xref: /openbmc/linux/block/blk-core.c (revision dc6a81c3)
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 	blk_queue_flag_set(QUEUE_FLAG_DYING, q);
350 
351 	/*
352 	 * Drain all requests queued before DYING marking. Set DEAD flag to
353 	 * prevent that blk_mq_run_hw_queues() accesses the hardware queues
354 	 * after draining finished.
355 	 */
356 	blk_freeze_queue(q);
357 
358 	rq_qos_exit(q);
359 
360 	blk_queue_flag_set(QUEUE_FLAG_DEAD, q);
361 
362 	/* for synchronous bio-based driver finish in-flight integrity i/o */
363 	blk_flush_integrity();
364 
365 	/* @q won't process any more request, flush async actions */
366 	del_timer_sync(&q->backing_dev_info->laptop_mode_wb_timer);
367 	blk_sync_queue(q);
368 
369 	if (queue_is_mq(q))
370 		blk_mq_exit_queue(q);
371 
372 	/*
373 	 * In theory, request pool of sched_tags belongs to request queue.
374 	 * However, the current implementation requires tag_set for freeing
375 	 * requests, so free the pool now.
376 	 *
377 	 * Queue has become frozen, there can't be any in-queue requests, so
378 	 * it is safe to free requests now.
379 	 */
380 	mutex_lock(&q->sysfs_lock);
381 	if (q->elevator)
382 		blk_mq_sched_free_requests(q);
383 	mutex_unlock(&q->sysfs_lock);
384 
385 	percpu_ref_exit(&q->q_usage_counter);
386 
387 	/* @q is and will stay empty, shutdown and put */
388 	blk_put_queue(q);
389 }
390 EXPORT_SYMBOL(blk_cleanup_queue);
391 
392 struct request_queue *blk_alloc_queue(gfp_t gfp_mask)
393 {
394 	return blk_alloc_queue_node(gfp_mask, NUMA_NO_NODE);
395 }
396 EXPORT_SYMBOL(blk_alloc_queue);
397 
398 /**
399  * blk_queue_enter() - try to increase q->q_usage_counter
400  * @q: request queue pointer
401  * @flags: BLK_MQ_REQ_NOWAIT and/or BLK_MQ_REQ_PREEMPT
402  */
403 int blk_queue_enter(struct request_queue *q, blk_mq_req_flags_t flags)
404 {
405 	const bool pm = flags & BLK_MQ_REQ_PREEMPT;
406 
407 	while (true) {
408 		bool success = false;
409 
410 		rcu_read_lock();
411 		if (percpu_ref_tryget_live(&q->q_usage_counter)) {
412 			/*
413 			 * The code that increments the pm_only counter is
414 			 * responsible for ensuring that that counter is
415 			 * globally visible before the queue is unfrozen.
416 			 */
417 			if (pm || !blk_queue_pm_only(q)) {
418 				success = true;
419 			} else {
420 				percpu_ref_put(&q->q_usage_counter);
421 			}
422 		}
423 		rcu_read_unlock();
424 
425 		if (success)
426 			return 0;
427 
428 		if (flags & BLK_MQ_REQ_NOWAIT)
429 			return -EBUSY;
430 
431 		/*
432 		 * read pair of barrier in blk_freeze_queue_start(),
433 		 * we need to order reading __PERCPU_REF_DEAD flag of
434 		 * .q_usage_counter and reading .mq_freeze_depth or
435 		 * queue dying flag, otherwise the following wait may
436 		 * never return if the two reads are reordered.
437 		 */
438 		smp_rmb();
439 
440 		wait_event(q->mq_freeze_wq,
441 			   (!q->mq_freeze_depth &&
442 			    (pm || (blk_pm_request_resume(q),
443 				    !blk_queue_pm_only(q)))) ||
444 			   blk_queue_dying(q));
445 		if (blk_queue_dying(q))
446 			return -ENODEV;
447 	}
448 }
449 
450 void blk_queue_exit(struct request_queue *q)
451 {
452 	percpu_ref_put(&q->q_usage_counter);
453 }
454 
455 static void blk_queue_usage_counter_release(struct percpu_ref *ref)
456 {
457 	struct request_queue *q =
458 		container_of(ref, struct request_queue, q_usage_counter);
459 
460 	wake_up_all(&q->mq_freeze_wq);
461 }
462 
463 static void blk_rq_timed_out_timer(struct timer_list *t)
464 {
465 	struct request_queue *q = from_timer(q, t, timeout);
466 
467 	kblockd_schedule_work(&q->timeout_work);
468 }
469 
470 static void blk_timeout_work(struct work_struct *work)
471 {
472 }
473 
474 /**
475  * blk_alloc_queue_node - allocate a request queue
476  * @gfp_mask: memory allocation flags
477  * @node_id: NUMA node to allocate memory from
478  */
479 struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id)
480 {
481 	struct request_queue *q;
482 	int ret;
483 
484 	q = kmem_cache_alloc_node(blk_requestq_cachep,
485 				gfp_mask | __GFP_ZERO, node_id);
486 	if (!q)
487 		return NULL;
488 
489 	q->last_merge = NULL;
490 
491 	q->id = ida_simple_get(&blk_queue_ida, 0, 0, gfp_mask);
492 	if (q->id < 0)
493 		goto fail_q;
494 
495 	ret = bioset_init(&q->bio_split, BIO_POOL_SIZE, 0, BIOSET_NEED_BVECS);
496 	if (ret)
497 		goto fail_id;
498 
499 	q->backing_dev_info = bdi_alloc_node(gfp_mask, node_id);
500 	if (!q->backing_dev_info)
501 		goto fail_split;
502 
503 	q->stats = blk_alloc_queue_stats();
504 	if (!q->stats)
505 		goto fail_stats;
506 
507 	q->backing_dev_info->ra_pages = VM_READAHEAD_PAGES;
508 	q->backing_dev_info->capabilities = BDI_CAP_CGROUP_WRITEBACK;
509 	q->backing_dev_info->name = "block";
510 	q->node = node_id;
511 
512 	timer_setup(&q->backing_dev_info->laptop_mode_wb_timer,
513 		    laptop_mode_timer_fn, 0);
514 	timer_setup(&q->timeout, blk_rq_timed_out_timer, 0);
515 	INIT_WORK(&q->timeout_work, blk_timeout_work);
516 	INIT_LIST_HEAD(&q->icq_list);
517 #ifdef CONFIG_BLK_CGROUP
518 	INIT_LIST_HEAD(&q->blkg_list);
519 #endif
520 
521 	kobject_init(&q->kobj, &blk_queue_ktype);
522 
523 #ifdef CONFIG_BLK_DEV_IO_TRACE
524 	mutex_init(&q->blk_trace_mutex);
525 #endif
526 	mutex_init(&q->sysfs_lock);
527 	mutex_init(&q->sysfs_dir_lock);
528 	spin_lock_init(&q->queue_lock);
529 
530 	init_waitqueue_head(&q->mq_freeze_wq);
531 	mutex_init(&q->mq_freeze_lock);
532 
533 	/*
534 	 * Init percpu_ref in atomic mode so that it's faster to shutdown.
535 	 * See blk_register_queue() for details.
