Line data Source code
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/psi.h>
41 : #include <linux/part_stat.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-sched.h"
50 : #include "blk-pm.h"
51 : #include "blk-cgroup.h"
52 : #include "blk-throttle.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 : DEFINE_IDA(blk_queue_ida);
64 :
65 : /*
66 : * For queue allocation
67 : */
68 : struct kmem_cache *blk_requestq_cachep;
69 : struct kmem_cache *blk_requestq_srcu_cachep;
70 :
71 : /*
72 : * Controlling structure to kblockd
73 : */
74 : static struct workqueue_struct *kblockd_workqueue;
75 :
76 : /**
77 : * blk_queue_flag_set - atomically set a queue flag
78 : * @flag: flag to be set
79 : * @q: request queue
80 : */
81 0 : void blk_queue_flag_set(unsigned int flag, struct request_queue *q)
82 : {
83 0 : set_bit(flag, &q->queue_flags);
84 0 : }
85 : EXPORT_SYMBOL(blk_queue_flag_set);
86 :
87 : /**
88 : * blk_queue_flag_clear - atomically clear a queue flag
89 : * @flag: flag to be cleared
90 : * @q: request queue
91 : */
92 0 : void blk_queue_flag_clear(unsigned int flag, struct request_queue *q)
93 : {
94 0 : clear_bit(flag, &q->queue_flags);
95 0 : }
96 : EXPORT_SYMBOL(blk_queue_flag_clear);
97 :
98 : /**
99 : * blk_queue_flag_test_and_set - atomically test and set a queue flag
100 : * @flag: flag to be set
101 : * @q: request queue
102 : *
103 : * Returns the previous value of @flag - 0 if the flag was not set and 1 if
104 : * the flag was already set.
105 : */
106 0 : bool blk_queue_flag_test_and_set(unsigned int flag, struct request_queue *q)
107 : {
108 0 : return test_and_set_bit(flag, &q->queue_flags);
109 : }
110 : EXPORT_SYMBOL_GPL(blk_queue_flag_test_and_set);
111 :
112 : #define REQ_OP_NAME(name) [REQ_OP_##name] = #name
113 : static const char *const blk_op_name[] = {
114 : REQ_OP_NAME(READ),
115 : REQ_OP_NAME(WRITE),
116 : REQ_OP_NAME(FLUSH),
117 : REQ_OP_NAME(DISCARD),
118 : REQ_OP_NAME(SECURE_ERASE),
119 : REQ_OP_NAME(ZONE_RESET),
120 : REQ_OP_NAME(ZONE_RESET_ALL),
121 : REQ_OP_NAME(ZONE_OPEN),
122 : REQ_OP_NAME(ZONE_CLOSE),
123 : REQ_OP_NAME(ZONE_FINISH),
124 : REQ_OP_NAME(ZONE_APPEND),
125 : REQ_OP_NAME(WRITE_ZEROES),
126 : REQ_OP_NAME(DRV_IN),
127 : REQ_OP_NAME(DRV_OUT),
128 : };
129 : #undef REQ_OP_NAME
130 :
131 : /**
132 : * blk_op_str - Return string XXX in the REQ_OP_XXX.
133 : * @op: REQ_OP_XXX.
134 : *
135 : * Description: Centralize block layer function to convert REQ_OP_XXX into
136 : * string format. Useful in the debugging and tracing bio or request. For
137 : * invalid REQ_OP_XXX it returns string "UNKNOWN".
138 : */
139 0 : inline const char *blk_op_str(unsigned int op)
140 : {
141 0 : const char *op_str = "UNKNOWN";
142 :
143 0 : if (op < ARRAY_SIZE(blk_op_name) && blk_op_name[op])
144 0 : op_str = blk_op_name[op];
145 :
146 0 : return op_str;
147 : }
148 : EXPORT_SYMBOL_GPL(blk_op_str);
149 :
150 : static const struct {
151 : int errno;
152 : const char *name;
153 : } blk_errors[] = {
154 : [BLK_STS_OK] = { 0, "" },
155 : [BLK_STS_NOTSUPP] = { -EOPNOTSUPP, "operation not supported" },
156 : [BLK_STS_TIMEOUT] = { -ETIMEDOUT, "timeout" },
157 : [BLK_STS_NOSPC] = { -ENOSPC, "critical space allocation" },
158 : [BLK_STS_TRANSPORT] = { -ENOLINK, "recoverable transport" },
159 : [BLK_STS_TARGET] = { -EREMOTEIO, "critical target" },
160 : [BLK_STS_NEXUS] = { -EBADE, "critical nexus" },
161 : [BLK_STS_MEDIUM] = { -ENODATA, "critical medium" },
162 : [BLK_STS_PROTECTION] = { -EILSEQ, "protection" },
163 : [BLK_STS_RESOURCE] = { -ENOMEM, "kernel resource" },
164 : [BLK_STS_DEV_RESOURCE] = { -EBUSY, "device resource" },
165 : [BLK_STS_AGAIN] = { -EAGAIN, "nonblocking retry" },
166 : [BLK_STS_OFFLINE] = { -ENODEV, "device offline" },
167 :
168 : /* device mapper special case, should not leak out: */
169 : [BLK_STS_DM_REQUEUE] = { -EREMCHG, "dm internal retry" },
170 :
171 : /* zone device specific errors */
172 : [BLK_STS_ZONE_OPEN_RESOURCE] = { -ETOOMANYREFS, "open zones exceeded" },
173 : [BLK_STS_ZONE_ACTIVE_RESOURCE] = { -EOVERFLOW, "active zones exceeded" },
174 :
175 : /* everything else not covered above: */
176 : [BLK_STS_IOERR] = { -EIO, "I/O" },
177 : };
178 :
179 0 : blk_status_t errno_to_blk_status(int errno)
180 : {
181 : int i;
182 :
183 0 : for (i = 0; i < ARRAY_SIZE(blk_errors); i++) {
184 0 : if (blk_errors[i].errno == errno)
185 0 : return (__force blk_status_t)i;
186 : }
187 :
188 : return BLK_STS_IOERR;
189 : }
190 : EXPORT_SYMBOL_GPL(errno_to_blk_status);
191 :
192 0 : int blk_status_to_errno(blk_status_t status)
193 : {
194 0 : int idx = (__force int)status;
195 :
196 0 : if (WARN_ON_ONCE(idx >= ARRAY_SIZE(blk_errors)))
197 : return -EIO;
198 0 : return blk_errors[idx].errno;
199 : }
200 : EXPORT_SYMBOL_GPL(blk_status_to_errno);
201 :
202 0 : const char *blk_status_to_str(blk_status_t status)
203 : {
204 0 : int idx = (__force int)status;
205 :
206 0 : if (WARN_ON_ONCE(idx >= ARRAY_SIZE(blk_errors)))
207 : return "<null>";
208 0 : return blk_errors[idx].name;
209 : }
210 :
211 : /**
212 : * blk_sync_queue - cancel any pending callbacks on a queue
213 : * @q: the queue
214 : *
215 : * Description:
216 : * The block layer may perform asynchronous callback activity
217 : * on a queue, such as calling the unplug function after a timeout.
