xref: /openbmc/linux/block/blk-core.c (revision e620a1e0)
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(WRITE_SAME),
136 	REQ_OP_NAME(WRITE_ZEROES),
137 	REQ_OP_NAME(SCSI_IN),
138 	REQ_OP_NAME(SCSI_OUT),
139 	REQ_OP_NAME(DRV_IN),
140 	REQ_OP_NAME(DRV_OUT),
141 };
142 #undef REQ_OP_NAME
143 
144 /**
145  * blk_op_str - Return string XXX in the REQ_OP_XXX.
146  * @op: REQ_OP_XXX.
147  *
148  * Description: Centralize block layer function to convert REQ_OP_XXX into
149  * string format. Useful in the debugging and tracing bio or request. For
150  * invalid REQ_OP_XXX it returns string "UNKNOWN".
151  */
152 inline const char *blk_op_str(unsigned int op)
153 {
154 	const char *op_str = "UNKNOWN";
155 
156 	if (op < ARRAY_SIZE(blk_op_name) && blk_op_name[op])
157 		op_str = blk_op_name[op];
158 
159 	return op_str;
160 }
161 EXPORT_SYMBOL_GPL(blk_op_str);
162 
163 static const struct {
164 	int		errno;
165 	const char	*name;
166 } blk_errors[] = {
167 	[BLK_STS_OK]		= { 0,		"" },
168 	[BLK_STS_NOTSUPP]	= { -EOPNOTSUPP, "operation not supported" },
169 	[BLK_STS_TIMEOUT]	= { -ETIMEDOUT,	"timeout" },
170 	[BLK_STS_NOSPC]		= { -ENOSPC,	"critical space allocation" },
171 	[BLK_STS_TRANSPORT]	= { -ENOLINK,	"recoverable transport" },
172 	[BLK_STS_TARGET]	= { -EREMOTEIO,	"critical target" },
173 	[BLK_STS_NEXUS]		= { -EBADE,	"critical nexus" },
174 	[BLK_STS_MEDIUM]	= { -ENODATA,	"critical medium" },
175 	[BLK_STS_PROTECTION]	= { -EILSEQ,	"protection" },
176 	[BLK_STS_RESOURCE]	= { -ENOMEM,	"kernel resource" },
177 	[BLK_STS_DEV_RESOURCE]	= { -EBUSY,	"device resource" },
178 	[BLK_STS_AGAIN]		= { -EAGAIN,	"nonblocking retry" },
179 
180 	/* device mapper special case, should not leak out: */
181 	[BLK_STS_DM_REQUEUE]	= { -EREMCHG, "dm internal retry" },
182 
183 	/* everything else not covered above: */
184 	[BLK_STS_IOERR]		= { -EIO,	"I/O" },
185 };
186 
187 blk_status_t errno_to_blk_status(int errno)
188 {
189 	int i;
190 
191 	for (i = 0; i < ARRAY_SIZE(blk_errors); i++) {
192 		if (blk_errors[i].errno == errno)
193 			return (__force blk_status_t)i;
194 	}
195 
196 	return BLK_STS_IOERR;
197 }
198 EXPORT_SYMBOL_GPL(errno_to_blk_status);
199 
200 int blk_status_to_errno(blk_status_t status)
201 {
202 	int idx = (__force int)status;
203 
204 	if (WARN_ON_ONCE(idx >= ARRAY_SIZE(blk_errors)))
205 		return -EIO;
206 	return blk_errors[idx].errno;
207 }
208 EXPORT_SYMBOL_GPL(blk_status_to_errno);
209 
210 static void print_req_error(struct request *req, blk_status_t status,
211 		const char *caller)
212 {
213 	int idx = (__force int)status;
214 
215 	if (WARN_ON_ONCE(idx >= ARRAY_SIZE(blk_errors)))
216 		return;
217 
218 	printk_ratelimited(KERN_ERR
219 		"%s: %s error, dev %s, sector %llu op 0x%x:(%s) flags 0x%x "
220 		"phys_seg %u prio class %u\n",
221 		caller, blk_errors[idx].name,
222 		req->rq_disk ? req->rq_disk->disk_name : "?",
223 		blk_rq_pos(req), req_op(req), blk_op_str(req_op(req)),
224 		req->cmd_flags & ~REQ_OP_MASK,
225 		req->nr_phys_segments,
226 		IOPRIO_PRIO_CLASS(req->ioprio));
227 }
228 
229 static void req_bio_endio(struct request *rq, struct bio *bio,
230 			  unsigned int nbytes, blk_status_t error)
231 {
232 	if (error)
233 		bio->bi_status = error;
234 
235 	if (unlikely(rq->rq_flags & RQF_QUIET))
236 		bio_set_flag(bio, BIO_QUIET);
237 
238 	bio_advance(bio, nbytes);
239 
240 	/* don't actually finish bio if it's part of flush sequence */
241 	if (bio->bi_iter.bi_size == 0 && !(rq->rq_flags & RQF_FLUSH_SEQ))
242 		bio_endio(bio);
243 }
244 
245 void blk_dump_rq_flags(struct request *rq, char *msg)
246 {
247 	printk(KERN_INFO "%s: dev %s: flags=%llx\n", msg,
248 		rq->rq_disk ? rq->rq_disk->disk_name : "?",
249 		(unsigned long long) rq->cmd_flags);
250 
251 	printk(KERN_INFO "  sector %llu, nr/cnr %u/%u\n",
252 	       (unsigned long long)blk_rq_pos(rq),
253 	       blk_rq_sectors(rq), blk_rq_cur_sectors(rq));
254 	printk(KERN_INFO "  bio %p, biotail %p, len %u\n",
255 	       rq->bio, rq->biotail, blk_rq_bytes(rq));
256 }
257 EXPORT_SYMBOL(blk_dump_rq_flags);
258 
259 /**
260  * blk_sync_queue - cancel any pending callbacks on a queue
261  * @q: the queue
262  *
263  * Description:
264  *     The block layer may perform asynchronous callback activity
265  *     on a queue, such as calling the unplug function after a timeout.
266  *     A block device may call blk_sync_queue to ensure that any
267  *     such activity is cancelled, thus allowing it to release resources
268  *     that the callbacks might use. The caller must already have made sure
269  *     that its ->make_request_fn will not re-add plugging prior to calling
270  *     this function.
271  *
272  *     This function does not cancel any asynchronous activity arising
273  *     out of elevator or throttling code. That would require elevator_exit()
274  *     and blkcg_exit_queue() to be called with queue lock initialized.
275  *
276  */
277 void blk_sync_queue(struct request_queue *q)
278 {
279 	del_timer_sync(&q->timeout);
280 	cancel_work_sync(&q->timeout_work);
281 }
282 EXPORT_SYMBOL(blk_sync_queue);
283 
284 /**
285  * blk_set_pm_only - increment pm_only counter
286  * @q: request queue pointer
287  */
288 void blk_set_pm_only(struct request_queue *q)
289 {
290 	atomic_inc(&q->pm_only);
291 }
292 EXPORT_SYMBOL_GPL(blk_set_pm_only);
293 
294 void blk_clear_pm_only(struct request_queue *q)
295 {
296 	int pm_only;
297 
298 	pm_only = atomic_dec_return(&q->pm_only);
299 	WARN_ON_ONCE(pm_only < 0);
300 	if (pm_only == 0)
301 		wake_up_all(&q->mq_freeze_wq);
302 }
303 EXPORT_SYMBOL_GPL(blk_clear_pm_only);
304 
305 void blk_put_queue(struct request_queue *q)
306 {
307 	kobject_put(&q->kobj);
308 }
309 EXPORT_SYMBOL(blk_put_queue);
310 
311 void blk_set_queue_dying(struct request_queue *q)
312 {
313 	blk_queue_flag_set(QUEUE_FLAG_DYING, q);
314 
315 	/*
316 	 * When queue DYING flag is set, we need to block new req
317 	 * entering queue, so we call blk_freeze_queue_start() to
318 	 * prevent I/O from crossing blk_queue_enter().
