xref: /openbmc/linux/drivers/md/dm.c (revision 9d1deb83d489364f8749a3a1ba1689efb07d94b0)
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 #include <linux/wait.h>
23 #include <linux/kthread.h>
24 
25 #include <trace/events/block.h>
26 
27 #define DM_MSG_PREFIX "core"
28 
29 #ifdef CONFIG_PRINTK
30 /*
31  * ratelimit state to be used in DMXXX_LIMIT().
32  */
33 DEFINE_RATELIMIT_STATE(dm_ratelimit_state,
34 		       DEFAULT_RATELIMIT_INTERVAL,
35 		       DEFAULT_RATELIMIT_BURST);
36 EXPORT_SYMBOL(dm_ratelimit_state);
37 #endif
38 
39 /*
40  * Cookies are numeric values sent with CHANGE and REMOVE
41  * uevents while resuming, removing or renaming the device.
42  */
43 #define DM_COOKIE_ENV_VAR_NAME "DM_COOKIE"
44 #define DM_COOKIE_LENGTH 24
45 
46 static const char *_name = DM_NAME;
47 
48 static unsigned int major = 0;
49 static unsigned int _major = 0;
50 
51 static DEFINE_IDR(_minor_idr);
52 
53 static DEFINE_SPINLOCK(_minor_lock);
54 
55 static void do_deferred_remove(struct work_struct *w);
56 
57 static DECLARE_WORK(deferred_remove_work, do_deferred_remove);
58 
59 static struct workqueue_struct *deferred_remove_workqueue;
60 
61 /*
62  * For bio-based dm.
63  * One of these is allocated per bio.
64  */
65 struct dm_io {
66 	struct mapped_device *md;
67 	int error;
68 	atomic_t io_count;
69 	struct bio *bio;
70 	unsigned long start_time;
71 	spinlock_t endio_lock;
72 	struct dm_stats_aux stats_aux;
73 };
74 
75 /*
76  * For request-based dm.
77  * One of these is allocated per request.
78  */
79 struct dm_rq_target_io {
80 	struct mapped_device *md;
81 	struct dm_target *ti;
82 	struct request *orig, *clone;
83 	struct kthread_work work;
84 	int error;
85 	union map_info info;
86 };
87 
88 /*
89  * For request-based dm - the bio clones we allocate are embedded in these
90  * structs.
91  *
92  * We allocate these with bio_alloc_bioset, using the front_pad parameter when
93  * the bioset is created - this means the bio has to come at the end of the
94  * struct.
95  */
96 struct dm_rq_clone_bio_info {
97 	struct bio *orig;
98 	struct dm_rq_target_io *tio;
99 	struct bio clone;
100 };
101 
102 union map_info *dm_get_rq_mapinfo(struct request *rq)
103 {
104 	if (rq && rq->end_io_data)
105 		return &((struct dm_rq_target_io *)rq->end_io_data)->info;
106 	return NULL;
107 }
108 EXPORT_SYMBOL_GPL(dm_get_rq_mapinfo);
109 
110 #define MINOR_ALLOCED ((void *)-1)
111 
112 /*
113  * Bits for the md->flags field.
114  */
115 #define DMF_BLOCK_IO_FOR_SUSPEND 0
116 #define DMF_SUSPENDED 1
117 #define DMF_FROZEN 2
118 #define DMF_FREEING 3
119 #define DMF_DELETING 4
120 #define DMF_NOFLUSH_SUSPENDING 5
121 #define DMF_MERGE_IS_OPTIONAL 6
122 #define DMF_DEFERRED_REMOVE 7
123 #define DMF_SUSPENDED_INTERNALLY 8
124 
125 /*
126  * A dummy definition to make RCU happy.
127  * struct dm_table should never be dereferenced in this file.
128  */
129 struct dm_table {
130 	int undefined__;
131 };
132 
133 /*
134  * Work processed by per-device workqueue.
135  */
136 struct mapped_device {
137 	struct srcu_struct io_barrier;
138 	struct mutex suspend_lock;
139 	atomic_t holders;
140 	atomic_t open_count;
141 
142 	/*
143 	 * The current mapping.
144 	 * Use dm_get_live_table{_fast} or take suspend_lock for
145 	 * dereference.
146 	 */
147 	struct dm_table __rcu *map;
148 
149 	struct list_head table_devices;
150 	struct mutex table_devices_lock;
151 
152 	unsigned long flags;
153 
154 	struct request_queue *queue;
155 	unsigned type;
156 	/* Protect queue and type against concurrent access. */
157 	struct mutex type_lock;
158 
159 	struct target_type *immutable_target_type;
160 
161 	struct gendisk *disk;
162 	char name[16];
163 
164 	void *interface_ptr;
165 
166 	/*
167 	 * A list of ios that arrived while we were suspended.