536 	 */
537 	if (percpu_ref_init(&q->q_usage_counter,
538 				blk_queue_usage_counter_release,
539 				PERCPU_REF_INIT_ATOMIC, GFP_KERNEL))
540 		goto fail_bdi;
541 
542 	if (blkcg_init_queue(q))
543 		goto fail_ref;
544 
545 	return q;
546 
547 fail_ref:
548 	percpu_ref_exit(&q->q_usage_counter);
549 fail_bdi:
550 	blk_free_queue_stats(q->stats);
551 fail_stats:
552 	bdi_put(q->backing_dev_info);
553 fail_split:
554 	bioset_exit(&q->bio_split);
555 fail_id:
556 	ida_simple_remove(&blk_queue_ida, q->id);
557 fail_q:
558 	kmem_cache_free(blk_requestq_cachep, q);
559 	return NULL;
560 }
561 EXPORT_SYMBOL(blk_alloc_queue_node);
562 
563 bool blk_get_queue(struct request_queue *q)
564 {
565 	if (likely(!blk_queue_dying(q))) {
566 		__blk_get_queue(q);
567 		return true;
568 	}
569 
570 	return false;
571 }
572 EXPORT_SYMBOL(blk_get_queue);
573 
574 /**
575  * blk_get_request - allocate a request
576  * @q: request queue to allocate a request for
577  * @op: operation (REQ_OP_*) and REQ_* flags, e.g. REQ_SYNC.
578  * @flags: BLK_MQ_REQ_* flags, e.g. BLK_MQ_REQ_NOWAIT.
579  */
580 struct request *blk_get_request(struct request_queue *q, unsigned int op,
581 				blk_mq_req_flags_t flags)
582 {
583 	struct request *req;
584 
585 	WARN_ON_ONCE(op & REQ_NOWAIT);
586 	WARN_ON_ONCE(flags & ~(BLK_MQ_REQ_NOWAIT | BLK_MQ_REQ_PREEMPT));
587 
588 	req = blk_mq_alloc_request(q, op, flags);
589 	if (!IS_ERR(req) && q->mq_ops->initialize_rq_fn)
590 		q->mq_ops->initialize_rq_fn(req);
591 
592 	return req;
593 }
594 EXPORT_SYMBOL(blk_get_request);
595 
596 void blk_put_request(struct request *req)
597 {
598 	blk_mq_free_request(req);
599 }
600 EXPORT_SYMBOL(blk_put_request);
601 
602 bool bio_attempt_back_merge(struct request *req, struct bio *bio,
603 		unsigned int nr_segs)
604 {
605 	const int ff = bio->bi_opf & REQ_FAILFAST_MASK;
606 
607 	if (!ll_back_merge_fn(req, bio, nr_segs))
608 		return false;
609 
610 	trace_block_bio_backmerge(req->q, req, bio);
611 	rq_qos_merge(req->q, req, bio);
612 
613 	if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
614 		blk_rq_set_mixed_merge(req);
615 
616 	req->biotail->bi_next = bio;
617 	req->biotail = bio;
618 	req->__data_len += bio->bi_iter.bi_size;
619 
620 	blk_account_io_start(req, false);
621 	return true;
622 }
623 
624 bool bio_attempt_front_merge(struct request *req, struct bio *bio,
625 		unsigned int nr_segs)
626 {
627 	const int ff = bio->bi_opf & REQ_FAILFAST_MASK;
628 
629 	if (!ll_front_merge_fn(req, bio, nr_segs))
630 		return false;
631 
632 	trace_block_bio_frontmerge(req->q, req, bio);
633 	rq_qos_merge(req->q, req, bio);
634 
635 	if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
636 		blk_rq_set_mixed_merge(req);
637 
638 	bio->bi_next = req->bio;
639 	req->bio = bio;
640 
641 	req->__sector = bio->bi_iter.bi_sector;
642 	req->__data_len += bio->bi_iter.bi_size;
643 
644 	blk_account_io_start(req, false);
645 	return true;
646 }
647 
648 bool bio_attempt_discard_merge(struct request_queue *q, struct request *req,
649 		struct bio *bio)
650 {
651 	unsigned short segments = blk_rq_nr_discard_segments(req);
652 
653 	if (segments >= queue_max_discard_segments(q))
654 		goto no_merge;
655 	if (blk_rq_sectors(req) + bio_sectors(bio) >
656 	    blk_rq_get_max_sectors(req, blk_rq_pos(req)))
657 		goto no_merge;
658 
659 	rq_qos_merge(q, req, bio);
660 
661 	req->biotail->bi_next = bio;
662 	req->biotail = bio;
663 	req->__data_len += bio->bi_iter.bi_size;
664 	req->nr_phys_segments = segments + 1;
665 
666 	blk_account_io_start(req, false);
667 	return true;
668 no_merge:
669 	req_set_nomerge(q, req);
670 	return false;
671 }
672 
673 /**
674  * blk_attempt_plug_merge - try to merge with %current's plugged list
675  * @q: request_queue new bio is being queued at
676  * @bio: new bio being queued
677  * @nr_segs: number of segments in @bio
678  * @same_queue_rq: pointer to &struct request that gets filled in when
679  * another request associated with @q is found on the plug list
680  * (optional, may be %NULL)
681  *
682  * Determine whether @bio being queued on @q can be merged with a request
683  * on %current's plugged list.  Returns %true if merge was successful,
684  * otherwise %false.
685  *
686  * Plugging coalesces IOs from the same issuer for the same purpose without
687  * going through @q->queue_lock.  As such it's more of an issuing mechanism
688  * than scheduling, and the request, while may have elvpriv data, is not
689  * added on the elevator at this point.  In addition, we don't have
690  * reliable access to the elevator outside queue lock.  Only check basic
691  * merging parameters without querying the elevator.
692  *
693  * Caller must ensure !blk_queue_nomerges(q) beforehand.