319 	 */
320 	blk_freeze_queue_start(q);
321 
322 	if (queue_is_mq(q))
323 		blk_mq_wake_waiters(q);
324 
325 	/* Make blk_queue_enter() reexamine the DYING flag. */
326 	wake_up_all(&q->mq_freeze_wq);
327 }
328 EXPORT_SYMBOL_GPL(blk_set_queue_dying);
329 
330 /**
331  * blk_cleanup_queue - shutdown a request queue
332  * @q: request queue to shutdown
333  *
334  * Mark @q DYING, drain all pending requests, mark @q DEAD, destroy and
335  * put it.  All future requests will be failed immediately with -ENODEV.
336  */
337 void blk_cleanup_queue(struct request_queue *q)
338 {
339 	/* mark @q DYING, no new request or merges will be allowed afterwards */
340 	mutex_lock(&q->sysfs_lock);
341 	blk_set_queue_dying(q);
342 
343 	blk_queue_flag_set(QUEUE_FLAG_NOMERGES, q);
344 	blk_queue_flag_set(QUEUE_FLAG_NOXMERGES, q);
345 	blk_queue_flag_set(QUEUE_FLAG_DYING, q);
346 	mutex_unlock(&q->sysfs_lock);
347 
348 	/*
349 	 * Drain all requests queued before DYING marking. Set DEAD flag to
350 	 * prevent that blk_mq_run_hw_queues() accesses the hardware queues
351 	 * after draining finished.
352 	 */
353 	blk_freeze_queue(q);
354 
355 	rq_qos_exit(q);
356 
357 	blk_queue_flag_set(QUEUE_FLAG_DEAD, q);
358 
359 	/* for synchronous bio-based driver finish in-flight integrity i/o */
360 	blk_flush_integrity();
361 
362 	/* @q won't process any more request, flush async actions */
363 	del_timer_sync(&q->backing_dev_info->laptop_mode_wb_timer);
364 	blk_sync_queue(q);
365 
366 	if (queue_is_mq(q))
367 		blk_mq_exit_queue(q);
368 
369 	/*
370 	 * In theory, request pool of sched_tags belongs to request queue.
371 	 * However, the current implementation requires tag_set for freeing
372 	 * requests, so free the pool now.
373 	 *
374 	 * Queue has become frozen, there can't be any in-queue requests, so
375 	 * it is safe to free requests now.
376 	 */
377 	mutex_lock(&q->sysfs_lock);
378 	if (q->elevator)
379 		blk_mq_sched_free_requests(q);
380 	mutex_unlock(&q->sysfs_lock);
381 
382 	percpu_ref_exit(&q->q_usage_counter);
383 
384 	/* @q is and will stay empty, shutdown and put */
385 	blk_put_queue(q);
386 }
387 EXPORT_SYMBOL(blk_cleanup_queue);
388 
389 struct request_queue *blk_alloc_queue(gfp_t gfp_mask)
390 {
391 	return blk_alloc_queue_node(gfp_mask, NUMA_NO_NODE);
392 }
393 EXPORT_SYMBOL(blk_alloc_queue);
394 
395 /**
396  * blk_queue_enter() - try to increase q->q_usage_counter
397  * @q: request queue pointer
398  * @flags: BLK_MQ_REQ_NOWAIT and/or BLK_MQ_REQ_PREEMPT
399  */
400 int blk_queue_enter(struct request_queue *q, blk_mq_req_flags_t flags)
401 {
402 	const bool pm = flags & BLK_MQ_REQ_PREEMPT;
403 
404 	while (true) {
405 		bool success = false;
406 
407 		rcu_read_lock();
408 		if (percpu_ref_tryget_live(&q->q_usage_counter)) {
409 			/*
410 			 * The code that increments the pm_only counter is
411 			 * responsible for ensuring that that counter is
412 			 * globally visible before the queue is unfrozen.
413 			 */
414 			if (pm || !blk_queue_pm_only(q)) {
415 				success = true;
416 			} else {
417 				percpu_ref_put(&q->q_usage_counter);
418 			}
419 		}
420 		rcu_read_unlock();
421 
422 		if (success)
423 			return 0;
424 
425 		if (flags & BLK_MQ_REQ_NOWAIT)
426 			return -EBUSY;
427 
428 		/*
429 		 * read pair of barrier in blk_freeze_queue_start(),
430 		 * we need to order reading __PERCPU_REF_DEAD flag of
431 		 * .q_usage_counter and reading .mq_freeze_depth or
432 		 * queue dying flag, otherwise the following wait may
433 		 * never return if the two reads are reordered.
434 		 */
435 		smp_rmb();
436 
437 		wait_event(q->mq_freeze_wq,
438 			   (!q->mq_freeze_depth &&
439 			    (pm || (blk_pm_request_resume(q),
440 				    !blk_queue_pm_only(q)))) ||
441 			   blk_queue_dying(q));
442 		if (blk_queue_dying(q))
443 			return -ENODEV;
444 	}
445 }
446 
447 void blk_queue_exit(struct request_queue *q)
448 {
449 	percpu_ref_put(&q->q_usage_counter);
450 }
451 
452 static void blk_queue_usage_counter_release(struct percpu_ref *ref)
453 {
454 	struct request_queue *q =
455 		container_of(ref, struct request_queue, q_usage_counter);
456 
457 	wake_up_all(&q->mq_freeze_wq);
458 }
459 
460 static void blk_rq_timed_out_timer(struct timer_list *t)
461 {
462 	struct request_queue *q = from_timer(q, t, timeout);
463 
464 	kblockd_schedule_work(&q->timeout_work);
465 }
466 
467 static void blk_timeout_work(struct work_struct *work)
468 {
469 }
470 
471 /**
472  * blk_alloc_queue_node - allocate a request queue
473  * @gfp_mask: memory allocation flags
474  * @node_id: NUMA node to allocate memory from
475  */
476 struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id)
477 {
478 	struct request_queue *q;
479 	int ret;
480 
481 	q = kmem_cache_alloc_node(blk_requestq_cachep,
482 				gfp_mask | __GFP_ZERO, node_id);
483 	if (!q)
484 		return NULL;
485 
486 	q->last_merge = NULL;
487 
488 	q->id = ida_simple_get(&blk_queue_ida, 0, 0, gfp_mask);
489 	if (q->id < 0)
490 		goto fail_q;
491 
492 	ret = bioset_init(&q->bio_split, BIO_POOL_SIZE, 0, BIOSET_NEED_BVECS);
493 	if (ret)
494 		goto fail_id;
495 
496 	q->backing_dev_info = bdi_alloc_node(gfp_mask, node_id);
497 	if (!q->backing_dev_info)
498 		goto fail_split;
499 
500 	q->stats = blk_alloc_queue_stats();
501 	if (!q->stats)
502 		goto fail_stats;
503 
504 	q->backing_dev_info->ra_pages = VM_READAHEAD_PAGES;
505 	q->backing_dev_info->capabilities = BDI_CAP_CGROUP_WRITEBACK;
506 	q->backing_dev_info->name = "block";
507 	q->node = node_id;
508 
509 	timer_setup(&q->backing_dev_info->laptop_mode_wb_timer,
510 		    laptop_mode_timer_fn, 0);
511 	timer_setup(&q->timeout, blk_rq_timed_out_timer, 0);
512 	INIT_WORK(&q->timeout_work, blk_timeout_work);
513 	INIT_LIST_HEAD(&q->icq_list);
514 #ifdef CONFIG_BLK_CGROUP
515 	INIT_LIST_HEAD(&q->blkg_list);
516 #endif
517 
518 	kobject_init(&q->kobj, &blk_queue_ktype);
519 
520 #ifdef CONFIG_BLK_DEV_IO_TRACE
521 	mutex_init(&q->blk_trace_mutex);
522 #endif
523 	mutex_init(&q->sysfs_lock);
524 	mutex_init(&q->sysfs_dir_lock);
525 	spin_lock_init(&q->queue_lock);
526 
527 	init_waitqueue_head(&q->mq_freeze_wq);
528 	mutex_init(&q->mq_freeze_lock);
529 
530 	/*
531 	 * Init percpu_ref in atomic mode so that it's faster to shutdown.
532 	 * See blk_register_queue() for details.
