xref: /openbmc/linux/block/blk-core.c (revision 8dde5715)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Copyright (C) 1991, 1992 Linus Torvalds
4  * Copyright (C) 1994,      Karl Keyte: Added support for disk statistics
5  * Elevator latency, (C) 2000  Andrea Arcangeli <andrea@suse.de> SuSE
6  * Queue request tables / lock, selectable elevator, Jens Axboe <axboe@suse.de>
7  * kernel-doc documentation started by NeilBrown <neilb@cse.unsw.edu.au>
8  *	-  July2000
9  * bio rewrite, highmem i/o, etc, Jens Axboe <axboe@suse.de> - may 2001
10  */
11 
12 /*
13  * This handles all read/write requests to block devices
14  */
15 #include <linux/kernel.h>
16 #include <linux/module.h>
17 #include <linux/backing-dev.h>
18 #include <linux/bio.h>
19 #include <linux/blkdev.h>
20 #include <linux/blk-mq.h>
21 #include <linux/highmem.h>
22 #include <linux/mm.h>
23 #include <linux/kernel_stat.h>
24 #include <linux/string.h>
25 #include <linux/init.h>
26 #include <linux/completion.h>
27 #include <linux/slab.h>
28 #include <linux/swap.h>
29 #include <linux/writeback.h>
30 #include <linux/task_io_accounting_ops.h>
31 #include <linux/fault-inject.h>
32 #include <linux/list_sort.h>
33 #include <linux/delay.h>
34 #include <linux/ratelimit.h>
35 #include <linux/pm_runtime.h>
36 #include <linux/blk-cgroup.h>
37 #include <linux/debugfs.h>
38 #include <linux/bpf.h>
39 
40 #define CREATE_TRACE_POINTS
41 #include <trace/events/block.h>
42 
43 #include "blk.h"
44 #include "blk-mq.h"
45 #include "blk-mq-sched.h"
46 #include "blk-pm.h"
47 #include "blk-rq-qos.h"
48 
49 #ifdef CONFIG_DEBUG_FS
50 struct dentry *blk_debugfs_root;
51 #endif
52 
53 EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_remap);
54 EXPORT_TRACEPOINT_SYMBOL_GPL(block_rq_remap);
55 EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_complete);
56 EXPORT_TRACEPOINT_SYMBOL_GPL(block_split);
57 EXPORT_TRACEPOINT_SYMBOL_GPL(block_unplug);
58 
59 DEFINE_IDA(blk_queue_ida);
60 
61 /*
62  * For queue allocation
63  */
64 struct kmem_cache *blk_requestq_cachep;
65 
66 /*
67  * Controlling structure to kblockd
68  */
69 static struct workqueue_struct *kblockd_workqueue;
70 
71 /**
72  * blk_queue_flag_set - atomically set a queue flag
73  * @flag: flag to be set
74  * @q: request queue
75  */
76 void blk_queue_flag_set(unsigned int flag, struct request_queue *q)
77 {
78 	set_bit(flag, &q->queue_flags);
79 }
80 EXPORT_SYMBOL(blk_queue_flag_set);
81 
82 /**
83  * blk_queue_flag_clear - atomically clear a queue flag
84  * @flag: flag to be cleared
85  * @q: request queue
86  */
87 void blk_queue_flag_clear(unsigned int flag, struct request_queue *q)
88 {
89 	clear_bit(flag, &q->queue_flags);
90 }
91 EXPORT_SYMBOL(blk_queue_flag_clear);
92 
93 /**
94  * blk_queue_flag_test_and_set - atomically test and set a queue flag
95  * @flag: flag to be set
96  * @q: request queue
97  *
98  * Returns the previous value of @flag - 0 if the flag was not set and 1 if
99  * the flag was already set.
100  */
101 bool blk_queue_flag_test_and_set(unsigned int flag, struct request_queue *q)
102 {
103 	return test_and_set_bit(flag, &q->queue_flags);
104 }
105 EXPORT_SYMBOL_GPL(blk_queue_flag_test_and_set);
106 
107 void blk_rq_init(struct request_queue *q, struct request *rq)
108 {
109 	memset(rq, 0, sizeof(*rq));
110 
111 	INIT_LIST_HEAD(&rq->queuelist);
112 	rq->q = q;
113 	rq->__sector = (sector_t) -1;
114 	INIT_HLIST_NODE(&rq->hash);
115 	RB_CLEAR_NODE(&rq->rb_node);
116 	rq->tag = -1;
117 	rq->internal_tag = -1;
118 	rq->start_time_ns = ktime_get_ns();
119 	rq->part = NULL;
120 }
121 EXPORT_SYMBOL(blk_rq_init);
122 
123 static const struct {
124 	int		errno;
125 	const char	*name;
126 } blk_errors[] = {
127 	[BLK_STS_OK]		= { 0,		"" },
128 	[BLK_STS_NOTSUPP]	= { -EOPNOTSUPP, "operation not supported" },
129 	[BLK_STS_TIMEOUT]	= { -ETIMEDOUT,	"timeout" },
130 	[BLK_STS_NOSPC]		= { -ENOSPC,	"critical space allocation" },
131 	[BLK_STS_TRANSPORT]	= { -ENOLINK,	"recoverable transport" },
132 	[BLK_STS_TARGET]	= { -EREMOTEIO,	"critical target" },
133 	[BLK_STS_NEXUS]		= { -EBADE,	"critical nexus" },
134 	[BLK_STS_MEDIUM]	= { -ENODATA,	"critical medium" },
135 	[BLK_STS_PROTECTION]	= { -EILSEQ,	"protection" },
136 	[BLK_STS_RESOURCE]	= { -ENOMEM,	"kernel resource" },
137 	[BLK_STS_DEV_RESOURCE]	= { -EBUSY,	"device resource" },
138 	[BLK_STS_AGAIN]		= { -EAGAIN,	"nonblocking retry" },
139 
140 	/* device mapper special case, should not leak out: */
141 	[BLK_STS_DM_REQUEUE]	= { -EREMCHG, "dm internal retry" },
142 
143 	/* everything else not covered above: */
144 	[BLK_STS_IOERR]		= { -EIO,	"I/O" },
145 };
146 
147 blk_status_t errno_to_blk_status(int errno)
148 {
149 	int i;
150 
151 	for (i = 0; i < ARRAY_SIZE(blk_errors); i++) {
152 		if (blk_errors[i].errno == errno)
153 			return (__force blk_status_t)i;
154 	}
155 
156 	return BLK_STS_IOERR;
157 }
158 EXPORT_SYMBOL_GPL(errno_to_blk_status);
159 
160 int blk_status_to_errno(blk_status_t status)
161 {
162 	int idx = (__force int)status;
163 
164 	if (WARN_ON_ONCE(idx >= ARRAY_SIZE(blk_errors)))
165 		return -EIO;
166 	return blk_errors[idx].errno;
167 }
168 EXPORT_SYMBOL_GPL(blk_status_to_errno);
169 
170 static void print_req_error(struct request *req, blk_status_t status)
171 {
172 	int idx = (__force int)status;
173 
174 	if (WARN_ON_ONCE(idx >= ARRAY_SIZE(blk_errors)))
175 		return;
176 
177 	printk_ratelimited(KERN_ERR "%s: %s error, dev %s, sector %llu flags %x\n",
178 				__func__, blk_errors[idx].name,
179 				req->rq_disk ?  req->rq_disk->disk_name : "?",
180 				(unsigned long long)blk_rq_pos(req),
181 				req->cmd_flags);
182 }
183 
184 static void req_bio_endio(struct request *rq, struct bio *bio,
185 			  unsigned int nbytes, blk_status_t error)
186 {
187 	if (error)
188 		bio->bi_status = error;
189 
190 	if (unlikely(rq->rq_flags & RQF_QUIET))
191 		bio_set_flag(bio, BIO_QUIET);
192 
193 	bio_advance(bio, nbytes);
194 
195 	/* don't actually finish bio if it's part of flush sequence */
196 	if (bio->bi_iter.bi_size == 0 && !(rq->rq_flags & RQF_FLUSH_SEQ))
197 		bio_endio(bio);
198 }
199 
200 void blk_dump_rq_flags(struct request *rq, char *msg)
201 {
202 	printk(KERN_INFO "%s: dev %s: flags=%llx\n", msg,
203 		rq->rq_disk ? rq->rq_disk->disk_name : "?",
204 		(unsigned long long) rq->cmd_flags);
205 
206 	printk(KERN_INFO "  sector %llu, nr/cnr %u/%u\n",
207 	       (unsigned long long)blk_rq_pos(rq),
208 	       blk_rq_sectors(rq), blk_rq_cur_sectors(rq));
209 	printk(KERN_INFO "  bio %p, biotail %p, len %u\n",
210 	       rq->bio, rq->biotail, blk_rq_bytes(rq));
211 }
212 EXPORT_SYMBOL(blk_dump_rq_flags);
213 
214 /**
215  * blk_sync_queue - cancel any pending callbacks on a queue
216  * @q: the queue
217  *
218  * Description:
219  *     The block layer may perform asynchronous callback activity
220  *     on a queue, such as calling the unplug function after a timeout.