694  */
695 bool blk_attempt_plug_merge(struct request_queue *q, struct bio *bio,
696 		unsigned int nr_segs, struct request **same_queue_rq)
697 {
698 	struct blk_plug *plug;
699 	struct request *rq;
700 	struct list_head *plug_list;
701 
702 	plug = blk_mq_plug(q, bio);
703 	if (!plug)
704 		return false;
705 
706 	plug_list = &plug->mq_list;
707 
708 	list_for_each_entry_reverse(rq, plug_list, queuelist) {
709 		bool merged = false;
710 
711 		if (rq->q == q && same_queue_rq) {
712 			/*
713 			 * Only blk-mq multiple hardware queues case checks the
714 			 * rq in the same queue, there should be only one such
715 			 * rq in a queue
716 			 **/
717 			*same_queue_rq = rq;
718 		}
719 
720 		if (rq->q != q || !blk_rq_merge_ok(rq, bio))
721 			continue;
722 
723 		switch (blk_try_merge(rq, bio)) {
724 		case ELEVATOR_BACK_MERGE:
725 			merged = bio_attempt_back_merge(rq, bio, nr_segs);
726 			break;
727 		case ELEVATOR_FRONT_MERGE:
728 			merged = bio_attempt_front_merge(rq, bio, nr_segs);
729 			break;
730 		case ELEVATOR_DISCARD_MERGE:
731 			merged = bio_attempt_discard_merge(q, rq, bio);
732 			break;
733 		default:
734 			break;
735 		}
736 
737 		if (merged)
738 			return true;
739 	}
740 
741 	return false;
742 }
743 
744 static void handle_bad_sector(struct bio *bio, sector_t maxsector)
745 {
746 	char b[BDEVNAME_SIZE];
747 
748 	printk(KERN_INFO "attempt to access beyond end of device\n");
749 	printk(KERN_INFO "%s: rw=%d, want=%Lu, limit=%Lu\n",
750 			bio_devname(bio, b), bio->bi_opf,
751 			(unsigned long long)bio_end_sector(bio),
752 			(long long)maxsector);
753 }
754 
755 #ifdef CONFIG_FAIL_MAKE_REQUEST
756 
757 static DECLARE_FAULT_ATTR(fail_make_request);
758 
759 static int __init setup_fail_make_request(char *str)
760 {
761 	return setup_fault_attr(&fail_make_request, str);
762 }
763 __setup("fail_make_request=", setup_fail_make_request);
764 
765 static bool should_fail_request(struct hd_struct *part, unsigned int bytes)
766 {
767 	return part->make_it_fail && should_fail(&fail_make_request, bytes);
768 }
769 
770 static int __init fail_make_request_debugfs(void)
771 {
772 	struct dentry *dir = fault_create_debugfs_attr("fail_make_request",
773 						NULL, &fail_make_request);
774 
775 	return PTR_ERR_OR_ZERO(dir);
776 }
777 
778 late_initcall(fail_make_request_debugfs);
779 
780 #else /* CONFIG_FAIL_MAKE_REQUEST */
781 
782 static inline bool should_fail_request(struct hd_struct *part,
783 					unsigned int bytes)
784 {
785 	return false;
786 }
787 
788 #endif /* CONFIG_FAIL_MAKE_REQUEST */
789 
790 static inline bool bio_check_ro(struct bio *bio, struct hd_struct *part)
791 {
792 	const int op = bio_op(bio);
793 
794 	if (part->policy && op_is_write(op)) {
795 		char b[BDEVNAME_SIZE];
796 
797 		if (op_is_flush(bio->bi_opf) && !bio_sectors(bio))
798 			return false;
799 
800 		WARN_ONCE(1,
801 		       "generic_make_request: Trying to write "
802 			"to read-only block-device %s (partno %d)\n",
803 			bio_devname(bio, b), part->partno);
804 		/* Older lvm-tools actually trigger this */
805 		return false;
806 	}
807 
808 	return false;
809 }
810 
811 static noinline int should_fail_bio(struct bio *bio)
812 {
813 	if (should_fail_request(&bio->bi_disk->part0, bio->bi_iter.bi_size))
814 		return -EIO;
815 	return 0;
816 }
817 ALLOW_ERROR_INJECTION(should_fail_bio, ERRNO);
818 
819 /*
820  * Check whether this bio extends beyond the end of the device or partition.
821  * This may well happen - the kernel calls bread() without checking the size of
822  * the device, e.g., when mounting a file system.
823  */
824 static inline int bio_check_eod(struct bio *bio, sector_t maxsector)
825 {
826 	unsigned int nr_sectors = bio_sectors(bio);
827 
828 	if (nr_sectors && maxsector &&
829 	    (nr_sectors > maxsector ||
830 	     bio->bi_iter.bi_sector > maxsector - nr_sectors)) {
831 		handle_bad_sector(bio, maxsector);
832 		return -EIO;
833 	}
834 	return 0;
835 }
836 
837 /*
838  * Remap block n of partition p to block n+start(p) of the disk.
839  */
840 static inline int blk_partition_remap(struct bio *bio)
841 {
842 	struct hd_struct *p;
843 	int ret = -EIO;
844 
845 	rcu_read_lock();
846 	p = __disk_get_part(bio->bi_disk, bio->bi_partno);
847 	if (unlikely(!p))
848 		goto out;
849 	if (unlikely(should_fail_request(p, bio->bi_iter.bi_size)))
850 		goto out;
851 	if (unlikely(bio_check_ro(bio, p)))
852 		goto out;
853 
854 	if (bio_sectors(bio)) {
855 		if (bio_check_eod(bio, part_nr_sects_read(p)))
856 			goto out;
857 		bio->bi_iter.bi_sector += p->start_sect;
858 		trace_block_bio_remap(bio->bi_disk->queue, bio, part_devt(p),
859 				      bio->bi_iter.bi_sector - p->start_sect);
860 	}
861 	bio->bi_partno = 0;
862 	ret = 0;
863 out:
864 	rcu_read_unlock();
865 	return ret;
866 }
867 
868 static noinline_for_stack bool
869 generic_make_request_checks(struct bio *bio)
870 {
871 	struct request_queue *q;
872 	int nr_sectors = bio_sectors(bio);
873 	blk_status_t status = BLK_STS_IOERR;
874 	char b[BDEVNAME_SIZE];
875 
876 	might_sleep();
877 
878 	q = bio->bi_disk->queue;
879 	if (unlikely(!q)) {
880 		printk(KERN_ERR
881 		       "generic_make_request: Trying to access "
882 			"nonexistent block-device %s (%Lu)\n",
883 			bio_devname(bio, b), (long long)bio->bi_iter.bi_sector);
884 		goto end_io;
885 	}
886 
887 	/*
888 	 * Non-mq queues do not honor REQ_NOWAIT, so complete a bio
889 	 * with BLK_STS_AGAIN status in order to catch -EAGAIN and
890 	 * to give a chance to the caller to repeat request gracefully.
891 	 */
892 	if ((bio->bi_opf & REQ_NOWAIT) && !queue_is_mq(q)) {
893 		status = BLK_STS_AGAIN;
894 		goto end_io;
895 	}
896 
897 	if (should_fail_bio(bio))
898 		goto end_io;
899 
900 	if (bio->bi_partno) {
901 		if (unlikely(blk_partition_remap(bio)))
902 			goto end_io;
903 	} else {
904 		if (unlikely(bio_check_ro(bio, &bio->bi_disk->part0)))
905 			goto end_io;
906 		if (unlikely(bio_check_eod(bio, get_capacity(bio->bi_disk))))
907 			goto end_io;
908 	}
909 
910 	/*
911 	 * Filter flush bio's early so that make_request based
912 	 * drivers without flush support don't have to worry
913 	 * about them.