533 	 */
534 	if (percpu_ref_init(&q->q_usage_counter,
535 				blk_queue_usage_counter_release,
536 				PERCPU_REF_INIT_ATOMIC, GFP_KERNEL))
537 		goto fail_bdi;
538 
539 	if (blkcg_init_queue(q))
540 		goto fail_ref;
541 
542 	return q;
543 
544 fail_ref:
545 	percpu_ref_exit(&q->q_usage_counter);
546 fail_bdi:
547 	blk_free_queue_stats(q->stats);
548 fail_stats:
549 	bdi_put(q->backing_dev_info);
550 fail_split:
551 	bioset_exit(&q->bio_split);
552 fail_id:
553 	ida_simple_remove(&blk_queue_ida, q->id);
554 fail_q:
555 	kmem_cache_free(blk_requestq_cachep, q);
556 	return NULL;
557 }
558 EXPORT_SYMBOL(blk_alloc_queue_node);
559 
560 bool blk_get_queue(struct request_queue *q)
561 {
562 	if (likely(!blk_queue_dying(q))) {
563 		__blk_get_queue(q);
564 		return true;
565 	}
566 
567 	return false;
568 }
569 EXPORT_SYMBOL(blk_get_queue);
570 
571 /**
572  * blk_get_request - allocate a request
573  * @q: request queue to allocate a request for
574  * @op: operation (REQ_OP_*) and REQ_* flags, e.g. REQ_SYNC.
575  * @flags: BLK_MQ_REQ_* flags, e.g. BLK_MQ_REQ_NOWAIT.
576  */
577 struct request *blk_get_request(struct request_queue *q, unsigned int op,
578 				blk_mq_req_flags_t flags)
579 {
580 	struct request *req;
581 
582 	WARN_ON_ONCE(op & REQ_NOWAIT);
583 	WARN_ON_ONCE(flags & ~(BLK_MQ_REQ_NOWAIT | BLK_MQ_REQ_PREEMPT));
584 
585 	req = blk_mq_alloc_request(q, op, flags);
586 	if (!IS_ERR(req) && q->mq_ops->initialize_rq_fn)
587 		q->mq_ops->initialize_rq_fn(req);
588 
589 	return req;
590 }
591 EXPORT_SYMBOL(blk_get_request);
592 
593 void blk_put_request(struct request *req)
594 {
595 	blk_mq_free_request(req);
596 }
597 EXPORT_SYMBOL(blk_put_request);
598 
599 bool bio_attempt_back_merge(struct request *req, struct bio *bio,
600 		unsigned int nr_segs)
601 {
602 	const int ff = bio->bi_opf & REQ_FAILFAST_MASK;
603 
604 	if (!ll_back_merge_fn(req, bio, nr_segs))
605 		return false;
606 
607 	trace_block_bio_backmerge(req->q, req, bio);
608 	rq_qos_merge(req->q, req, bio);
609 
610 	if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
611 		blk_rq_set_mixed_merge(req);
612 
613 	req->biotail->bi_next = bio;
614 	req->biotail = bio;
615 	req->__data_len += bio->bi_iter.bi_size;
616 
617 	blk_account_io_start(req, false);
618 	return true;
619 }
620 
621 bool bio_attempt_front_merge(struct request *req, struct bio *bio,
622 		unsigned int nr_segs)
623 {
624 	const int ff = bio->bi_opf & REQ_FAILFAST_MASK;
625 
626 	if (!ll_front_merge_fn(req, bio, nr_segs))
627 		return false;
628 
629 	trace_block_bio_frontmerge(req->q, req, bio);
630 	rq_qos_merge(req->q, req, bio);
631 
632 	if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
633 		blk_rq_set_mixed_merge(req);
634 
635 	bio->bi_next = req->bio;
636 	req->bio = bio;
637 
638 	req->__sector = bio->bi_iter.bi_sector;
639 	req->__data_len += bio->bi_iter.bi_size;
640 
641 	blk_account_io_start(req, false);
642 	return true;
643 }
644 
645 bool bio_attempt_discard_merge(struct request_queue *q, struct request *req,
646 		struct bio *bio)
647 {
648 	unsigned short segments = blk_rq_nr_discard_segments(req);
649 
650 	if (segments >= queue_max_discard_segments(q))
651 		goto no_merge;
652 	if (blk_rq_sectors(req) + bio_sectors(bio) >
653 	    blk_rq_get_max_sectors(req, blk_rq_pos(req)))
654 		goto no_merge;
655 
656 	rq_qos_merge(q, req, bio);
657 
658 	req->biotail->bi_next = bio;
659 	req->biotail = bio;
660 	req->__data_len += bio->bi_iter.bi_size;
661 	req->nr_phys_segments = segments + 1;
662 
663 	blk_account_io_start(req, false);
664 	return true;
665 no_merge:
666 	req_set_nomerge(q, req);
667 	return false;
668 }
669 
670 /**
671  * blk_attempt_plug_merge - try to merge with %current's plugged list
672  * @q: request_queue new bio is being queued at
673  * @bio: new bio being queued
674  * @nr_segs: number of segments in @bio
675  * @same_queue_rq: pointer to &struct request that gets filled in when
676  * another request associated with @q is found on the plug list
677  * (optional, may be %NULL)
678  *
679  * Determine whether @bio being queued on @q can be merged with a request
680  * on %current's plugged list.  Returns %true if merge was successful,
681  * otherwise %false.
682  *
683  * Plugging coalesces IOs from the same issuer for the same purpose without
684  * going through @q->queue_lock.  As such it's more of an issuing mechanism
685  * than scheduling, and the request, while may have elvpriv data, is not
686  * added on the elevator at this point.  In addition, we don't have
687  * reliable access to the elevator outside queue lock.  Only check basic
688  * merging parameters without querying the elevator.
689  *
690  * Caller must ensure !blk_queue_nomerges(q) beforehand.
691  */
692 bool blk_attempt_plug_merge(struct request_queue *q, struct bio *bio,
693 		unsigned int nr_segs, struct request **same_queue_rq)
694 {
695 	struct blk_plug *plug;
696 	struct request *rq;
697 	struct list_head *plug_list;
698 
699 	plug = blk_mq_plug(q, bio);
700 	if (!plug)
701 		return false;
702 
703 	plug_list = &plug->mq_list;
704 
705 	list_for_each_entry_reverse(rq, plug_list, queuelist) {
706 		bool merged = false;
707 
708 		if (rq->q == q && same_queue_rq) {
709 			/*
710 			 * Only blk-mq multiple hardware queues case checks the
711 			 * rq in the same queue, there should be only one such
712 			 * rq in a queue
713 			 **/
714 			*same_queue_rq = rq;
715 		}
716 
717 		if (rq->q != q || !blk_rq_merge_ok(rq, bio))
718 			continue;
719 
720 		switch (blk_try_merge(rq, bio)) {
721 		case ELEVATOR_BACK_MERGE:
722 			merged = bio_attempt_back_merge(rq, bio, nr_segs);
723 			break;
724 		case ELEVATOR_FRONT_MERGE:
725 			merged = bio_attempt_front_merge(rq, bio, nr_segs);
726 			break;
727 		case ELEVATOR_DISCARD_MERGE:
728 			merged = bio_attempt_discard_merge(q, rq, bio);
729 			break;
730 		default:
731 			break;
732 		}
733 
734 		if (merged)
735 			return true;
736 	}
737 
738 	return false;
739 }
740 
741 static void handle_bad_sector(struct bio *bio, sector_t maxsector)
742 {
743 	char b[BDEVNAME_SIZE];
744 
745 	printk(KERN_INFO "attempt to access beyond end of device\n");
746 	printk(KERN_INFO "%s: rw=%d, want=%Lu, limit=%Lu\n",
747 			bio_devname(bio, b), bio->bi_opf,
748 			(unsigned long long)bio_end_sector(bio),
749 			(long long)maxsector);
750 }
751 
752 #ifdef CONFIG_FAIL_MAKE_REQUEST
753 
754 static DECLARE_FAULT_ATTR(fail_make_request);
755 
756 static int __init setup_fail_make_request(char *str)
757 {
758 	return setup_fault_attr(&fail_make_request, str);
759 }
760 __setup("fail_make_request=", setup_fail_make_request);
761 
762 static bool should_fail_request(struct hd_struct *part, unsigned int bytes)
763 {
764 	return part->make_it_fail && should_fail(&fail_make_request, bytes);
765 }
766 
767 static int __init fail_make_request_debugfs(void)
768 {
769 	struct dentry *dir = fault_create_debugfs_attr("fail_make_request",
770 						NULL, &fail_make_request);
771 
772 	return PTR_ERR_OR_ZERO(dir);
773 }
774 
775 late_initcall(fail_make_request_debugfs);
776 
777 #else /* CONFIG_FAIL_MAKE_REQUEST */
778 
779 static inline bool should_fail_request(struct hd_struct *part,
780 					unsigned int bytes)
781 {
782 	return false;
783 }
784 
785 #endif /* CONFIG_FAIL_MAKE_REQUEST */
786 
787 static inline bool bio_check_ro(struct bio *bio, struct hd_struct *part)
788 {
789 	const int op = bio_op(bio);
790 
791 	if (part->policy && op_is_write(op)) {
792 		char b[BDEVNAME_SIZE];
793 
794 		if (op_is_flush(bio->bi_opf) && !bio_sectors(bio))
795 			return false;
796 
797 		WARN_ONCE(1,
798 		       "generic_make_request: Trying to write "
799 			"to read-only block-device %s (partno %d)\n",
800 			bio_devname(bio, b), part->partno);
801 		/* Older lvm-tools actually trigger this */
802 		return false;
803 	}
804 
805 	return false;
806 }
807 
808 static noinline int should_fail_bio(struct bio *bio)
809 {
810 	if (should_fail_request(&bio->bi_disk->part0, bio->bi_iter.bi_size))
811 		return -EIO;
812 	return 0;
813 }
814 ALLOW_ERROR_INJECTION(should_fail_bio, ERRNO);
815 
816 /*
817  * Check whether this bio extends beyond the end of the device or partition.
