xref: /openbmc/linux/block/blk-core.c (revision 37744fee)
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/t10-pi.h>
38 #include <linux/debugfs.h>
39 #include <linux/bpf.h>
40 #include <linux/psi.h>
41 
42 #define CREATE_TRACE_POINTS
43 #include <trace/events/block.h>
44 
45 #include "blk.h"
46 #include "blk-mq.h"
47 #include "blk-mq-sched.h"
48 #include "blk-pm.h"
49 #include "blk-rq-qos.h"
50 
51 #ifdef CONFIG_DEBUG_FS
52 struct dentry *blk_debugfs_root;
53 #endif
54 
55 EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_remap);
56 EXPORT_TRACEPOINT_SYMBOL_GPL(block_rq_remap);
57 EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_complete);
58 EXPORT_TRACEPOINT_SYMBOL_GPL(block_split);
59 EXPORT_TRACEPOINT_SYMBOL_GPL(block_unplug);
60 
61 DEFINE_IDA(blk_queue_ida);
62 
63 /*
64  * For queue allocation
65  */
66 struct kmem_cache *blk_requestq_cachep;
67 
68 /*
69  * Controlling structure to kblockd
70  */
71 static struct workqueue_struct *kblockd_workqueue;
72 
73 /**
74  * blk_queue_flag_set - atomically set a queue flag
75  * @flag: flag to be set
76  * @q: request queue
77  */
78 void blk_queue_flag_set(unsigned int flag, struct request_queue *q)
79 {
80 	set_bit(flag, &q->queue_flags);
81 }
82 EXPORT_SYMBOL(blk_queue_flag_set);
83 
84 /**
85  * blk_queue_flag_clear - atomically clear a queue flag
86  * @flag: flag to be cleared
87  * @q: request queue
88  */
89 void blk_queue_flag_clear(unsigned int flag, struct request_queue *q)
90 {
91 	clear_bit(flag, &q->queue_flags);
92 }
93 EXPORT_SYMBOL(blk_queue_flag_clear);
94 
95 /**
96  * blk_queue_flag_test_and_set - atomically test and set a queue flag
97  * @flag: flag to be set
98  * @q: request queue
99  *
100  * Returns the previous value of @flag - 0 if the flag was not set and 1 if
101  * the flag was already set.
102  */
103 bool blk_queue_flag_test_and_set(unsigned int flag, struct request_queue *q)
104 {
105 	return test_and_set_bit(flag, &q->queue_flags);
106 }
107 EXPORT_SYMBOL_GPL(blk_queue_flag_test_and_set);
108 
109 void blk_rq_init(struct request_queue *q, struct request *rq)
110 {
111 	memset(rq, 0, sizeof(*rq));
112 
113 	INIT_LIST_HEAD(&rq->queuelist);
114 	rq->q = q;
115 	rq->__sector = (sector_t) -1;
116 	INIT_HLIST_NODE(&rq->hash);
117 	RB_CLEAR_NODE(&rq->rb_node);
118 	rq->tag = -1;
119 	rq->internal_tag = -1;
120 	rq->start_time_ns = ktime_get_ns();
121 	rq->part = NULL;
122 	refcount_set(&rq->ref, 1);
123 }
124 EXPORT_SYMBOL(blk_rq_init);
125 
126 #define REQ_OP_NAME(name) [REQ_OP_##name] = #name
127 static const char *const blk_op_name[] = {
128 	REQ_OP_NAME(READ),
129 	REQ_OP_NAME(WRITE),
130 	REQ_OP_NAME(FLUSH),
131 	REQ_OP_NAME(DISCARD),
132 	REQ_OP_NAME(SECURE_ERASE),
133 	REQ_OP_NAME(ZONE_RESET),
134 	REQ_OP_NAME(ZONE_RESET_ALL),
135 	REQ_OP_NAME(ZONE_OPEN),
136 	REQ_OP_NAME(ZONE_CLOSE),
137 	REQ_OP_NAME(ZONE_FINISH),
138 	REQ_OP_NAME(WRITE_SAME),
139 	REQ_OP_NAME(WRITE_ZEROES),
140 	REQ_OP_NAME(SCSI_IN),
141 	REQ_OP_NAME(SCSI_OUT),
142 	REQ_OP_NAME(DRV_IN),
143 	REQ_OP_NAME(DRV_OUT),
144 };
145 #undef REQ_OP_NAME
146 
147 /**
148  * blk_op_str - Return string XXX in the REQ_OP_XXX.
149  * @op: REQ_OP_XXX.
150  *
151  * Description: Centralize block layer function to convert REQ_OP_XXX into
152  * string format. Useful in the debugging and tracing bio or request. For
153  * invalid REQ_OP_XXX it returns string "UNKNOWN".
154  */
155 inline const char *blk_op_str(unsigned int op)
156 {
157 	const char *op_str = "UNKNOWN";
158 
159 	if (op < ARRAY_SIZE(blk_op_name) && blk_op_name[op])
160 		op_str = blk_op_name[op];
161 
162 	return op_str;
163 }
164 EXPORT_SYMBOL_GPL(blk_op_str);
165 
166 static const struct {
167 	int		errno;
168 	const char	*name;
169 } blk_errors[] = {
170 	[BLK_STS_OK]		= { 0,		"" },
171 	[BLK_STS_NOTSUPP]	= { -EOPNOTSUPP, "operation not supported" },
172 	[BLK_STS_TIMEOUT]	= { -ETIMEDOUT,	"timeout" },
173 	[BLK_STS_NOSPC]		= { -ENOSPC,	"critical space allocation" },
174 	[BLK_STS_TRANSPORT]	= { -ENOLINK,	"recoverable transport" },
175 	[BLK_STS_TARGET]	= { -EREMOTEIO,	"critical target" },
176 	[BLK_STS_NEXUS]		= { -EBADE,	"critical nexus" },
177 	[BLK_STS_MEDIUM]	= { -ENODATA,	"critical medium" },
178 	[BLK_STS_PROTECTION]	= { -EILSEQ,	"protection" },
179 	[BLK_STS_RESOURCE]	= { -ENOMEM,	"kernel resource" },
180 	[BLK_STS_DEV_RESOURCE]	= { -EBUSY,	"device resource" },
181 	[BLK_STS_AGAIN]		= { -EAGAIN,	"nonblocking retry" },
182 
183 	/* device mapper special case, should not leak out: */
184 	[BLK_STS_DM_REQUEUE]	= { -EREMCHG, "dm internal retry" },
185 
186 	/* everything else not covered above: */
187 	[BLK_STS_IOERR]		= { -EIO,	"I/O" },
188 };
189 
190 blk_status_t errno_to_blk_status(int errno)
191 {
192 	int i;
193 
194 	for (i = 0; i < ARRAY_SIZE(blk_errors); i++) {
195 		if (blk_errors[i].errno == errno)
196 			return (__force blk_status_t)i;
197 	}
198 
199 	return BLK_STS_IOERR;
200 }
201 EXPORT_SYMBOL_GPL(errno_to_blk_status);
202 
203 int blk_status_to_errno(blk_status_t status)
204 {
205 	int idx = (__force int)status;
206 
207 	if (WARN_ON_ONCE(idx >= ARRAY_SIZE(blk_errors)))
208 		return -EIO;
209 	return blk_errors[idx].errno;
210 }
211 EXPORT_SYMBOL_GPL(blk_status_to_errno);
212 
213 static void print_req_error(struct request *req, blk_status_t status,
214 		const char *caller)
215 {
216 	int idx = (__force int)status;
217 
218 	if (WARN_ON_ONCE(idx >= ARRAY_SIZE(blk_errors)))
219 		return;
220 
221 	printk_ratelimited(KERN_ERR
222 		"%s: %s error, dev %s, sector %llu op 0x%x:(%s) flags 0x%x "
223 		"phys_seg %u prio class %u\n",
224 		caller, blk_errors[idx].name,
225 		req->rq_disk ? req->rq_disk->disk_name : "?",
226 		blk_rq_pos(req), req_op(req), blk_op_str(req_op(req)),
227 		req->cmd_flags & ~REQ_OP_MASK,
228 		req->nr_phys_segments,
229 		IOPRIO_PRIO_CLASS(req->ioprio));
230 }
231 
232 static void req_bio_endio(struct request *rq, struct bio *bio,
233 			  unsigned int nbytes, blk_status_t error)
234 {
235 	if (error)
236 		bio->bi_status = error;
237 
238 	if (unlikely(rq->rq_flags & RQF_QUIET))
239 		bio_set_flag(bio, BIO_QUIET);
240 
241 	bio_advance(bio, nbytes);
242 
243 	/* don't actually finish bio if it's part of flush sequence */
244 	if (bio->bi_iter.bi_size == 0 && !(rq->rq_flags & RQF_FLUSH_SEQ))
245 		bio_endio(bio);
246 }
247 
248 void blk_dump_rq_flags(struct request *rq, char *msg)
249 {
250 	printk(KERN_INFO "%s: dev %s: flags=%llx\n", msg,
251 		rq->rq_disk ? rq->rq_disk->disk_name : "?",
252 		(unsigned long long) rq->cmd_flags);
253 
254 	printk(KERN_INFO "  sector %llu, nr/cnr %u/%u\n",
255 	       (unsigned long long)blk_rq_pos(rq),
256 	       blk_rq_sectors(rq), blk_rq_cur_sectors(rq));
257 	printk(KERN_INFO "  bio %p, biotail %p, len %u\n",
258 	       rq->bio, rq->biotail, blk_rq_bytes(rq));
259 }
260 EXPORT_SYMBOL(blk_dump_rq_flags);
261 
262 /**
263  * blk_sync_queue - cancel any pending callbacks on a queue
264  * @q: the queue
265  *
266  * Description:
267  *     The block layer may perform asynchronous callback activity
268  *     on a queue, such as calling the unplug function after a timeout.
