xref: /openbmc/linux/drivers/md/dm.c (revision 8a10bc9d)
1 /*
2  * Copyright (C) 2001, 2002 Sistina Software (UK) Limited.
3  * Copyright (C) 2004-2008 Red Hat, Inc. All rights reserved.
4  *
5  * This file is released under the GPL.
6  */
7 
8 #include "dm.h"
9 #include "dm-uevent.h"
10 
11 #include <linux/init.h>
12 #include <linux/module.h>
13 #include <linux/mutex.h>
14 #include <linux/moduleparam.h>
15 #include <linux/blkpg.h>
16 #include <linux/bio.h>
17 #include <linux/mempool.h>
18 #include <linux/slab.h>
19 #include <linux/idr.h>
20 #include <linux/hdreg.h>
21 #include <linux/delay.h>
22 
23 #include <trace/events/block.h>
24 
25 #define DM_MSG_PREFIX "core"
26 
27 #ifdef CONFIG_PRINTK
28 /*
29  * ratelimit state to be used in DMXXX_LIMIT().
30  */
31 DEFINE_RATELIMIT_STATE(dm_ratelimit_state,
32 		       DEFAULT_RATELIMIT_INTERVAL,
33 		       DEFAULT_RATELIMIT_BURST);
34 EXPORT_SYMBOL(dm_ratelimit_state);
35 #endif
36 
37 /*
38  * Cookies are numeric values sent with CHANGE and REMOVE
39  * uevents while resuming, removing or renaming the device.
40  */
41 #define DM_COOKIE_ENV_VAR_NAME "DM_COOKIE"
42 #define DM_COOKIE_LENGTH 24
43 
44 static const char *_name = DM_NAME;
45 
46 static unsigned int major = 0;
47 static unsigned int _major = 0;
48 
49 static DEFINE_IDR(_minor_idr);
50 
51 static DEFINE_SPINLOCK(_minor_lock);
52 
53 static void do_deferred_remove(struct work_struct *w);
54 
55 static DECLARE_WORK(deferred_remove_work, do_deferred_remove);
56 
57 /*
58  * For bio-based dm.
59  * One of these is allocated per bio.
60  */
61 struct dm_io {
62 	struct mapped_device *md;
63 	int error;
64 	atomic_t io_count;
65 	struct bio *bio;
66 	unsigned long start_time;
67 	spinlock_t endio_lock;
68 	struct dm_stats_aux stats_aux;
69 };
70 
71 /*
72  * For request-based dm.
73  * One of these is allocated per request.
74  */
75 struct dm_rq_target_io {
76 	struct mapped_device *md;
77 	struct dm_target *ti;
78 	struct request *orig, clone;
79 	int error;
80 	union map_info info;
81 };
82 
83 /*
84  * For request-based dm - the bio clones we allocate are embedded in these
85  * structs.
86  *
87  * We allocate these with bio_alloc_bioset, using the front_pad parameter when
88  * the bioset is created - this means the bio has to come at the end of the
89  * struct.
90  */
91 struct dm_rq_clone_bio_info {
92 	struct bio *orig;
93 	struct dm_rq_target_io *tio;
94 	struct bio clone;
95 };
96 
97 union map_info *dm_get_mapinfo(struct bio *bio)
98 {
99 	if (bio && bio->bi_private)
100 		return &((struct dm_target_io *)bio->bi_private)->info;
101 	return NULL;
102 }
103 
104 union map_info *dm_get_rq_mapinfo(struct request *rq)
105 {
106 	if (rq && rq->end_io_data)
107 		return &((struct dm_rq_target_io *)rq->end_io_data)->info;
108 	return NULL;
109 }
110 EXPORT_SYMBOL_GPL(dm_get_rq_mapinfo);
111 
112 #define MINOR_ALLOCED ((void *)-1)
113 
114 /*
115  * Bits for the md->flags field.
116  */
117 #define DMF_BLOCK_IO_FOR_SUSPEND 0
118 #define DMF_SUSPENDED 1
119 #define DMF_FROZEN 2
120 #define DMF_FREEING 3
121 #define DMF_DELETING 4
122 #define DMF_NOFLUSH_SUSPENDING 5
123 #define DMF_MERGE_IS_OPTIONAL 6
124 #define DMF_DEFERRED_REMOVE 7
125 
126 /*
127  * A dummy definition to make RCU happy.
128  * struct dm_table should never be dereferenced in this file.
129  */
130 struct dm_table {
131 	int undefined__;
132 };
133 
134 /*
135  * Work processed by per-device workqueue.
136  */
137 struct mapped_device {
138 	struct srcu_struct io_barrier;
139 	struct mutex suspend_lock;
140 	atomic_t holders;
141 	atomic_t open_count;
142 
143 	/*
144 	 * The current mapping.
145 	 * Use dm_get_live_table{_fast} or take suspend_lock for
146 	 * dereference.
147 	 */
148 	struct dm_table *map;
149 
150 	unsigned long flags;
151 
152 	struct request_queue *queue;
153 	unsigned type;
154 	/* Protect queue and type against concurrent access. */
155 	struct mutex type_lock;
156 
157 	struct target_type *immutable_target_type;
158 
159 	struct gendisk *disk;
160 	char name[16];
161 
162 	void *interface_ptr;
163 
164 	/*
165 	 * A list of ios that arrived while we were suspended.
166 	 */
167 	atomic_t pending[2];
168 	wait_queue_head_t wait;
169 	struct work_struct work;
170 	struct bio_list deferred;
171 	spinlock_t deferred_lock;
172 
173 	/*
174 	 * Processing queue (flush)
175 	 */
176 	struct workqueue_struct *wq;
177 
178 	/*
179 	 * io objects are allocated from here.
180 	 */
181 	mempool_t *io_pool;
182 
183 	struct bio_set *bs;
184 
185 	/*
186 	 * Event handling.
187 	 */
188 	atomic_t event_nr;
189 	wait_queue_head_t eventq;
190 	atomic_t uevent_seq;
191 	struct list_head uevent_list;
192 	spinlock_t uevent_lock; /* Protect access to uevent_list */
193 
194 	/*
195 	 * freeze/thaw support require holding onto a super block
196 	 */
197 	struct super_block *frozen_sb;
198 	struct block_device *bdev;
199 
200 	/* forced geometry settings */
201 	struct hd_geometry geometry;
202 
203 	/* kobject and completion */
204 	struct dm_kobject_holder kobj_holder;
205 
206 	/* zero-length flush that will be cloned and submitted to targets */
207 	struct bio flush_bio;
208 
209 	struct dm_stats stats;
210 };
211 
212 /*
213  * For mempools pre-allocation at the table loading time.
214  */
215 struct dm_md_mempools {
216 	mempool_t *io_pool;
217 	struct bio_set *bs;
218 };
219 
220 #define RESERVED_BIO_BASED_IOS		16
221 #define RESERVED_REQUEST_BASED_IOS	256
222 #define RESERVED_MAX_IOS		1024
223 static struct kmem_cache *_io_cache;
224 static struct kmem_cache *_rq_tio_cache;
225 
226 /*
227  * Bio-based DM's mempools' reserved IOs set by the user.
228  */
229 static unsigned reserved_bio_based_ios = RESERVED_BIO_BASED_IOS;
230 
231 /*
232  * Request-based DM's mempools' reserved IOs set by the user.
233  */
234 static unsigned reserved_rq_based_ios = RESERVED_REQUEST_BASED_IOS;
235 
236 static unsigned __dm_get_reserved_ios(unsigned *reserved_ios,
237 				      unsigned def, unsigned max)
238 {
239 	unsigned ios = ACCESS_ONCE(*reserved_ios);
240 	unsigned modified_ios = 0;
241 
242 	if (!ios)
243 		modified_ios = def;
244 	else if (ios > max)
245 		modified_ios = max;
246 
247 	if (modified_ios) {
248 		(void)cmpxchg(reserved_ios, ios, modified_ios);
249 		ios = modified_ios;
250 	}
251 
252 	return ios;
253 }
254 
255 unsigned dm_get_reserved_bio_based_ios(void)
256 {
257 	return __dm_get_reserved_ios(&reserved_bio_based_ios,
258 				     RESERVED_BIO_BASED_IOS, RESERVED_MAX_IOS);
259 }
260 EXPORT_SYMBOL_GPL(dm_get_reserved_bio_based_ios);
261 
262 unsigned dm_get_reserved_rq_based_ios(void)
263 {
264 	return __dm_get_reserved_ios(&reserved_rq_based_ios,
265 				     RESERVED_REQUEST_BASED_IOS, RESERVED_MAX_IOS);
266 }
267 EXPORT_SYMBOL_GPL(dm_get_reserved_rq_based_ios);
268 
269 static int __init local_init(void)
270 {
271 	int r = -ENOMEM;
272 
273 	/* allocate a slab for the dm_ios */
274 	_io_cache = KMEM_CACHE(dm_io, 0);
275 	if (!_io_cache)
276 		return r;
277 
278 	_rq_tio_cache = KMEM_CACHE(dm_rq_target_io, 0);
279 	if (!_rq_tio_cache)
280 		goto out_free_io_cache;
281 
282 	r = dm_uevent_init();
283 	if (r)
284 		goto out_free_rq_tio_cache;
285 
286 	_major = major;
287 	r = register_blkdev(_major, _name);
288 	if (r < 0)
289 		goto out_uevent_exit;
290 
291 	if (!_major)
292 		_major = r;
293 
294 	return 0;
295 
296 out_uevent_exit:
297 	dm_uevent_exit();
298 out_free_rq_tio_cache:
299 	kmem_cache_destroy(_rq_tio_cache);
300 out_free_io_cache:
301 	kmem_cache_destroy(_io_cache);
302 
303 	return r;
304 }
305 
306 static void local_exit(void)
307 {
308 	flush_scheduled_work();
309 
310 	kmem_cache_destroy(_rq_tio_cache);
311 	kmem_cache_destroy(_io_cache);
312 	unregister_blkdev(_major, _name);
313 	dm_uevent_exit();
314 
315 	_major = 0;
316 
317 	DMINFO("cleaned up");
318 }
319 
320 static int (*_inits[])(void) __initdata = {
321 	local_init,
322 	dm_target_init,
323 	dm_linear_init,
324 	dm_stripe_init,
325 	dm_io_init,
326 	dm_kcopyd_init,
327 	dm_interface_init,
328 	dm_statistics_init,
329 };
330 
331 static void (*_exits[])(void) = {
332 	local_exit,
333 	dm_target_exit,
334 	dm_linear_exit,
335 	dm_stripe_exit,
336 	dm_io_exit,
337 	dm_kcopyd_exit,
338 	dm_interface_exit,
339 	dm_statistics_exit,
340 };
341 
342 static int __init dm_init(void)
343 {
344 	const int count = ARRAY_SIZE(_inits);
345 
346 	int r, i;
347 
348 	for (i = 0; i < count; i++) {
349 		r = _inits[i]();
350 		if (r)
351 			goto bad;
352 	}
353 
354 	return 0;
355 
356       bad:
357 	while (i--)
358 		_exits[i]();
359 
360 	return r;
361 }
362 
363 static void __exit dm_exit(void)
364 {
365 	int i = ARRAY_SIZE(_exits);
366 
367 	while (i--)
368 		_exits[i]();
369 
370 	/*
371 	 * Should be empty by this point.
372 	 */
373 	idr_destroy(&_minor_idr);
374 }
375 
376 /*
377  * Block device functions
378  */
379 int dm_deleting_md(struct mapped_device *md)
380 {
381 	return test_bit(DMF_DELETING, &md->flags);
382 }
383 
384 static int dm_blk_open(struct block_device *bdev, fmode_t mode)
385 {
386 	struct mapped_device *md;
387 
388 	spin_lock(&_minor_lock);
389 
390 	md = bdev->bd_disk->private_data;
391 	if (!md)
392 		goto out;
393 
394 	if (test_bit(DMF_FREEING, &md->flags) ||
395 	    dm_deleting_md(md)) {
396 		md = NULL;
397 		goto out;
398 	}
399 
400 	dm_get(md);
401 	atomic_inc(&md->open_count);
402 
403 out:
404 	spin_unlock(&_minor_lock);
405 
406 	return md ? 0 : -ENXIO;
407 }
408 
409 static void dm_blk_close(struct gendisk *disk, fmode_t mode)
410 {
411 	struct mapped_device *md = disk->private_data;
412 
413 	spin_lock(&_minor_lock);
414 
415 	if (atomic_dec_and_test(&md->open_count) &&
416 	    (test_bit(DMF_DEFERRED_REMOVE, &md->flags)))
417 		schedule_work(&deferred_remove_work);
418 
419 	dm_put(md);
420 
421 	spin_unlock(&_minor_lock);
422 }
423 
424 int dm_open_count(struct mapped_device *md)
425 {
426 	return atomic_read(&md->open_count);
427 }
428 
429 /*
430  * Guarantees nothing is using the device before it's deleted.