221  *     A block device may call blk_sync_queue to ensure that any
222  *     such activity is cancelled, thus allowing it to release resources
223  *     that the callbacks might use. The caller must already have made sure
224  *     that its ->make_request_fn will not re-add plugging prior to calling
225  *     this function.
226  *
227  *     This function does not cancel any asynchronous activity arising
228  *     out of elevator or throttling code. That would require elevator_exit()
229  *     and blkcg_exit_queue() to be called with queue lock initialized.
230  *
231  */
232 void blk_sync_queue(struct request_queue *q)
233 {
234 	del_timer_sync(&q->timeout);
235 	cancel_work_sync(&q->timeout_work);
236 }
237 EXPORT_SYMBOL(blk_sync_queue);
238 
239 /**
240  * blk_set_pm_only - increment pm_only counter
241  * @q: request queue pointer
242  */
243 void blk_set_pm_only(struct request_queue *q)
244 {
245 	atomic_inc(&q->pm_only);
246 }
247 EXPORT_SYMBOL_GPL(blk_set_pm_only);
248 
249 void blk_clear_pm_only(struct request_queue *q)
250 {
251 	int pm_only;
252 
253 	pm_only = atomic_dec_return(&q->pm_only);
254 	WARN_ON_ONCE(pm_only < 0);
255 	if (pm_only == 0)
256 		wake_up_all(&q->mq_freeze_wq);
257 }
258 EXPORT_SYMBOL_GPL(blk_clear_pm_only);
259 
260 void blk_put_queue(struct request_queue *q)
261 {
262 	kobject_put(&q->kobj);
263 }
264 EXPORT_SYMBOL(blk_put_queue);
265 
266 void blk_set_queue_dying(struct request_queue *q)
267 {
268 	blk_queue_flag_set(QUEUE_FLAG_DYING, q);
269 
270 	/*
271 	 * When queue DYING flag is set, we need to block new req
272 	 * entering queue, so we call blk_freeze_queue_start() to
273 	 * prevent I/O from crossing blk_queue_enter().
274 	 */
275 	blk_freeze_queue_start(q);
276 
277 	if (queue_is_mq(q))
278 		blk_mq_wake_waiters(q);
279 
280 	/* Make blk_queue_enter() reexamine the DYING flag. */
281 	wake_up_all(&q->mq_freeze_wq);
282 }
283 EXPORT_SYMBOL_GPL(blk_set_queue_dying);
284 
285 /**
286  * blk_cleanup_queue - shutdown a request queue
287  * @q: request queue to shutdown
288  *
289  * Mark @q DYING, drain all pending requests, mark @q DEAD, destroy and
290  * put it.  All future requests will be failed immediately with -ENODEV.
291  */
292 void blk_cleanup_queue(struct request_queue *q)
293 {
294 	/* mark @q DYING, no new request or merges will be allowed afterwards */
295 	mutex_lock(&q->sysfs_lock);
296 	blk_set_queue_dying(q);
297 
298 	blk_queue_flag_set(QUEUE_FLAG_NOMERGES, q);
299 	blk_queue_flag_set(QUEUE_FLAG_NOXMERGES, q);
300 	blk_queue_flag_set(QUEUE_FLAG_DYING, q);
301 	mutex_unlock(&q->sysfs_lock);
302 
303 	/*
304 	 * Drain all requests queued before DYING marking. Set DEAD flag to
305 	 * prevent that q->request_fn() gets invoked after draining finished.
306 	 */
307 	blk_freeze_queue(q);
308 
309 	rq_qos_exit(q);
310 
311 	blk_queue_flag_set(QUEUE_FLAG_DEAD, q);
312 
313 	/* for synchronous bio-based driver finish in-flight integrity i/o */
314 	blk_flush_integrity();
315 
316 	/* @q won't process any more request, flush async actions */
317 	del_timer_sync(&q->backing_dev_info->laptop_mode_wb_timer);
318 	blk_sync_queue(q);
319 
320 	if (queue_is_mq(q))
321 		blk_mq_exit_queue(q);
322 
323 	percpu_ref_exit(&q->q_usage_counter);
324 
325 	/* @q is and will stay empty, shutdown and put */
326 	blk_put_queue(q);
327 }
328 EXPORT_SYMBOL(blk_cleanup_queue);
329 
330 struct request_queue *blk_alloc_queue(gfp_t gfp_mask)
331 {
332 	return blk_alloc_queue_node(gfp_mask, NUMA_NO_NODE);
333 }
334 EXPORT_SYMBOL(blk_alloc_queue);
335 
336 /**
337  * blk_queue_enter() - try to increase q->q_usage_counter
338  * @q: request queue pointer
339  * @flags: BLK_MQ_REQ_NOWAIT and/or BLK_MQ_REQ_PREEMPT
340  */
341 int blk_queue_enter(struct request_queue *q, blk_mq_req_flags_t flags)
342 {
343 	const bool pm = flags & BLK_MQ_REQ_PREEMPT;
344 
345 	while (true) {
346 		bool success = false;
347 
348 		rcu_read_lock();
349 		if (percpu_ref_tryget_live(&q->q_usage_counter)) {
350 			/*
351 			 * The code that increments the pm_only counter is
352 			 * responsible for ensuring that that counter is
353 			 * globally visible before the queue is unfrozen.
354 			 */
355 			if (pm || !blk_queue_pm_only(q)) {
356 				success = true;
357 			} else {
358 				percpu_ref_put(&q->q_usage_counter);
359 			}
360 		}
361 		rcu_read_unlock();
362 
363 		if (success)
364 			return 0;
365 
366 		if (flags & BLK_MQ_REQ_NOWAIT)
367 			return -EBUSY;
368 
369 		/*
370 		 * read pair of barrier in blk_freeze_queue_start(),
371 		 * we need to order reading __PERCPU_REF_DEAD flag of
372 		 * .q_usage_counter and reading .mq_freeze_depth or
373 		 * queue dying flag, otherwise the following wait may
374 		 * never return if the two reads are reordered.
375 		 */
376 		smp_rmb();
377 
378 		wait_event(q->mq_freeze_wq,
379 			   (!q->mq_freeze_depth &&
380 			    (pm || (blk_pm_request_resume(q),
381 				    !blk_queue_pm_only(q)))) ||
382 			   blk_queue_dying(q));
383 		if (blk_queue_dying(q))
384 			return -ENODEV;
385 	}
386 }
387 
388 void blk_queue_exit(struct request_queue *q)
389 {
390 	percpu_ref_put(&q->q_usage_counter);
391 }
392 
393 static void blk_queue_usage_counter_release(struct percpu_ref *ref)
394 {
395 	struct request_queue *q =
396 		container_of(ref, struct request_queue, q_usage_counter);
397 
398 	wake_up_all(&q->mq_freeze_wq);
399 }
400 
401 static void blk_rq_timed_out_timer(struct timer_list *t)
402 {
403 	struct request_queue *q = from_timer(q, t, timeout);
404 
405 	kblockd_schedule_work(&q->timeout_work);
406 }
407 
408 static void blk_timeout_work(struct work_struct *work)
409 {
410 }
411 
412 /**
413  * blk_alloc_queue_node - allocate a request queue
414  * @gfp_mask: memory allocation flags
415  * @node_id: NUMA node to allocate memory from
416  */
417 struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id)
418 {
419 	struct request_queue *q;
420 	int ret;
421 
422 	q = kmem_cache_alloc_node(blk_requestq_cachep,
423 				gfp_mask | __GFP_ZERO, node_id);
424 	if (!q)
425 		return NULL;
426 
427 	INIT_LIST_HEAD(&q->queue_head);
428 	q->last_merge = NULL;
429 
430 	q->id = ida_simple_get(&blk_queue_ida, 0, 0, gfp_mask);
431 	if (q->id < 0)
432 		goto fail_q;
433 
434 	ret = bioset_init(&q->bio_split, BIO_POOL_SIZE, 0, BIOSET_NEED_BVECS);
435 	if (ret)
436 		goto fail_id;
437 
438 	q->backing_dev_info = bdi_alloc_node(gfp_mask, node_id);
439 	if (!q->backing_dev_info)
440 		goto fail_split;
441 
442 	q->stats = blk_alloc_queue_stats();
443 	if (!q->stats)
444 		goto fail_stats;
445 
446 	q->backing_dev_info->ra_pages = VM_READAHEAD_PAGES;
447 	q->backing_dev_info->capabilities = BDI_CAP_CGROUP_WRITEBACK;
448 	q->backing_dev_info->name = "block";
449 	q->node = node_id;
450 
451 	timer_setup(&q->backing_dev_info->laptop_mode_wb_timer,
452 		    laptop_mode_timer_fn, 0);
453 	timer_setup(&q->timeout, blk_rq_timed_out_timer, 0);
454 	INIT_WORK(&q->timeout_work, blk_timeout_work);
455 	INIT_LIST_HEAD(&q->icq_list);
456 #ifdef CONFIG_BLK_CGROUP
457 	INIT_LIST_HEAD(&q->blkg_list);
458 #endif
459 
460 	kobject_init(&q->kobj, &blk_queue_ktype);
461 
462 #ifdef CONFIG_BLK_DEV_IO_TRACE
463 	mutex_init(&q->blk_trace_mutex);
464 #endif
465 	mutex_init(&q->sysfs_lock);
466 	spin_lock_init(&q->queue_lock);
467 
468 	init_waitqueue_head(&q->mq_freeze_wq);
469 	mutex_init(&q->mq_freeze_lock);
470 
471 	/*
472 	 * Init percpu_ref in atomic mode so that it's faster to shutdown.
