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