431  */
432 int dm_lock_for_deletion(struct mapped_device *md, bool mark_deferred, bool only_deferred)
433 {
434 	int r = 0;
435 
436 	spin_lock(&_minor_lock);
437 
438 	if (dm_open_count(md)) {
439 		r = -EBUSY;
440 		if (mark_deferred)
441 			set_bit(DMF_DEFERRED_REMOVE, &md->flags);
442 	} else if (only_deferred && !test_bit(DMF_DEFERRED_REMOVE, &md->flags))
443 		r = -EEXIST;
444 	else
445 		set_bit(DMF_DELETING, &md->flags);
446 
447 	spin_unlock(&_minor_lock);
448 
449 	return r;
450 }
451 
452 int dm_cancel_deferred_remove(struct mapped_device *md)
453 {
454 	int r = 0;
455 
456 	spin_lock(&_minor_lock);
457 
458 	if (test_bit(DMF_DELETING, &md->flags))
459 		r = -EBUSY;
460 	else
461 		clear_bit(DMF_DEFERRED_REMOVE, &md->flags);
462 
463 	spin_unlock(&_minor_lock);
464 
465 	return r;
466 }
467 
468 static void do_deferred_remove(struct work_struct *w)
469 {
470 	dm_deferred_remove();
471 }
472 
473 sector_t dm_get_size(struct mapped_device *md)
474 {
475 	return get_capacity(md->disk);
476 }
477 
478 struct dm_stats *dm_get_stats(struct mapped_device *md)
479 {
480 	return &md->stats;
481 }
482 
483 static int dm_blk_getgeo(struct block_device *bdev, struct hd_geometry *geo)
484 {
485 	struct mapped_device *md = bdev->bd_disk->private_data;
486 
487 	return dm_get_geometry(md, geo);
488 }
489 
490 static int dm_blk_ioctl(struct block_device *bdev, fmode_t mode,
491 			unsigned int cmd, unsigned long arg)
492 {
493 	struct mapped_device *md = bdev->bd_disk->private_data;
494 	int srcu_idx;
495 	struct dm_table *map;
496 	struct dm_target *tgt;
497 	int r = -ENOTTY;
498 
499 retry:
500 	map = dm_get_live_table(md, &srcu_idx);
501 
502 	if (!map || !dm_table_get_size(map))
503 		goto out;
504 
505 	/* We only support devices that have a single target */
506 	if (dm_table_get_num_targets(map) != 1)
507 		goto out;
508 
509 	tgt = dm_table_get_target(map, 0);
510 
511 	if (dm_suspended_md(md)) {
512 		r = -EAGAIN;
513 		goto out;
514 	}
515 
516 	if (tgt->type->ioctl)
517 		r = tgt->type->ioctl(tgt, cmd, arg);
518 
519 out:
520 	dm_put_live_table(md, srcu_idx);
521 
522 	if (r == -ENOTCONN) {
523 		msleep(10);
524 		goto retry;
525 	}
526 
527 	return r;
528 }
529 
530 static struct dm_io *alloc_io(struct mapped_device *md)
531 {
532 	return mempool_alloc(md->io_pool, GFP_NOIO);
533 }
534 
535 static void free_io(struct mapped_device *md, struct dm_io *io)
536 {
537 	mempool_free(io, md->io_pool);
538 }
539 
540 static void free_tio(struct mapped_device *md, struct dm_target_io *tio)
541 {
542 	bio_put(&tio->clone);
543 }
544 
545 static struct dm_rq_target_io *alloc_rq_tio(struct mapped_device *md,
546 					    gfp_t gfp_mask)
547 {
548 	return mempool_alloc(md->io_pool, gfp_mask);
549 }
550 
551 static void free_rq_tio(struct dm_rq_target_io *tio)
552 {
553 	mempool_free(tio, tio->md->io_pool);
554 }
555 
556 static int md_in_flight(struct mapped_device *md)
557 {
558 	return atomic_read(&md->pending[READ]) +
559 	       atomic_read(&md->pending[WRITE]);
560 }
561 
562 static void start_io_acct(struct dm_io *io)
563 {
564 	struct mapped_device *md = io->md;
565 	struct bio *bio = io->bio;
566 	int cpu;
567 	int rw = bio_data_dir(bio);
568 
569 	io->start_time = jiffies;
570 
571 	cpu = part_stat_lock();
572 	part_round_stats(cpu, &dm_disk(md)->part0);
573 	part_stat_unlock();
574 	atomic_set(&dm_disk(md)->part0.in_flight[rw],
575 		atomic_inc_return(&md->pending[rw]));
576 
577 	if (unlikely(dm_stats_used(&md->stats)))
578 		dm_stats_account_io(&md->stats, bio->bi_rw, bio->bi_iter.bi_sector,
579 				    bio_sectors(bio), false, 0, &io->stats_aux);
580 }
581 
582 static void end_io_acct(struct dm_io *io)
583 {
584 	struct mapped_device *md = io->md;
585 	struct bio *bio = io->bio;
586 	unsigned long duration = jiffies - io->start_time;
587 	int pending, cpu;
588 	int rw = bio_data_dir(bio);
589 
590 	cpu = part_stat_lock();
591 	part_round_stats(cpu, &dm_disk(md)->part0);
592 	part_stat_add(cpu, &dm_disk(md)->part0, ticks[rw], duration);
593 	part_stat_unlock();
594 
595 	if (unlikely(dm_stats_used(&md->stats)))
596 		dm_stats_account_io(&md->stats, bio->bi_rw, bio->bi_iter.bi_sector,
597 				    bio_sectors(bio), true, duration, &io->stats_aux);
598 
599 	/*
600 	 * After this is decremented the bio must not be touched if it is
601 	 * a flush.
602 	 */
603 	pending = atomic_dec_return(&md->pending[rw]);
604 	atomic_set(&dm_disk(md)->part0.in_flight[rw], pending);
605 	pending += atomic_read(&md->pending[rw^0x1]);
606 
607 	/* nudge anyone waiting on suspend queue */
608 	if (!pending)
609 		wake_up(&md->wait);
610 }
611 
612 /*
613  * Add the bio to the list of deferred io.
614  */
615 static void queue_io(struct mapped_device *md, struct bio *bio)
616 {
617 	unsigned long flags;
618 
619 	spin_lock_irqsave(&md->deferred_lock, flags);
620 	bio_list_add(&md->deferred, bio);
621 	spin_unlock_irqrestore(&md->deferred_lock, flags);
622 	queue_work(md->wq, &md->work);
623 }
624 
625 /*
626  * Everyone (including functions in this file), should use this
627  * function to access the md->map field, and make sure they call
628  * dm_put_live_table() when finished.
629  */
630 struct dm_table *dm_get_live_table(struct mapped_device *md, int *srcu_idx) __acquires(md->io_barrier)
631 {
632 	*srcu_idx = srcu_read_lock(&md->io_barrier);
633 
634 	return srcu_dereference(md->map, &md->io_barrier);
635 }
636 
637 void dm_put_live_table(struct mapped_device *md, int srcu_idx) __releases(md->io_barrier)
638 {
639 	srcu_read_unlock(&md->io_barrier, srcu_idx);
640 }
641 
642 void dm_sync_table(struct mapped_device *md)
643 {
644 	synchronize_srcu(&md->io_barrier);
645 	synchronize_rcu_expedited();
646 }
647 
648 /*
649  * A fast alternative to dm_get_live_table/dm_put_live_table.
650  * The caller must not block between these two functions.
651  */
652 static struct dm_table *dm_get_live_table_fast(struct mapped_device *md) __acquires(RCU)
653 {
654 	rcu_read_lock();
655 	return rcu_dereference(md->map);
656 }
657 
658 static void dm_put_live_table_fast(struct mapped_device *md) __releases(RCU)
659 {
660 	rcu_read_unlock();
661 }
662 
663 /*
664  * Get the geometry associated with a dm device
665  */
666 int dm_get_geometry(struct mapped_device *md, struct hd_geometry *geo)
667 {
668 	*geo = md->geometry;
669 
670 	return 0;
671 }
672 
673 /*
674  * Set the geometry of a device.
675  */
676 int dm_set_geometry(struct mapped_device *md, struct hd_geometry *geo)
677 {
678 	sector_t sz = (sector_t)geo->cylinders * geo->heads * geo->sectors;
679 
680 	if (geo->start > sz) {
681 		DMWARN("Start sector is beyond the geometry limits.");
682 		return -EINVAL;
683 	}
684 
685 	md->geometry = *geo;
686 
687 	return 0;
688 }
689 
690 /*-----------------------------------------------------------------
691  * CRUD START:
692  *   A more elegant soln is in the works that uses the queue
693  *   merge fn, unfortunately there are a couple of changes to
694  *   the block layer that I want to make for this.  So in the
695  *   interests of getting something for people to use I give
696  *   you this clearly demarcated crap.
697  *---------------------------------------------------------------*/
698 
699 static int __noflush_suspending(struct mapped_device *md)
700 {
701 	return test_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
702 }
703 
704 /*
705  * Decrements the number of outstanding ios that a bio has been
706  * cloned into, completing the original io if necc.
707  */
708 static void dec_pending(struct dm_io *io, int error)
709 {
710 	unsigned long flags;
711 	int io_error;
712 	struct bio *bio;
713 	struct mapped_device *md = io->md;
714 
715 	/* Push-back supersedes any I/O errors */
716 	if (unlikely(error)) {
717 		spin_lock_irqsave(&io->endio_lock, flags);
718 		if (!(io->error > 0 && __noflush_suspending(md)))
719 			io->error = error;
720 		spin_unlock_irqrestore(&io->endio_lock, flags);
721 	}
722 
723 	if (atomic_dec_and_test(&io->io_count)) {
724 		if (io->error == DM_ENDIO_REQUEUE) {
725 			/*
726 			 * Target requested pushing back the I/O.
727 			 */
728 			spin_lock_irqsave(&md->deferred_lock, flags);
729 			if (__noflush_suspending(md))
730 				bio_list_add_head(&md->deferred, io->bio);
731 			else
732 				/* noflush suspend was interrupted. */
733 				io->error = -EIO;
734 			spin_unlock_irqrestore(&md->deferred_lock, flags);
735 		}
736 
737 		io_error = io->error;
738 		bio = io->bio;
739 		end_io_acct(io);
740 		free_io(md, io);
741 
742 		if (io_error == DM_ENDIO_REQUEUE)
743 			return;
744 
745 		if ((bio->bi_rw & REQ_FLUSH) && bio->bi_iter.bi_size) {
746 			/*
747 			 * Preflush done for flush with data, reissue
748 			 * without REQ_FLUSH.
749 			 */
750 			bio->bi_rw &= ~REQ_FLUSH;
751 			queue_io(md, bio);
752 		} else {
753 			/* done with normal IO or empty flush */
754 			trace_block_bio_complete(md->queue, bio, io_error);
755 			bio_endio(bio, io_error);
756 		}
757 	}
758 }
759 
760 static void clone_endio(struct bio *bio, int error)
761 {
762 	int r = 0;
763 	struct dm_target_io *tio = bio->bi_private;
764 	struct dm_io *io = tio->io;
765 	struct mapped_device *md = tio->io->md;
766 	dm_endio_fn endio = tio->ti->type->end_io;
767 
768 	if (!bio_flagged(bio, BIO_UPTODATE) && !error)
769 		error = -EIO;
770 
771 	if (endio) {
772 		r = endio(tio->ti, bio, error);
773 		if (r < 0 || r == DM_ENDIO_REQUEUE)
774 			/*
775 			 * error and requeue request are handled
776 			 * in dec_pending().
