xref: /linux/block/blk-core.c (revision 22d55f02b8922a097cd4be1e2f131dfa7ef65901)
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 	/*
324 	 * In theory, request pool of sched_tags belongs to request queue.
325 	 * However, the current implementation requires tag_set for freeing
326 	 * requests, so free the pool now.
327 	 *
328 	 * Queue has become frozen, there can't be any in-queue requests, so
329 	 * it is safe to free requests now.
330 	 */
331 	mutex_lock(&q->sysfs_lock);
332 	if (q->elevator)
333 		blk_mq_sched_free_requests(q);
334 	mutex_unlock(&q->sysfs_lock);
335 
336 	percpu_ref_exit(&q->q_usage_counter);
337 
338 	/* @q is and will stay empty, shutdown and put */
339 	blk_put_queue(q);
340 }
341 EXPORT_SYMBOL(blk_cleanup_queue);
342 
343 struct request_queue *blk_alloc_queue(gfp_t gfp_mask)
344 {
345 	return blk_alloc_queue_node(gfp_mask, NUMA_NO_NODE);
346 }
347 EXPORT_SYMBOL(blk_alloc_queue);
348 
349 /**
350  * blk_queue_enter() - try to increase q->q_usage_counter
351  * @q: request queue pointer
352  * @flags: BLK_MQ_REQ_NOWAIT and/or BLK_MQ_REQ_PREEMPT
353  */
354 int blk_queue_enter(struct request_queue *q, blk_mq_req_flags_t flags)
355 {
356 	const bool pm = flags & BLK_MQ_REQ_PREEMPT;
357 
358 	while (true) {
359 		bool success = false;
360 
361 		rcu_read_lock();
362 		if (percpu_ref_tryget_live(&q->q_usage_counter)) {
363 			/*
364 			 * The code that increments the pm_only counter is
365 			 * responsible for ensuring that that counter is
366 			 * globally visible before the queue is unfrozen.
367 			 */
368 			if (pm || !blk_queue_pm_only(q)) {
369 				success = true;
370 			} else {
371 				percpu_ref_put(&q->q_usage_counter);
372 			}
373 		}
374 		rcu_read_unlock();
375 
376 		if (success)
377 			return 0;
378 
379 		if (flags & BLK_MQ_REQ_NOWAIT)
380 			return -EBUSY;
381 
382 		/*
383 		 * read pair of barrier in blk_freeze_queue_start(),
384 		 * we need to order reading __PERCPU_REF_DEAD flag of
385 		 * .q_usage_counter and reading .mq_freeze_depth or
386 		 * queue dying flag, otherwise the following wait may
387 		 * never return if the two reads are reordered.
388 		 */
389 		smp_rmb();
390 
391 		wait_event(q->mq_freeze_wq,
392 			   (!q->mq_freeze_depth &&
393 			    (pm || (blk_pm_request_resume(q),
394 				    !blk_queue_pm_only(q)))) ||
395 			   blk_queue_dying(q));
396 		if (blk_queue_dying(q))
397 			return -ENODEV;
398 	}
399 }
400 
401 void blk_queue_exit(struct request_queue *q)
402 {
403 	percpu_ref_put(&q->q_usage_counter);
404 }
405 
406 static void blk_queue_usage_counter_release(struct percpu_ref *ref)
407 {
408 	struct request_queue *q =
409 		container_of(ref, struct request_queue, q_usage_counter);
410 
411 	wake_up_all(&q->mq_freeze_wq);
412 }
413 
414 static void blk_rq_timed_out_timer(struct timer_list *t)
415 {
416 	struct request_queue *q = from_timer(q, t, timeout);
417 
418 	kblockd_schedule_work(&q->timeout_work);
419 }
420 
421 static void blk_timeout_work(struct work_struct *work)
422 {
423 }
424 
425 /**
426  * blk_alloc_queue_node - allocate a request queue
427  * @gfp_mask: memory allocation flags
428  * @node_id: NUMA node to allocate memory from
429  */
430 struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id)
431 {
432 	struct request_queue *q;
433 	int ret;
434 
435 	q = kmem_cache_alloc_node(blk_requestq_cachep,
436 				gfp_mask | __GFP_ZERO, node_id);
437 	if (!q)
438 		return NULL;
439 
440 	INIT_LIST_HEAD(&q->queue_head);
441 	q->last_merge = NULL;
442 
443 	q->id = ida_simple_get(&blk_queue_ida, 0, 0, gfp_mask);
444 	if (q->id < 0)
445 		goto fail_q;
446 
447 	ret = bioset_init(&q->bio_split, BIO_POOL_SIZE, 0, BIOSET_NEED_BVECS);
448 	if (ret)
449 		goto fail_id;
450 
451 	q->backing_dev_info = bdi_alloc_node(gfp_mask, node_id);
452 	if (!q->backing_dev_info)
453 		goto fail_split;
454 
455 	q->stats = blk_alloc_queue_stats();
456 	if (!q->stats)
457 		goto fail_stats;
458 
459 	q->backing_dev_info->ra_pages = VM_READAHEAD_PAGES;
460 	q->backing_dev_info->capabilities = BDI_CAP_CGROUP_WRITEBACK;
461 	q->backing_dev_info->name = "block";
462 	q->node = node_id;
463 
464 	timer_setup(&q->backing_dev_info->laptop_mode_wb_timer,
465 		    laptop_mode_timer_fn, 0);
466 	timer_setup(&q->timeout, blk_rq_timed_out_timer, 0);
467 	INIT_WORK(&q->timeout_work, blk_timeout_work);
468 	INIT_LIST_HEAD(&q->icq_list);
469 #ifdef CONFIG_BLK_CGROUP
470 	INIT_LIST_HEAD(&q->blkg_list);
471 #endif
472 
473 	kobject_init(&q->kobj, &blk_queue_ktype);
474 
475 #ifdef CONFIG_BLK_DEV_IO_TRACE
476 	mutex_init(&q->blk_trace_mutex);
477 #endif
478 	mutex_init(&q->sysfs_lock);
479 	spin_lock_init(&q->queue_lock);
480 
481 	init_waitqueue_head(&q->mq_freeze_wq);
482 	mutex_init(&q->mq_freeze_lock);
483 
484 	/*
485 	 * Init percpu_ref in atomic mode so that it's faster to shutdown.
486 	 * See blk_register_queue() for details.