269  *     A block device may call blk_sync_queue to ensure that any
270  *     such activity is cancelled, thus allowing it to release resources
271  *     that the callbacks might use. The caller must already have made sure
272  *     that its ->make_request_fn will not re-add plugging prior to calling
273  *     this function.
274  *
275  *     This function does not cancel any asynchronous activity arising
276  *     out of elevator or throttling code. That would require elevator_exit()
277  *     and blkcg_exit_queue() to be called with queue lock initialized.
278  *
279  */
280 void blk_sync_queue(struct request_queue *q)
281 {
282 	del_timer_sync(&q->timeout);
283 	cancel_work_sync(&q->timeout_work);
284 }
285 EXPORT_SYMBOL(blk_sync_queue);
286 
287 /**
288  * blk_set_pm_only - increment pm_only counter
289  * @q: request queue pointer
290  */
291 void blk_set_pm_only(struct request_queue *q)
292 {
293 	atomic_inc(&q->pm_only);
294 }
295 EXPORT_SYMBOL_GPL(blk_set_pm_only);
296 
297 void blk_clear_pm_only(struct request_queue *q)
298 {
299 	int pm_only;
300 
301 	pm_only = atomic_dec_return(&q->pm_only);
302 	WARN_ON_ONCE(pm_only < 0);
303 	if (pm_only == 0)
304 		wake_up_all(&q->mq_freeze_wq);
305 }
306 EXPORT_SYMBOL_GPL(blk_clear_pm_only);
307 
308 void blk_put_queue(struct request_queue *q)
309 {
310 	kobject_put(&q->kobj);
311 }
312 EXPORT_SYMBOL(blk_put_queue);
313 
314 void blk_set_queue_dying(struct request_queue *q)
315 {
316 	blk_queue_flag_set(QUEUE_FLAG_DYING, q);
317 
318 	/*
319 	 * When queue DYING flag is set, we need to block new req
320 	 * entering queue, so we call blk_freeze_queue_start() to
321 	 * prevent I/O from crossing blk_queue_enter().
322 	 */
323 	blk_freeze_queue_start(q);
324 
325 	if (queue_is_mq(q))
326 		blk_mq_wake_waiters(q);
327 
328 	/* Make blk_queue_enter() reexamine the DYING flag. */
329 	wake_up_all(&q->mq_freeze_wq);
330 }
331 EXPORT_SYMBOL_GPL(blk_set_queue_dying);
332 
333 /**
334  * blk_cleanup_queue - shutdown a request queue
335  * @q: request queue to shutdown
336  *
337  * Mark @q DYING, drain all pending requests, mark @q DEAD, destroy and
338  * put it.  All future requests will be failed immediately with -ENODEV.
339  */
340 void blk_cleanup_queue(struct request_queue *q)
341 {
342 	WARN_ON_ONCE(blk_queue_registered(q));
343 
344 	/* mark @q DYING, no new request or merges will be allowed afterwards */
345 	blk_set_queue_dying(q);
346 
347 	blk_queue_flag_set(QUEUE_FLAG_NOMERGES, q);
348 	blk_queue_flag_set(QUEUE_FLAG_NOXMERGES, q);
349 
350 	/*
351 	 * Drain all requests queued before DYING marking. Set DEAD flag to
352 	 * prevent that blk_mq_run_hw_queues() accesses the hardware queues
353 	 * after draining finished.
354 	 */
355 	blk_freeze_queue(q);
356 
357 	rq_qos_exit(q);
358 
359 	blk_queue_flag_set(QUEUE_FLAG_DEAD, q);
360 
361 	/* for synchronous bio-based driver finish in-flight integrity i/o */
362 	blk_flush_integrity();
363 
364 	/* @q won't process any more request, flush async actions */
365 	del_timer_sync(&q->backing_dev_info->laptop_mode_wb_timer);
366 	blk_sync_queue(q);
367 
368 	if (queue_is_mq(q))
369 		blk_mq_exit_queue(q);
370 
371 	/*
372 	 * In theory, request pool of sched_tags belongs to request queue.
373 	 * However, the current implementation requires tag_set for freeing
374 	 * requests, so free the pool now.
375 	 *
376 	 * Queue has become frozen, there can't be any in-queue requests, so
377 	 * it is safe to free requests now.
378 	 */
379 	mutex_lock(&q->sysfs_lock);
380 	if (q->elevator)
381 		blk_mq_sched_free_requests(q);
382 	mutex_unlock(&q->sysfs_lock);
383 
384 	percpu_ref_exit(&q->q_usage_counter);
385 
386 	/* @q is and will stay empty, shutdown and put */
387 	blk_put_queue(q);
388 }
389 EXPORT_SYMBOL(blk_cleanup_queue);
390 
391 /**
392  * blk_queue_enter() - try to increase q->q_usage_counter
393  * @q: request queue pointer
394  * @flags: BLK_MQ_REQ_NOWAIT and/or BLK_MQ_REQ_PREEMPT
395  */
396 int blk_queue_enter(struct request_queue *q, blk_mq_req_flags_t flags)
397 {
398 	const bool pm = flags & BLK_MQ_REQ_PREEMPT;
399 
400 	while (true) {
401 		bool success = false;
402 
403 		rcu_read_lock();
404 		if (percpu_ref_tryget_live(&q->q_usage_counter)) {
405 			/*
406 			 * The code that increments the pm_only counter is
407 			 * responsible for ensuring that that counter is
408 			 * globally visible before the queue is unfrozen.
409 			 */
410 			if (pm || !blk_queue_pm_only(q)) {
411 				success = true;
412 			} else {
413 				percpu_ref_put(&q->q_usage_counter);
414 			}
415 		}
416 		rcu_read_unlock();
417 
418 		if (success)
419 			return 0;
420 
421 		if (flags & BLK_MQ_REQ_NOWAIT)
422 			return -EBUSY;
423 
424 		/*
425 		 * read pair of barrier in blk_freeze_queue_start(),
426 		 * we need to order reading __PERCPU_REF_DEAD flag of
427 		 * .q_usage_counter and reading .mq_freeze_depth or
428 		 * queue dying flag, otherwise the following wait may
429 		 * never return if the two reads are reordered.
