write to read-only block-device %pg\n",
518 bio->bi_bdev);
519 }
520 }
521
522 static noinline int should_fail_bio(struct bio *bio)
523 {
524 if (should_fail_request(bdev_whole(bio->bi_bdev), bio->bi_iter.bi_size))
525 return -EIO;
526 return 0;
527 }
528 ALLOW_ERROR_INJECTION(should_fail_bio, ERRNO);
529
530 /*
531 * Check whether this bio extends beyond the end of the device or partition.
532 * This may well happen - the kernel calls bread() without checking the size of
533 * the device, e.g., when mounting a file system.
534 */
535 static inline int bio_check_eod(struct bio *bio)
536 {
537 sector_t maxsector = bdev_nr_sectors(bio->bi_bdev);
538 unsigned int nr_sectors = bio_sectors(bio);
539
540 if (nr_sectors &&
541 (nr_sectors > maxsector ||
542 bio->bi_iter.bi_sector > maxsector - nr_sectors)) {
543 pr_info_ratelimited("%s: attempt to access beyond end of device\n"
544 "%pg: rw=%d, sector=%llu, nr_sectors = %u limit=%llu\n",
545 current->comm, bio->bi_bdev, bio->bi_opf,
546 bio->bi_iter.bi_sector, nr_sectors, maxsector);
547 return -EIO;
548 }
549 return 0;
550 }
551
552 /*
553 * Remap block n of partition p to block n+start(p) of the disk.
554 */
555 static int blk_partition_remap(struct bio *bio)
556 {
557 struct block_device *p = bio->bi_bdev;
558
559 if (unlikely(should_fail_request(p, bio->bi_iter.bi_size)))
560 return -EIO;
561 if (bio_sectors(bio)) {
562 bio->bi_iter.bi_sector += p->bd_start_sect;
563 trace_block_bio_remap(bio, p->bd_dev,
564 bio->bi_iter.bi_sector -
565 p->bd_start_sect);
566 }
567 bio_set_flag(bio, BIO_REMAPPED);
568 return 0;
569 }
570
571 /*
572 * Check write append to a zoned block device.
573 */
574 static inline blk_status_t blk_check_zone_append(struct request_queue *q,
575 struct bio *bio)
576 {
577 int nr_sectors = bio_sectors(bio);
578
579 /* Only applicable to zoned block devices */
580 if (!bdev_is_zoned(bio->bi_bdev))
581 return BLK_STS_NOTSUPP;
582
583 /* The bio sector must point to the start of a sequential zone */
584 if (!bdev_is_zone_start(bio->bi_bdev, bio->bi_iter.bi_sector))
585 return BLK_STS_IOERR;
586
587 /*
588 * Not allowed to cross zone boundaries. Otherwise, the BIO will be
589 * split and could result in non-contiguous sectors being written in
590 * different zones.
591 */
592 if (nr_sectors > q->limits.chunk_sectors)
593 return BLK_STS_IOERR;
594
595 /* Make sure the BIO is small enough and will not get split */
596 if (nr_sectors > queue_max_zone_append_sectors(q))
597 return BLK_STS_IOERR;
598
599 bio->bi_opf |= REQ_NOMERGE;
600
601 return BLK_STS_OK;
602 }
603
604 static void __submit_bio(struct bio *bio)
605 {
606 /* If plug is not used, add new plug here to cache nsecs time. */
607 struct blk_plug plug;
608
609 if (unlikely(!blk_crypto_bio_prep(&bio)))
610 return;
611
612 blk_start_plug(&plug);
613
614 if (!bdev_test_flag(bio->bi_bdev, BD_HAS_SUBMIT_BIO)) {
615 blk_mq_submit_bio(bio);
616 } else if (likely(bio_queue_enter(bio) == 0)) {
617 struct gendisk *disk = bio->bi_bdev->bd_disk;
618
619 disk->fops->submit_bio(bio);
620 blk_queue_exit(disk->queue);
621 }
622
623 blk_finish_plug(&plug);
624 }
625
626 /*
627 * The loop in this function may be a bit non-obvious, and so deserves some
628 * explanation:
629 *
630 * - Before entering the loop, bio->bi_next is NULL (as all callers ensure
631 * that), so we have a list with a single bio.