487 	 */
488 	if (percpu_ref_init(&q->q_usage_counter,
489 				blk_queue_usage_counter_release,
490 				PERCPU_REF_INIT_ATOMIC, GFP_KERNEL))
491 		goto fail_bdi;
492 
493 	if (blkcg_init_queue(q))
494 		goto fail_ref;
495 
496 	return q;
497 
498 fail_ref:
499 	percpu_ref_exit(&q->q_usage_counter);
500 fail_bdi:
501 	blk_free_queue_stats(q->stats);
502 fail_stats:
503 	bdi_put(q->backing_dev_info);
504 fail_split:
505 	bioset_exit(&q->bio_split);
506 fail_id:
507 	ida_simple_remove(&blk_queue_ida, q->id);
508 fail_q:
509 	kmem_cache_free(blk_requestq_cachep, q);
510 	return NULL;
511 }
512 EXPORT_SYMBOL(blk_alloc_queue_node);
513 
514 bool blk_get_queue(struct request_queue *q)
515 {
516 	if (likely(!blk_queue_dying(q))) {
517 		__blk_get_queue(q);
518 		return true;
519 	}
520 
521 	return false;
522 }
523 EXPORT_SYMBOL(blk_get_queue);
524 
525 /**
526  * blk_get_request - allocate a request
527  * @q: request queue to allocate a request for
528  * @op: operation (REQ_OP_*) and REQ_* flags, e.g. REQ_SYNC.
529  * @flags: BLK_MQ_REQ_* flags, e.g. BLK_MQ_REQ_NOWAIT.
530  */
531 struct request *blk_get_request(struct request_queue *q, unsigned int op,
532 				blk_mq_req_flags_t flags)
533 {
534 	struct request *req;
535 
536 	WARN_ON_ONCE(op & REQ_NOWAIT);
537 	WARN_ON_ONCE(flags & ~(BLK_MQ_REQ_NOWAIT | BLK_MQ_REQ_PREEMPT));
538 
539 	req = blk_mq_alloc_request(q, op, flags);
540 	if (!IS_ERR(req) && q->mq_ops->initialize_rq_fn)
541 		q->mq_ops->initialize_rq_fn(req);
542 
543 	return req;
544 }
545 EXPORT_SYMBOL(blk_get_request);
546 
547 void blk_put_request(struct request *req)
548 {
549 	blk_mq_free_request(req);
550 }
551 EXPORT_SYMBOL(blk_put_request);
552 
553 bool bio_attempt_back_merge(struct request_queue *q, struct request *req,
554 			    struct bio *bio)
555 {
556 	const int ff = bio->bi_opf & REQ_FAILFAST_MASK;
557 
558 	if (!ll_back_merge_fn(q, req, bio))
559 		return false;
560 
561 	trace_block_bio_backmerge(q, req, bio);
562 
563 	if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
564 		blk_rq_set_mixed_merge(req);
565 
566 	req->biotail->bi_next = bio;
567 	req->biotail = bio;
568 	req->__data_len += bio->bi_iter.bi_size;
569 
570 	blk_account_io_start(req, false);
571 	return true;
572 }
573 
574 bool bio_attempt_front_merge(struct request_queue *q, struct request *req,
575 			     struct bio *bio)
576 {
577 	const int ff = bio->bi_opf & REQ_FAILFAST_MASK;
578 
579 	if (!ll_front_merge_fn(q, req, bio))
580 		return false;
581 
582 	trace_block_bio_frontmerge(q, req, bio);
583 
584 	if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
585 		blk_rq_set_mixed_merge(req);
586 
587 	bio->bi_next = req->bio;
588 	req->bio = bio;
589 
590 	req->__sector = bio->bi_iter.bi_sector;
591 	req->__data_len += bio->bi_iter.bi_size;
592 
593 	blk_account_io_start(req, false);
594 	return true;
595 }
596 
597 bool bio_attempt_discard_merge(struct request_queue *q, struct request *req,
598 		struct bio *bio)
599 {
600 	unsigned short segments = blk_rq_nr_discard_segments(req);
601 
602 	if (segments >= queue_max_discard_segments(q))
603 		goto no_merge;
604 	if (blk_rq_sectors(req) + bio_sectors(bio) >
605 	    blk_rq_get_max_sectors(req, blk_rq_pos(req)))
606 		goto no_merge;
607 
608 	req->biotail->bi_next = bio;
609 	req->biotail = bio;
610 	req->__data_len += bio->bi_iter.bi_size;
611 	req->nr_phys_segments = segments + 1;
612 
613 	blk_account_io_start(req, false);
614 	return true;
615 no_merge:
616 	req_set_nomerge(q, req);
617 	return false;
618 }
619 
620 /**
621  * blk_attempt_plug_merge - try to merge with %current's plugged list
622  * @q: request_queue new bio is being queued at
623  * @bio: new bio being queued
624  * @same_queue_rq: pointer to &struct request that gets filled in when
625  * another request associated with @q is found on the plug list
626  * (optional, may be %NULL)
627  *
628  * Determine whether @bio being queued on @q can be merged with a request
629  * on %current's plugged list.  Returns %true if merge was successful,
630  * otherwise %false.
631  *
632  * Plugging coalesces IOs from the same issuer for the same purpose without
633  * going through @q->queue_lock.  As such it's more of an issuing mechanism
634  * than scheduling, and the request, while may have elvpriv data, is not
635  * added on the elevator at this point.  In addition, we don't have
636  * reliable access to the elevator outside queue lock.  Only check basic
637  * merging parameters without querying the elevator.
638  *
639  * Caller must ensure !blk_queue_nomerges(q) beforehand.