430 		 */
431 		smp_rmb();
432 
433 		wait_event(q->mq_freeze_wq,
434 			   (!q->mq_freeze_depth &&
435 			    (pm || (blk_pm_request_resume(q),
436 				    !blk_queue_pm_only(q)))) ||
437 			   blk_queue_dying(q));
438 		if (blk_queue_dying(q))
439 			return -ENODEV;
440 	}
441 }
442 
443 void blk_queue_exit(struct request_queue *q)
444 {
445 	percpu_ref_put(&q->q_usage_counter);
446 }
447 
448 static void blk_queue_usage_counter_release(struct percpu_ref *ref)
449 {
450 	struct request_queue *q =
451 		container_of(ref, struct request_queue, q_usage_counter);
452 
453 	wake_up_all(&q->mq_freeze_wq);
454 }
455 
456 static void blk_rq_timed_out_timer(struct timer_list *t)
457 {
458 	struct request_queue *q = from_timer(q, t, timeout);
459 
460 	kblockd_schedule_work(&q->timeout_work);
461 }
462 
463 static void blk_timeout_work(struct work_struct *work)
464 {
465 }
466 
467 struct request_queue *__blk_alloc_queue(int node_id)
468 {
469 	struct request_queue *q;
470 	int ret;
471 
472 	q = kmem_cache_alloc_node(blk_requestq_cachep,
473 				GFP_KERNEL | __GFP_ZERO, node_id);
474 	if (!q)
475 		return NULL;
476 
477 	q->last_merge = NULL;
478 
479 	q->id = ida_simple_get(&blk_queue_ida, 0, 0, GFP_KERNEL);
480 	if (q->id < 0)
481 		goto fail_q;
482 
483 	ret = bioset_init(&q->bio_split, BIO_POOL_SIZE, 0, BIOSET_NEED_BVECS);
484 	if (ret)
485 		goto fail_id;
486 
487 	q->backing_dev_info = bdi_alloc_node(GFP_KERNEL, node_id);
488 	if (!q->backing_dev_info)
489 		goto fail_split;
490 
491 	q->stats = blk_alloc_queue_stats();
492 	if (!q->stats)
493 		goto fail_stats;
494 
495 	q->backing_dev_info->ra_pages = VM_READAHEAD_PAGES;
496 	q->backing_dev_info->capabilities = BDI_CAP_CGROUP_WRITEBACK;
497 	q->backing_dev_info->name = "block";
498 	q->node = node_id;
499 
500 	timer_setup(&q->backing_dev_info->laptop_mode_wb_timer,
501 		    laptop_mode_timer_fn, 0);
502 	timer_setup(&q->timeout, blk_rq_timed_out_timer, 0);
503 	INIT_WORK(&q->timeout_work, blk_timeout_work);
504 	INIT_LIST_HEAD(&q->icq_list);
505 #ifdef CONFIG_BLK_CGROUP
506 	INIT_LIST_HEAD(&q->blkg_list);
507 #endif
508 
509 	kobject_init(&q->kobj, &blk_queue_ktype);
510 
511 #ifdef CONFIG_BLK_DEV_IO_TRACE
512 	mutex_init(&q->blk_trace_mutex);
513 #endif
514 	mutex_init(&q->sysfs_lock);
515 	mutex_init(&q->sysfs_dir_lock);
516 	spin_lock_init(&q->queue_lock);
517 
518 	init_waitqueue_head(&q->mq_freeze_wq);
519 	mutex_init(&q->mq_freeze_lock);
520 
521 	/*
522 	 * Init percpu_ref in atomic mode so that it's faster to shutdown.
523 	 * See blk_register_queue() for details.
524 	 */
525 	if (percpu_ref_init(&q->q_usage_counter,
526 				blk_queue_usage_counter_release,
527 				PERCPU_REF_INIT_ATOMIC, GFP_KERNEL))
528 		goto fail_bdi;
529 
530 	if (blkcg_init_queue(q))
531 		goto fail_ref;
532 
533 	blk_queue_dma_alignment(q, 511);
534 	blk_set_default_limits(&q->limits);
535 
536 	return q;
537 
538 fail_ref:
539 	percpu_ref_exit(&q->q_usage_counter);
540 fail_bdi:
541 	blk_free_queue_stats(q->stats);
542 fail_stats:
543 	bdi_put(q->backing_dev_info);
544 fail_split:
545 	bioset_exit(&q->bio_split);
546 fail_id:
547 	ida_simple_remove(&blk_queue_ida, q->id);
548 fail_q:
549 	kmem_cache_free(blk_requestq_cachep, q);
550 	return NULL;
551 }
552 
553 struct request_queue *blk_alloc_queue(make_request_fn make_request, int node_id)
554 {
555 	struct request_queue *q;
556 
557 	if (WARN_ON_ONCE(!make_request))
558 		return NULL;
559 
560 	q = __blk_alloc_queue(node_id);
561 	if (!q)
562 		return NULL;
563 	q->make_request_fn = make_request;
564 	q->nr_requests = BLKDEV_MAX_RQ;
565 	return q;
566 }
567 EXPORT_SYMBOL(blk_alloc_queue);
568 
569 bool blk_get_queue(struct request_queue *q)
570 {
571 	if (likely(!blk_queue_dying(q))) {
572 		__blk_get_queue(q);
573 		return true;
574 	}
575 
576 	return false;
577 }
578 EXPORT_SYMBOL(blk_get_queue);
579 
580 /**
581  * blk_get_request - allocate a request
582  * @q: request queue to allocate a request for
583  * @op: operation (REQ_OP_*) and REQ_* flags, e.g. REQ_SYNC.
584  * @flags: BLK_MQ_REQ_* flags, e.g. BLK_MQ_REQ_NOWAIT.
585  */
586 struct request *blk_get_request(struct request_queue *q, unsigned int op,
587 				blk_mq_req_flags_t flags)
588 {
589 	struct request *req;
590 
591 	WARN_ON_ONCE(op & REQ_NOWAIT);
592 	WARN_ON_ONCE(flags & ~(BLK_MQ_REQ_NOWAIT | BLK_MQ_REQ_PREEMPT));
593 
594 	req = blk_mq_alloc_request(q, op, flags);
595 	if (!IS_ERR(req) && q->mq_ops->initialize_rq_fn)
596 		q->mq_ops->initialize_rq_fn(req);
597 
598 	return req;
599 }
600 EXPORT_SYMBOL(blk_get_request);
601 
602 void blk_put_request(struct request *req)
603 {
604 	blk_mq_free_request(req);
605 }
606 EXPORT_SYMBOL(blk_put_request);
607 
608 bool bio_attempt_back_merge(struct request *req, struct bio *bio,
609 		unsigned int nr_segs)
610 {
611 	const int ff = bio->bi_opf & REQ_FAILFAST_MASK;
612 
613 	if (!ll_back_merge_fn(req, bio, nr_segs))
614 		return false;
615 
616 	trace_block_bio_backmerge(req->q, req, bio);
617 	rq_qos_merge(req->q, req, bio);
618 
619 	if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
620 		blk_rq_set_mixed_merge(req);
621 
622 	req->biotail->bi_next = bio;
623 	req->biotail = bio;
624 	req->__data_len += bio->bi_iter.bi_size;
625 
626 	blk_account_io_start(req, false);
627 	return true;
628 }
629 
630 bool bio_attempt_front_merge(struct request *req, struct bio *bio,
631 		unsigned int nr_segs)
632 {
633 	const int ff = bio->bi_opf & REQ_FAILFAST_MASK;
634 
635 	if (!ll_front_merge_fn(req, bio, nr_segs))
636 		return false;
637 
638 	trace_block_bio_frontmerge(req->q, req, bio);
639 	rq_qos_merge(req->q, req, bio);
640 
641 	if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
642 		blk_rq_set_mixed_merge(req);
643 
644 	bio->bi_next = req->bio;
645 	req->bio = bio;
646 
647 	req->__sector = bio->bi_iter.bi_sector;
648 	req->__data_len += bio->bi_iter.bi_size;
649 
650 	blk_account_io_start(req, false);
651 	return true;
652 }
653 
654 bool bio_attempt_discard_merge(struct request_queue *q, struct request *req,
655 		struct bio *bio)
656 {
657 	unsigned short segments = blk_rq_nr_discard_segments(req);
658 
659 	if (segments >= queue_max_discard_segments(q))
660 		goto no_merge;
661 	if (blk_rq_sectors(req) + bio_sectors(bio) >
662 	    blk_rq_get_max_sectors(req, blk_rq_pos(req)))
663 		goto no_merge;
664 
665 	rq_qos_merge(q, req, bio);
666 
667 	req->biotail->bi_next = bio;
668 	req->biotail = bio;
669 	req->__data_len += bio->bi_iter.bi_size;
670 	req->nr_phys_segments = segments + 1;
671 
672 	blk_account_io_start(req, false);
673 	return true;
674 no_merge:
675 	req_set_nomerge(q, req);
676 	return false;
677 }
678 
679 /**
680  * blk_attempt_plug_merge - try to merge with %current's plugged list
681  * @q: request_queue new bio is being queued at
682  * @bio: new bio being queued
683  * @nr_segs: number of segments in @bio
684  * @same_queue_rq: pointer to &struct request that gets filled in when
685  * another request associated with @q is found on the plug list
686  * (optional, may be %NULL)
687  *
688  * Determine whether @bio being queued on @q can be merged with a request
689  * on %current's plugged list.  Returns %true if merge was successful,
690  * otherwise %false.
691  *
692  * Plugging coalesces IOs from the same issuer for the same purpose without
693  * going through @q->queue_lock.  As such it's more of an issuing mechanism
694  * than scheduling, and the request, while may have elvpriv data, is not
695  * added on the elevator at this point.  In addition, we don't have
696  * reliable access to the elevator outside queue lock.  Only check basic
697  * merging parameters without querying the elevator.
698  *
699  * Caller must ensure !blk_queue_nomerges(q) beforehand.