218 : * A block device may call blk_sync_queue to ensure that any
219 : * such activity is cancelled, thus allowing it to release resources
220 : * that the callbacks might use. The caller must already have made sure
221 : * that its ->submit_bio will not re-add plugging prior to calling
222 : * this function.
223 : *
224 : * This function does not cancel any asynchronous activity arising
225 : * out of elevator or throttling code. That would require elevator_exit()
226 : * and blkcg_exit_queue() to be called with queue lock initialized.
227 : *
228 : */
229 0 : void blk_sync_queue(struct request_queue *q)
230 : {
231 0 : del_timer_sync(&q->timeout);
232 0 : cancel_work_sync(&q->timeout_work);
233 0 : }
234 : EXPORT_SYMBOL(blk_sync_queue);
235 :
236 : /**
237 : * blk_set_pm_only - increment pm_only counter
238 : * @q: request queue pointer
239 : */
240 0 : void blk_set_pm_only(struct request_queue *q)
241 : {
242 0 : atomic_inc(&q->pm_only);
243 0 : }
244 : EXPORT_SYMBOL_GPL(blk_set_pm_only);
245 :
246 0 : void blk_clear_pm_only(struct request_queue *q)
247 : {
248 : int pm_only;
249 :
250 0 : pm_only = atomic_dec_return(&q->pm_only);
251 0 : WARN_ON_ONCE(pm_only < 0);
252 0 : if (pm_only == 0)
253 0 : wake_up_all(&q->mq_freeze_wq);
254 0 : }
255 : EXPORT_SYMBOL_GPL(blk_clear_pm_only);
256 :
257 : /**
258 : * blk_put_queue - decrement the request_queue refcount
259 : * @q: the request_queue structure to decrement the refcount for
260 : *
261 : * Decrements the refcount of the request_queue kobject. When this reaches 0
262 : * we'll have blk_release_queue() called.
263 : *
264 : * Context: Any context, but the last reference must not be dropped from
265 : * atomic context.
266 : */
267 0 : void blk_put_queue(struct request_queue *q)
268 : {
269 0 : kobject_put(&q->kobj);
270 0 : }
271 : EXPORT_SYMBOL(blk_put_queue);
272 :
273 0 : void blk_queue_start_drain(struct request_queue *q)
274 : {
275 : /*
276 : * When queue DYING flag is set, we need to block new req
277 : * entering queue, so we call blk_freeze_queue_start() to
278 : * prevent I/O from crossing blk_queue_enter().
279 : */
280 0 : blk_freeze_queue_start(q);
281 0 : if (queue_is_mq(q))
282 0 : blk_mq_wake_waiters(q);
283 : /* Make blk_queue_enter() reexamine the DYING flag. */
284 0 : wake_up_all(&q->mq_freeze_wq);
285 0 : }
286 :
287 : /**
288 : * blk_cleanup_queue - shutdown a request queue
289 : * @q: request queue to shutdown
290 : *
291 : * Mark @q DYING, drain all pending requests, mark @q DEAD, destroy and
292 : * put it. All future requests will be failed immediately with -ENODEV.
293 : *
294 : * Context: can sleep
295 : */
296 0 : void blk_cleanup_queue(struct request_queue *q)
297 : {
298 : /* cannot be called from atomic context */
299 : might_sleep();
300 :
301 0 : WARN_ON_ONCE(blk_queue_registered(q));
302 :
303 : /* mark @q DYING, no new request or merges will be allowed afterwards */
304 0 : blk_queue_flag_set(QUEUE_FLAG_DYING, q);
305 0 : blk_queue_start_drain(q);
306 :
307 0 : blk_queue_flag_set(QUEUE_FLAG_NOMERGES, q);
308 0 : blk_queue_flag_set(QUEUE_FLAG_NOXMERGES, q);
309 :
310 : /*
311 : * Drain all requests queued before DYING marking. Set DEAD flag to
312 : * prevent that blk_mq_run_hw_queues() accesses the hardware queues
313 : * after draining finished.
314 : */
315 0 : blk_freeze_queue(q);
316 :
317 0 : blk_queue_flag_set(QUEUE_FLAG_DEAD, q);
318 :
319 0 : blk_sync_queue(q);
320 0 : if (queue_is_mq(q)) {
321 0 : blk_mq_cancel_work_sync(q);
322 0 : blk_mq_exit_queue(q);
323 : }
324 :
325 : /*
326 : * In theory, request pool of sched_tags belongs to request queue.
327 : * However, the current implementation requires tag_set for freeing
328 : * requests, so free the pool now.
329 : *
330 : * Queue has become frozen, there can't be any in-queue requests, so
331 : * it is safe to free requests now.
332 : */
333 0 : mutex_lock(&q->sysfs_lock);
334 0 : if (q->elevator)
335 0 : blk_mq_sched_free_rqs(q);
336 0 : mutex_unlock(&q->sysfs_lock);
337 :
338 : /* @q is and will stay empty, shutdown and put */
339 0 : blk_put_queue(q);
340 0 : }
341 : EXPORT_SYMBOL(blk_cleanup_queue);
342 :
343 : /**
344 : * blk_queue_enter() - try to increase q->q_usage_counter
345 : * @q: request queue pointer
346 : * @flags: BLK_MQ_REQ_NOWAIT and/or BLK_MQ_REQ_PM
347 : */
348 0 : int blk_queue_enter(struct request_queue *q, blk_mq_req_flags_t flags)
349 : {
350 0 : const bool pm = flags & BLK_MQ_REQ_PM;
351 :
352 0 : while (!blk_try_enter_queue(q, pm)) {
353 0 : if (flags & BLK_MQ_REQ_NOWAIT)
354 : return -EBUSY;
355 :
356 : /*
357 : * read pair of barrier in blk_freeze_queue_start(), we need to
358 : * order reading __PERCPU_REF_DEAD flag of .q_usage_counter and
359 : * reading .mq_freeze_depth or queue dying flag, otherwise the
360 : * following wait may never return if the two reads are
361 : * reordered.