640  */
641 bool blk_attempt_plug_merge(struct request_queue *q, struct bio *bio,
642 			    struct request **same_queue_rq)
643 {
644 	struct blk_plug *plug;
645 	struct request *rq;
646 	struct list_head *plug_list;
647 
648 	plug = current->plug;
649 	if (!plug)
650 		return false;
651 
652 	plug_list = &plug->mq_list;
653 
654 	list_for_each_entry_reverse(rq, plug_list, queuelist) {
655 		bool merged = false;
656 
657 		if (rq->q == q && same_queue_rq) {
658 			/*
659 			 * Only blk-mq multiple hardware queues case checks the
660 			 * rq in the same queue, there should be only one such
661 			 * rq in a queue
662 			 **/
663 			*same_queue_rq = rq;
664 		}
665 
666 		if (rq->q != q || !blk_rq_merge_ok(rq, bio))
667 			continue;
668 
669 		switch (blk_try_merge(rq, bio)) {
670 		case ELEVATOR_BACK_MERGE:
671 			merged = bio_attempt_back_merge(q, rq, bio);
672 			break;
673 		case ELEVATOR_FRONT_MERGE:
674 			merged = bio_attempt_front_merge(q, rq, bio);
675 			break;
676 		case ELEVATOR_DISCARD_MERGE:
677 			merged = bio_attempt_discard_merge(q, rq, bio);
678 			break;
679 		default:
680 			break;
681 		}
682 
683 		if (merged)
684 			return true;
685 	}
686 
687 	return false;
688 }
689 
690 void blk_init_request_from_bio(struct request *req, struct bio *bio)
691 {
692 	if (bio->bi_opf & REQ_RAHEAD)
693 		req->cmd_flags |= REQ_FAILFAST_MASK;
694 
695 	req->__sector = bio->bi_iter.bi_sector;
696 	req->ioprio = bio_prio(bio);
697 	req->write_hint = bio->bi_write_hint;
698 	blk_rq_bio_prep(req->q, req, bio);
699 }
700 EXPORT_SYMBOL_GPL(blk_init_request_from_bio);
701 
702 static void handle_bad_sector(struct bio *bio, sector_t maxsector)
703 {
704 	char b[BDEVNAME_SIZE];
705 
706 	printk(KERN_INFO "attempt to access beyond end of device\n");
707 	printk(KERN_INFO "%s: rw=%d, want=%Lu, limit=%Lu\n",
708 			bio_devname(bio, b), bio->bi_opf,
709 			(unsigned long long)bio_end_sector(bio),
710 			(long long)maxsector);
711 }
712 
713 #ifdef CONFIG_FAIL_MAKE_REQUEST
714 
715 static DECLARE_FAULT_ATTR(fail_make_request);
716 
717 static int __init setup_fail_make_request(char *str)
718 {
719 	return setup_fault_attr(&fail_make_request, str);
720 }
721 __setup("fail_make_request=", setup_fail_make_request);
722 
723 static bool should_fail_request(struct hd_struct *part, unsigned int bytes)
724 {
725 	return part->make_it_fail && should_fail(&fail_make_request, bytes);
726 }
727 
728 static int __init fail_make_request_debugfs(void)
729 {
730 	struct dentry *dir = fault_create_debugfs_attr("fail_make_request",
731 						NULL, &fail_make_request);
732 
733 	return PTR_ERR_OR_ZERO(dir);
734 }
735 
736 late_initcall(fail_make_request_debugfs);
737 
738 #else /* CONFIG_FAIL_MAKE_REQUEST */
739 
740 static inline bool should_fail_request(struct hd_struct *part,
741 					unsigned int bytes)
742 {
743 	return false;
744 }
745 
746 #endif /* CONFIG_FAIL_MAKE_REQUEST */
747 
748 static inline bool bio_check_ro(struct bio *bio, struct hd_struct *part)
749 {
750 	const int op = bio_op(bio);
751 
752 	if (part->policy && op_is_write(op)) {
753 		char b[BDEVNAME_SIZE];
754 
755 		if (op_is_flush(bio->bi_opf) && !bio_sectors(bio))
756 			return false;
757 
758 		WARN_ONCE(1,
759 		       "generic_make_request: Trying to write "
760 			"to read-only block-device %s (partno %d)\n",
761 			bio_devname(bio, b), part->partno);
762 		/* Older lvm-tools actually trigger this */
763 		return false;
764 	}
765 
766 	return false;
767 }
768 
769 static noinline int should_fail_bio(struct bio *bio)
770 {
771 	if (should_fail_request(&bio->bi_disk->part0, bio->bi_iter.bi_size))
772 		return -EIO;
773 	return 0;
774 }
775 ALLOW_ERROR_INJECTION(should_fail_bio, ERRNO);
776 
777 /*
778  * Check whether this bio extends beyond the end of the device or partition.
779  * This may well happen - the kernel calls bread() without checking the size of
780  * the device, e.g., when mounting a file system.
781  */
782 static inline int bio_check_eod(struct bio *bio, sector_t maxsector)
783 {
784 	unsigned int nr_sectors = bio_sectors(bio);
785 
786 	if (nr_sectors && maxsector &&
787 	    (nr_sectors > maxsector ||
788 	     bio->bi_iter.bi_sector > maxsector - nr_sectors)) {
789 		handle_bad_sector(bio, maxsector);
790 		return -EIO;
791 	}
792 	return 0;
793 }
794 
795 /*
796  * Remap block n of partition p to block n+start(p) of the disk.
797  */
798 static inline int blk_partition_remap(struct bio *bio)
799 {
800 	struct hd_struct *p;
801 	int ret = -EIO;
802 
803 	rcu_read_lock();
804 	p = __disk_get_part(bio->bi_disk, bio->bi_partno);
805 	if (unlikely(!p))
806 		goto out;
807 	if (unlikely(should_fail_request(p, bio->bi_iter.bi_size)))
808 		goto out;
809 	if (unlikely(bio_check_ro(bio, p)))
810 		goto out;
811 
812 	/*
813 	 * Zone reset does not include bi_size so bio_sectors() is always 0.
814 	 * Include a test for the reset op code and perform the remap if needed.
815 	 */
816 	if (bio_sectors(bio) || bio_op(bio) == REQ_OP_ZONE_RESET) {
817 		if (bio_check_eod(bio, part_nr_sects_read(p)))
818 			goto out;
819 		bio->bi_iter.bi_sector += p->start_sect;
820 		trace_block_bio_remap(bio->bi_disk->queue, bio, part_devt(p),
821 				      bio->bi_iter.bi_sector - p->start_sect);
822 	}
823 	bio->bi_partno = 0;
824 	ret = 0;
825 out:
826 	rcu_read_unlock();
827 	return ret;
828 }
829 
830 static noinline_for_stack bool
831 generic_make_request_checks(struct bio *bio)
832 {
833 	struct request_queue *q;
834 	int nr_sectors = bio_sectors(bio);
835 	blk_status_t status = BLK_STS_IOERR;
836 	char b[BDEVNAME_SIZE];
837 
838 	might_sleep();
839 
840 	q = bio->bi_disk->queue;
841 	if (unlikely(!q)) {
842 		printk(KERN_ERR
843 		       "generic_make_request: Trying to access "
844 			"nonexistent block-device %s (%Lu)\n",
845 			bio_devname(bio, b), (long long)bio->bi_iter.bi_sector);
846 		goto end_io;
847 	}
848 
849 	/*
850 	 * For a REQ_NOWAIT based request, return -EOPNOTSUPP
851 	 * if queue is not a request based queue.
852 	 */
853 	if ((bio->bi_opf & REQ_NOWAIT) && !queue_is_mq(q))
854 		goto not_supported;
855 
856 	if (should_fail_bio(bio))
857 		goto end_io;
858 
859 	if (bio->bi_partno) {
860 		if (unlikely(blk_partition_remap(bio)))
861 			goto end_io;
862 	} else {
863 		if (unlikely(bio_check_ro(bio, &bio->bi_disk->part0)))
864 			goto end_io;
865 		if (unlikely(bio_check_eod(bio, get_capacity(bio->bi_disk))))
866 			goto end_io;
867 	}
868 
869 	/*
870 	 * Filter flush bio's early so that make_request based
871 	 * drivers without flush support don't have to worry
872 	 * about them.