168 	 */
169 	atomic_t pending[2];
170 	wait_queue_head_t wait;
171 	struct work_struct work;
172 	struct bio_list deferred;
173 	spinlock_t deferred_lock;
174 
175 	/*
176 	 * Processing queue (flush)
177 	 */
178 	struct workqueue_struct *wq;
179 
180 	/*
181 	 * io objects are allocated from here.
182 	 */
183 	mempool_t *io_pool;
184 	mempool_t *rq_pool;
185 
186 	struct bio_set *bs;
187 
188 	/*
189 	 * Event handling.
190 	 */
191 	atomic_t event_nr;
192 	wait_queue_head_t eventq;
193 	atomic_t uevent_seq;
194 	struct list_head uevent_list;
195 	spinlock_t uevent_lock; /* Protect access to uevent_list */
196 
197 	/*
198 	 * freeze/thaw support require holding onto a super block
199 	 */
200 	struct super_block *frozen_sb;
201 	struct block_device *bdev;
202 
203 	/* forced geometry settings */
204 	struct hd_geometry geometry;
205 
206 	/* kobject and completion */
207 	struct dm_kobject_holder kobj_holder;
208 
209 	/* zero-length flush that will be cloned and submitted to targets */
210 	struct bio flush_bio;
211 
212 	/* the number of internal suspends */
213 	unsigned internal_suspend_count;
214 
215 	struct dm_stats stats;
216 
217 	struct kthread_worker kworker;
218 	struct task_struct *kworker_task;
219 };
220 
221 /*
222  * For mempools pre-allocation at the table loading time.
223  */
224 struct dm_md_mempools {
225 	mempool_t *io_pool;
226 	mempool_t *rq_pool;
227 	struct bio_set *bs;
228 };
229 
230 struct table_device {
231 	struct list_head list;
232 	atomic_t count;
233 	struct dm_dev dm_dev;
234 };
235 
236 #define RESERVED_BIO_BASED_IOS		16
237 #define RESERVED_REQUEST_BASED_IOS	256
238 #define RESERVED_MAX_IOS		1024
239 static struct kmem_cache *_io_cache;
240 static struct kmem_cache *_rq_tio_cache;
241 static struct kmem_cache *_rq_cache;
242 
243 /*
244  * Bio-based DM's mempools' reserved IOs set by the user.
245  */
246 static unsigned reserved_bio_based_ios = RESERVED_BIO_BASED_IOS;
247 
248 /*
249  * Request-based DM's mempools' reserved IOs set by the user.
250  */
251 static unsigned reserved_rq_based_ios = RESERVED_REQUEST_BASED_IOS;
252 
253 static unsigned __dm_get_module_param(unsigned *module_param,
254 				      unsigned def, unsigned max)
255 {
256 	unsigned param = ACCESS_ONCE(*module_param);
257 	unsigned modified_param = 0;
258 
259 	if (!param)
260 		modified_param = def;
261 	else if (param > max)
262 		modified_param = max;
263 
264 	if (modified_param) {
265 		(void)cmpxchg(module_param, param, modified_param);
266 		param = modified_param;
267 	}
268 
269 	return param;
270 }
271 
272 unsigned dm_get_reserved_bio_based_ios(void)
273 {
274 	return __dm_get_module_param(&reserved_bio_based_ios,
275 				     RESERVED_BIO_BASED_IOS, RESERVED_MAX_IOS);
276 }
277 EXPORT_SYMBOL_GPL(dm_get_reserved_bio_based_ios);
278 
279 unsigned dm_get_reserved_rq_based_ios(void)
280 {
281 	return __dm_get_module_param(&reserved_rq_based_ios,
282 				     RESERVED_REQUEST_BASED_IOS, RESERVED_MAX_IOS);
283 }
284 EXPORT_SYMBOL_GPL(dm_get_reserved_rq_based_ios);
285 
286 static int __init local_init(void)
287 {
288 	int r = -ENOMEM;
289 
290 	/* allocate a slab for the dm_ios */
291 	_io_cache = KMEM_CACHE(dm_io, 0);
292 	if (!_io_cache)
293 		return r;
294 
295 	_rq_tio_cache = KMEM_CACHE(dm_rq_target_io, 0);
296 	if (!_rq_tio_cache)
297 		goto out_free_io_cache;
298 
299 	_rq_cache = kmem_cache_create("dm_clone_request", sizeof(struct request),
300 				      __alignof__(struct request), 0, NULL);