818  * This may well happen - the kernel calls bread() without checking the size of
819  * the device, e.g., when mounting a file system.
820  */
821 static inline int bio_check_eod(struct bio *bio, sector_t maxsector)
822 {
823 	unsigned int nr_sectors = bio_sectors(bio);
824 
825 	if (nr_sectors && maxsector &&
826 	    (nr_sectors > maxsector ||
827 	     bio->bi_iter.bi_sector > maxsector - nr_sectors)) {
828 		handle_bad_sector(bio, maxsector);
829 		return -EIO;
830 	}
831 	return 0;
832 }
833 
834 /*
835  * Remap block n of partition p to block n+start(p) of the disk.
836  */
837 static inline int blk_partition_remap(struct bio *bio)
838 {
839 	struct hd_struct *p;
840 	int ret = -EIO;
841 
842 	rcu_read_lock();
843 	p = __disk_get_part(bio->bi_disk, bio->bi_partno);
844 	if (unlikely(!p))
845 		goto out;
846 	if (unlikely(should_fail_request(p, bio->bi_iter.bi_size)))
847 		goto out;
848 	if (unlikely(bio_check_ro(bio, p)))
849 		goto out;
850 
851 	/*
852 	 * Zone reset does not include bi_size so bio_sectors() is always 0.
853 	 * Include a test for the reset op code and perform the remap if needed.
854 	 */
855 	if (bio_sectors(bio) || bio_op(bio) == REQ_OP_ZONE_RESET) {
856 		if (bio_check_eod(bio, part_nr_sects_read(p)))
857 			goto out;
858 		bio->bi_iter.bi_sector += p->start_sect;
859 		trace_block_bio_remap(bio->bi_disk->queue, bio, part_devt(p),
860 				      bio->bi_iter.bi_sector - p->start_sect);
861 	}
862 	bio->bi_partno = 0;
863 	ret = 0;
864 out:
865 	rcu_read_unlock();
866 	return ret;
867 }
868 
869 static noinline_for_stack bool
870 generic_make_request_checks(struct bio *bio)
871 {
872 	struct request_queue *q;
873 	int nr_sectors = bio_sectors(bio);
874 	blk_status_t status = BLK_STS_IOERR;
875 	char b[BDEVNAME_SIZE];
876 
877 	might_sleep();
878 
879 	q = bio->bi_disk->queue;
880 	if (unlikely(!q)) {
881 		printk(KERN_ERR
882 		       "generic_make_request: Trying to access "
883 			"nonexistent block-device %s (%Lu)\n",
884 			bio_devname(bio, b), (long long)bio->bi_iter.bi_sector);
885 		goto end_io;
886 	}
887 
888 	/*
889 	 * For a REQ_NOWAIT based request, return -EOPNOTSUPP
890 	 * if queue is not a request based queue.
891 	 */
892 	if ((bio->bi_opf & REQ_NOWAIT) && !queue_is_mq(q))
893 		goto not_supported;
894 
895 	if (should_fail_bio(bio))
896 		goto end_io;
897 
898 	if (bio->bi_partno) {
899 		if (unlikely(blk_partition_remap(bio)))
900 			goto end_io;
901 	} else {
902 		if (unlikely(bio_check_ro(bio, &bio->bi_disk->part0)))
903 			goto end_io;
904 		if (unlikely(bio_check_eod(bio, get_capacity(bio->bi_disk))))
905 			goto end_io;
906 	}
907 
908 	/*
909 	 * Filter flush bio's early so that make_request based
910 	 * drivers without flush support don't have to worry
911 	 * about them.
912 	 */
913 	if (op_is_flush(bio->bi_opf) &&
914 	    !test_bit(QUEUE_FLAG_WC, &q->queue_flags)) {
915 		bio->bi_opf &= ~(REQ_PREFLUSH | REQ_FUA);
916 		if (!nr_sectors) {
917 			status = BLK_STS_OK;
918 			goto end_io;
919 		}
920 	}
921 
922 	if (!test_bit(QUEUE_FLAG_POLL, &q->queue_flags))
923 		bio->bi_opf &= ~REQ_HIPRI;
924 
925 	switch (bio_op(bio)) {
926 	case REQ_OP_DISCARD:
927 		if (!blk_queue_discard(q))
928 			goto not_supported;
929 		break;
930 	case REQ_OP_SECURE_ERASE:
931 		if (!blk_queue_secure_erase(q))
932 			goto not_supported;
933 		break;
934 	case REQ_OP_WRITE_SAME:
935 		if (!q->limits.max_write_same_sectors)
936 			goto not_supported;
937 		break;
938 	case REQ_OP_ZONE_RESET:
939 		if (!blk_queue_is_zoned(q))
940 			goto not_supported;
941 		break;
942 	case REQ_OP_ZONE_RESET_ALL:
943 		if (!blk_queue_is_zoned(q) || !blk_queue_zone_resetall(q))
944 			goto not_supported;
945 		break;
946 	case REQ_OP_WRITE_ZEROES:
947 		if (!q->limits.max_write_zeroes_sectors)
948 			goto not_supported;
949 		break;
950 	default:
951 		break;
952 	}
953 
954 	/*
955 	 * Various block parts want %current->io_context and lazy ioc
956 	 * allocation ends up trading a lot of pain for a small amount of
957 	 * memory.  Just allocate it upfront.  This may fail and block
958 	 * layer knows how to live with it.
959 	 */
960 	create_io_context(GFP_ATOMIC, q->node);
961 
962 	if (!blkcg_bio_issue_check(q, bio))
963 		return false;
964 
965 	if (!bio_flagged(bio, BIO_TRACE_COMPLETION)) {
966 		trace_block_bio_queue(q, bio);
967 		/* Now that enqueuing has been traced, we need to trace
968 		 * completion as well.
969 		 */
970 		bio_set_flag(bio, BIO_TRACE_COMPLETION);
971 	}
972 	return true;
973 
974 not_supported:
975 	status = BLK_STS_NOTSUPP;
976 end_io:
977 	bio->bi_status = status;
978 	bio_endio(bio);
979 	return false;
980 }
981 
982 /**
983  * generic_make_request - hand a buffer to its device driver for I/O
984  * @bio:  The bio describing the location in memory and on the device.
985  *
986  * generic_make_request() is used to make I/O requests of block
987  * devices. It is passed a &struct bio, which describes the I/O that needs
988  * to be done.
989  *
990  * generic_make_request() does not return any status.  The
991  * success/failure status of the request, along with notification of
992  * completion, is delivered asynchronously through the bio->bi_end_io
993  * function described (one day) else where.
994  *
995  * The caller of generic_make_request must make sure that bi_io_vec
996  * are set to describe the memory buffer, and that bi_dev and bi_sector are
997  * set to describe the device address, and the
998  * bi_end_io and optionally bi_private are set to describe how
999  * completion notification should be signaled.
