xref: /linux/block/blk-core.c (revision ed00aabd5eb9fb44d6aff1173234a2e911b9fead)
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/pagemap.h>
24 #include <linux/kernel_stat.h>
25 #include <linux/string.h>
26 #include <linux/init.h>
27 #include <linux/completion.h>
28 #include <linux/slab.h>
29 #include <linux/swap.h>
30 #include <linux/writeback.h>
31 #include <linux/task_io_accounting_ops.h>
32 #include <linux/fault-inject.h>
33 #include <linux/list_sort.h>
34 #include <linux/delay.h>
35 #include <linux/ratelimit.h>
36 #include <linux/pm_runtime.h>
37 #include <linux/blk-cgroup.h>
38 #include <linux/t10-pi.h>
39 #include <linux/debugfs.h>
40 #include <linux/bpf.h>
41 #include <linux/psi.h>
42 #include <linux/sched/sysctl.h>
43 #include <linux/blk-crypto.h>
44 
45 #define CREATE_TRACE_POINTS
46 #include <trace/events/block.h>
47 
48 #include "blk.h"
49 #include "blk-mq.h"
50 #include "blk-mq-sched.h"
51 #include "blk-pm.h"
52 #include "blk-rq-qos.h"
53 
54 struct dentry *blk_debugfs_root;
55 
56 EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_remap);
57 EXPORT_TRACEPOINT_SYMBOL_GPL(block_rq_remap);
58 EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_complete);
59 EXPORT_TRACEPOINT_SYMBOL_GPL(block_split);
60 EXPORT_TRACEPOINT_SYMBOL_GPL(block_unplug);
61 
62 DEFINE_IDA(blk_queue_ida);
63 
64 /*
65  * For queue allocation
66  */
67 struct kmem_cache *blk_requestq_cachep;
68 
69 /*
70  * Controlling structure to kblockd
71  */
72 static struct workqueue_struct *kblockd_workqueue;
73 
74 /**
75  * blk_queue_flag_set - atomically set a queue flag
76  * @flag: flag to be set
77  * @q: request queue
78  */
79 void blk_queue_flag_set(unsigned int flag, struct request_queue *q)
80 {
81 	set_bit(flag, &q->queue_flags);
82 }
83 EXPORT_SYMBOL(blk_queue_flag_set);
84 
85 /**
86  * blk_queue_flag_clear - atomically clear a queue flag
87  * @flag: flag to be cleared
88  * @q: request queue
89  */
90 void blk_queue_flag_clear(unsigned int flag, struct request_queue *q)
91 {
92 	clear_bit(flag, &q->queue_flags);
93 }
94 EXPORT_SYMBOL(blk_queue_flag_clear);
95 
96 /**
97  * blk_queue_flag_test_and_set - atomically test and set a queue flag
98  * @flag: flag to be set
99  * @q: request queue
100  *
101  * Returns the previous value of @flag - 0 if the flag was not set and 1 if
102  * the flag was already set.
103  */
104 bool blk_queue_flag_test_and_set(unsigned int flag, struct request_queue *q)
105 {
106 	return test_and_set_bit(flag, &q->queue_flags);
107 }
108 EXPORT_SYMBOL_GPL(blk_queue_flag_test_and_set);
109 
110 void blk_rq_init(struct request_queue *q, struct request *rq)
111 {
112 	memset(rq, 0, sizeof(*rq));
113 
114 	INIT_LIST_HEAD(&rq->queuelist);
115 	rq->q = q;
116 	rq->__sector = (sector_t) -1;
117 	INIT_HLIST_NODE(&rq->hash);
118 	RB_CLEAR_NODE(&rq->rb_node);
119 	rq->tag = -1;
120 	rq->internal_tag = -1;
121 	rq->start_time_ns = ktime_get_ns();
122 	rq->part = NULL;
123 	refcount_set(&rq->ref, 1);
124 	blk_crypto_rq_set_defaults(rq);
125 }
126 EXPORT_SYMBOL(blk_rq_init);
127 
128 #define REQ_OP_NAME(name) [REQ_OP_##name] = #name
129 static const char *const blk_op_name[] = {
130 	REQ_OP_NAME(READ),
131 	REQ_OP_NAME(WRITE),
132 	REQ_OP_NAME(FLUSH),
133 	REQ_OP_NAME(DISCARD),
134 	REQ_OP_NAME(SECURE_ERASE),
135 	REQ_OP_NAME(ZONE_RESET),
136 	REQ_OP_NAME(ZONE_RESET_ALL),
137 	REQ_OP_NAME(ZONE_OPEN),
138 	REQ_OP_NAME(ZONE_CLOSE),
139 	REQ_OP_NAME(ZONE_FINISH),
140 	REQ_OP_NAME(ZONE_APPEND),
141 	REQ_OP_NAME(WRITE_SAME),
142 	REQ_OP_NAME(WRITE_ZEROES),
143 	REQ_OP_NAME(SCSI_IN),
144 	REQ_OP_NAME(SCSI_OUT),
145 	REQ_OP_NAME(DRV_IN),
146 	REQ_OP_NAME(DRV_OUT),
147 };
148 #undef REQ_OP_NAME
149 
150 /**
151  * blk_op_str - Return string XXX in the REQ_OP_XXX.
152  * @op: REQ_OP_XXX.
153  *
154  * Description: Centralize block layer function to convert REQ_OP_XXX into
155  * string format. Useful in the debugging and tracing bio or request. For
156  * invalid REQ_OP_XXX it returns string "UNKNOWN".
157  */
158 inline const char *blk_op_str(unsigned int op)
159 {
160 	const char *op_str = "UNKNOWN";
161 
162 	if (op < ARRAY_SIZE(blk_op_name) && blk_op_name[op])
163 		op_str = blk_op_name[op];
164 
165 	return op_str;
166 }
167 EXPORT_SYMBOL_GPL(blk_op_str);
168 
169 static const struct {
170 	int		errno;
171 	const char	*name;
172 } blk_errors[] = {
173 	[BLK_STS_OK]		= { 0,		"" },
174 	[BLK_STS_NOTSUPP]	= { -EOPNOTSUPP, "operation not supported" },
175 	[BLK_STS_TIMEOUT]	= { -ETIMEDOUT,	"timeout" },
176 	[BLK_STS_NOSPC]		= { -ENOSPC,	"critical space allocation" },
177 	[BLK_STS_TRANSPORT]	= { -ENOLINK,	"recoverable transport" },
178 	[BLK_STS_TARGET]	= { -EREMOTEIO,	"critical target" },
179 	[BLK_STS_NEXUS]		= { -EBADE,	"critical nexus" },
180 	[BLK_STS_MEDIUM]	= { -ENODATA,	"critical medium" },
181 	[BLK_STS_PROTECTION]	= { -EILSEQ,	"protection" },
182 	[BLK_STS_RESOURCE]	= { -ENOMEM,	"kernel resource" },
183 	[BLK_STS_DEV_RESOURCE]	= { -EBUSY,	"device resource" },
184 	[BLK_STS_AGAIN]		= { -EAGAIN,	"nonblocking retry" },
185 
186 	/* device mapper special case, should not leak out: */
187 	[BLK_STS_DM_REQUEUE]	= { -EREMCHG, "dm internal retry" },
188 
189 	/* everything else not covered above: */
190 	[BLK_STS_IOERR]		= { -EIO,	"I/O" },
191 };
192 
193 blk_status_t errno_to_blk_status(int errno)
194 {
195 	int i;
196 
197 	for (i = 0; i < ARRAY_SIZE(blk_errors); i++) {
198 		if (blk_errors[i].errno == errno)
199 			return (__force blk_status_t)i;
200 	}
201 
202 	return BLK_STS_IOERR;
203 }
204 EXPORT_SYMBOL_GPL(errno_to_blk_status);
205 
206 int blk_status_to_errno(blk_status_t status)
207 {
208 	int idx = (__force int)status;
209 
210 	if (WARN_ON_ONCE(idx >= ARRAY_SIZE(blk_errors)))
211 		return -EIO;
212 	return blk_errors[idx].errno;
213 }
214 EXPORT_SYMBOL_GPL(blk_status_to_errno);
215 
216 static void print_req_error(struct request *req, blk_status_t status,
217 		const char *caller)
218 {
219 	int idx = (__force int)status;
220 
221 	if (WARN_ON_ONCE(idx >= ARRAY_SIZE(blk_errors)))
222 		return;
223 
224 	printk_ratelimited(KERN_ERR
225 		"%s: %s error, dev %s, sector %llu op 0x%x:(%s) flags 0x%x "
226 		"phys_seg %u prio class %u\n",
227 		caller, blk_errors[idx].name,
228 		req->rq_disk ? req->rq_disk->disk_name : "?",
229 		blk_rq_pos(req), req_op(req), blk_op_str(req_op(req)),
230 		req->cmd_flags & ~REQ_OP_MASK,
231 		req->nr_phys_segments,
232 		IOPRIO_PRIO_CLASS(req->ioprio));
233 }
234 
235 static void req_bio_endio(struct request *rq, struct bio *bio,
236 			  unsigned int nbytes, blk_status_t error)
237 {
238 	if (error)
239 		bio->bi_status = error;
240 
241 	if (unlikely(rq->rq_flags & RQF_QUIET))
242 		bio_set_flag(bio, BIO_QUIET);
243 
244 	bio_advance(bio, nbytes);
245 
246 	if (req_op(rq) == REQ_OP_ZONE_APPEND && error == BLK_STS_OK) {
247 		/*
248 		 * Partial zone append completions cannot be supported as the
249 		 * BIO fragments may end up not being written sequentially.