301 	if (!_rq_cache)
302 		goto out_free_rq_tio_cache;
303 
304 	r = dm_uevent_init();
305 	if (r)
306 		goto out_free_rq_cache;
307 
308 	deferred_remove_workqueue = alloc_workqueue("kdmremove", WQ_UNBOUND, 1);
309 	if (!deferred_remove_workqueue) {
310 		r = -ENOMEM;
311 		goto out_uevent_exit;
312 	}
313 
314 	_major = major;
315 	r = register_blkdev(_major, _name);
316 	if (r < 0)
317 		goto out_free_workqueue;
318 
319 	if (!_major)
320 		_major = r;
321 
322 	return 0;
323 
324 out_free_workqueue:
325 	destroy_workqueue(deferred_remove_workqueue);
326 out_uevent_exit:
327 	dm_uevent_exit();
328 out_free_rq_cache:
329 	kmem_cache_destroy(_rq_cache);
330 out_free_rq_tio_cache:
331 	kmem_cache_destroy(_rq_tio_cache);
332 out_free_io_cache:
333 	kmem_cache_destroy(_io_cache);
334 
335 	return r;
336 }
337 
338 static void local_exit(void)
339 {
340 	flush_scheduled_work();
341 	destroy_workqueue(deferred_remove_workqueue);
342 
343 	kmem_cache_destroy(_rq_cache);
344 	kmem_cache_destroy(_rq_tio_cache);
345 	kmem_cache_destroy(_io_cache);
346 	unregister_blkdev(_major, _name);
347 	dm_uevent_exit();
348 
349 	_major = 0;
350 
351 	DMINFO("cleaned up");
352 }
353 
354 static int (*_inits[])(void) __initdata = {
355 	local_init,
356 	dm_target_init,
357 	dm_linear_init,
358 	dm_stripe_init,
359 	dm_io_init,
360 	dm_kcopyd_init,
361 	dm_interface_init,
362 	dm_statistics_init,
363 };
364 
365 static void (*_exits[])(void) = {
366 	local_exit,
367 	dm_target_exit,
368 	dm_linear_exit,
369 	dm_stripe_exit,
370 	dm_io_exit,
371 	dm_kcopyd_exit,
372 	dm_interface_exit,
373 	dm_statistics_exit,
374 };
375 
376 static int __init dm_init(void)
377 {
378 	const int count = ARRAY_SIZE(_inits);
379 
380 	int r, i;
381 
382 	for (i = 0; i < count; i++) {
383 		r = _inits[i]();
384 		if (r)
385 			goto bad;
386 	}
387 
388 	return 0;
389 
390       bad:
391 	while (i--)
392 		_exits[i]();
393 
394 	return r;
395 }
396 
397 static void __exit dm_exit(void)
398 {
399 	int i = ARRAY_SIZE(_exits);
400 
401 	while (i--)
402 		_exits[i]();
403 
404 	/*
405 	 * Should be empty by this point.
406 	 */
407 	idr_destroy(&_minor_idr);
408 }
409 
410 /*
411  * Block device functions
412  */
413 int dm_deleting_md(struct mapped_device *md)
414 {
415 	return test_bit(DMF_DELETING, &md->flags);
416 }
417 
418 static int dm_blk_open(struct block_device *bdev, fmode_t mode)
419 {
420 	struct mapped_device *md;
421 
422 	spin_lock(&_minor_lock);
423 
424 	md = bdev->bd_disk->private_data;
425 	if (!md)
426 		goto out;
427 
428 	if (test_bit(DMF_FREEING, &md->flags) ||
429 	    dm_deleting_md(md)) {
430 		md = NULL;
431 		goto out;
432 	}
433 
434 	dm_get(md);
435 	atomic_inc(&md->open_count);
436 out:
437 	spin_unlock(&_minor_lock);
438 
439 	return md ? 0 : -ENXIO;
440 }
441 
442 static void dm_blk_close(struct gendisk *disk, fmode_t mode)
443 {
444 	struct mapped_device *md;
445 
446 	spin_lock(&_minor_lock);
447 
448 	md = disk->private_data;
449 	if (WARN_ON(!md))
450 		goto out;
451 
452 	if (atomic_dec_and_test(&md->open_count) &&
453 	    (test_bit(DMF_DEFERRED_REMOVE, &md->flags)))
454 		queue_work(deferred_remove_workqueue, &deferred_remove_work);
455 
456 	dm_put(md);
457 out:
458 	spin_unlock(&_minor_lock);
459 }
460 
461 int dm_open_count(struct mapped_device *md)
462 {
463 	return atomic_read(&md->open_count);
464 }
465 
466 /*
467  * Guarantees nothing is using the device before it's deleted.
