xref: /openbmc/linux/block/blk-core.c (revision 7fc38225363dd8f19e667ad7c77b63bc4a5c065d)
1 /*
2  * Copyright (C) 1991, 1992 Linus Torvalds
3  * Copyright (C) 1994,      Karl Keyte: Added support for disk statistics
4  * Elevator latency, (C) 2000  Andrea Arcangeli <andrea@suse.de> SuSE
5  * Queue request tables / lock, selectable elevator, Jens Axboe <axboe@suse.de>
6  * kernel-doc documentation started by NeilBrown <neilb@cse.unsw.edu.au>
7  *	-  July2000
8  * bio rewrite, highmem i/o, etc, Jens Axboe <axboe@suse.de> - may 2001
9  */
10 
11 /*
12  * This handles all read/write requests to block devices
13  */
14 #include <linux/kernel.h>
15 #include <linux/module.h>
16 #include <linux/backing-dev.h>
17 #include <linux/bio.h>
18 #include <linux/blkdev.h>
19 #include <linux/blk-mq.h>
20 #include <linux/highmem.h>
21 #include <linux/mm.h>
22 #include <linux/kernel_stat.h>
23 #include <linux/string.h>
24 #include <linux/init.h>
25 #include <linux/completion.h>
26 #include <linux/slab.h>
27 #include <linux/swap.h>
28 #include <linux/writeback.h>
29 #include <linux/task_io_accounting_ops.h>
30 #include <linux/fault-inject.h>
31 #include <linux/list_sort.h>
32 #include <linux/delay.h>
33 #include <linux/ratelimit.h>
34 #include <linux/pm_runtime.h>
35 #include <linux/blk-cgroup.h>
36 #include <linux/debugfs.h>
37 #include <linux/bpf.h>
38 
39 #define CREATE_TRACE_POINTS
40 #include <trace/events/block.h>
41 
42 #include "blk.h"
43 #include "blk-mq.h"
44 #include "blk-mq-sched.h"
45 #include "blk-pm.h"
46 #include "blk-rq-qos.h"
47 
48 #ifdef CONFIG_DEBUG_FS
49 struct dentry *blk_debugfs_root;
50 #endif
51 
52 EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_remap);
53 EXPORT_TRACEPOINT_SYMBOL_GPL(block_rq_remap);
54 EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_complete);
55 EXPORT_TRACEPOINT_SYMBOL_GPL(block_split);
56 EXPORT_TRACEPOINT_SYMBOL_GPL(block_unplug);
57 
58 DEFINE_IDA(blk_queue_ida);
59 
60 /*
61  * For queue allocation
62  */
63 struct kmem_cache *blk_requestq_cachep;
64 
65 /*
66  * Controlling structure to kblockd
67  */
68 static struct workqueue_struct *kblockd_workqueue;
69 
70 /**
71  * blk_queue_flag_set - atomically set a queue flag
72  * @flag: flag to be set
73  * @q: request queue
74  */
75 void blk_queue_flag_set(unsigned int flag, struct request_queue *q)
76 {
77 	set_bit(flag, &q->queue_flags);
78 }
79 EXPORT_SYMBOL(blk_queue_flag_set);
80 
81 /**
82  * blk_queue_flag_clear - atomically clear a queue flag
83  * @flag: flag to be cleared
84  * @q: request queue
85  */
86 void blk_queue_flag_clear(unsigned int flag, struct request_queue *q)
87 {
88 	clear_bit(flag, &q->queue_flags);
89 }
90 EXPORT_SYMBOL(blk_queue_flag_clear);
91 
92 /**
93  * blk_queue_flag_test_and_set - atomically test and set a queue flag
94  * @flag: flag to be set
95  * @q: request queue
96  *
97  * Returns the previous value of @flag - 0 if the flag was not set and 1 if
98  * the flag was already set.
99  */
100 bool blk_queue_flag_test_and_set(unsigned int flag, struct request_queue *q)
101 {
102 	return test_and_set_bit(flag, &q->queue_flags);
103 }
104 EXPORT_SYMBOL_GPL(blk_queue_flag_test_and_set);
105 
106 void blk_rq_init(struct request_queue *q, struct request *rq)
107 {
108 	memset(rq, 0, sizeof(*rq));
109 
110 	INIT_LIST_HEAD(&rq->queuelist);
111 	rq->q = q;
112 	rq->__sector = (sector_t) -1;
113 	INIT_HLIST_NODE(&rq->hash);
114 	RB_CLEAR_NODE(&rq->rb_node);
115 	rq->tag = -1;
116 	rq->internal_tag = -1;
117 	rq->start_time_ns = ktime_get_ns();
118 	rq->part = NULL;
119 }
120 EXPORT_SYMBOL(blk_rq_init);
121 
122 static const struct {
123 	int		errno;
124 	const char	*name;
125 } blk_errors[] = {
126 	[BLK_STS_OK]		= { 0,		"" },
127 	[BLK_STS_NOTSUPP]	= { -EOPNOTSUPP, "operation not supported" },
128 	[BLK_STS_TIMEOUT]	= { -ETIMEDOUT,	"timeout" },
129 	[BLK_STS_NOSPC]		= { -ENOSPC,	"critical space allocation" },
130 	[BLK_STS_TRANSPORT]	= { -ENOLINK,	"recoverable transport" },
131 	[BLK_STS_TARGET]	= { -EREMOTEIO,	"critical target" },
132 	[BLK_STS_NEXUS]		= { -EBADE,	"critical nexus" },
133 	[BLK_STS_MEDIUM]	= { -ENODATA,	"critical medium" },
134 	[BLK_STS_PROTECTION]	= { -EILSEQ,	"protection" },
135 	[BLK_STS_RESOURCE]	= { -ENOMEM,	"kernel resource" },
136 	[BLK_STS_DEV_RESOURCE]	= { -EBUSY,	"device resource" },
137 	[BLK_STS_AGAIN]		= { -EAGAIN,	"nonblocking retry" },
138 
139 	/* device mapper special case, should not leak out: */
140 	[BLK_STS_DM_REQUEUE]	= { -EREMCHG, "dm internal retry" },
141 
142 	/* everything else not covered above: */
143 	[BLK_STS_IOERR]		= { -EIO,	"I/O" },
144 };
145 
146 blk_status_t errno_to_blk_status(int errno)
147 {
148 	int i;
149 
150 	for (i = 0; i < ARRAY_SIZE(blk_errors); i++) {
151 		if (blk_errors[i].errno == errno)
152 			return (__force blk_status_t)i;
153 	}
154 
155 	return BLK_STS_IOERR;
156 }
157 EXPORT_SYMBOL_GPL(errno_to_blk_status);
158 
159 int blk_status_to_errno(blk_status_t status)
160 {
161 	int idx = (__force int)status;
162 
163 	if (WARN_ON_ONCE(idx >= ARRAY_SIZE(blk_errors)))
164 		return -EIO;
165 	return blk_errors[idx].errno;
166 }
167 EXPORT_SYMBOL_GPL(blk_status_to_errno);
168 
169 static void print_req_error(struct request *req, blk_status_t status)
170 {
171 	int idx = (__force int)status;
172 
173 	if (WARN_ON_ONCE(idx >= ARRAY_SIZE(blk_errors)))
174 		return;
175 
176 	printk_ratelimited(KERN_ERR "%s: %s error, dev %s, sector %llu flags %x\n",
177 				__func__, blk_errors[idx].name,
178 				req->rq_disk ?  req->rq_disk->disk_name : "?",
179 				(unsigned long long)blk_rq_pos(req),
180 				req->cmd_flags);
181 }
182 
183 static void req_bio_endio(struct request *rq, struct bio *bio,
184 			  unsigned int nbytes, blk_status_t error)
185 {
186 	if (error)
187 		bio->bi_status = error;
188 
189 	if (unlikely(rq->rq_flags & RQF_QUIET))
190 		bio_set_flag(bio, BIO_QUIET);
191 
192 	bio_advance(bio, nbytes);
193 
194 	/* don't actually finish bio if it's part of flush sequence */
195 	if (bio->bi_iter.bi_size == 0 && !(rq->rq_flags & RQF_FLUSH_SEQ))
196 		bio_endio(bio);
197 }
198 
199 void blk_dump_rq_flags(struct request *rq, char *msg)
200 {
201 	printk(KERN_INFO "%s: dev %s: flags=%llx\n", msg,
202 		rq->rq_disk ? rq->rq_disk->disk_name : "?",
203 		(unsigned long long) rq->cmd_flags);
204 
205 	printk(KERN_INFO "  sector %llu, nr/cnr %u/%u\n",
206 	       (unsigned long long)blk_rq_pos(rq),
207 	       blk_rq_sectors(rq), blk_rq_cur_sectors(rq));
208 	printk(KERN_INFO "  bio %p, biotail %p, len %u\n",
209 	       rq->bio, rq->biotail, blk_rq_bytes(rq));
210 }
211 EXPORT_SYMBOL(blk_dump_rq_flags);
212 
213 /**
214  * blk_sync_queue - cancel any pending callbacks on a queue
215  * @q: the queue
216  *
217  * Description:
218  *     The block layer may perform asynchronous callback activity
219  *     on a queue, such as calling the unplug function after a timeout.