1000  *
1001  * generic_make_request and the drivers it calls may use bi_next if this
1002  * bio happens to be merged with someone else, and may resubmit the bio to
1003  * a lower device by calling into generic_make_request recursively, which
1004  * means the bio should NOT be touched after the call to ->make_request_fn.
1005  */
1006 blk_qc_t generic_make_request(struct bio *bio)
1007 {
1008 	/*
1009 	 * bio_list_on_stack[0] contains bios submitted by the current
1010 	 * make_request_fn.
1011 	 * bio_list_on_stack[1] contains bios that were submitted before
1012 	 * the current make_request_fn, but that haven't been processed
1013 	 * yet.
1014 	 */
1015 	struct bio_list bio_list_on_stack[2];
1016 	blk_qc_t ret = BLK_QC_T_NONE;
1017 
1018 	if (!generic_make_request_checks(bio))
1019 		goto out;
1020 
1021 	/*
1022 	 * We only want one ->make_request_fn to be active at a time, else
1023 	 * stack usage with stacked devices could be a problem.  So use
1024 	 * current->bio_list to keep a list of requests submited by a
1025 	 * make_request_fn function.  current->bio_list is also used as a
1026 	 * flag to say if generic_make_request is currently active in this
1027 	 * task or not.  If it is NULL, then no make_request is active.  If
1028 	 * it is non-NULL, then a make_request is active, and new requests
1029 	 * should be added at the tail
1030 	 */
1031 	if (current->bio_list) {
1032 		bio_list_add(&current->bio_list[0], bio);
1033 		goto out;
1034 	}
1035 
1036 	/* following loop may be a bit non-obvious, and so deserves some
1037 	 * explanation.
1038 	 * Before entering the loop, bio->bi_next is NULL (as all callers
1039 	 * ensure that) so we have a list with a single bio.
1040 	 * We pretend that we have just taken it off a longer list, so
1041 	 * we assign bio_list to a pointer to the bio_list_on_stack,
1042 	 * thus initialising the bio_list of new bios to be
1043 	 * added.  ->make_request() may indeed add some more bios
1044 	 * through a recursive call to generic_make_request.  If it
1045 	 * did, we find a non-NULL value in bio_list and re-enter the loop
1046 	 * from the top.  In this case we really did just take the bio
1047 	 * of the top of the list (no pretending) and so remove it from
1048 	 * bio_list, and call into ->make_request() again.
1049 	 */
1050 	BUG_ON(bio->bi_next);
1051 	bio_list_init(&bio_list_on_stack[0]);
1052 	current->bio_list = bio_list_on_stack;
1053 	do {
1054 		struct request_queue *q = bio->bi_disk->queue;
1055 		blk_mq_req_flags_t flags = bio->bi_opf & REQ_NOWAIT ?
1056 			BLK_MQ_REQ_NOWAIT : 0;
1057 
1058 		if (likely(blk_queue_enter(q, flags) == 0)) {
1059 			struct bio_list lower, same;
1060 
1061 			/* Create a fresh bio_list for all subordinate requests */
1062 			bio_list_on_stack[1] = bio_list_on_stack[0];
1063 			bio_list_init(&bio_list_on_stack[0]);
1064 			ret = q->make_request_fn(q, bio);
1065 
1066 			blk_queue_exit(q);
1067 
1068 			/* sort new bios into those for a lower level
1069 			 * and those for the same level
1070 			 */
1071 			bio_list_init(&lower);
1072 			bio_list_init(&same);
1073 			while ((bio = bio_list_pop(&bio_list_on_stack[0])) != NULL)
1074 				if (q == bio->bi_disk->queue)
1075 					bio_list_add(&same, bio);
1076 				else
1077 					bio_list_add(&lower, bio);
1078 			/* now assemble so we handle the lowest level first */
1079 			bio_list_merge(&bio_list_on_stack[0], &lower);
1080 			bio_list_merge(&bio_list_on_stack[0], &same);
1081 			bio_list_merge(&bio_list_on_stack[0], &bio_list_on_stack[1]);
1082 		} else {
1083 			if (unlikely(!blk_queue_dying(q) &&
1084 					(bio->bi_opf & REQ_NOWAIT)))
1085 				bio_wouldblock_error(bio);
1086 			else
1087 				bio_io_error(bio);
1088 		}
1089 		bio = bio_list_pop(&bio_list_on_stack[0]);
1090 	} while (bio);
1091 	current->bio_list = NULL; /* deactivate */
1092 
1093 out:
1094 	return ret;
1095 }
1096 EXPORT_SYMBOL(generic_make_request);
1097 
1098 /**
1099  * direct_make_request - hand a buffer directly to its device driver for I/O
1100  * @bio:  The bio describing the location in memory and on the device.
1101  *
1102  * This function behaves like generic_make_request(), but does not protect
1103  * against recursion.  Must only be used if the called driver is known
1104  * to not call generic_make_request (or direct_make_request) again from
1105  * its make_request function.  (Calling direct_make_request again from
1106  * a workqueue is perfectly fine as that doesn't recurse).
1107  */
1108 blk_qc_t direct_make_request(struct bio *bio)
1109 {
1110 	struct request_queue *q = bio->bi_disk->queue;
1111 	bool nowait = bio->bi_opf & REQ_NOWAIT;
1112 	blk_qc_t ret;
1113 
1114 	if (!generic_make_request_checks(bio))
1115 		return BLK_QC_T_NONE;
1116 
1117 	if (unlikely(blk_queue_enter(q, nowait ? BLK_MQ_REQ_NOWAIT : 0))) {
1118 		if (nowait && !blk_queue_dying(q))
1119 			bio->bi_status = BLK_STS_AGAIN;
1120 		else
1121 			bio->bi_status = BLK_STS_IOERR;
1122 		bio_endio(bio);
1123 		return BLK_QC_T_NONE;
1124 	}
1125 
1126 	ret = q->make_request_fn(q, bio);
1127 	blk_queue_exit(q);
1128 	return ret;
1129 }
1130 EXPORT_SYMBOL_GPL(direct_make_request);
1131 
1132 /**
1133  * submit_bio - submit a bio to the block device layer for I/O
1134  * @bio: The &struct bio which describes the I/O
1135  *
1136  * submit_bio() is very similar in purpose to generic_make_request(), and
1137  * uses that function to do most of the work. Both are fairly rough
1138  * interfaces; @bio must be presetup and ready for I/O.
1139  *
1140  */
1141 blk_qc_t submit_bio(struct bio *bio)
1142 {
1143 	bool workingset_read = false;
1144 	unsigned long pflags;
1145 	blk_qc_t ret;
1146 
1147 	if (blkcg_punt_bio_submit(bio))
1148 		return BLK_QC_T_NONE;
1149 
1150 	/*
1151 	 * If it's a regular read/write or a barrier with data attached,
1152 	 * go through the normal accounting stuff before submission.
1153 	 */
1154 	if (bio_has_data(bio)) {
1155 		unsigned int count;
1156 
1157 		if (unlikely(bio_op(bio) == REQ_OP_WRITE_SAME))
1158 			count = queue_logical_block_size(bio->bi_disk->queue) >> 9;
1159 		else
1160 			count = bio_sectors(bio);
1161 
1162 		if (op_is_write(bio_op(bio))) {
1163 			count_vm_events(PGPGOUT, count);
1164 		} else {
1165 			if (bio_flagged(bio, BIO_WORKINGSET))
1166 				workingset_read = true;
1167 			task_io_account_read(bio->bi_iter.bi_size);
1168 			count_vm_events(PGPGIN, count);
1169 		}
1170 
1171 		if (unlikely(block_dump)) {
1172 			char b[BDEVNAME_SIZE];
1173 			printk(KERN_DEBUG "%s(%d): %s block %Lu on %s (%u sectors)\n",
1174 			current->comm, task_pid_nr(current),
1175 				op_is_write(bio_op(bio)) ? "WRITE" : "READ",
1176 				(unsigned long long)bio->bi_iter.bi_sector,
1177 				bio_devname(bio, b), count);
1178 		}
1179 	}
1180 
1181 	/*
1182 	 * If we're reading data that is part of the userspace
1183 	 * workingset, count submission time as memory stall. When the
1184 	 * device is congested, or the submitting cgroup IO-throttled,
1185 	 * submission can be a significant part of overall IO time.