468  */
469 int dm_lock_for_deletion(struct mapped_device *md, bool mark_deferred, bool only_deferred)
470 {
471 	int r = 0;
472 
473 	spin_lock(&_minor_lock);
474 
475 	if (dm_open_count(md)) {
476 		r = -EBUSY;
477 		if (mark_deferred)
478 			set_bit(DMF_DEFERRED_REMOVE, &md->flags);
479 	} else if (only_deferred && !test_bit(DMF_DEFERRED_REMOVE, &md->flags))
480 		r = -EEXIST;
481 	else
482 		set_bit(DMF_DELETING, &md->flags);
483 
484 	spin_unlock(&_minor_lock);
485 
486 	return r;
487 }
488 
489 int dm_cancel_deferred_remove(struct mapped_device *md)
490 {
491 	int r = 0;
492 
493 	spin_lock(&_minor_lock);
494 
495 	if (test_bit(DMF_DELETING, &md->flags))
496 		r = -EBUSY;
497 	else
498 		clear_bit(DMF_DEFERRED_REMOVE, &md->flags);
499 
500 	spin_unlock(&_minor_lock);
501 
502 	return r;
503 }
504 
505 static void do_deferred_remove(struct work_struct *w)
506 {
507 	dm_deferred_remove();
508 }
509 
510 sector_t dm_get_size(struct mapped_device *md)
511 {
512 	return get_capacity(md->disk);
513 }
514 
515 struct request_queue *dm_get_md_queue(struct mapped_device *md)
516 {
517 	return md->queue;
518 }
519 
520 struct dm_stats *dm_get_stats(struct mapped_device *md)
521 {
522 	return &md->stats;
523 }
524 
525 static int dm_blk_getgeo(struct block_device *bdev, struct hd_geometry *geo)
526 {
527 	struct mapped_device *md = bdev->bd_disk->private_data;
528 
529 	return dm_get_geometry(md, geo);
530 }
531 
532 static int dm_blk_ioctl(struct block_device *bdev, fmode_t mode,
533 			unsigned int cmd, unsigned long arg)
534 {
535 	struct mapped_device *md = bdev->bd_disk->private_data;
536 	int srcu_idx;
537 	struct dm_table *map;
538 	struct dm_target *tgt;
539 	int r = -ENOTTY;
540 
541 retry:
542 	map = dm_get_live_table(md, &srcu_idx);
543 
544 	if (!map || !dm_table_get_size(map))
545 		goto out;
546 
547 	/* We only support devices that have a single target */
548 	if (dm_table_get_num_targets(map) != 1)
549 		goto out;
550 
551 	tgt = dm_table_get_target(map, 0);
552 	if (!tgt->type->ioctl)
553 		goto out;
554 
555 	if (dm_suspended_md(md)) {
556 		r = -EAGAIN;
557 		goto out;
558 	}
559 
560 	r = tgt->type->ioctl(tgt, cmd, arg);
561 
562 out:
563 	dm_put_live_table(md, srcu_idx);
564 
565 	if (r == -ENOTCONN) {
566 		msleep(10);
567 		goto retry;
568 	}
569 
570 	return r;
571 }
572 
573 static struct dm_io *alloc_io(struct mapped_device *md)
574 {
575 	return mempool_alloc(md->io_pool, GFP_NOIO);
576 }
577 
578 static void free_io(struct mapped_device *md, struct dm_io *io)
579 {
580 	mempool_free(io, md->io_pool);
581 }
582 
583 static void free_tio(struct mapped_device *md, struct dm_target_io *tio)
584 {
585 	bio_put(&tio->clone);
586 }
587 
588 static struct dm_rq_target_io *alloc_rq_tio(struct mapped_device *md,
589 					    gfp_t gfp_mask)
590 {
591 	return mempool_alloc(md->io_pool, gfp_mask);
592 }
593 
594 static void free_rq_tio(struct dm_rq_target_io *tio)
595 {
596 	mempool_free(tio, tio->md->io_pool);
597 }
598 
599 static struct request *alloc_clone_request(struct mapped_device *md,
600 					   gfp_t gfp_mask)
601 {
602 	return mempool_alloc(md->rq_pool, gfp_mask);
603 }
604 
605 static void free_clone_request(struct mapped_device *md, struct request *rq)
606 {
607 	mempool_free(rq, md->rq_pool);
608 }
609 
610 static int md_in_flight(struct mapped_device *md)
611 {
612 	return atomic_read(&md->pending[READ]) +
613 	       atomic_read(&md->pending[WRITE]);
614 }
615 
616 static void start_io_acct(struct dm_io *io)
617 {
618 	struct mapped_device *md = io->md;
619 	struct bio *bio = io->bio;
620 	int cpu;