777 			 */
778 			error = r;
779 		else if (r == DM_ENDIO_INCOMPLETE)
780 			/* The target will handle the io */
781 			return;
782 		else if (r) {
783 			DMWARN("unimplemented target endio return value: %d", r);
784 			BUG();
785 		}
786 	}
787 
788 	free_tio(md, tio);
789 	dec_pending(io, error);
790 }
791 
792 /*
793  * Partial completion handling for request-based dm
794  */
795 static void end_clone_bio(struct bio *clone, int error)
796 {
797 	struct dm_rq_clone_bio_info *info = clone->bi_private;
798 	struct dm_rq_target_io *tio = info->tio;
799 	struct bio *bio = info->orig;
800 	unsigned int nr_bytes = info->orig->bi_iter.bi_size;
801 
802 	bio_put(clone);
803 
804 	if (tio->error)
805 		/*
806 		 * An error has already been detected on the request.
807 		 * Once error occurred, just let clone->end_io() handle
808 		 * the remainder.
809 		 */
810 		return;
811 	else if (error) {
812 		/*
813 		 * Don't notice the error to the upper layer yet.
814 		 * The error handling decision is made by the target driver,
815 		 * when the request is completed.
816 		 */
817 		tio->error = error;
818 		return;
819 	}
820 
821 	/*
822 	 * I/O for the bio successfully completed.
823 	 * Notice the data completion to the upper layer.
824 	 */
825 
826 	/*
827 	 * bios are processed from the head of the list.
828 	 * So the completing bio should always be rq->bio.
829 	 * If it's not, something wrong is happening.
830 	 */
831 	if (tio->orig->bio != bio)
832 		DMERR("bio completion is going in the middle of the request");
833 
834 	/*
835 	 * Update the original request.
836 	 * Do not use blk_end_request() here, because it may complete
837 	 * the original request before the clone, and break the ordering.
838 	 */
839 	blk_update_request(tio->orig, 0, nr_bytes);
840 }
841 
842 /*
843  * Don't touch any member of the md after calling this function because
844  * the md may be freed in dm_put() at the end of this function.
845  * Or do dm_get() before calling this function and dm_put() later.
846  */
847 static void rq_completed(struct mapped_device *md, int rw, int run_queue)
848 {
849 	atomic_dec(&md->pending[rw]);
850 
851 	/* nudge anyone waiting on suspend queue */
852 	if (!md_in_flight(md))
853 		wake_up(&md->wait);
854 
855 	/*
856 	 * Run this off this callpath, as drivers could invoke end_io while
857 	 * inside their request_fn (and holding the queue lock). Calling
858 	 * back into ->request_fn() could deadlock attempting to grab the
859 	 * queue lock again.
860 	 */
861 	if (run_queue)
862 		blk_run_queue_async(md->queue);
863 
864 	/*
865 	 * dm_put() must be at the end of this function. See the comment above
866 	 */
867 	dm_put(md);
868 }
869 
870 static void free_rq_clone(struct request *clone)
871 {
872 	struct dm_rq_target_io *tio = clone->end_io_data;
873 
874 	blk_rq_unprep_clone(clone);
875 	free_rq_tio(tio);
876 }
877 
878 /*
879  * Complete the clone and the original request.
880  * Must be called without queue lock.
881  */
882 static void dm_end_request(struct request *clone, int error)
883 {
884 	int rw = rq_data_dir(clone);
885 	struct dm_rq_target_io *tio = clone->end_io_data;
886 	struct mapped_device *md = tio->md;
887 	struct request *rq = tio->orig;
888 
889 	if (rq->cmd_type == REQ_TYPE_BLOCK_PC) {
890 		rq->errors = clone->errors;
891 		rq->resid_len = clone->resid_len;
892 
893 		if (rq->sense)
894 			/*
895 			 * We are using the sense buffer of the original
896 			 * request.
897 			 * So setting the length of the sense data is enough.
898 			 */
899 			rq->sense_len = clone->sense_len;
900 	}
901 
902 	free_rq_clone(clone);
903 	blk_end_request_all(rq, error);
904 	rq_completed(md, rw, true);
905 }
906 
907 static void dm_unprep_request(struct request *rq)
908 {
909 	struct request *clone = rq->special;
910 
911 	rq->special = NULL;
912 	rq->cmd_flags &= ~REQ_DONTPREP;
913 
914 	free_rq_clone(clone);
915 }
916 
917 /*
918  * Requeue the original request of a clone.
919  */
920 void dm_requeue_unmapped_request(struct request *clone)
921 {
922 	int rw = rq_data_dir(clone);
923 	struct dm_rq_target_io *tio = clone->end_io_data;
924 	struct mapped_device *md = tio->md;
925 	struct request *rq = tio->orig;
926 	struct request_queue *q = rq->q;
927 	unsigned long flags;
928 
929 	dm_unprep_request(rq);
930 
931 	spin_lock_irqsave(q->queue_lock, flags);
932 	blk_requeue_request(q, rq);
933 	spin_unlock_irqrestore(q->queue_lock, flags);
934 
935 	rq_completed(md, rw, 0);
936 }
937 EXPORT_SYMBOL_GPL(dm_requeue_unmapped_request);
938 
939 static void __stop_queue(struct request_queue *q)
940 {
941 	blk_stop_queue(q);
942 }
943 
944 static void stop_queue(struct request_queue *q)
945 {
946 	unsigned long flags;
947 
948 	spin_lock_irqsave(q->queue_lock, flags);
949 	__stop_queue(q);
950 	spin_unlock_irqrestore(q->queue_lock, flags);
951 }
952 
953 static void __start_queue(struct request_queue *q)
954 {
955 	if (blk_queue_stopped(q))
956 		blk_start_queue(q);
957 }
958 
959 static void start_queue(struct request_queue *q)
960 {
961 	unsigned long flags;
962 
963 	spin_lock_irqsave(q->queue_lock, flags);
964 	__start_queue(q);
965 	spin_unlock_irqrestore(q->queue_lock, flags);
966 }
967 
968 static void dm_done(struct request *clone, int error, bool mapped)
969 {
970 	int r = error;
971 	struct dm_rq_target_io *tio = clone->end_io_data;
972 	dm_request_endio_fn rq_end_io = NULL;
973 
974 	if (tio->ti) {
975 		rq_end_io = tio->ti->type->rq_end_io;
976 
977 		if (mapped && rq_end_io)
978 			r = rq_end_io(tio->ti, clone, error, &tio->info);
979 	}
980 
981 	if (r <= 0)
982 		/* The target wants to complete the I/O */
983 		dm_end_request(clone, r);
984 	else if (r == DM_ENDIO_INCOMPLETE)
985 		/* The target will handle the I/O */
986 		return;
987 	else if (r == DM_ENDIO_REQUEUE)
988 		/* The target wants to requeue the I/O */
989 		dm_requeue_unmapped_request(clone);
990 	else {
991 		DMWARN("unimplemented target endio return value: %d", r);
992 		BUG();
993 	}
994 }
995 
996 /*
997  * Request completion handler for request-based dm
998  */
999 static void dm_softirq_done(struct request *rq)
1000 {
1001 	bool mapped = true;
1002 	struct request *clone = rq->completion_data;
1003 	struct dm_rq_target_io *tio = clone->end_io_data;
1004 
1005 	if (rq->cmd_flags & REQ_FAILED)
1006 		mapped = false;
1007 
1008 	dm_done(clone, tio->error, mapped);
1009 }
1010 
1011 /*
1012  * Complete the clone and the original request with the error status
1013  * through softirq context.
1014  */
1015 static void dm_complete_request(struct request *clone, int error)
1016 {
1017 	struct dm_rq_target_io *tio = clone->end_io_data;
1018 	struct request *rq = tio->orig;
1019 
1020 	tio->error = error;
1021 	rq->completion_data = clone;
1022 	blk_complete_request(rq);
1023 }
1024 
1025 /*
1026  * Complete the not-mapped clone and the original request with the error status
1027  * through softirq context.
1028  * Target's rq_end_io() function isn't called.
1029  * This may be used when the target's map_rq() function fails.
1030  */
1031 void dm_kill_unmapped_request(struct request *clone, int error)
1032 {
1033 	struct dm_rq_target_io *tio = clone->end_io_data;
1034 	struct request *rq = tio->orig;
1035 
1036 	rq->cmd_flags |= REQ_FAILED;
1037 	dm_complete_request(clone, error);
1038 }
1039 EXPORT_SYMBOL_GPL(dm_kill_unmapped_request);
1040 
1041 /*
1042  * Called with the queue lock held
1043  */
1044 static void end_clone_request(struct request *clone, int error)
1045 {
1046 	/*
1047 	 * For just cleaning up the information of the queue in which
1048 	 * the clone was dispatched.
1049 	 * The clone is *NOT* freed actually here because it is alloced from
1050 	 * dm own mempool and REQ_ALLOCED isn't set in clone->cmd_flags.
1051 	 */
1052 	__blk_put_request(clone->q, clone);
1053 
1054 	/*
1055 	 * Actual request completion is done in a softirq context which doesn't
1056 	 * hold the queue lock.  Otherwise, deadlock could occur because:
1057 	 *     - another request may be submitted by the upper level driver
1058 	 *       of the stacking during the completion
1059 	 *     - the submission which requires queue lock may be done
1060 	 *       against this queue
1061 	 */
1062 	dm_complete_request(clone, error);
1063 }
1064 
1065 /*
1066  * Return maximum size of I/O possible at the supplied sector up to the current
1067  * target boundary.
1068  */
1069 static sector_t max_io_len_target_boundary(sector_t sector, struct dm_target *ti)
1070 {
1071 	sector_t target_offset = dm_target_offset(ti, sector);
1072 
1073 	return ti->len - target_offset;
1074 }
1075 
1076 static sector_t max_io_len(sector_t sector, struct dm_target *ti)
1077 {
1078 	sector_t len = max_io_len_target_boundary(sector, ti);
1079 	sector_t offset, max_len;
1080 
1081 	/*
1082 	 * Does the target need to split even further?
1083 	 */
1084 	if (ti->max_io_len) {
1085 		offset = dm_target_offset(ti, sector);
1086 		if (unlikely(ti->max_io_len & (ti->max_io_len - 1)))
1087 			max_len = sector_div(offset, ti->max_io_len);
1088 		else
1089 			max_len = offset & (ti->max_io_len - 1);
1090 		max_len = ti->max_io_len - max_len;
1091 
1092 		if (len > max_len)
1093 			len = max_len;
1094 	}
1095 
1096 	return len;
1097 }
1098 
1099 int dm_set_target_max_io_len(struct dm_target *ti, sector_t len)
1100 {
1101 	if (len > UINT_MAX) {
1102 		DMERR("Specified maximum size of target IO (%llu) exceeds limit (%u)",
1103 		      (unsigned long long)len, UINT_MAX);
1104 		ti->error = "Maximum size of target IO is too large";
1105 		return -EINVAL;
1106 	}
1107 
1108 	ti->max_io_len = (uint32_t) len;
1109 
1110 	return 0;
1111 }
1112 EXPORT_SYMBOL_GPL(dm_set_target_max_io_len);
1113 
1114 static void __map_bio(struct dm_target_io *tio)
1115 {
1116 	int r;
1117 	sector_t sector;
1118 	struct mapped_device *md;
1119 	struct bio *clone = &tio->clone;
1120 	struct dm_target *ti = tio->ti;
1121 
1122 	clone->bi_end_io = clone_endio;
1123 	clone->bi_private = tio;
1124 
1125 	/*
1126 	 * Map the clone.  If r == 0 we don't need to do
1127 	 * anything, the target has assumed ownership of
1128 	 * this io.
1129 	 */
1130 	atomic_inc(&tio->io->io_count);
1131 	sector = clone->bi_iter.bi_sector;
1132 	r = ti->type->map(ti, clone);
1133 	if (r == DM_MAPIO_REMAPPED) {
1134 		/* the bio has been remapped so dispatch it */
1135 
1136 		trace_block_bio_remap(bdev_get_queue(clone->bi_bdev), clone,
1137 				      tio->io->bio->bi_bdev->bd_dev, sector);
1138 
1139 		generic_make_request(clone);
1140 	} else if (r < 0 || r == DM_MAPIO_REQUEUE) {
1141 		/* error the io and bail out, or requeue it if needed */
1142 		md = tio->io->md;
1143 		dec_pending(tio->io, r);
1144 		free_tio(md, tio);
1145 	} else if (r) {
1146 		DMWARN("unimplemented target map return value: %d", r);
1147 		BUG();
1148 	}
1149 }
1150 
1151 struct clone_info {
1152 	struct mapped_device *md;
1153 	struct dm_table *map;
1154 	struct bio *bio;
1155 	struct dm_io *io;
1156 	sector_t sector;
1157 	sector_t sector_count;
1158 };
1159 
1160 static void bio_setup_sector(struct bio *bio, sector_t sector, sector_t len)
1161 {
1162 	bio->bi_iter.bi_sector = sector;
1163 	bio->bi_iter.bi_size = to_bytes(len);
1164 }
1165 
1166 /*
1167  * Creates a bio that consists of range of complete bvecs.