914 	 */
915 	if (op_is_flush(bio->bi_opf) &&
916 	    !test_bit(QUEUE_FLAG_WC, &q->queue_flags)) {
917 		bio->bi_opf &= ~(REQ_PREFLUSH | REQ_FUA);
918 		if (!nr_sectors) {
919 			status = BLK_STS_OK;
920 			goto end_io;
921 		}
922 	}
923 
924 	if (!test_bit(QUEUE_FLAG_POLL, &q->queue_flags))
925 		bio->bi_opf &= ~REQ_HIPRI;
926 
927 	switch (bio_op(bio)) {
928 	case REQ_OP_DISCARD:
929 		if (!blk_queue_discard(q))
930 			goto not_supported;
931 		break;
932 	case REQ_OP_SECURE_ERASE:
933 		if (!blk_queue_secure_erase(q))
934 			goto not_supported;
935 		break;
936 	case REQ_OP_WRITE_SAME:
937 		if (!q->limits.max_write_same_sectors)
938 			goto not_supported;
939 		break;
940 	case REQ_OP_ZONE_RESET:
941 	case REQ_OP_ZONE_OPEN:
942 	case REQ_OP_ZONE_CLOSE:
943 	case REQ_OP_ZONE_FINISH:
944 		if (!blk_queue_is_zoned(q))
945 			goto not_supported;
946 		break;
947 	case REQ_OP_ZONE_RESET_ALL:
948 		if (!blk_queue_is_zoned(q) || !blk_queue_zone_resetall(q))
949 			goto not_supported;
950 		break;
951 	case REQ_OP_WRITE_ZEROES:
952 		if (!q->limits.max_write_zeroes_sectors)
953 			goto not_supported;
954 		break;
955 	default:
956 		break;
957 	}
958 
959 	/*
960 	 * Various block parts want %current->io_context and lazy ioc
961 	 * allocation ends up trading a lot of pain for a small amount of
962 	 * memory.  Just allocate it upfront.  This may fail and block
963 	 * layer knows how to live with it.
964 	 */
965 	create_io_context(GFP_ATOMIC, q->node);
966 
967 	if (!blkcg_bio_issue_check(q, bio))
968 		return false;
969 
970 	if (!bio_flagged(bio, BIO_TRACE_COMPLETION)) {
971 		trace_block_bio_queue(q, bio);
972 		/* Now that enqueuing has been traced, we need to trace
973 		 * completion as well.
974 		 */
975 		bio_set_flag(bio, BIO_TRACE_COMPLETION);
976 	}
977 	return true;
978 
979 not_supported:
980 	status = BLK_STS_NOTSUPP;
981 end_io:
982 	bio->bi_status = status;
983 	bio_endio(bio);
984 	return false;
985 }
986 
987 /**
988  * generic_make_request - hand a buffer to its device driver for I/O
989  * @bio:  The bio describing the location in memory and on the device.
990  *
991  * generic_make_request() is used to make I/O requests of block
992  * devices. It is passed a &struct bio, which describes the I/O that needs
993  * to be done.
994  *
995  * generic_make_request() does not return any status.  The
996  * success/failure status of the request, along with notification of
997  * completion, is delivered asynchronously through the bio->bi_end_io
998  * function described (one day) else where.
999  *
1000  * The caller of generic_make_request must make sure that bi_io_vec
1001  * are set to describe the memory buffer, and that bi_dev and bi_sector are
1002  * set to describe the device address, and the
1003  * bi_end_io and optionally bi_private are set to describe how
1004  * completion notification should be signaled.
1005  *
1006  * generic_make_request and the drivers it calls may use bi_next if this
1007  * bio happens to be merged with someone else, and may resubmit the bio to
1008  * a lower device by calling into generic_make_request recursively, which
1009  * means the bio should NOT be touched after the call to ->make_request_fn.
1010  */
1011 blk_qc_t generic_make_request(struct bio *bio)
1012 {
1013 	/*
1014 	 * bio_list_on_stack[0] contains bios submitted by the current
1015 	 * make_request_fn.
1016 	 * bio_list_on_stack[1] contains bios that were submitted before
1017 	 * the current make_request_fn, but that haven't been processed
1018 	 * yet.
1019 	 */
1020 	struct bio_list bio_list_on_stack[2];
1021 	blk_qc_t ret = BLK_QC_T_NONE;
1022 
1023 	if (!generic_make_request_checks(bio))
1024 		goto out;
1025 
1026 	/*
1027 	 * We only want one ->make_request_fn to be active at a time, else
1028 	 * stack usage with stacked devices could be a problem.  So use
1029 	 * current->bio_list to keep a list of requests submited by a
1030 	 * make_request_fn function.  current->bio_list is also used as a
1031 	 * flag to say if generic_make_request is currently active in this
1032 	 * task or not.  If it is NULL, then no make_request is active.  If
1033 	 * it is non-NULL, then a make_request is active, and new requests
1034 	 * should be added at the tail
1035 	 */
1036 	if (current->bio_list) {
1037 		bio_list_add(&current->bio_list[0], bio);
1038 		goto out;
1039 	}
1040 
1041 	/* following loop may be a bit non-obvious, and so deserves some
1042 	 * explanation.
1043 	 * Before entering the loop, bio->bi_next is NULL (as all callers
1044 	 * ensure that) so we have a list with a single bio.
1045 	 * We pretend that we have just taken it off a longer list, so
1046 	 * we assign bio_list to a pointer to the bio_list_on_stack,
1047 	 * thus initialising the bio_list of new bios to be
1048 	 * added.  ->make_request() may indeed add some more bios
1049 	 * through a recursive call to generic_make_request.  If it
1050 	 * did, we find a non-NULL value in bio_list and re-enter the loop
1051 	 * from the top.  In this case we really did just take the bio
1052 	 * of the top of the list (no pretending) and so remove it from
1053 	 * bio_list, and call into ->make_request() again.
1054 	 */
1055 	BUG_ON(bio->bi_next);
1056 	bio_list_init(&bio_list_on_stack[0]);
1057 	current->bio_list = bio_list_on_stack;
1058 	do {
1059 		struct request_queue *q = bio->bi_disk->queue;
1060 		blk_mq_req_flags_t flags = bio->bi_opf & REQ_NOWAIT ?
1061 			BLK_MQ_REQ_NOWAIT : 0;
1062 
1063 		if (likely(blk_queue_enter(q, flags) == 0)) {
1064 			struct bio_list lower, same;
1065 
1066 			/* Create a fresh bio_list for all subordinate requests */
1067 			bio_list_on_stack[1] = bio_list_on_stack[0];
1068 			bio_list_init(&bio_list_on_stack[0]);
1069 			ret = q->make_request_fn(q, bio);
1070 
1071 			blk_queue_exit(q);
1072 
1073 			/* sort new bios into those for a lower level
1074 			 * and those for the same level
1075 			 */
1076 			bio_list_init(&lower);
1077 			bio_list_init(&same);
1078 			while ((bio = bio_list_pop(&bio_list_on_stack[0])) != NULL)
1079 				if (q == bio->bi_disk->queue)
1080 					bio_list_add(&same, bio);
1081 				else
1082 					bio_list_add(&lower, bio);
1083 			/* now assemble so we handle the lowest level first */
1084 			bio_list_merge(&bio_list_on_stack[0], &lower);
1085 			bio_list_merge(&bio_list_on_stack[0], &same);
1086 			bio_list_merge(&bio_list_on_stack[0], &bio_list_on_stack[1]);
1087 		} else {
1088 			if (unlikely(!blk_queue_dying(q) &&
1089 					(bio->bi_opf & REQ_NOWAIT)))
1090 				bio_wouldblock_error(bio);
1091 			else
1092 				bio_io_error(bio);
1093 		}
1094 		bio = bio_list_pop(&bio_list_on_stack[0]);
1095 	} while (bio);
1096 	current->bio_list = NULL; /* deactivate */
1097 
1098 out:
1099 	return ret;
1100 }
1101 EXPORT_SYMBOL(generic_make_request);
1102 
1103 /**
1104  * direct_make_request - hand a buffer directly to its device driver for I/O
1105  * @bio:  The bio describing the location in memory and on the device.