621 	int rw = bio_data_dir(bio);
622 
623 	io->start_time = jiffies;
624 
625 	cpu = part_stat_lock();
626 	part_round_stats(cpu, &dm_disk(md)->part0);
627 	part_stat_unlock();
628 	atomic_set(&dm_disk(md)->part0.in_flight[rw],
629 		atomic_inc_return(&md->pending[rw]));
630 
631 	if (unlikely(dm_stats_used(&md->stats)))
632 		dm_stats_account_io(&md->stats, bio->bi_rw, bio->bi_iter.bi_sector,
633 				    bio_sectors(bio), false, 0, &io->stats_aux);
634 }
635 
636 static void end_io_acct(struct dm_io *io)
637 {
638 	struct mapped_device *md = io->md;
639 	struct bio *bio = io->bio;
640 	unsigned long duration = jiffies - io->start_time;
641 	int pending;
642 	int rw = bio_data_dir(bio);
643 
644 	generic_end_io_acct(rw, &dm_disk(md)->part0, io->start_time);
645 
646 	if (unlikely(dm_stats_used(&md->stats)))
647 		dm_stats_account_io(&md->stats, bio->bi_rw, bio->bi_iter.bi_sector,
648 				    bio_sectors(bio), true, duration, &io->stats_aux);
649 
650 	/*
651 	 * After this is decremented the bio must not be touched if it is
652 	 * a flush.
653 	 */
654 	pending = atomic_dec_return(&md->pending[rw]);
655 	atomic_set(&dm_disk(md)->part0.in_flight[rw], pending);
656 	pending += atomic_read(&md->pending[rw^0x1]);
657 
658 	/* nudge anyone waiting on suspend queue */
659 	if (!pending)
660 		wake_up(&md->wait);
661 }
662 
663 /*
664  * Add the bio to the list of deferred io.
665  */
666 static void queue_io(struct mapped_device *md, struct bio *bio)
667 {
668 	unsigned long flags;
669 
670 	spin_lock_irqsave(&md->deferred_lock, flags);
671 	bio_list_add(&md->deferred, bio);
672 	spin_unlock_irqrestore(&md->deferred_lock, flags);
673 	queue_work(md->wq, &md->work);
674 }
675 
676 /*
677  * Everyone (including functions in this file), should use this
678  * function to access the md->map field, and make sure they call
679  * dm_put_live_table() when finished.
680  */
681 struct dm_table *dm_get_live_table(struct mapped_device *md, int *srcu_idx) __acquires(md->io_barrier)
682 {
683 	*srcu_idx = srcu_read_lock(&md->io_barrier);
684 
685 	return srcu_dereference(md->map, &md->io_barrier);
686 }
687 
688 void dm_put_live_table(struct mapped_device *md, int srcu_idx) __releases(md->io_barrier)
689 {
690 	srcu_read_unlock(&md->io_barrier, srcu_idx);
691 }
692 
693 void dm_sync_table(struct mapped_device *md)
694 {
695 	synchronize_srcu(&md->io_barrier);
696 	synchronize_rcu_expedited();
697 }
698 
699 /*
700  * A fast alternative to dm_get_live_table/dm_put_live_table.
701  * The caller must not block between these two functions.
702  */
703 static struct dm_table *dm_get_live_table_fast(struct mapped_device *md) __acquires(RCU)
704 {
705 	rcu_read_lock();
706 	return rcu_dereference(md->map);
707 }
708 
709 static void dm_put_live_table_fast(struct mapped_device *md) __releases(RCU)
710 {
711 	rcu_read_unlock();
712 }
713 
714 /*
715  * Open a table device so we can use it as a map destination.
716  */
717 static int open_table_device(struct table_device *td, dev_t dev,
718 			     struct mapped_device *md)
719 {
720 	static char *_claim_ptr = "I belong to device-mapper";
721 	struct block_device *bdev;
722 
723 	int r;
724 
725 	BUG_ON(td->dm_dev.bdev);
726 
727 	bdev = blkdev_get_by_dev(dev, td->dm_dev.mode | FMODE_EXCL, _claim_ptr);
728 	if (IS_ERR(bdev))
729 		return PTR_ERR(bdev);
730 
731 	r = bd_link_disk_holder(bdev, dm_disk(md));
732 	if (r) {
733 		blkdev_put(bdev, td->dm_dev.mode | FMODE_EXCL);
734 		return r;
735 	}
736 
737 	td->dm_dev.bdev = bdev;
738 	return 0;
739 }
740 
741 /*
742  * Close a table device that we've been using.