250 		 */
251 		if (bio->bi_iter.bi_size)
252 			bio->bi_status = BLK_STS_IOERR;
253 		else
254 			bio->bi_iter.bi_sector = rq->__sector;
255 	}
256 
257 	/* don't actually finish bio if it's part of flush sequence */
258 	if (bio->bi_iter.bi_size == 0 && !(rq->rq_flags & RQF_FLUSH_SEQ))
259 		bio_endio(bio);
260 }
261 
262 void blk_dump_rq_flags(struct request *rq, char *msg)
263 {
264 	printk(KERN_INFO "%s: dev %s: flags=%llx\n", msg,
265 		rq->rq_disk ? rq->rq_disk->disk_name : "?",
266 		(unsigned long long) rq->cmd_flags);
267 
268 	printk(KERN_INFO "  sector %llu, nr/cnr %u/%u\n",
269 	       (unsigned long long)blk_rq_pos(rq),
270 	       blk_rq_sectors(rq), blk_rq_cur_sectors(rq));
271 	printk(KERN_INFO "  bio %p, biotail %p, len %u\n",
272 	       rq->bio, rq->biotail, blk_rq_bytes(rq));
273 }
274 EXPORT_SYMBOL(blk_dump_rq_flags);
275 
276 /**
277  * blk_sync_queue - cancel any pending callbacks on a queue
278  * @q: the queue
279  *
280  * Description:
281  *     The block layer may perform asynchronous callback activity
282  *     on a queue, such as calling the unplug function after a timeout.
283  *     A block device may call blk_sync_queue to ensure that any
284  *     such activity is cancelled, thus allowing it to release resources
285  *     that the callbacks might use. The caller must already have made sure
286  *     that its ->submit_bio will not re-add plugging prior to calling
287  *     this function.
288  *
289  *     This function does not cancel any asynchronous activity arising
290  *     out of elevator or throttling code. That would require elevator_exit()
291  *     and blkcg_exit_queue() to be called with queue lock initialized.
292  *
293  */
294 void blk_sync_queue(struct request_queue *q)
295 {
296 	del_timer_sync(&q->timeout);
297 	cancel_work_sync(&q->timeout_work);
298 }
299 EXPORT_SYMBOL(blk_sync_queue);
300 
301 /**
302  * blk_set_pm_only - increment pm_only counter
303  * @q: request queue pointer
304  */
305 void blk_set_pm_only(struct request_queue *q)
306 {
307 	atomic_inc(&q->pm_only);
308 }
309 EXPORT_SYMBOL_GPL(blk_set_pm_only);
310 
311 void blk_clear_pm_only(struct request_queue *q)
312 {
313 	int pm_only;
314 
315 	pm_only = atomic_dec_return(&q->pm_only);
316 	WARN_ON_ONCE(pm_only < 0);
317 	if (pm_only == 0)
318 		wake_up_all(&q->mq_freeze_wq);
319 }
320 EXPORT_SYMBOL_GPL(blk_clear_pm_only);
321 
322 /**
323  * blk_put_queue - decrement the request_queue refcount
324  * @q: the request_queue structure to decrement the refcount for
325  *
326  * Decrements the refcount of the request_queue kobject. When this reaches 0
327  * we'll have blk_release_queue() called.
328  *
329  * Context: Any context, but the last reference must not be dropped from
330  *          atomic context.
331  */
332 void blk_put_queue(struct request_queue *q)
333 {
334 	kobject_put(&q->kobj);
335 }
336 EXPORT_SYMBOL(blk_put_queue);
337 
338 void blk_set_queue_dying(struct request_queue *q)
339 {
340 	blk_queue_flag_set(QUEUE_FLAG_DYING, q);
341 
342 	/*
343 	 * When queue DYING flag is set, we need to block new req
344 	 * entering queue, so we call blk_freeze_queue_start() to
345 	 * prevent I/O from crossing blk_queue_enter().
346 	 */
347 	blk_freeze_queue_start(q);
348 
349 	if (queue_is_mq(q))
350 		blk_mq_wake_waiters(q);
351 
352 	/* Make blk_queue_enter() reexamine the DYING flag. */
353 	wake_up_all(&q->mq_freeze_wq);
354 }
355 EXPORT_SYMBOL_GPL(blk_set_queue_dying);
356 
357 /**
358  * blk_cleanup_queue - shutdown a request queue
359  * @q: request queue to shutdown
360  *
361  * Mark @q DYING, drain all pending requests, mark @q DEAD, destroy and
362  * put it.  All future requests will be failed immediately with -ENODEV.
363  *
364  * Context: can sleep
365  */
366 void blk_cleanup_queue(struct request_queue *q)
367 {
368 	/* cannot be called from atomic context */
369 	might_sleep();
370 
371 	WARN_ON_ONCE(blk_queue_registered(q));
372 
373 	/* mark @q DYING, no new request or merges will be allowed afterwards */
374 	blk_set_queue_dying(q);
375 
376 	blk_queue_flag_set(QUEUE_FLAG_NOMERGES, q);
377 	blk_queue_flag_set(QUEUE_FLAG_NOXMERGES, q);
378 
379 	/*
380 	 * Drain all requests queued before DYING marking. Set DEAD flag to
381 	 * prevent that blk_mq_run_hw_queues() accesses the hardware queues
382 	 * after draining finished.
383 	 */
384 	blk_freeze_queue(q);
385 
386 	rq_qos_exit(q);
387 
388 	blk_queue_flag_set(QUEUE_FLAG_DEAD, q);
389 
390 	/* for synchronous bio-based driver finish in-flight integrity i/o */
391 	blk_flush_integrity();
392 
393 	/* @q won't process any more request, flush async actions */
394 	del_timer_sync(&q->backing_dev_info->laptop_mode_wb_timer);
395 	blk_sync_queue(q);
396 
397 	if (queue_is_mq(q))
398 		blk_mq_exit_queue(q);
399 
400 	/*
401 	 * In theory, request pool of sched_tags belongs to request queue.
402 	 * However, the current implementation requires tag_set for freeing
403 	 * requests, so free the pool now.
404 	 *
405 	 * Queue has become frozen, there can't be any in-queue requests, so
406 	 * it is safe to free requests now.
407 	 */
408 	mutex_lock(&q->sysfs_lock);
409 	if (q->elevator)
410 		blk_mq_sched_free_requests(q);
411 	mutex_unlock(&q->sysfs_lock);
412 
413 	percpu_ref_exit(&q->q_usage_counter);
414 
415 	/* @q is and will stay empty, shutdown and put */
416 	blk_put_queue(q);
417 }
418 EXPORT_SYMBOL(blk_cleanup_queue);
419 
420 /**
421  * blk_queue_enter() - try to increase q->q_usage_counter
422  * @q: request queue pointer
423  * @flags: BLK_MQ_REQ_NOWAIT and/or BLK_MQ_REQ_PREEMPT
424  */
425 int blk_queue_enter(struct request_queue *q, blk_mq_req_flags_t flags)
426 {
427 	const bool pm = flags & BLK_MQ_REQ_PREEMPT;
428 
429 	while (true) {
430 		bool success = false;
431 
432 		rcu_read_lock();
433 		if (percpu_ref_tryget_live(&q->q_usage_counter)) {
434 			/*
435 			 * The code that increments the pm_only counter is
436 			 * responsible for ensuring that that counter is
437 			 * globally visible before the queue is unfrozen.
438 			 */
439 			if (pm || !blk_queue_pm_only(q)) {
440 				success = true;
441 			} else {
442 				percpu_ref_put(&q->q_usage_counter);
443 			}
444 		}
445 		rcu_read_unlock();
446 
447 		if (success)
448 			return 0;
449 
450 		if (flags & BLK_MQ_REQ_NOWAIT)
451 			return -EBUSY;
452 
453 		/*
454 		 * read pair of barrier in blk_freeze_queue_start(),
455 		 * we need to order reading __PERCPU_REF_DEAD flag of
456 		 * .q_usage_counter and reading .mq_freeze_depth or
457 		 * queue dying flag, otherwise the following wait may
458 		 * never return if the two reads are reordered.