220  *     A block device may call blk_sync_queue to ensure that any
221  *     such activity is cancelled, thus allowing it to release resources
222  *     that the callbacks might use. The caller must already have made sure
223  *     that its ->make_request_fn will not re-add plugging prior to calling
224  *     this function.
225  *
226  *     This function does not cancel any asynchronous activity arising
227  *     out of elevator or throttling code. That would require elevator_exit()
228  *     and blkcg_exit_queue() to be called with queue lock initialized.
229  *
230  */
231 void blk_sync_queue(struct request_queue *q)
232 {
233 	del_timer_sync(&q->timeout);
234 	cancel_work_sync(&q->timeout_work);
235 
236 	if (queue_is_mq(q)) {
237 		struct blk_mq_hw_ctx *hctx;
238 		int i;
239 
240 		cancel_delayed_work_sync(&q->requeue_work);
241 		queue_for_each_hw_ctx(q, hctx, i)
242 			cancel_delayed_work_sync(&hctx->run_work);
243 	}
244 }
245 EXPORT_SYMBOL(blk_sync_queue);
246 
247 /**
248  * blk_set_pm_only - increment pm_only counter
249  * @q: request queue pointer
250  */
251 void blk_set_pm_only(struct request_queue *q)
252 {
253 	atomic_inc(&q->pm_only);
254 }
255 EXPORT_SYMBOL_GPL(blk_set_pm_only);
256 
257 void blk_clear_pm_only(struct request_queue *q)
258 {
259 	int pm_only;
260 
261 	pm_only = atomic_dec_return(&q->pm_only);
262 	WARN_ON_ONCE(pm_only < 0);
263 	if (pm_only == 0)
264 		wake_up_all(&q->mq_freeze_wq);
265 }
266 EXPORT_SYMBOL_GPL(blk_clear_pm_only);
267 
268 void blk_put_queue(struct request_queue *q)
269 {
270 	kobject_put(&q->kobj);
271 }
272 EXPORT_SYMBOL(blk_put_queue);
273 
274 void blk_set_queue_dying(struct request_queue *q)
275 {
276 	blk_queue_flag_set(QUEUE_FLAG_DYING, q);
277 
278 	/*
279 	 * When queue DYING flag is set, we need to block new req
280 	 * entering queue, so we call blk_freeze_queue_start() to
281 	 * prevent I/O from crossing blk_queue_enter().
282 	 */
283 	blk_freeze_queue_start(q);
284 
285 	if (queue_is_mq(q))
286 		blk_mq_wake_waiters(q);
287 
288 	/* Make blk_queue_enter() reexamine the DYING flag. */
289 	wake_up_all(&q->mq_freeze_wq);
290 }
291 EXPORT_SYMBOL_GPL(blk_set_queue_dying);
292 
293 /* Unconfigure the I/O scheduler and dissociate from the cgroup controller. */
294 void blk_exit_queue(struct request_queue *q)
295 {
296 	/*
297 	 * Since the I/O scheduler exit code may access cgroup information,
298 	 * perform I/O scheduler exit before disassociating from the block
299 	 * cgroup controller.
300 	 */
301 	if (q->elevator) {
302 		ioc_clear_queue(q);
303 		elevator_exit(q, q->elevator);
304 		q->elevator = NULL;
305 	}
306 
307 	/*
308 	 * Remove all references to @q from the block cgroup controller before
309 	 * restoring @q->queue_lock to avoid that restoring this pointer causes
310 	 * e.g. blkcg_print_blkgs() to crash.
311 	 */
312 	blkcg_exit_queue(q);
313 
314 	/*
315 	 * Since the cgroup code may dereference the @q->backing_dev_info
316 	 * pointer, only decrease its reference count after having removed the
317 	 * association with the block cgroup controller.
318 	 */
319 	bdi_put(q->backing_dev_info);
320 }
321 
322 /**
323  * blk_cleanup_queue - shutdown a request queue
324  * @q: request queue to shutdown
325  *
326  * Mark @q DYING, drain all pending requests, mark @q DEAD, destroy and
327  * put it.  All future requests will be failed immediately with -ENODEV.
328  */
329 void blk_cleanup_queue(struct request_queue *q)
330 {
331 	/* mark @q DYING, no new request or merges will be allowed afterwards */
332 	mutex_lock(&q->sysfs_lock);
333 	blk_set_queue_dying(q);
334 
335 	blk_queue_flag_set(QUEUE_FLAG_NOMERGES, q);
336 	blk_queue_flag_set(QUEUE_FLAG_NOXMERGES, q);
337 	blk_queue_flag_set(QUEUE_FLAG_DYING, q);
338 	mutex_unlock(&q->sysfs_lock);
339 
340 	/*
341 	 * Drain all requests queued before DYING marking. Set DEAD flag to
342 	 * prevent that q->request_fn() gets invoked after draining finished.
343 	 */
344 	blk_freeze_queue(q);
345 
346 	rq_qos_exit(q);
347 
348 	blk_queue_flag_set(QUEUE_FLAG_DEAD, q);
349 
350 	/*
351 	 * make sure all in-progress dispatch are completed because
352 	 * blk_freeze_queue() can only complete all requests, and
353 	 * dispatch may still be in-progress since we dispatch requests
354 	 * from more than one contexts.
355 	 *
356 	 * We rely on driver to deal with the race in case that queue
357 	 * initialization isn't done.
358 	 */
359 	if (queue_is_mq(q) && blk_queue_init_done(q))
360 		blk_mq_quiesce_queue(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 	/*
370 	 * I/O scheduler exit is only safe after the sysfs scheduler attribute
371 	 * has been removed.
372 	 */
373 	WARN_ON_ONCE(q->kobj.state_in_sysfs);
374 
375 	blk_exit_queue(q);
376 
377 	if (queue_is_mq(q))
378 		blk_mq_free_queue(q);
379 
380 	percpu_ref_exit(&q->q_usage_counter);
381 
382 	/* @q is and will stay empty, shutdown and put */
383 	blk_put_queue(q);
384 }
385 EXPORT_SYMBOL(blk_cleanup_queue);
386 
387 struct request_queue *blk_alloc_queue(gfp_t gfp_mask)
388 {
389 	return blk_alloc_queue_node(gfp_mask, NUMA_NO_NODE);
390 }
391 EXPORT_SYMBOL(blk_alloc_queue);
392 
393 /**
394  * blk_queue_enter() - try to increase q->q_usage_counter
395  * @q: request queue pointer
396  * @flags: BLK_MQ_REQ_NOWAIT and/or BLK_MQ_REQ_PREEMPT
397  */
398 int blk_queue_enter(struct request_queue *q, blk_mq_req_flags_t flags)
399 {
400 	const bool pm = flags & BLK_MQ_REQ_PREEMPT;
401 
402 	while (true) {
403 		bool success = false;
404 
405 		rcu_read_lock();
406 		if (percpu_ref_tryget_live(&q->q_usage_counter)) {
407 			/*
408 			 * The code that increments the pm_only counter is
409 			 * responsible for ensuring that that counter is
410 			 * globally visible before the queue is unfrozen.
411 			 */
412 			if (pm || !blk_queue_pm_only(q)) {
413 				success = true;
414 			} else {
415 				percpu_ref_put(&q->q_usage_counter);
416 			}
417 		}
418 		rcu_read_unlock();
419 
420 		if (success)
421 			return 0;
422 
423 		if (flags & BLK_MQ_REQ_NOWAIT)
424 			return -EBUSY;
425 
426 		/*
427 		 * read pair of barrier in blk_freeze_queue_start(),
428 		 * we need to order reading __PERCPU_REF_DEAD flag of
429 		 * .q_usage_counter and reading .mq_freeze_depth or
430 		 * queue dying flag, otherwise the following wait may
431 		 * never return if the two reads are reordered.