362 : */
363 0 : smp_rmb();
364 0 : wait_event(q->mq_freeze_wq,
365 : (!q->mq_freeze_depth &&
366 : blk_pm_resume_queue(pm, q)) ||
367 : blk_queue_dying(q));
368 0 : if (blk_queue_dying(q))
369 : return -ENODEV;
370 : }
371 :
372 : return 0;
373 : }
374 :
375 0 : int __bio_queue_enter(struct request_queue *q, struct bio *bio)
376 : {
377 0 : while (!blk_try_enter_queue(q, false)) {
378 0 : struct gendisk *disk = bio->bi_bdev->bd_disk;
379 :
380 0 : if (bio->bi_opf & REQ_NOWAIT) {
381 0 : if (test_bit(GD_DEAD, &disk->state))
382 : goto dead;
383 0 : bio_wouldblock_error(bio);
384 0 : return -EBUSY;
385 : }
386 :
387 : /*
388 : * read pair of barrier in blk_freeze_queue_start(), we need to
389 : * order reading __PERCPU_REF_DEAD flag of .q_usage_counter and
390 : * reading .mq_freeze_depth or queue dying flag, otherwise the
391 : * following wait may never return if the two reads are
392 : * reordered.
393 : */
394 0 : smp_rmb();
395 0 : wait_event(q->mq_freeze_wq,
396 : (!q->mq_freeze_depth &&
397 : blk_pm_resume_queue(false, q)) ||
398 : test_bit(GD_DEAD, &disk->state));
399 0 : if (test_bit(GD_DEAD, &disk->state))
400 : goto dead;
401 : }
402 :
403 : return 0;
404 : dead:
405 0 : bio_io_error(bio);
406 0 : return -ENODEV;
407 : }
408 :
409 0 : void blk_queue_exit(struct request_queue *q)
410 : {
411 0 : percpu_ref_put(&q->q_usage_counter);
412 0 : }
413 :
414 0 : static void blk_queue_usage_counter_release(struct percpu_ref *ref)
415 : {
416 0 : struct request_queue *q =
417 0 : container_of(ref, struct request_queue, q_usage_counter);
418 :
419 0 : wake_up_all(&q->mq_freeze_wq);
420 0 : }
421 :
422 0 : static void blk_rq_timed_out_timer(struct timer_list *t)
423 : {
424 0 : struct request_queue *q = from_timer(q, t, timeout);
425 :
426 0 : kblockd_schedule_work(&q->timeout_work);
427 0 : }
428 :
429 0 : static void blk_timeout_work(struct work_struct *work)
430 : {
431 0 : }
432 :
433 0 : struct request_queue *blk_alloc_queue(int node_id, bool alloc_srcu)
434 : {
435 : struct request_queue *q;
436 : int ret;
437 :
438 0 : q = kmem_cache_alloc_node(blk_get_queue_kmem_cache(alloc_srcu),
439 : GFP_KERNEL | __GFP_ZERO, node_id);
440 0 : if (!q)
441 : return NULL;
442 :
443 0 : if (alloc_srcu) {
444 0 : blk_queue_flag_set(QUEUE_FLAG_HAS_SRCU, q);
445 0 : if (init_srcu_struct(q->srcu) != 0)
446 : goto fail_q;
447 : }
448 :
449 0 : q->last_merge = NULL;
450 :
451 0 : q->id = ida_simple_get(&blk_queue_ida, 0, 0, GFP_KERNEL);
452 0 : if (q->id < 0)
453 : goto fail_srcu;
454 :
455 0 : ret = bioset_init(&q->bio_split, BIO_POOL_SIZE, 0, 0);
456 0 : if (ret)
457 : goto fail_id;
458 :
459 0 : q->stats = blk_alloc_queue_stats();
460 0 : if (!q->stats)
461 : goto fail_split;
462 :
463 0 : q->node = node_id;
464 :
465 0 : atomic_set(&q->nr_active_requests_shared_tags, 0);
466 :
467 0 : timer_setup(&q->timeout, blk_rq_timed_out_timer, 0);
468 0 : INIT_WORK(&q->timeout_work, blk_timeout_work);
469 0 : INIT_LIST_HEAD(&q->icq_list);
470 :
471 0 : kobject_init(&q->kobj, &blk_queue_ktype);
472 :
473 0 : mutex_init(&q->debugfs_mutex);
474 0 : mutex_init(&q->sysfs_lock);
475 0 : mutex_init(&q->sysfs_dir_lock);
476 0 : spin_lock_init(&q->queue_lock);
477 :
478 0 : init_waitqueue_head(&q->mq_freeze_wq);
479 0 : mutex_init(&q->mq_freeze_lock);
480 :
481 : /*
482 : * Init percpu_ref in atomic mode so that it's faster to shutdown.
483 : * See blk_register_queue() for details.
484 : */
485 0 : if (percpu_ref_init(&q->q_usage_counter,
486 : blk_queue_usage_counter_release,
487 : PERCPU_REF_INIT_ATOMIC, GFP_KERNEL))
488 : goto fail_stats;
489 :
490 0 : blk_queue_dma_alignment(q, 511);
491 0 : blk_set_default_limits(&q->limits);
492 0 : q->nr_requests = BLKDEV_DEFAULT_RQ;
493 :
494 0 : return q;
495 :
496 : fail_stats:
497 0 : blk_free_queue_stats(q->stats);
498 : fail_split:
499 0 : bioset_exit(&q->bio_split);
500 : fail_id:
501 0 : ida_simple_remove(&blk_queue_ida, q->id);
502 : fail_srcu:
503 0 : if (alloc_srcu)
504 0 : cleanup_srcu_struct(q->srcu);
505 : fail_q:
506 0 : kmem_cache_free(blk_get_queue_kmem_cache(alloc_srcu), q);
507 0 : return NULL;
508 : }
509 :
510 : /**
511 : * blk_get_queue - increment the request_queue refcount
512 : * @q: the request_queue structure to increment the refcount for
513 : *
514 : * Increment the refcount of the request_queue kobject.
515 : *
516 : * Context: Any context.