873 	 */
874 	if (op_is_flush(bio->bi_opf) &&
875 	    !test_bit(QUEUE_FLAG_WC, &q->queue_flags)) {
876 		bio->bi_opf &= ~(REQ_PREFLUSH | REQ_FUA);
877 		if (!nr_sectors) {
878 			status = BLK_STS_OK;
879 			goto end_io;
880 		}
881 	}
882 
883 	if (!test_bit(QUEUE_FLAG_POLL, &q->queue_flags))
884 		bio->bi_opf &= ~REQ_HIPRI;
885 
886 	switch (bio_op(bio)) {
887 	case REQ_OP_DISCARD:
888 		if (!blk_queue_discard(q))
889 			goto not_supported;
890 		break;
891 	case REQ_OP_SECURE_ERASE:
892 		if (!blk_queue_secure_erase(q))
893 			goto not_supported;
894 		break;
895 	case REQ_OP_WRITE_SAME:
896 		if (!q->limits.max_write_same_sectors)
897 			goto not_supported;
898 		break;
899 	case REQ_OP_ZONE_RESET:
900 		if (!blk_queue_is_zoned(q))
901 			goto not_supported;
902 		break;
903 	case REQ_OP_WRITE_ZEROES:
904 		if (!q->limits.max_write_zeroes_sectors)
905 			goto not_supported;
906 		break;
907 	default:
908 		break;
909 	}
910 
911 	/*
912 	 * Various block parts want %current->io_context and lazy ioc
913 	 * allocation ends up trading a lot of pain for a small amount of
914 	 * memory.  Just allocate it upfront.  This may fail and block
915 	 * layer knows how to live with it.
916 	 */
917 	create_io_context(GFP_ATOMIC, q->node);
918 
919 	if (!blkcg_bio_issue_check(q, bio))
920 		return false;
921 
922 	if (!bio_flagged(bio, BIO_TRACE_COMPLETION)) {
923 		trace_block_bio_queue(q, bio);
924 		/* Now that enqueuing has been traced, we need to trace
925 		 * completion as well.
926 		 */
927 		bio_set_flag(bio, BIO_TRACE_COMPLETION);
928 	}
929 	return true;
930 
931 not_supported:
932 	status = BLK_STS_NOTSUPP;
933 end_io:
934 	bio->bi_status = status;
935 	bio_endio(bio);
936 	return false;
937 }
938 
939 /**
940  * generic_make_request - hand a buffer to its device driver for I/O
941  * @bio:  The bio describing the location in memory and on the device.
942  *
943  * generic_make_request() is used to make I/O requests of block
944  * devices. It is passed a &struct bio, which describes the I/O that needs
945  * to be done.
946  *
947  * generic_make_request() does not return any status.  The
948  * success/failure status of the request, along with notification of
949  * completion, is delivered asynchronously through the bio->bi_end_io
950  * function described (one day) else where.
951  *
952  * The caller of generic_make_request must make sure that bi_io_vec
953  * are set to describe the memory buffer, and that bi_dev and bi_sector are
954  * set to describe the device address, and the
955  * bi_end_io and optionally bi_private are set to describe how
956  * completion notification should be signaled.
957  *
958  * generic_make_request and the drivers it calls may use bi_next if this
959  * bio happens to be merged with someone else, and may resubmit the bio to
960  * a lower device by calling into generic_make_request recursively, which
961  * means the bio should NOT be touched after the call to ->make_request_fn.
962  */
963 blk_qc_t generic_make_request(struct bio *bio)
964 {
965 	/*
966 	 * bio_list_on_stack[0] contains bios submitted by the current
967 	 * make_request_fn.
968 	 * bio_list_on_stack[1] contains bios that were submitted before
969 	 * the current make_request_fn, but that haven't been processed
970 	 * yet.
971 	 */
972 	struct bio_list bio_list_on_stack[2];
973 	blk_qc_t ret = BLK_QC_T_NONE;
974 
975 	if (!generic_make_request_checks(bio))
976 		goto out;
977 
978 	/*
979 	 * We only want one ->make_request_fn to be active at a time, else
980 	 * stack usage with stacked devices could be a problem.  So use
981 	 * current->bio_list to keep a list of requests submited by a
982 	 * make_request_fn function.  current->bio_list is also used as a
983 	 * flag to say if generic_make_request is currently active in this
984 	 * task or not.  If it is NULL, then no make_request is active.  If
985 	 * it is non-NULL, then a make_request is active, and new requests
986 	 * should be added at the tail
987 	 */
988 	if (current->bio_list) {
989 		bio_list_add(&current->bio_list[0], bio);
990 		goto out;
991 	}
992 
993 	/* following loop may be a bit non-obvious, and so deserves some
994 	 * explanation.
995 	 * Before entering the loop, bio->bi_next is NULL (as all callers
996 	 * ensure that) so we have a list with a single bio.
997 	 * We pretend that we have just taken it off a longer list, so
998 	 * we assign bio_list to a pointer to the bio_list_on_stack,
999 	 * thus initialising the bio_list of new bios to be
1000 	 * added.  ->make_request() may indeed add some more bios
1001 	 * through a recursive call to generic_make_request.  If it
1002 	 * did, we find a non-NULL value in bio_list and re-enter the loop
1003 	 * from the top.  In this case we really did just take the bio
1004 	 * of the top of the list (no pretending) and so remove it from
1005 	 * bio_list, and call into ->make_request() again.
1006 	 */
1007 	BUG_ON(bio->bi_next);
1008 	bio_list_init(&bio_list_on_stack[0]);
1009 	current->bio_list = bio_list_on_stack;
1010 	do {
1011 		struct request_queue *q = bio->bi_disk->queue;
1012 		blk_mq_req_flags_t flags = bio->bi_opf & REQ_NOWAIT ?
1013 			BLK_MQ_REQ_NOWAIT : 0;
1014 
1015 		if (likely(blk_queue_enter(q, flags) == 0)) {
1016 			struct bio_list lower, same;
1017 
1018 			/* Create a fresh bio_list for all subordinate requests */
1019 			bio_list_on_stack[1] = bio_list_on_stack[0];
1020 			bio_list_init(&bio_list_on_stack[0]);
1021 			ret = q->make_request_fn(q, bio);
1022 
1023 			blk_queue_exit(q);
1024 
1025 			/* sort new bios into those for a lower level
1026 			 * and those for the same level
1027 			 */
1028 			bio_list_init(&lower);
1029 			bio_list_init(&same);
1030 			while ((bio = bio_list_pop(&bio_list_on_stack[0])) != NULL)
1031 				if (q == bio->bi_disk->queue)
1032 					bio_list_add(&same, bio);
1033 				else
1034 					bio_list_add(&lower, bio);
1035 			/* now assemble so we handle the lowest level first */
1036 			bio_list_merge(&bio_list_on_stack[0], &lower);
1037 			bio_list_merge(&bio_list_on_stack[0], &same);
1038 			bio_list_merge(&bio_list_on_stack[0], &bio_list_on_stack[1]);
1039 		} else {
1040 			if (unlikely(!blk_queue_dying(q) &&
1041 					(bio->bi_opf & REQ_NOWAIT)))
1042 				bio_wouldblock_error(bio);
1043 			else
1044 				bio_io_error(bio);
1045 		}
1046 		bio = bio_list_pop(&bio_list_on_stack[0]);
1047 	} while (bio);
1048 	current->bio_list = NULL; /* deactivate */
1049 
1050 out:
1051 	return ret;
1052 }
1053 EXPORT_SYMBOL(generic_make_request);
1054 
1055 /**
1056  * direct_make_request - hand a buffer directly to its device driver for I/O
1057  * @bio:  The bio describing the location in memory and on the device.