459 		 */
460 		smp_rmb();
461 
462 		wait_event(q->mq_freeze_wq,
463 			   (!q->mq_freeze_depth &&
464 			    (pm || (blk_pm_request_resume(q),
465 				    !blk_queue_pm_only(q)))) ||
466 			   blk_queue_dying(q));
467 		if (blk_queue_dying(q))
468 			return -ENODEV;
469 	}
470 }
471 
472 static inline int bio_queue_enter(struct bio *bio)
473 {
474 	struct request_queue *q = bio->bi_disk->queue;
475 	bool nowait = bio->bi_opf & REQ_NOWAIT;
476 	int ret;
477 
478 	ret = blk_queue_enter(q, nowait ? BLK_MQ_REQ_NOWAIT : 0);
479 	if (unlikely(ret)) {
480 		if (nowait && !blk_queue_dying(q))
481 			bio_wouldblock_error(bio);
482 		else
483 			bio_io_error(bio);
484 	}
485 
486 	return ret;
487 }
488 
489 void blk_queue_exit(struct request_queue *q)
490 {
491 	percpu_ref_put(&q->q_usage_counter);
492 }
493 
494 static void blk_queue_usage_counter_release(struct percpu_ref *ref)
495 {
496 	struct request_queue *q =
497 		container_of(ref, struct request_queue, q_usage_counter);
498 
499 	wake_up_all(&q->mq_freeze_wq);
500 }
501 
502 static void blk_rq_timed_out_timer(struct timer_list *t)
503 {
504 	struct request_queue *q = from_timer(q, t, timeout);
505 
506 	kblockd_schedule_work(&q->timeout_work);
507 }
508 
509 static void blk_timeout_work(struct work_struct *work)
510 {
511 }
512 
513 struct request_queue *blk_alloc_queue(int node_id)
514 {
515 	struct request_queue *q;
516 	int ret;
517 
518 	q = kmem_cache_alloc_node(blk_requestq_cachep,
519 				GFP_KERNEL | __GFP_ZERO, node_id);
520 	if (!q)
521 		return NULL;
522 
523 	q->last_merge = NULL;
524 
525 	q->id = ida_simple_get(&blk_queue_ida, 0, 0, GFP_KERNEL);
526 	if (q->id < 0)
527 		goto fail_q;
528 
529 	ret = bioset_init(&q->bio_split, BIO_POOL_SIZE, 0, BIOSET_NEED_BVECS);
530 	if (ret)
531 		goto fail_id;
532 
533 	q->backing_dev_info = bdi_alloc(node_id);
534 	if (!q->backing_dev_info)
535 		goto fail_split;
536 
537 	q->stats = blk_alloc_queue_stats();
538 	if (!q->stats)
539 		goto fail_stats;
540 
541 	q->backing_dev_info->ra_pages = VM_READAHEAD_PAGES;
542 	q->backing_dev_info->capabilities = BDI_CAP_CGROUP_WRITEBACK;
543 	q->node = node_id;
544 
545 	timer_setup(&q->backing_dev_info->laptop_mode_wb_timer,
546 		    laptop_mode_timer_fn, 0);
547 	timer_setup(&q->timeout, blk_rq_timed_out_timer, 0);
548 	INIT_WORK(&q->timeout_work, blk_timeout_work);
549 	INIT_LIST_HEAD(&q->icq_list);
550 #ifdef CONFIG_BLK_CGROUP
551 	INIT_LIST_HEAD(&q->blkg_list);
552 #endif
553 
554 	kobject_init(&q->kobj, &blk_queue_ktype);
555 
556 	mutex_init(&q->debugfs_mutex);
557 	mutex_init(&q->sysfs_lock);
558 	mutex_init(&q->sysfs_dir_lock);
559 	spin_lock_init(&q->queue_lock);
560 
561 	init_waitqueue_head(&q->mq_freeze_wq);
562 	mutex_init(&q->mq_freeze_lock);
563 
564 	/*
565 	 * Init percpu_ref in atomic mode so that it's faster to shutdown.
566 	 * See blk_register_queue() for details.
567 	 */
568 	if (percpu_ref_init(&q->q_usage_counter,
569 				blk_queue_usage_counter_release,
570 				PERCPU_REF_INIT_ATOMIC, GFP_KERNEL))
571 		goto fail_bdi;
572 
573 	if (blkcg_init_queue(q))
574 		goto fail_ref;
575 
576 	blk_queue_dma_alignment(q, 511);
577 	blk_set_default_limits(&q->limits);
578 	q->nr_requests = BLKDEV_MAX_RQ;
579 
580 	return q;
581 
582 fail_ref:
583 	percpu_ref_exit(&q->q_usage_counter);
584 fail_bdi:
585 	blk_free_queue_stats(q->stats);
586 fail_stats:
587 	bdi_put(q->backing_dev_info);
588 fail_split:
589 	bioset_exit(&q->bio_split);
590 fail_id:
591 	ida_simple_remove(&blk_queue_ida, q->id);
592 fail_q:
593 	kmem_cache_free(blk_requestq_cachep, q);
594 	return NULL;
595 }
596 EXPORT_SYMBOL(blk_alloc_queue);
597 
598 /**
599  * blk_get_queue - increment the request_queue refcount
600  * @q: the request_queue structure to increment the refcount for
601  *
602  * Increment the refcount of the request_queue kobject.
603  *
604  * Context: Any context.
605  */
606 bool blk_get_queue(struct request_queue *q)
607 {
608 	if (likely(!blk_queue_dying(q))) {
609 		__blk_get_queue(q);
610 		return true;
611 	}
612 
613 	return false;
614 }
615 EXPORT_SYMBOL(blk_get_queue);
616 
617 /**
618  * blk_get_request - allocate a request
619  * @q: request queue to allocate a request for
620  * @op: operation (REQ_OP_*) and REQ_* flags, e.g. REQ_SYNC.
621  * @flags: BLK_MQ_REQ_* flags, e.g. BLK_MQ_REQ_NOWAIT.
622  */
623 struct request *blk_get_request(struct request_queue *q, unsigned int op,
624 				blk_mq_req_flags_t flags)
625 {
626 	struct request *req;
627 
628 	WARN_ON_ONCE(op & REQ_NOWAIT);
629 	WARN_ON_ONCE(flags & ~(BLK_MQ_REQ_NOWAIT | BLK_MQ_REQ_PREEMPT));
630 
631 	req = blk_mq_alloc_request(q, op, flags);
632 	if (!IS_ERR(req) && q->mq_ops->initialize_rq_fn)
633 		q->mq_ops->initialize_rq_fn(req);
634 
635 	return req;
636 }
637 EXPORT_SYMBOL(blk_get_request);
638 
639 void blk_put_request(struct request *req)
640 {
641 	blk_mq_free_request(req);
642 }
643 EXPORT_SYMBOL(blk_put_request);
644 
645 static void blk_account_io_merge_bio(struct request *req)
646 {
647 	if (!blk_do_io_stat(req))
648 		return;
649 
650 	part_stat_lock();
651 	part_stat_inc(req->part, merges[op_stat_group(req_op(req))]);
652 	part_stat_unlock();
653 }
654 
655 bool bio_attempt_back_merge(struct request *req, struct bio *bio,
656 		unsigned int nr_segs)
657 {
658 	const int ff = bio->bi_opf & REQ_FAILFAST_MASK;
659 
660 	if (!ll_back_merge_fn(req, bio, nr_segs))
661 		return false;
662 
663 	trace_block_bio_backmerge(req->q, req, bio);
664 	rq_qos_merge(req->q, req, bio);
665 
666 	if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
667 		blk_rq_set_mixed_merge(req);
668 
669 	req->biotail->bi_next = bio;
670 	req->biotail = bio;
671 	req->__data_len += bio->bi_iter.bi_size;
672 
673 	bio_crypt_free_ctx(bio);
674 
675 	blk_account_io_merge_bio(req);
676 	return true;
677 }
678 
679 bool bio_attempt_front_merge(struct request *req, struct bio *bio,
680 		unsigned int nr_segs)
681 {
682 	const int ff = bio->bi_opf & REQ_FAILFAST_MASK;
683 
684 	if (!ll_front_merge_fn(req, bio, nr_segs))
685 		return false;
686 
687 	trace_block_bio_frontmerge(req->q, req, bio);
688 	rq_qos_merge(req->q, req, bio);
689 
690 	if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
691 		blk_rq_set_mixed_merge(req);
692 
693 	bio->bi_next = req->bio;
694 	req->bio = bio;
695 
696 	req->__sector = bio->bi_iter.bi_sector;
697 	req->__data_len += bio->bi_iter.bi_size;
698 
699 	bio_crypt_do_front_merge(req, bio);
700 
701 	blk_account_io_merge_bio(req);
702 	return true;
703 }
704 
705 bool bio_attempt_discard_merge(struct request_queue *q, struct request *req,
706 		struct bio *bio)
707 {
708 	unsigned short segments = blk_rq_nr_discard_segments(req);
709 
710 	if (segments >= queue_max_discard_segments(q))
711 		goto no_merge;
712 	if (blk_rq_sectors(req) + bio_sectors(bio) >
713 	    blk_rq_get_max_sectors(req, blk_rq_pos(req)))
714 		goto no_merge;
715 
716 	rq_qos_merge(q, req, bio);
717 
718 	req->biotail->bi_next = bio;
719 	req->biotail = bio;
720 	req->__data_len += bio->bi_iter.bi_size;
721 	req->nr_phys_segments = segments + 1;
722 
723 	blk_account_io_merge_bio(req);
724 	return true;
725 no_merge:
726 	req_set_nomerge(q, req);
727 	return false;
728 }
729 
730 /**
731  * blk_attempt_plug_merge - try to merge with %current's plugged list
732  * @q: request_queue new bio is being queued at
733  * @bio: new bio being queued
734  * @nr_segs: number of segments in @bio
735  * @same_queue_rq: pointer to &struct request that gets filled in when
736  * another request associated with @q is found on the plug list
737  * (optional, may be %NULL)
738  *
739  * Determine whether @bio being queued on @q can be merged with a request
740  * on %current's plugged list.  Returns %true if merge was successful,
741  * otherwise %false.