1186 	 */
1187 	if (workingset_read)
1188 		psi_memstall_enter(&pflags);
1189 
1190 	ret = generic_make_request(bio);
1191 
1192 	if (workingset_read)
1193 		psi_memstall_leave(&pflags);
1194 
1195 	return ret;
1196 }
1197 EXPORT_SYMBOL(submit_bio);
1198 
1199 /**
1200  * blk_cloned_rq_check_limits - Helper function to check a cloned request
1201  *                              for new the queue limits
1202  * @q:  the queue
1203  * @rq: the request being checked
1204  *
1205  * Description:
1206  *    @rq may have been made based on weaker limitations of upper-level queues
1207  *    in request stacking drivers, and it may violate the limitation of @q.
1208  *    Since the block layer and the underlying device driver trust @rq
1209  *    after it is inserted to @q, it should be checked against @q before
1210  *    the insertion using this generic function.
1211  *
1212  *    Request stacking drivers like request-based dm may change the queue
1213  *    limits when retrying requests on other queues. Those requests need
1214  *    to be checked against the new queue limits again during dispatch.
1215  */
1216 static int blk_cloned_rq_check_limits(struct request_queue *q,
1217 				      struct request *rq)
1218 {
1219 	if (blk_rq_sectors(rq) > blk_queue_get_max_sectors(q, req_op(rq))) {
1220 		printk(KERN_ERR "%s: over max size limit. (%u > %u)\n",
1221 			__func__, blk_rq_sectors(rq),
1222 			blk_queue_get_max_sectors(q, req_op(rq)));
1223 		return -EIO;
1224 	}
1225 
1226 	/*
1227 	 * queue's settings related to segment counting like q->bounce_pfn
1228 	 * may differ from that of other stacking queues.
1229 	 * Recalculate it to check the request correctly on this queue's
1230 	 * limitation.
1231 	 */
1232 	rq->nr_phys_segments = blk_recalc_rq_segments(rq);
1233 	if (rq->nr_phys_segments > queue_max_segments(q)) {
1234 		printk(KERN_ERR "%s: over max segments limit. (%hu > %hu)\n",
1235 			__func__, rq->nr_phys_segments, queue_max_segments(q));
1236 		return -EIO;
1237 	}
1238 
1239 	return 0;
1240 }
1241 
1242 /**
1243  * blk_insert_cloned_request - Helper for stacking drivers to submit a request
1244  * @q:  the queue to submit the request
1245  * @rq: the request being queued
1246  */
1247 blk_status_t blk_insert_cloned_request(struct request_queue *q, struct request *rq)
1248 {
1249 	if (blk_cloned_rq_check_limits(q, rq))
1250 		return BLK_STS_IOERR;
1251 
1252 	if (rq->rq_disk &&
1253 	    should_fail_request(&rq->rq_disk->part0, blk_rq_bytes(rq)))
1254 		return BLK_STS_IOERR;
1255 
1256 	if (blk_queue_io_stat(q))
1257 		blk_account_io_start(rq, true);
1258 
1259 	/*
1260 	 * Since we have a scheduler attached on the top device,
1261 	 * bypass a potential scheduler on the bottom device for
1262 	 * insert.
1263 	 */
1264 	return blk_mq_request_issue_directly(rq, true);
1265 }
1266 EXPORT_SYMBOL_GPL(blk_insert_cloned_request);
1267 
1268 /**
1269  * blk_rq_err_bytes - determine number of bytes till the next failure boundary
1270  * @rq: request to examine
1271  *
1272  * Description:
1273  *     A request could be merge of IOs which require different failure
1274  *     handling.  This function determines the number of bytes which
1275  *     can be failed from the beginning of the request without
1276  *     crossing into area which need to be retried further.
1277  *
1278  * Return:
1279  *     The number of bytes to fail.
1280  */
1281 unsigned int blk_rq_err_bytes(const struct request *rq)
1282 {
1283 	unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK;
1284 	unsigned int bytes = 0;
1285 	struct bio *bio;
1286 
1287 	if (!(rq->rq_flags & RQF_MIXED_MERGE))
1288 		return blk_rq_bytes(rq);
1289 
1290 	/*
1291 	 * Currently the only 'mixing' which can happen is between
1292 	 * different fastfail types.  We can safely fail portions
1293 	 * which have all the failfast bits that the first one has -
1294 	 * the ones which are at least as eager to fail as the first
1295 	 * one.
1296 	 */
1297 	for (bio = rq->bio; bio; bio = bio->bi_next) {
1298 		if ((bio->bi_opf & ff) != ff)
1299 			break;
1300 		bytes += bio->bi_iter.bi_size;
1301 	}
1302 
1303 	/* this could lead to infinite loop */
1304 	BUG_ON(blk_rq_bytes(rq) && !bytes);
1305 	return bytes;
1306 }
1307 EXPORT_SYMBOL_GPL(blk_rq_err_bytes);
1308 
1309 void blk_account_io_completion(struct request *req, unsigned int bytes)
1310 {
1311 	if (blk_do_io_stat(req)) {
1312 		const int sgrp = op_stat_group(req_op(req));
1313 		struct hd_struct *part;
1314 
1315 		part_stat_lock();
1316 		part = req->part;
1317 		part_stat_add(part, sectors[sgrp], bytes >> 9);
1318 		part_stat_unlock();
1319 	}
1320 }
1321 
1322 void blk_account_io_done(struct request *req, u64 now)
1323 {
1324 	/*
1325 	 * Account IO completion.  flush_rq isn't accounted as a
1326 	 * normal IO on queueing nor completion.  Accounting the
1327 	 * containing request is enough.
1328 	 */
1329 	if (blk_do_io_stat(req) && !(req->rq_flags & RQF_FLUSH_SEQ)) {
1330 		const int sgrp = op_stat_group(req_op(req));
1331 		struct hd_struct *part;
1332 
1333 		part_stat_lock();
1334 		part = req->part;
1335 
1336 		update_io_ticks(part, jiffies);
1337 		part_stat_inc(part, ios[sgrp]);
1338 		part_stat_add(part, nsecs[sgrp], now - req->start_time_ns);
1339 		part_stat_add(part, time_in_queue, nsecs_to_jiffies64(now - req->start_time_ns));
1340 		part_dec_in_flight(req->q, part, rq_data_dir(req));
1341 
1342 		hd_struct_put(part);
1343 		part_stat_unlock();
1344 	}
1345 }
1346 
1347 void blk_account_io_start(struct request *rq, bool new_io)
1348 {
1349 	struct hd_struct *part;
1350 	int rw = rq_data_dir(rq);
1351 
1352 	if (!blk_do_io_stat(rq))
1353 		return;
1354 
1355 	part_stat_lock();
1356 
1357 	if (!new_io) {
1358 		part = rq->part;
1359 		part_stat_inc(part, merges[rw]);
1360 	} else {
1361 		part = disk_map_sector_rcu(rq->rq_disk, blk_rq_pos(rq));
1362 		if (!hd_struct_try_get(part)) {
1363 			/*
1364 			 * The partition is already being removed,
1365 			 * the request will be accounted on the disk only
1366 			 *
1367 			 * We take a reference on disk->part0 although that
1368 			 * partition will never be deleted, so we can treat
1369 			 * it as any other partition.
1370 			 */
1371 			part = &rq->rq_disk->part0;
1372 			hd_struct_get(part);
1373 		}
1374 		part_inc_in_flight(rq->q, part, rw);
1375 		rq->part = part;
1376 	}
1377 
1378 	update_io_ticks(part, jiffies);
1379 
1380 	part_stat_unlock();
1381 }
1382 
1383 /*
1384  * Steal bios from a request and add them to a bio list.
1385  * The request must not have been partially completed before.
1386  */
1387 void blk_steal_bios(struct bio_list *list, struct request *rq)
1388 {
1389 	if (rq->bio) {
1390 		if (list->tail)
1391 			list->tail->bi_next = rq->bio;
1392 		else
1393 			list->head = rq->bio;
1394 		list->tail = rq->biotail;
1395 
1396 		rq->bio = NULL;
1397 		rq->biotail = NULL;
1398 	}
1399 
1400 	rq->__data_len = 0;
1401 }
1402 EXPORT_SYMBOL_GPL(blk_steal_bios);
1403 
1404 /**
1405  * blk_update_request - Special helper function for request stacking drivers
1406  * @req:      the request being processed
1407  * @error:    block status code
1408  * @nr_bytes: number of bytes to complete @req
1409  *
1410  * Description:
1411  *     Ends I/O on a number of bytes attached to @req, but doesn't complete
1412  *     the request structure even if @req doesn't have leftover.