632 * - We pretend that we have just taken it off a longer list, so we assign
633 * bio_list to a pointer to the bio_list_on_stack, thus initialising the
634 * bio_list of new bios to be added. ->submit_bio() may indeed add some more
635 * bios through a recursive call to submit_bio_noacct. If it did, we find a
636 * non-NULL value in bio_list and re-enter the loop from the top.
637 * - In this case we really did just take the bio of the top of the list (no
638 * pretending) and so remove it from bio_list, and call into ->submit_bio()
639 * again.
640 *
641 * bio_list_on_stack[0] contains bios submitted by the current ->submit_bio.
642 * bio_list_on_stack[1] contains bios that were submitted before the current
643 * ->submit_bio, but that haven't been processed yet.
644 */
645 static void __submit_bio_noacct(struct bio *bio)
646 {
647 struct bio_list bio_list_on_stack[2];
648
649 BUG_ON(bio->bi_next);
650
651 bio_list_init(&bio_list_on_stack[0]);
652 current->bio_list = bio_list_on_stack;
653
654 do {
655 struct request_queue *q = bdev_get_queue(bio->bi_bdev);
656 struct bio_list lower, same;
657
658 /*
659 * Create a fresh bio_list for all subordinate requests.
660 */
661 bio_list_on_stack[1] = bio_list_on_stack[0];
662 bio_list_init(&bio_list_on_stack[0]);
663
664 __submit_bio(bio);
665
666 /*
667 * Sort new bios into those for a lower level and those for the
668 * same level.
669 */
670 bio_list_init(&lower);
671 bio_list_init(&same);
672 while ((bio = bio_list_pop(&bio_list_on_stack[0])) != NULL)
673 if (q == bdev_get_queue(bio->bi_bdev))
674 bio_list_add(&same, bio);
675 else
676 bio_list_add(&lower, bio);
677
678 /*
679 * Now assemble so we handle the lowest level first.
680 */
681 bio_list_merge(&bio_list_on_stack[0], &lower);
682 bio_list_merge(&bio_list_on_stack[0], &same);
683 bio_list_merge(&bio_list_on_stack[0], &bio_list_on_stack[1]);
684 } while ((bio = bio_list_pop(&bio_list_on_stack[0])));
685
686 current->bio_list = NULL;
687 }
688
689 static void __submit_bio_noacct_mq(struct bio *bio)
690 {
691 struct bio_list bio_list[2] = { };
692
693 current->bio_list = bio_list;
694
695 do {
696 __submit_bio(bio);
697 } while ((bio = bio_list_pop(&bio_list[0])));
698
699 current->bio_list = NULL;
700 }
701
702 void submit_bio_noacct_nocheck(struct bio *bio)
703 {
704 blk_cgroup_bio_start(bio);
705 blkcg_bio_issue_init(bio);
706
707 if (!bio_flagged(bio, BIO_TRACE_COMPLETION)) {
708 trace_block_bio_queue(bio);
709 /*
710 * Now that enqueuing has been traced, we need to trace
711 * completion as well.
712 */
713 bio_set_flag(bio, BIO_TRACE_COMPLETION);
714 }
715
716 /*
717 * We only want one ->submit_bio to be active at a time, else stack
718 * usage with stacked devices could be a problem. Use current->bio_list
719 * to collect a list of requests submited by a ->submit_bio method while
720 * it is active, and then process them after it returned.
721 */
722 if (current->bio_list)
723 bio_list_add(¤t->bio_list[0], bio);
724 else if (!bdev_test_flag(bio->bi_bdev, BD_HAS_SUBMIT_BIO))
725 __submit_bio_noacct_mq(bio);
726 else
727 __submit_bio_noacct(bio);
728 }
729
730 static blk_status_t blk_validate_atomic_write_op_size(struct request_queue *q,
731 struct bio *bio)
732 {
733 if (bio->bi_iter.bi_size > queue_atomic_write_unit_max_bytes(q))
734 return BLK_STS_INVAL;
735
736 if (bio->bi_iter.bi_size % queue_atomic_write_unit_min_bytes(q))
737 return BLK_STS_INVAL;
738
739 return BLK_STS_OK;
740 }
741
742 /**
743 * submit_bio_noacct - re-submit a bio to the block device layer for I/O
744 * @bio: The bio describing the location in memory and on the device.