742  *
743  * Plugging coalesces IOs from the same issuer for the same purpose without
744  * going through @q->queue_lock.  As such it's more of an issuing mechanism
745  * than scheduling, and the request, while may have elvpriv data, is not
746  * added on the elevator at this point.  In addition, we don't have
747  * reliable access to the elevator outside queue lock.  Only check basic
748  * merging parameters without querying the elevator.
749  *
750  * Caller must ensure !blk_queue_nomerges(q) beforehand.
751  */
752 bool blk_attempt_plug_merge(struct request_queue *q, struct bio *bio,
753 		unsigned int nr_segs, struct request **same_queue_rq)
754 {
755 	struct blk_plug *plug;
756 	struct request *rq;
757 	struct list_head *plug_list;
758 
759 	plug = blk_mq_plug(q, bio);
760 	if (!plug)
761 		return false;
762 
763 	plug_list = &plug->mq_list;
764 
765 	list_for_each_entry_reverse(rq, plug_list, queuelist) {
766 		bool merged = false;
767 
768 		if (rq->q == q && same_queue_rq) {
769 			/*
770 			 * Only blk-mq multiple hardware queues case checks the
771 			 * rq in the same queue, there should be only one such
772 			 * rq in a queue
773 			 **/
774 			*same_queue_rq = rq;
775 		}
776 
777 		if (rq->q != q || !blk_rq_merge_ok(rq, bio))
778 			continue;
779 
780 		switch (blk_try_merge(rq, bio)) {
781 		case ELEVATOR_BACK_MERGE:
782 			merged = bio_attempt_back_merge(rq, bio, nr_segs);
783 			break;
784 		case ELEVATOR_FRONT_MERGE:
785 			merged = bio_attempt_front_merge(rq, bio, nr_segs);
786 			break;
787 		case ELEVATOR_DISCARD_MERGE:
788 			merged = bio_attempt_discard_merge(q, rq, bio);
789 			break;
790 		default:
791 			break;
792 		}
793 
794 		if (merged)
795 			return true;
796 	}
797 
798 	return false;
799 }
800 
801 static void handle_bad_sector(struct bio *bio, sector_t maxsector)
802 {
803 	char b[BDEVNAME_SIZE];
804 
805 	printk(KERN_INFO "attempt to access beyond end of device\n");
806 	printk(KERN_INFO "%s: rw=%d, want=%Lu, limit=%Lu\n",
807 			bio_devname(bio, b), bio->bi_opf,
808 			(unsigned long long)bio_end_sector(bio),
809 			(long long)maxsector);
810 }
811 
812 #ifdef CONFIG_FAIL_MAKE_REQUEST
813 
814 static DECLARE_FAULT_ATTR(fail_make_request);
815 
816 static int __init setup_fail_make_request(char *str)
817 {
818 	return setup_fault_attr(&fail_make_request, str);
819 }
820 __setup("fail_make_request=", setup_fail_make_request);
821 
822 static bool should_fail_request(struct hd_struct *part, unsigned int bytes)
823 {
824 	return part->make_it_fail && should_fail(&fail_make_request, bytes);
825 }
826 
827 static int __init fail_make_request_debugfs(void)
828 {
829 	struct dentry *dir = fault_create_debugfs_attr("fail_make_request",
830 						NULL, &fail_make_request);
831 
832 	return PTR_ERR_OR_ZERO(dir);
833 }
834 
835 late_initcall(fail_make_request_debugfs);
836 
837 #else /* CONFIG_FAIL_MAKE_REQUEST */
838 
839 static inline bool should_fail_request(struct hd_struct *part,
840 					unsigned int bytes)
841 {
842 	return false;
843 }
844 
845 #endif /* CONFIG_FAIL_MAKE_REQUEST */
846 
847 static inline bool bio_check_ro(struct bio *bio, struct hd_struct *part)
848 {
849 	const int op = bio_op(bio);
850 
851 	if (part->policy && op_is_write(op)) {
852 		char b[BDEVNAME_SIZE];
853 
854 		if (op_is_flush(bio->bi_opf) && !bio_sectors(bio))
855 			return false;
856 
857 		WARN_ONCE(1,
858 		       "Trying to write to read-only block-device %s (partno %d)\n",
859 			bio_devname(bio, b), part->partno);
860 		/* Older lvm-tools actually trigger this */
861 		return false;
862 	}
863 
864 	return false;
865 }
866 
867 static noinline int should_fail_bio(struct bio *bio)
868 {
869 	if (should_fail_request(&bio->bi_disk->part0, bio->bi_iter.bi_size))
870 		return -EIO;
871 	return 0;
872 }
873 ALLOW_ERROR_INJECTION(should_fail_bio, ERRNO);
874 
875 /*
876  * Check whether this bio extends beyond the end of the device or partition.
877  * This may well happen - the kernel calls bread() without checking the size of
878  * the device, e.g., when mounting a file system.
879  */
880 static inline int bio_check_eod(struct bio *bio, sector_t maxsector)
881 {
882 	unsigned int nr_sectors = bio_sectors(bio);
883 
884 	if (nr_sectors && maxsector &&
885 	    (nr_sectors > maxsector ||
886 	     bio->bi_iter.bi_sector > maxsector - nr_sectors)) {
887 		handle_bad_sector(bio, maxsector);
888 		return -EIO;
889 	}
890 	return 0;
891 }
892 
893 /*
894  * Remap block n of partition p to block n+start(p) of the disk.
895  */
896 static inline int blk_partition_remap(struct bio *bio)
897 {
898 	struct hd_struct *p;
899 	int ret = -EIO;
900 
901 	rcu_read_lock();
902 	p = __disk_get_part(bio->bi_disk, bio->bi_partno);
903 	if (unlikely(!p))
904 		goto out;
905 	if (unlikely(should_fail_request(p, bio->bi_iter.bi_size)))
906 		goto out;
907 	if (unlikely(bio_check_ro(bio, p)))
908 		goto out;
909 
910 	if (bio_sectors(bio)) {
911 		if (bio_check_eod(bio, part_nr_sects_read(p)))
912 			goto out;
913 		bio->bi_iter.bi_sector += p->start_sect;
914 		trace_block_bio_remap(bio->bi_disk->queue, bio, part_devt(p),
915 				      bio->bi_iter.bi_sector - p->start_sect);
916 	}
917 	bio->bi_partno = 0;
918 	ret = 0;
919 out:
920 	rcu_read_unlock();
921 	return ret;
922 }
923 
924 /*
925  * Check write append to a zoned block device.
926  */
927 static inline blk_status_t blk_check_zone_append(struct request_queue *q,
928 						 struct bio *bio)
929 {
930 	sector_t pos = bio->bi_iter.bi_sector;
931 	int nr_sectors = bio_sectors(bio);
932 
933 	/* Only applicable to zoned block devices */
934 	if (!blk_queue_is_zoned(q))
935 		return BLK_STS_NOTSUPP;
936 
937 	/* The bio sector must point to the start of a sequential zone */
938 	if (pos & (blk_queue_zone_sectors(q) - 1) ||
939 	    !blk_queue_zone_is_seq(q, pos))
940 		return BLK_STS_IOERR;
941 
942 	/*
943 	 * Not allowed to cross zone boundaries. Otherwise, the BIO will be
944 	 * split and could result in non-contiguous sectors being written in
945 	 * different zones.
946 	 */
947 	if (nr_sectors > q->limits.chunk_sectors)
948 		return BLK_STS_IOERR;
949 
950 	/* Make sure the BIO is small enough and will not get split */
951 	if (nr_sectors > q->limits.max_zone_append_sectors)
952 		return BLK_STS_IOERR;
953 
954 	bio->bi_opf |= REQ_NOMERGE;
955 
956 	return BLK_STS_OK;
957 }
958 
959 static noinline_for_stack bool submit_bio_checks(struct bio *bio)
960 {
961 	struct request_queue *q = bio->bi_disk->queue;
962 	blk_status_t status = BLK_STS_IOERR;
963 
964 	might_sleep();
965 
966 	/*
967 	 * For a REQ_NOWAIT based request, return -EOPNOTSUPP
968 	 * if queue is not a request based queue.
969 	 */
970 	if ((bio->bi_opf & REQ_NOWAIT) && !queue_is_mq(q))
971 		goto not_supported;
972 
973 	if (should_fail_bio(bio))
974 		goto end_io;
975 
976 	if (bio->bi_partno) {
977 		if (unlikely(blk_partition_remap(bio)))
978 			goto end_io;
979 	} else {
980 		if (unlikely(bio_check_ro(bio, &bio->bi_disk->part0)))
981 			goto end_io;
982 		if (unlikely(bio_check_eod(bio, get_capacity(bio->bi_disk))))
983 			goto end_io;
984 	}
985 
986 	/*
987 	 * Filter flush bio's early so that bio based drivers without flush
988 	 * support don't have to worry about them.