1168  */
1169 static void clone_bio(struct dm_target_io *tio, struct bio *bio,
1170 		      sector_t sector, unsigned len)
1171 {
1172 	struct bio *clone = &tio->clone;
1173 
1174 	__bio_clone_fast(clone, bio);
1175 
1176 	if (bio_integrity(bio))
1177 		bio_integrity_clone(clone, bio, GFP_NOIO);
1178 
1179 	bio_advance(clone, to_bytes(sector - clone->bi_iter.bi_sector));
1180 	clone->bi_iter.bi_size = to_bytes(len);
1181 
1182 	if (bio_integrity(bio))
1183 		bio_integrity_trim(clone, 0, len);
1184 }
1185 
1186 static struct dm_target_io *alloc_tio(struct clone_info *ci,
1187 				      struct dm_target *ti, int nr_iovecs,
1188 				      unsigned target_bio_nr)
1189 {
1190 	struct dm_target_io *tio;
1191 	struct bio *clone;
1192 
1193 	clone = bio_alloc_bioset(GFP_NOIO, nr_iovecs, ci->md->bs);
1194 	tio = container_of(clone, struct dm_target_io, clone);
1195 
1196 	tio->io = ci->io;
1197 	tio->ti = ti;
1198 	memset(&tio->info, 0, sizeof(tio->info));
1199 	tio->target_bio_nr = target_bio_nr;
1200 
1201 	return tio;
1202 }
1203 
1204 static void __clone_and_map_simple_bio(struct clone_info *ci,
1205 				       struct dm_target *ti,
1206 				       unsigned target_bio_nr, sector_t len)
1207 {
1208 	struct dm_target_io *tio = alloc_tio(ci, ti, ci->bio->bi_max_vecs, target_bio_nr);
1209 	struct bio *clone = &tio->clone;
1210 
1211 	/*
1212 	 * Discard requests require the bio's inline iovecs be initialized.
1213 	 * ci->bio->bi_max_vecs is BIO_INLINE_VECS anyway, for both flush
1214 	 * and discard, so no need for concern about wasted bvec allocations.
1215 	 */
1216 	 __bio_clone_fast(clone, ci->bio);
1217 	if (len)
1218 		bio_setup_sector(clone, ci->sector, len);
1219 
1220 	__map_bio(tio);
1221 }
1222 
1223 static void __send_duplicate_bios(struct clone_info *ci, struct dm_target *ti,
1224 				  unsigned num_bios, sector_t len)
1225 {
1226 	unsigned target_bio_nr;
1227 
1228 	for (target_bio_nr = 0; target_bio_nr < num_bios; target_bio_nr++)
1229 		__clone_and_map_simple_bio(ci, ti, target_bio_nr, len);
1230 }
1231 
1232 static int __send_empty_flush(struct clone_info *ci)
1233 {
1234 	unsigned target_nr = 0;
1235 	struct dm_target *ti;
1236 
1237 	BUG_ON(bio_has_data(ci->bio));
1238 	while ((ti = dm_table_get_target(ci->map, target_nr++)))
1239 		__send_duplicate_bios(ci, ti, ti->num_flush_bios, 0);
1240 
1241 	return 0;
1242 }
1243 
1244 static void __clone_and_map_data_bio(struct clone_info *ci, struct dm_target *ti,
1245 				     sector_t sector, unsigned len)
1246 {
1247 	struct bio *bio = ci->bio;
1248 	struct dm_target_io *tio;
1249 	unsigned target_bio_nr;
1250 	unsigned num_target_bios = 1;
1251 
1252 	/*
1253 	 * Does the target want to receive duplicate copies of the bio?
1254 	 */
1255 	if (bio_data_dir(bio) == WRITE && ti->num_write_bios)
1256 		num_target_bios = ti->num_write_bios(ti, bio);
1257 
1258 	for (target_bio_nr = 0; target_bio_nr < num_target_bios; target_bio_nr++) {
1259 		tio = alloc_tio(ci, ti, 0, target_bio_nr);
1260 		clone_bio(tio, bio, sector, len);
1261 		__map_bio(tio);
1262 	}
1263 }
1264 
1265 typedef unsigned (*get_num_bios_fn)(struct dm_target *ti);
1266 
1267 static unsigned get_num_discard_bios(struct dm_target *ti)
1268 {
1269 	return ti->num_discard_bios;
1270 }
1271 
1272 static unsigned get_num_write_same_bios(struct dm_target *ti)
1273 {
1274 	return ti->num_write_same_bios;
1275 }
1276 
1277 typedef bool (*is_split_required_fn)(struct dm_target *ti);
1278 
1279 static bool is_split_required_for_discard(struct dm_target *ti)
1280 {
1281 	return ti->split_discard_bios;
1282 }
1283 
1284 static int __send_changing_extent_only(struct clone_info *ci,
1285 				       get_num_bios_fn get_num_bios,
1286 				       is_split_required_fn is_split_required)
1287 {
1288 	struct dm_target *ti;
1289 	sector_t len;
1290 	unsigned num_bios;
1291 
1292 	do {
1293 		ti = dm_table_find_target(ci->map, ci->sector);
1294 		if (!dm_target_is_valid(ti))
1295 			return -EIO;
1296 
1297 		/*
1298 		 * Even though the device advertised support for this type of
1299 		 * request, that does not mean every target supports it, and
1300 		 * reconfiguration might also have changed that since the
1301 		 * check was performed.
1302 		 */
1303 		num_bios = get_num_bios ? get_num_bios(ti) : 0;
1304 		if (!num_bios)
1305 			return -EOPNOTSUPP;
1306 
1307 		if (is_split_required && !is_split_required(ti))
1308 			len = min(ci->sector_count, max_io_len_target_boundary(ci->sector, ti));
1309 		else
1310 			len = min(ci->sector_count, max_io_len(ci->sector, ti));
1311 
1312 		__send_duplicate_bios(ci, ti, num_bios, len);
1313 
1314 		ci->sector += len;
1315 	} while (ci->sector_count -= len);
1316 
1317 	return 0;
1318 }
1319 
1320 static int __send_discard(struct clone_info *ci)
1321 {
1322 	return __send_changing_extent_only(ci, get_num_discard_bios,
1323 					   is_split_required_for_discard);
1324 }
1325 
1326 static int __send_write_same(struct clone_info *ci)
1327 {
1328 	return __send_changing_extent_only(ci, get_num_write_same_bios, NULL);
1329 }
1330 
1331 /*
1332  * Select the correct strategy for processing a non-flush bio.
1333  */
1334 static int __split_and_process_non_flush(struct clone_info *ci)
1335 {
1336 	struct bio *bio = ci->bio;
1337 	struct dm_target *ti;
1338 	unsigned len;
1339 
1340 	if (unlikely(bio->bi_rw & REQ_DISCARD))
1341 		return __send_discard(ci);
1342 	else if (unlikely(bio->bi_rw & REQ_WRITE_SAME))
1343 		return __send_write_same(ci);
1344 
1345 	ti = dm_table_find_target(ci->map, ci->sector);
1346 	if (!dm_target_is_valid(ti))
1347 		return -EIO;
1348 
1349 	len = min_t(sector_t, max_io_len(ci->sector, ti), ci->sector_count);
1350 
1351 	__clone_and_map_data_bio(ci, ti, ci->sector, len);
1352 
1353 	ci->sector += len;
1354 	ci->sector_count -= len;
1355 
1356 	return 0;
1357 }
1358 
1359 /*
1360  * Entry point to split a bio into clones and submit them to the targets.
1361  */
1362 static void __split_and_process_bio(struct mapped_device *md,
1363 				    struct dm_table *map, struct bio *bio)
1364 {
1365 	struct clone_info ci;
1366 	int error = 0;
1367 
1368 	if (unlikely(!map)) {
1369 		bio_io_error(bio);
1370 		return;
1371 	}
1372 
1373 	ci.map = map;
1374 	ci.md = md;
1375 	ci.io = alloc_io(md);
1376 	ci.io->error = 0;
1377 	atomic_set(&ci.io->io_count, 1);
1378 	ci.io->bio = bio;
1379 	ci.io->md = md;
1380 	spin_lock_init(&ci.io->endio_lock);
1381 	ci.sector = bio->bi_iter.bi_sector;
1382 
1383 	start_io_acct(ci.io);
1384 
1385 	if (bio->bi_rw & REQ_FLUSH) {
1386 		ci.bio = &ci.md->flush_bio;
1387 		ci.sector_count = 0;
1388 		error = __send_empty_flush(&ci);
1389 		/* dec_pending submits any data associated with flush */
1390 	} else {
1391 		ci.bio = bio;
1392 		ci.sector_count = bio_sectors(bio);
1393 		while (ci.sector_count && !error)
1394 			error = __split_and_process_non_flush(&ci);
1395 	}
1396 
1397 	/* drop the extra reference count */
1398 	dec_pending(ci.io, error);
1399 }
1400 /*-----------------------------------------------------------------
1401  * CRUD END
1402  *---------------------------------------------------------------*/
1403 
1404 static int dm_merge_bvec(struct request_queue *q,
1405 			 struct bvec_merge_data *bvm,
1406 			 struct bio_vec *biovec)
1407 {
1408 	struct mapped_device *md = q->queuedata;
1409 	struct dm_table *map = dm_get_live_table_fast(md);
1410 	struct dm_target *ti;
1411 	sector_t max_sectors;
1412 	int max_size = 0;
1413 
1414 	if (unlikely(!map))
1415 		goto out;
1416 
1417 	ti = dm_table_find_target(map, bvm->bi_sector);
1418 	if (!dm_target_is_valid(ti))
1419 		goto out;
1420 
1421 	/*
1422 	 * Find maximum amount of I/O that won't need splitting
1423 	 */
1424 	max_sectors = min(max_io_len(bvm->bi_sector, ti),
1425 			  (sector_t) BIO_MAX_SECTORS);
1426 	max_size = (max_sectors << SECTOR_SHIFT) - bvm->bi_size;
1427 	if (max_size < 0)
1428 		max_size = 0;
1429 
1430 	/*
1431 	 * merge_bvec_fn() returns number of bytes
1432 	 * it can accept at this offset
1433 	 * max is precomputed maximal io size
1434 	 */
1435 	if (max_size && ti->type->merge)
1436 		max_size = ti->type->merge(ti, bvm, biovec, max_size);
1437 	/*
1438 	 * If the target doesn't support merge method and some of the devices
1439 	 * provided their merge_bvec method (we know this by looking at
1440 	 * queue_max_hw_sectors), then we can't allow bios with multiple vector
1441 	 * entries.  So always set max_size to 0, and the code below allows
1442 	 * just one page.
1443 	 */
1444 	else if (queue_max_hw_sectors(q) <= PAGE_SIZE >> 9)
1445 
1446 		max_size = 0;
1447 
1448 out:
1449 	dm_put_live_table_fast(md);
1450 	/*
1451 	 * Always allow an entire first page
1452 	 */
1453 	if (max_size <= biovec->bv_len && !(bvm->bi_size >> SECTOR_SHIFT))
1454 		max_size = biovec->bv_len;
1455 
1456 	return max_size;
1457 }
1458 
1459 /*
1460  * The request function that just remaps the bio built up by
1461  * dm_merge_bvec.