745 *
746 * This is a version of submit_bio() that shall only be used for I/O that is
747 * resubmitted to lower level drivers by stacking block drivers. All file
748 * systems and other upper level users of the block layer should use
749 * submit_bio() instead.
750 */
751 void submit_bio_noacct(struct bio *bio)
752 {
753 struct block_device *bdev = bio->bi_bdev;
754 struct request_queue *q = bdev_get_queue(bdev);
755 blk_status_t status = BLK_STS_IOERR;
756
757 might_sleep();
758
759 /*
760 * For a REQ_NOWAIT based request, return -EOPNOTSUPP
761 * if queue does not support NOWAIT.
762 */
763 if ((bio->bi_opf & REQ_NOWAIT) && !bdev_nowait(bdev))
764 goto not_supported;
765
766 if (should_fail_bio(bio))
767 goto end_io;
768 bio_check_ro(bio);
769 if (!bio_flagged(bio, BIO_REMAPPED)) {
770 if (unlikely(bio_check_eod(bio)))
771 goto end_io;
772 if (bdev_is_partition(bdev) &&
773 unlikely(blk_partition_remap(bio)))
774 goto end_io;
775 }
776
777 /*
778 * Filter flush bio's early so that bio based drivers without flush
779 * support don't have to worry about them.
780 */
781 if (op_is_flush(bio->bi_opf)) {
782 if (WARN_ON_ONCE(bio_op(bio) != REQ_OP_WRITE &&
783 bio_op(bio) != REQ_OP_ZONE_APPEND))
784 goto end_io;
785 if (!bdev_write_cache(bdev)) {
786 bio->bi_opf &= ~(REQ_PREFLUSH | REQ_FUA);
787 if (!bio_sectors(bio)) {
788 status = BLK_STS_OK;
789 goto end_io;
790 }
791 }
792 }
793
794 if (!(q->limits.features & BLK_FEAT_POLL) &&
795 (bio->bi_opf & REQ_POLLED)) {
796 bio_clear_polled(bio);
797 goto not_supported;
798 }
799
800 switch (bio_op(bio)) {
801 case REQ_OP_READ:
802 break;
803 case REQ_OP_WRITE:
804 if (bio->bi_opf & REQ_ATOMIC) {
805 status = blk_validate_atomic_write_op_size(q, bio);
806 if (status != BLK_STS_OK)
807 goto end_io;
808 }
809 break;
810 case REQ_OP_FLUSH:
811 /*
812 * REQ_OP_FLUSH can't be submitted through bios, it is only
813 * synthetized in struct request by the flush state machine.
814 */
815 goto not_supported;
816 case REQ_OP_DISCARD:
817 if (!bdev_max_discard_sectors(bdev))
818 goto not_supported;
819 break;
820 case REQ_OP_SECURE_ERASE:
821 if (!bdev_max_secure_erase_sectors(bdev))
822 goto not_supported;
823 break;
824 case REQ_OP_ZONE_APPEND:
825 status = blk_check_zone_append(q, bio);
826 if (status != BLK_STS_OK)
827 goto end_io;
828 break;
829 case REQ_OP_WRITE_ZEROES:
830 if (!q->limits.max_write_zeroes_sectors)
831 goto not_supported;
832 break;
833 case REQ_OP_ZONE_RESET:
834 case REQ_OP_ZONE_OPEN:
835 case REQ_OP_ZONE_CLOSE:
836 case REQ_OP_ZONE_FINISH:
837 case REQ_OP_ZONE_RESET_ALL:
838 if (!bdev_is_zoned(bio->bi_bdev))
839 goto not_supported;
840 break;
841 case REQ_OP_DRV_IN:
842 case REQ_OP_DRV_OUT:
843 /*
844 * Driver private operations are only used with passthrough
845 * requests.