1106  *
1107  * This function behaves like generic_make_request(), but does not protect
1108  * against recursion.  Must only be used if the called driver is known
1109  * to not call generic_make_request (or direct_make_request) again from
1110  * its make_request function.  (Calling direct_make_request again from
1111  * a workqueue is perfectly fine as that doesn't recurse).
1112  */
1113 blk_qc_t direct_make_request(struct bio *bio)
1114 {
1115 	struct request_queue *q = bio->bi_disk->queue;
1116 	bool nowait = bio->bi_opf & REQ_NOWAIT;
1117 	blk_qc_t ret;
1118 
1119 	if (!generic_make_request_checks(bio))
1120 		return BLK_QC_T_NONE;
1121 
1122 	if (unlikely(blk_queue_enter(q, nowait ? BLK_MQ_REQ_NOWAIT : 0))) {
1123 		if (nowait && !blk_queue_dying(q))
1124 			bio->bi_status = BLK_STS_AGAIN;
1125 		else
1126 			bio->bi_status = BLK_STS_IOERR;
1127 		bio_endio(bio);
1128 		return BLK_QC_T_NONE;
1129 	}
1130 
1131 	ret = q->make_request_fn(q, bio);
1132 	blk_queue_exit(q);
1133 	return ret;
1134 }
1135 EXPORT_SYMBOL_GPL(direct_make_request);
1136 
1137 /**
1138  * submit_bio - submit a bio to the block device layer for I/O
1139  * @bio: The &struct bio which describes the I/O
1140  *
1141  * submit_bio() is very similar in purpose to generic_make_request(), and
1142  * uses that function to do most of the work. Both are fairly rough
1143  * interfaces; @bio must be presetup and ready for I/O.
1144  *
1145  */
1146 blk_qc_t submit_bio(struct bio *bio)
1147 {
1148 	bool workingset_read = false;
1149 	unsigned long pflags;
1150 	blk_qc_t ret;
1151 
1152 	if (blkcg_punt_bio_submit(bio))
1153 		return BLK_QC_T_NONE;
1154 
1155 	/*
1156 	 * If it's a regular read/write or a barrier with data attached,
1157 	 * go through the normal accounting stuff before submission.
1158 	 */
1159 	if (bio_has_data(bio)) {
1160 		unsigned int count;
1161 
1162 		if (unlikely(bio_op(bio) == REQ_OP_WRITE_SAME))
1163 			count = queue_logical_block_size(bio->bi_disk->queue) >> 9;
1164 		else
1165 			count = bio_sectors(bio);
1166 
1167 		if (op_is_write(bio_op(bio))) {
1168 			count_vm_events(PGPGOUT, count);
1169 		} else {
1170 			if (bio_flagged(bio, BIO_WORKINGSET))
1171 				workingset_read = true;
1172 			task_io_account_read(bio->bi_iter.bi_size);
1173 			count_vm_events(PGPGIN, count);
1174 		}
1175 
1176 		if (unlikely(block_dump)) {
1177 			char b[BDEVNAME_SIZE];
1178 			printk(KERN_DEBUG "%s(%d): %s block %Lu on %s (%u sectors)\n",
1179 			current->comm, task_pid_nr(current),
1180 				op_is_write(bio_op(bio)) ? "WRITE" : "READ",
1181 				(unsigned long long)bio->bi_iter.bi_sector,
1182 				bio_devname(bio, b), count);
1183 		}
1184 	}
1185 
1186 	/*
1187 	 * If we're reading data that is part of the userspace
1188 	 * workingset, count submission time as memory stall. When the
1189 	 * device is congested, or the submitting cgroup IO-throttled,
1190 	 * submission can be a significant part of overall IO time.
1191 	 */
1192 	if (workingset_read)
1193 		psi_memstall_enter(&pflags);
1194 
1195 	ret = generic_make_request(bio);
1196 
1197 	if (workingset_read)
1198 		psi_memstall_leave(&pflags);
1199 
1200 	return ret;
1201 }
1202 EXPORT_SYMBOL(submit_bio);
1203 
1204 /**
1205  * blk_cloned_rq_check_limits - Helper function to check a cloned request
1206  *                              for new the queue limits
1207  * @q:  the queue
1208  * @rq: the request being checked
1209  *
1210  * Description:
1211  *    @rq may have been made based on weaker limitations of upper-level queues
1212  *    in request stacking drivers, and it may violate the limitation of @q.
1213  *    Since the block layer and the underlying device driver trust @rq
1214  *    after it is inserted to @q, it should be checked against @q before
1215  *    the insertion using this generic function.
1216  *
1217  *    Request stacking drivers like request-based dm may change the queue
1218  *    limits when retrying requests on other queues. Those requests need
1219  *    to be checked against the new queue limits again during dispatch.
1220  */
1221 static int blk_cloned_rq_check_limits(struct request_queue *q,
1222 				      struct request *rq)
1223 {
1224 	if (blk_rq_sectors(rq) > blk_queue_get_max_sectors(q, req_op(rq))) {
1225 		printk(KERN_ERR "%s: over max size limit. (%u > %u)\n",
1226 			__func__, blk_rq_sectors(rq),
1227 			blk_queue_get_max_sectors(q, req_op(rq)));
1228 		return -EIO;
1229 	}
1230 
1231 	/*
1232 	 * queue's settings related to segment counting like q->bounce_pfn
1233 	 * may differ from that of other stacking queues.
1234 	 * Recalculate it to check the request correctly on this queue's
1235 	 * limitation.
1236 	 */
1237 	rq->nr_phys_segments = blk_recalc_rq_segments(rq);
1238 	if (rq->nr_phys_segments > queue_max_segments(q)) {
1239 		printk(KERN_ERR "%s: over max segments limit. (%hu > %hu)\n",
1240 			__func__, rq->nr_phys_segments, queue_max_segments(q));
1241 		return -EIO;
1242 	}
1243 
1244 	return 0;
1245 }
1246 
1247 /**
1248  * blk_insert_cloned_request - Helper for stacking drivers to submit a request
1249  * @q:  the queue to submit the request
1250  * @rq: the request being queued
1251  */
1252 blk_status_t blk_insert_cloned_request(struct request_queue *q, struct request *rq)
1253 {
1254 	if (blk_cloned_rq_check_limits(q, rq))
1255 		return BLK_STS_IOERR;
1256 
1257 	if (rq->rq_disk &&
1258 	    should_fail_request(&rq->rq_disk->part0, blk_rq_bytes(rq)))
1259 		return BLK_STS_IOERR;
1260 
1261 	if (blk_queue_io_stat(q))
1262 		blk_account_io_start(rq, true);
1263 
1264 	/*
1265 	 * Since we have a scheduler attached on the top device,
1266 	 * bypass a potential scheduler on the bottom device for
1267 	 * insert.