1462  */
1463 static void _dm_request(struct request_queue *q, struct bio *bio)
1464 {
1465 	int rw = bio_data_dir(bio);
1466 	struct mapped_device *md = q->queuedata;
1467 	int cpu;
1468 	int srcu_idx;
1469 	struct dm_table *map;
1470 
1471 	map = dm_get_live_table(md, &srcu_idx);
1472 
1473 	cpu = part_stat_lock();
1474 	part_stat_inc(cpu, &dm_disk(md)->part0, ios[rw]);
1475 	part_stat_add(cpu, &dm_disk(md)->part0, sectors[rw], bio_sectors(bio));
1476 	part_stat_unlock();
1477 
1478 	/* if we're suspended, we have to queue this io for later */
1479 	if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags))) {
1480 		dm_put_live_table(md, srcu_idx);
1481 
1482 		if (bio_rw(bio) != READA)
1483 			queue_io(md, bio);
1484 		else
1485 			bio_io_error(bio);
1486 		return;
1487 	}
1488 
1489 	__split_and_process_bio(md, map, bio);
1490 	dm_put_live_table(md, srcu_idx);
1491 	return;
1492 }
1493 
1494 int dm_request_based(struct mapped_device *md)
1495 {
1496 	return blk_queue_stackable(md->queue);
1497 }
1498 
1499 static void dm_request(struct request_queue *q, struct bio *bio)
1500 {
1501 	struct mapped_device *md = q->queuedata;
1502 
1503 	if (dm_request_based(md))
1504 		blk_queue_bio(q, bio);
1505 	else
1506 		_dm_request(q, bio);
1507 }
1508 
1509 void dm_dispatch_request(struct request *rq)
1510 {
1511 	int r;
1512 
1513 	if (blk_queue_io_stat(rq->q))
1514 		rq->cmd_flags |= REQ_IO_STAT;
1515 
1516 	rq->start_time = jiffies;
1517 	r = blk_insert_cloned_request(rq->q, rq);
1518 	if (r)
1519 		dm_complete_request(rq, r);
1520 }
1521 EXPORT_SYMBOL_GPL(dm_dispatch_request);
1522 
1523 static int dm_rq_bio_constructor(struct bio *bio, struct bio *bio_orig,
1524 				 void *data)
1525 {
1526 	struct dm_rq_target_io *tio = data;
1527 	struct dm_rq_clone_bio_info *info =
1528 		container_of(bio, struct dm_rq_clone_bio_info, clone);
1529 
1530 	info->orig = bio_orig;
1531 	info->tio = tio;
1532 	bio->bi_end_io = end_clone_bio;
1533 	bio->bi_private = info;
1534 
1535 	return 0;
1536 }
1537 
1538 static int setup_clone(struct request *clone, struct request *rq,
1539 		       struct dm_rq_target_io *tio)
1540 {
1541 	int r;
1542 
1543 	r = blk_rq_prep_clone(clone, rq, tio->md->bs, GFP_ATOMIC,
1544 			      dm_rq_bio_constructor, tio);
1545 	if (r)
1546 		return r;
1547 
1548 	clone->cmd = rq->cmd;
1549 	clone->cmd_len = rq->cmd_len;
1550 	clone->sense = rq->sense;
1551 	clone->buffer = rq->buffer;
1552 	clone->end_io = end_clone_request;
1553 	clone->end_io_data = tio;
1554 
1555 	return 0;
1556 }
1557 
1558 static struct request *clone_rq(struct request *rq, struct mapped_device *md,
1559 				gfp_t gfp_mask)
1560 {
1561 	struct request *clone;
1562 	struct dm_rq_target_io *tio;
1563 
1564 	tio = alloc_rq_tio(md, gfp_mask);
1565 	if (!tio)
1566 		return NULL;
1567 
1568 	tio->md = md;
1569 	tio->ti = NULL;
1570 	tio->orig = rq;
1571 	tio->error = 0;
1572 	memset(&tio->info, 0, sizeof(tio->info));
1573 
1574 	clone = &tio->clone;
1575 	if (setup_clone(clone, rq, tio)) {
1576 		/* -ENOMEM */
1577 		free_rq_tio(tio);
1578 		return NULL;
1579 	}
1580 
1581 	return clone;
1582 }
1583 
1584 /*
1585  * Called with the queue lock held.
1586  */
1587 static int dm_prep_fn(struct request_queue *q, struct request *rq)
1588 {
1589 	struct mapped_device *md = q->queuedata;
1590 	struct request *clone;
1591 
1592 	if (unlikely(rq->special)) {
1593 		DMWARN("Already has something in rq->special.");
1594 		return BLKPREP_KILL;
1595 	}
1596 
1597 	clone = clone_rq(rq, md, GFP_ATOMIC);
1598 	if (!clone)
1599 		return BLKPREP_DEFER;
1600 
1601 	rq->special = clone;
1602 	rq->cmd_flags |= REQ_DONTPREP;
1603 
1604 	return BLKPREP_OK;
1605 }
1606 
1607 /*
1608  * Returns:
1609  * 0  : the request has been processed (not requeued)
1610  * !0 : the request has been requeued
1611  */
1612 static int map_request(struct dm_target *ti, struct request *clone,
1613 		       struct mapped_device *md)
1614 {
1615 	int r, requeued = 0;
1616 	struct dm_rq_target_io *tio = clone->end_io_data;
1617 
1618 	tio->ti = ti;
1619 	r = ti->type->map_rq(ti, clone, &tio->info);
1620 	switch (r) {
1621 	case DM_MAPIO_SUBMITTED:
1622 		/* The target has taken the I/O to submit by itself later */
1623 		break;
1624 	case DM_MAPIO_REMAPPED:
1625 		/* The target has remapped the I/O so dispatch it */
1626 		trace_block_rq_remap(clone->q, clone, disk_devt(dm_disk(md)),
1627 				     blk_rq_pos(tio->orig));
1628 		dm_dispatch_request(clone);
1629 		break;
1630 	case DM_MAPIO_REQUEUE:
1631 		/* The target wants to requeue the I/O */
1632 		dm_requeue_unmapped_request(clone);
1633 		requeued = 1;
1634 		break;
1635 	default:
1636 		if (r > 0) {
1637 			DMWARN("unimplemented target map return value: %d", r);
1638 			BUG();
1639 		}
1640 
1641 		/* The target wants to complete the I/O */
1642 		dm_kill_unmapped_request(clone, r);
1643 		break;
1644 	}
1645 
1646 	return requeued;
1647 }
1648 
1649 static struct request *dm_start_request(struct mapped_device *md, struct request *orig)
1650 {
1651 	struct request *clone;
1652 
1653 	blk_start_request(orig);
1654 	clone = orig->special;
1655 	atomic_inc(&md->pending[rq_data_dir(clone)]);
1656 
1657 	/*
1658 	 * Hold the md reference here for the in-flight I/O.
1659 	 * We can't rely on the reference count by device opener,
1660 	 * because the device may be closed during the request completion
1661 	 * when all bios are completed.
1662 	 * See the comment in rq_completed() too.
1663 	 */
1664 	dm_get(md);
1665 
1666 	return clone;
1667 }
1668 
1669 /*
1670  * q->request_fn for request-based dm.
1671  * Called with the queue lock held.
1672  */
1673 static void dm_request_fn(struct request_queue *q)
1674 {
1675 	struct mapped_device *md = q->queuedata;
1676 	int srcu_idx;
1677 	struct dm_table *map = dm_get_live_table(md, &srcu_idx);
1678 	struct dm_target *ti;
1679 	struct request *rq, *clone;
1680 	sector_t pos;
1681 
1682 	/*
1683 	 * For suspend, check blk_queue_stopped() and increment
1684 	 * ->pending within a single queue_lock not to increment the
1685 	 * number of in-flight I/Os after the queue is stopped in
1686 	 * dm_suspend().
1687 	 */
1688 	while (!blk_queue_stopped(q)) {
1689 		rq = blk_peek_request(q);
1690 		if (!rq)
1691 			goto delay_and_out;
1692 
1693 		/* always use block 0 to find the target for flushes for now */
1694 		pos = 0;
1695 		if (!(rq->cmd_flags & REQ_FLUSH))
1696 			pos = blk_rq_pos(rq);
1697 
1698 		ti = dm_table_find_target(map, pos);
1699 		if (!dm_target_is_valid(ti)) {
1700 			/*
1701 			 * Must perform setup, that dm_done() requires,
1702 			 * before calling dm_kill_unmapped_request
1703 			 */
1704 			DMERR_LIMIT("request attempted access beyond the end of device");
1705 			clone = dm_start_request(md, rq);
1706 			dm_kill_unmapped_request(clone, -EIO);
1707 			continue;
1708 		}
1709 
1710 		if (ti->type->busy && ti->type->busy(ti))
1711 			goto delay_and_out;
1712 
1713 		clone = dm_start_request(md, rq);
1714 
1715 		spin_unlock(q->queue_lock);
1716 		if (map_request(ti, clone, md))
1717 			goto requeued;
1718 
1719 		BUG_ON(!irqs_disabled());
1720 		spin_lock(q->queue_lock);
1721 	}
1722 
1723 	goto out;
1724 
1725 requeued:
1726 	BUG_ON(!irqs_disabled());
1727 	spin_lock(q->queue_lock);
1728 
1729 delay_and_out:
1730 	blk_delay_queue(q, HZ / 10);
1731 out:
1732 	dm_put_live_table(md, srcu_idx);
1733 }
1734 
1735 int dm_underlying_device_busy(struct request_queue *q)
1736 {
1737 	return blk_lld_busy(q);
1738 }
1739 EXPORT_SYMBOL_GPL(dm_underlying_device_busy);
1740 
1741 static int dm_lld_busy(struct request_queue *q)
1742 {
1743 	int r;
1744 	struct mapped_device *md = q->queuedata;
1745 	struct dm_table *map = dm_get_live_table_fast(md);
1746 
1747 	if (!map || test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags))
1748 		r = 1;
1749 	else
1750 		r = dm_table_any_busy_target(map);
1751 
1752 	dm_put_live_table_fast(md);
1753 
1754 	return r;
1755 }
1756 
1757 static int dm_any_congested(void *congested_data, int bdi_bits)
1758 {
1759 	int r = bdi_bits;
1760 	struct mapped_device *md = congested_data;
1761 	struct dm_table *map;
1762 
1763 	if (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
1764 		map = dm_get_live_table_fast(md);
1765 		if (map) {
1766 			/*
1767 			 * Request-based dm cares about only own queue for
1768 			 * the query about congestion status of request_queue
1769 			 */
1770 			if (dm_request_based(md))
1771 				r = md->queue->backing_dev_info.state &
1772 				    bdi_bits;
1773 			else
1774 				r = dm_table_any_congested(map, bdi_bits);
1775 		}
1776 		dm_put_live_table_fast(md);
1777 	}
1778 
1779 	return r;
1780 }
1781 
1782 /*-----------------------------------------------------------------
1783  * An IDR is used to keep track of allocated minor numbers.
1784  *---------------------------------------------------------------*/
1785 static void free_minor(int minor)
1786 {
1787 	spin_lock(&_minor_lock);
1788 	idr_remove(&_minor_idr, minor);
1789 	spin_unlock(&_minor_lock);
1790 }
1791 
1792 /*
1793  * See if the device with a specific minor # is free.
1794  */
1795 static int specific_minor(int minor)
1796 {
1797 	int r;
1798 
1799 	if (minor >= (1 << MINORBITS))
1800 		return -EINVAL;
1801 
1802 	idr_preload(GFP_KERNEL);
1803 	spin_lock(&_minor_lock);
1804 
1805 	r = idr_alloc(&_minor_idr, MINOR_ALLOCED, minor, minor + 1, GFP_NOWAIT);
1806 
1807 	spin_unlock(&_minor_lock);
1808 	idr_preload_end();
1809 	if (r < 0)
1810 		return r == -ENOSPC ? -EBUSY : r;
1811 	return 0;
1812 }
1813 
1814 static int next_free_minor(int *minor)
1815 {
1816 	int r;
1817 
1818 	idr_preload(GFP_KERNEL);
1819 	spin_lock(&_minor_lock);
1820 
1821 	r = idr_alloc(&_minor_idr, MINOR_ALLOCED, 0, 1 << MINORBITS, GFP_NOWAIT);
1822 
1823 	spin_unlock(&_minor_lock);
1824 	idr_preload_end();
1825 	if (r < 0)
1826 		return r;
1827 	*minor = r;
1828 	return 0;
1829 }
1830 
1831 static const struct block_device_operations dm_blk_dops;
1832 
1833 static void dm_wq_work(struct work_struct *work);
1834 
1835 static void dm_init_md_queue(struct mapped_device *md)
1836 {
1837 	/*
1838 	 * Request-based dm devices cannot be stacked on top of bio-based dm
1839 	 * devices.  The type of this dm device has not been decided yet.
1840 	 * The type is decided at the first table loading time.
1841 	 * To prevent problematic device stacking, clear the queue flag
1842 	 * for request stacking support until then.
1843 	 *
1844 	 * This queue is new, so no concurrency on the queue_flags.