473 	 * See blk_register_queue() for details.
474 	 */
475 	if (percpu_ref_init(&q->q_usage_counter,
476 				blk_queue_usage_counter_release,
477 				PERCPU_REF_INIT_ATOMIC, GFP_KERNEL))
478 		goto fail_bdi;
479 
480 	if (blkcg_init_queue(q))
481 		goto fail_ref;
482 
483 	return q;
484 
485 fail_ref:
486 	percpu_ref_exit(&q->q_usage_counter);
487 fail_bdi:
488 	blk_free_queue_stats(q->stats);
489 fail_stats:
490 	bdi_put(q->backing_dev_info);
491 fail_split:
492 	bioset_exit(&q->bio_split);
493 fail_id:
494 	ida_simple_remove(&blk_queue_ida, q->id);
495 fail_q:
496 	kmem_cache_free(blk_requestq_cachep, q);
497 	return NULL;
498 }
499 EXPORT_SYMBOL(blk_alloc_queue_node);
500 
501 bool blk_get_queue(struct request_queue *q)
502 {
503 	if (likely(!blk_queue_dying(q))) {
504 		__blk_get_queue(q);
505 		return true;
506 	}
507 
508 	return false;
509 }
510 EXPORT_SYMBOL(blk_get_queue);
511 
512 /**
513  * blk_get_request - allocate a request
514  * @q: request queue to allocate a request for
515  * @op: operation (REQ_OP_*) and REQ_* flags, e.g. REQ_SYNC.
516  * @flags: BLK_MQ_REQ_* flags, e.g. BLK_MQ_REQ_NOWAIT.
517  */
518 struct request *blk_get_request(struct request_queue *q, unsigned int op,
519 				blk_mq_req_flags_t flags)
520 {
521 	struct request *req;
522 
523 	WARN_ON_ONCE(op & REQ_NOWAIT);
524 	WARN_ON_ONCE(flags & ~(BLK_MQ_REQ_NOWAIT | BLK_MQ_REQ_PREEMPT));
525 
526 	req = blk_mq_alloc_request(q, op, flags);
527 	if (!IS_ERR(req) && q->mq_ops->initialize_rq_fn)
528 		q->mq_ops->initialize_rq_fn(req);
529 
530 	return req;
531 }
532 EXPORT_SYMBOL(blk_get_request);
533 
534 void blk_put_request(struct request *req)
535 {
536 	blk_mq_free_request(req);
537 }
538 EXPORT_SYMBOL(blk_put_request);
539 
540 bool bio_attempt_back_merge(struct request_queue *q, struct request *req,
541 			    struct bio *bio)
542 {
543 	const int ff = bio->bi_opf & REQ_FAILFAST_MASK;
544 
545 	if (!ll_back_merge_fn(q, req, bio))
546 		return false;
547 
548 	trace_block_bio_backmerge(q, req, bio);
549 
550 	if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
551 		blk_rq_set_mixed_merge(req);
552 
553 	req->biotail->bi_next = bio;
554 	req->biotail = bio;
555 	req->__data_len += bio->bi_iter.bi_size;
556 
557 	blk_account_io_start(req, false);
558 	return true;
559 }
560 
561 bool bio_attempt_front_merge(struct request_queue *q, struct request *req,
562 			     struct bio *bio)
563 {
564 	const int ff = bio->bi_opf & REQ_FAILFAST_MASK;
565 
566 	if (!ll_front_merge_fn(q, req, bio))
567 		return false;
568 
569 	trace_block_bio_frontmerge(q, req, bio);
570 
571 	if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
572 		blk_rq_set_mixed_merge(req);
573 
574 	bio->bi_next = req->bio;
575 	req->bio = bio;
576 
577 	req->__sector = bio->bi_iter.bi_sector;
578 	req->__data_len += bio->bi_iter.bi_size;
579 
580 	blk_account_io_start(req, false);
581 	return true;
582 }
583 
584 bool bio_attempt_discard_merge(struct request_queue *q, struct request *req,
585 		struct bio *bio)
586 {
587 	unsigned short segments = blk_rq_nr_discard_segments(req);
588 
589 	if (segments >= queue_max_discard_segments(q))
590 		goto no_merge;
591 	if (blk_rq_sectors(req) + bio_sectors(bio) >
592 	    blk_rq_get_max_sectors(req, blk_rq_pos(req)))
593 		goto no_merge;
594 
595 	req->biotail->bi_next = bio;
596 	req->biotail = bio;
597 	req->__data_len += bio->bi_iter.bi_size;
598 	req->nr_phys_segments = segments + 1;
599 
600 	blk_account_io_start(req, false);
601 	return true;
602 no_merge:
603 	req_set_nomerge(q, req);
604 	return false;
605 }
606 
607 /**
608  * blk_attempt_plug_merge - try to merge with %current's plugged list
609  * @q: request_queue new bio is being queued at
610  * @bio: new bio being queued
611  * @same_queue_rq: pointer to &struct request that gets filled in when
612  * another request associated with @q is found on the plug list
613  * (optional, may be %NULL)
614  *
615  * Determine whether @bio being queued on @q can be merged with a request
616  * on %current's plugged list.  Returns %true if merge was successful,
617  * otherwise %false.
618  *
619  * Plugging coalesces IOs from the same issuer for the same purpose without
620  * going through @q->queue_lock.  As such it's more of an issuing mechanism
621  * than scheduling, and the request, while may have elvpriv data, is not
622  * added on the elevator at this point.  In addition, we don't have
623  * reliable access to the elevator outside queue lock.  Only check basic
624  * merging parameters without querying the elevator.
625  *
626  * Caller must ensure !blk_queue_nomerges(q) beforehand.