700  */
701 bool blk_attempt_plug_merge(struct request_queue *q, struct bio *bio,
702 		unsigned int nr_segs, struct request **same_queue_rq)
703 {
704 	struct blk_plug *plug;
705 	struct request *rq;
706 	struct list_head *plug_list;
707 
708 	plug = blk_mq_plug(q, bio);
709 	if (!plug)
710 		return false;
711 
712 	plug_list = &plug->mq_list;
713 
714 	list_for_each_entry_reverse(rq, plug_list, queuelist) {
715 		bool merged = false;
716 
717 		if (rq->q == q && same_queue_rq) {
718 			/*
719 			 * Only blk-mq multiple hardware queues case checks the
720 			 * rq in the same queue, there should be only one such
721 			 * rq in a queue
722 			 **/
723 			*same_queue_rq = rq;
724 		}
725 
726 		if (rq->q != q || !blk_rq_merge_ok(rq, bio))
727 			continue;
728 
729 		switch (blk_try_merge(rq, bio)) {
730 		case ELEVATOR_BACK_MERGE:
731 			merged = bio_attempt_back_merge(rq, bio, nr_segs);
732 			break;
733 		case ELEVATOR_FRONT_MERGE:
734 			merged = bio_attempt_front_merge(rq, bio, nr_segs);
735 			break;
736 		case ELEVATOR_DISCARD_MERGE:
737 			merged = bio_attempt_discard_merge(q, rq, bio);
738 			break;
739 		default:
740 			break;
741 		}
742 
743 		if (merged)
744 			return true;
745 	}
746 
747 	return false;
748 }
749 
750 static void handle_bad_sector(struct bio *bio, sector_t maxsector)
751 {
752 	char b[BDEVNAME_SIZE];
753 
754 	printk(KERN_INFO "attempt to access beyond end of device\n");
755 	printk(KERN_INFO "%s: rw=%d, want=%Lu, limit=%Lu\n",
756 			bio_devname(bio, b), bio->bi_opf,
757 			(unsigned long long)bio_end_sector(bio),
758 			(long long)maxsector);
759 }
760 
761 #ifdef CONFIG_FAIL_MAKE_REQUEST
762 
763 static DECLARE_FAULT_ATTR(fail_make_request);
764 
765 static int __init setup_fail_make_request(char *str)
766 {
767 	return setup_fault_attr(&fail_make_request, str);
768 }
769 __setup("fail_make_request=", setup_fail_make_request);
770 
771 static bool should_fail_request(struct hd_struct *part, unsigned int bytes)
772 {
773 	return part->make_it_fail && should_fail(&fail_make_request, bytes);
774 }
775 
776 static int __init fail_make_request_debugfs(void)
777 {
778 	struct dentry *dir = fault_create_debugfs_attr("fail_make_request",
779 						NULL, &fail_make_request);
780 
781 	return PTR_ERR_OR_ZERO(dir);
782 }
783 
784 late_initcall(fail_make_request_debugfs);
785 
786 #else /* CONFIG_FAIL_MAKE_REQUEST */
787 
788 static inline bool should_fail_request(struct hd_struct *part,
789 					unsigned int bytes)
790 {
791 	return false;
792 }
793 
794 #endif /* CONFIG_FAIL_MAKE_REQUEST */
795 
796 static inline bool bio_check_ro(struct bio *bio, struct hd_struct *part)
797 {
798 	const int op = bio_op(bio);
799 
800 	if (part->policy && op_is_write(op)) {
801 		char b[BDEVNAME_SIZE];
802 
803 		if (op_is_flush(bio->bi_opf) && !bio_sectors(bio))
804 			return false;
805 
806 		WARN_ONCE(1,
807 		       "generic_make_request: Trying to write "
808 			"to read-only block-device %s (partno %d)\n",
809 			bio_devname(bio, b), part->partno);
810 		/* Older lvm-tools actually trigger this */
811 		return false;
812 	}
813 
814 	return false;
815 }
816 
817 static noinline int should_fail_bio(struct bio *bio)
818 {
819 	if (should_fail_request(&bio->bi_disk->part0, bio->bi_iter.bi_size))
820 		return -EIO;
821 	return 0;
822 }
823 ALLOW_ERROR_INJECTION(should_fail_bio, ERRNO);
824 
825 /*
826  * Check whether this bio extends beyond the end of the device or partition.
827  * This may well happen - the kernel calls bread() without checking the size of
828  * the device, e.g., when mounting a file system.
829  */
830 static inline int bio_check_eod(struct bio *bio, sector_t maxsector)
831 {
832 	unsigned int nr_sectors = bio_sectors(bio);
833 
834 	if (nr_sectors && maxsector &&
835 	    (nr_sectors > maxsector ||
836 	     bio->bi_iter.bi_sector > maxsector - nr_sectors)) {
837 		handle_bad_sector(bio, maxsector);
838 		return -EIO;
839 	}
840 	return 0;
841 }
842 
843 /*
844  * Remap block n of partition p to block n+start(p) of the disk.
845  */
846 static inline int blk_partition_remap(struct bio *bio)
847 {
848 	struct hd_struct *p;
849 	int ret = -EIO;
850 
851 	rcu_read_lock();
852 	p = __disk_get_part(bio->bi_disk, bio->bi_partno);
853 	if (unlikely(!p))
854 		goto out;
855 	if (unlikely(should_fail_request(p, bio->bi_iter.bi_size)))
856 		goto out;
857 	if (unlikely(bio_check_ro(bio, p)))
858 		goto out;
859 
860 	if (bio_sectors(bio)) {
861 		if (bio_check_eod(bio, part_nr_sects_read(p)))
862 			goto out;
863 		bio->bi_iter.bi_sector += p->start_sect;
864 		trace_block_bio_remap(bio->bi_disk->queue, bio, part_devt(p),
865 				      bio->bi_iter.bi_sector - p->start_sect);
866 	}
867 	bio->bi_partno = 0;
868 	ret = 0;
869 out:
870 	rcu_read_unlock();
871 	return ret;
872 }
873 
874 static noinline_for_stack bool
875 generic_make_request_checks(struct bio *bio)
876 {
877 	struct request_queue *q;
878 	int nr_sectors = bio_sectors(bio);
879 	blk_status_t status = BLK_STS_IOERR;
880 	char b[BDEVNAME_SIZE];
881 
882 	might_sleep();
883 
884 	q = bio->bi_disk->queue;
885 	if (unlikely(!q)) {
886 		printk(KERN_ERR
887 		       "generic_make_request: Trying to access "
888 			"nonexistent block-device %s (%Lu)\n",
889 			bio_devname(bio, b), (long long)bio->bi_iter.bi_sector);
890 		goto end_io;
891 	}
892 
893 	/*
894 	 * For a REQ_NOWAIT based request, return -EOPNOTSUPP
895 	 * if queue is not a request based queue.
896 	 */
897 	if ((bio->bi_opf & REQ_NOWAIT) && !queue_is_mq(q))
898 		goto not_supported;
899 
900 	if (should_fail_bio(bio))
901 		goto end_io;
902 
903 	if (bio->bi_partno) {
904 		if (unlikely(blk_partition_remap(bio)))
905 			goto end_io;
906 	} else {
907 		if (unlikely(bio_check_ro(bio, &bio->bi_disk->part0)))
908 			goto end_io;
909 		if (unlikely(bio_check_eod(bio, get_capacity(bio->bi_disk))))
910 			goto end_io;
911 	}
912 
913 	/*
914 	 * Filter flush bio's early so that make_request based
915 	 * drivers without flush support don't have to worry
916 	 * about them.
917 	 */
918 	if (op_is_flush(bio->bi_opf) &&
919 	    !test_bit(QUEUE_FLAG_WC, &q->queue_flags)) {
920 		bio->bi_opf &= ~(REQ_PREFLUSH | REQ_FUA);
921 		if (!nr_sectors) {
922 			status = BLK_STS_OK;
923 			goto end_io;
924 		}
925 	}
926 
927 	if (!test_bit(QUEUE_FLAG_POLL, &q->queue_flags))
928 		bio->bi_opf &= ~REQ_HIPRI;
929 
930 	switch (bio_op(bio)) {
931 	case REQ_OP_DISCARD:
932 		if (!blk_queue_discard(q))
933 			goto not_supported;
934 		break;
935 	case REQ_OP_SECURE_ERASE:
936 		if (!blk_queue_secure_erase(q))
937 			goto not_supported;
938 		break;
939 	case REQ_OP_WRITE_SAME:
940 		if (!q->limits.max_write_same_sectors)
941 			goto not_supported;
942 		break;
943 	case REQ_OP_ZONE_RESET:
944 	case REQ_OP_ZONE_OPEN:
945 	case REQ_OP_ZONE_CLOSE:
946 	case REQ_OP_ZONE_FINISH:
947 		if (!blk_queue_is_zoned(q))
948 			goto not_supported;
949 		break;
950 	case REQ_OP_ZONE_RESET_ALL:
951 		if (!blk_queue_is_zoned(q) || !blk_queue_zone_resetall(q))
952 			goto not_supported;
953 		break;
954 	case REQ_OP_WRITE_ZEROES:
955 		if (!q->limits.max_write_zeroes_sectors)
956 			goto not_supported;
957 		break;
958 	default:
959 		break;
960 	}
961 
962 	/*
963 	 * Various block parts want %current->io_context and lazy ioc
964 	 * allocation ends up trading a lot of pain for a small amount of
965 	 * memory.  Just allocate it upfront.  This may fail and block
966 	 * layer knows how to live with it.