432 		 */
433 		smp_rmb();
434 
435 		wait_event(q->mq_freeze_wq,
436 			   (atomic_read(&q->mq_freeze_depth) == 0 &&
437 			    (pm || (blk_pm_request_resume(q),
438 				    !blk_queue_pm_only(q)))) ||
439 			   blk_queue_dying(q));
440 		if (blk_queue_dying(q))
441 			return -ENODEV;
442 	}
443 }
444 
445 void blk_queue_exit(struct request_queue *q)
446 {
447 	percpu_ref_put(&q->q_usage_counter);
448 }
449 
450 static void blk_queue_usage_counter_release(struct percpu_ref *ref)
451 {
452 	struct request_queue *q =
453 		container_of(ref, struct request_queue, q_usage_counter);
454 
455 	wake_up_all(&q->mq_freeze_wq);
456 }
457 
458 static void blk_rq_timed_out_timer(struct timer_list *t)
459 {
460 	struct request_queue *q = from_timer(q, t, timeout);
461 
462 	kblockd_schedule_work(&q->timeout_work);
463 }
464 
465 /**
466  * blk_alloc_queue_node - allocate a request queue
467  * @gfp_mask: memory allocation flags
468  * @node_id: NUMA node to allocate memory from
469  */
470 struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id)
471 {
472 	struct request_queue *q;
473 	int ret;
474 
475 	q = kmem_cache_alloc_node(blk_requestq_cachep,
476 				gfp_mask | __GFP_ZERO, node_id);
477 	if (!q)
478 		return NULL;
479 
480 	INIT_LIST_HEAD(&q->queue_head);
481 	q->last_merge = NULL;
482 
483 	q->id = ida_simple_get(&blk_queue_ida, 0, 0, gfp_mask);
484 	if (q->id < 0)
485 		goto fail_q;
486 
487 	ret = bioset_init(&q->bio_split, BIO_POOL_SIZE, 0, BIOSET_NEED_BVECS);
488 	if (ret)
489 		goto fail_id;
490 
491 	q->backing_dev_info = bdi_alloc_node(gfp_mask, node_id);
492 	if (!q->backing_dev_info)
493 		goto fail_split;
494 
495 	q->stats = blk_alloc_queue_stats();
496 	if (!q->stats)
497 		goto fail_stats;
498 
499 	q->backing_dev_info->ra_pages =
500 			(VM_MAX_READAHEAD * 1024) / PAGE_SIZE;
501 	q->backing_dev_info->capabilities = BDI_CAP_CGROUP_WRITEBACK;
502 	q->backing_dev_info->name = "block";
503 	q->node = node_id;
504 
505 	timer_setup(&q->backing_dev_info->laptop_mode_wb_timer,
506 		    laptop_mode_timer_fn, 0);
507 	timer_setup(&q->timeout, blk_rq_timed_out_timer, 0);
508 	INIT_WORK(&q->timeout_work, NULL);
509 	INIT_LIST_HEAD(&q->icq_list);
510 #ifdef CONFIG_BLK_CGROUP
511 	INIT_LIST_HEAD(&q->blkg_list);
512 #endif
513 
514 	kobject_init(&q->kobj, &blk_queue_ktype);
515 
516 #ifdef CONFIG_BLK_DEV_IO_TRACE
517 	mutex_init(&q->blk_trace_mutex);
518 #endif
519 	mutex_init(&q->sysfs_lock);
520 	spin_lock_init(&q->queue_lock);
521 
522 	init_waitqueue_head(&q->mq_freeze_wq);
523 
524 	/*
525 	 * Init percpu_ref in atomic mode so that it's faster to shutdown.
526 	 * See blk_register_queue() for details.
527 	 */
528 	if (percpu_ref_init(&q->q_usage_counter,
529 				blk_queue_usage_counter_release,
530 				PERCPU_REF_INIT_ATOMIC, GFP_KERNEL))
531 		goto fail_bdi;
532 
533 	if (blkcg_init_queue(q))
534 		goto fail_ref;
535 
536 	return q;
537 
538 fail_ref:
539 	percpu_ref_exit(&q->q_usage_counter);
540 fail_bdi:
541 	blk_free_queue_stats(q->stats);
542 fail_stats:
543 	bdi_put(q->backing_dev_info);
544 fail_split:
545 	bioset_exit(&q->bio_split);
546 fail_id:
547 	ida_simple_remove(&blk_queue_ida, q->id);
548 fail_q:
549 	kmem_cache_free(blk_requestq_cachep, q);
550 	return NULL;
551 }
552 EXPORT_SYMBOL(blk_alloc_queue_node);
553 
554 bool blk_get_queue(struct request_queue *q)
555 {
556 	if (likely(!blk_queue_dying(q))) {
557 		__blk_get_queue(q);
558 		return true;
559 	}
560 
561 	return false;
562 }
563 EXPORT_SYMBOL(blk_get_queue);
564 
565 /**
566  * blk_get_request - allocate a request
567  * @q: request queue to allocate a request for
568  * @op: operation (REQ_OP_*) and REQ_* flags, e.g. REQ_SYNC.
569  * @flags: BLK_MQ_REQ_* flags, e.g. BLK_MQ_REQ_NOWAIT.
570  */
571 struct request *blk_get_request(struct request_queue *q, unsigned int op,
572 				blk_mq_req_flags_t flags)
573 {
574 	struct request *req;
575 
576 	WARN_ON_ONCE(op & REQ_NOWAIT);
577 	WARN_ON_ONCE(flags & ~(BLK_MQ_REQ_NOWAIT | BLK_MQ_REQ_PREEMPT));
578 
579 	req = blk_mq_alloc_request(q, op, flags);
580 	if (!IS_ERR(req) && q->mq_ops->initialize_rq_fn)
581 		q->mq_ops->initialize_rq_fn(req);
582 
583 	return req;
584 }
585 EXPORT_SYMBOL(blk_get_request);
586 
587 void blk_put_request(struct request *req)
588 {
589 	blk_mq_free_request(req);
590 }
591 EXPORT_SYMBOL(blk_put_request);
592 
593 bool bio_attempt_back_merge(struct request_queue *q, struct request *req,
594 			    struct bio *bio)
595 {
596 	const int ff = bio->bi_opf & REQ_FAILFAST_MASK;
597 
598 	if (!ll_back_merge_fn(q, req, bio))
599 		return false;
600 
601 	trace_block_bio_backmerge(q, req, bio);
602 
603 	if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
604 		blk_rq_set_mixed_merge(req);
605 
606 	req->biotail->bi_next = bio;
607 	req->biotail = bio;
608 	req->__data_len += bio->bi_iter.bi_size;
609 
610 	blk_account_io_start(req, false);
611 	return true;
612 }
613 
614 bool bio_attempt_front_merge(struct request_queue *q, struct request *req,
615 			     struct bio *bio)
616 {
617 	const int ff = bio->bi_opf & REQ_FAILFAST_MASK;
618 
619 	if (!ll_front_merge_fn(q, req, bio))
620 		return false;
621 
622 	trace_block_bio_frontmerge(q, req, bio);
623 
624 	if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
625 		blk_rq_set_mixed_merge(req);
626 
627 	bio->bi_next = req->bio;
628 	req->bio = bio;
629 
630 	req->__sector = bio->bi_iter.bi_sector;
631 	req->__data_len += bio->bi_iter.bi_size;
632 
633 	blk_account_io_start(req, false);
634 	return true;
635 }
636 
637 bool bio_attempt_discard_merge(struct request_queue *q, struct request *req,
638 		struct bio *bio)
639 {
640 	unsigned short segments = blk_rq_nr_discard_segments(req);
641 
642 	if (segments >= queue_max_discard_segments(q))
643 		goto no_merge;
644 	if (blk_rq_sectors(req) + bio_sectors(bio) >
645 	    blk_rq_get_max_sectors(req, blk_rq_pos(req)))
646 		goto no_merge;
647 
648 	req->biotail->bi_next = bio;
649 	req->biotail = bio;
650 	req->__data_len += bio->bi_iter.bi_size;
651 	req->nr_phys_segments = segments + 1;
652 
653 	blk_account_io_start(req, false);
654 	return true;
655 no_merge:
656 	req_set_nomerge(q, req);
657 	return false;
658 }
659 
660 /**
661  * blk_attempt_plug_merge - try to merge with %current's plugged list
662  * @q: request_queue new bio is being queued at
663  * @bio: new bio being queued
664  * @same_queue_rq: pointer to &struct request that gets filled in when
665  * another request associated with @q is found on the plug list
666  * (optional, may be %NULL)
667  *
668  * Determine whether @bio being queued on @q can be merged with a request
669  * on %current's plugged list.  Returns %true if merge was successful,
670  * otherwise %false.
671  *
672  * Plugging coalesces IOs from the same issuer for the same purpose without
673  * going through @q->queue_lock.  As such it's more of an issuing mechanism
674  * than scheduling, and the request, while may have elvpriv data, is not
675  * added on the elevator at this point.  In addition, we don't have
676  * reliable access to the elevator outside queue lock.  Only check basic
677  * merging parameters without querying the elevator.
678  *
679  * Caller must ensure !blk_queue_nomerges(q) beforehand.