989 	 */
990 	if (op_is_flush(bio->bi_opf) &&
991 	    !test_bit(QUEUE_FLAG_WC, &q->queue_flags)) {
992 		bio->bi_opf &= ~(REQ_PREFLUSH | REQ_FUA);
993 		if (!bio_sectors(bio)) {
994 			status = BLK_STS_OK;
995 			goto end_io;
996 		}
997 	}
998 
999 	if (!test_bit(QUEUE_FLAG_POLL, &q->queue_flags))
1000 		bio->bi_opf &= ~REQ_HIPRI;
1001 
1002 	switch (bio_op(bio)) {
1003 	case REQ_OP_DISCARD:
1004 		if (!blk_queue_discard(q))
1005 			goto not_supported;
1006 		break;
1007 	case REQ_OP_SECURE_ERASE:
1008 		if (!blk_queue_secure_erase(q))
1009 			goto not_supported;
1010 		break;
1011 	case REQ_OP_WRITE_SAME:
1012 		if (!q->limits.max_write_same_sectors)
1013 			goto not_supported;
1014 		break;
1015 	case REQ_OP_ZONE_APPEND:
1016 		status = blk_check_zone_append(q, bio);
1017 		if (status != BLK_STS_OK)
1018 			goto end_io;
1019 		break;
1020 	case REQ_OP_ZONE_RESET:
1021 	case REQ_OP_ZONE_OPEN:
1022 	case REQ_OP_ZONE_CLOSE:
1023 	case REQ_OP_ZONE_FINISH:
1024 		if (!blk_queue_is_zoned(q))
1025 			goto not_supported;
1026 		break;
1027 	case REQ_OP_ZONE_RESET_ALL:
1028 		if (!blk_queue_is_zoned(q) || !blk_queue_zone_resetall(q))
1029 			goto not_supported;
1030 		break;
1031 	case REQ_OP_WRITE_ZEROES:
1032 		if (!q->limits.max_write_zeroes_sectors)
1033 			goto not_supported;
1034 		break;
1035 	default:
1036 		break;
1037 	}
1038 
1039 	/*
1040 	 * Various block parts want %current->io_context, so allocate it up
1041 	 * front rather than dealing with lots of pain to allocate it only
1042 	 * where needed. This may fail and the block layer knows how to live
1043 	 * with it.
1044 	 */
1045 	if (unlikely(!current->io_context))
1046 		create_task_io_context(current, GFP_ATOMIC, q->node);
1047 
1048 	if (blk_throtl_bio(bio)) {
1049 		blkcg_bio_issue_init(bio);
1050 		return false;
1051 	}
1052 
1053 	blk_cgroup_bio_start(bio);
1054 	blkcg_bio_issue_init(bio);
1055 
1056 	if (!bio_flagged(bio, BIO_TRACE_COMPLETION)) {
1057 		trace_block_bio_queue(q, bio);
1058 		/* Now that enqueuing has been traced, we need to trace
1059 		 * completion as well.
1060 		 */
1061 		bio_set_flag(bio, BIO_TRACE_COMPLETION);
1062 	}
1063 	return true;
1064 
1065 not_supported:
1066 	status = BLK_STS_NOTSUPP;
1067 end_io:
1068 	bio->bi_status = status;
1069 	bio_endio(bio);
1070 	return false;
1071 }
1072 
1073 static blk_qc_t __submit_bio(struct bio *bio)
1074 {
1075 	struct gendisk *disk = bio->bi_disk;
1076 	blk_qc_t ret = BLK_QC_T_NONE;
1077 
1078 	if (blk_crypto_bio_prep(&bio)) {
1079 		if (!disk->fops->submit_bio)
1080 			return blk_mq_submit_bio(bio);
1081 		ret = disk->fops->submit_bio(bio);
1082 	}
1083 	blk_queue_exit(disk->queue);
1084 	return ret;
1085 }
1086 
1087 /**
1088  * submit_bio_noacct - re-submit a bio to the block device layer for I/O
1089  * @bio:  The bio describing the location in memory and on the device.
1090  *
1091  * This is a version of submit_bio() that shall only be used for I/O that is
1092  * resubmitted to lower level drivers by stacking block drivers.  All file
1093  * systems and other upper level users of the block layer should use
1094  * submit_bio() instead.
1095  */
1096 blk_qc_t submit_bio_noacct(struct bio *bio)
1097 {
1098 	/*
1099 	 * bio_list_on_stack[0] contains bios submitted by the current
1100 	 * ->submit_bio.
1101 	 * bio_list_on_stack[1] contains bios that were submitted before the
1102 	 * current ->submit_bio_bio, but that haven't been processed yet.
1103 	 */
1104 	struct bio_list bio_list_on_stack[2];
1105 	blk_qc_t ret = BLK_QC_T_NONE;
1106 
1107 	if (!submit_bio_checks(bio))
1108 		goto out;
1109 
1110 	/*
1111 	 * We only want one ->submit_bio to be active at a time, else
1112 	 * stack usage with stacked devices could be a problem.  So use
1113 	 * current->bio_list to keep a list of requests submited by a
1114 	 * ->submit_bio method.  current->bio_list is also used as a
1115 	 * flag to say if submit_bio_noacct is currently active in this
1116 	 * task or not.  If it is NULL, then no make_request is active.  If
1117 	 * it is non-NULL, then a make_request is active, and new requests
1118 	 * should be added at the tail
1119 	 */
1120 	if (current->bio_list) {
1121 		bio_list_add(&current->bio_list[0], bio);
1122 		goto out;
1123 	}
1124 
1125 	/* following loop may be a bit non-obvious, and so deserves some
1126 	 * explanation.
1127 	 * Before entering the loop, bio->bi_next is NULL (as all callers
1128 	 * ensure that) so we have a list with a single bio.
1129 	 * We pretend that we have just taken it off a longer list, so
1130 	 * we assign bio_list to a pointer to the bio_list_on_stack,
1131 	 * thus initialising the bio_list of new bios to be
1132 	 * added.  ->submit_bio() may indeed add some more bios
1133 	 * through a recursive call to submit_bio_noacct.  If it
1134 	 * did, we find a non-NULL value in bio_list and re-enter the loop
1135 	 * from the top.  In this case we really did just take the bio
1136 	 * of the top of the list (no pretending) and so remove it from
1137 	 * bio_list, and call into ->submit_bio() again.
1138 	 */
1139 	BUG_ON(bio->bi_next);
1140 	bio_list_init(&bio_list_on_stack[0]);
1141 	current->bio_list = bio_list_on_stack;
1142 	do {
1143 		struct request_queue *q = bio->bi_disk->queue;
1144 
1145 		if (likely(bio_queue_enter(bio) == 0)) {
1146 			struct bio_list lower, same;
1147 
1148 			/* Create a fresh bio_list for all subordinate requests */
1149 			bio_list_on_stack[1] = bio_list_on_stack[0];
1150 			bio_list_init(&bio_list_on_stack[0]);
1151 			ret = __submit_bio(bio);
1152 
1153 			/* sort new bios into those for a lower level
1154 			 * and those for the same level
1155 			 */
1156 			bio_list_init(&lower);
1157 			bio_list_init(&same);
1158 			while ((bio = bio_list_pop(&bio_list_on_stack[0])) != NULL)
1159 				if (q == bio->bi_disk->queue)
1160 					bio_list_add(&same, bio);
1161 				else
1162 					bio_list_add(&lower, bio);
1163 			/* now assemble so we handle the lowest level first */
1164 			bio_list_merge(&bio_list_on_stack[0], &lower);
1165 			bio_list_merge(&bio_list_on_stack[0], &same);
1166 			bio_list_merge(&bio_list_on_stack[0], &bio_list_on_stack[1]);
1167 		}
1168 		bio = bio_list_pop(&bio_list_on_stack[0]);
1169 	} while (bio);
1170 	current->bio_list = NULL; /* deactivate */
1171 
1172 out:
1173 	return ret;
1174 }
1175 EXPORT_SYMBOL(submit_bio_noacct);
1176 
1177 /**
1178  * direct_make_request - hand a buffer directly to its device driver for I/O
1179  * @bio:  The bio describing the location in memory and on the device.
1180  *
1181  * This function behaves like submit_bio_noacct(), but does not protect
1182  * against recursion.  Must only be used if the called driver is known
1183  * to be blk-mq based.
1184  */
1185 blk_qc_t direct_make_request(struct bio *bio)
1186 {
1187 	struct gendisk *disk = bio->bi_disk;
1188 
1189 	if (WARN_ON_ONCE(!disk->queue->mq_ops)) {
1190 		bio_io_error(bio);
1191 		return BLK_QC_T_NONE;
1192 	}
1193 	if (!submit_bio_checks(bio))
1194 		return BLK_QC_T_NONE;
1195 	if (unlikely(bio_queue_enter(bio)))
1196 		return BLK_QC_T_NONE;
1197 	if (!blk_crypto_bio_prep(&bio)) {
1198 		blk_queue_exit(disk->queue);
1199 		return BLK_QC_T_NONE;
1200 	}
1201 	return blk_mq_submit_bio(bio);
1202 }
1203 EXPORT_SYMBOL_GPL(direct_make_request);
1204 
1205 /**
1206  * submit_bio - submit a bio to the block device layer for I/O
1207  * @bio: The &struct bio which describes the I/O
1208  *
1209  * submit_bio() is used to submit I/O requests to block devices.  It is passed a
1210  * fully set up &struct bio that describes the I/O that needs to be done.  The
1211  * bio will be send to the device described by the bi_disk and bi_partno fields.