967 	 */
968 	create_io_context(GFP_ATOMIC, q->node);
969 
970 	if (!blkcg_bio_issue_check(q, bio))
971 		return false;
972 
973 	if (!bio_flagged(bio, BIO_TRACE_COMPLETION)) {
974 		trace_block_bio_queue(q, bio);
975 		/* Now that enqueuing has been traced, we need to trace
976 		 * completion as well.
977 		 */
978 		bio_set_flag(bio, BIO_TRACE_COMPLETION);
979 	}
980 	return true;
981 
982 not_supported:
983 	status = BLK_STS_NOTSUPP;
984 end_io:
985 	bio->bi_status = status;
986 	bio_endio(bio);
987 	return false;
988 }
989 
990 /**
991  * generic_make_request - hand a buffer to its device driver for I/O
992  * @bio:  The bio describing the location in memory and on the device.
993  *
994  * generic_make_request() is used to make I/O requests of block
995  * devices. It is passed a &struct bio, which describes the I/O that needs
996  * to be done.
997  *
998  * generic_make_request() does not return any status.  The
999  * success/failure status of the request, along with notification of
1000  * completion, is delivered asynchronously through the bio->bi_end_io
1001  * function described (one day) else where.
1002  *
1003  * The caller of generic_make_request must make sure that bi_io_vec
1004  * are set to describe the memory buffer, and that bi_dev and bi_sector are
1005  * set to describe the device address, and the
1006  * bi_end_io and optionally bi_private are set to describe how
1007  * completion notification should be signaled.
1008  *
1009  * generic_make_request and the drivers it calls may use bi_next if this
1010  * bio happens to be merged with someone else, and may resubmit the bio to
1011  * a lower device by calling into generic_make_request recursively, which
1012  * means the bio should NOT be touched after the call to ->make_request_fn.
1013  */
1014 blk_qc_t generic_make_request(struct bio *bio)
1015 {
1016 	/*
1017 	 * bio_list_on_stack[0] contains bios submitted by the current
1018 	 * make_request_fn.
1019 	 * bio_list_on_stack[1] contains bios that were submitted before
1020 	 * the current make_request_fn, but that haven't been processed
1021 	 * yet.
1022 	 */
1023 	struct bio_list bio_list_on_stack[2];
1024 	blk_qc_t ret = BLK_QC_T_NONE;
1025 
1026 	if (!generic_make_request_checks(bio))
1027 		goto out;
1028 
1029 	/*
1030 	 * We only want one ->make_request_fn to be active at a time, else
1031 	 * stack usage with stacked devices could be a problem.  So use
1032 	 * current->bio_list to keep a list of requests submited by a
1033 	 * make_request_fn function.  current->bio_list is also used as a
1034 	 * flag to say if generic_make_request is currently active in this
1035 	 * task or not.  If it is NULL, then no make_request is active.  If
1036 	 * it is non-NULL, then a make_request is active, and new requests
1037 	 * should be added at the tail
1038 	 */
1039 	if (current->bio_list) {
1040 		bio_list_add(&current->bio_list[0], bio);
1041 		goto out;
1042 	}
1043 
1044 	/* following loop may be a bit non-obvious, and so deserves some
1045 	 * explanation.
1046 	 * Before entering the loop, bio->bi_next is NULL (as all callers
1047 	 * ensure that) so we have a list with a single bio.
1048 	 * We pretend that we have just taken it off a longer list, so
1049 	 * we assign bio_list to a pointer to the bio_list_on_stack,
1050 	 * thus initialising the bio_list of new bios to be
1051 	 * added.  ->make_request() may indeed add some more bios
1052 	 * through a recursive call to generic_make_request.  If it
1053 	 * did, we find a non-NULL value in bio_list and re-enter the loop
1054 	 * from the top.  In this case we really did just take the bio
1055 	 * of the top of the list (no pretending) and so remove it from
1056 	 * bio_list, and call into ->make_request() again.
1057 	 */
1058 	BUG_ON(bio->bi_next);
1059 	bio_list_init(&bio_list_on_stack[0]);
1060 	current->bio_list = bio_list_on_stack;
1061 	do {
1062 		struct request_queue *q = bio->bi_disk->queue;
1063 		blk_mq_req_flags_t flags = bio->bi_opf & REQ_NOWAIT ?
1064 			BLK_MQ_REQ_NOWAIT : 0;
1065 
1066 		if (likely(blk_queue_enter(q, flags) == 0)) {
1067 			struct bio_list lower, same;
1068 
1069 			/* Create a fresh bio_list for all subordinate requests */
1070 			bio_list_on_stack[1] = bio_list_on_stack[0];
1071 			bio_list_init(&bio_list_on_stack[0]);
1072 			ret = q->make_request_fn(q, bio);
1073 
1074 			blk_queue_exit(q);
1075 
1076 			/* sort new bios into those for a lower level
1077 			 * and those for the same level
1078 			 */
1079 			bio_list_init(&lower);
1080 			bio_list_init(&same);
1081 			while ((bio = bio_list_pop(&bio_list_on_stack[0])) != NULL)
1082 				if (q == bio->bi_disk->queue)
1083 					bio_list_add(&same, bio);
1084 				else
1085 					bio_list_add(&lower, bio);
1086 			/* now assemble so we handle the lowest level first */
1087 			bio_list_merge(&bio_list_on_stack[0], &lower);
1088 			bio_list_merge(&bio_list_on_stack[0], &same);
1089 			bio_list_merge(&bio_list_on_stack[0], &bio_list_on_stack[1]);
1090 		} else {
1091 			if (unlikely(!blk_queue_dying(q) &&
1092 					(bio->bi_opf & REQ_NOWAIT)))
1093 				bio_wouldblock_error(bio);
1094 			else
1095 				bio_io_error(bio);
1096 		}
1097 		bio = bio_list_pop(&bio_list_on_stack[0]);
1098 	} while (bio);
1099 	current->bio_list = NULL; /* deactivate */
1100 
1101 out:
1102 	return ret;
1103 }
1104 EXPORT_SYMBOL(generic_make_request);
1105 
1106 /**
1107  * direct_make_request - hand a buffer directly to its device driver for I/O
1108  * @bio:  The bio describing the location in memory and on the device.
1109  *
1110  * This function behaves like generic_make_request(), but does not protect
1111  * against recursion.  Must only be used if the called driver is known
1112  * to not call generic_make_request (or direct_make_request) again from
1113  * its make_request function.  (Calling direct_make_request again from
1114  * a workqueue is perfectly fine as that doesn't recurse).
1115  */
1116 blk_qc_t direct_make_request(struct bio *bio)
1117 {
1118 	struct request_queue *q = bio->bi_disk->queue;
1119 	bool nowait = bio->bi_opf & REQ_NOWAIT;
1120 	blk_qc_t ret;
1121 
1122 	if (!generic_make_request_checks(bio))
1123 		return BLK_QC_T_NONE;
1124 
1125 	if (unlikely(blk_queue_enter(q, nowait ? BLK_MQ_REQ_NOWAIT : 0))) {
1126 		if (nowait && !blk_queue_dying(q))
1127 			bio_wouldblock_error(bio);
1128 		else
1129 			bio_io_error(bio);
1130 		return BLK_QC_T_NONE;
1131 	}
1132 
1133 	ret = q->make_request_fn(q, bio);
1134 	blk_queue_exit(q);
1135 	return ret;
1136 }
1137 EXPORT_SYMBOL_GPL(direct_make_request);
1138 
1139 /**
1140  * submit_bio - submit a bio to the block device layer for I/O
1141  * @bio: The &struct bio which describes the I/O
1142  *
1143  * submit_bio() is very similar in purpose to generic_make_request(), and
1144  * uses that function to do most of the work. Both are fairly rough
1145  * interfaces; @bio must be presetup and ready for I/O.