846 */
847 fallthrough;
848 default:
849 goto not_supported;
850 }
851
852 if (blk_throtl_bio(bio))
853 return;
854 submit_bio_noacct_nocheck(bio);
855 return;
856
857 not_supported:
858 status = BLK_STS_NOTSUPP;
859 end_io:
860 bio->bi_status = status;
861 bio_endio(bio);
862 }
863 EXPORT_SYMBOL(submit_bio_noacct);
864
865 static void bio_set_ioprio(struct bio *bio)
866 {
867 /* Nobody set ioprio so far? Initialize it based on task's nice value */
868 if (IOPRIO_PRIO_CLASS(bio->bi_ioprio) == IOPRIO_CLASS_NONE)
869 bio->bi_ioprio = get_current_ioprio();
870 blkcg_set_ioprio(bio);
871 }
872
873 /**
874 * submit_bio - submit a bio to the block device layer for I/O
875 * @bio: The &struct bio which describes the I/O
876 *
877 * submit_bio() is used to submit I/O requests to block devices. It is passed a
878 * fully set up &struct bio that describes the I/O that needs to be done. The
879 * bio will be send to the device described by the bi_bdev field.
880 *
881 * The success/failure status of the request, along with notification of
882 * completion, is delivered asynchronously through the ->bi_end_io() callback
883 * in @bio. The bio must NOT be touched by the caller until ->bi_end_io() has
884 * been called.
885 */
886 void submit_bio(struct bio *bio)
887 {
888 if (bio_op(bio) == REQ_OP_READ) {
889 task_io_account_read(bio->bi_iter.bi_size);
890 count_vm_events(PGPGIN, bio_sectors(bio));
891 } else if (bio_op(bio) == REQ_OP_WRITE) {
892 count_vm_events(PGPGOUT, bio_sectors(bio));
893 }
894
895 bio_set_ioprio(bio);
896 submit_bio_noacct(bio);
897 }
898 EXPORT_SYMBOL(submit_bio);
899
900 /**
901 * bio_poll - poll for BIO completions
902 * @bio: bio to poll for
903 * @iob: batches of IO
904 * @flags: BLK_POLL_* flags that control the behavior
905 *
906 * Poll for completions on queue associated with the bio. Returns number of
907 * completed entries found.
908 *
909 * Note: the caller must either be the context that submitted @bio, or
910 * be in a RCU critical section to prevent freeing of @bio.
911 */
912 int bio_poll(struct bio *bio, struct io_comp_batch *iob, unsigned int flags)
913 {
914 blk_qc_t cookie = READ_ONCE(bio->bi_cookie);
915 struct block_device *bdev;
916 struct request_queue *q;
917 int ret = 0;
918
919 bdev = READ_ONCE(bio->bi_bdev);
920 if (!bdev)
921 return 0;
922
923 q = bdev_get_queue(bdev);
924 if (cookie == BLK_QC_T_NONE || !(q->limits.features & BLK_FEAT_POLL))
925 return 0;
926
927 blk_flush_plug(current->plug, false);
928
929 /*
930 * We need to be able to enter a frozen queue, similar to how
931 * timeouts also need to do that. If that is blocked, then we can
932 * have pending IO when a queue freeze is started, and then the
933 * wait for the freeze to finish will wait for polled requests to
934 * timeout as the poller is preventer from entering the queue and
935 * completing them. As long as we prevent new IO from being queued,
936 * that should be all that matters.
937 */
938 if (!percpu_ref_tryget(&q->q_usage_counter))
939 return 0;
940 if (queue_is_mq(q)) {
941 ret = blk_mq_poll(q, cookie, iob, flags);
942 } else {
943 struct gendisk *disk = q->disk;
944
945 if (disk && disk->fops->poll_bio)
946 ret = disk->fops->poll_bio(bio, iob, flags);
947 }
948 blk_queue_exit(q);
949 return ret;
950 }
951 EXPORT_SYMBOL_GPL(bio_poll);
952
953 /*
954 * Helper to implement file_operations.iopoll. Requires the bio to be stored
955 * in iocb->private, and cleared before freeing the bio.