743  */
744 static void close_table_device(struct table_device *td, struct mapped_device *md)
745 {
746 	if (!td->dm_dev.bdev)
747 		return;
748 
749 	bd_unlink_disk_holder(td->dm_dev.bdev, dm_disk(md));
750 	blkdev_put(td->dm_dev.bdev, td->dm_dev.mode | FMODE_EXCL);
751 	td->dm_dev.bdev = NULL;
752 }
753 
754 static struct table_device *find_table_device(struct list_head *l, dev_t dev,
755 					      fmode_t mode) {
756 	struct table_device *td;
757 
758 	list_for_each_entry(td, l, list)
759 		if (td->dm_dev.bdev->bd_dev == dev && td->dm_dev.mode == mode)
760 			return td;
761 
762 	return NULL;
763 }
764 
765 int dm_get_table_device(struct mapped_device *md, dev_t dev, fmode_t mode,
766 			struct dm_dev **result) {
767 	int r;
768 	struct table_device *td;
769 
770 	mutex_lock(&md->table_devices_lock);
771 	td = find_table_device(&md->table_devices, dev, mode);
772 	if (!td) {
773 		td = kmalloc(sizeof(*td), GFP_KERNEL);
774 		if (!td) {
775 			mutex_unlock(&md->table_devices_lock);
776 			return -ENOMEM;
777 		}
778 
779 		td->dm_dev.mode = mode;
780 		td->dm_dev.bdev = NULL;
781 
782 		if ((r = open_table_device(td, dev, md))) {
783 			mutex_unlock(&md->table_devices_lock);
784 			kfree(td);
785 			return r;
786 		}
787 
788 		format_dev_t(td->dm_dev.name, dev);
789 
790 		atomic_set(&td->count, 0);
791 		list_add(&td->list, &md->table_devices);
792 	}
793 	atomic_inc(&td->count);
794 	mutex_unlock(&md->table_devices_lock);
795 
796 	*result = &td->dm_dev;
797 	return 0;
798 }
799 EXPORT_SYMBOL_GPL(dm_get_table_device);
800 
801 void dm_put_table_device(struct mapped_device *md, struct dm_dev *d)
802 {
803 	struct table_device *td = container_of(d, struct table_device, dm_dev);
804 
805 	mutex_lock(&md->table_devices_lock);
806 	if (atomic_dec_and_test(&td->count)) {
807 		close_table_device(td, md);
808 		list_del(&td->list);
809 		kfree(td);
810 	}
811 	mutex_unlock(&md->table_devices_lock);
812 }
813 EXPORT_SYMBOL(dm_put_table_device);
814 
815 static void free_table_devices(struct list_head *devices)
816 {
817 	struct list_head *tmp, *next;
818 
819 	list_for_each_safe(tmp, next, devices) {
820 		struct table_device *td = list_entry(tmp, struct table_device, list);
821 
822 		DMWARN("dm_destroy: %s still exists with %d references",
823 		       td->dm_dev.name, atomic_read(&td->count));
824 		kfree(td);
825 	}
826 }
827 
828 /*
829  * Get the geometry associated with a dm device
830  */
831 int dm_get_geometry(struct mapped_device *md, struct hd_geometry *geo)
832 {
833 	*geo = md->geometry;
834 
835 	return 0;
836 }
837 
838 /*
839  * Set the geometry of a device.
840  */
841 int dm_set_geometry(struct mapped_device *md, struct hd_geometry *geo)
842 {
843 	sector_t sz = (sector_t)geo->cylinders * geo->heads * geo->sectors;
844 
845 	if (geo->start > sz) {
846 		DMWARN("Start sector is beyond the geometry limits.");
847 		return -EINVAL;
848 	}
849 
850 	md->geometry = *geo;
851 
852 	return 0;
853 }
854 
855 /*-----------------------------------------------------------------
856  * CRUD START:
857  *   A more elegant soln is in the works that uses the queue
858  *   merge fn, unfortunately there are a couple of changes to
859  *   the block layer that I want to make for this.  So in the
860  *   interests of getting something for people to use I give
861  *   you this clearly demarcated crap.
862  *---------------------------------------------------------------*/
863 
864 static int __noflush_suspending(struct mapped_device *md)
865 {
866 	return test_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
867 }
868 
869 /*
870  * Decrements the number of outstanding ios that a bio has been
871  * cloned into, completing the original io if necc.