1146  *
1147  */
1148 blk_qc_t submit_bio(struct bio *bio)
1149 {
1150 	bool workingset_read = false;
1151 	unsigned long pflags;
1152 	blk_qc_t ret;
1153 
1154 	if (blkcg_punt_bio_submit(bio))
1155 		return BLK_QC_T_NONE;
1156 
1157 	/*
1158 	 * If it's a regular read/write or a barrier with data attached,
1159 	 * go through the normal accounting stuff before submission.
1160 	 */
1161 	if (bio_has_data(bio)) {
1162 		unsigned int count;
1163 
1164 		if (unlikely(bio_op(bio) == REQ_OP_WRITE_SAME))
1165 			count = queue_logical_block_size(bio->bi_disk->queue) >> 9;
1166 		else
1167 			count = bio_sectors(bio);
1168 
1169 		if (op_is_write(bio_op(bio))) {
1170 			count_vm_events(PGPGOUT, count);
1171 		} else {
1172 			if (bio_flagged(bio, BIO_WORKINGSET))
1173 				workingset_read = true;
1174 			task_io_account_read(bio->bi_iter.bi_size);
1175 			count_vm_events(PGPGIN, count);
1176 		}
1177 
1178 		if (unlikely(block_dump)) {
1179 			char b[BDEVNAME_SIZE];
1180 			printk(KERN_DEBUG "%s(%d): %s block %Lu on %s (%u sectors)\n",
1181 			current->comm, task_pid_nr(current),
1182 				op_is_write(bio_op(bio)) ? "WRITE" : "READ",
1183 				(unsigned long long)bio->bi_iter.bi_sector,
1184 				bio_devname(bio, b), count);
1185 		}
1186 	}
1187 
1188 	/*
1189 	 * If we're reading data that is part of the userspace
1190 	 * workingset, count submission time as memory stall. When the
1191 	 * device is congested, or the submitting cgroup IO-throttled,
1192 	 * submission can be a significant part of overall IO time.
1193 	 */
1194 	if (workingset_read)
1195 		psi_memstall_enter(&pflags);
1196 
1197 	ret = generic_make_request(bio);
1198 
1199 	if (workingset_read)
1200 		psi_memstall_leave(&pflags);
1201 
1202 	return ret;
1203 }
1204 EXPORT_SYMBOL(submit_bio);
1205 
1206 /**
1207  * blk_cloned_rq_check_limits - Helper function to check a cloned request
1208  *                              for the new queue limits
1209  * @q:  the queue
1210  * @rq: the request being checked
1211  *
1212  * Description:
1213  *    @rq may have been made based on weaker limitations of upper-level queues
1214  *    in request stacking drivers, and it may violate the limitation of @q.
1215  *    Since the block layer and the underlying device driver trust @rq
1216  *    after it is inserted to @q, it should be checked against @q before
1217  *    the insertion using this generic function.
1218  *
1219  *    Request stacking drivers like request-based dm may change the queue
1220  *    limits when retrying requests on other queues. Those requests need
1221  *    to be checked against the new queue limits again during dispatch.
1222  */
1223 static int blk_cloned_rq_check_limits(struct request_queue *q,
1224 				      struct request *rq)
1225 {
1226 	if (blk_rq_sectors(rq) > blk_queue_get_max_sectors(q, req_op(rq))) {
1227 		printk(KERN_ERR "%s: over max size limit. (%u > %u)\n",
1228 			__func__, blk_rq_sectors(rq),
1229 			blk_queue_get_max_sectors(q, req_op(rq)));
1230 		return -EIO;
1231 	}
1232 
1233 	/*
1234 	 * queue's settings related to segment counting like q->bounce_pfn
1235 	 * may differ from that of other stacking queues.
1236 	 * Recalculate it to check the request correctly on this queue's
1237 	 * limitation.
1238 	 */
1239 	rq->nr_phys_segments = blk_recalc_rq_segments(rq);
1240 	if (rq->nr_phys_segments > queue_max_segments(q)) {
1241 		printk(KERN_ERR "%s: over max segments limit. (%hu > %hu)\n",
1242 			__func__, rq->nr_phys_segments, queue_max_segments(q));
1243 		return -EIO;
1244 	}
1245 
1246 	return 0;
1247 }
1248 
1249 /**
1250  * blk_insert_cloned_request - Helper for stacking drivers to submit a request
1251  * @q:  the queue to submit the request
1252  * @rq: the request being queued
1253  */
1254 blk_status_t blk_insert_cloned_request(struct request_queue *q, struct request *rq)
1255 {
1256 	if (blk_cloned_rq_check_limits(q, rq))
1257 		return BLK_STS_IOERR;
1258 
1259 	if (rq->rq_disk &&
1260 	    should_fail_request(&rq->rq_disk->part0, blk_rq_bytes(rq)))
1261 		return BLK_STS_IOERR;
1262 
1263 	if (blk_queue_io_stat(q))
1264 		blk_account_io_start(rq, true);
1265 
1266 	/*
1267 	 * Since we have a scheduler attached on the top device,
1268 	 * bypass a potential scheduler on the bottom device for
1269 	 * insert.
1270 	 */
1271 	return blk_mq_request_issue_directly(rq, true);
1272 }
1273 EXPORT_SYMBOL_GPL(blk_insert_cloned_request);
1274 
1275 /**
1276  * blk_rq_err_bytes - determine number of bytes till the next failure boundary
1277  * @rq: request to examine
1278  *
1279  * Description:
1280  *     A request could be merge of IOs which require different failure
1281  *     handling.  This function determines the number of bytes which
1282  *     can be failed from the beginning of the request without
1283  *     crossing into area which need to be retried further.
1284  *
1285  * Return:
1286  *     The number of bytes to fail.
1287  */
1288 unsigned int blk_rq_err_bytes(const struct request *rq)
1289 {
1290 	unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK;
1291 	unsigned int bytes = 0;
1292 	struct bio *bio;
1293 
1294 	if (!(rq->rq_flags & RQF_MIXED_MERGE))
1295 		return blk_rq_bytes(rq);
1296 
1297 	/*
1298 	 * Currently the only 'mixing' which can happen is between
1299 	 * different fastfail types.  We can safely fail portions
1300 	 * which have all the failfast bits that the first one has -
1301 	 * the ones which are at least as eager to fail as the first
1302 	 * one.
1303 	 */
1304 	for (bio = rq->bio; bio; bio = bio->bi_next) {
1305 		if ((bio->bi_opf & ff) != ff)
1306 			break;
1307 		bytes += bio->bi_iter.bi_size;
1308 	}
1309 
1310 	/* this could lead to infinite loop */
1311 	BUG_ON(blk_rq_bytes(rq) && !bytes);
1312 	return bytes;
1313 }
1314 EXPORT_SYMBOL_GPL(blk_rq_err_bytes);
1315 
1316 void blk_account_io_completion(struct request *req, unsigned int bytes)
1317 {
1318 	if (req->part && blk_do_io_stat(req)) {
1319 		const int sgrp = op_stat_group(req_op(req));
1320 		struct hd_struct *part;
1321 
1322 		part_stat_lock();
1323 		part = req->part;
1324 		part_stat_add(part, sectors[sgrp], bytes >> 9);
1325 		part_stat_unlock();
1326 	}
1327 }
1328 
1329 void blk_account_io_done(struct request *req, u64 now)
1330 {
1331 	/*
1332 	 * Account IO completion.  flush_rq isn't accounted as a
1333 	 * normal IO on queueing nor completion.  Accounting the
1334 	 * containing request is enough.
1335 	 */
1336 	if (req->part && blk_do_io_stat(req) &&
1337 	    !(req->rq_flags & RQF_FLUSH_SEQ)) {
1338 		const int sgrp = op_stat_group(req_op(req));
1339 		struct hd_struct *part;
1340 
1341 		part_stat_lock();
1342 		part = req->part;
1343 
1344 		update_io_ticks(part, jiffies, true);
1345 		part_stat_inc(part, ios[sgrp]);
1346 		part_stat_add(part, nsecs[sgrp], now - req->start_time_ns);
1347 		part_dec_in_flight(req->q, part, rq_data_dir(req));
1348 
1349 		hd_struct_put(part);
1350 		part_stat_unlock();
1351 	}
1352 }
1353 
1354 void blk_account_io_start(struct request *rq, bool new_io)
1355 {
1356 	struct hd_struct *part;
1357 	int rw = rq_data_dir(rq);
1358 
1359 	if (!blk_do_io_stat(rq))
1360 		return;
1361 
1362 	part_stat_lock();
1363 
1364 	if (!new_io) {
1365 		part = rq->part;
1366 		part_stat_inc(part, merges[rw]);
1367 	} else {
1368 		part = disk_map_sector_rcu(rq->rq_disk, blk_rq_pos(rq));
1369 		if (!hd_struct_try_get(part)) {
1370 			/*
1371 			 * The partition is already being removed,
1372 			 * the request will be accounted on the disk only
1373 			 *
1374 			 * We take a reference on disk->part0 although that
1375 			 * partition will never be deleted, so we can treat
1376 			 * it as any other partition.