1058  *
1059  * This function behaves like generic_make_request(), but does not protect
1060  * against recursion.  Must only be used if the called driver is known
1061  * to not call generic_make_request (or direct_make_request) again from
1062  * its make_request function.  (Calling direct_make_request again from
1063  * a workqueue is perfectly fine as that doesn't recurse).
1064  */
1065 blk_qc_t direct_make_request(struct bio *bio)
1066 {
1067 	struct request_queue *q = bio->bi_disk->queue;
1068 	bool nowait = bio->bi_opf & REQ_NOWAIT;
1069 	blk_qc_t ret;
1070 
1071 	if (!generic_make_request_checks(bio))
1072 		return BLK_QC_T_NONE;
1073 
1074 	if (unlikely(blk_queue_enter(q, nowait ? BLK_MQ_REQ_NOWAIT : 0))) {
1075 		if (nowait && !blk_queue_dying(q))
1076 			bio->bi_status = BLK_STS_AGAIN;
1077 		else
1078 			bio->bi_status = BLK_STS_IOERR;
1079 		bio_endio(bio);
1080 		return BLK_QC_T_NONE;
1081 	}
1082 
1083 	ret = q->make_request_fn(q, bio);
1084 	blk_queue_exit(q);
1085 	return ret;
1086 }
1087 EXPORT_SYMBOL_GPL(direct_make_request);
1088 
1089 /**
1090  * submit_bio - submit a bio to the block device layer for I/O
1091  * @bio: The &struct bio which describes the I/O
1092  *
1093  * submit_bio() is very similar in purpose to generic_make_request(), and
1094  * uses that function to do most of the work. Both are fairly rough
1095  * interfaces; @bio must be presetup and ready for I/O.
1096  *
1097  */
1098 blk_qc_t submit_bio(struct bio *bio)
1099 {
1100 	/*
1101 	 * If it's a regular read/write or a barrier with data attached,
1102 	 * go through the normal accounting stuff before submission.
1103 	 */
1104 	if (bio_has_data(bio)) {
1105 		unsigned int count;
1106 
1107 		if (unlikely(bio_op(bio) == REQ_OP_WRITE_SAME))
1108 			count = queue_logical_block_size(bio->bi_disk->queue) >> 9;
1109 		else
1110 			count = bio_sectors(bio);
1111 
1112 		if (op_is_write(bio_op(bio))) {
1113 			count_vm_events(PGPGOUT, count);
1114 		} else {
1115 			task_io_account_read(bio->bi_iter.bi_size);
1116 			count_vm_events(PGPGIN, count);
1117 		}
1118 
1119 		if (unlikely(block_dump)) {
1120 			char b[BDEVNAME_SIZE];
1121 			printk(KERN_DEBUG "%s(%d): %s block %Lu on %s (%u sectors)\n",
1122 			current->comm, task_pid_nr(current),
1123 				op_is_write(bio_op(bio)) ? "WRITE" : "READ",
1124 				(unsigned long long)bio->bi_iter.bi_sector,
1125 				bio_devname(bio, b), count);
1126 		}
1127 	}
1128 
1129 	return generic_make_request(bio);
1130 }
1131 EXPORT_SYMBOL(submit_bio);
1132 
1133 /**
1134  * blk_cloned_rq_check_limits - Helper function to check a cloned request
1135  *                              for new the queue limits
1136  * @q:  the queue
1137  * @rq: the request being checked
1138  *
1139  * Description:
1140  *    @rq may have been made based on weaker limitations of upper-level queues
1141  *    in request stacking drivers, and it may violate the limitation of @q.
1142  *    Since the block layer and the underlying device driver trust @rq
1143  *    after it is inserted to @q, it should be checked against @q before
1144  *    the insertion using this generic function.
1145  *
1146  *    Request stacking drivers like request-based dm may change the queue
1147  *    limits when retrying requests on other queues. Those requests need
1148  *    to be checked against the new queue limits again during dispatch.
1149  */
1150 static int blk_cloned_rq_check_limits(struct request_queue *q,
1151 				      struct request *rq)
1152 {
1153 	if (blk_rq_sectors(rq) > blk_queue_get_max_sectors(q, req_op(rq))) {
1154 		printk(KERN_ERR "%s: over max size limit. (%u > %u)\n",
1155 			__func__, blk_rq_sectors(rq),
1156 			blk_queue_get_max_sectors(q, req_op(rq)));
1157 		return -EIO;
1158 	}
1159 
1160 	/*
1161 	 * queue's settings related to segment counting like q->bounce_pfn
1162 	 * may differ from that of other stacking queues.
1163 	 * Recalculate it to check the request correctly on this queue's
1164 	 * limitation.
1165 	 */
1166 	blk_recalc_rq_segments(rq);
1167 	if (rq->nr_phys_segments > queue_max_segments(q)) {
1168 		printk(KERN_ERR "%s: over max segments limit. (%hu > %hu)\n",
1169 			__func__, rq->nr_phys_segments, queue_max_segments(q));
1170 		return -EIO;
1171 	}
1172 
1173 	return 0;
1174 }
1175 
1176 /**
1177  * blk_insert_cloned_request - Helper for stacking drivers to submit a request
1178  * @q:  the queue to submit the request
1179  * @rq: the request being queued
1180  */
1181 blk_status_t blk_insert_cloned_request(struct request_queue *q, struct request *rq)
1182 {
1183 	if (blk_cloned_rq_check_limits(q, rq))
1184 		return BLK_STS_IOERR;
1185 
1186 	if (rq->rq_disk &&
1187 	    should_fail_request(&rq->rq_disk->part0, blk_rq_bytes(rq)))
1188 		return BLK_STS_IOERR;
1189 
1190 	if (blk_queue_io_stat(q))
1191 		blk_account_io_start(rq, true);
1192 
1193 	/*
1194 	 * Since we have a scheduler attached on the top device,
1195 	 * bypass a potential scheduler on the bottom device for
1196 	 * insert.
1197 	 */
1198 	return blk_mq_request_issue_directly(rq, true);
1199 }
1200 EXPORT_SYMBOL_GPL(blk_insert_cloned_request);
1201 
1202 /**
1203  * blk_rq_err_bytes - determine number of bytes till the next failure boundary
1204  * @rq: request to examine
1205  *
1206  * Description:
1207  *     A request could be merge of IOs which require different failure
1208  *     handling.  This function determines the number of bytes which
1209  *     can be failed from the beginning of the request without
1210  *     crossing into area which need to be retried further.
1211  *
1212  * Return:
1213  *     The number of bytes to fail.
1214  */
1215 unsigned int blk_rq_err_bytes(const struct request *rq)
1216 {
1217 	unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK;
1218 	unsigned int bytes = 0;
1219 	struct bio *bio;
1220 
1221 	if (!(rq->rq_flags & RQF_MIXED_MERGE))
1222 		return blk_rq_bytes(rq);
1223 
1224 	/*
1225 	 * Currently the only 'mixing' which can happen is between
1226 	 * different fastfail types.  We can safely fail portions
1227 	 * which have all the failfast bits that the first one has -
1228 	 * the ones which are at least as eager to fail as the first
1229 	 * one.