680  */
681 bool blk_attempt_plug_merge(struct request_queue *q, struct bio *bio,
682 			    struct request **same_queue_rq)
683 {
684 	struct blk_plug *plug;
685 	struct request *rq;
686 	struct list_head *plug_list;
687 
688 	plug = current->plug;
689 	if (!plug)
690 		return false;
691 
692 	plug_list = &plug->mq_list;
693 
694 	list_for_each_entry_reverse(rq, plug_list, queuelist) {
695 		bool merged = false;
696 
697 		if (rq->q == q && same_queue_rq) {
698 			/*
699 			 * Only blk-mq multiple hardware queues case checks the
700 			 * rq in the same queue, there should be only one such
701 			 * rq in a queue
702 			 **/
703 			*same_queue_rq = rq;
704 		}
705 
706 		if (rq->q != q || !blk_rq_merge_ok(rq, bio))
707 			continue;
708 
709 		switch (blk_try_merge(rq, bio)) {
710 		case ELEVATOR_BACK_MERGE:
711 			merged = bio_attempt_back_merge(q, rq, bio);
712 			break;
713 		case ELEVATOR_FRONT_MERGE:
714 			merged = bio_attempt_front_merge(q, rq, bio);
715 			break;
716 		case ELEVATOR_DISCARD_MERGE:
717 			merged = bio_attempt_discard_merge(q, rq, bio);
718 			break;
719 		default:
720 			break;
721 		}
722 
723 		if (merged)
724 			return true;
725 	}
726 
727 	return false;
728 }
729 
730 void blk_init_request_from_bio(struct request *req, struct bio *bio)
731 {
732 	if (bio->bi_opf & REQ_RAHEAD)
733 		req->cmd_flags |= REQ_FAILFAST_MASK;
734 
735 	req->__sector = bio->bi_iter.bi_sector;
736 	req->ioprio = bio_prio(bio);
737 	req->write_hint = bio->bi_write_hint;
738 	blk_rq_bio_prep(req->q, req, bio);
739 }
740 EXPORT_SYMBOL_GPL(blk_init_request_from_bio);
741 
742 static void handle_bad_sector(struct bio *bio, sector_t maxsector)
743 {
744 	char b[BDEVNAME_SIZE];
745 
746 	printk(KERN_INFO "attempt to access beyond end of device\n");
747 	printk(KERN_INFO "%s: rw=%d, want=%Lu, limit=%Lu\n",
748 			bio_devname(bio, b), bio->bi_opf,
749 			(unsigned long long)bio_end_sector(bio),
750 			(long long)maxsector);
751 }
752 
753 #ifdef CONFIG_FAIL_MAKE_REQUEST
754 
755 static DECLARE_FAULT_ATTR(fail_make_request);
756 
757 static int __init setup_fail_make_request(char *str)
758 {
759 	return setup_fault_attr(&fail_make_request, str);
760 }
761 __setup("fail_make_request=", setup_fail_make_request);
762 
763 static bool should_fail_request(struct hd_struct *part, unsigned int bytes)
764 {
765 	return part->make_it_fail && should_fail(&fail_make_request, bytes);
766 }
767 
768 static int __init fail_make_request_debugfs(void)
769 {
770 	struct dentry *dir = fault_create_debugfs_attr("fail_make_request",
771 						NULL, &fail_make_request);
772 
773 	return PTR_ERR_OR_ZERO(dir);
774 }
775 
776 late_initcall(fail_make_request_debugfs);
777 
778 #else /* CONFIG_FAIL_MAKE_REQUEST */
779 
780 static inline bool should_fail_request(struct hd_struct *part,
781 					unsigned int bytes)
782 {
783 	return false;
784 }
785 
786 #endif /* CONFIG_FAIL_MAKE_REQUEST */
787 
788 static inline bool bio_check_ro(struct bio *bio, struct hd_struct *part)
789 {
790 	const int op = bio_op(bio);
791 
792 	if (part->policy && op_is_write(op)) {
793 		char b[BDEVNAME_SIZE];
794 
795 		if (op_is_flush(bio->bi_opf) && !bio_sectors(bio))
796 			return false;
797 
798 		WARN_ONCE(1,
799 		       "generic_make_request: Trying to write "
800 			"to read-only block-device %s (partno %d)\n",
801 			bio_devname(bio, b), part->partno);
802 		/* Older lvm-tools actually trigger this */
803 		return false;
804 	}
805 
806 	return false;
807 }
808 
809 static noinline int should_fail_bio(struct bio *bio)
810 {
811 	if (should_fail_request(&bio->bi_disk->part0, bio->bi_iter.bi_size))
812 		return -EIO;
813 	return 0;
814 }
815 ALLOW_ERROR_INJECTION(should_fail_bio, ERRNO);
816 
817 /*
818  * Check whether this bio extends beyond the end of the device or partition.
819  * This may well happen - the kernel calls bread() without checking the size of
820  * the device, e.g., when mounting a file system.
821  */
822 static inline int bio_check_eod(struct bio *bio, sector_t maxsector)
823 {
824 	unsigned int nr_sectors = bio_sectors(bio);
825 
826 	if (nr_sectors && maxsector &&
827 	    (nr_sectors > maxsector ||
828 	     bio->bi_iter.bi_sector > maxsector - nr_sectors)) {
829 		handle_bad_sector(bio, maxsector);
830 		return -EIO;
831 	}
832 	return 0;
833 }
834 
835 /*
836  * Remap block n of partition p to block n+start(p) of the disk.
837  */
838 static inline int blk_partition_remap(struct bio *bio)
839 {
840 	struct hd_struct *p;
841 	int ret = -EIO;
842 
843 	rcu_read_lock();
844 	p = __disk_get_part(bio->bi_disk, bio->bi_partno);
845 	if (unlikely(!p))
846 		goto out;
847 	if (unlikely(should_fail_request(p, bio->bi_iter.bi_size)))
848 		goto out;
849 	if (unlikely(bio_check_ro(bio, p)))
850 		goto out;
851 
852 	/*
853 	 * Zone reset does not include bi_size so bio_sectors() is always 0.
854 	 * Include a test for the reset op code and perform the remap if needed.
855 	 */
856 	if (bio_sectors(bio) || bio_op(bio) == REQ_OP_ZONE_RESET) {
857 		if (bio_check_eod(bio, part_nr_sects_read(p)))
858 			goto out;
859 		bio->bi_iter.bi_sector += p->start_sect;
860 		trace_block_bio_remap(bio->bi_disk->queue, bio, part_devt(p),
861 				      bio->bi_iter.bi_sector - p->start_sect);
862 	}
863 	bio->bi_partno = 0;
864 	ret = 0;
865 out:
866 	rcu_read_unlock();
867 	return ret;
868 }
869 
870 static noinline_for_stack bool
871 generic_make_request_checks(struct bio *bio)
872 {
873 	struct request_queue *q;
874 	int nr_sectors = bio_sectors(bio);
875 	blk_status_t status = BLK_STS_IOERR;
876 	char b[BDEVNAME_SIZE];
877 
878 	might_sleep();
879 
880 	q = bio->bi_disk->queue;
881 	if (unlikely(!q)) {
882 		printk(KERN_ERR
883 		       "generic_make_request: Trying to access "
884 			"nonexistent block-device %s (%Lu)\n",
885 			bio_devname(bio, b), (long long)bio->bi_iter.bi_sector);
886 		goto end_io;
887 	}
888 
889 	/*
890 	 * For a REQ_NOWAIT based request, return -EOPNOTSUPP
891 	 * if queue is not a request based queue.
892 	 */
893 	if ((bio->bi_opf & REQ_NOWAIT) && !queue_is_mq(q))
894 		goto not_supported;
895 
896 	if (should_fail_bio(bio))
897 		goto end_io;
898 
899 	if (bio->bi_partno) {
900 		if (unlikely(blk_partition_remap(bio)))
901 			goto end_io;
902 	} else {
903 		if (unlikely(bio_check_ro(bio, &bio->bi_disk->part0)))
904 			goto end_io;
905 		if (unlikely(bio_check_eod(bio, get_capacity(bio->bi_disk))))
906 			goto end_io;
907 	}
908 
909 	/*
910 	 * Filter flush bio's early so that make_request based
911 	 * drivers without flush support don't have to worry
912 	 * about them.
913 	 */
914 	if (op_is_flush(bio->bi_opf) &&
915 	    !test_bit(QUEUE_FLAG_WC, &q->queue_flags)) {
916 		bio->bi_opf &= ~(REQ_PREFLUSH | REQ_FUA);
917 		if (!nr_sectors) {
918 			status = BLK_STS_OK;
919 			goto end_io;
920 		}
921 	}
922 
923 	if (!test_bit(QUEUE_FLAG_POLL, &q->queue_flags))
924 		bio->bi_opf &= ~REQ_HIPRI;
925 
926 	switch (bio_op(bio)) {
927 	case REQ_OP_DISCARD:
928 		if (!blk_queue_discard(q))
929 			goto not_supported;
930 		break;
931 	case REQ_OP_SECURE_ERASE:
932 		if (!blk_queue_secure_erase(q))
933 			goto not_supported;
934 		break;
935 	case REQ_OP_WRITE_SAME:
936 		if (!q->limits.max_write_same_sectors)
937 			goto not_supported;
938 		break;
939 	case REQ_OP_ZONE_RESET:
940 		if (!blk_queue_is_zoned(q))
941 			goto not_supported;
942 		break;
943 	case REQ_OP_WRITE_ZEROES:
944 		if (!q->limits.max_write_zeroes_sectors)
945 			goto not_supported;
946 		break;
947 	default:
948 		break;
949 	}
950 
951 	/*
952 	 * Various block parts want %current->io_context and lazy ioc
953 	 * allocation ends up trading a lot of pain for a small amount of
954 	 * memory.  Just allocate it upfront.  This may fail and block
955 	 * layer knows how to live with it.