1268 	 */
1269 	return blk_mq_request_issue_directly(rq, true);
1270 }
1271 EXPORT_SYMBOL_GPL(blk_insert_cloned_request);
1272 
1273 /**
1274  * blk_rq_err_bytes - determine number of bytes till the next failure boundary
1275  * @rq: request to examine
1276  *
1277  * Description:
1278  *     A request could be merge of IOs which require different failure
1279  *     handling.  This function determines the number of bytes which
1280  *     can be failed from the beginning of the request without
1281  *     crossing into area which need to be retried further.
1282  *
1283  * Return:
1284  *     The number of bytes to fail.
1285  */
1286 unsigned int blk_rq_err_bytes(const struct request *rq)
1287 {
1288 	unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK;
1289 	unsigned int bytes = 0;
1290 	struct bio *bio;
1291 
1292 	if (!(rq->rq_flags & RQF_MIXED_MERGE))
1293 		return blk_rq_bytes(rq);
1294 
1295 	/*
1296 	 * Currently the only 'mixing' which can happen is between
1297 	 * different fastfail types.  We can safely fail portions
1298 	 * which have all the failfast bits that the first one has -
1299 	 * the ones which are at least as eager to fail as the first
1300 	 * one.
1301 	 */
1302 	for (bio = rq->bio; bio; bio = bio->bi_next) {
1303 		if ((bio->bi_opf & ff) != ff)
1304 			break;
1305 		bytes += bio->bi_iter.bi_size;
1306 	}
1307 
1308 	/* this could lead to infinite loop */
1309 	BUG_ON(blk_rq_bytes(rq) && !bytes);
1310 	return bytes;
1311 }
1312 EXPORT_SYMBOL_GPL(blk_rq_err_bytes);
1313 
1314 void blk_account_io_completion(struct request *req, unsigned int bytes)
1315 {
1316 	if (req->part && blk_do_io_stat(req)) {
1317 		const int sgrp = op_stat_group(req_op(req));
1318 		struct hd_struct *part;
1319 
1320 		part_stat_lock();
1321 		part = req->part;
1322 		part_stat_add(part, sectors[sgrp], bytes >> 9);
1323 		part_stat_unlock();
1324 	}
1325 }
1326 
1327 void blk_account_io_done(struct request *req, u64 now)
1328 {
1329 	/*
1330 	 * Account IO completion.  flush_rq isn't accounted as a
1331 	 * normal IO on queueing nor completion.  Accounting the
1332 	 * containing request is enough.
1333 	 */
1334 	if (req->part && blk_do_io_stat(req) &&
1335 	    !(req->rq_flags & RQF_FLUSH_SEQ)) {
1336 		const int sgrp = op_stat_group(req_op(req));
1337 		struct hd_struct *part;
1338 
1339 		part_stat_lock();
1340 		part = req->part;
1341 
1342 		update_io_ticks(part, jiffies);
1343 		part_stat_inc(part, ios[sgrp]);
1344 		part_stat_add(part, nsecs[sgrp], now - req->start_time_ns);
1345 		part_stat_add(part, time_in_queue, nsecs_to_jiffies64(now - req->start_time_ns));
1346 		part_dec_in_flight(req->q, part, rq_data_dir(req));
1347 
1348 		hd_struct_put(part);
1349 		part_stat_unlock();
1350 	}
1351 }
1352 
1353 void blk_account_io_start(struct request *rq, bool new_io)
1354 {
1355 	struct hd_struct *part;
1356 	int rw = rq_data_dir(rq);
1357 
1358 	if (!blk_do_io_stat(rq))
1359 		return;
1360 
1361 	part_stat_lock();
1362 
1363 	if (!new_io) {
1364 		part = rq->part;
1365 		part_stat_inc(part, merges[rw]);
1366 	} else {
1367 		part = disk_map_sector_rcu(rq->rq_disk, blk_rq_pos(rq));
1368 		if (!hd_struct_try_get(part)) {
1369 			/*
1370 			 * The partition is already being removed,
1371 			 * the request will be accounted on the disk only
1372 			 *
1373 			 * We take a reference on disk->part0 although that
1374 			 * partition will never be deleted, so we can treat
1375 			 * it as any other partition.
1376 			 */
1377 			part = &rq->rq_disk->part0;
1378 			hd_struct_get(part);
1379 		}
1380 		part_inc_in_flight(rq->q, part, rw);
1381 		rq->part = part;
1382 	}
1383 
1384 	update_io_ticks(part, jiffies);
1385 
1386 	part_stat_unlock();
1387 }
1388 
1389 /*
1390  * Steal bios from a request and add them to a bio list.
1391  * The request must not have been partially completed before.
1392  */
1393 void blk_steal_bios(struct bio_list *list, struct request *rq)
1394 {
1395 	if (rq->bio) {
1396 		if (list->tail)
1397 			list->tail->bi_next = rq->bio;
1398 		else
1399 			list->head = rq->bio;
1400 		list->tail = rq->biotail;
1401 
1402 		rq->bio = NULL;
1403 		rq->biotail = NULL;
1404 	}
1405 
1406 	rq->__data_len = 0;
1407 }
1408 EXPORT_SYMBOL_GPL(blk_steal_bios);
1409 
1410 /**
1411  * blk_update_request - Special helper function for request stacking drivers
1412  * @req:      the request being processed
1413  * @error:    block status code
1414  * @nr_bytes: number of bytes to complete @req
1415  *
1416  * Description:
1417  *     Ends I/O on a number of bytes attached to @req, but doesn't complete
1418  *     the request structure even if @req doesn't have leftover.
1419  *     If @req has leftover, sets it up for the next range of segments.
1420  *
1421  *     This special helper function is only for request stacking drivers
1422  *     (e.g. request-based dm) so that they can handle partial completion.
1423  *     Actual device drivers should use blk_mq_end_request instead.
1424  *
1425  *     Passing the result of blk_rq_bytes() as @nr_bytes guarantees
1426  *     %false return from this function.
1427  *
1428  * Note:
1429  *	The RQF_SPECIAL_PAYLOAD flag is ignored on purpose in both
1430  *	blk_rq_bytes() and in blk_update_request().
1431  *
1432  * Return:
1433  *     %false - this request doesn't have any more data
1434  *     %true  - this request has more data
1435  **/
1436 bool blk_update_request(struct request *req, blk_status_t error,
1437 		unsigned int nr_bytes)
1438 {
1439 	int total_bytes;
1440 
1441 	trace_block_rq_complete(req, blk_status_to_errno(error), nr_bytes);
1442 
1443 	if (!req->bio)
1444 		return false;
1445 
1446 #ifdef CONFIG_BLK_DEV_INTEGRITY
1447 	if (blk_integrity_rq(req) && req_op(req) == REQ_OP_READ &&
1448 	    error == BLK_STS_OK)
1449 		req->q->integrity.profile->complete_fn(req, nr_bytes);
1450 #endif
1451 
1452 	if (unlikely(error && !blk_rq_is_passthrough(req) &&
1453 		     !(req->rq_flags & RQF_QUIET)))
1454 		print_req_error(req, error, __func__);
1455 
1456 	blk_account_io_completion(req, nr_bytes);
1457 
1458 	total_bytes = 0;
1459 	while (req->bio) {
1460 		struct bio *bio = req->bio;
1461 		unsigned bio_bytes = min(bio->bi_iter.bi_size, nr_bytes);
1462 
1463 		if (bio_bytes == bio->bi_iter.bi_size)
1464 			req->bio = bio->bi_next;
1465 
1466 		/* Completion has already been traced */
1467 		bio_clear_flag(bio, BIO_TRACE_COMPLETION);
1468 		req_bio_endio(req, bio, bio_bytes, error);
1469 
1470 		total_bytes += bio_bytes;
1471 		nr_bytes -= bio_bytes;
1472 
1473 		if (!nr_bytes)
1474 			break;
1475 	}
1476 
1477 	/*
1478 	 * completely done
1479 	 */
1480 	if (!req->bio) {
1481 		/*
1482 		 * Reset counters so that the request stacking driver
1483 		 * can find how many bytes remain in the request
1484 		 * later.