1230 	 */
1231 	for (bio = rq->bio; bio; bio = bio->bi_next) {
1232 		if ((bio->bi_opf & ff) != ff)
1233 			break;
1234 		bytes += bio->bi_iter.bi_size;
1235 	}
1236 
1237 	/* this could lead to infinite loop */
1238 	BUG_ON(blk_rq_bytes(rq) && !bytes);
1239 	return bytes;
1240 }
1241 EXPORT_SYMBOL_GPL(blk_rq_err_bytes);
1242 
1243 void blk_account_io_completion(struct request *req, unsigned int bytes)
1244 {
1245 	if (blk_do_io_stat(req)) {
1246 		const int sgrp = op_stat_group(req_op(req));
1247 		struct hd_struct *part;
1248 
1249 		part_stat_lock();
1250 		part = req->part;
1251 		part_stat_add(part, sectors[sgrp], bytes >> 9);
1252 		part_stat_unlock();
1253 	}
1254 }
1255 
1256 void blk_account_io_done(struct request *req, u64 now)
1257 {
1258 	/*
1259 	 * Account IO completion.  flush_rq isn't accounted as a
1260 	 * normal IO on queueing nor completion.  Accounting the
1261 	 * containing request is enough.
1262 	 */
1263 	if (blk_do_io_stat(req) && !(req->rq_flags & RQF_FLUSH_SEQ)) {
1264 		const int sgrp = op_stat_group(req_op(req));
1265 		struct hd_struct *part;
1266 
1267 		part_stat_lock();
1268 		part = req->part;
1269 
1270 		update_io_ticks(part, jiffies);
1271 		part_stat_inc(part, ios[sgrp]);
1272 		part_stat_add(part, nsecs[sgrp], now - req->start_time_ns);
1273 		part_stat_add(part, time_in_queue, nsecs_to_jiffies64(now - req->start_time_ns));
1274 		part_dec_in_flight(req->q, part, rq_data_dir(req));
1275 
1276 		hd_struct_put(part);
1277 		part_stat_unlock();
1278 	}
1279 }
1280 
1281 void blk_account_io_start(struct request *rq, bool new_io)
1282 {
1283 	struct hd_struct *part;
1284 	int rw = rq_data_dir(rq);
1285 
1286 	if (!blk_do_io_stat(rq))
1287 		return;
1288 
1289 	part_stat_lock();
1290 
1291 	if (!new_io) {
1292 		part = rq->part;
1293 		part_stat_inc(part, merges[rw]);
1294 	} else {
1295 		part = disk_map_sector_rcu(rq->rq_disk, blk_rq_pos(rq));
1296 		if (!hd_struct_try_get(part)) {
1297 			/*
1298 			 * The partition is already being removed,
1299 			 * the request will be accounted on the disk only
1300 			 *
1301 			 * We take a reference on disk->part0 although that
1302 			 * partition will never be deleted, so we can treat
1303 			 * it as any other partition.
1304 			 */
1305 			part = &rq->rq_disk->part0;
1306 			hd_struct_get(part);
1307 		}
1308 		part_inc_in_flight(rq->q, part, rw);
1309 		rq->part = part;
1310 	}
1311 
1312 	update_io_ticks(part, jiffies);
1313 
1314 	part_stat_unlock();
1315 }
1316 
1317 /*
1318  * Steal bios from a request and add them to a bio list.
1319  * The request must not have been partially completed before.
1320  */
1321 void blk_steal_bios(struct bio_list *list, struct request *rq)
1322 {
1323 	if (rq->bio) {
1324 		if (list->tail)
1325 			list->tail->bi_next = rq->bio;
1326 		else
1327 			list->head = rq->bio;
1328 		list->tail = rq->biotail;
1329 
1330 		rq->bio = NULL;
1331 		rq->biotail = NULL;
1332 	}
1333 
1334 	rq->__data_len = 0;
1335 }
1336 EXPORT_SYMBOL_GPL(blk_steal_bios);
1337 
1338 /**
1339  * blk_update_request - Special helper function for request stacking drivers
1340  * @req:      the request being processed
1341  * @error:    block status code
1342  * @nr_bytes: number of bytes to complete @req
1343  *
1344  * Description:
1345  *     Ends I/O on a number of bytes attached to @req, but doesn't complete
1346  *     the request structure even if @req doesn't have leftover.
1347  *     If @req has leftover, sets it up for the next range of segments.
1348  *
1349  *     This special helper function is only for request stacking drivers
1350  *     (e.g. request-based dm) so that they can handle partial completion.
1351  *     Actual device drivers should use blk_end_request instead.
1352  *
1353  *     Passing the result of blk_rq_bytes() as @nr_bytes guarantees
1354  *     %false return from this function.
1355  *
1356  * Note:
1357  *	The RQF_SPECIAL_PAYLOAD flag is ignored on purpose in both
1358  *	blk_rq_bytes() and in blk_update_request().
1359  *
1360  * Return:
1361  *     %false - this request doesn't have any more data
1362  *     %true  - this request has more data
1363  **/
1364 bool blk_update_request(struct request *req, blk_status_t error,
1365 		unsigned int nr_bytes)
1366 {
1367 	int total_bytes;
1368 
1369 	trace_block_rq_complete(req, blk_status_to_errno(error), nr_bytes);
1370 
1371 	if (!req->bio)
1372 		return false;
1373 
1374 	if (unlikely(error && !blk_rq_is_passthrough(req) &&
1375 		     !(req->rq_flags & RQF_QUIET)))
1376 		print_req_error(req, error);
1377 
1378 	blk_account_io_completion(req, nr_bytes);
1379 
1380 	total_bytes = 0;
1381 	while (req->bio) {
1382 		struct bio *bio = req->bio;
1383 		unsigned bio_bytes = min(bio->bi_iter.bi_size, nr_bytes);
1384 
1385 		if (bio_bytes == bio->bi_iter.bi_size)
1386 			req->bio = bio->bi_next;
1387 
1388 		/* Completion has already been traced */
1389 		bio_clear_flag(bio, BIO_TRACE_COMPLETION);
1390 		req_bio_endio(req, bio, bio_bytes, error);
1391 
1392 		total_bytes += bio_bytes;
1393 		nr_bytes -= bio_bytes;
1394 
1395 		if (!nr_bytes)
1396 			break;
1397 	}
1398 
1399 	/*
1400 	 * completely done
1401 	 */
1402 	if (!req->bio) {
1403 		/*
1404 		 * Reset counters so that the request stacking driver
1405 		 * can find how many bytes remain in the request
1406 		 * later.