1377 			 */
1378 			part = &rq->rq_disk->part0;
1379 			hd_struct_get(part);
1380 		}
1381 		part_inc_in_flight(rq->q, part, rw);
1382 		rq->part = part;
1383 	}
1384 
1385 	update_io_ticks(part, jiffies, false);
1386 
1387 	part_stat_unlock();
1388 }
1389 
1390 /*
1391  * Steal bios from a request and add them to a bio list.
1392  * The request must not have been partially completed before.
1393  */
1394 void blk_steal_bios(struct bio_list *list, struct request *rq)
1395 {
1396 	if (rq->bio) {
1397 		if (list->tail)
1398 			list->tail->bi_next = rq->bio;
1399 		else
1400 			list->head = rq->bio;
1401 		list->tail = rq->biotail;
1402 
1403 		rq->bio = NULL;
1404 		rq->biotail = NULL;
1405 	}
1406 
1407 	rq->__data_len = 0;
1408 }
1409 EXPORT_SYMBOL_GPL(blk_steal_bios);
1410 
1411 /**
1412  * blk_update_request - Special helper function for request stacking drivers
1413  * @req:      the request being processed
1414  * @error:    block status code
1415  * @nr_bytes: number of bytes to complete @req
1416  *
1417  * Description:
1418  *     Ends I/O on a number of bytes attached to @req, but doesn't complete
1419  *     the request structure even if @req doesn't have leftover.
1420  *     If @req has leftover, sets it up for the next range of segments.
1421  *
1422  *     This special helper function is only for request stacking drivers
1423  *     (e.g. request-based dm) so that they can handle partial completion.
1424  *     Actual device drivers should use blk_mq_end_request instead.
1425  *
1426  *     Passing the result of blk_rq_bytes() as @nr_bytes guarantees
1427  *     %false return from this function.
1428  *
1429  * Note:
1430  *	The RQF_SPECIAL_PAYLOAD flag is ignored on purpose in both
1431  *	blk_rq_bytes() and in blk_update_request().
1432  *
1433  * Return:
1434  *     %false - this request doesn't have any more data
1435  *     %true  - this request has more data
1436  **/
1437 bool blk_update_request(struct request *req, blk_status_t error,
1438 		unsigned int nr_bytes)
1439 {
1440 	int total_bytes;
1441 
1442 	trace_block_rq_complete(req, blk_status_to_errno(error), nr_bytes);
1443 
1444 	if (!req->bio)
1445 		return false;
1446 
1447 #ifdef CONFIG_BLK_DEV_INTEGRITY
1448 	if (blk_integrity_rq(req) && req_op(req) == REQ_OP_READ &&
1449 	    error == BLK_STS_OK)
1450 		req->q->integrity.profile->complete_fn(req, nr_bytes);
1451 #endif
1452 
1453 	if (unlikely(error && !blk_rq_is_passthrough(req) &&
1454 		     !(req->rq_flags & RQF_QUIET)))
1455 		print_req_error(req, error, __func__);
1456 
1457 	blk_account_io_completion(req, nr_bytes);
1458 
1459 	total_bytes = 0;
1460 	while (req->bio) {
1461 		struct bio *bio = req->bio;
1462 		unsigned bio_bytes = min(bio->bi_iter.bi_size, nr_bytes);
1463 
1464 		if (bio_bytes == bio->bi_iter.bi_size)
1465 			req->bio = bio->bi_next;
1466 
1467 		/* Completion has already been traced */
1468 		bio_clear_flag(bio, BIO_TRACE_COMPLETION);
1469 		req_bio_endio(req, bio, bio_bytes, error);
1470 
1471 		total_bytes += bio_bytes;
1472 		nr_bytes -= bio_bytes;
1473 
1474 		if (!nr_bytes)
1475 			break;
1476 	}
1477 
1478 	/*
1479 	 * completely done
1480 	 */
1481 	if (!req->bio) {
1482 		/*
1483 		 * Reset counters so that the request stacking driver
1484 		 * can find how many bytes remain in the request
1485 		 * later.
1486 		 */
1487 		req->__data_len = 0;
1488 		return false;
1489 	}
1490 
1491 	req->__data_len -= total_bytes;
1492 
1493 	/* update sector only for requests with clear definition of sector */
1494 	if (!blk_rq_is_passthrough(req))
1495 		req->__sector += total_bytes >> 9;
1496 
1497 	/* mixed attributes always follow the first bio */
1498 	if (req->rq_flags & RQF_MIXED_MERGE) {
1499 		req->cmd_flags &= ~REQ_FAILFAST_MASK;
1500 		req->cmd_flags |= req->bio->bi_opf & REQ_FAILFAST_MASK;
1501 	}
1502 
1503 	if (!(req->rq_flags & RQF_SPECIAL_PAYLOAD)) {
1504 		/*
1505 		 * If total number of sectors is less than the first segment
1506 		 * size, something has gone terribly wrong.
1507 		 */
1508 		if (blk_rq_bytes(req) < blk_rq_cur_bytes(req)) {
1509 			blk_dump_rq_flags(req, "request botched");
1510 			req->__data_len = blk_rq_cur_bytes(req);
1511 		}
1512 
1513 		/* recalculate the number of segments */
1514 		req->nr_phys_segments = blk_recalc_rq_segments(req);
1515 	}
1516 
1517 	return true;
1518 }
1519 EXPORT_SYMBOL_GPL(blk_update_request);
1520 
1521 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
1522 /**
1523  * rq_flush_dcache_pages - Helper function to flush all pages in a request
1524  * @rq: the request to be flushed
1525  *
1526  * Description:
1527  *     Flush all pages in @rq.
1528  */
1529 void rq_flush_dcache_pages(struct request *rq)
1530 {
1531 	struct req_iterator iter;
1532 	struct bio_vec bvec;
1533 
1534 	rq_for_each_segment(bvec, rq, iter)
1535 		flush_dcache_page(bvec.bv_page);
1536 }
1537 EXPORT_SYMBOL_GPL(rq_flush_dcache_pages);
1538 #endif
1539 
1540 /**
1541  * blk_lld_busy - Check if underlying low-level drivers of a device are busy
1542  * @q : the queue of the device being checked
1543  *
1544  * Description:
1545  *    Check if underlying low-level drivers of a device are busy.
1546  *    If the drivers want to export their busy state, they must set own
1547  *    exporting function using blk_queue_lld_busy() first.
1548  *
1549  *    Basically, this function is used only by request stacking drivers
1550  *    to stop dispatching requests to underlying devices when underlying
1551  *    devices are busy.  This behavior helps more I/O merging on the queue
1552  *    of the request stacking driver and prevents I/O throughput regression
1553  *    on burst I/O load.
1554  *
1555  * Return:
1556  *    0 - Not busy (The request stacking driver should dispatch request)
1557  *    1 - Busy (The request stacking driver should stop dispatching request)
1558  */
1559 int blk_lld_busy(struct request_queue *q)
1560 {
1561 	if (queue_is_mq(q) && q->mq_ops->busy)
1562 		return q->mq_ops->busy(q);
1563 
1564 	return 0;
1565 }
1566 EXPORT_SYMBOL_GPL(blk_lld_busy);
1567 
1568 /**
1569  * blk_rq_unprep_clone - Helper function to free all bios in a cloned request
1570  * @rq: the clone request to be cleaned up
1571  *
1572  * Description:
1573  *     Free all bios in @rq for a cloned request.
1574  */
1575 void blk_rq_unprep_clone(struct request *rq)
1576 {
1577 	struct bio *bio;
1578 
1579 	while ((bio = rq->bio) != NULL) {
1580 		rq->bio = bio->bi_next;
1581 
1582 		bio_put(bio);
1583 	}
1584 }
1585 EXPORT_SYMBOL_GPL(blk_rq_unprep_clone);
1586 
1587 /**
1588  * blk_rq_prep_clone - Helper function to setup clone request
1589  * @rq: the request to be setup
1590  * @rq_src: original request to be cloned
1591  * @bs: bio_set that bios for clone are allocated from
1592  * @gfp_mask: memory allocation mask for bio
1593  * @bio_ctr: setup function to be called for each clone bio.