1212  *
1213  * The success/failure status of the request, along with notification of
1214  * completion, is delivered asynchronously through the ->bi_end_io() callback
1215  * in @bio.  The bio must NOT be touched by thecaller until ->bi_end_io() has
1216  * been called.
1217  */
1218 blk_qc_t submit_bio(struct bio *bio)
1219 {
1220 	if (blkcg_punt_bio_submit(bio))
1221 		return BLK_QC_T_NONE;
1222 
1223 	/*
1224 	 * If it's a regular read/write or a barrier with data attached,
1225 	 * go through the normal accounting stuff before submission.
1226 	 */
1227 	if (bio_has_data(bio)) {
1228 		unsigned int count;
1229 
1230 		if (unlikely(bio_op(bio) == REQ_OP_WRITE_SAME))
1231 			count = queue_logical_block_size(bio->bi_disk->queue) >> 9;
1232 		else
1233 			count = bio_sectors(bio);
1234 
1235 		if (op_is_write(bio_op(bio))) {
1236 			count_vm_events(PGPGOUT, count);
1237 		} else {
1238 			task_io_account_read(bio->bi_iter.bi_size);
1239 			count_vm_events(PGPGIN, count);
1240 		}
1241 
1242 		if (unlikely(block_dump)) {
1243 			char b[BDEVNAME_SIZE];
1244 			printk(KERN_DEBUG "%s(%d): %s block %Lu on %s (%u sectors)\n",
1245 			current->comm, task_pid_nr(current),
1246 				op_is_write(bio_op(bio)) ? "WRITE" : "READ",
1247 				(unsigned long long)bio->bi_iter.bi_sector,
1248 				bio_devname(bio, b), count);
1249 		}
1250 	}
1251 
1252 	/*
1253 	 * If we're reading data that is part of the userspace workingset, count
1254 	 * submission time as memory stall.  When the device is congested, or
1255 	 * the submitting cgroup IO-throttled, submission can be a significant
1256 	 * part of overall IO time.
1257 	 */
1258 	if (unlikely(bio_op(bio) == REQ_OP_READ &&
1259 	    bio_flagged(bio, BIO_WORKINGSET))) {
1260 		unsigned long pflags;
1261 		blk_qc_t ret;
1262 
1263 		psi_memstall_enter(&pflags);
1264 		ret = submit_bio_noacct(bio);
1265 		psi_memstall_leave(&pflags);
1266 
1267 		return ret;
1268 	}
1269 
1270 	return submit_bio_noacct(bio);
1271 }
1272 EXPORT_SYMBOL(submit_bio);
1273 
1274 /**
1275  * blk_cloned_rq_check_limits - Helper function to check a cloned request
1276  *                              for the new queue limits
1277  * @q:  the queue
1278  * @rq: the request being checked
1279  *
1280  * Description:
1281  *    @rq may have been made based on weaker limitations of upper-level queues
1282  *    in request stacking drivers, and it may violate the limitation of @q.
1283  *    Since the block layer and the underlying device driver trust @rq
1284  *    after it is inserted to @q, it should be checked against @q before
1285  *    the insertion using this generic function.
1286  *
1287  *    Request stacking drivers like request-based dm may change the queue
1288  *    limits when retrying requests on other queues. Those requests need
1289  *    to be checked against the new queue limits again during dispatch.
1290  */
1291 static int blk_cloned_rq_check_limits(struct request_queue *q,
1292 				      struct request *rq)
1293 {
1294 	if (blk_rq_sectors(rq) > blk_queue_get_max_sectors(q, req_op(rq))) {
1295 		printk(KERN_ERR "%s: over max size limit. (%u > %u)\n",
1296 			__func__, blk_rq_sectors(rq),
1297 			blk_queue_get_max_sectors(q, req_op(rq)));
1298 		return -EIO;
1299 	}
1300 
1301 	/*
1302 	 * queue's settings related to segment counting like q->bounce_pfn
1303 	 * may differ from that of other stacking queues.
1304 	 * Recalculate it to check the request correctly on this queue's
1305 	 * limitation.
1306 	 */
1307 	rq->nr_phys_segments = blk_recalc_rq_segments(rq);
1308 	if (rq->nr_phys_segments > queue_max_segments(q)) {
1309 		printk(KERN_ERR "%s: over max segments limit. (%hu > %hu)\n",
1310 			__func__, rq->nr_phys_segments, queue_max_segments(q));
1311 		return -EIO;
1312 	}
1313 
1314 	return 0;
1315 }
1316 
1317 /**
1318  * blk_insert_cloned_request - Helper for stacking drivers to submit a request
1319  * @q:  the queue to submit the request
1320  * @rq: the request being queued
1321  */
1322 blk_status_t blk_insert_cloned_request(struct request_queue *q, struct request *rq)
1323 {
1324 	if (blk_cloned_rq_check_limits(q, rq))
1325 		return BLK_STS_IOERR;
1326 
1327 	if (rq->rq_disk &&
1328 	    should_fail_request(&rq->rq_disk->part0, blk_rq_bytes(rq)))
1329 		return BLK_STS_IOERR;
1330 
1331 	if (blk_crypto_insert_cloned_request(rq))
1332 		return BLK_STS_IOERR;
1333 
1334 	if (blk_queue_io_stat(q))
1335 		blk_account_io_start(rq);
1336 
1337 	/*
1338 	 * Since we have a scheduler attached on the top device,
1339 	 * bypass a potential scheduler on the bottom device for
1340 	 * insert.
1341 	 */
1342 	return blk_mq_request_issue_directly(rq, true);
1343 }
1344 EXPORT_SYMBOL_GPL(blk_insert_cloned_request);
1345 
1346 /**
1347  * blk_rq_err_bytes - determine number of bytes till the next failure boundary
1348  * @rq: request to examine
1349  *
1350  * Description:
1351  *     A request could be merge of IOs which require different failure
1352  *     handling.  This function determines the number of bytes which
1353  *     can be failed from the beginning of the request without
1354  *     crossing into area which need to be retried further.
1355  *
1356  * Return:
1357  *     The number of bytes to fail.
1358  */
1359 unsigned int blk_rq_err_bytes(const struct request *rq)
1360 {
1361 	unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK;
1362 	unsigned int bytes = 0;
1363 	struct bio *bio;
1364 
1365 	if (!(rq->rq_flags & RQF_MIXED_MERGE))
1366 		return blk_rq_bytes(rq);
1367 
1368 	/*
1369 	 * Currently the only 'mixing' which can happen is between
1370 	 * different fastfail types.  We can safely fail portions
1371 	 * which have all the failfast bits that the first one has -
1372 	 * the ones which are at least as eager to fail as the first
1373 	 * one.
1374 	 */
1375 	for (bio = rq->bio; bio; bio = bio->bi_next) {
1376 		if ((bio->bi_opf & ff) != ff)
1377 			break;
1378 		bytes += bio->bi_iter.bi_size;
1379 	}
1380 
1381 	/* this could lead to infinite loop */
1382 	BUG_ON(blk_rq_bytes(rq) && !bytes);
1383 	return bytes;
1384 }
1385 EXPORT_SYMBOL_GPL(blk_rq_err_bytes);
1386 
1387 static void update_io_ticks(struct hd_struct *part, unsigned long now, bool end)
1388 {
1389 	unsigned long stamp;
1390 again:
1391 	stamp = READ_ONCE(part->stamp);
1392 	if (unlikely(stamp != now)) {
1393 		if (likely(cmpxchg(&part->stamp, stamp, now) == stamp))
1394 			__part_stat_add(part, io_ticks, end ? now - stamp : 1);
1395 	}
1396 	if (part->partno) {
1397 		part = &part_to_disk(part)->part0;
1398 		goto again;
1399 	}
1400 }
1401 
1402 static void blk_account_io_completion(struct request *req, unsigned int bytes)
1403 {
1404 	if (req->part && blk_do_io_stat(req)) {
1405 		const int sgrp = op_stat_group(req_op(req));
1406 		struct hd_struct *part;
1407 
1408 		part_stat_lock();
1409 		part = req->part;
1410 		part_stat_add(part, sectors[sgrp], bytes >> 9);
1411 		part_stat_unlock();
1412 	}
1413 }
1414 
1415 void blk_account_io_done(struct request *req, u64 now)
1416 {
1417 	/*
1418 	 * Account IO completion.  flush_rq isn't accounted as a
1419 	 * normal IO on queueing nor completion.  Accounting the
1420 	 * containing request is enough.