1413  *     If @req has leftover, sets it up for the next range of segments.
1414  *
1415  *     This special helper function is only for request stacking drivers
1416  *     (e.g. request-based dm) so that they can handle partial completion.
1417  *     Actual device drivers should use blk_mq_end_request instead.
1418  *
1419  *     Passing the result of blk_rq_bytes() as @nr_bytes guarantees
1420  *     %false return from this function.
1421  *
1422  * Note:
1423  *	The RQF_SPECIAL_PAYLOAD flag is ignored on purpose in both
1424  *	blk_rq_bytes() and in blk_update_request().
1425  *
1426  * Return:
1427  *     %false - this request doesn't have any more data
1428  *     %true  - this request has more data
1429  **/
1430 bool blk_update_request(struct request *req, blk_status_t error,
1431 		unsigned int nr_bytes)
1432 {
1433 	int total_bytes;
1434 
1435 	trace_block_rq_complete(req, blk_status_to_errno(error), nr_bytes);
1436 
1437 	if (!req->bio)
1438 		return false;
1439 
1440 #ifdef CONFIG_BLK_DEV_INTEGRITY
1441 	if (blk_integrity_rq(req) && req_op(req) == REQ_OP_READ &&
1442 	    error == BLK_STS_OK)
1443 		req->q->integrity.profile->complete_fn(req, nr_bytes);
1444 #endif
1445 
1446 	if (unlikely(error && !blk_rq_is_passthrough(req) &&
1447 		     !(req->rq_flags & RQF_QUIET)))
1448 		print_req_error(req, error, __func__);
1449 
1450 	blk_account_io_completion(req, nr_bytes);
1451 
1452 	total_bytes = 0;
1453 	while (req->bio) {
1454 		struct bio *bio = req->bio;
1455 		unsigned bio_bytes = min(bio->bi_iter.bi_size, nr_bytes);
1456 
1457 		if (bio_bytes == bio->bi_iter.bi_size)
1458 			req->bio = bio->bi_next;
1459 
1460 		/* Completion has already been traced */
1461 		bio_clear_flag(bio, BIO_TRACE_COMPLETION);
1462 		req_bio_endio(req, bio, bio_bytes, error);
1463 
1464 		total_bytes += bio_bytes;
1465 		nr_bytes -= bio_bytes;
1466 
1467 		if (!nr_bytes)
1468 			break;
1469 	}
1470 
1471 	/*
1472 	 * completely done
1473 	 */
1474 	if (!req->bio) {
1475 		/*
1476 		 * Reset counters so that the request stacking driver
1477 		 * can find how many bytes remain in the request
1478 		 * later.
1479 		 */
1480 		req->__data_len = 0;
1481 		return false;
1482 	}
1483 
1484 	req->__data_len -= total_bytes;
1485 
1486 	/* update sector only for requests with clear definition of sector */
1487 	if (!blk_rq_is_passthrough(req))
1488 		req->__sector += total_bytes >> 9;
1489 
1490 	/* mixed attributes always follow the first bio */
1491 	if (req->rq_flags & RQF_MIXED_MERGE) {
1492 		req->cmd_flags &= ~REQ_FAILFAST_MASK;
1493 		req->cmd_flags |= req->bio->bi_opf & REQ_FAILFAST_MASK;
1494 	}
1495 
1496 	if (!(req->rq_flags & RQF_SPECIAL_PAYLOAD)) {
1497 		/*
1498 		 * If total number of sectors is less than the first segment
1499 		 * size, something has gone terribly wrong.
1500 		 */
1501 		if (blk_rq_bytes(req) < blk_rq_cur_bytes(req)) {
1502 			blk_dump_rq_flags(req, "request botched");
1503 			req->__data_len = blk_rq_cur_bytes(req);
1504 		}
1505 
1506 		/* recalculate the number of segments */
1507 		req->nr_phys_segments = blk_recalc_rq_segments(req);
1508 	}
1509 
1510 	return true;
1511 }
1512 EXPORT_SYMBOL_GPL(blk_update_request);
1513 
1514 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
1515 /**
1516  * rq_flush_dcache_pages - Helper function to flush all pages in a request
1517  * @rq: the request to be flushed
1518  *
1519  * Description:
1520  *     Flush all pages in @rq.
1521  */
1522 void rq_flush_dcache_pages(struct request *rq)
1523 {
1524 	struct req_iterator iter;
1525 	struct bio_vec bvec;
1526 
1527 	rq_for_each_segment(bvec, rq, iter)
1528 		flush_dcache_page(bvec.bv_page);
1529 }
1530 EXPORT_SYMBOL_GPL(rq_flush_dcache_pages);
1531 #endif
1532 
1533 /**
1534  * blk_lld_busy - Check if underlying low-level drivers of a device are busy
1535  * @q : the queue of the device being checked
1536  *
1537  * Description:
1538  *    Check if underlying low-level drivers of a device are busy.
1539  *    If the drivers want to export their busy state, they must set own
1540  *    exporting function using blk_queue_lld_busy() first.
1541  *
1542  *    Basically, this function is used only by request stacking drivers
1543  *    to stop dispatching requests to underlying devices when underlying
1544  *    devices are busy.  This behavior helps more I/O merging on the queue
1545  *    of the request stacking driver and prevents I/O throughput regression
1546  *    on burst I/O load.
1547  *
1548  * Return:
1549  *    0 - Not busy (The request stacking driver should dispatch request)
1550  *    1 - Busy (The request stacking driver should stop dispatching request)
1551  */
1552 int blk_lld_busy(struct request_queue *q)
1553 {
1554 	if (queue_is_mq(q) && q->mq_ops->busy)
1555 		return q->mq_ops->busy(q);
1556 
1557 	return 0;
1558 }
1559 EXPORT_SYMBOL_GPL(blk_lld_busy);
1560 
1561 /**
1562  * blk_rq_unprep_clone - Helper function to free all bios in a cloned request
1563  * @rq: the clone request to be cleaned up
1564  *
1565  * Description:
1566  *     Free all bios in @rq for a cloned request.
1567  */
1568 void blk_rq_unprep_clone(struct request *rq)
1569 {
1570 	struct bio *bio;
1571 
1572 	while ((bio = rq->bio) != NULL) {
1573 		rq->bio = bio->bi_next;
1574 
1575 		bio_put(bio);
1576 	}
1577 }
1578 EXPORT_SYMBOL_GPL(blk_rq_unprep_clone);
1579 
1580 /*
1581  * Copy attributes of the original request to the clone request.
1582  * The actual data parts (e.g. ->cmd, ->sense) are not copied.
1583  */
1584 static void __blk_rq_prep_clone(struct request *dst, struct request *src)
1585 {
1586 	dst->__sector = blk_rq_pos(src);
1587 	dst->__data_len = blk_rq_bytes(src);
1588 	if (src->rq_flags & RQF_SPECIAL_PAYLOAD) {
1589 		dst->rq_flags |= RQF_SPECIAL_PAYLOAD;
1590 		dst->special_vec = src->special_vec;
1591 	}
1592 	dst->nr_phys_segments = src->nr_phys_segments;
1593 	dst->ioprio = src->ioprio;
1594 	dst->extra_len = src->extra_len;
1595 }
1596 
1597 /**
1598  * blk_rq_prep_clone - Helper function to setup clone request
1599  * @rq: the request to be setup
1600  * @rq_src: original request to be cloned
1601  * @bs: bio_set that bios for clone are allocated from
1602  * @gfp_mask: memory allocation mask for bio
1603  * @bio_ctr: setup function to be called for each clone bio.
1604  *           Returns %0 for success, non %0 for failure.
1605  * @data: private data to be passed to @bio_ctr
1606  *
1607  * Description:
1608  *     Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq.
1609  *     The actual data parts of @rq_src (e.g. ->cmd, ->sense)
1610  *     are not copied, and copying such parts is the caller's responsibility.