956 */
957 int iocb_bio_iopoll(struct kiocb *kiocb, struct io_comp_batch *iob,
958 unsigned int flags)
959 {
960 struct bio *bio;
961 int ret = 0;
962
963 /*
964 * Note: the bio cache only uses SLAB_TYPESAFE_BY_RCU, so bio can
965 * point to a freshly allocated bio at this point. If that happens
966 * we have a few cases to consider:
967 *
968 * 1) the bio is beeing initialized and bi_bdev is NULL. We can just
969 * simply nothing in this case
970 * 2) the bio points to a not poll enabled device. bio_poll will catch
971 * this and return 0
972 * 3) the bio points to a poll capable device, including but not
973 * limited to the one that the original bio pointed to. In this
974 * case we will call into the actual poll method and poll for I/O,
975 * even if we don't need to, but it won't cause harm either.
976 *
977 * For cases 2) and 3) above the RCU grace period ensures that bi_bdev
978 * is still allocated. Because partitions hold a reference to the whole
979 * device bdev and thus disk, the disk is also still valid. Grabbing
980 * a reference to the queue in bio_poll() ensures the hctxs and requests
981 * are still valid as well.
982 */
983 rcu_read_lock();
984 bio = READ_ONCE(kiocb->private);
985 if (bio)
986 ret = bio_poll(bio, iob, flags);
987 rcu_read_unlock();
988
989 return ret;
990 }
991 EXPORT_SYMBOL_GPL(iocb_bio_iopoll);
992
993 void update_io_ticks(struct block_device *part, unsigned long now, bool end)
994 {
995 unsigned long stamp;
996 again:
997 stamp = READ_ONCE(part->bd_stamp);
998 if (unlikely(time_after(now, stamp)) &&
999 likely(try_cmpxchg(&part->bd_stamp, &stamp, now)) &&
1000 (end || part_in_flight(part)))
1001 __part_stat_add(part, io_ticks, now - stamp);
1002
1003 if (bdev_is_partition(part)) {
1004 part = bdev_whole(part);
1005 goto again;
1006 }
1007 }
1008
1009 unsigned long bdev_start_io_acct(struct block_device *bdev, enum req_op op,
1010 unsigned long start_time)
1011 {
1012 part_stat_lock();
1013 update_io_ticks(bdev, start_time, false);
1014 part_stat_local_inc(bdev, in_flight[op_is_write(op)]);
1015 part_stat_unlock();
1016
1017 return start_time;
1018 }
1019 EXPORT_SYMBOL(bdev_start_io_acct);
1020
1021 /**
1022 * bio_start_io_acct - start I/O accounting for bio based drivers
1023 * @bio: bio to start account for
1024 *
1025 * Returns the start time that should be passed back to bio_end_io_acct().
1026 */
1027 unsigned long bio_start_io_acct(struct bio *bio)
1028 {
1029 return bdev_start_io_acct(bio->bi_bdev, bio_op(bio), jiffies);
1030 }
1031 EXPORT_SYMBOL_GPL(bio_start_io_acct);
1032
1033 void bdev_end_io_acct(struct block_device *bdev, enum req_op op,
1034 unsigned int sectors, unsigned long start_time)
1035 {
1036 const int sgrp = op_stat_group(op);
1037 unsigned long now = READ_ONCE(jiffies);
1038 unsigned long duration = now - start_time;
1039
1040 part_stat_lock();
1041 update_io_ticks(bdev, now, true);
1042 part_stat_inc(bdev, ios[sgrp]);
1043 part_stat_add(bdev, sectors[sgrp], sectors);
1044 part_stat_add(bdev, nsecs[sgrp], jiffies_to_nsecs(duration));
1045 part_stat_local_dec(bdev, in_flight[op_is_write(op)]);
1046 part_stat_unlock();
1047 }
1048 EXPORT_SYMBOL(bdev_end_io_acct);
1049
1050 void bio_end_io_acct_remapped(struct bio *bio, unsigned long start_time,
1051 struct block_device *orig_bdev)
1052 {
1053 bdev_end_io_acct(orig_bdev, bio_op(bio), bio_sectors(bio), start_time);
1054 }
1055 EXPORT_SYMBOL_GPL(bio_end_io_acct_remapped);
1056
1057 /**
1058 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
1059 * @q : the queue of the device being checked
1060 *
1061 * Description:
1062 * Check if underlying low-level drivers of a device are busy.