1845 	 */
1846 	queue_flag_clear_unlocked(QUEUE_FLAG_STACKABLE, md->queue);
1847 
1848 	md->queue->queuedata = md;
1849 	md->queue->backing_dev_info.congested_fn = dm_any_congested;
1850 	md->queue->backing_dev_info.congested_data = md;
1851 	blk_queue_make_request(md->queue, dm_request);
1852 	blk_queue_bounce_limit(md->queue, BLK_BOUNCE_ANY);
1853 	blk_queue_merge_bvec(md->queue, dm_merge_bvec);
1854 }
1855 
1856 /*
1857  * Allocate and initialise a blank device with a given minor.
1858  */
1859 static struct mapped_device *alloc_dev(int minor)
1860 {
1861 	int r;
1862 	struct mapped_device *md = kzalloc(sizeof(*md), GFP_KERNEL);
1863 	void *old_md;
1864 
1865 	if (!md) {
1866 		DMWARN("unable to allocate device, out of memory.");
1867 		return NULL;
1868 	}
1869 
1870 	if (!try_module_get(THIS_MODULE))
1871 		goto bad_module_get;
1872 
1873 	/* get a minor number for the dev */
1874 	if (minor == DM_ANY_MINOR)
1875 		r = next_free_minor(&minor);
1876 	else
1877 		r = specific_minor(minor);
1878 	if (r < 0)
1879 		goto bad_minor;
1880 
1881 	r = init_srcu_struct(&md->io_barrier);
1882 	if (r < 0)
1883 		goto bad_io_barrier;
1884 
1885 	md->type = DM_TYPE_NONE;
1886 	mutex_init(&md->suspend_lock);
1887 	mutex_init(&md->type_lock);
1888 	spin_lock_init(&md->deferred_lock);
1889 	atomic_set(&md->holders, 1);
1890 	atomic_set(&md->open_count, 0);
1891 	atomic_set(&md->event_nr, 0);
1892 	atomic_set(&md->uevent_seq, 0);
1893 	INIT_LIST_HEAD(&md->uevent_list);
1894 	spin_lock_init(&md->uevent_lock);
1895 
1896 	md->queue = blk_alloc_queue(GFP_KERNEL);
1897 	if (!md->queue)
1898 		goto bad_queue;
1899 
1900 	dm_init_md_queue(md);
1901 
1902 	md->disk = alloc_disk(1);
1903 	if (!md->disk)
1904 		goto bad_disk;
1905 
1906 	atomic_set(&md->pending[0], 0);
1907 	atomic_set(&md->pending[1], 0);
1908 	init_waitqueue_head(&md->wait);
1909 	INIT_WORK(&md->work, dm_wq_work);
1910 	init_waitqueue_head(&md->eventq);
1911 	init_completion(&md->kobj_holder.completion);
1912 
1913 	md->disk->major = _major;
1914 	md->disk->first_minor = minor;
1915 	md->disk->fops = &dm_blk_dops;
1916 	md->disk->queue = md->queue;
1917 	md->disk->private_data = md;
1918 	sprintf(md->disk->disk_name, "dm-%d", minor);
1919 	add_disk(md->disk);
1920 	format_dev_t(md->name, MKDEV(_major, minor));
1921 
1922 	md->wq = alloc_workqueue("kdmflush", WQ_MEM_RECLAIM, 0);
1923 	if (!md->wq)
1924 		goto bad_thread;
1925 
1926 	md->bdev = bdget_disk(md->disk, 0);
1927 	if (!md->bdev)
1928 		goto bad_bdev;
1929 
1930 	bio_init(&md->flush_bio);
1931 	md->flush_bio.bi_bdev = md->bdev;
1932 	md->flush_bio.bi_rw = WRITE_FLUSH;
1933 
1934 	dm_stats_init(&md->stats);
1935 
1936 	/* Populate the mapping, nobody knows we exist yet */
1937 	spin_lock(&_minor_lock);
1938 	old_md = idr_replace(&_minor_idr, md, minor);
1939 	spin_unlock(&_minor_lock);
1940 
1941 	BUG_ON(old_md != MINOR_ALLOCED);
1942 
1943 	return md;
1944 
1945 bad_bdev:
1946 	destroy_workqueue(md->wq);
1947 bad_thread:
1948 	del_gendisk(md->disk);
1949 	put_disk(md->disk);
1950 bad_disk:
1951 	blk_cleanup_queue(md->queue);
1952 bad_queue:
1953 	cleanup_srcu_struct(&md->io_barrier);
1954 bad_io_barrier:
1955 	free_minor(minor);
1956 bad_minor:
1957 	module_put(THIS_MODULE);
1958 bad_module_get:
1959 	kfree(md);
1960 	return NULL;
1961 }
1962 
1963 static void unlock_fs(struct mapped_device *md);
1964 
1965 static void free_dev(struct mapped_device *md)
1966 {
1967 	int minor = MINOR(disk_devt(md->disk));
1968 
1969 	unlock_fs(md);
1970 	bdput(md->bdev);
1971 	destroy_workqueue(md->wq);
1972 	if (md->io_pool)
1973 		mempool_destroy(md->io_pool);
1974 	if (md->bs)
1975 		bioset_free(md->bs);
1976 	blk_integrity_unregister(md->disk);
1977 	del_gendisk(md->disk);
1978 	cleanup_srcu_struct(&md->io_barrier);
1979 	free_minor(minor);
1980 
1981 	spin_lock(&_minor_lock);
1982 	md->disk->private_data = NULL;
1983 	spin_unlock(&_minor_lock);
1984 
1985 	put_disk(md->disk);
1986 	blk_cleanup_queue(md->queue);
1987 	dm_stats_cleanup(&md->stats);
1988 	module_put(THIS_MODULE);
1989 	kfree(md);
1990 }
1991 
1992 static void __bind_mempools(struct mapped_device *md, struct dm_table *t)
1993 {
1994 	struct dm_md_mempools *p = dm_table_get_md_mempools(t);
1995 
1996 	if (md->io_pool && md->bs) {
1997 		/* The md already has necessary mempools. */
1998 		if (dm_table_get_type(t) == DM_TYPE_BIO_BASED) {
1999 			/*
2000 			 * Reload bioset because front_pad may have changed
2001 			 * because a different table was loaded.
2002 			 */
2003 			bioset_free(md->bs);
2004 			md->bs = p->bs;
2005 			p->bs = NULL;
2006 		} else if (dm_table_get_type(t) == DM_TYPE_REQUEST_BASED) {
2007 			/*
2008 			 * There's no need to reload with request-based dm
2009 			 * because the size of front_pad doesn't change.
2010 			 * Note for future: If you are to reload bioset,
2011 			 * prep-ed requests in the queue may refer
2012 			 * to bio from the old bioset, so you must walk
2013 			 * through the queue to unprep.
2014 			 */
2015 		}
2016 		goto out;
2017 	}
2018 
2019 	BUG_ON(!p || md->io_pool || md->bs);
2020 
2021 	md->io_pool = p->io_pool;
2022 	p->io_pool = NULL;
2023 	md->bs = p->bs;
2024 	p->bs = NULL;
2025 
2026 out:
2027 	/* mempool bind completed, now no need any mempools in the table */
2028 	dm_table_free_md_mempools(t);
2029 }
2030 
2031 /*
2032  * Bind a table to the device.
2033  */
2034 static void event_callback(void *context)
2035 {
2036 	unsigned long flags;
2037 	LIST_HEAD(uevents);
2038 	struct mapped_device *md = (struct mapped_device *) context;
2039 
2040 	spin_lock_irqsave(&md->uevent_lock, flags);
2041 	list_splice_init(&md->uevent_list, &uevents);
2042 	spin_unlock_irqrestore(&md->uevent_lock, flags);
2043 
2044 	dm_send_uevents(&uevents, &disk_to_dev(md->disk)->kobj);
2045 
2046 	atomic_inc(&md->event_nr);
2047 	wake_up(&md->eventq);
2048 }
2049 
2050 /*
2051  * Protected by md->suspend_lock obtained by dm_swap_table().
2052  */
2053 static void __set_size(struct mapped_device *md, sector_t size)
2054 {
2055 	set_capacity(md->disk, size);
2056 
2057 	i_size_write(md->bdev->bd_inode, (loff_t)size << SECTOR_SHIFT);
2058 }
2059 
2060 /*
2061  * Return 1 if the queue has a compulsory merge_bvec_fn function.
2062  *
2063  * If this function returns 0, then the device is either a non-dm
2064  * device without a merge_bvec_fn, or it is a dm device that is
2065  * able to split any bios it receives that are too big.
2066  */
2067 int dm_queue_merge_is_compulsory(struct request_queue *q)
2068 {
2069 	struct mapped_device *dev_md;
2070 
2071 	if (!q->merge_bvec_fn)
2072 		return 0;
2073 
2074 	if (q->make_request_fn == dm_request) {
2075 		dev_md = q->queuedata;
2076 		if (test_bit(DMF_MERGE_IS_OPTIONAL, &dev_md->flags))
2077 			return 0;
2078 	}
2079 
2080 	return 1;
2081 }
2082 
2083 static int dm_device_merge_is_compulsory(struct dm_target *ti,
2084 					 struct dm_dev *dev, sector_t start,
2085 					 sector_t len, void *data)
2086 {
2087 	struct block_device *bdev = dev->bdev;
2088 	struct request_queue *q = bdev_get_queue(bdev);
2089 
2090 	return dm_queue_merge_is_compulsory(q);
2091 }
2092 
2093 /*
2094  * Return 1 if it is acceptable to ignore merge_bvec_fn based
2095  * on the properties of the underlying devices.
2096  */
2097 static int dm_table_merge_is_optional(struct dm_table *table)
2098 {
2099 	unsigned i = 0;
2100 	struct dm_target *ti;
2101 
2102 	while (i < dm_table_get_num_targets(table)) {
2103 		ti = dm_table_get_target(table, i++);
2104 
2105 		if (ti->type->iterate_devices &&
2106 		    ti->type->iterate_devices(ti, dm_device_merge_is_compulsory, NULL))
2107 			return 0;
2108 	}
2109 
2110 	return 1;
2111 }
2112 
2113 /*
2114  * Returns old map, which caller must destroy.
2115  */
2116 static struct dm_table *__bind(struct mapped_device *md, struct dm_table *t,
2117 			       struct queue_limits *limits)
2118 {
2119 	struct dm_table *old_map;
2120 	struct request_queue *q = md->queue;
2121 	sector_t size;
2122 	int merge_is_optional;
2123 
2124 	size = dm_table_get_size(t);
2125 
2126 	/*
2127 	 * Wipe any geometry if the size of the table changed.
2128 	 */
2129 	if (size != dm_get_size(md))
2130 		memset(&md->geometry, 0, sizeof(md->geometry));
2131 
2132 	__set_size(md, size);
2133 
2134 	dm_table_event_callback(t, event_callback, md);
2135 
2136 	/*
2137 	 * The queue hasn't been stopped yet, if the old table type wasn't
2138 	 * for request-based during suspension.  So stop it to prevent
2139 	 * I/O mapping before resume.
2140 	 * This must be done before setting the queue restrictions,
2141 	 * because request-based dm may be run just after the setting.
2142 	 */
2143 	if (dm_table_request_based(t) && !blk_queue_stopped(q))
2144 		stop_queue(q);
2145 
2146 	__bind_mempools(md, t);
2147 
2148 	merge_is_optional = dm_table_merge_is_optional(t);
2149 
2150 	old_map = md->map;
2151 	rcu_assign_pointer(md->map, t);
2152 	md->immutable_target_type = dm_table_get_immutable_target_type(t);
2153 
2154 	dm_table_set_restrictions(t, q, limits);
2155 	if (merge_is_optional)
2156 		set_bit(DMF_MERGE_IS_OPTIONAL, &md->flags);
2157 	else
2158 		clear_bit(DMF_MERGE_IS_OPTIONAL, &md->flags);
2159 	dm_sync_table(md);
2160 
2161 	return old_map;
2162 }
2163 
2164 /*
2165  * Returns unbound table for the caller to free.
2166  */
2167 static struct dm_table *__unbind(struct mapped_device *md)
2168 {
2169 	struct dm_table *map = md->map;
2170 
2171 	if (!map)
2172 		return NULL;
2173 
2174 	dm_table_event_callback(map, NULL, NULL);
2175 	rcu_assign_pointer(md->map, NULL);
2176 	dm_sync_table(md);
2177 
2178 	return map;
2179 }
2180 
2181 /*
2182  * Constructor for a new device.