872  */
873 static void dec_pending(struct dm_io *io, int error)
874 {
875 	unsigned long flags;
876 	int io_error;
877 	struct bio *bio;
878 	struct mapped_device *md = io->md;
879 
880 	/* Push-back supersedes any I/O errors */
881 	if (unlikely(error)) {
882 		spin_lock_irqsave(&io->endio_lock, flags);
883 		if (!(io->error > 0 && __noflush_suspending(md)))
884 			io->error = error;
885 		spin_unlock_irqrestore(&io->endio_lock, flags);
886 	}
887 
888 	if (atomic_dec_and_test(&io->io_count)) {
889 		if (io->error == DM_ENDIO_REQUEUE) {
890 			/*
891 			 * Target requested pushing back the I/O.
892 			 */
893 			spin_lock_irqsave(&md->deferred_lock, flags);
894 			if (__noflush_suspending(md))
895 				bio_list_add_head(&md->deferred, io->bio);
896 			else
897 				/* noflush suspend was interrupted. */
898 				io->error = -EIO;
899 			spin_unlock_irqrestore(&md->deferred_lock, flags);
900 		}
901 
902 		io_error = io->error;
903 		bio = io->bio;
904 		end_io_acct(io);
905 		free_io(md, io);
906 
907 		if (io_error == DM_ENDIO_REQUEUE)
908 			return;
909 
910 		if ((bio->bi_rw & REQ_FLUSH) && bio->bi_iter.bi_size) {
911 			/*
912 			 * Preflush done for flush with data, reissue
913 			 * without REQ_FLUSH.
914 			 */
915 			bio->bi_rw &= ~REQ_FLUSH;
916 			queue_io(md, bio);
917 		} else {
918 			/* done with normal IO or empty flush */
919 			trace_block_bio_complete(md->queue, bio, io_error);
920 			bio_endio(bio, io_error);
921 		}
922 	}
923 }
924 
925 static void disable_write_same(struct mapped_device *md)
926 {
927 	struct queue_limits *limits = dm_get_queue_limits(md);
928 
929 	/* device doesn't really support WRITE SAME, disable it */
930 	limits->max_write_same_sectors = 0;
931 }
932 
933 static void clone_endio(struct bio *bio, int error)
934 {
935 	int r = error;
936 	struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone);
937 	struct dm_io *io = tio->io;
938 	struct mapped_device *md = tio->io->md;
939 	dm_endio_fn endio = tio->ti->type->end_io;
940 
941 	if (!bio_flagged(bio, BIO_UPTODATE) && !error)
942 		error = -EIO;
943 
944 	if (endio) {
945 		r = endio(tio->ti, bio, error);
946 		if (r < 0 || r == DM_ENDIO_REQUEUE)
947 			/*
948 			 * error and requeue request are handled
949 			 * in dec_pending().
950 			 */
951 			error = r;
952 		else if (r == DM_ENDIO_INCOMPLETE)
953 			/* The target will handle the io */
954 			return;
955 		else if (r) {
956 			DMWARN("unimplemented target endio return value: %d", r);
957 			BUG();
958 		}
959 	}
960 
961 	if (unlikely(r == -EREMOTEIO && (bio->bi_rw & REQ_WRITE_SAME) &&
962 		     !bdev_get_queue(bio->bi_bdev)->limits.max_write_same_sectors))
963 		disable_write_same(md);
964 
965 	free_tio(md, tio);
966 	dec_pending(io, error);
967 }
968 
969 /*
970  * Partial completion handling for request-based dm
971  */
972 static void end_clone_bio(struct bio *clone, int error)
973 {
974 	struct dm_rq_clone_bio_info *info =
975 		container_of(clone, struct dm_rq_clone_bio_info, clone);
976 	struct dm_rq_target_io *tio = info->tio;
977 	struct bio *bio = info->orig;
978 	unsigned int nr_bytes = info->orig->bi_iter.bi_size;
979 
980 	bio_put(clone);
981 
982 	if (tio->error)
983 		/*
984 		 * An error has already been detected on the request.
985 		 * Once error occurred, just let clone->end_io() handle
986 		 * the remainder.
987 		 */
988 		return;
989 	else if (error) {
990 		/*
991 		 * Don't notice the error to the upper layer yet.
992 		 * The error handling decision is made by the target driver,
993 		 * when the request is completed.
994 		 */
995 		tio->error = error;
996 		return;
997 	}
998 
999 	/*
1000 	 * I/O for the bio successfully completed.
1001 	 * Notice the data completion to the upper layer.
1002 	 */
1003 
1004 	/*
1005 	 * bios are processed from the head of the list.
1006 	 * So the completing bio should always be rq->bio.
1007 	 * If it's not, something wrong is happening.
1008 	 */
1009 	if (tio->orig->bio != bio)
1010 		DMERR("bio completion is going in the middle of the request");
1011 
1012 	/*
1013 	 * Update the original request.
1014 	 * Do not use blk_end_request() here, because it may complete
1015 	 * the original request before the clone, and break the ordering.