956 	 */
957 	create_io_context(GFP_ATOMIC, q->node);
958 
959 	if (!blkcg_bio_issue_check(q, bio))
960 		return false;
961 
962 	if (!bio_flagged(bio, BIO_TRACE_COMPLETION)) {
963 		trace_block_bio_queue(q, bio);
964 		/* Now that enqueuing has been traced, we need to trace
965 		 * completion as well.
966 		 */
967 		bio_set_flag(bio, BIO_TRACE_COMPLETION);
968 	}
969 	return true;
970 
971 not_supported:
972 	status = BLK_STS_NOTSUPP;
973 end_io:
974 	bio->bi_status = status;
975 	bio_endio(bio);
976 	return false;
977 }
978 
979 /**
980  * generic_make_request - hand a buffer to its device driver for I/O
981  * @bio:  The bio describing the location in memory and on the device.
982  *
983  * generic_make_request() is used to make I/O requests of block
984  * devices. It is passed a &struct bio, which describes the I/O that needs
985  * to be done.
986  *
987  * generic_make_request() does not return any status.  The
988  * success/failure status of the request, along with notification of
989  * completion, is delivered asynchronously through the bio->bi_end_io
990  * function described (one day) else where.
991  *
992  * The caller of generic_make_request must make sure that bi_io_vec
993  * are set to describe the memory buffer, and that bi_dev and bi_sector are
994  * set to describe the device address, and the
995  * bi_end_io and optionally bi_private are set to describe how
996  * completion notification should be signaled.
997  *
998  * generic_make_request and the drivers it calls may use bi_next if this
999  * bio happens to be merged with someone else, and may resubmit the bio to
1000  * a lower device by calling into generic_make_request recursively, which
1001  * means the bio should NOT be touched after the call to ->make_request_fn.
1002  */
1003 blk_qc_t generic_make_request(struct bio *bio)
1004 {
1005 	/*
1006 	 * bio_list_on_stack[0] contains bios submitted by the current
1007 	 * make_request_fn.
1008 	 * bio_list_on_stack[1] contains bios that were submitted before
1009 	 * the current make_request_fn, but that haven't been processed
1010 	 * yet.
1011 	 */
1012 	struct bio_list bio_list_on_stack[2];
1013 	blk_mq_req_flags_t flags = 0;
1014 	struct request_queue *q = bio->bi_disk->queue;
1015 	blk_qc_t ret = BLK_QC_T_NONE;
1016 
1017 	if (bio->bi_opf & REQ_NOWAIT)
1018 		flags = BLK_MQ_REQ_NOWAIT;
1019 	if (bio_flagged(bio, BIO_QUEUE_ENTERED))
1020 		blk_queue_enter_live(q);
1021 	else if (blk_queue_enter(q, flags) < 0) {
1022 		if (!blk_queue_dying(q) && (bio->bi_opf & REQ_NOWAIT))
1023 			bio_wouldblock_error(bio);
1024 		else
1025 			bio_io_error(bio);
1026 		return ret;
1027 	}
1028 
1029 	if (!generic_make_request_checks(bio))
1030 		goto out;
1031 
1032 	/*
1033 	 * We only want one ->make_request_fn to be active at a time, else
1034 	 * stack usage with stacked devices could be a problem.  So use
1035 	 * current->bio_list to keep a list of requests submited by a
1036 	 * make_request_fn function.  current->bio_list is also used as a
1037 	 * flag to say if generic_make_request is currently active in this
1038 	 * task or not.  If it is NULL, then no make_request is active.  If
1039 	 * it is non-NULL, then a make_request is active, and new requests
1040 	 * should be added at the tail
1041 	 */
1042 	if (current->bio_list) {
1043 		bio_list_add(&current->bio_list[0], bio);
1044 		goto out;
1045 	}
1046 
1047 	/* following loop may be a bit non-obvious, and so deserves some
1048 	 * explanation.
1049 	 * Before entering the loop, bio->bi_next is NULL (as all callers
1050 	 * ensure that) so we have a list with a single bio.
1051 	 * We pretend that we have just taken it off a longer list, so
1052 	 * we assign bio_list to a pointer to the bio_list_on_stack,
1053 	 * thus initialising the bio_list of new bios to be
1054 	 * added.  ->make_request() may indeed add some more bios
1055 	 * through a recursive call to generic_make_request.  If it
1056 	 * did, we find a non-NULL value in bio_list and re-enter the loop
1057 	 * from the top.  In this case we really did just take the bio
1058 	 * of the top of the list (no pretending) and so remove it from
1059 	 * bio_list, and call into ->make_request() again.
1060 	 */
1061 	BUG_ON(bio->bi_next);
1062 	bio_list_init(&bio_list_on_stack[0]);
1063 	current->bio_list = bio_list_on_stack;
1064 	do {
1065 		bool enter_succeeded = true;
1066 
1067 		if (unlikely(q != bio->bi_disk->queue)) {
1068 			if (q)
1069 				blk_queue_exit(q);
1070 			q = bio->bi_disk->queue;
1071 			flags = 0;
1072 			if (bio->bi_opf & REQ_NOWAIT)
1073 				flags = BLK_MQ_REQ_NOWAIT;
1074 			if (blk_queue_enter(q, flags) < 0) {
1075 				enter_succeeded = false;
1076 				q = NULL;
1077 			}
1078 		}
1079 
1080 		if (enter_succeeded) {
1081 			struct bio_list lower, same;
1082 
1083 			/* Create a fresh bio_list for all subordinate requests */
1084 			bio_list_on_stack[1] = bio_list_on_stack[0];
1085 			bio_list_init(&bio_list_on_stack[0]);
1086 
1087 			/*
1088 			 * Since we're recursing into make_request here, ensure
1089 			 * that we mark this bio as already having entered the queue.
1090 			 * If not, and the queue is going away, we can get stuck
1091 			 * forever on waiting for the queue reference to drop. But
1092 			 * that will never happen, as we're already holding a
1093 			 * reference to it.
1094 			 */
1095 			bio_set_flag(bio, BIO_QUEUE_ENTERED);
1096 			ret = q->make_request_fn(q, bio);
1097 			bio_clear_flag(bio, BIO_QUEUE_ENTERED);
1098 
1099 			/* sort new bios into those for a lower level
1100 			 * and those for the same level
1101 			 */
1102 			bio_list_init(&lower);
1103 			bio_list_init(&same);
1104 			while ((bio = bio_list_pop(&bio_list_on_stack[0])) != NULL)
1105 				if (q == bio->bi_disk->queue)
1106 					bio_list_add(&same, bio);
1107 				else
1108 					bio_list_add(&lower, bio);
1109 			/* now assemble so we handle the lowest level first */
1110 			bio_list_merge(&bio_list_on_stack[0], &lower);
1111 			bio_list_merge(&bio_list_on_stack[0], &same);
1112 			bio_list_merge(&bio_list_on_stack[0], &bio_list_on_stack[1]);
1113 		} else {
1114 			if (unlikely(!blk_queue_dying(q) &&
1115 					(bio->bi_opf & REQ_NOWAIT)))
1116 				bio_wouldblock_error(bio);
1117 			else
1118 				bio_io_error(bio);
1119 		}
1120 		bio = bio_list_pop(&bio_list_on_stack[0]);
1121 	} while (bio);
1122 	current->bio_list = NULL; /* deactivate */
1123 
1124 out:
1125 	if (q)
1126 		blk_queue_exit(q);
1127 	return ret;
1128 }
1129 EXPORT_SYMBOL(generic_make_request);
1130 
1131 /**
1132  * direct_make_request - hand a buffer directly to its device driver for I/O
1133  * @bio:  The bio describing the location in memory and on the device.
1134  *
1135  * This function behaves like generic_make_request(), but does not protect
1136  * against recursion.  Must only be used if the called driver is known
1137  * to not call generic_make_request (or direct_make_request) again from
1138  * its make_request function.  (Calling direct_make_request again from
1139  * a workqueue is perfectly fine as that doesn't recurse).