517 : */
518 0 : bool blk_get_queue(struct request_queue *q)
519 : {
520 0 : if (likely(!blk_queue_dying(q))) {
521 0 : __blk_get_queue(q);
522 0 : return true;
523 : }
524 :
525 : return false;
526 : }
527 : EXPORT_SYMBOL(blk_get_queue);
528 :
529 : #ifdef CONFIG_FAIL_MAKE_REQUEST
530 :
531 : static DECLARE_FAULT_ATTR(fail_make_request);
532 :
533 : static int __init setup_fail_make_request(char *str)
534 : {
535 : return setup_fault_attr(&fail_make_request, str);
536 : }
537 : __setup("fail_make_request=", setup_fail_make_request);
538 :
539 : bool should_fail_request(struct block_device *part, unsigned int bytes)
540 : {
541 : return part->bd_make_it_fail && should_fail(&fail_make_request, bytes);
542 : }
543 :
544 : static int __init fail_make_request_debugfs(void)
545 : {
546 : struct dentry *dir = fault_create_debugfs_attr("fail_make_request",
547 : NULL, &fail_make_request);
548 :
549 : return PTR_ERR_OR_ZERO(dir);
550 : }
551 :
552 : late_initcall(fail_make_request_debugfs);
553 : #endif /* CONFIG_FAIL_MAKE_REQUEST */
554 :
555 0 : static inline bool bio_check_ro(struct bio *bio)
556 : {
557 0 : if (op_is_write(bio_op(bio)) && bdev_read_only(bio->bi_bdev)) {
558 0 : if (op_is_flush(bio->bi_opf) && !bio_sectors(bio))
559 : return false;
560 0 : pr_warn("Trying to write to read-only block-device %pg\n",
561 : bio->bi_bdev);
562 : /* Older lvm-tools actually trigger this */
563 0 : return false;
564 : }
565 :
566 : return false;
567 : }
568 :
569 0 : static noinline int should_fail_bio(struct bio *bio)
570 : {
571 0 : if (should_fail_request(bdev_whole(bio->bi_bdev), bio->bi_iter.bi_size))
572 : return -EIO;
573 : return 0;
574 : }
575 : ALLOW_ERROR_INJECTION(should_fail_bio, ERRNO);
576 :
577 : /*
578 : * Check whether this bio extends beyond the end of the device or partition.
579 : * This may well happen - the kernel calls bread() without checking the size of
580 : * the device, e.g., when mounting a file system.
581 : */
582 0 : static inline int bio_check_eod(struct bio *bio)
583 : {
584 0 : sector_t maxsector = bdev_nr_sectors(bio->bi_bdev);
585 0 : unsigned int nr_sectors = bio_sectors(bio);
586 :
587 0 : if (nr_sectors && maxsector &&
588 0 : (nr_sectors > maxsector ||
589 0 : bio->bi_iter.bi_sector > maxsector - nr_sectors)) {
590 0 : pr_info_ratelimited("%s: attempt to access beyond end of device\n"
591 : "%pg: rw=%d, want=%llu, limit=%llu\n",
592 : current->comm,
593 : bio->bi_bdev, bio->bi_opf,
594 : bio_end_sector(bio), maxsector);
595 : return -EIO;
596 : }
597 : return 0;
598 : }
599 :
600 : /*
601 : * Remap block n of partition p to block n+start(p) of the disk.
602 : */
603 : static int blk_partition_remap(struct bio *bio)
604 : {
605 0 : struct block_device *p = bio->bi_bdev;
606 :
607 0 : if (unlikely(should_fail_request(p, bio->bi_iter.bi_size)))
608 : return -EIO;
609 0 : if (bio_sectors(bio)) {
610 0 : bio->bi_iter.bi_sector += p->bd_start_sect;
611 0 : trace_block_bio_remap(bio, p->bd_dev,
612 : bio->bi_iter.bi_sector -
613 0 : p->bd_start_sect);
614 : }
615 0 : bio_set_flag(bio, BIO_REMAPPED);
616 : return 0;
617 : }
618 :
619 : /*
620 : * Check write append to a zoned block device.
621 : */
622 : static inline blk_status_t blk_check_zone_append(struct request_queue *q,
623 : struct bio *bio)
624 : {
625 : sector_t pos = bio->bi_iter.bi_sector;
626 : int nr_sectors = bio_sectors(bio);
627 :
628 : /* Only applicable to zoned block devices */
629 : if (!blk_queue_is_zoned(q))
630 : return BLK_STS_NOTSUPP;
631 :
632 : /* The bio sector must point to the start of a sequential zone */
633 : if (pos & (blk_queue_zone_sectors(q) - 1) ||
634 : !blk_queue_zone_is_seq(q, pos))
635 : return BLK_STS_IOERR;
636 :
637 : /*
638 : * Not allowed to cross zone boundaries. Otherwise, the BIO will be
639 : * split and could result in non-contiguous sectors being written in
640 : * different zones.
641 : */
642 : if (nr_sectors > q->limits.chunk_sectors)
643 : return BLK_STS_IOERR;
644 :
645 : /* Make sure the BIO is small enough and will not get split */
646 : if (nr_sectors > q->limits.max_zone_append_sectors)
647 : return BLK_STS_IOERR;
648 :
649 : bio->bi_opf |= REQ_NOMERGE;
650 :
651 : return BLK_STS_OK;
652 : }
653 :
654 0 : static void __submit_bio(struct bio *bio)
655 : {
656 0 : struct gendisk *disk = bio->bi_bdev->bd_disk;
657 :
658 0 : if (unlikely(!blk_crypto_bio_prep(&bio)))
659 : return;
660 :
661 0 : if (!disk->fops->submit_bio) {
662 0 : blk_mq_submit_bio(bio);
663 0 : } else if (likely(bio_queue_enter(bio) == 0)) {
664 0 : disk->fops->submit_bio(bio);
665 0 : blk_queue_exit(disk->queue);
666 : }
667 : }
668 :
669 : /*
670 : * The loop in this function may be a bit non-obvious, and so deserves some
671 : * explanation:
672 : *
673 : * - Before entering the loop, bio->bi_next is NULL (as all callers ensure
674 : * that), so we have a list with a single bio.
675 : * - We pretend that we have just taken it off a longer list, so we assign
676 : * bio_list to a pointer to the bio_list_on_stack, thus initialising the
677 : * bio_list of new bios to be added. ->submit_bio() may indeed add some more
678 : * bios through a recursive call to submit_bio_noacct. If it did, we find a
679 : * non-NULL value in bio_list and re-enter the loop from the top.
680 : * - In this case we really did just take the bio of the top of the list (no
681 : * pretending) and so remove it from bio_list, and call into ->submit_bio()
682 : * again.
683 : *
684 : * bio_list_on_stack[0] contains bios submitted by the current ->submit_bio.
685 : * bio_list_on_stack[1] contains bios that were submitted before the current
686 : * ->submit_bio, but that haven't been processed yet.
687 : */
688 0 : static void __submit_bio_noacct(struct bio *bio)
689 : {
690 : struct bio_list bio_list_on_stack[2];
691 :
692 0 : BUG_ON(bio->bi_next);
693 :
694 0 : bio_list_init(&bio_list_on_stack[0]);
695 0 : current->bio_list = bio_list_on_stack;
696 :
697 : do {
698 0 : struct request_queue *q = bdev_get_queue(bio->bi_bdev);
699 : struct bio_list lower, same;
700 :
701 : /*
702 : * Create a fresh bio_list for all subordinate requests.