1485 		 */
1486 		req->__data_len = 0;
1487 		return false;
1488 	}
1489 
1490 	req->__data_len -= total_bytes;
1491 
1492 	/* update sector only for requests with clear definition of sector */
1493 	if (!blk_rq_is_passthrough(req))
1494 		req->__sector += total_bytes >> 9;
1495 
1496 	/* mixed attributes always follow the first bio */
1497 	if (req->rq_flags & RQF_MIXED_MERGE) {
1498 		req->cmd_flags &= ~REQ_FAILFAST_MASK;
1499 		req->cmd_flags |= req->bio->bi_opf & REQ_FAILFAST_MASK;
1500 	}
1501 
1502 	if (!(req->rq_flags & RQF_SPECIAL_PAYLOAD)) {
1503 		/*
1504 		 * If total number of sectors is less than the first segment
1505 		 * size, something has gone terribly wrong.
1506 		 */
1507 		if (blk_rq_bytes(req) < blk_rq_cur_bytes(req)) {
1508 			blk_dump_rq_flags(req, "request botched");
1509 			req->__data_len = blk_rq_cur_bytes(req);
1510 		}
1511 
1512 		/* recalculate the number of segments */
1513 		req->nr_phys_segments = blk_recalc_rq_segments(req);
1514 	}
1515 
1516 	return true;
1517 }
1518 EXPORT_SYMBOL_GPL(blk_update_request);
1519 
1520 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
1521 /**
1522  * rq_flush_dcache_pages - Helper function to flush all pages in a request
1523  * @rq: the request to be flushed
1524  *
1525  * Description:
1526  *     Flush all pages in @rq.
1527  */
1528 void rq_flush_dcache_pages(struct request *rq)
1529 {
1530 	struct req_iterator iter;
1531 	struct bio_vec bvec;
1532 
1533 	rq_for_each_segment(bvec, rq, iter)
1534 		flush_dcache_page(bvec.bv_page);
1535 }
1536 EXPORT_SYMBOL_GPL(rq_flush_dcache_pages);
1537 #endif
1538 
1539 /**
1540  * blk_lld_busy - Check if underlying low-level drivers of a device are busy
1541  * @q : the queue of the device being checked
1542  *
1543  * Description:
1544  *    Check if underlying low-level drivers of a device are busy.
1545  *    If the drivers want to export their busy state, they must set own
1546  *    exporting function using blk_queue_lld_busy() first.
1547  *
1548  *    Basically, this function is used only by request stacking drivers
1549  *    to stop dispatching requests to underlying devices when underlying
1550  *    devices are busy.  This behavior helps more I/O merging on the queue
1551  *    of the request stacking driver and prevents I/O throughput regression
1552  *    on burst I/O load.
1553  *
1554  * Return:
1555  *    0 - Not busy (The request stacking driver should dispatch request)
1556  *    1 - Busy (The request stacking driver should stop dispatching request)
1557  */
1558 int blk_lld_busy(struct request_queue *q)
1559 {
1560 	if (queue_is_mq(q) && q->mq_ops->busy)
1561 		return q->mq_ops->busy(q);
1562 
1563 	return 0;
1564 }
1565 EXPORT_SYMBOL_GPL(blk_lld_busy);
1566 
1567 /**
1568  * blk_rq_unprep_clone - Helper function to free all bios in a cloned request
1569  * @rq: the clone request to be cleaned up
1570  *
1571  * Description:
1572  *     Free all bios in @rq for a cloned request.
1573  */
1574 void blk_rq_unprep_clone(struct request *rq)
1575 {
1576 	struct bio *bio;
1577 
1578 	while ((bio = rq->bio) != NULL) {
1579 		rq->bio = bio->bi_next;
1580 
1581 		bio_put(bio);
1582 	}
1583 }
1584 EXPORT_SYMBOL_GPL(blk_rq_unprep_clone);
1585 
1586 /*
1587  * Copy attributes of the original request to the clone request.
1588  * The actual data parts (e.g. ->cmd, ->sense) are not copied.
1589  */
1590 static void __blk_rq_prep_clone(struct request *dst, struct request *src)
1591 {
1592 	dst->__sector = blk_rq_pos(src);
1593 	dst->__data_len = blk_rq_bytes(src);
1594 	if (src->rq_flags & RQF_SPECIAL_PAYLOAD) {
1595 		dst->rq_flags |= RQF_SPECIAL_PAYLOAD;
1596 		dst->special_vec = src->special_vec;
1597 	}
1598 	dst->nr_phys_segments = src->nr_phys_segments;
1599 	dst->ioprio = src->ioprio;
1600 	dst->extra_len = src->extra_len;
1601 }
1602 
1603 /**
1604  * blk_rq_prep_clone - Helper function to setup clone request
1605  * @rq: the request to be setup
1606  * @rq_src: original request to be cloned
1607  * @bs: bio_set that bios for clone are allocated from
1608  * @gfp_mask: memory allocation mask for bio
1609  * @bio_ctr: setup function to be called for each clone bio.
1610  *           Returns %0 for success, non %0 for failure.
1611  * @data: private data to be passed to @bio_ctr
1612  *
1613  * Description:
1614  *     Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq.
1615  *     The actual data parts of @rq_src (e.g. ->cmd, ->sense)
1616  *     are not copied, and copying such parts is the caller's responsibility.
1617  *     Also, pages which the original bios are pointing to are not copied
1618  *     and the cloned bios just point same pages.
1619  *     So cloned bios must be completed before original bios, which means
1620  *     the caller must complete @rq before @rq_src.