1421 	 */
1422 	if (req->part && blk_do_io_stat(req) &&
1423 	    !(req->rq_flags & RQF_FLUSH_SEQ)) {
1424 		const int sgrp = op_stat_group(req_op(req));
1425 		struct hd_struct *part;
1426 
1427 		part_stat_lock();
1428 		part = req->part;
1429 
1430 		update_io_ticks(part, jiffies, true);
1431 		part_stat_inc(part, ios[sgrp]);
1432 		part_stat_add(part, nsecs[sgrp], now - req->start_time_ns);
1433 		part_stat_unlock();
1434 
1435 		hd_struct_put(part);
1436 	}
1437 }
1438 
1439 void blk_account_io_start(struct request *rq)
1440 {
1441 	if (!blk_do_io_stat(rq))
1442 		return;
1443 
1444 	rq->part = disk_map_sector_rcu(rq->rq_disk, blk_rq_pos(rq));
1445 
1446 	part_stat_lock();
1447 	update_io_ticks(rq->part, jiffies, false);
1448 	part_stat_unlock();
1449 }
1450 
1451 unsigned long disk_start_io_acct(struct gendisk *disk, unsigned int sectors,
1452 		unsigned int op)
1453 {
1454 	struct hd_struct *part = &disk->part0;
1455 	const int sgrp = op_stat_group(op);
1456 	unsigned long now = READ_ONCE(jiffies);
1457 
1458 	part_stat_lock();
1459 	update_io_ticks(part, now, false);
1460 	part_stat_inc(part, ios[sgrp]);
1461 	part_stat_add(part, sectors[sgrp], sectors);
1462 	part_stat_local_inc(part, in_flight[op_is_write(op)]);
1463 	part_stat_unlock();
1464 
1465 	return now;
1466 }
1467 EXPORT_SYMBOL(disk_start_io_acct);
1468 
1469 void disk_end_io_acct(struct gendisk *disk, unsigned int op,
1470 		unsigned long start_time)
1471 {
1472 	struct hd_struct *part = &disk->part0;
1473 	const int sgrp = op_stat_group(op);
1474 	unsigned long now = READ_ONCE(jiffies);
1475 	unsigned long duration = now - start_time;
1476 
1477 	part_stat_lock();
1478 	update_io_ticks(part, now, true);
1479 	part_stat_add(part, nsecs[sgrp], jiffies_to_nsecs(duration));
1480 	part_stat_local_dec(part, in_flight[op_is_write(op)]);
1481 	part_stat_unlock();
1482 }
1483 EXPORT_SYMBOL(disk_end_io_acct);
1484 
1485 /*
1486  * Steal bios from a request and add them to a bio list.
1487  * The request must not have been partially completed before.
1488  */
1489 void blk_steal_bios(struct bio_list *list, struct request *rq)
1490 {
1491 	if (rq->bio) {
1492 		if (list->tail)
1493 			list->tail->bi_next = rq->bio;
1494 		else
1495 			list->head = rq->bio;
1496 		list->tail = rq->biotail;
1497 
1498 		rq->bio = NULL;
1499 		rq->biotail = NULL;
1500 	}
1501 
1502 	rq->__data_len = 0;
1503 }
1504 EXPORT_SYMBOL_GPL(blk_steal_bios);
1505 
1506 /**
1507  * blk_update_request - Special helper function for request stacking drivers
1508  * @req:      the request being processed
1509  * @error:    block status code
1510  * @nr_bytes: number of bytes to complete @req
1511  *
1512  * Description:
1513  *     Ends I/O on a number of bytes attached to @req, but doesn't complete
1514  *     the request structure even if @req doesn't have leftover.
1515  *     If @req has leftover, sets it up for the next range of segments.
1516  *
1517  *     This special helper function is only for request stacking drivers
1518  *     (e.g. request-based dm) so that they can handle partial completion.
1519  *     Actual device drivers should use blk_mq_end_request instead.
1520  *
1521  *     Passing the result of blk_rq_bytes() as @nr_bytes guarantees
1522  *     %false return from this function.
1523  *
1524  * Note:
1525  *	The RQF_SPECIAL_PAYLOAD flag is ignored on purpose in both
1526  *	blk_rq_bytes() and in blk_update_request().
1527  *
1528  * Return:
1529  *     %false - this request doesn't have any more data
1530  *     %true  - this request has more data
1531  **/
1532 bool blk_update_request(struct request *req, blk_status_t error,
1533 		unsigned int nr_bytes)
1534 {
1535 	int total_bytes;
1536 
1537 	trace_block_rq_complete(req, blk_status_to_errno(error), nr_bytes);
1538 
1539 	if (!req->bio)
1540 		return false;
1541 
1542 #ifdef CONFIG_BLK_DEV_INTEGRITY
1543 	if (blk_integrity_rq(req) && req_op(req) == REQ_OP_READ &&
1544 	    error == BLK_STS_OK)
1545 		req->q->integrity.profile->complete_fn(req, nr_bytes);
1546 #endif
1547 
1548 	if (unlikely(error && !blk_rq_is_passthrough(req) &&
1549 		     !(req->rq_flags & RQF_QUIET)))
1550 		print_req_error(req, error, __func__);
1551 
1552 	blk_account_io_completion(req, nr_bytes);
1553 
1554 	total_bytes = 0;
1555 	while (req->bio) {
1556 		struct bio *bio = req->bio;
1557 		unsigned bio_bytes = min(bio->bi_iter.bi_size, nr_bytes);
1558 
1559 		if (bio_bytes == bio->bi_iter.bi_size)
1560 			req->bio = bio->bi_next;
1561 
1562 		/* Completion has already been traced */
1563 		bio_clear_flag(bio, BIO_TRACE_COMPLETION);
1564 		req_bio_endio(req, bio, bio_bytes, error);
1565 
1566 		total_bytes += bio_bytes;
1567 		nr_bytes -= bio_bytes;
1568 
1569 		if (!nr_bytes)
1570 			break;
1571 	}
1572 
1573 	/*
1574 	 * completely done
1575 	 */
1576 	if (!req->bio) {
1577 		/*
1578 		 * Reset counters so that the request stacking driver
1579 		 * can find how many bytes remain in the request
1580 		 * later.
1581 		 */
1582 		req->__data_len = 0;
1583 		return false;
1584 	}
1585 
1586 	req->__data_len -= total_bytes;
1587 
1588 	/* update sector only for requests with clear definition of sector */
1589 	if (!blk_rq_is_passthrough(req))
1590 		req->__sector += total_bytes >> 9;
1591 
1592 	/* mixed attributes always follow the first bio */
1593 	if (req->rq_flags & RQF_MIXED_MERGE) {
1594 		req->cmd_flags &= ~REQ_FAILFAST_MASK;
1595 		req->cmd_flags |= req->bio->bi_opf & REQ_FAILFAST_MASK;
1596 	}
1597 
1598 	if (!(req->rq_flags & RQF_SPECIAL_PAYLOAD)) {
1599 		/*
1600 		 * If total number of sectors is less than the first segment
1601 		 * size, something has gone terribly wrong.
1602 		 */
1603 		if (blk_rq_bytes(req) < blk_rq_cur_bytes(req)) {
1604 			blk_dump_rq_flags(req, "request botched");
1605 			req->__data_len = blk_rq_cur_bytes(req);
1606 		}
1607 
1608 		/* recalculate the number of segments */
1609 		req->nr_phys_segments = blk_recalc_rq_segments(req);
1610 	}
1611 
1612 	return true;
1613 }
1614 EXPORT_SYMBOL_GPL(blk_update_request);
1615 
1616 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
1617 /**
1618  * rq_flush_dcache_pages - Helper function to flush all pages in a request
1619  * @rq: the request to be flushed
1620  *
1621  * Description:
1622  *     Flush all pages in @rq.
1623  */
1624 void rq_flush_dcache_pages(struct request *rq)
1625 {
1626 	struct req_iterator iter;
1627 	struct bio_vec bvec;
1628 
1629 	rq_for_each_segment(bvec, rq, iter)
1630 		flush_dcache_page(bvec.bv_page);
1631 }
1632 EXPORT_SYMBOL_GPL(rq_flush_dcache_pages);
1633 #endif
1634 
1635 /**
1636  * blk_lld_busy - Check if underlying low-level drivers of a device are busy
1637  * @q : the queue of the device being checked
1638  *
1639  * Description:
1640  *    Check if underlying low-level drivers of a device are busy.
1641  *    If the drivers want to export their busy state, they must set own
1642  *    exporting function using blk_queue_lld_busy() first.
1643  *
1644  *    Basically, this function is used only by request stacking drivers
1645  *    to stop dispatching requests to underlying devices when underlying
1646  *    devices are busy.  This behavior helps more I/O merging on the queue
1647  *    of the request stacking driver and prevents I/O throughput regression
1648  *    on burst I/O load.
1649  *
1650  * Return:
1651  *    0 - Not busy (The request stacking driver should dispatch request)
1652  *    1 - Busy (The request stacking driver should stop dispatching request)
1653  */
1654 int blk_lld_busy(struct request_queue *q)
1655 {
1656 	if (queue_is_mq(q) && q->mq_ops->busy)
1657 		return q->mq_ops->busy(q);
1658 
1659 	return 0;
1660 }
1661 EXPORT_SYMBOL_GPL(blk_lld_busy);
1662 
1663 /**
1664  * blk_rq_unprep_clone - Helper function to free all bios in a cloned request
1665  * @rq: the clone request to be cleaned up
1666  *
1667  * Description:
1668  *     Free all bios in @rq for a cloned request.
1669  */
1670 void blk_rq_unprep_clone(struct request *rq)
1671 {
1672 	struct bio *bio;
1673 
1674 	while ((bio = rq->bio) != NULL) {
1675 		rq->bio = bio->bi_next;
1676 
1677 		bio_put(bio);
1678 	}
1679 }
1680 EXPORT_SYMBOL_GPL(blk_rq_unprep_clone);
1681 
1682 /**
1683  * blk_rq_prep_clone - Helper function to setup clone request
1684  * @rq: the request to be setup
1685  * @rq_src: original request to be cloned
1686  * @bs: bio_set that bios for clone are allocated from
1687  * @gfp_mask: memory allocation mask for bio
1688  * @bio_ctr: setup function to be called for each clone bio.