627  */
628 bool blk_attempt_plug_merge(struct request_queue *q, struct bio *bio,
629 			    struct request **same_queue_rq)
630 {
631 	struct blk_plug *plug;
632 	struct request *rq;
633 	struct list_head *plug_list;
634 
635 	plug = current->plug;
636 	if (!plug)
637 		return false;
638 
639 	plug_list = &plug->mq_list;
640 
641 	list_for_each_entry_reverse(rq, plug_list, queuelist) {
642 		bool merged = false;
643 
644 		if (rq->q == q && same_queue_rq) {
645 			/*
646 			 * Only blk-mq multiple hardware queues case checks the
647 			 * rq in the same queue, there should be only one such
648 			 * rq in a queue
649 			 **/
650 			*same_queue_rq = rq;
651 		}
652 
653 		if (rq->q != q || !blk_rq_merge_ok(rq, bio))
654 			continue;
655 
656 		switch (blk_try_merge(rq, bio)) {
657 		case ELEVATOR_BACK_MERGE:
658 			merged = bio_attempt_back_merge(q, rq, bio);
659 			break;
660 		case ELEVATOR_FRONT_MERGE:
661 			merged = bio_attempt_front_merge(q, rq, bio);
662 			break;
663 		case ELEVATOR_DISCARD_MERGE:
664 			merged = bio_attempt_discard_merge(q, rq, bio);
665 			break;
666 		default:
667 			break;
668 		}
669 
670 		if (merged)
671 			return true;
672 	}
673 
674 	return false;
675 }
676 
677 void blk_init_request_from_bio(struct request *req, struct bio *bio)
678 {
679 	if (bio->bi_opf & REQ_RAHEAD)
680 		req->cmd_flags |= REQ_FAILFAST_MASK;
681 
682 	req->__sector = bio->bi_iter.bi_sector;
683 	req->ioprio = bio_prio(bio);
684 	req->write_hint = bio->bi_write_hint;
685 	blk_rq_bio_prep(req->q, req, bio);
686 }
687 EXPORT_SYMBOL_GPL(blk_init_request_from_bio);
688 
689 static void handle_bad_sector(struct bio *bio, sector_t maxsector)
690 {
691 	char b[BDEVNAME_SIZE];
692 
693 	printk(KERN_INFO "attempt to access beyond end of device\n");
694 	printk(KERN_INFO "%s: rw=%d, want=%Lu, limit=%Lu\n",
695 			bio_devname(bio, b), bio->bi_opf,
696 			(unsigned long long)bio_end_sector(bio),
697 			(long long)maxsector);
698 }
699 
700 #ifdef CONFIG_FAIL_MAKE_REQUEST
701 
702 static DECLARE_FAULT_ATTR(fail_make_request);
703 
704 static int __init setup_fail_make_request(char *str)
705 {
706 	return setup_fault_attr(&fail_make_request, str);
707 }
708 __setup("fail_make_request=", setup_fail_make_request);
709 
710 static bool should_fail_request(struct hd_struct *part, unsigned int bytes)
711 {
712 	return part->make_it_fail && should_fail(&fail_make_request, bytes);
713 }
714 
715 static int __init fail_make_request_debugfs(void)
716 {
717 	struct dentry *dir = fault_create_debugfs_attr("fail_make_request",
718 						NULL, &fail_make_request);
719 
720 	return PTR_ERR_OR_ZERO(dir);
721 }
722 
723 late_initcall(fail_make_request_debugfs);
724 
725 #else /* CONFIG_FAIL_MAKE_REQUEST */
726 
727 static inline bool should_fail_request(struct hd_struct *part,
728 					unsigned int bytes)
729 {
730 	return false;
731 }
732 
733 #endif /* CONFIG_FAIL_MAKE_REQUEST */
734 
735 static inline bool bio_check_ro(struct bio *bio, struct hd_struct *part)
736 {
737 	const int op = bio_op(bio);
738 
739 	if (part->policy && op_is_write(op)) {
740 		char b[BDEVNAME_SIZE];
741 
742 		if (op_is_flush(bio->bi_opf) && !bio_sectors(bio))
743 			return false;
744 
745 		WARN_ONCE(1,
746 		       "generic_make_request: Trying to write "
747 			"to read-only block-device %s (partno %d)\n",
748 			bio_devname(bio, b), part->partno);
749 		/* Older lvm-tools actually trigger this */
750 		return false;
751 	}
752 
753 	return false;
754 }
755 
756 static noinline int should_fail_bio(struct bio *bio)
757 {
758 	if (should_fail_request(&bio->bi_disk->part0, bio->bi_iter.bi_size))
759 		return -EIO;
760 	return 0;
761 }
762 ALLOW_ERROR_INJECTION(should_fail_bio, ERRNO);
763 
764 /*
765  * Check whether this bio extends beyond the end of the device or partition.
766  * This may well happen - the kernel calls bread() without checking the size of
767  * the device, e.g., when mounting a file system.
768  */
769 static inline int bio_check_eod(struct bio *bio, sector_t maxsector)
770 {
771 	unsigned int nr_sectors = bio_sectors(bio);
772 
773 	if (nr_sectors && maxsector &&
774 	    (nr_sectors > maxsector ||
775 	     bio->bi_iter.bi_sector > maxsector - nr_sectors)) {
776 		handle_bad_sector(bio, maxsector);
777 		return -EIO;
778 	}
779 	return 0;
780 }
781 
782 /*
783  * Remap block n of partition p to block n+start(p) of the disk.
784  */
785 static inline int blk_partition_remap(struct bio *bio)
786 {
787 	struct hd_struct *p;
788 	int ret = -EIO;
789 
790 	rcu_read_lock();
791 	p = __disk_get_part(bio->bi_disk, bio->bi_partno);
792 	if (unlikely(!p))
793 		goto out;
794 	if (unlikely(should_fail_request(p, bio->bi_iter.bi_size)))
795 		goto out;
796 	if (unlikely(bio_check_ro(bio, p)))
797 		goto out;
798 
799 	/*
800 	 * Zone reset does not include bi_size so bio_sectors() is always 0.
801 	 * Include a test for the reset op code and perform the remap if needed.
802 	 */
803 	if (bio_sectors(bio) || bio_op(bio) == REQ_OP_ZONE_RESET) {
804 		if (bio_check_eod(bio, part_nr_sects_read(p)))
805 			goto out;
806 		bio->bi_iter.bi_sector += p->start_sect;
807 		trace_block_bio_remap(bio->bi_disk->queue, bio, part_devt(p),
808 				      bio->bi_iter.bi_sector - p->start_sect);
809 	}
810 	bio->bi_partno = 0;
811 	ret = 0;
812 out:
813 	rcu_read_unlock();
814 	return ret;
815 }
816 
817 static noinline_for_stack bool
818 generic_make_request_checks(struct bio *bio)
819 {
820 	struct request_queue *q;
821 	int nr_sectors = bio_sectors(bio);
822 	blk_status_t status = BLK_STS_IOERR;
823 	char b[BDEVNAME_SIZE];
824 
825 	might_sleep();
826 
827 	q = bio->bi_disk->queue;
828 	if (unlikely(!q)) {
829 		printk(KERN_ERR
830 		       "generic_make_request: Trying to access "
831 			"nonexistent block-device %s (%Lu)\n",
832 			bio_devname(bio, b), (long long)bio->bi_iter.bi_sector);
833 		goto end_io;
834 	}
835 
836 	/*
837 	 * For a REQ_NOWAIT based request, return -EOPNOTSUPP
838 	 * if queue is not a request based queue.
839 	 */
840 	if ((bio->bi_opf & REQ_NOWAIT) && !queue_is_mq(q))
841 		goto not_supported;
842 
843 	if (should_fail_bio(bio))
844 		goto end_io;
845 
846 	if (bio->bi_partno) {
847 		if (unlikely(blk_partition_remap(bio)))
848 			goto end_io;
849 	} else {
850 		if (unlikely(bio_check_ro(bio, &bio->bi_disk->part0)))
851 			goto end_io;
852 		if (unlikely(bio_check_eod(bio, get_capacity(bio->bi_disk))))
853 			goto end_io;
854 	}
855 
856 	/*
857 	 * Filter flush bio's early so that make_request based
858 	 * drivers without flush support don't have to worry
859 	 * about them.
860 	 */
861 	if (op_is_flush(bio->bi_opf) &&
862 	    !test_bit(QUEUE_FLAG_WC, &q->queue_flags)) {
863 		bio->bi_opf &= ~(REQ_PREFLUSH | REQ_FUA);
864 		if (!nr_sectors) {
865 			status = BLK_STS_OK;
866 			goto end_io;
867 		}
868 	}
869 
870 	if (!test_bit(QUEUE_FLAG_POLL, &q->queue_flags))
871 		bio->bi_opf &= ~REQ_HIPRI;
872 
873 	switch (bio_op(bio)) {
874 	case REQ_OP_DISCARD:
875 		if (!blk_queue_discard(q))
876 			goto not_supported;
877 		break;
878 	case REQ_OP_SECURE_ERASE:
879 		if (!blk_queue_secure_erase(q))
880 			goto not_supported;
881 		break;
882 	case REQ_OP_WRITE_SAME:
883 		if (!q->limits.max_write_same_sectors)
884 			goto not_supported;
885 		break;
886 	case REQ_OP_ZONE_RESET:
887 		if (!blk_queue_is_zoned(q))
888 			goto not_supported;
889 		break;
890 	case REQ_OP_WRITE_ZEROES:
891 		if (!q->limits.max_write_zeroes_sectors)
892 			goto not_supported;
893 		break;
894 	default:
895 		break;
896 	}
897 
898 	/*
899 	 * Various block parts want %current->io_context and lazy ioc
900 	 * allocation ends up trading a lot of pain for a small amount of
901 	 * memory.  Just allocate it upfront.  This may fail and block
902 	 * layer knows how to live with it.