1621  */
1622 int blk_rq_prep_clone(struct request *rq, struct request *rq_src,
1623 		      struct bio_set *bs, gfp_t gfp_mask,
1624 		      int (*bio_ctr)(struct bio *, struct bio *, void *),
1625 		      void *data)
1626 {
1627 	struct bio *bio, *bio_src;
1628 
1629 	if (!bs)
1630 		bs = &fs_bio_set;
1631 
1632 	__rq_for_each_bio(bio_src, rq_src) {
1633 		bio = bio_clone_fast(bio_src, gfp_mask, bs);
1634 		if (!bio)
1635 			goto free_and_out;
1636 
1637 		if (bio_ctr && bio_ctr(bio, bio_src, data))
1638 			goto free_and_out;
1639 
1640 		if (rq->bio) {
1641 			rq->biotail->bi_next = bio;
1642 			rq->biotail = bio;
1643 		} else
1644 			rq->bio = rq->biotail = bio;
1645 	}
1646 
1647 	__blk_rq_prep_clone(rq, rq_src);
1648 
1649 	return 0;
1650 
1651 free_and_out:
1652 	if (bio)
1653 		bio_put(bio);
1654 	blk_rq_unprep_clone(rq);
1655 
1656 	return -ENOMEM;
1657 }
1658 EXPORT_SYMBOL_GPL(blk_rq_prep_clone);
1659 
1660 int kblockd_schedule_work(struct work_struct *work)
1661 {
1662 	return queue_work(kblockd_workqueue, work);
1663 }
1664 EXPORT_SYMBOL(kblockd_schedule_work);
1665 
1666 int kblockd_schedule_work_on(int cpu, struct work_struct *work)
1667 {
1668 	return queue_work_on(cpu, kblockd_workqueue, work);
1669 }
1670 EXPORT_SYMBOL(kblockd_schedule_work_on);
1671 
1672 int kblockd_mod_delayed_work_on(int cpu, struct delayed_work *dwork,
1673 				unsigned long delay)
1674 {
1675 	return mod_delayed_work_on(cpu, kblockd_workqueue, dwork, delay);
1676 }
1677 EXPORT_SYMBOL(kblockd_mod_delayed_work_on);
1678 
1679 /**
1680  * blk_start_plug - initialize blk_plug and track it inside the task_struct
1681  * @plug:	The &struct blk_plug that needs to be initialized
1682  *
1683  * Description:
1684  *   blk_start_plug() indicates to the block layer an intent by the caller
1685  *   to submit multiple I/O requests in a batch.  The block layer may use
1686  *   this hint to defer submitting I/Os from the caller until blk_finish_plug()
1687  *   is called.  However, the block layer may choose to submit requests
1688  *   before a call to blk_finish_plug() if the number of queued I/Os
1689  *   exceeds %BLK_MAX_REQUEST_COUNT, or if the size of the I/O is larger than
1690  *   %BLK_PLUG_FLUSH_SIZE.  The queued I/Os may also be submitted early if
1691  *   the task schedules (see below).
1692  *
1693  *   Tracking blk_plug inside the task_struct will help with auto-flushing the
1694  *   pending I/O should the task end up blocking between blk_start_plug() and
1695  *   blk_finish_plug(). This is important from a performance perspective, but
1696  *   also ensures that we don't deadlock. For instance, if the task is blocking
1697  *   for a memory allocation, memory reclaim could end up wanting to free a
1698  *   page belonging to that request that is currently residing in our private
1699  *   plug. By flushing the pending I/O when the process goes to sleep, we avoid
1700  *   this kind of deadlock.
1701  */
1702 void blk_start_plug(struct blk_plug *plug)
1703 {
1704 	struct task_struct *tsk = current;
1705 
1706 	/*
1707 	 * If this is a nested plug, don't actually assign it.
1708 	 */
1709 	if (tsk->plug)
1710 		return;
1711 
1712 	INIT_LIST_HEAD(&plug->mq_list);
1713 	INIT_LIST_HEAD(&plug->cb_list);
1714 	plug->rq_count = 0;
1715 	plug->multiple_queues = false;
1716 
1717 	/*
1718 	 * Store ordering should not be needed here, since a potential
1719 	 * preempt will imply a full memory barrier
1720 	 */
1721 	tsk->plug = plug;
1722 }
1723 EXPORT_SYMBOL(blk_start_plug);
1724 
1725 static void flush_plug_callbacks(struct blk_plug *plug, bool from_schedule)
1726 {
1727 	LIST_HEAD(callbacks);
1728 
1729 	while (!list_empty(&plug->cb_list)) {
1730 		list_splice_init(&plug->cb_list, &callbacks);
1731 
1732 		while (!list_empty(&callbacks)) {
1733 			struct blk_plug_cb *cb = list_first_entry(&callbacks,
1734 							  struct blk_plug_cb,
1735 							  list);
1736 			list_del(&cb->list);
1737 			cb->callback(cb, from_schedule);
1738 		}
1739 	}
1740 }
1741 
1742 struct blk_plug_cb *blk_check_plugged(blk_plug_cb_fn unplug, void *data,
1743 				      int size)
1744 {
1745 	struct blk_plug *plug = current->plug;
1746 	struct blk_plug_cb *cb;
1747 
1748 	if (!plug)
1749 		return NULL;
1750 
1751 	list_for_each_entry(cb, &plug->cb_list, list)
1752 		if (cb->callback == unplug && cb->data == data)
1753 			return cb;
1754 
1755 	/* Not currently on the callback list */
1756 	BUG_ON(size < sizeof(*cb));
1757 	cb = kzalloc(size, GFP_ATOMIC);
1758 	if (cb) {
1759 		cb->data = data;
1760 		cb->callback = unplug;
1761 		list_add(&cb->list, &plug->cb_list);
1762 	}
1763 	return cb;
1764 }
1765 EXPORT_SYMBOL(blk_check_plugged);
1766 
1767 void blk_flush_plug_list(struct blk_plug *plug, bool from_schedule)
1768 {
1769 	flush_plug_callbacks(plug, from_schedule);
1770 
1771 	if (!list_empty(&plug->mq_list))
1772 		blk_mq_flush_plug_list(plug, from_schedule);
1773 }
1774 
1775 /**
1776  * blk_finish_plug - mark the end of a batch of submitted I/O
1777  * @plug:	The &struct blk_plug passed to blk_start_plug()
1778  *
1779  * Description:
1780  * Indicate that a batch of I/O submissions is complete.  This function
1781  * must be paired with an initial call to blk_start_plug().  The intent
1782  * is to allow the block layer to optimize I/O submission.  See the
1783  * documentation for blk_start_plug() for more information.
1784  */
1785 void blk_finish_plug(struct blk_plug *plug)
1786 {
1787 	if (plug != current->plug)
1788 		return;
1789 	blk_flush_plug_list(plug, false);
1790 
1791 	current->plug = NULL;
1792 }
1793 EXPORT_SYMBOL(blk_finish_plug);
1794 
1795 int __init blk_dev_init(void)
1796 {
1797 	BUILD_BUG_ON(REQ_OP_LAST >= (1 << REQ_OP_BITS));
1798 	BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 *
1799 			sizeof_field(struct request, cmd_flags));
1800 	BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 *
1801 			sizeof_field(struct bio, bi_opf));
1802 
1803 	/* used for unplugging and affects IO latency/throughput - HIGHPRI */
1804 	kblockd_workqueue = alloc_workqueue("kblockd",
1805 					    WQ_MEM_RECLAIM | WQ_HIGHPRI, 0);
1806 	if (!kblockd_workqueue)
1807 		panic("Failed to create kblockd\n");
1808 
1809 	blk_requestq_cachep = kmem_cache_create("request_queue",
1810 			sizeof(struct request_queue), 0, SLAB_PANIC, NULL);
1811 
1812 #ifdef CONFIG_DEBUG_FS
1813 	blk_debugfs_root = debugfs_create_dir("block", NULL);
1814 #endif
1815 
1816 	return 0;
1817 }
1818