1140  */
1141 blk_qc_t direct_make_request(struct bio *bio)
1142 {
1143 	struct request_queue *q = bio->bi_disk->queue;
1144 	bool nowait = bio->bi_opf & REQ_NOWAIT;
1145 	blk_qc_t ret;
1146 
1147 	if (!generic_make_request_checks(bio))
1148 		return BLK_QC_T_NONE;
1149 
1150 	if (unlikely(blk_queue_enter(q, nowait ? BLK_MQ_REQ_NOWAIT : 0))) {
1151 		if (nowait && !blk_queue_dying(q))
1152 			bio->bi_status = BLK_STS_AGAIN;
1153 		else
1154 			bio->bi_status = BLK_STS_IOERR;
1155 		bio_endio(bio);
1156 		return BLK_QC_T_NONE;
1157 	}
1158 
1159 	ret = q->make_request_fn(q, bio);
1160 	blk_queue_exit(q);
1161 	return ret;
1162 }
1163 EXPORT_SYMBOL_GPL(direct_make_request);
1164 
1165 /**
1166  * submit_bio - submit a bio to the block device layer for I/O
1167  * @bio: The &struct bio which describes the I/O
1168  *
1169  * submit_bio() is very similar in purpose to generic_make_request(), and
1170  * uses that function to do most of the work. Both are fairly rough
1171  * interfaces; @bio must be presetup and ready for I/O.
1172  *
1173  */
1174 blk_qc_t submit_bio(struct bio *bio)
1175 {
1176 	/*
1177 	 * If it's a regular read/write or a barrier with data attached,
1178 	 * go through the normal accounting stuff before submission.
1179 	 */
1180 	if (bio_has_data(bio)) {
1181 		unsigned int count;
1182 
1183 		if (unlikely(bio_op(bio) == REQ_OP_WRITE_SAME))
1184 			count = queue_logical_block_size(bio->bi_disk->queue) >> 9;
1185 		else
1186 			count = bio_sectors(bio);
1187 
1188 		if (op_is_write(bio_op(bio))) {
1189 			count_vm_events(PGPGOUT, count);
1190 		} else {
1191 			task_io_account_read(bio->bi_iter.bi_size);
1192 			count_vm_events(PGPGIN, count);
1193 		}
1194 
1195 		if (unlikely(block_dump)) {
1196 			char b[BDEVNAME_SIZE];
1197 			printk(KERN_DEBUG "%s(%d): %s block %Lu on %s (%u sectors)\n",
1198 			current->comm, task_pid_nr(current),
1199 				op_is_write(bio_op(bio)) ? "WRITE" : "READ",
1200 				(unsigned long long)bio->bi_iter.bi_sector,
1201 				bio_devname(bio, b), count);
1202 		}
1203 	}
1204 
1205 	return generic_make_request(bio);
1206 }
1207 EXPORT_SYMBOL(submit_bio);
1208 
1209 /**
1210  * blk_cloned_rq_check_limits - Helper function to check a cloned request
1211  *                              for new the queue limits
1212  * @q:  the queue
1213  * @rq: the request being checked
1214  *
1215  * Description:
1216  *    @rq may have been made based on weaker limitations of upper-level queues
1217  *    in request stacking drivers, and it may violate the limitation of @q.
1218  *    Since the block layer and the underlying device driver trust @rq
1219  *    after it is inserted to @q, it should be checked against @q before
1220  *    the insertion using this generic function.
1221  *
1222  *    Request stacking drivers like request-based dm may change the queue
1223  *    limits when retrying requests on other queues. Those requests need
1224  *    to be checked against the new queue limits again during dispatch.
1225  */
1226 static int blk_cloned_rq_check_limits(struct request_queue *q,
1227 				      struct request *rq)
1228 {
1229 	if (blk_rq_sectors(rq) > blk_queue_get_max_sectors(q, req_op(rq))) {
1230 		printk(KERN_ERR "%s: over max size limit.\n", __func__);
1231 		return -EIO;
1232 	}
1233 
1234 	/*
1235 	 * queue's settings related to segment counting like q->bounce_pfn
1236 	 * may differ from that of other stacking queues.
1237 	 * Recalculate it to check the request correctly on this queue's
1238 	 * limitation.
1239 	 */
1240 	blk_recalc_rq_segments(rq);
1241 	if (rq->nr_phys_segments > queue_max_segments(q)) {
1242 		printk(KERN_ERR "%s: over max segments limit.\n", __func__);
1243 		return -EIO;
1244 	}
1245 
1246 	return 0;
1247 }
1248 
1249 /**
1250  * blk_insert_cloned_request - Helper for stacking drivers to submit a request
1251  * @q:  the queue to submit the request
1252  * @rq: the request being queued
1253  */
1254 blk_status_t blk_insert_cloned_request(struct request_queue *q, struct request *rq)
1255 {
1256 	blk_qc_t unused;
1257 
1258 	if (blk_cloned_rq_check_limits(q, rq))
1259 		return BLK_STS_IOERR;
1260 
1261 	if (rq->rq_disk &&
1262 	    should_fail_request(&rq->rq_disk->part0, blk_rq_bytes(rq)))
1263 		return BLK_STS_IOERR;
1264 
1265 	if (blk_queue_io_stat(q))
1266 		blk_account_io_start(rq, true);
1267 
1268 	/*
1269 	 * Since we have a scheduler attached on the top device,
1270 	 * bypass a potential scheduler on the bottom device for
1271 	 * insert.
1272 	 */
1273 	return blk_mq_try_issue_directly(rq->mq_hctx, rq, &unused, true, true);
1274 }
1275 EXPORT_SYMBOL_GPL(blk_insert_cloned_request);
1276 
1277 /**
1278  * blk_rq_err_bytes - determine number of bytes till the next failure boundary
1279  * @rq: request to examine
1280  *
1281  * Description:
1282  *     A request could be merge of IOs which require different failure
1283  *     handling.  This function determines the number of bytes which
1284  *     can be failed from the beginning of the request without
1285  *     crossing into area which need to be retried further.
1286  *
1287  * Return:
1288  *     The number of bytes to fail.
1289  */
1290 unsigned int blk_rq_err_bytes(const struct request *rq)
1291 {
1292 	unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK;
1293 	unsigned int bytes = 0;
1294 	struct bio *bio;
1295 
1296 	if (!(rq->rq_flags & RQF_MIXED_MERGE))
1297 		return blk_rq_bytes(rq);
1298 
1299 	/*
1300 	 * Currently the only 'mixing' which can happen is between
1301 	 * different fastfail types.  We can safely fail portions
1302 	 * which have all the failfast bits that the first one has -
1303 	 * the ones which are at least as eager to fail as the first
1304 	 * one.
1305 	 */
1306 	for (bio = rq->bio; bio; bio = bio->bi_next) {
1307 		if ((bio->bi_opf & ff) != ff)
1308 			break;
1309 		bytes += bio->bi_iter.bi_size;
1310 	}
1311 
1312 	/* this could lead to infinite loop */
1313 	BUG_ON(blk_rq_bytes(rq) && !bytes);
1314 	return bytes;
1315 }
1316 EXPORT_SYMBOL_GPL(blk_rq_err_bytes);
1317 
1318 void blk_account_io_completion(struct request *req, unsigned int bytes)
1319 {
1320 	if (blk_do_io_stat(req)) {
1321 		const int sgrp = op_stat_group(req_op(req));
1322 		struct hd_struct *part;
1323 
1324 		part_stat_lock();
1325 		part = req->part;
1326 		part_stat_add(part, sectors[sgrp], bytes >> 9);
1327 		part_stat_unlock();
1328 	}
1329 }
1330 
1331 void blk_account_io_done(struct request *req, u64 now)
1332 {
1333 	/*
1334 	 * Account IO completion.  flush_rq isn't accounted as a
1335 	 * normal IO on queueing nor completion.  Accounting the
1336 	 * containing request is enough.
1337 	 */
1338 	if (blk_do_io_stat(req) && !(req->rq_flags & RQF_FLUSH_SEQ)) {
1339 		const int sgrp = op_stat_group(req_op(req));
1340 		struct hd_struct *part;
1341 
1342 		part_stat_lock();
1343 		part = req->part;
1344 
1345 		update_io_ticks(part, jiffies);
1346 		part_stat_inc(part, ios[sgrp]);
1347 		part_stat_add(part, nsecs[sgrp], now - req->start_time_ns);
1348 		part_stat_add(part, time_in_queue, nsecs_to_jiffies64(now - req->start_time_ns));
1349 		part_dec_in_flight(req->q, part, rq_data_dir(req));
1350 
1351 		hd_struct_put(part);
1352 		part_stat_unlock();
1353 	}
1354 }
1355 
1356 void blk_account_io_start(struct request *rq, bool new_io)
1357 {
1358 	struct hd_struct *part;
1359 	int rw = rq_data_dir(rq);
1360 
1361 	if (!blk_do_io_stat(rq))
1362 		return;
1363 
1364 	part_stat_lock();
1365 
1366 	if (!new_io) {
1367 		part = rq->part;
1368 		part_stat_inc(part, merges[rw]);
1369 	} else {
1370 		part = disk_map_sector_rcu(rq->rq_disk, blk_rq_pos(rq));
1371 		if (!hd_struct_try_get(part)) {
1372 			/*
1373 			 * The partition is already being removed,
1374 			 * the request will be accounted on the disk only
1375 			 *
1376 			 * We take a reference on disk->part0 although that
1377 			 * partition will never be deleted, so we can treat
1378 			 * it as any other partition.