703 : */
704 0 : bio_list_on_stack[1] = bio_list_on_stack[0];
705 0 : bio_list_init(&bio_list_on_stack[0]);
706 :
707 0 : __submit_bio(bio);
708 :
709 : /*
710 : * Sort new bios into those for a lower level and those for the
711 : * same level.
712 : */
713 0 : bio_list_init(&lower);
714 : bio_list_init(&same);
715 0 : while ((bio = bio_list_pop(&bio_list_on_stack[0])) != NULL)
716 0 : if (q == bdev_get_queue(bio->bi_bdev))
717 : bio_list_add(&same, bio);
718 : else
719 : bio_list_add(&lower, bio);
720 :
721 : /*
722 : * Now assemble so we handle the lowest level first.
723 : */
724 0 : bio_list_merge(&bio_list_on_stack[0], &lower);
725 0 : bio_list_merge(&bio_list_on_stack[0], &same);
726 0 : bio_list_merge(&bio_list_on_stack[0], &bio_list_on_stack[1]);
727 0 : } while ((bio = bio_list_pop(&bio_list_on_stack[0])));
728 :
729 0 : current->bio_list = NULL;
730 0 : }
731 :
732 0 : static void __submit_bio_noacct_mq(struct bio *bio)
733 : {
734 0 : struct bio_list bio_list[2] = { };
735 :
736 0 : current->bio_list = bio_list;
737 :
738 : do {
739 0 : __submit_bio(bio);
740 0 : } while ((bio = bio_list_pop(&bio_list[0])));
741 :
742 0 : current->bio_list = NULL;
743 0 : }
744 :
745 0 : void submit_bio_noacct_nocheck(struct bio *bio)
746 : {
747 : /*
748 : * We only want one ->submit_bio to be active at a time, else stack
749 : * usage with stacked devices could be a problem. Use current->bio_list
750 : * to collect a list of requests submited by a ->submit_bio method while
751 : * it is active, and then process them after it returned.
752 : */
753 0 : if (current->bio_list)
754 0 : bio_list_add(¤t->bio_list[0], bio);
755 0 : else if (!bio->bi_bdev->bd_disk->fops->submit_bio)
756 0 : __submit_bio_noacct_mq(bio);
757 : else
758 0 : __submit_bio_noacct(bio);
759 0 : }
760 :
761 : /**
762 : * submit_bio_noacct - re-submit a bio to the block device layer for I/O
763 : * @bio: The bio describing the location in memory and on the device.
764 : *
765 : * This is a version of submit_bio() that shall only be used for I/O that is
766 : * resubmitted to lower level drivers by stacking block drivers. All file
767 : * systems and other upper level users of the block layer should use
768 : * submit_bio() instead.
769 : */
770 0 : void submit_bio_noacct(struct bio *bio)
771 : {
772 0 : struct block_device *bdev = bio->bi_bdev;
773 0 : struct request_queue *q = bdev_get_queue(bdev);
774 0 : blk_status_t status = BLK_STS_IOERR;
775 : struct blk_plug *plug;
776 :
777 : might_sleep();
778 :
779 0 : plug = blk_mq_plug(q, bio);
780 0 : if (plug && plug->nowait)
781 0 : bio->bi_opf |= REQ_NOWAIT;
782 :
783 : /*
784 : * For a REQ_NOWAIT based request, return -EOPNOTSUPP
785 : * if queue does not support NOWAIT.
786 : */
787 0 : if ((bio->bi_opf & REQ_NOWAIT) && !blk_queue_nowait(q))
788 : goto not_supported;
789 :
790 0 : if (should_fail_bio(bio))
791 : goto end_io;
792 0 : if (unlikely(bio_check_ro(bio)))
793 : goto end_io;
794 0 : if (!bio_flagged(bio, BIO_REMAPPED)) {
795 0 : if (unlikely(bio_check_eod(bio)))
796 : goto end_io;
797 0 : if (bdev->bd_partno && unlikely(blk_partition_remap(bio)))
798 : goto end_io;
799 : }
800 :
801 : /*
802 : * Filter flush bio's early so that bio based drivers without flush
803 : * support don't have to worry about them.
804 : */
805 0 : if (op_is_flush(bio->bi_opf) &&
806 0 : !test_bit(QUEUE_FLAG_WC, &q->queue_flags)) {
807 0 : bio->bi_opf &= ~(REQ_PREFLUSH | REQ_FUA);
808 0 : if (!bio_sectors(bio)) {
809 : status = BLK_STS_OK;
810 : goto end_io;
811 : }
812 : }
813 :
814 0 : if (!test_bit(QUEUE_FLAG_POLL, &q->queue_flags))
815 : bio_clear_polled(bio);
816 :
817 0 : switch (bio_op(bio)) {
818 : case REQ_OP_DISCARD:
819 0 : if (!blk_queue_discard(q))
820 : goto not_supported;
821 : break;
822 : case REQ_OP_SECURE_ERASE:
823 0 : if (!blk_queue_secure_erase(q))
824 : goto not_supported;
825 : break;
826 : case REQ_OP_ZONE_APPEND:
827 : status = blk_check_zone_append(q, bio);
828 : if (status != BLK_STS_OK)
829 : goto end_io;
830 : break;
831 : case REQ_OP_ZONE_RESET:
832 : case REQ_OP_ZONE_OPEN:
833 : case REQ_OP_ZONE_CLOSE:
834 : case REQ_OP_ZONE_FINISH:
835 : if (!blk_queue_is_zoned(q))
836 : goto not_supported;
837 : break;
838 : case REQ_OP_ZONE_RESET_ALL:
839 : if (!blk_queue_is_zoned(q) || !blk_queue_zone_resetall(q))
840 : goto not_supported;
841 : break;
842 : case REQ_OP_WRITE_ZEROES:
843 0 : if (!q->limits.max_write_zeroes_sectors)
844 : goto not_supported;
845 : break;
846 : default:
847 : break;
848 : }
849 :
850 0 : if (blk_throtl_bio(bio))
851 : return;
852 :
853 0 : blk_cgroup_bio_start(bio);
854 0 : blkcg_bio_issue_init(bio);
855 :
856 0 : if (!bio_flagged(bio, BIO_TRACE_COMPLETION)) {
857 0 : trace_block_bio_queue(bio);
858 : /* Now that enqueuing has been traced, we need to trace
859 : * completion as well.