1063 * If the drivers want to export their busy state, they must set own
1064 * exporting function using blk_queue_lld_busy() first.
1065 *
1066 * Basically, this function is used only by request stacking drivers
1067 * to stop dispatching requests to underlying devices when underlying
1068 * devices are busy. This behavior helps more I/O merging on the queue
1069 * of the request stacking driver and prevents I/O throughput regression
1070 * on burst I/O load.
1071 *
1072 * Return:
1073 * 0 - Not busy (The request stacking driver should dispatch request)
1074 * 1 - Busy (The request stacking driver should stop dispatching request)
1075 */
1076 int blk_lld_busy(struct request_queue *q)
1077 {
1078 if (queue_is_mq(q) && q->mq_ops->busy)
1079 return q->mq_ops->busy(q);
1080
1081 return 0;
1082 }
1083 EXPORT_SYMBOL_GPL(blk_lld_busy);
1084
1085 int kblockd_schedule_work(struct work_struct *work)
1086 {
1087 return queue_work(kblockd_workqueue, work);
1088 }
1089 EXPORT_SYMBOL(kblockd_schedule_work);
1090
1091 int kblockd_mod_delayed_work_on(int cpu, struct delayed_work *dwork,
1092 unsigned long delay)
1093 {
1094 return mod_delayed_work_on(cpu, kblockd_workqueue, dwork, delay);
1095 }
1096 EXPORT_SYMBOL(kblockd_mod_delayed_work_on);
1097
1098 void blk_start_plug_nr_ios(struct blk_plug *plug, unsigned short nr_ios)
1099 {
1100 struct task_struct *tsk = current;
1101
1102 /*
1103 * If this is a nested plug, don't actually assign it.
1104 */
1105 if (tsk->plug)
1106 return;
1107
1108 plug->cur_ktime = 0;
1109 plug->mq_list = NULL;
1110 plug->cached_rq = NULL;
1111 plug->nr_ios = min_t(unsigned short, nr_ios, BLK_MAX_REQUEST_COUNT);
1112 plug->rq_count = 0;
1113 plug->multiple_queues = false;
1114 plug->has_elevator = false;
1115 INIT_LIST_HEAD(&plug->cb_list);
1116
1117 /*
1118 * Store ordering should not be needed here, since a potential
1119 * preempt will imply a full memory barrier
1120 */
1121 tsk->plug = plug;
1122 }
1123
1124 /**
1125 * blk_start_plug - initialize blk_plug and track it inside the task_struct
1126 * @plug: The &struct blk_plug that needs to be initialized
1127 *
1128 * Description:
1129 * blk_start_plug() indicates to the block layer an intent by the caller
1130 * to submit multiple I/O requests in a batch. The block layer may use
1131 * this hint to defer submitting I/Os from the caller until blk_finish_plug()
1132 * is called. However, the block layer may choose to submit requests
1133 * before a call to blk_finish_plug() if the number of queued I/Os
1134 * exceeds %BLK_MAX_REQUEST_COUNT, or if the size of the I/O is larger than
1135 * %BLK_PLUG_FLUSH_SIZE. The queued I/Os may also be submitted early if
1136 * the task schedules (see below).
1137 *
1138 * Tracking blk_plug inside the task_struct will help with auto-flushing the
1139 * pending I/O should the task end up blocking between blk_start_plug() and
1140 * blk_finish_plug(). This is important from a performance perspective, but
1141 * also ensures that we don't deadlock. For instance, if the task is blocking
1142 * for a memory allocation, memory reclaim could end up wanting to free a
1143 * page belonging to that request that is currently residing in our private
1144 * plug. By flushing the pending I/O when the process goes to sleep, we avoid
1145 * this kind of deadlock.