1594  *           Returns %0 for success, non %0 for failure.
1595  * @data: private data to be passed to @bio_ctr
1596  *
1597  * Description:
1598  *     Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq.
1599  *     Also, pages which the original bios are pointing to are not copied
1600  *     and the cloned bios just point same pages.
1601  *     So cloned bios must be completed before original bios, which means
1602  *     the caller must complete @rq before @rq_src.
1603  */
1604 int blk_rq_prep_clone(struct request *rq, struct request *rq_src,
1605 		      struct bio_set *bs, gfp_t gfp_mask,
1606 		      int (*bio_ctr)(struct bio *, struct bio *, void *),
1607 		      void *data)
1608 {
1609 	struct bio *bio, *bio_src;
1610 
1611 	if (!bs)
1612 		bs = &fs_bio_set;
1613 
1614 	__rq_for_each_bio(bio_src, rq_src) {
1615 		bio = bio_clone_fast(bio_src, gfp_mask, bs);
1616 		if (!bio)
1617 			goto free_and_out;
1618 
1619 		if (bio_ctr && bio_ctr(bio, bio_src, data))
1620 			goto free_and_out;
1621 
1622 		if (rq->bio) {
1623 			rq->biotail->bi_next = bio;
1624 			rq->biotail = bio;
1625 		} else
1626 			rq->bio = rq->biotail = bio;
1627 	}
1628 
1629 	/* Copy attributes of the original request to the clone request. */
1630 	rq->__sector = blk_rq_pos(rq_src);
1631 	rq->__data_len = blk_rq_bytes(rq_src);
1632 	if (rq_src->rq_flags & RQF_SPECIAL_PAYLOAD) {
1633 		rq->rq_flags |= RQF_SPECIAL_PAYLOAD;
1634 		rq->special_vec = rq_src->special_vec;
1635 	}
1636 	rq->nr_phys_segments = rq_src->nr_phys_segments;
1637 	rq->ioprio = rq_src->ioprio;
1638 	rq->extra_len = rq_src->extra_len;
1639 
1640 	return 0;
1641 
1642 free_and_out:
1643 	if (bio)
1644 		bio_put(bio);
1645 	blk_rq_unprep_clone(rq);
1646 
1647 	return -ENOMEM;
1648 }
1649 EXPORT_SYMBOL_GPL(blk_rq_prep_clone);
1650 
1651 int kblockd_schedule_work(struct work_struct *work)
1652 {
1653 	return queue_work(kblockd_workqueue, work);
1654 }
1655 EXPORT_SYMBOL(kblockd_schedule_work);
1656 
1657 int kblockd_mod_delayed_work_on(int cpu, struct delayed_work *dwork,
1658 				unsigned long delay)
1659 {
1660 	return mod_delayed_work_on(cpu, kblockd_workqueue, dwork, delay);
1661 }
1662 EXPORT_SYMBOL(kblockd_mod_delayed_work_on);
1663 
1664 /**
1665  * blk_start_plug - initialize blk_plug and track it inside the task_struct
1666  * @plug:	The &struct blk_plug that needs to be initialized
1667  *
1668  * Description:
1669  *   blk_start_plug() indicates to the block layer an intent by the caller
1670  *   to submit multiple I/O requests in a batch.  The block layer may use
1671  *   this hint to defer submitting I/Os from the caller until blk_finish_plug()
1672  *   is called.  However, the block layer may choose to submit requests
1673  *   before a call to blk_finish_plug() if the number of queued I/Os
1674  *   exceeds %BLK_MAX_REQUEST_COUNT, or if the size of the I/O is larger than
1675  *   %BLK_PLUG_FLUSH_SIZE.  The queued I/Os may also be submitted early if
1676  *   the task schedules (see below).
1677  *
1678  *   Tracking blk_plug inside the task_struct will help with auto-flushing the
1679  *   pending I/O should the task end up blocking between blk_start_plug() and
1680  *   blk_finish_plug(). This is important from a performance perspective, but
1681  *   also ensures that we don't deadlock. For instance, if the task is blocking
1682  *   for a memory allocation, memory reclaim could end up wanting to free a
1683  *   page belonging to that request that is currently residing in our private
1684  *   plug. By flushing the pending I/O when the process goes to sleep, we avoid
1685  *   this kind of deadlock.
1686  */
1687 void blk_start_plug(struct blk_plug *plug)
1688 {
1689 	struct task_struct *tsk = current;
1690 
1691 	/*
1692 	 * If this is a nested plug, don't actually assign it.
1693 	 */
1694 	if (tsk->plug)
1695 		return;
1696 
1697 	INIT_LIST_HEAD(&plug->mq_list);
1698 	INIT_LIST_HEAD(&plug->cb_list);
1699 	plug->rq_count = 0;
1700 	plug->multiple_queues = false;
1701 
1702 	/*
1703 	 * Store ordering should not be needed here, since a potential
1704 	 * preempt will imply a full memory barrier
1705 	 */
1706 	tsk->plug = plug;
1707 }
1708 EXPORT_SYMBOL(blk_start_plug);
1709 
1710 static void flush_plug_callbacks(struct blk_plug *plug, bool from_schedule)
1711 {
1712 	LIST_HEAD(callbacks);
1713 
1714 	while (!list_empty(&plug->cb_list)) {
1715 		list_splice_init(&plug->cb_list, &callbacks);
1716 
1717 		while (!list_empty(&callbacks)) {
1718 			struct blk_plug_cb *cb = list_first_entry(&callbacks,
1719 							  struct blk_plug_cb,
1720 							  list);
1721 			list_del(&cb->list);
1722 			cb->callback(cb, from_schedule);
1723 		}
1724 	}
1725 }
1726 
1727 struct blk_plug_cb *blk_check_plugged(blk_plug_cb_fn unplug, void *data,
1728 				      int size)
1729 {
1730 	struct blk_plug *plug = current->plug;
1731 	struct blk_plug_cb *cb;
1732 
1733 	if (!plug)
1734 		return NULL;
1735 
1736 	list_for_each_entry(cb, &plug->cb_list, list)
1737 		if (cb->callback == unplug && cb->data == data)
1738 			return cb;
1739 
1740 	/* Not currently on the callback list */
1741 	BUG_ON(size < sizeof(*cb));
1742 	cb = kzalloc(size, GFP_ATOMIC);
1743 	if (cb) {
1744 		cb->data = data;
1745 		cb->callback = unplug;
1746 		list_add(&cb->list, &plug->cb_list);
1747 	}
1748 	return cb;
1749 }
1750 EXPORT_SYMBOL(blk_check_plugged);
1751 
1752 void blk_flush_plug_list(struct blk_plug *plug, bool from_schedule)
1753 {
1754 	flush_plug_callbacks(plug, from_schedule);
1755 
1756 	if (!list_empty(&plug->mq_list))
1757 		blk_mq_flush_plug_list(plug, from_schedule);
1758 }
1759 
1760 /**
1761  * blk_finish_plug - mark the end of a batch of submitted I/O
1762  * @plug:	The &struct blk_plug passed to blk_start_plug()
1763  *
1764  * Description:
1765  * Indicate that a batch of I/O submissions is complete.  This function
1766  * must be paired with an initial call to blk_start_plug().  The intent
1767  * is to allow the block layer to optimize I/O submission.  See the
1768  * documentation for blk_start_plug() for more information.
1769  */
1770 void blk_finish_plug(struct blk_plug *plug)
1771 {
1772 	if (plug != current->plug)
1773 		return;
1774 	blk_flush_plug_list(plug, false);
1775 
1776 	current->plug = NULL;
1777 }
1778 EXPORT_SYMBOL(blk_finish_plug);
1779 
1780 int __init blk_dev_init(void)
1781 {
1782 	BUILD_BUG_ON(REQ_OP_LAST >= (1 << REQ_OP_BITS));
1783 	BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 *
1784 			sizeof_field(struct request, cmd_flags));
1785 	BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 *
1786 			sizeof_field(struct bio, bi_opf));
1787 
1788 	/* used for unplugging and affects IO latency/throughput - HIGHPRI */
1789 	kblockd_workqueue = alloc_workqueue("kblockd",
1790 					    WQ_MEM_RECLAIM | WQ_HIGHPRI, 0);
1791 	if (!kblockd_workqueue)
1792 		panic("Failed to create kblockd\n");
1793 
1794 	blk_requestq_cachep = kmem_cache_create("request_queue",
1795 			sizeof(struct request_queue), 0, SLAB_PANIC, NULL);
1796 
1797 #ifdef CONFIG_DEBUG_FS
1798 	blk_debugfs_root = debugfs_create_dir("block", NULL);
1799 #endif
1800 
1801 	return 0;
1802 }
1803