860 : */
861 : bio_set_flag(bio, BIO_TRACE_COMPLETION);
862 : }
863 0 : submit_bio_noacct_nocheck(bio);
864 0 : return;
865 :
866 : not_supported:
867 : status = BLK_STS_NOTSUPP;
868 : end_io:
869 0 : bio->bi_status = status;
870 0 : bio_endio(bio);
871 : }
872 : EXPORT_SYMBOL(submit_bio_noacct);
873 :
874 : /**
875 : * submit_bio - submit a bio to the block device layer for I/O
876 : * @bio: The &struct bio which describes the I/O
877 : *
878 : * submit_bio() is used to submit I/O requests to block devices. It is passed a
879 : * fully set up &struct bio that describes the I/O that needs to be done. The
880 : * bio will be send to the device described by the bi_bdev field.
881 : *
882 : * The success/failure status of the request, along with notification of
883 : * completion, is delivered asynchronously through the ->bi_end_io() callback
884 : * in @bio. The bio must NOT be touched by thecaller until ->bi_end_io() has
885 : * been called.
886 : */
887 0 : void submit_bio(struct bio *bio)
888 : {
889 0 : if (blkcg_punt_bio_submit(bio))
890 : return;
891 :
892 : /*
893 : * If it's a regular read/write or a barrier with data attached,
894 : * go through the normal accounting stuff before submission.
895 : */
896 0 : if (bio_has_data(bio)) {
897 0 : unsigned int count = bio_sectors(bio);
898 :
899 0 : if (op_is_write(bio_op(bio))) {
900 0 : count_vm_events(PGPGOUT, count);
901 : } else {
902 0 : task_io_account_read(bio->bi_iter.bi_size);
903 0 : count_vm_events(PGPGIN, count);
904 : }
905 : }
906 :
907 : /*
908 : * If we're reading data that is part of the userspace workingset, count
909 : * submission time as memory stall. When the device is congested, or
910 : * the submitting cgroup IO-throttled, submission can be a significant
911 : * part of overall IO time.
912 : */
913 0 : if (unlikely(bio_op(bio) == REQ_OP_READ &&
914 : bio_flagged(bio, BIO_WORKINGSET))) {
915 : unsigned long pflags;
916 :
917 0 : psi_memstall_enter(&pflags);
918 0 : submit_bio_noacct(bio);
919 0 : psi_memstall_leave(&pflags);
920 : return;
921 : }
922 :
923 0 : submit_bio_noacct(bio);
924 : }
925 : EXPORT_SYMBOL(submit_bio);
926 :
927 : /**
928 : * bio_poll - poll for BIO completions
929 : * @bio: bio to poll for
930 : * @iob: batches of IO
931 : * @flags: BLK_POLL_* flags that control the behavior
932 : *
933 : * Poll for completions on queue associated with the bio. Returns number of
934 : * completed entries found.
935 : *
936 : * Note: the caller must either be the context that submitted @bio, or
937 : * be in a RCU critical section to prevent freeing of @bio.
938 : */
939 0 : int bio_poll(struct bio *bio, struct io_comp_batch *iob, unsigned int flags)
940 : {
941 0 : struct request_queue *q = bdev_get_queue(bio->bi_bdev);
942 0 : blk_qc_t cookie = READ_ONCE(bio->bi_cookie);
943 0 : int ret = 0;
944 :
945 0 : if (cookie == BLK_QC_T_NONE ||
946 0 : !test_bit(QUEUE_FLAG_POLL, &q->queue_flags))
947 : return 0;
948 :
949 0 : blk_flush_plug(current->plug, false);
950 :
951 0 : if (blk_queue_enter(q, BLK_MQ_REQ_NOWAIT))
952 : return 0;
953 0 : if (queue_is_mq(q)) {
954 0 : ret = blk_mq_poll(q, cookie, iob, flags);
955 : } else {
956 0 : struct gendisk *disk = q->disk;
957 :
958 0 : if (disk && disk->fops->poll_bio)
959 0 : ret = disk->fops->poll_bio(bio, iob, flags);
960 : }
961 0 : blk_queue_exit(q);
962 0 : return ret;
963 : }
964 : EXPORT_SYMBOL_GPL(bio_poll);
965 :
966 : /*
967 : * Helper to implement file_operations.iopoll. Requires the bio to be stored
968 : * in iocb->private, and cleared before freeing the bio.
969 : */
970 0 : int iocb_bio_iopoll(struct kiocb *kiocb, struct io_comp_batch *iob,
971 : unsigned int flags)
972 : {
973 : struct bio *bio;
974 0 : int ret = 0;
975 :
976 : /*
977 : * Note: the bio cache only uses SLAB_TYPESAFE_BY_RCU, so bio can
978 : * point to a freshly allocated bio at this point. If that happens
979 : * we have a few cases to consider:
980 : *
981 : * 1) the bio is beeing initialized and bi_bdev is NULL. We can just
982 : * simply nothing in this case
983 : * 2) the bio points to a not poll enabled device. bio_poll will catch
984 : * this and return 0
985 : * 3) the bio points to a poll capable device, including but not
986 : * limited to the one that the original bio pointed to. In this
987 : * case we will call into the actual poll method and poll for I/O,
988 : * even if we don't need to, but it won't cause harm either.
989 : *
990 : * For cases 2) and 3) above the RCU grace period ensures that bi_bdev
991 : * is still allocated. Because partitions hold a reference to the whole
992 : * device bdev and thus disk, the disk is also still valid. Grabbing
993 : * a reference to the queue in bio_poll() ensures the hctxs and requests
994 : * are still valid as well.
995 : */
996 : rcu_read_lock();
997 0 : bio = READ_ONCE(kiocb->private);
998 0 : if (bio && bio->bi_bdev)
999 0 : ret = bio_poll(bio, iob, flags);
1000 : rcu_read_unlock();
1001 :
1002 0 : return ret;
1003 : }
1004 : EXPORT_SYMBOL_GPL(iocb_bio_iopoll);
1005 :
1006 0 : void update_io_ticks(struct block_device *part, unsigned long now, bool end)
1007 : {
1008 : unsigned long stamp;
1009 : again:
1010 0 : stamp = READ_ONCE(part->bd_stamp);
1011 0 : if (unlikely(time_after(now, stamp))) {
1012 0 : if (likely(cmpxchg(&part->bd_stamp, stamp, now) == stamp))
1013 0 : __part_stat_add(part, io_ticks, end ? now - stamp : 1);
1014 : }
1015 0 : if (part->bd_partno) {
1016 0 : part = bdev_whole(part);
1017 0 : goto again;
1018 : }
1019 0 : }
1020 :
1021 0 : static unsigned long __part_start_io_acct(struct block_device *part,
1022 : unsigned int sectors, unsigned int op,
1023 : unsigned long start_time)
1024 : {
1025 0 : const int sgrp = op_stat_group(op);
1026 :
1027 0 : part_stat_lock();
1028 0 : update_io_ticks(part, start_time, false);
1029 0 : part_stat_inc(part, ios[sgrp]);
1030 0 : part_stat_add(part, sectors[sgrp], sectors);
1031 0 : part_stat_local_inc(part, in_flight[op_is_write(op)]);
1032 0 : part_stat_unlock();
1033 :
1034 0 : return start_time;
1035 : }
1036 :
1037 : /**
1038 : * bio_start_io_acct_time - start I/O accounting for bio based drivers
1039 : * @bio: bio to start account for
1040 : * @start_time: start time that should be passed back to bio_end_io_acct().