1379 			 */
1380 			part = &rq->rq_disk->part0;
1381 			hd_struct_get(part);
1382 		}
1383 		part_inc_in_flight(rq->q, part, rw);
1384 		rq->part = part;
1385 	}
1386 
1387 	update_io_ticks(part, jiffies);
1388 
1389 	part_stat_unlock();
1390 }
1391 
1392 /*
1393  * Steal bios from a request and add them to a bio list.
1394  * The request must not have been partially completed before.
1395  */
1396 void blk_steal_bios(struct bio_list *list, struct request *rq)
1397 {
1398 	if (rq->bio) {
1399 		if (list->tail)
1400 			list->tail->bi_next = rq->bio;
1401 		else
1402 			list->head = rq->bio;
1403 		list->tail = rq->biotail;
1404 
1405 		rq->bio = NULL;
1406 		rq->biotail = NULL;
1407 	}
1408 
1409 	rq->__data_len = 0;
1410 }
1411 EXPORT_SYMBOL_GPL(blk_steal_bios);
1412 
1413 /**
1414  * blk_update_request - Special helper function for request stacking drivers
1415  * @req:      the request being processed
1416  * @error:    block status code
1417  * @nr_bytes: number of bytes to complete @req
1418  *
1419  * Description:
1420  *     Ends I/O on a number of bytes attached to @req, but doesn't complete
1421  *     the request structure even if @req doesn't have leftover.
1422  *     If @req has leftover, sets it up for the next range of segments.
1423  *
1424  *     This special helper function is only for request stacking drivers
1425  *     (e.g. request-based dm) so that they can handle partial completion.
1426  *     Actual device drivers should use blk_end_request instead.
1427  *
1428  *     Passing the result of blk_rq_bytes() as @nr_bytes guarantees
1429  *     %false return from this function.
1430  *
1431  * Note:
1432  *	The RQF_SPECIAL_PAYLOAD flag is ignored on purpose in both
1433  *	blk_rq_bytes() and in blk_update_request().
1434  *
1435  * Return:
1436  *     %false - this request doesn't have any more data
1437  *     %true  - this request has more data
1438  **/
1439 bool blk_update_request(struct request *req, blk_status_t error,
1440 		unsigned int nr_bytes)
1441 {
1442 	int total_bytes;
1443 
1444 	trace_block_rq_complete(req, blk_status_to_errno(error), nr_bytes);
1445 
1446 	if (!req->bio)
1447 		return false;
1448 
1449 	if (unlikely(error && !blk_rq_is_passthrough(req) &&
1450 		     !(req->rq_flags & RQF_QUIET)))
1451 		print_req_error(req, error);
1452 
1453 	blk_account_io_completion(req, nr_bytes);
1454 
1455 	total_bytes = 0;
1456 	while (req->bio) {
1457 		struct bio *bio = req->bio;
1458 		unsigned bio_bytes = min(bio->bi_iter.bi_size, nr_bytes);
1459 
1460 		if (bio_bytes == bio->bi_iter.bi_size)
1461 			req->bio = bio->bi_next;
1462 
1463 		/* Completion has already been traced */
1464 		bio_clear_flag(bio, BIO_TRACE_COMPLETION);
1465 		req_bio_endio(req, bio, bio_bytes, error);
1466 
1467 		total_bytes += bio_bytes;
1468 		nr_bytes -= bio_bytes;
1469 
1470 		if (!nr_bytes)
1471 			break;
1472 	}
1473 
1474 	/*
1475 	 * completely done
1476 	 */
1477 	if (!req->bio) {
1478 		/*
1479 		 * Reset counters so that the request stacking driver
1480 		 * can find how many bytes remain in the request
1481 		 * later.
1482 		 */
1483 		req->__data_len = 0;
1484 		return false;
1485 	}
1486 
1487 	req->__data_len -= total_bytes;
1488 
1489 	/* update sector only for requests with clear definition of sector */
1490 	if (!blk_rq_is_passthrough(req))
1491 		req->__sector += total_bytes >> 9;
1492 
1493 	/* mixed attributes always follow the first bio */
1494 	if (req->rq_flags & RQF_MIXED_MERGE) {
1495 		req->cmd_flags &= ~REQ_FAILFAST_MASK;
1496 		req->cmd_flags |= req->bio->bi_opf & REQ_FAILFAST_MASK;
1497 	}
1498 
1499 	if (!(req->rq_flags & RQF_SPECIAL_PAYLOAD)) {
1500 		/*
1501 		 * If total number of sectors is less than the first segment
1502 		 * size, something has gone terribly wrong.
1503 		 */
1504 		if (blk_rq_bytes(req) < blk_rq_cur_bytes(req)) {
1505 			blk_dump_rq_flags(req, "request botched");
1506 			req->__data_len = blk_rq_cur_bytes(req);
1507 		}
1508 
1509 		/* recalculate the number of segments */
1510 		blk_recalc_rq_segments(req);
1511 	}
1512 
1513 	return true;
1514 }
1515 EXPORT_SYMBOL_GPL(blk_update_request);
1516 
1517 void blk_rq_bio_prep(struct request_queue *q, struct request *rq,
1518 		     struct bio *bio)
1519 {
1520 	if (bio_has_data(bio))
1521 		rq->nr_phys_segments = bio_phys_segments(q, bio);
1522 	else if (bio_op(bio) == REQ_OP_DISCARD)
1523 		rq->nr_phys_segments = 1;
1524 
1525 	rq->__data_len = bio->bi_iter.bi_size;
1526 	rq->bio = rq->biotail = bio;
1527 
1528 	if (bio->bi_disk)
1529 		rq->rq_disk = bio->bi_disk;
1530 }
1531 
1532 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
1533 /**
1534  * rq_flush_dcache_pages - Helper function to flush all pages in a request
1535  * @rq: the request to be flushed
1536  *
1537  * Description:
1538  *     Flush all pages in @rq.
1539  */
1540 void rq_flush_dcache_pages(struct request *rq)
1541 {
1542 	struct req_iterator iter;
1543 	struct bio_vec bvec;
1544 
1545 	rq_for_each_segment(bvec, rq, iter)
1546 		flush_dcache_page(bvec.bv_page);
1547 }
1548 EXPORT_SYMBOL_GPL(rq_flush_dcache_pages);
1549 #endif
1550 
1551 /**
1552  * blk_lld_busy - Check if underlying low-level drivers of a device are busy
1553  * @q : the queue of the device being checked
1554  *
1555  * Description:
1556  *    Check if underlying low-level drivers of a device are busy.
1557  *    If the drivers want to export their busy state, they must set own
1558  *    exporting function using blk_queue_lld_busy() first.
1559  *
1560  *    Basically, this function is used only by request stacking drivers
1561  *    to stop dispatching requests to underlying devices when underlying
1562  *    devices are busy.  This behavior helps more I/O merging on the queue
1563  *    of the request stacking driver and prevents I/O throughput regression
1564  *    on burst I/O load.
1565  *
1566  * Return:
1567  *    0 - Not busy (The request stacking driver should dispatch request)
1568  *    1 - Busy (The request stacking driver should stop dispatching request)
1569  */
1570 int blk_lld_busy(struct request_queue *q)
1571 {
1572 	if (queue_is_mq(q) && q->mq_ops->busy)
1573 		return q->mq_ops->busy(q);
1574 
1575 	return 0;
1576 }
1577 EXPORT_SYMBOL_GPL(blk_lld_busy);
1578 
1579 /**
1580  * blk_rq_unprep_clone - Helper function to free all bios in a cloned request
1581  * @rq: the clone request to be cleaned up
1582  *
1583  * Description:
1584  *     Free all bios in @rq for a cloned request.
1585  */
1586 void blk_rq_unprep_clone(struct request *rq)
1587 {
1588 	struct bio *bio;
1589 
1590 	while ((bio = rq->bio) != NULL) {
1591 		rq->bio = bio->bi_next;
1592 
1593 		bio_put(bio);
1594 	}
1595 }
1596 EXPORT_SYMBOL_GPL(blk_rq_unprep_clone);
1597 
1598 /*
1599  * Copy attributes of the original request to the clone request.
1600  * The actual data parts (e.g. ->cmd, ->sense) are not copied.
1601  */
1602 static void __blk_rq_prep_clone(struct request *dst, struct request *src)
1603 {
1604 	dst->__sector = blk_rq_pos(src);
1605 	dst->__data_len = blk_rq_bytes(src);
1606 	if (src->rq_flags & RQF_SPECIAL_PAYLOAD) {
1607 		dst->rq_flags |= RQF_SPECIAL_PAYLOAD;
1608 		dst->special_vec = src->special_vec;
1609 	}
1610 	dst->nr_phys_segments = src->nr_phys_segments;
1611 	dst->ioprio = src->ioprio;
1612 	dst->extra_len = src->extra_len;
1613 }
1614 
1615 /**
1616  * blk_rq_prep_clone - Helper function to setup clone request
1617  * @rq: the request to be setup
1618  * @rq_src: original request to be cloned
1619  * @bs: bio_set that bios for clone are allocated from
1620  * @gfp_mask: memory allocation mask for bio
1621  * @bio_ctr: setup function to be called for each clone bio.