903 	 */
904 	create_io_context(GFP_ATOMIC, q->node);
905 
906 	if (!blkcg_bio_issue_check(q, bio))
907 		return false;
908 
909 	if (!bio_flagged(bio, BIO_TRACE_COMPLETION)) {
910 		trace_block_bio_queue(q, bio);
911 		/* Now that enqueuing has been traced, we need to trace
912 		 * completion as well.
913 		 */
914 		bio_set_flag(bio, BIO_TRACE_COMPLETION);
915 	}
916 	return true;
917 
918 not_supported:
919 	status = BLK_STS_NOTSUPP;
920 end_io:
921 	bio->bi_status = status;
922 	bio_endio(bio);
923 	return false;
924 }
925 
926 /**
927  * generic_make_request - hand a buffer to its device driver for I/O
928  * @bio:  The bio describing the location in memory and on the device.
929  *
930  * generic_make_request() is used to make I/O requests of block
931  * devices. It is passed a &struct bio, which describes the I/O that needs
932  * to be done.
933  *
934  * generic_make_request() does not return any status.  The
935  * success/failure status of the request, along with notification of
936  * completion, is delivered asynchronously through the bio->bi_end_io
937  * function described (one day) else where.
938  *
939  * The caller of generic_make_request must make sure that bi_io_vec
940  * are set to describe the memory buffer, and that bi_dev and bi_sector are
941  * set to describe the device address, and the
942  * bi_end_io and optionally bi_private are set to describe how
943  * completion notification should be signaled.
944  *
945  * generic_make_request and the drivers it calls may use bi_next if this
946  * bio happens to be merged with someone else, and may resubmit the bio to
947  * a lower device by calling into generic_make_request recursively, which
948  * means the bio should NOT be touched after the call to ->make_request_fn.
949  */
950 blk_qc_t generic_make_request(struct bio *bio)
951 {
952 	/*
953 	 * bio_list_on_stack[0] contains bios submitted by the current
954 	 * make_request_fn.
955 	 * bio_list_on_stack[1] contains bios that were submitted before
956 	 * the current make_request_fn, but that haven't been processed
957 	 * yet.
958 	 */
959 	struct bio_list bio_list_on_stack[2];
960 	blk_qc_t ret = BLK_QC_T_NONE;
961 
962 	if (!generic_make_request_checks(bio))
963 		goto out;
964 
965 	/*
966 	 * We only want one ->make_request_fn to be active at a time, else
967 	 * stack usage with stacked devices could be a problem.  So use
968 	 * current->bio_list to keep a list of requests submited by a
969 	 * make_request_fn function.  current->bio_list is also used as a
970 	 * flag to say if generic_make_request is currently active in this
971 	 * task or not.  If it is NULL, then no make_request is active.  If
972 	 * it is non-NULL, then a make_request is active, and new requests
973 	 * should be added at the tail
974 	 */
975 	if (current->bio_list) {
976 		bio_list_add(&current->bio_list[0], bio);
977 		goto out;
978 	}
979 
980 	/* following loop may be a bit non-obvious, and so deserves some
981 	 * explanation.
982 	 * Before entering the loop, bio->bi_next is NULL (as all callers
983 	 * ensure that) so we have a list with a single bio.
984 	 * We pretend that we have just taken it off a longer list, so
985 	 * we assign bio_list to a pointer to the bio_list_on_stack,
986 	 * thus initialising the bio_list of new bios to be
987 	 * added.  ->make_request() may indeed add some more bios
988 	 * through a recursive call to generic_make_request.  If it
989 	 * did, we find a non-NULL value in bio_list and re-enter the loop
990 	 * from the top.  In this case we really did just take the bio
991 	 * of the top of the list (no pretending) and so remove it from
992 	 * bio_list, and call into ->make_request() again.
993 	 */
994 	BUG_ON(bio->bi_next);
995 	bio_list_init(&bio_list_on_stack[0]);
996 	current->bio_list = bio_list_on_stack;
997 	do {
998 		struct request_queue *q = bio->bi_disk->queue;
999 		blk_mq_req_flags_t flags = bio->bi_opf & REQ_NOWAIT ?
1000 			BLK_MQ_REQ_NOWAIT : 0;
1001 
1002 		if (likely(blk_queue_enter(q, flags) == 0)) {
1003 			struct bio_list lower, same;
1004 
1005 			/* Create a fresh bio_list for all subordinate requests */
1006 			bio_list_on_stack[1] = bio_list_on_stack[0];
1007 			bio_list_init(&bio_list_on_stack[0]);
1008 			ret = q->make_request_fn(q, bio);
1009 
1010 			blk_queue_exit(q);
1011 
1012 			/* sort new bios into those for a lower level
1013 			 * and those for the same level
1014 			 */
1015 			bio_list_init(&lower);
1016 			bio_list_init(&same);
1017 			while ((bio = bio_list_pop(&bio_list_on_stack[0])) != NULL)
1018 				if (q == bio->bi_disk->queue)
1019 					bio_list_add(&same, bio);
1020 				else
1021 					bio_list_add(&lower, bio);
1022 			/* now assemble so we handle the lowest level first */
1023 			bio_list_merge(&bio_list_on_stack[0], &lower);
1024 			bio_list_merge(&bio_list_on_stack[0], &same);
1025 			bio_list_merge(&bio_list_on_stack[0], &bio_list_on_stack[1]);
1026 		} else {
1027 			if (unlikely(!blk_queue_dying(q) &&
1028 					(bio->bi_opf & REQ_NOWAIT)))
1029 				bio_wouldblock_error(bio);
1030 			else
1031 				bio_io_error(bio);
1032 		}
1033 		bio = bio_list_pop(&bio_list_on_stack[0]);
1034 	} while (bio);
1035 	current->bio_list = NULL; /* deactivate */
1036 
1037 out:
1038 	return ret;
1039 }
1040 EXPORT_SYMBOL(generic_make_request);
1041 
1042 /**
1043  * direct_make_request - hand a buffer directly to its device driver for I/O
1044  * @bio:  The bio describing the location in memory and on the device.
1045  *
1046  * This function behaves like generic_make_request(), but does not protect
1047  * against recursion.  Must only be used if the called driver is known
1048  * to not call generic_make_request (or direct_make_request) again from
1049  * its make_request function.  (Calling direct_make_request again from
1050  * a workqueue is perfectly fine as that doesn't recurse).
1051  */
1052 blk_qc_t direct_make_request(struct bio *bio)
1053 {
1054 	struct request_queue *q = bio->bi_disk->queue;
1055 	bool nowait = bio->bi_opf & REQ_NOWAIT;
1056 	blk_qc_t ret;
1057 
1058 	if (!generic_make_request_checks(bio))
1059 		return BLK_QC_T_NONE;
1060 
1061 	if (unlikely(blk_queue_enter(q, nowait ? BLK_MQ_REQ_NOWAIT : 0))) {
1062 		if (nowait && !blk_queue_dying(q))
1063 			bio->bi_status = BLK_STS_AGAIN;
1064 		else
1065 			bio->bi_status = BLK_STS_IOERR;
1066 		bio_endio(bio);
1067 		return BLK_QC_T_NONE;
1068 	}
1069 
1070 	ret = q->make_request_fn(q, bio);
1071 	blk_queue_exit(q);
1072 	return ret;
1073 }
1074 EXPORT_SYMBOL_GPL(direct_make_request);
1075 
1076 /**
1077  * submit_bio - submit a bio to the block device layer for I/O
1078  * @bio: The &struct bio which describes the I/O
1079  *
1080  * submit_bio() is very similar in purpose to generic_make_request(), and
1081  * uses that function to do most of the work. Both are fairly rough
1082  * interfaces; @bio must be presetup and ready for I/O.
1083  *
1084  */
1085 blk_qc_t submit_bio(struct bio *bio)
1086 {
1087 	/*
1088 	 * If it's a regular read/write or a barrier with data attached,
1089 	 * go through the normal accounting stuff before submission.