1041 : */
1042 0 : void bio_start_io_acct_time(struct bio *bio, unsigned long start_time)
1043 : {
1044 0 : __part_start_io_acct(bio->bi_bdev, bio_sectors(bio),
1045 0 : bio_op(bio), start_time);
1046 0 : }
1047 : EXPORT_SYMBOL_GPL(bio_start_io_acct_time);
1048 :
1049 : /**
1050 : * bio_start_io_acct - start I/O accounting for bio based drivers
1051 : * @bio: bio to start account for
1052 : *
1053 : * Returns the start time that should be passed back to bio_end_io_acct().
1054 : */
1055 0 : unsigned long bio_start_io_acct(struct bio *bio)
1056 : {
1057 0 : return __part_start_io_acct(bio->bi_bdev, bio_sectors(bio),
1058 0 : bio_op(bio), jiffies);
1059 : }
1060 : EXPORT_SYMBOL_GPL(bio_start_io_acct);
1061 :
1062 0 : unsigned long disk_start_io_acct(struct gendisk *disk, unsigned int sectors,
1063 : unsigned int op)
1064 : {
1065 0 : return __part_start_io_acct(disk->part0, sectors, op, jiffies);
1066 : }
1067 : EXPORT_SYMBOL(disk_start_io_acct);
1068 :
1069 0 : static void __part_end_io_acct(struct block_device *part, unsigned int op,
1070 : unsigned long start_time)
1071 : {
1072 0 : const int sgrp = op_stat_group(op);
1073 0 : unsigned long now = READ_ONCE(jiffies);
1074 0 : unsigned long duration = now - start_time;
1075 :
1076 0 : part_stat_lock();
1077 0 : update_io_ticks(part, now, true);
1078 0 : part_stat_add(part, nsecs[sgrp], jiffies_to_nsecs(duration));
1079 0 : part_stat_local_dec(part, in_flight[op_is_write(op)]);
1080 0 : part_stat_unlock();
1081 0 : }
1082 :
1083 0 : void bio_end_io_acct_remapped(struct bio *bio, unsigned long start_time,
1084 : struct block_device *orig_bdev)
1085 : {
1086 0 : __part_end_io_acct(orig_bdev, bio_op(bio), start_time);
1087 0 : }
1088 : EXPORT_SYMBOL_GPL(bio_end_io_acct_remapped);
1089 :
1090 0 : void disk_end_io_acct(struct gendisk *disk, unsigned int op,
1091 : unsigned long start_time)
1092 : {
1093 0 : __part_end_io_acct(disk->part0, op, start_time);
1094 0 : }
1095 : EXPORT_SYMBOL(disk_end_io_acct);
1096 :
1097 : /**
1098 : * blk_lld_busy - Check if underlying low-level drivers of a device are busy
1099 : * @q : the queue of the device being checked
1100 : *
1101 : * Description:
1102 : * Check if underlying low-level drivers of a device are busy.
1103 : * If the drivers want to export their busy state, they must set own
1104 : * exporting function using blk_queue_lld_busy() first.
1105 : *
1106 : * Basically, this function is used only by request stacking drivers
1107 : * to stop dispatching requests to underlying devices when underlying
1108 : * devices are busy. This behavior helps more I/O merging on the queue
1109 : * of the request stacking driver and prevents I/O throughput regression
1110 : * on burst I/O load.
1111 : *
1112 : * Return:
1113 : * 0 - Not busy (The request stacking driver should dispatch request)
1114 : * 1 - Busy (The request stacking driver should stop dispatching request)
1115 : */
1116 0 : int blk_lld_busy(struct request_queue *q)
1117 : {
1118 0 : if (queue_is_mq(q) && q->mq_ops->busy)
1119 0 : return q->mq_ops->busy(q);
1120 :
1121 : return 0;
1122 : }
1123 : EXPORT_SYMBOL_GPL(blk_lld_busy);
1124 :
1125 0 : int kblockd_schedule_work(struct work_struct *work)
1126 : {
1127 0 : return queue_work(kblockd_workqueue, work);
1128 : }
1129 : EXPORT_SYMBOL(kblockd_schedule_work);
1130 :
1131 0 : int kblockd_mod_delayed_work_on(int cpu, struct delayed_work *dwork,
1132 : unsigned long delay)
1133 : {
1134 0 : return mod_delayed_work_on(cpu, kblockd_workqueue, dwork, delay);
1135 : }
1136 : EXPORT_SYMBOL(kblockd_mod_delayed_work_on);
1137 :
1138 0 : void blk_start_plug_nr_ios(struct blk_plug *plug, unsigned short nr_ios)
1139 : {
1140 0 : struct task_struct *tsk = current;
1141 :
1142 : /*
1143 : * If this is a nested plug, don't actually assign it.
1144 : */
1145 0 : if (tsk->plug)
1146 : return;
1147 :
1148 0 : plug->mq_list = NULL;
1149 0 : plug->cached_rq = NULL;
1150 0 : plug->nr_ios = min_t(unsigned short, nr_ios, BLK_MAX_REQUEST_COUNT);
1151 0 : plug->rq_count = 0;
1152 0 : plug->multiple_queues = false;
1153 0 : plug->has_elevator = false;
1154 0 : plug->nowait = false;
1155 0 : INIT_LIST_HEAD(&plug->cb_list);
1156 :
1157 : /*
1158 : * Store ordering should not be needed here, since a potential
1159 : * preempt will imply a full memory barrier
1160 : */
1161 0 : tsk->plug = plug;
1162 : }
1163 :
1164 : /**
1165 : * blk_start_plug - initialize blk_plug and track it inside the task_struct
1166 : * @plug: The &struct blk_plug that needs to be initialized
1167 : *
1168 : * Description:
1169 : * blk_start_plug() indicates to the block layer an intent by the caller
1170 : * to submit multiple I/O requests in a batch. The block layer may use
1171 : * this hint to defer submitting I/Os from the caller until blk_finish_plug()
1172 : * is called. However, the block layer may choose to submit requests
1173 : * before a call to blk_finish_plug() if the number of queued I/Os
1174 : * exceeds %BLK_MAX_REQUEST_COUNT, or if the size of the I/O is larger than
1175 : * %BLK_PLUG_FLUSH_SIZE. The queued I/Os may also be submitted early if
1176 : * the task schedules (see below).