2183  */
2184 int dm_create(int minor, struct mapped_device **result)
2185 {
2186 	struct mapped_device *md;
2187 
2188 	md = alloc_dev(minor);
2189 	if (!md)
2190 		return -ENXIO;
2191 
2192 	dm_sysfs_init(md);
2193 
2194 	*result = md;
2195 	return 0;
2196 }
2197 
2198 /*
2199  * Functions to manage md->type.
2200  * All are required to hold md->type_lock.
2201  */
2202 void dm_lock_md_type(struct mapped_device *md)
2203 {
2204 	mutex_lock(&md->type_lock);
2205 }
2206 
2207 void dm_unlock_md_type(struct mapped_device *md)
2208 {
2209 	mutex_unlock(&md->type_lock);
2210 }
2211 
2212 void dm_set_md_type(struct mapped_device *md, unsigned type)
2213 {
2214 	BUG_ON(!mutex_is_locked(&md->type_lock));
2215 	md->type = type;
2216 }
2217 
2218 unsigned dm_get_md_type(struct mapped_device *md)
2219 {
2220 	BUG_ON(!mutex_is_locked(&md->type_lock));
2221 	return md->type;
2222 }
2223 
2224 struct target_type *dm_get_immutable_target_type(struct mapped_device *md)
2225 {
2226 	return md->immutable_target_type;
2227 }
2228 
2229 /*
2230  * The queue_limits are only valid as long as you have a reference
2231  * count on 'md'.
2232  */
2233 struct queue_limits *dm_get_queue_limits(struct mapped_device *md)
2234 {
2235 	BUG_ON(!atomic_read(&md->holders));
2236 	return &md->queue->limits;
2237 }
2238 EXPORT_SYMBOL_GPL(dm_get_queue_limits);
2239 
2240 /*
2241  * Fully initialize a request-based queue (->elevator, ->request_fn, etc).
2242  */
2243 static int dm_init_request_based_queue(struct mapped_device *md)
2244 {
2245 	struct request_queue *q = NULL;
2246 
2247 	if (md->queue->elevator)
2248 		return 1;
2249 
2250 	/* Fully initialize the queue */
2251 	q = blk_init_allocated_queue(md->queue, dm_request_fn, NULL);
2252 	if (!q)
2253 		return 0;
2254 
2255 	md->queue = q;
2256 	dm_init_md_queue(md);
2257 	blk_queue_softirq_done(md->queue, dm_softirq_done);
2258 	blk_queue_prep_rq(md->queue, dm_prep_fn);
2259 	blk_queue_lld_busy(md->queue, dm_lld_busy);
2260 
2261 	elv_register_queue(md->queue);
2262 
2263 	return 1;
2264 }
2265 
2266 /*
2267  * Setup the DM device's queue based on md's type
2268  */
2269 int dm_setup_md_queue(struct mapped_device *md)
2270 {
2271 	if ((dm_get_md_type(md) == DM_TYPE_REQUEST_BASED) &&
2272 	    !dm_init_request_based_queue(md)) {
2273 		DMWARN("Cannot initialize queue for request-based mapped device");
2274 		return -EINVAL;
2275 	}
2276 
2277 	return 0;
2278 }
2279 
2280 static struct mapped_device *dm_find_md(dev_t dev)
2281 {
2282 	struct mapped_device *md;
2283 	unsigned minor = MINOR(dev);
2284 
2285 	if (MAJOR(dev) != _major || minor >= (1 << MINORBITS))
2286 		return NULL;
2287 
2288 	spin_lock(&_minor_lock);
2289 
2290 	md = idr_find(&_minor_idr, minor);
2291 	if (md && (md == MINOR_ALLOCED ||
2292 		   (MINOR(disk_devt(dm_disk(md))) != minor) ||
2293 		   dm_deleting_md(md) ||
2294 		   test_bit(DMF_FREEING, &md->flags))) {
2295 		md = NULL;
2296 		goto out;
2297 	}
2298 
2299 out:
2300 	spin_unlock(&_minor_lock);
2301 
2302 	return md;
2303 }
2304 
2305 struct mapped_device *dm_get_md(dev_t dev)
2306 {
2307 	struct mapped_device *md = dm_find_md(dev);
2308 
2309 	if (md)
2310 		dm_get(md);
2311 
2312 	return md;
2313 }
2314 EXPORT_SYMBOL_GPL(dm_get_md);
2315 
2316 void *dm_get_mdptr(struct mapped_device *md)
2317 {
2318 	return md->interface_ptr;
2319 }
2320 
2321 void dm_set_mdptr(struct mapped_device *md, void *ptr)
2322 {
2323 	md->interface_ptr = ptr;
2324 }
2325 
2326 void dm_get(struct mapped_device *md)
2327 {
2328 	atomic_inc(&md->holders);
2329 	BUG_ON(test_bit(DMF_FREEING, &md->flags));
2330 }
2331 
2332 const char *dm_device_name(struct mapped_device *md)
2333 {
2334 	return md->name;
2335 }
2336 EXPORT_SYMBOL_GPL(dm_device_name);
2337 
2338 static void __dm_destroy(struct mapped_device *md, bool wait)
2339 {
2340 	struct dm_table *map;
2341 	int srcu_idx;
2342 
2343 	might_sleep();
2344 
2345 	spin_lock(&_minor_lock);
2346 	map = dm_get_live_table(md, &srcu_idx);
2347 	idr_replace(&_minor_idr, MINOR_ALLOCED, MINOR(disk_devt(dm_disk(md))));
2348 	set_bit(DMF_FREEING, &md->flags);
2349 	spin_unlock(&_minor_lock);
2350 
2351 	if (!dm_suspended_md(md)) {
2352 		dm_table_presuspend_targets(map);
2353 		dm_table_postsuspend_targets(map);
2354 	}
2355 
2356 	/* dm_put_live_table must be before msleep, otherwise deadlock is possible */
2357 	dm_put_live_table(md, srcu_idx);
2358 
2359 	/*
2360 	 * Rare, but there may be I/O requests still going to complete,
2361 	 * for example.  Wait for all references to disappear.
2362 	 * No one should increment the reference count of the mapped_device,
2363 	 * after the mapped_device state becomes DMF_FREEING.
2364 	 */
2365 	if (wait)
2366 		while (atomic_read(&md->holders))
2367 			msleep(1);
2368 	else if (atomic_read(&md->holders))
2369 		DMWARN("%s: Forcibly removing mapped_device still in use! (%d users)",
2370 		       dm_device_name(md), atomic_read(&md->holders));
2371 
2372 	dm_sysfs_exit(md);
2373 	dm_table_destroy(__unbind(md));
2374 	free_dev(md);
2375 }
2376 
2377 void dm_destroy(struct mapped_device *md)
2378 {
2379 	__dm_destroy(md, true);
2380 }
2381 
2382 void dm_destroy_immediate(struct mapped_device *md)
2383 {
2384 	__dm_destroy(md, false);
2385 }
2386 
2387 void dm_put(struct mapped_device *md)
2388 {
2389 	atomic_dec(&md->holders);
2390 }
2391 EXPORT_SYMBOL_GPL(dm_put);
2392 
2393 static int dm_wait_for_completion(struct mapped_device *md, int interruptible)
2394 {
2395 	int r = 0;
2396 	DECLARE_WAITQUEUE(wait, current);
2397 
2398 	add_wait_queue(&md->wait, &wait);
2399 
2400 	while (1) {
2401 		set_current_state(interruptible);
2402 
2403 		if (!md_in_flight(md))
2404 			break;
2405 
2406 		if (interruptible == TASK_INTERRUPTIBLE &&
2407 		    signal_pending(current)) {
2408 			r = -EINTR;
2409 			break;
2410 		}
2411 
2412 		io_schedule();
2413 	}
2414 	set_current_state(TASK_RUNNING);
2415 
2416 	remove_wait_queue(&md->wait, &wait);
2417 
2418 	return r;
2419 }
2420 
2421 /*
2422  * Process the deferred bios
2423  */
2424 static void dm_wq_work(struct work_struct *work)
2425 {
2426 	struct mapped_device *md = container_of(work, struct mapped_device,
2427 						work);
2428 	struct bio *c;
2429 	int srcu_idx;
2430 	struct dm_table *map;
2431 
2432 	map = dm_get_live_table(md, &srcu_idx);
2433 
2434 	while (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
2435 		spin_lock_irq(&md->deferred_lock);
2436 		c = bio_list_pop(&md->deferred);
2437 		spin_unlock_irq(&md->deferred_lock);
2438 
2439 		if (!c)
2440 			break;
2441 
2442 		if (dm_request_based(md))
2443 			generic_make_request(c);
2444 		else
2445 			__split_and_process_bio(md, map, c);
2446 	}
2447 
2448 	dm_put_live_table(md, srcu_idx);
2449 }
2450 
2451 static void dm_queue_flush(struct mapped_device *md)
2452 {
2453 	clear_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2454 	smp_mb__after_clear_bit();
2455 	queue_work(md->wq, &md->work);
2456 }
2457 
2458 /*
2459  * Swap in a new table, returning the old one for the caller to destroy.
2460  */
2461 struct dm_table *dm_swap_table(struct mapped_device *md, struct dm_table *table)
2462 {
2463 	struct dm_table *live_map = NULL, *map = ERR_PTR(-EINVAL);
2464 	struct queue_limits limits;
2465 	int r;
2466 
2467 	mutex_lock(&md->suspend_lock);
2468 
2469 	/* device must be suspended */
2470 	if (!dm_suspended_md(md))
2471 		goto out;
2472 
2473 	/*
2474 	 * If the new table has no data devices, retain the existing limits.
2475 	 * This helps multipath with queue_if_no_path if all paths disappear,
2476 	 * then new I/O is queued based on these limits, and then some paths
2477 	 * reappear.
2478 	 */
2479 	if (dm_table_has_no_data_devices(table)) {
2480 		live_map = dm_get_live_table_fast(md);
2481 		if (live_map)
2482 			limits = md->queue->limits;
2483 		dm_put_live_table_fast(md);
2484 	}
2485 
2486 	if (!live_map) {
2487 		r = dm_calculate_queue_limits(table, &limits);
2488 		if (r) {
2489 			map = ERR_PTR(r);
2490 			goto out;
2491 		}
2492 	}
2493 
2494 	map = __bind(md, table, &limits);
2495 
2496 out:
2497 	mutex_unlock(&md->suspend_lock);
2498 	return map;
2499 }
2500 
2501 /*
2502  * Functions to lock and unlock any filesystem running on the
2503  * device.
2504  */
2505 static int lock_fs(struct mapped_device *md)
2506 {
2507 	int r;
2508 
2509 	WARN_ON(md->frozen_sb);
2510 
2511 	md->frozen_sb = freeze_bdev(md->bdev);
2512 	if (IS_ERR(md->frozen_sb)) {
2513 		r = PTR_ERR(md->frozen_sb);
2514 		md->frozen_sb = NULL;
2515 		return r;
2516 	}
2517 
2518 	set_bit(DMF_FROZEN, &md->flags);
2519 
2520 	return 0;
2521 }
2522 
2523 static void unlock_fs(struct mapped_device *md)
2524 {
2525 	if (!test_bit(DMF_FROZEN, &md->flags))
2526 		return;
2527 
2528 	thaw_bdev(md->bdev, md->frozen_sb);
2529 	md->frozen_sb = NULL;
2530 	clear_bit(DMF_FROZEN, &md->flags);
2531 }
2532 
2533 /*
2534  * We need to be able to change a mapping table under a mounted
2535  * filesystem.  For example we might want to move some data in
2536  * the background.  Before the table can be swapped with
2537  * dm_bind_table, dm_suspend must be called to flush any in
2538  * flight bios and ensure that any further io gets deferred.
2539  */
2540 /*
2541  * Suspend mechanism in request-based dm.
2542  *
2543  * 1. Flush all I/Os by lock_fs() if needed.
2544  * 2. Stop dispatching any I/O by stopping the request_queue.
2545  * 3. Wait for all in-flight I/Os to be completed or requeued.
2546  *
2547  * To abort suspend, start the request_queue.