1090 	 */
1091 	if (bio_has_data(bio)) {
1092 		unsigned int count;
1093 
1094 		if (unlikely(bio_op(bio) == REQ_OP_WRITE_SAME))
1095 			count = queue_logical_block_size(bio->bi_disk->queue) >> 9;
1096 		else
1097 			count = bio_sectors(bio);
1098 
1099 		if (op_is_write(bio_op(bio))) {
1100 			count_vm_events(PGPGOUT, count);
1101 		} else {
1102 			task_io_account_read(bio->bi_iter.bi_size);
1103 			count_vm_events(PGPGIN, count);
1104 		}
1105 
1106 		if (unlikely(block_dump)) {
1107 			char b[BDEVNAME_SIZE];
1108 			printk(KERN_DEBUG "%s(%d): %s block %Lu on %s (%u sectors)\n",
1109 			current->comm, task_pid_nr(current),
1110 				op_is_write(bio_op(bio)) ? "WRITE" : "READ",
1111 				(unsigned long long)bio->bi_iter.bi_sector,
1112 				bio_devname(bio, b), count);
1113 		}
1114 	}
1115 
1116 	return generic_make_request(bio);
1117 }
1118 EXPORT_SYMBOL(submit_bio);
1119 
1120 /**
1121  * blk_cloned_rq_check_limits - Helper function to check a cloned request
1122  *                              for new the queue limits
1123  * @q:  the queue
1124  * @rq: the request being checked
1125  *
1126  * Description:
1127  *    @rq may have been made based on weaker limitations of upper-level queues
1128  *    in request stacking drivers, and it may violate the limitation of @q.
1129  *    Since the block layer and the underlying device driver trust @rq
1130  *    after it is inserted to @q, it should be checked against @q before
1131  *    the insertion using this generic function.
1132  *
1133  *    Request stacking drivers like request-based dm may change the queue
1134  *    limits when retrying requests on other queues. Those requests need
1135  *    to be checked against the new queue limits again during dispatch.
1136  */
1137 static int blk_cloned_rq_check_limits(struct request_queue *q,
1138 				      struct request *rq)
1139 {
1140 	if (blk_rq_sectors(rq) > blk_queue_get_max_sectors(q, req_op(rq))) {
1141 		printk(KERN_ERR "%s: over max size limit. (%u > %u)\n",
1142 			__func__, blk_rq_sectors(rq),
1143 			blk_queue_get_max_sectors(q, req_op(rq)));
1144 		return -EIO;
1145 	}
1146 
1147 	/*
1148 	 * queue's settings related to segment counting like q->bounce_pfn
1149 	 * may differ from that of other stacking queues.
1150 	 * Recalculate it to check the request correctly on this queue's
1151 	 * limitation.
1152 	 */
1153 	blk_recalc_rq_segments(rq);
1154 	if (rq->nr_phys_segments > queue_max_segments(q)) {
1155 		printk(KERN_ERR "%s: over max segments limit. (%hu > %hu)\n",
1156 			__func__, rq->nr_phys_segments, queue_max_segments(q));
1157 		return -EIO;
1158 	}
1159 
1160 	return 0;
1161 }
1162 
1163 /**
1164  * blk_insert_cloned_request - Helper for stacking drivers to submit a request
1165  * @q:  the queue to submit the request
1166  * @rq: the request being queued
1167  */
1168 blk_status_t blk_insert_cloned_request(struct request_queue *q, struct request *rq)
1169 {
1170 	if (blk_cloned_rq_check_limits(q, rq))
1171 		return BLK_STS_IOERR;
1172 
1173 	if (rq->rq_disk &&
1174 	    should_fail_request(&rq->rq_disk->part0, blk_rq_bytes(rq)))
1175 		return BLK_STS_IOERR;
1176 
1177 	if (blk_queue_io_stat(q))
1178 		blk_account_io_start(rq, true);
1179 
1180 	/*
1181 	 * Since we have a scheduler attached on the top device,
1182 	 * bypass a potential scheduler on the bottom device for
1183 	 * insert.
1184 	 */
1185 	return blk_mq_request_issue_directly(rq, true);
1186 }
1187 EXPORT_SYMBOL_GPL(blk_insert_cloned_request);
1188 
1189 /**
1190  * blk_rq_err_bytes - determine number of bytes till the next failure boundary
1191  * @rq: request to examine
1192  *
1193  * Description:
1194  *     A request could be merge of IOs which require different failure
1195  *     handling.  This function determines the number of bytes which
1196  *     can be failed from the beginning of the request without
1197  *     crossing into area which need to be retried further.
1198  *
1199  * Return:
1200  *     The number of bytes to fail.
1201  */
1202 unsigned int blk_rq_err_bytes(const struct request *rq)
1203 {
1204 	unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK;
1205 	unsigned int bytes = 0;
1206 	struct bio *bio;
1207 
1208 	if (!(rq->rq_flags & RQF_MIXED_MERGE))
1209 		return blk_rq_bytes(rq);
1210 
1211 	/*
1212 	 * Currently the only 'mixing' which can happen is between
1213 	 * different fastfail types.  We can safely fail portions
1214 	 * which have all the failfast bits that the first one has -
1215 	 * the ones which are at least as eager to fail as the first
1216 	 * one.
1217 	 */
1218 	for (bio = rq->bio; bio; bio = bio->bi_next) {
1219 		if ((bio->bi_opf & ff) != ff)
1220 			break;
1221 		bytes += bio->bi_iter.bi_size;
1222 	}
1223 
1224 	/* this could lead to infinite loop */
1225 	BUG_ON(blk_rq_bytes(rq) && !bytes);
1226 	return bytes;
1227 }
1228 EXPORT_SYMBOL_GPL(blk_rq_err_bytes);
1229 
1230 void blk_account_io_completion(struct request *req, unsigned int bytes)
1231 {
1232 	if (blk_do_io_stat(req)) {
1233 		const int sgrp = op_stat_group(req_op(req));
1234 		struct hd_struct *part;
1235 
1236 		part_stat_lock();
1237 		part = req->part;
1238 		part_stat_add(part, sectors[sgrp], bytes >> 9);
1239 		part_stat_unlock();
1240 	}
1241 }
1242 
1243 void blk_account_io_done(struct request *req, u64 now)
1244 {
1245 	/*
1246 	 * Account IO completion.  flush_rq isn't accounted as a
1247 	 * normal IO on queueing nor completion.  Accounting the
1248 	 * containing request is enough.
1249 	 */
1250 	if (blk_do_io_stat(req) && !(req->rq_flags & RQF_FLUSH_SEQ)) {
1251 		const int sgrp = op_stat_group(req_op(req));
1252 		struct hd_struct *part;
1253 
1254 		part_stat_lock();
1255 		part = req->part;
1256 
1257 		update_io_ticks(part, jiffies);
1258 		part_stat_inc(part, ios[sgrp]);
1259 		part_stat_add(part, nsecs[sgrp], now - req->start_time_ns);
1260 		part_stat_add(part, time_in_queue, nsecs_to_jiffies64(now - req->start_time_ns));
1261 		part_dec_in_flight(req->q, part, rq_data_dir(req));
1262 
1263 		hd_struct_put(part);
1264 		part_stat_unlock();
1265 	}
1266 }
1267 
1268 void blk_account_io_start(struct request *rq, bool new_io)
1269 {
1270 	struct hd_struct *part;
1271 	int rw = rq_data_dir(rq);
1272 
1273 	if (!blk_do_io_stat(rq))
1274 		return;
1275 
1276 	part_stat_lock();
1277 
1278 	if (!new_io) {
1279 		part = rq->part;
1280 		part_stat_inc(part, merges[rw]);
1281 	} else {
1282 		part = disk_map_sector_rcu(rq->rq_disk, blk_rq_pos(rq));
1283 		if (!hd_struct_try_get(part)) {
1284 			/*
1285 			 * The partition is already being removed,
1286 			 * the request will be accounted on the disk only
1287 			 *
1288 			 * We take a reference on disk->part0 although that
1289 			 * partition will never be deleted, so we can treat
1290 			 * it as any other partition.
1291 			 */
1292 			part = &rq->rq_disk->part0;
1293 			hd_struct_get(part);
1294 		}
1295 		part_inc_in_flight(rq->q, part, rw);
1296 		rq->part = part;
1297 	}
1298 
1299 	update_io_ticks(part, jiffies);
1300 
1301 	part_stat_unlock();
1302 }
1303 
1304 /*
1305  * Steal bios from a request and add them to a bio list.
1306  * The request must not have been partially completed before.