1177 : *
1178 : * Tracking blk_plug inside the task_struct will help with auto-flushing the
1179 : * pending I/O should the task end up blocking between blk_start_plug() and
1180 : * blk_finish_plug(). This is important from a performance perspective, but
1181 : * also ensures that we don't deadlock. For instance, if the task is blocking
1182 : * for a memory allocation, memory reclaim could end up wanting to free a
1183 : * page belonging to that request that is currently residing in our private
1184 : * plug. By flushing the pending I/O when the process goes to sleep, we avoid
1185 : * this kind of deadlock.
1186 : */
1187 0 : void blk_start_plug(struct blk_plug *plug)
1188 : {
1189 0 : blk_start_plug_nr_ios(plug, 1);
1190 0 : }
1191 : EXPORT_SYMBOL(blk_start_plug);
1192 :
1193 0 : static void flush_plug_callbacks(struct blk_plug *plug, bool from_schedule)
1194 : {
1195 0 : LIST_HEAD(callbacks);
1196 :
1197 0 : while (!list_empty(&plug->cb_list)) {
1198 0 : list_splice_init(&plug->cb_list, &callbacks);
1199 :
1200 0 : while (!list_empty(&callbacks)) {
1201 0 : struct blk_plug_cb *cb = list_first_entry(&callbacks,
1202 : struct blk_plug_cb,
1203 : list);
1204 0 : list_del(&cb->list);
1205 0 : cb->callback(cb, from_schedule);
1206 : }
1207 : }
1208 0 : }
1209 :
1210 0 : struct blk_plug_cb *blk_check_plugged(blk_plug_cb_fn unplug, void *data,
1211 : int size)
1212 : {
1213 0 : struct blk_plug *plug = current->plug;
1214 : struct blk_plug_cb *cb;
1215 :
1216 0 : if (!plug)
1217 : return NULL;
1218 :
1219 0 : list_for_each_entry(cb, &plug->cb_list, list)
1220 0 : if (cb->callback == unplug && cb->data == data)
1221 : return cb;
1222 :
1223 : /* Not currently on the callback list */
1224 0 : BUG_ON(size < sizeof(*cb));
1225 0 : cb = kzalloc(size, GFP_ATOMIC);
1226 0 : if (cb) {
1227 0 : cb->data = data;
1228 0 : cb->callback = unplug;
1229 0 : list_add(&cb->list, &plug->cb_list);
1230 : }
1231 : return cb;
1232 : }
1233 : EXPORT_SYMBOL(blk_check_plugged);
1234 :
1235 0 : void __blk_flush_plug(struct blk_plug *plug, bool from_schedule)
1236 : {
1237 0 : if (!list_empty(&plug->cb_list))
1238 0 : flush_plug_callbacks(plug, from_schedule);
1239 0 : if (!rq_list_empty(plug->mq_list))
1240 0 : blk_mq_flush_plug_list(plug, from_schedule);
1241 : /*
1242 : * Unconditionally flush out cached requests, even if the unplug
1243 : * event came from schedule. Since we know hold references to the
1244 : * queue for cached requests, we don't want a blocked task holding
1245 : * up a queue freeze/quiesce event.
1246 : */
1247 0 : if (unlikely(!rq_list_empty(plug->cached_rq)))
1248 0 : blk_mq_free_plug_rqs(plug);
1249 0 : }
1250 :
1251 : /**
1252 : * blk_finish_plug - mark the end of a batch of submitted I/O
1253 : * @plug: The &struct blk_plug passed to blk_start_plug()
1254 : *
1255 : * Description:
1256 : * Indicate that a batch of I/O submissions is complete. This function
1257 : * must be paired with an initial call to blk_start_plug(). The intent
1258 : * is to allow the block layer to optimize I/O submission. See the
1259 : * documentation for blk_start_plug() for more information.
1260 : */
1261 0 : void blk_finish_plug(struct blk_plug *plug)
1262 : {
1263 0 : if (plug == current->plug) {
1264 0 : __blk_flush_plug(plug, false);
1265 0 : current->plug = NULL;
1266 : }
1267 0 : }
1268 : EXPORT_SYMBOL(blk_finish_plug);
1269 :
1270 0 : void blk_io_schedule(void)
1271 : {
1272 : /* Prevent hang_check timer from firing at us during very long I/O */
1273 0 : unsigned long timeout = sysctl_hung_task_timeout_secs * HZ / 2;
1274 :
1275 : if (timeout)
1276 : io_schedule_timeout(timeout);
1277 : else
1278 0 : io_schedule();
1279 0 : }
1280 : EXPORT_SYMBOL_GPL(blk_io_schedule);
1281 :
1282 1 : int __init blk_dev_init(void)
1283 : {
1284 : BUILD_BUG_ON(REQ_OP_LAST >= (1 << REQ_OP_BITS));
1285 : BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 *
1286 : sizeof_field(struct request, cmd_flags));
1287 : BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 *
1288 : sizeof_field(struct bio, bi_opf));
1289 : BUILD_BUG_ON(ALIGN(offsetof(struct request_queue, srcu),
1290 : __alignof__(struct request_queue)) !=
1291 : sizeof(struct request_queue));
1292 :
1293 : /* used for unplugging and affects IO latency/throughput - HIGHPRI */
1294 1 : kblockd_workqueue = alloc_workqueue("kblockd",
1295 : WQ_MEM_RECLAIM | WQ_HIGHPRI, 0);
1296 1 : if (!kblockd_workqueue)
1297 0 : panic("Failed to create kblockd\n");
1298 :
1299 1 : blk_requestq_cachep = kmem_cache_create("request_queue",
1300 : sizeof(struct request_queue), 0, SLAB_PANIC, NULL);
1301 :
1302 1 : blk_requestq_srcu_cachep = kmem_cache_create("request_queue_srcu",
1303 : sizeof(struct request_queue) +
1304 : sizeof(struct srcu_struct), 0, SLAB_PANIC, NULL);
1305 :
1306 2 : blk_debugfs_root = debugfs_create_dir("block", NULL);
1307 :
1308 1 : return 0;
1309 : }
|