1622  *           Returns %0 for success, non %0 for failure.
1623  * @data: private data to be passed to @bio_ctr
1624  *
1625  * Description:
1626  *     Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq.
1627  *     The actual data parts of @rq_src (e.g. ->cmd, ->sense)
1628  *     are not copied, and copying such parts is the caller's responsibility.
1629  *     Also, pages which the original bios are pointing to are not copied
1630  *     and the cloned bios just point same pages.
1631  *     So cloned bios must be completed before original bios, which means
1632  *     the caller must complete @rq before @rq_src.
1633  */
1634 int blk_rq_prep_clone(struct request *rq, struct request *rq_src,
1635 		      struct bio_set *bs, gfp_t gfp_mask,
1636 		      int (*bio_ctr)(struct bio *, struct bio *, void *),
1637 		      void *data)
1638 {
1639 	struct bio *bio, *bio_src;
1640 
1641 	if (!bs)
1642 		bs = &fs_bio_set;
1643 
1644 	__rq_for_each_bio(bio_src, rq_src) {
1645 		bio = bio_clone_fast(bio_src, gfp_mask, bs);
1646 		if (!bio)
1647 			goto free_and_out;
1648 
1649 		if (bio_ctr && bio_ctr(bio, bio_src, data))
1650 			goto free_and_out;
1651 
1652 		if (rq->bio) {
1653 			rq->biotail->bi_next = bio;
1654 			rq->biotail = bio;
1655 		} else
1656 			rq->bio = rq->biotail = bio;
1657 	}
1658 
1659 	__blk_rq_prep_clone(rq, rq_src);
1660 
1661 	return 0;
1662 
1663 free_and_out:
1664 	if (bio)
1665 		bio_put(bio);
1666 	blk_rq_unprep_clone(rq);
1667 
1668 	return -ENOMEM;
1669 }
1670 EXPORT_SYMBOL_GPL(blk_rq_prep_clone);
1671 
1672 int kblockd_schedule_work(struct work_struct *work)
1673 {
1674 	return queue_work(kblockd_workqueue, work);
1675 }
1676 EXPORT_SYMBOL(kblockd_schedule_work);
1677 
1678 int kblockd_schedule_work_on(int cpu, struct work_struct *work)
1679 {
1680 	return queue_work_on(cpu, kblockd_workqueue, work);
1681 }
1682 EXPORT_SYMBOL(kblockd_schedule_work_on);
1683 
1684 int kblockd_mod_delayed_work_on(int cpu, struct delayed_work *dwork,
1685 				unsigned long delay)
1686 {
1687 	return mod_delayed_work_on(cpu, kblockd_workqueue, dwork, delay);
1688 }
1689 EXPORT_SYMBOL(kblockd_mod_delayed_work_on);
1690 
1691 /**
1692  * blk_start_plug - initialize blk_plug and track it inside the task_struct
1693  * @plug:	The &struct blk_plug that needs to be initialized
1694  *
1695  * Description:
1696  *   blk_start_plug() indicates to the block layer an intent by the caller
1697  *   to submit multiple I/O requests in a batch.  The block layer may use
1698  *   this hint to defer submitting I/Os from the caller until blk_finish_plug()
1699  *   is called.  However, the block layer may choose to submit requests
1700  *   before a call to blk_finish_plug() if the number of queued I/Os
1701  *   exceeds %BLK_MAX_REQUEST_COUNT, or if the size of the I/O is larger than
1702  *   %BLK_PLUG_FLUSH_SIZE.  The queued I/Os may also be submitted early if
1703  *   the task schedules (see below).
1704  *
1705  *   Tracking blk_plug inside the task_struct will help with auto-flushing the
1706  *   pending I/O should the task end up blocking between blk_start_plug() and
1707  *   blk_finish_plug(). This is important from a performance perspective, but
1708  *   also ensures that we don't deadlock. For instance, if the task is blocking
1709  *   for a memory allocation, memory reclaim could end up wanting to free a
1710  *   page belonging to that request that is currently residing in our private
1711  *   plug. By flushing the pending I/O when the process goes to sleep, we avoid
1712  *   this kind of deadlock.
1713  */
1714 void blk_start_plug(struct blk_plug *plug)
1715 {
1716 	struct task_struct *tsk = current;
1717 
1718 	/*
1719 	 * If this is a nested plug, don't actually assign it.
1720 	 */
1721 	if (tsk->plug)
1722 		return;
1723 
1724 	INIT_LIST_HEAD(&plug->mq_list);
1725 	INIT_LIST_HEAD(&plug->cb_list);
1726 	plug->rq_count = 0;
1727 	plug->multiple_queues = false;
1728 
1729 	/*
1730 	 * Store ordering should not be needed here, since a potential
1731 	 * preempt will imply a full memory barrier
1732 	 */
1733 	tsk->plug = plug;
1734 }
1735 EXPORT_SYMBOL(blk_start_plug);
1736 
1737 static void flush_plug_callbacks(struct blk_plug *plug, bool from_schedule)
1738 {
1739 	LIST_HEAD(callbacks);
1740 
1741 	while (!list_empty(&plug->cb_list)) {
1742 		list_splice_init(&plug->cb_list, &callbacks);
1743 
1744 		while (!list_empty(&callbacks)) {
1745 			struct blk_plug_cb *cb = list_first_entry(&callbacks,
1746 							  struct blk_plug_cb,
1747 							  list);
1748 			list_del(&cb->list);
1749 			cb->callback(cb, from_schedule);
1750 		}
1751 	}
1752 }
1753 
1754 struct blk_plug_cb *blk_check_plugged(blk_plug_cb_fn unplug, void *data,
1755 				      int size)
1756 {
1757 	struct blk_plug *plug = current->plug;
1758 	struct blk_plug_cb *cb;
1759 
1760 	if (!plug)
1761 		return NULL;
1762 
1763 	list_for_each_entry(cb, &plug->cb_list, list)
1764 		if (cb->callback == unplug && cb->data == data)
1765 			return cb;
1766 
1767 	/* Not currently on the callback list */
1768 	BUG_ON(size < sizeof(*cb));
1769 	cb = kzalloc(size, GFP_ATOMIC);
1770 	if (cb) {
1771 		cb->data = data;
1772 		cb->callback = unplug;
1773 		list_add(&cb->list, &plug->cb_list);
1774 	}
1775 	return cb;
1776 }
1777 EXPORT_SYMBOL(blk_check_plugged);
1778 
1779 void blk_flush_plug_list(struct blk_plug *plug, bool from_schedule)
1780 {
1781 	flush_plug_callbacks(plug, from_schedule);
1782 
1783 	if (!list_empty(&plug->mq_list))
1784 		blk_mq_flush_plug_list(plug, from_schedule);
1785 }
1786 
1787 /**
1788  * blk_finish_plug - mark the end of a batch of submitted I/O
1789  * @plug:	The &struct blk_plug passed to blk_start_plug()
1790  *
1791  * Description:
1792  * Indicate that a batch of I/O submissions is complete.  This function
1793  * must be paired with an initial call to blk_start_plug().  The intent
1794  * is to allow the block layer to optimize I/O submission.  See the
1795  * documentation for blk_start_plug() for more information.
1796  */
1797 void blk_finish_plug(struct blk_plug *plug)
1798 {
1799 	if (plug != current->plug)
1800 		return;
1801 	blk_flush_plug_list(plug, false);
1802 
1803 	current->plug = NULL;
1804 }
1805 EXPORT_SYMBOL(blk_finish_plug);
1806 
1807 int __init blk_dev_init(void)
1808 {
1809 	BUILD_BUG_ON(REQ_OP_LAST >= (1 << REQ_OP_BITS));
1810 	BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 *
1811 			FIELD_SIZEOF(struct request, cmd_flags));
1812 	BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 *
1813 			FIELD_SIZEOF(struct bio, bi_opf));
1814 
1815 	/* used for unplugging and affects IO latency/throughput - HIGHPRI */
1816 	kblockd_workqueue = alloc_workqueue("kblockd",
1817 					    WQ_MEM_RECLAIM | WQ_HIGHPRI, 0);
1818 	if (!kblockd_workqueue)
1819 		panic("Failed to create kblockd\n");
1820 
1821 	blk_requestq_cachep = kmem_cache_create("request_queue",
1822 			sizeof(struct request_queue), 0, SLAB_PANIC, NULL);
1823 
1824 #ifdef CONFIG_DEBUG_FS
1825 	blk_debugfs_root = debugfs_create_dir("block", NULL);
1826 #endif
1827 
1828 	return 0;
1829 }
1830