1611  *     Also, pages which the original bios are pointing to are not copied
1612  *     and the cloned bios just point same pages.
1613  *     So cloned bios must be completed before original bios, which means
1614  *     the caller must complete @rq before @rq_src.
1615  */
1616 int blk_rq_prep_clone(struct request *rq, struct request *rq_src,
1617 		      struct bio_set *bs, gfp_t gfp_mask,
1618 		      int (*bio_ctr)(struct bio *, struct bio *, void *),
1619 		      void *data)
1620 {
1621 	struct bio *bio, *bio_src;
1622 
1623 	if (!bs)
1624 		bs = &fs_bio_set;
1625 
1626 	__rq_for_each_bio(bio_src, rq_src) {
1627 		bio = bio_clone_fast(bio_src, gfp_mask, bs);
1628 		if (!bio)
1629 			goto free_and_out;
1630 
1631 		if (bio_ctr && bio_ctr(bio, bio_src, data))
1632 			goto free_and_out;
1633 
1634 		if (rq->bio) {
1635 			rq->biotail->bi_next = bio;
1636 			rq->biotail = bio;
1637 		} else
1638 			rq->bio = rq->biotail = bio;
1639 	}
1640 
1641 	__blk_rq_prep_clone(rq, rq_src);
1642 
1643 	return 0;
1644 
1645 free_and_out:
1646 	if (bio)
1647 		bio_put(bio);
1648 	blk_rq_unprep_clone(rq);
1649 
1650 	return -ENOMEM;
1651 }
1652 EXPORT_SYMBOL_GPL(blk_rq_prep_clone);
1653 
1654 int kblockd_schedule_work(struct work_struct *work)
1655 {
1656 	return queue_work(kblockd_workqueue, work);
1657 }
1658 EXPORT_SYMBOL(kblockd_schedule_work);
1659 
1660 int kblockd_schedule_work_on(int cpu, struct work_struct *work)
1661 {
1662 	return queue_work_on(cpu, kblockd_workqueue, work);
1663 }
1664 EXPORT_SYMBOL(kblockd_schedule_work_on);
1665 
1666 int kblockd_mod_delayed_work_on(int cpu, struct delayed_work *dwork,
1667 				unsigned long delay)
1668 {
1669 	return mod_delayed_work_on(cpu, kblockd_workqueue, dwork, delay);
1670 }
1671 EXPORT_SYMBOL(kblockd_mod_delayed_work_on);
1672 
1673 /**
1674  * blk_start_plug - initialize blk_plug and track it inside the task_struct
1675  * @plug:	The &struct blk_plug that needs to be initialized
1676  *
1677  * Description:
1678  *   blk_start_plug() indicates to the block layer an intent by the caller
1679  *   to submit multiple I/O requests in a batch.  The block layer may use
1680  *   this hint to defer submitting I/Os from the caller until blk_finish_plug()
1681  *   is called.  However, the block layer may choose to submit requests
1682  *   before a call to blk_finish_plug() if the number of queued I/Os
1683  *   exceeds %BLK_MAX_REQUEST_COUNT, or if the size of the I/O is larger than
1684  *   %BLK_PLUG_FLUSH_SIZE.  The queued I/Os may also be submitted early if
1685  *   the task schedules (see below).
1686  *
1687  *   Tracking blk_plug inside the task_struct will help with auto-flushing the
1688  *   pending I/O should the task end up blocking between blk_start_plug() and
1689  *   blk_finish_plug(). This is important from a performance perspective, but
1690  *   also ensures that we don't deadlock. For instance, if the task is blocking
1691  *   for a memory allocation, memory reclaim could end up wanting to free a
1692  *   page belonging to that request that is currently residing in our private
1693  *   plug. By flushing the pending I/O when the process goes to sleep, we avoid
1694  *   this kind of deadlock.
1695  */
1696 void blk_start_plug(struct blk_plug *plug)
1697 {
1698 	struct task_struct *tsk = current;
1699 
1700 	/*
1701 	 * If this is a nested plug, don't actually assign it.
1702 	 */
1703 	if (tsk->plug)
1704 		return;
1705 
1706 	INIT_LIST_HEAD(&plug->mq_list);
1707 	INIT_LIST_HEAD(&plug->cb_list);
1708 	plug->rq_count = 0;
1709 	plug->multiple_queues = false;
1710 
1711 	/*
1712 	 * Store ordering should not be needed here, since a potential
1713 	 * preempt will imply a full memory barrier
1714 	 */
1715 	tsk->plug = plug;
1716 }
1717 EXPORT_SYMBOL(blk_start_plug);
1718 
1719 static void flush_plug_callbacks(struct blk_plug *plug, bool from_schedule)
1720 {
1721 	LIST_HEAD(callbacks);
1722 
1723 	while (!list_empty(&plug->cb_list)) {
1724 		list_splice_init(&plug->cb_list, &callbacks);
1725 
1726 		while (!list_empty(&callbacks)) {
1727 			struct blk_plug_cb *cb = list_first_entry(&callbacks,
1728 							  struct blk_plug_cb,
1729 							  list);
1730 			list_del(&cb->list);
1731 			cb->callback(cb, from_schedule);
1732 		}
1733 	}
1734 }
1735 
1736 struct blk_plug_cb *blk_check_plugged(blk_plug_cb_fn unplug, void *data,
1737 				      int size)
1738 {
1739 	struct blk_plug *plug = current->plug;
1740 	struct blk_plug_cb *cb;
1741 
1742 	if (!plug)
1743 		return NULL;
1744 
1745 	list_for_each_entry(cb, &plug->cb_list, list)
1746 		if (cb->callback == unplug && cb->data == data)
1747 			return cb;
1748 
1749 	/* Not currently on the callback list */
1750 	BUG_ON(size < sizeof(*cb));
1751 	cb = kzalloc(size, GFP_ATOMIC);
1752 	if (cb) {
1753 		cb->data = data;
1754 		cb->callback = unplug;
1755 		list_add(&cb->list, &plug->cb_list);
1756 	}
1757 	return cb;
1758 }
1759 EXPORT_SYMBOL(blk_check_plugged);
1760 
1761 void blk_flush_plug_list(struct blk_plug *plug, bool from_schedule)
1762 {
1763 	flush_plug_callbacks(plug, from_schedule);
1764 
1765 	if (!list_empty(&plug->mq_list))
1766 		blk_mq_flush_plug_list(plug, from_schedule);
1767 }
1768 
1769 /**
1770  * blk_finish_plug - mark the end of a batch of submitted I/O
1771  * @plug:	The &struct blk_plug passed to blk_start_plug()
1772  *
1773  * Description:
1774  * Indicate that a batch of I/O submissions is complete.  This function
1775  * must be paired with an initial call to blk_start_plug().  The intent
1776  * is to allow the block layer to optimize I/O submission.  See the
1777  * documentation for blk_start_plug() for more information.
1778  */
1779 void blk_finish_plug(struct blk_plug *plug)
1780 {
1781 	if (plug != current->plug)
1782 		return;
1783 	blk_flush_plug_list(plug, false);
1784 
1785 	current->plug = NULL;
1786 }
1787 EXPORT_SYMBOL(blk_finish_plug);
1788 
1789 int __init blk_dev_init(void)
1790 {
1791 	BUILD_BUG_ON(REQ_OP_LAST >= (1 << REQ_OP_BITS));
1792 	BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 *
1793 			FIELD_SIZEOF(struct request, cmd_flags));
1794 	BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 *
1795 			FIELD_SIZEOF(struct bio, bi_opf));
1796 
1797 	/* used for unplugging and affects IO latency/throughput - HIGHPRI */
1798 	kblockd_workqueue = alloc_workqueue("kblockd",
1799 					    WQ_MEM_RECLAIM | WQ_HIGHPRI, 0);
1800 	if (!kblockd_workqueue)
1801 		panic("Failed to create kblockd\n");
1802 
1803 	blk_requestq_cachep = kmem_cache_create("request_queue",
1804 			sizeof(struct request_queue), 0, SLAB_PANIC, NULL);
1805 
1806 #ifdef CONFIG_DEBUG_FS
1807 	blk_debugfs_root = debugfs_create_dir("block", NULL);
1808 #endif
1809 
1810 	return 0;
1811 }
1812