2548  */
2549 int dm_suspend(struct mapped_device *md, unsigned suspend_flags)
2550 {
2551 	struct dm_table *map = NULL;
2552 	int r = 0;
2553 	int do_lockfs = suspend_flags & DM_SUSPEND_LOCKFS_FLAG ? 1 : 0;
2554 	int noflush = suspend_flags & DM_SUSPEND_NOFLUSH_FLAG ? 1 : 0;
2555 
2556 	mutex_lock(&md->suspend_lock);
2557 
2558 	if (dm_suspended_md(md)) {
2559 		r = -EINVAL;
2560 		goto out_unlock;
2561 	}
2562 
2563 	map = md->map;
2564 
2565 	/*
2566 	 * DMF_NOFLUSH_SUSPENDING must be set before presuspend.
2567 	 * This flag is cleared before dm_suspend returns.
2568 	 */
2569 	if (noflush)
2570 		set_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
2571 
2572 	/* This does not get reverted if there's an error later. */
2573 	dm_table_presuspend_targets(map);
2574 
2575 	/*
2576 	 * Flush I/O to the device.
2577 	 * Any I/O submitted after lock_fs() may not be flushed.
2578 	 * noflush takes precedence over do_lockfs.
2579 	 * (lock_fs() flushes I/Os and waits for them to complete.)
2580 	 */
2581 	if (!noflush && do_lockfs) {
2582 		r = lock_fs(md);
2583 		if (r)
2584 			goto out_unlock;
2585 	}
2586 
2587 	/*
2588 	 * Here we must make sure that no processes are submitting requests
2589 	 * to target drivers i.e. no one may be executing
2590 	 * __split_and_process_bio. This is called from dm_request and
2591 	 * dm_wq_work.
2592 	 *
2593 	 * To get all processes out of __split_and_process_bio in dm_request,
2594 	 * we take the write lock. To prevent any process from reentering
2595 	 * __split_and_process_bio from dm_request and quiesce the thread
2596 	 * (dm_wq_work), we set BMF_BLOCK_IO_FOR_SUSPEND and call
2597 	 * flush_workqueue(md->wq).
2598 	 */
2599 	set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2600 	synchronize_srcu(&md->io_barrier);
2601 
2602 	/*
2603 	 * Stop md->queue before flushing md->wq in case request-based
2604 	 * dm defers requests to md->wq from md->queue.
2605 	 */
2606 	if (dm_request_based(md))
2607 		stop_queue(md->queue);
2608 
2609 	flush_workqueue(md->wq);
2610 
2611 	/*
2612 	 * At this point no more requests are entering target request routines.
2613 	 * We call dm_wait_for_completion to wait for all existing requests
2614 	 * to finish.
2615 	 */
2616 	r = dm_wait_for_completion(md, TASK_INTERRUPTIBLE);
2617 
2618 	if (noflush)
2619 		clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
2620 	synchronize_srcu(&md->io_barrier);
2621 
2622 	/* were we interrupted ? */
2623 	if (r < 0) {
2624 		dm_queue_flush(md);
2625 
2626 		if (dm_request_based(md))
2627 			start_queue(md->queue);
2628 
2629 		unlock_fs(md);
2630 		goto out_unlock; /* pushback list is already flushed, so skip flush */
2631 	}
2632 
2633 	/*
2634 	 * If dm_wait_for_completion returned 0, the device is completely
2635 	 * quiescent now. There is no request-processing activity. All new
2636 	 * requests are being added to md->deferred list.
2637 	 */
2638 
2639 	set_bit(DMF_SUSPENDED, &md->flags);
2640 
2641 	dm_table_postsuspend_targets(map);
2642 
2643 out_unlock:
2644 	mutex_unlock(&md->suspend_lock);
2645 	return r;
2646 }
2647 
2648 int dm_resume(struct mapped_device *md)
2649 {
2650 	int r = -EINVAL;
2651 	struct dm_table *map = NULL;
2652 
2653 	mutex_lock(&md->suspend_lock);
2654 	if (!dm_suspended_md(md))
2655 		goto out;
2656 
2657 	map = md->map;
2658 	if (!map || !dm_table_get_size(map))
2659 		goto out;
2660 
2661 	r = dm_table_resume_targets(map);
2662 	if (r)
2663 		goto out;
2664 
2665 	dm_queue_flush(md);
2666 
2667 	/*
2668 	 * Flushing deferred I/Os must be done after targets are resumed
2669 	 * so that mapping of targets can work correctly.
2670 	 * Request-based dm is queueing the deferred I/Os in its request_queue.
2671 	 */
2672 	if (dm_request_based(md))
2673 		start_queue(md->queue);
2674 
2675 	unlock_fs(md);
2676 
2677 	clear_bit(DMF_SUSPENDED, &md->flags);
2678 
2679 	r = 0;
2680 out:
2681 	mutex_unlock(&md->suspend_lock);
2682 
2683 	return r;
2684 }
2685 
2686 /*
2687  * Internal suspend/resume works like userspace-driven suspend. It waits
2688  * until all bios finish and prevents issuing new bios to the target drivers.
2689  * It may be used only from the kernel.
2690  *
2691  * Internal suspend holds md->suspend_lock, which prevents interaction with
2692  * userspace-driven suspend.
2693  */
2694 
2695 void dm_internal_suspend(struct mapped_device *md)
2696 {
2697 	mutex_lock(&md->suspend_lock);
2698 	if (dm_suspended_md(md))
2699 		return;
2700 
2701 	set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2702 	synchronize_srcu(&md->io_barrier);
2703 	flush_workqueue(md->wq);
2704 	dm_wait_for_completion(md, TASK_UNINTERRUPTIBLE);
2705 }
2706 
2707 void dm_internal_resume(struct mapped_device *md)
2708 {
2709 	if (dm_suspended_md(md))
2710 		goto done;
2711 
2712 	dm_queue_flush(md);
2713 
2714 done:
2715 	mutex_unlock(&md->suspend_lock);
2716 }
2717 
2718 /*-----------------------------------------------------------------
2719  * Event notification.
2720  *---------------------------------------------------------------*/
2721 int dm_kobject_uevent(struct mapped_device *md, enum kobject_action action,
2722 		       unsigned cookie)
2723 {
2724 	char udev_cookie[DM_COOKIE_LENGTH];
2725 	char *envp[] = { udev_cookie, NULL };
2726 
2727 	if (!cookie)
2728 		return kobject_uevent(&disk_to_dev(md->disk)->kobj, action);
2729 	else {
2730 		snprintf(udev_cookie, DM_COOKIE_LENGTH, "%s=%u",
2731 			 DM_COOKIE_ENV_VAR_NAME, cookie);
2732 		return kobject_uevent_env(&disk_to_dev(md->disk)->kobj,
2733 					  action, envp);
2734 	}
2735 }
2736 
2737 uint32_t dm_next_uevent_seq(struct mapped_device *md)
2738 {
2739 	return atomic_add_return(1, &md->uevent_seq);
2740 }
2741 
2742 uint32_t dm_get_event_nr(struct mapped_device *md)
2743 {
2744 	return atomic_read(&md->event_nr);
2745 }
2746 
2747 int dm_wait_event(struct mapped_device *md, int event_nr)
2748 {
2749 	return wait_event_interruptible(md->eventq,
2750 			(event_nr != atomic_read(&md->event_nr)));
2751 }
2752 
2753 void dm_uevent_add(struct mapped_device *md, struct list_head *elist)
2754 {
2755 	unsigned long flags;
2756 
2757 	spin_lock_irqsave(&md->uevent_lock, flags);
2758 	list_add(elist, &md->uevent_list);
2759 	spin_unlock_irqrestore(&md->uevent_lock, flags);
2760 }
2761 
2762 /*
2763  * The gendisk is only valid as long as you have a reference
2764  * count on 'md'.
2765  */
2766 struct gendisk *dm_disk(struct mapped_device *md)
2767 {
2768 	return md->disk;
2769 }
2770 
2771 struct kobject *dm_kobject(struct mapped_device *md)
2772 {
2773 	return &md->kobj_holder.kobj;
2774 }
2775 
2776 struct mapped_device *dm_get_from_kobject(struct kobject *kobj)
2777 {
2778 	struct mapped_device *md;
2779 
2780 	md = container_of(kobj, struct mapped_device, kobj_holder.kobj);
2781 
2782 	if (test_bit(DMF_FREEING, &md->flags) ||
2783 	    dm_deleting_md(md))
2784 		return NULL;
2785 
2786 	dm_get(md);
2787 	return md;
2788 }
2789 
2790 int dm_suspended_md(struct mapped_device *md)
2791 {
2792 	return test_bit(DMF_SUSPENDED, &md->flags);
2793 }
2794 
2795 int dm_test_deferred_remove_flag(struct mapped_device *md)
2796 {
2797 	return test_bit(DMF_DEFERRED_REMOVE, &md->flags);
2798 }
2799 
2800 int dm_suspended(struct dm_target *ti)
2801 {
2802 	return dm_suspended_md(dm_table_get_md(ti->table));
2803 }
2804 EXPORT_SYMBOL_GPL(dm_suspended);
2805 
2806 int dm_noflush_suspending(struct dm_target *ti)
2807 {
2808 	return __noflush_suspending(dm_table_get_md(ti->table));
2809 }
2810 EXPORT_SYMBOL_GPL(dm_noflush_suspending);
2811 
2812 struct dm_md_mempools *dm_alloc_md_mempools(unsigned type, unsigned integrity, unsigned per_bio_data_size)
2813 {
2814 	struct dm_md_mempools *pools = kzalloc(sizeof(*pools), GFP_KERNEL);
2815 	struct kmem_cache *cachep;
2816 	unsigned int pool_size;
2817 	unsigned int front_pad;
2818 
2819 	if (!pools)
2820 		return NULL;
2821 
2822 	if (type == DM_TYPE_BIO_BASED) {
2823 		cachep = _io_cache;
2824 		pool_size = dm_get_reserved_bio_based_ios();
2825 		front_pad = roundup(per_bio_data_size, __alignof__(struct dm_target_io)) + offsetof(struct dm_target_io, clone);
2826 	} else if (type == DM_TYPE_REQUEST_BASED) {
2827 		cachep = _rq_tio_cache;
2828 		pool_size = dm_get_reserved_rq_based_ios();
2829 		front_pad = offsetof(struct dm_rq_clone_bio_info, clone);
2830 		/* per_bio_data_size is not used. See __bind_mempools(). */
2831 		WARN_ON(per_bio_data_size != 0);
2832 	} else
2833 		goto out;
2834 
2835 	pools->io_pool = mempool_create_slab_pool(pool_size, cachep);
2836 	if (!pools->io_pool)
2837 		goto out;
2838 
2839 	pools->bs = bioset_create(pool_size, front_pad);
2840 	if (!pools->bs)
2841 		goto out;
2842 
2843 	if (integrity && bioset_integrity_create(pools->bs, pool_size))
2844 		goto out;
2845 
2846 	return pools;
2847 
2848 out:
2849 	dm_free_md_mempools(pools);
2850 
2851 	return NULL;
2852 }
2853 
2854 void dm_free_md_mempools(struct dm_md_mempools *pools)
2855 {
2856 	if (!pools)
2857 		return;
2858 
2859 	if (pools->io_pool)
2860 		mempool_destroy(pools->io_pool);
2861 
2862 	if (pools->bs)
2863 		bioset_free(pools->bs);
2864 
2865 	kfree(pools);
2866 }
2867 
2868 static const struct block_device_operations dm_blk_dops = {
2869 	.open = dm_blk_open,
2870 	.release = dm_blk_close,
2871 	.ioctl = dm_blk_ioctl,
2872 	.getgeo = dm_blk_getgeo,
2873 	.owner = THIS_MODULE
2874 };
2875 
2876 EXPORT_SYMBOL(dm_get_mapinfo);
2877 
2878 /*
2879  * module hooks
2880  */
2881 module_init(dm_init);
2882 module_exit(dm_exit);
2883 
2884 module_param(major, uint, 0);
2885 MODULE_PARM_DESC(major, "The major number of the device mapper");
2886 
2887 module_param(reserved_bio_based_ios, uint, S_IRUGO | S_IWUSR);
2888 MODULE_PARM_DESC(reserved_bio_based_ios, "Reserved IOs in bio-based mempools");
2889 
2890 module_param(reserved_rq_based_ios, uint, S_IRUGO | S_IWUSR);
2891 MODULE_PARM_DESC(reserved_rq_based_ios, "Reserved IOs in request-based mempools");
2892 
2893 MODULE_DESCRIPTION(DM_NAME " driver");
2894 MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
2895 MODULE_LICENSE("GPL");
2896