1307  */
1308 void blk_steal_bios(struct bio_list *list, struct request *rq)
1309 {
1310 	if (rq->bio) {
1311 		if (list->tail)
1312 			list->tail->bi_next = rq->bio;
1313 		else
1314 			list->head = rq->bio;
1315 		list->tail = rq->biotail;
1316 
1317 		rq->bio = NULL;
1318 		rq->biotail = NULL;
1319 	}
1320 
1321 	rq->__data_len = 0;
1322 }
1323 EXPORT_SYMBOL_GPL(blk_steal_bios);
1324 
1325 /**
1326  * blk_update_request - Special helper function for request stacking drivers
1327  * @req:      the request being processed
1328  * @error:    block status code
1329  * @nr_bytes: number of bytes to complete @req
1330  *
1331  * Description:
1332  *     Ends I/O on a number of bytes attached to @req, but doesn't complete
1333  *     the request structure even if @req doesn't have leftover.
1334  *     If @req has leftover, sets it up for the next range of segments.
1335  *
1336  *     This special helper function is only for request stacking drivers
1337  *     (e.g. request-based dm) so that they can handle partial completion.
1338  *     Actual device drivers should use blk_end_request instead.
1339  *
1340  *     Passing the result of blk_rq_bytes() as @nr_bytes guarantees
1341  *     %false return from this function.
1342  *
1343  * Note:
1344  *	The RQF_SPECIAL_PAYLOAD flag is ignored on purpose in both
1345  *	blk_rq_bytes() and in blk_update_request().
1346  *
1347  * Return:
1348  *     %false - this request doesn't have any more data
1349  *     %true  - this request has more data
1350  **/
1351 bool blk_update_request(struct request *req, blk_status_t error,
1352 		unsigned int nr_bytes)
1353 {
1354 	int total_bytes;
1355 
1356 	trace_block_rq_complete(req, blk_status_to_errno(error), nr_bytes);
1357 
1358 	if (!req->bio)
1359 		return false;
1360 
1361 	if (unlikely(error && !blk_rq_is_passthrough(req) &&
1362 		     !(req->rq_flags & RQF_QUIET)))
1363 		print_req_error(req, error);
1364 
1365 	blk_account_io_completion(req, nr_bytes);
1366 
1367 	total_bytes = 0;
1368 	while (req->bio) {
1369 		struct bio *bio = req->bio;
1370 		unsigned bio_bytes = min(bio->bi_iter.bi_size, nr_bytes);
1371 
1372 		if (bio_bytes == bio->bi_iter.bi_size)
1373 			req->bio = bio->bi_next;
1374 
1375 		/* Completion has already been traced */
1376 		bio_clear_flag(bio, BIO_TRACE_COMPLETION);
1377 		req_bio_endio(req, bio, bio_bytes, error);
1378 
1379 		total_bytes += bio_bytes;
1380 		nr_bytes -= bio_bytes;
1381 
1382 		if (!nr_bytes)
1383 			break;
1384 	}
1385 
1386 	/*
1387 	 * completely done
1388 	 */
1389 	if (!req->bio) {
1390 		/*
1391 		 * Reset counters so that the request stacking driver
1392 		 * can find how many bytes remain in the request
1393 		 * later.
1394 		 */
1395 		req->__data_len = 0;
1396 		return false;
1397 	}
1398 
1399 	req->__data_len -= total_bytes;
1400 
1401 	/* update sector only for requests with clear definition of sector */
1402 	if (!blk_rq_is_passthrough(req))
1403 		req->__sector += total_bytes >> 9;
1404 
1405 	/* mixed attributes always follow the first bio */
1406 	if (req->rq_flags & RQF_MIXED_MERGE) {
1407 		req->cmd_flags &= ~REQ_FAILFAST_MASK;
1408 		req->cmd_flags |= req->bio->bi_opf & REQ_FAILFAST_MASK;
1409 	}
1410 
1411 	if (!(req->rq_flags & RQF_SPECIAL_PAYLOAD)) {
1412 		/*
1413 		 * If total number of sectors is less than the first segment
1414 		 * size, something has gone terribly wrong.
1415 		 */
1416 		if (blk_rq_bytes(req) < blk_rq_cur_bytes(req)) {
1417 			blk_dump_rq_flags(req, "request botched");
1418 			req->__data_len = blk_rq_cur_bytes(req);
1419 		}
1420 
1421 		/* recalculate the number of segments */
1422 		blk_recalc_rq_segments(req);
1423 	}
1424 
1425 	return true;
1426 }
1427 EXPORT_SYMBOL_GPL(blk_update_request);
1428 
1429 void blk_rq_bio_prep(struct request_queue *q, struct request *rq,
1430 		     struct bio *bio)
1431 {
1432 	if (bio_has_data(bio))
1433 		rq->nr_phys_segments = bio_phys_segments(q, bio);
1434 	else if (bio_op(bio) == REQ_OP_DISCARD)
1435 		rq->nr_phys_segments = 1;
1436 
1437 	rq->__data_len = bio->bi_iter.bi_size;
1438 	rq->bio = rq->biotail = bio;
1439 
1440 	if (bio->bi_disk)
1441 		rq->rq_disk = bio->bi_disk;
1442 }
1443 
1444 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
1445 /**
1446  * rq_flush_dcache_pages - Helper function to flush all pages in a request
1447  * @rq: the request to be flushed
1448  *
1449  * Description:
1450  *     Flush all pages in @rq.
1451  */
1452 void rq_flush_dcache_pages(struct request *rq)
1453 {
1454 	struct req_iterator iter;
1455 	struct bio_vec bvec;
1456 
1457 	rq_for_each_segment(bvec, rq, iter)
1458 		flush_dcache_page(bvec.bv_page);
1459 }
1460 EXPORT_SYMBOL_GPL(rq_flush_dcache_pages);
1461 #endif
1462 
1463 /**
1464  * blk_lld_busy - Check if underlying low-level drivers of a device are busy
1465  * @q : the queue of the device being checked
1466  *
1467  * Description:
1468  *    Check if underlying low-level drivers of a device are busy.
1469  *    If the drivers want to export their busy state, they must set own
1470  *    exporting function using blk_queue_lld_busy() first.
1471  *
1472  *    Basically, this function is used only by request stacking drivers
1473  *    to stop dispatching requests to underlying devices when underlying
1474  *    devices are busy.  This behavior helps more I/O merging on the queue
1475  *    of the request stacking driver and prevents I/O throughput regression
1476  *    on burst I/O load.
1477  *
1478  * Return:
1479  *    0 - Not busy (The request stacking driver should dispatch request)
1480  *    1 - Busy (The request stacking driver should stop dispatching request)
1481  */
1482 int blk_lld_busy(struct request_queue *q)
1483 {
1484 	if (queue_is_mq(q) && q->mq_ops->busy)
1485 		return q->mq_ops->busy(q);
1486 
1487 	return 0;
1488 }
1489 EXPORT_SYMBOL_GPL(blk_lld_busy);
1490 
1491 /**
1492  * blk_rq_unprep_clone - Helper function to free all bios in a cloned request
1493  * @rq: the clone request to be cleaned up
1494  *
1495  * Description:
1496  *     Free all bios in @rq for a cloned request.
1497  */
1498 void blk_rq_unprep_clone(struct request *rq)
1499 {
1500 	struct bio *bio;
1501 
1502 	while ((bio = rq->bio) != NULL) {
1503 		rq->bio = bio->bi_next;
1504 
1505 		bio_put(bio);
1506 	}
1507 }
1508 EXPORT_SYMBOL_GPL(blk_rq_unprep_clone);
1509 
1510 /*
1511  * Copy attributes of the original request to the clone request.
1512  * The actual data parts (e.g. ->cmd, ->sense) are not copied.
1513  */
1514 static void __blk_rq_prep_clone(struct request *dst, struct request *src)
1515 {
1516 	dst->__sector = blk_rq_pos(src);
1517 	dst->__data_len = blk_rq_bytes(src);
1518 	if (src->rq_flags & RQF_SPECIAL_PAYLOAD) {
1519 		dst->rq_flags |= RQF_SPECIAL_PAYLOAD;
1520 		dst->special_vec = src->special_vec;
1521 	}
1522 	dst->nr_phys_segments = src->nr_phys_segments;
1523 	dst->ioprio = src->ioprio;
1524 	dst->extra_len = src->extra_len;
1525 }
1526 
1527 /**
1528  * blk_rq_prep_clone - Helper function to setup clone request
1529  * @rq: the request to be setup
1530  * @rq_src: original request to be cloned
1531  * @bs: bio_set that bios for clone are allocated from
1532  * @gfp_mask: memory allocation mask for bio
1533  * @bio_ctr: setup function to be called for each clone bio.
1534  *           Returns %0 for success, non %0 for failure.