1407 		 */
1408 		req->__data_len = 0;
1409 		return false;
1410 	}
1411 
1412 	req->__data_len -= total_bytes;
1413 
1414 	/* update sector only for requests with clear definition of sector */
1415 	if (!blk_rq_is_passthrough(req))
1416 		req->__sector += total_bytes >> 9;
1417 
1418 	/* mixed attributes always follow the first bio */
1419 	if (req->rq_flags & RQF_MIXED_MERGE) {
1420 		req->cmd_flags &= ~REQ_FAILFAST_MASK;
1421 		req->cmd_flags |= req->bio->bi_opf & REQ_FAILFAST_MASK;
1422 	}
1423 
1424 	if (!(req->rq_flags & RQF_SPECIAL_PAYLOAD)) {
1425 		/*
1426 		 * If total number of sectors is less than the first segment
1427 		 * size, something has gone terribly wrong.
1428 		 */
1429 		if (blk_rq_bytes(req) < blk_rq_cur_bytes(req)) {
1430 			blk_dump_rq_flags(req, "request botched");
1431 			req->__data_len = blk_rq_cur_bytes(req);
1432 		}
1433 
1434 		/* recalculate the number of segments */
1435 		blk_recalc_rq_segments(req);
1436 	}
1437 
1438 	return true;
1439 }
1440 EXPORT_SYMBOL_GPL(blk_update_request);
1441 
1442 void blk_rq_bio_prep(struct request_queue *q, struct request *rq,
1443 		     struct bio *bio)
1444 {
1445 	if (bio_has_data(bio))
1446 		rq->nr_phys_segments = bio_phys_segments(q, bio);
1447 	else if (bio_op(bio) == REQ_OP_DISCARD)
1448 		rq->nr_phys_segments = 1;
1449 
1450 	rq->__data_len = bio->bi_iter.bi_size;
1451 	rq->bio = rq->biotail = bio;
1452 
1453 	if (bio->bi_disk)
1454 		rq->rq_disk = bio->bi_disk;
1455 }
1456 
1457 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
1458 /**
1459  * rq_flush_dcache_pages - Helper function to flush all pages in a request
1460  * @rq: the request to be flushed
1461  *
1462  * Description:
1463  *     Flush all pages in @rq.
1464  */
1465 void rq_flush_dcache_pages(struct request *rq)
1466 {
1467 	struct req_iterator iter;
1468 	struct bio_vec bvec;
1469 
1470 	rq_for_each_segment(bvec, rq, iter)
1471 		flush_dcache_page(bvec.bv_page);
1472 }
1473 EXPORT_SYMBOL_GPL(rq_flush_dcache_pages);
1474 #endif
1475 
1476 /**
1477  * blk_lld_busy - Check if underlying low-level drivers of a device are busy
1478  * @q : the queue of the device being checked
1479  *
1480  * Description:
1481  *    Check if underlying low-level drivers of a device are busy.
1482  *    If the drivers want to export their busy state, they must set own
1483  *    exporting function using blk_queue_lld_busy() first.
1484  *
1485  *    Basically, this function is used only by request stacking drivers
1486  *    to stop dispatching requests to underlying devices when underlying
1487  *    devices are busy.  This behavior helps more I/O merging on the queue
1488  *    of the request stacking driver and prevents I/O throughput regression
1489  *    on burst I/O load.
1490  *
1491  * Return:
1492  *    0 - Not busy (The request stacking driver should dispatch request)
1493  *    1 - Busy (The request stacking driver should stop dispatching request)
1494  */
1495 int blk_lld_busy(struct request_queue *q)
1496 {
1497 	if (queue_is_mq(q) && q->mq_ops->busy)
1498 		return q->mq_ops->busy(q);
1499 
1500 	return 0;
1501 }
1502 EXPORT_SYMBOL_GPL(blk_lld_busy);
1503 
1504 /**
1505  * blk_rq_unprep_clone - Helper function to free all bios in a cloned request
1506  * @rq: the clone request to be cleaned up
1507  *
1508  * Description:
1509  *     Free all bios in @rq for a cloned request.
1510  */
1511 void blk_rq_unprep_clone(struct request *rq)
1512 {
1513 	struct bio *bio;
1514 
1515 	while ((bio = rq->bio) != NULL) {
1516 		rq->bio = bio->bi_next;
1517 
1518 		bio_put(bio);
1519 	}
1520 }
1521 EXPORT_SYMBOL_GPL(blk_rq_unprep_clone);
1522 
1523 /*
1524  * Copy attributes of the original request to the clone request.
1525  * The actual data parts (e.g. ->cmd, ->sense) are not copied.
1526  */
1527 static void __blk_rq_prep_clone(struct request *dst, struct request *src)
1528 {
1529 	dst->__sector = blk_rq_pos(src);
1530 	dst->__data_len = blk_rq_bytes(src);
1531 	if (src->rq_flags & RQF_SPECIAL_PAYLOAD) {
1532 		dst->rq_flags |= RQF_SPECIAL_PAYLOAD;
1533 		dst->special_vec = src->special_vec;
1534 	}
1535 	dst->nr_phys_segments = src->nr_phys_segments;
1536 	dst->ioprio = src->ioprio;
1537 	dst->extra_len = src->extra_len;
1538 }
1539 
1540 /**
1541  * blk_rq_prep_clone - Helper function to setup clone request
1542  * @rq: the request to be setup
1543  * @rq_src: original request to be cloned
1544  * @bs: bio_set that bios for clone are allocated from
1545  * @gfp_mask: memory allocation mask for bio
1546  * @bio_ctr: setup function to be called for each clone bio.
1547  *           Returns %0 for success, non %0 for failure.
1548  * @data: private data to be passed to @bio_ctr
1549  *
1550  * Description:
1551  *     Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq.
1552  *     The actual data parts of @rq_src (e.g. ->cmd, ->sense)
1553  *     are not copied, and copying such parts is the caller's responsibility.
1554  *     Also, pages which the original bios are pointing to are not copied
1555  *     and the cloned bios just point same pages.
1556  *     So cloned bios must be completed before original bios, which means
1557  *     the caller must complete @rq before @rq_src.