1689  *           Returns %0 for success, non %0 for failure.
1690  * @data: private data to be passed to @bio_ctr
1691  *
1692  * Description:
1693  *     Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq.
1694  *     Also, pages which the original bios are pointing to are not copied
1695  *     and the cloned bios just point same pages.
1696  *     So cloned bios must be completed before original bios, which means
1697  *     the caller must complete @rq before @rq_src.
1698  */
1699 int blk_rq_prep_clone(struct request *rq, struct request *rq_src,
1700 		      struct bio_set *bs, gfp_t gfp_mask,
1701 		      int (*bio_ctr)(struct bio *, struct bio *, void *),
1702 		      void *data)
1703 {
1704 	struct bio *bio, *bio_src;
1705 
1706 	if (!bs)
1707 		bs = &fs_bio_set;
1708 
1709 	__rq_for_each_bio(bio_src, rq_src) {
1710 		bio = bio_clone_fast(bio_src, gfp_mask, bs);
1711 		if (!bio)
1712 			goto free_and_out;
1713 
1714 		if (bio_ctr && bio_ctr(bio, bio_src, data))
1715 			goto free_and_out;
1716 
1717 		if (rq->bio) {
1718 			rq->biotail->bi_next = bio;
1719 			rq->biotail = bio;
1720 		} else
1721 			rq->bio = rq->biotail = bio;
1722 	}
1723 
1724 	/* Copy attributes of the original request to the clone request. */
1725 	rq->__sector = blk_rq_pos(rq_src);
1726 	rq->__data_len = blk_rq_bytes(rq_src);
1727 	if (rq_src->rq_flags & RQF_SPECIAL_PAYLOAD) {
1728 		rq->rq_flags |= RQF_SPECIAL_PAYLOAD;
1729 		rq->special_vec = rq_src->special_vec;
1730 	}
1731 	rq->nr_phys_segments = rq_src->nr_phys_segments;
1732 	rq->ioprio = rq_src->ioprio;
1733 
1734 	if (rq->bio)
1735 		blk_crypto_rq_bio_prep(rq, rq->bio, gfp_mask);
1736 
1737 	return 0;
1738 
1739 free_and_out:
1740 	if (bio)
1741 		bio_put(bio);
1742 	blk_rq_unprep_clone(rq);
1743 
1744 	return -ENOMEM;
1745 }
1746 EXPORT_SYMBOL_GPL(blk_rq_prep_clone);
1747 
1748 int kblockd_schedule_work(struct work_struct *work)
1749 {
1750 	return queue_work(kblockd_workqueue, work);
1751 }
1752 EXPORT_SYMBOL(kblockd_schedule_work);
1753 
1754 int kblockd_mod_delayed_work_on(int cpu, struct delayed_work *dwork,
1755 				unsigned long delay)
1756 {
1757 	return mod_delayed_work_on(cpu, kblockd_workqueue, dwork, delay);
1758 }
1759 EXPORT_SYMBOL(kblockd_mod_delayed_work_on);
1760 
1761 /**
1762  * blk_start_plug - initialize blk_plug and track it inside the task_struct
1763  * @plug:	The &struct blk_plug that needs to be initialized
1764  *
1765  * Description:
1766  *   blk_start_plug() indicates to the block layer an intent by the caller
1767  *   to submit multiple I/O requests in a batch.  The block layer may use
1768  *   this hint to defer submitting I/Os from the caller until blk_finish_plug()
1769  *   is called.  However, the block layer may choose to submit requests
1770  *   before a call to blk_finish_plug() if the number of queued I/Os
1771  *   exceeds %BLK_MAX_REQUEST_COUNT, or if the size of the I/O is larger than
1772  *   %BLK_PLUG_FLUSH_SIZE.  The queued I/Os may also be submitted early if
1773  *   the task schedules (see below).
1774  *
1775  *   Tracking blk_plug inside the task_struct will help with auto-flushing the
1776  *   pending I/O should the task end up blocking between blk_start_plug() and
1777  *   blk_finish_plug(). This is important from a performance perspective, but
1778  *   also ensures that we don't deadlock. For instance, if the task is blocking
1779  *   for a memory allocation, memory reclaim could end up wanting to free a
1780  *   page belonging to that request that is currently residing in our private
1781  *   plug. By flushing the pending I/O when the process goes to sleep, we avoid
1782  *   this kind of deadlock.
1783  */
1784 void blk_start_plug(struct blk_plug *plug)
1785 {
1786 	struct task_struct *tsk = current;
1787 
1788 	/*
1789 	 * If this is a nested plug, don't actually assign it.
1790 	 */
1791 	if (tsk->plug)
1792 		return;
1793 
1794 	INIT_LIST_HEAD(&plug->mq_list);
1795 	INIT_LIST_HEAD(&plug->cb_list);
1796 	plug->rq_count = 0;
1797 	plug->multiple_queues = false;
1798 
1799 	/*
1800 	 * Store ordering should not be needed here, since a potential
1801 	 * preempt will imply a full memory barrier
1802 	 */
1803 	tsk->plug = plug;
1804 }
1805 EXPORT_SYMBOL(blk_start_plug);
1806 
1807 static void flush_plug_callbacks(struct blk_plug *plug, bool from_schedule)
1808 {
1809 	LIST_HEAD(callbacks);
1810 
1811 	while (!list_empty(&plug->cb_list)) {
1812 		list_splice_init(&plug->cb_list, &callbacks);
1813 
1814 		while (!list_empty(&callbacks)) {
1815 			struct blk_plug_cb *cb = list_first_entry(&callbacks,
1816 							  struct blk_plug_cb,
1817 							  list);
1818 			list_del(&cb->list);
1819 			cb->callback(cb, from_schedule);
1820 		}
1821 	}
1822 }
1823 
1824 struct blk_plug_cb *blk_check_plugged(blk_plug_cb_fn unplug, void *data,
1825 				      int size)
1826 {
1827 	struct blk_plug *plug = current->plug;
1828 	struct blk_plug_cb *cb;
1829 
1830 	if (!plug)
1831 		return NULL;
1832 
1833 	list_for_each_entry(cb, &plug->cb_list, list)
1834 		if (cb->callback == unplug && cb->data == data)
1835 			return cb;
1836 
1837 	/* Not currently on the callback list */
1838 	BUG_ON(size < sizeof(*cb));
1839 	cb = kzalloc(size, GFP_ATOMIC);
1840 	if (cb) {
1841 		cb->data = data;
1842 		cb->callback = unplug;
1843 		list_add(&cb->list, &plug->cb_list);
1844 	}
1845 	return cb;
1846 }
1847 EXPORT_SYMBOL(blk_check_plugged);
1848 
1849 void blk_flush_plug_list(struct blk_plug *plug, bool from_schedule)
1850 {
1851 	flush_plug_callbacks(plug, from_schedule);
1852 
1853 	if (!list_empty(&plug->mq_list))
1854 		blk_mq_flush_plug_list(plug, from_schedule);
1855 }
1856 
1857 /**
1858  * blk_finish_plug - mark the end of a batch of submitted I/O
1859  * @plug:	The &struct blk_plug passed to blk_start_plug()
1860  *
1861  * Description:
1862  * Indicate that a batch of I/O submissions is complete.  This function
1863  * must be paired with an initial call to blk_start_plug().  The intent
1864  * is to allow the block layer to optimize I/O submission.  See the
1865  * documentation for blk_start_plug() for more information.
1866  */
1867 void blk_finish_plug(struct blk_plug *plug)
1868 {
1869 	if (plug != current->plug)
1870 		return;
1871 	blk_flush_plug_list(plug, false);
1872 
1873 	current->plug = NULL;
1874 }
1875 EXPORT_SYMBOL(blk_finish_plug);
1876 
1877 void blk_io_schedule(void)
1878 {
1879 	/* Prevent hang_check timer from firing at us during very long I/O */
1880 	unsigned long timeout = sysctl_hung_task_timeout_secs * HZ / 2;
1881 
1882 	if (timeout)
1883 		io_schedule_timeout(timeout);
1884 	else
1885 		io_schedule();
1886 }
1887 EXPORT_SYMBOL_GPL(blk_io_schedule);
1888 
1889 int __init blk_dev_init(void)
1890 {
1891 	BUILD_BUG_ON(REQ_OP_LAST >= (1 << REQ_OP_BITS));
1892 	BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 *
1893 			sizeof_field(struct request, cmd_flags));
1894 	BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 *
1895 			sizeof_field(struct bio, bi_opf));
1896 
1897 	/* used for unplugging and affects IO latency/throughput - HIGHPRI */
1898 	kblockd_workqueue = alloc_workqueue("kblockd",
1899 					    WQ_MEM_RECLAIM | WQ_HIGHPRI, 0);
1900 	if (!kblockd_workqueue)
1901 		panic("Failed to create kblockd\n");
1902 
1903 	blk_requestq_cachep = kmem_cache_create("request_queue",
1904 			sizeof(struct request_queue), 0, SLAB_PANIC, NULL);
1905 
1906 	blk_debugfs_root = debugfs_create_dir("block", NULL);
1907 
1908 	return 0;
1909 }
1910