1535  * @data: private data to be passed to @bio_ctr
1536  *
1537  * Description:
1538  *     Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq.
1539  *     The actual data parts of @rq_src (e.g. ->cmd, ->sense)
1540  *     are not copied, and copying such parts is the caller's responsibility.
1541  *     Also, pages which the original bios are pointing to are not copied
1542  *     and the cloned bios just point same pages.
1543  *     So cloned bios must be completed before original bios, which means
1544  *     the caller must complete @rq before @rq_src.
1545  */
1546 int blk_rq_prep_clone(struct request *rq, struct request *rq_src,
1547 		      struct bio_set *bs, gfp_t gfp_mask,
1548 		      int (*bio_ctr)(struct bio *, struct bio *, void *),
1549 		      void *data)
1550 {
1551 	struct bio *bio, *bio_src;
1552 
1553 	if (!bs)
1554 		bs = &fs_bio_set;
1555 
1556 	__rq_for_each_bio(bio_src, rq_src) {
1557 		bio = bio_clone_fast(bio_src, gfp_mask, bs);
1558 		if (!bio)
1559 			goto free_and_out;
1560 
1561 		if (bio_ctr && bio_ctr(bio, bio_src, data))
1562 			goto free_and_out;
1563 
1564 		if (rq->bio) {
1565 			rq->biotail->bi_next = bio;
1566 			rq->biotail = bio;
1567 		} else
1568 			rq->bio = rq->biotail = bio;
1569 	}
1570 
1571 	__blk_rq_prep_clone(rq, rq_src);
1572 
1573 	return 0;
1574 
1575 free_and_out:
1576 	if (bio)
1577 		bio_put(bio);
1578 	blk_rq_unprep_clone(rq);
1579 
1580 	return -ENOMEM;
1581 }
1582 EXPORT_SYMBOL_GPL(blk_rq_prep_clone);
1583 
1584 int kblockd_schedule_work(struct work_struct *work)
1585 {
1586 	return queue_work(kblockd_workqueue, work);
1587 }
1588 EXPORT_SYMBOL(kblockd_schedule_work);
1589 
1590 int kblockd_schedule_work_on(int cpu, struct work_struct *work)
1591 {
1592 	return queue_work_on(cpu, kblockd_workqueue, work);
1593 }
1594 EXPORT_SYMBOL(kblockd_schedule_work_on);
1595 
1596 int kblockd_mod_delayed_work_on(int cpu, struct delayed_work *dwork,
1597 				unsigned long delay)
1598 {
1599 	return mod_delayed_work_on(cpu, kblockd_workqueue, dwork, delay);
1600 }
1601 EXPORT_SYMBOL(kblockd_mod_delayed_work_on);
1602 
1603 /**
1604  * blk_start_plug - initialize blk_plug and track it inside the task_struct
1605  * @plug:	The &struct blk_plug that needs to be initialized
1606  *
1607  * Description:
1608  *   blk_start_plug() indicates to the block layer an intent by the caller
1609  *   to submit multiple I/O requests in a batch.  The block layer may use
1610  *   this hint to defer submitting I/Os from the caller until blk_finish_plug()
1611  *   is called.  However, the block layer may choose to submit requests
1612  *   before a call to blk_finish_plug() if the number of queued I/Os
1613  *   exceeds %BLK_MAX_REQUEST_COUNT, or if the size of the I/O is larger than
1614  *   %BLK_PLUG_FLUSH_SIZE.  The queued I/Os may also be submitted early if
1615  *   the task schedules (see below).
1616  *
1617  *   Tracking blk_plug inside the task_struct will help with auto-flushing the
1618  *   pending I/O should the task end up blocking between blk_start_plug() and
1619  *   blk_finish_plug(). This is important from a performance perspective, but
1620  *   also ensures that we don't deadlock. For instance, if the task is blocking
1621  *   for a memory allocation, memory reclaim could end up wanting to free a
1622  *   page belonging to that request that is currently residing in our private
1623  *   plug. By flushing the pending I/O when the process goes to sleep, we avoid
1624  *   this kind of deadlock.
1625  */
1626 void blk_start_plug(struct blk_plug *plug)
1627 {
1628 	struct task_struct *tsk = current;
1629 
1630 	/*
1631 	 * If this is a nested plug, don't actually assign it.
1632 	 */
1633 	if (tsk->plug)
1634 		return;
1635 
1636 	INIT_LIST_HEAD(&plug->mq_list);
1637 	INIT_LIST_HEAD(&plug->cb_list);
1638 	plug->rq_count = 0;
1639 	plug->multiple_queues = false;
1640 
1641 	/*
1642 	 * Store ordering should not be needed here, since a potential
1643 	 * preempt will imply a full memory barrier
1644 	 */
1645 	tsk->plug = plug;
1646 }
1647 EXPORT_SYMBOL(blk_start_plug);
1648 
1649 static void flush_plug_callbacks(struct blk_plug *plug, bool from_schedule)
1650 {
1651 	LIST_HEAD(callbacks);
1652 
1653 	while (!list_empty(&plug->cb_list)) {
1654 		list_splice_init(&plug->cb_list, &callbacks);
1655 
1656 		while (!list_empty(&callbacks)) {
1657 			struct blk_plug_cb *cb = list_first_entry(&callbacks,
1658 							  struct blk_plug_cb,
1659 							  list);
1660 			list_del(&cb->list);
1661 			cb->callback(cb, from_schedule);
1662 		}
1663 	}
1664 }
1665 
1666 struct blk_plug_cb *blk_check_plugged(blk_plug_cb_fn unplug, void *data,
1667 				      int size)
1668 {
1669 	struct blk_plug *plug = current->plug;
1670 	struct blk_plug_cb *cb;
1671 
1672 	if (!plug)
1673 		return NULL;
1674 
1675 	list_for_each_entry(cb, &plug->cb_list, list)
1676 		if (cb->callback == unplug && cb->data == data)
1677 			return cb;
1678 
1679 	/* Not currently on the callback list */
1680 	BUG_ON(size < sizeof(*cb));
1681 	cb = kzalloc(size, GFP_ATOMIC);
1682 	if (cb) {
1683 		cb->data = data;
1684 		cb->callback = unplug;
1685 		list_add(&cb->list, &plug->cb_list);
1686 	}
1687 	return cb;
1688 }
1689 EXPORT_SYMBOL(blk_check_plugged);
1690 
1691 void blk_flush_plug_list(struct blk_plug *plug, bool from_schedule)
1692 {
1693 	flush_plug_callbacks(plug, from_schedule);
1694 
1695 	if (!list_empty(&plug->mq_list))
1696 		blk_mq_flush_plug_list(plug, from_schedule);
1697 }
1698 
1699 /**
1700  * blk_finish_plug - mark the end of a batch of submitted I/O
1701  * @plug:	The &struct blk_plug passed to blk_start_plug()
1702  *
1703  * Description:
1704  * Indicate that a batch of I/O submissions is complete.  This function
1705  * must be paired with an initial call to blk_start_plug().  The intent
1706  * is to allow the block layer to optimize I/O submission.  See the
1707  * documentation for blk_start_plug() for more information.
1708  */
1709 void blk_finish_plug(struct blk_plug *plug)
1710 {
1711 	if (plug != current->plug)
1712 		return;
1713 	blk_flush_plug_list(plug, false);
1714 
1715 	current->plug = NULL;
1716 }
1717 EXPORT_SYMBOL(blk_finish_plug);
1718 
1719 int __init blk_dev_init(void)
1720 {
1721 	BUILD_BUG_ON(REQ_OP_LAST >= (1 << REQ_OP_BITS));
1722 	BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 *
1723 			FIELD_SIZEOF(struct request, cmd_flags));
1724 	BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 *
1725 			FIELD_SIZEOF(struct bio, bi_opf));
1726 
1727 	/* used for unplugging and affects IO latency/throughput - HIGHPRI */
1728 	kblockd_workqueue = alloc_workqueue("kblockd",
1729 					    WQ_MEM_RECLAIM | WQ_HIGHPRI, 0);
1730 	if (!kblockd_workqueue)
1731 		panic("Failed to create kblockd\n");
1732 
1733 	blk_requestq_cachep = kmem_cache_create("request_queue",
1734 			sizeof(struct request_queue), 0, SLAB_PANIC, NULL);
1735 
1736 #ifdef CONFIG_DEBUG_FS
1737 	blk_debugfs_root = debugfs_create_dir("block", NULL);
1738 #endif
1739 
1740 	return 0;
1741 }
1742