1558  */
1559 int blk_rq_prep_clone(struct request *rq, struct request *rq_src,
1560 		      struct bio_set *bs, gfp_t gfp_mask,
1561 		      int (*bio_ctr)(struct bio *, struct bio *, void *),
1562 		      void *data)
1563 {
1564 	struct bio *bio, *bio_src;
1565 
1566 	if (!bs)
1567 		bs = &fs_bio_set;
1568 
1569 	__rq_for_each_bio(bio_src, rq_src) {
1570 		bio = bio_clone_fast(bio_src, gfp_mask, bs);
1571 		if (!bio)
1572 			goto free_and_out;
1573 
1574 		if (bio_ctr && bio_ctr(bio, bio_src, data))
1575 			goto free_and_out;
1576 
1577 		if (rq->bio) {
1578 			rq->biotail->bi_next = bio;
1579 			rq->biotail = bio;
1580 		} else
1581 			rq->bio = rq->biotail = bio;
1582 	}
1583 
1584 	__blk_rq_prep_clone(rq, rq_src);
1585 
1586 	return 0;
1587 
1588 free_and_out:
1589 	if (bio)
1590 		bio_put(bio);
1591 	blk_rq_unprep_clone(rq);
1592 
1593 	return -ENOMEM;
1594 }
1595 EXPORT_SYMBOL_GPL(blk_rq_prep_clone);
1596 
1597 int kblockd_schedule_work(struct work_struct *work)
1598 {
1599 	return queue_work(kblockd_workqueue, work);
1600 }
1601 EXPORT_SYMBOL(kblockd_schedule_work);
1602 
1603 int kblockd_schedule_work_on(int cpu, struct work_struct *work)
1604 {
1605 	return queue_work_on(cpu, kblockd_workqueue, work);
1606 }
1607 EXPORT_SYMBOL(kblockd_schedule_work_on);
1608 
1609 int kblockd_mod_delayed_work_on(int cpu, struct delayed_work *dwork,
1610 				unsigned long delay)
1611 {
1612 	return mod_delayed_work_on(cpu, kblockd_workqueue, dwork, delay);
1613 }
1614 EXPORT_SYMBOL(kblockd_mod_delayed_work_on);
1615 
1616 /**
1617  * blk_start_plug - initialize blk_plug and track it inside the task_struct
1618  * @plug:	The &struct blk_plug that needs to be initialized
1619  *
1620  * Description:
1621  *   blk_start_plug() indicates to the block layer an intent by the caller
1622  *   to submit multiple I/O requests in a batch.  The block layer may use
1623  *   this hint to defer submitting I/Os from the caller until blk_finish_plug()
1624  *   is called.  However, the block layer may choose to submit requests
1625  *   before a call to blk_finish_plug() if the number of queued I/Os
1626  *   exceeds %BLK_MAX_REQUEST_COUNT, or if the size of the I/O is larger than
1627  *   %BLK_PLUG_FLUSH_SIZE.  The queued I/Os may also be submitted early if
1628  *   the task schedules (see below).
1629  *
1630  *   Tracking blk_plug inside the task_struct will help with auto-flushing the
1631  *   pending I/O should the task end up blocking between blk_start_plug() and
1632  *   blk_finish_plug(). This is important from a performance perspective, but
1633  *   also ensures that we don't deadlock. For instance, if the task is blocking
1634  *   for a memory allocation, memory reclaim could end up wanting to free a
1635  *   page belonging to that request that is currently residing in our private
1636  *   plug. By flushing the pending I/O when the process goes to sleep, we avoid
1637  *   this kind of deadlock.
1638  */
1639 void blk_start_plug(struct blk_plug *plug)
1640 {
1641 	struct task_struct *tsk = current;
1642 
1643 	/*
1644 	 * If this is a nested plug, don't actually assign it.
1645 	 */
1646 	if (tsk->plug)
1647 		return;
1648 
1649 	INIT_LIST_HEAD(&plug->mq_list);
1650 	INIT_LIST_HEAD(&plug->cb_list);
1651 	plug->rq_count = 0;
1652 	plug->multiple_queues = false;
1653 
1654 	/*
1655 	 * Store ordering should not be needed here, since a potential
1656 	 * preempt will imply a full memory barrier
1657 	 */
1658 	tsk->plug = plug;
1659 }
1660 EXPORT_SYMBOL(blk_start_plug);
1661 
1662 static void flush_plug_callbacks(struct blk_plug *plug, bool from_schedule)
1663 {
1664 	LIST_HEAD(callbacks);
1665 
1666 	while (!list_empty(&plug->cb_list)) {
1667 		list_splice_init(&plug->cb_list, &callbacks);
1668 
1669 		while (!list_empty(&callbacks)) {
1670 			struct blk_plug_cb *cb = list_first_entry(&callbacks,
1671 							  struct blk_plug_cb,
1672 							  list);
1673 			list_del(&cb->list);
1674 			cb->callback(cb, from_schedule);
1675 		}
1676 	}
1677 }
1678 
1679 struct blk_plug_cb *blk_check_plugged(blk_plug_cb_fn unplug, void *data,
1680 				      int size)
1681 {
1682 	struct blk_plug *plug = current->plug;
1683 	struct blk_plug_cb *cb;
1684 
1685 	if (!plug)
1686 		return NULL;
1687 
1688 	list_for_each_entry(cb, &plug->cb_list, list)
1689 		if (cb->callback == unplug && cb->data == data)
1690 			return cb;
1691 
1692 	/* Not currently on the callback list */
1693 	BUG_ON(size < sizeof(*cb));
1694 	cb = kzalloc(size, GFP_ATOMIC);
1695 	if (cb) {
1696 		cb->data = data;
1697 		cb->callback = unplug;
1698 		list_add(&cb->list, &plug->cb_list);
1699 	}
1700 	return cb;
1701 }
1702 EXPORT_SYMBOL(blk_check_plugged);
1703 
1704 void blk_flush_plug_list(struct blk_plug *plug, bool from_schedule)
1705 {
1706 	flush_plug_callbacks(plug, from_schedule);
1707 
1708 	if (!list_empty(&plug->mq_list))
1709 		blk_mq_flush_plug_list(plug, from_schedule);
1710 }
1711 
1712 /**
1713  * blk_finish_plug - mark the end of a batch of submitted I/O
1714  * @plug:	The &struct blk_plug passed to blk_start_plug()
1715  *
1716  * Description:
1717  * Indicate that a batch of I/O submissions is complete.  This function
1718  * must be paired with an initial call to blk_start_plug().  The intent
1719  * is to allow the block layer to optimize I/O submission.  See the
1720  * documentation for blk_start_plug() for more information.
1721  */
1722 void blk_finish_plug(struct blk_plug *plug)
1723 {
1724 	if (plug != current->plug)
1725 		return;
1726 	blk_flush_plug_list(plug, false);
1727 
1728 	current->plug = NULL;
1729 }
1730 EXPORT_SYMBOL(blk_finish_plug);
1731 
1732 int __init blk_dev_init(void)
1733 {
1734 	BUILD_BUG_ON(REQ_OP_LAST >= (1 << REQ_OP_BITS));
1735 	BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 *
1736 			FIELD_SIZEOF(struct request, cmd_flags));
1737 	BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 *
1738 			FIELD_SIZEOF(struct bio, bi_opf));
1739 
1740 	/* used for unplugging and affects IO latency/throughput - HIGHPRI */
1741 	kblockd_workqueue = alloc_workqueue("kblockd",
1742 					    WQ_MEM_RECLAIM | WQ_HIGHPRI, 0);
1743 	if (!kblockd_workqueue)
1744 		panic("Failed to create kblockd\n");
1745 
1746 	blk_requestq_cachep = kmem_cache_create("request_queue",
1747 			sizeof(struct request_queue), 0, SLAB_PANIC, NULL);
1748 
1749 #ifdef CONFIG_DEBUG_FS
1750 	blk_debugfs_root = debugfs_create_dir("block", NULL);
1751 #endif
1752 
1753 	return 0;
1754 }
1755