1146 */
1147 void blk_start_plug(struct blk_plug *plug)
1148 {
1149 blk_start_plug_nr_ios(plug, 1);
1150 }
1151 EXPORT_SYMBOL(blk_start_plug);
1152
1153 static void flush_plug_callbacks(struct blk_plug *plug, bool from_schedule)
1154 {
1155 LIST_HEAD(callbacks);
1156
1157 while (!list_empty(&plug->cb_list)) {
1158 list_splice_init(&plug->cb_list, &callbacks);
1159
1160 while (!list_empty(&callbacks)) {
1161 struct blk_plug_cb *cb = list_first_entry(&callbacks,
1162 struct blk_plug_cb,
1163 list);
1164 list_del(&cb->list);
1165 cb->callback(cb, from_schedule);
1166 }
1167 }
1168 }
1169
1170 struct blk_plug_cb *blk_check_plugged(blk_plug_cb_fn unplug, void *data,
1171 int size)
1172 {
1173 struct blk_plug *plug = current->plug;
1174 struct blk_plug_cb *cb;
1175
1176 if (!plug)
1177 return NULL;
1178
1179 list_for_each_entry(cb, &plug->cb_list, list)
1180 if (cb->callback == unplug && cb->data == data)
1181 return cb;
1182
1183 /* Not currently on the callback list */
1184 BUG_ON(size < sizeof(*cb));
1185 cb = kzalloc(size, GFP_ATOMIC);
1186 if (cb) {
1187 cb->data = data;
1188 cb->callback = unplug;
1189 list_add(&cb->list, &plug->cb_list);
1190 }
1191 return cb;
1192 }
1193 EXPORT_SYMBOL(blk_check_plugged);
1194
1195 void __blk_flush_plug(struct blk_plug *plug, bool from_schedule)
1196 {
1197 if (!list_empty(&plug->cb_list))
1198 flush_plug_callbacks(plug, from_schedule);
1199 blk_mq_flush_plug_list(plug, from_schedule);
1200 /*
1201 * Unconditionally flush out cached requests, even if the unplug
1202 * event came from schedule. Since we know hold references to the
1203 * queue for cached requests, we don't want a blocked task holding
1204 * up a queue freeze/quiesce event.
1205 */
1206 if (unlikely(!rq_list_empty(plug->cached_rq)))
1207 blk_mq_free_plug_rqs(plug);
1208
1209 plug->cur_ktime = 0;
1210 current->flags &= ~PF_BLOCK_TS;
1211 }
1212
1213 /**
1214 * blk_finish_plug - mark the end of a batch of submitted I/O
1215 * @plug: The &struct blk_plug passed to blk_start_plug()
1216 *
1217 * Description:
1218 * Indicate that a batch of I/O submissions is complete. This function
1219 * must be paired with an initial call to blk_start_plug(). The intent
1220 * is to allow the block layer to optimize I/O submission. See the
1221 * documentation for blk_start_plug() for more information.
1222 */
1223 void blk_finish_plug(struct blk_plug *plug)
1224 {
1225 if (plug == current->plug) {
1226 __blk_flush_plug(plug, false);
1227 current->plug = NULL;
1228 }
1229 }
1230 EXPORT_SYMBOL(blk_finish_plug);
1231
1232 void blk_io_schedule(void)
1233 {
1234 /* Prevent hang_check timer from firing at us during very long I/O */
1235 unsigned long timeout = sysctl_hung_task_timeout_secs * HZ / 2;
1236
1237 if (timeout)
1238 io_schedule_timeout(timeout);
1239 else
1240 io_schedule();
1241 }
1242 EXPORT_SYMBOL_GPL(blk_io_schedule);
1243
1244 int __init blk_dev_init(void)
1245 {
1246 BUILD_BUG_ON((__force u32)REQ_OP_LAST >= (1 << REQ_OP_BITS));
1247 BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 *
1248 sizeof_field(struct request, cmd_flags));
1249 BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 *
1250 sizeof_field(struct bio, bi_opf));
1251
1252 /* used for unplugging and affects IO latency/throughput - HIGHPRI */
1253 kblockd_workqueue = alloc_workqueue("kblockd",
1254 WQ_MEM_RECLAIM | WQ_HIGHPRI, 0);
1255 if (!kblockd_workqueue)
1256 panic("Failed to create kblockd\n");
1257
1258 blk_requestq_cachep = KMEM_CACHE(request_queue, SLAB_PANIC);
1259
1260 blk_debugfs_root = debugfs_create_dir("block", NULL);
1261
1262 return 0;
1263 }
1264
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.