1016 	 */
1017 	blk_update_request(tio->orig, 0, nr_bytes);
1018 }
1019 
1020 /*
1021  * Don't touch any member of the md after calling this function because
1022  * the md may be freed in dm_put() at the end of this function.
1023  * Or do dm_get() before calling this function and dm_put() later.
1024  */
1025 static void rq_completed(struct mapped_device *md, int rw, bool run_queue)
1026 {
1027 	int nr_requests_pending;
1028 
1029 	atomic_dec(&md->pending[rw]);
1030 
1031 	/* nudge anyone waiting on suspend queue */
1032 	nr_requests_pending = md_in_flight(md);
1033 	if (!nr_requests_pending)
1034 		wake_up(&md->wait);
1035 
1036 	/*
1037 	 * Run this off this callpath, as drivers could invoke end_io while
1038 	 * inside their request_fn (and holding the queue lock). Calling
1039 	 * back into ->request_fn() could deadlock attempting to grab the
1040 	 * queue lock again.
1041 	 */
1042 	if (run_queue) {
1043 		if (!nr_requests_pending ||
1044 		    (nr_requests_pending >= md->queue->nr_congestion_on))
1045 			blk_run_queue_async(md->queue);
1046 	}
1047 
1048 	/*
1049 	 * dm_put() must be at the end of this function. See the comment above
1050 	 */
1051 	dm_put(md);
1052 }
1053 
1054 static void free_rq_clone(struct request *clone)
1055 {
1056 	struct dm_rq_target_io *tio = clone->end_io_data;
1057 
1058 	blk_rq_unprep_clone(clone);
1059 	if (clone->q && clone->q->mq_ops)
1060 		tio->ti->type->release_clone_rq(clone);
1061 	else
1062 		free_clone_request(tio->md, clone);
1063 	free_rq_tio(tio);
1064 }
1065 
1066 /*
1067  * Complete the clone and the original request.
1068  * Must be called without clone's queue lock held,
1069  * see end_clone_request() for more details.
1070  */
1071 static void dm_end_request(struct request *clone, int error)
1072 {
1073 	int rw = rq_data_dir(clone);
1074 	struct dm_rq_target_io *tio = clone->end_io_data;
1075 	struct mapped_device *md = tio->md;
1076 	struct request *rq = tio->orig;
1077 
1078 	if (rq->cmd_type == REQ_TYPE_BLOCK_PC) {
1079 		rq->errors = clone->errors;
1080 		rq->resid_len = clone->resid_len;
1081 
1082 		if (rq->sense)
1083 			/*
1084 			 * We are using the sense buffer of the original
1085 			 * request.
1086 			 * So setting the length of the sense data is enough.
1087 			 */
1088 			rq->sense_len = clone->sense_len;
1089 	}
1090 
1091 	free_rq_clone(clone);
1092 	blk_end_request_all(rq, error);
1093 	rq_completed(md, rw, true);
1094 }
1095 
1096 static void dm_unprep_request(struct request *rq)
1097 {
1098 	struct dm_rq_target_io *tio = rq->special;
1099 	struct request *clone = tio->clone;
1100 
1101 	rq->special = NULL;
1102 	rq->cmd_flags &= ~REQ_DONTPREP;
1103 
1104 	if (clone)
1105 		free_rq_clone(clone);
1106 }
1107 
1108 /*
1109  * Requeue the original request of a clone.
1110  */
1111 static void dm_requeue_unmapped_original_request(struct mapped_device *md,
1112 						 struct request *rq)
1113 {
1114 	int rw = rq_data_dir(rq);
1115 	struct request_queue *q = rq->q;
1116 	unsigned long flags;
1117 
1118 	dm_unprep_request(rq);
1119 
1120 	spin_lock_irqsave(q->queue_lock, flags);
1121 	blk_requeue_request(q, rq);
1122 	spin_unlock_irqrestore(q->queue_lock, flags);
1123 
1124 	rq_completed(md, rw, false);
1125 }
1126 
1127 static void dm_requeue_unmapped_request(struct request *clone)
1128 {
1129 	struct dm_rq_target_io *tio = clone->end_io_data;
1130 
1131 	dm_requeue_unmapped_original_request(tio->md, tio->orig);
1132 }
1133 
1134 static void __stop_queue(struct request_queue *q)
1135 {
1136 	blk_stop_queue(q);
1137 }
1138 
1139 static void stop_queue(struct request_queue *q)
1140 {
1141 	unsigned long flags;
1142 
1143 	spin_lock_irqsave(q->queue_lock, flags);
1144 	__stop_queue(q);
1145 	spin_unlock_irqrestore(q->queue_lock, flags);
1146 }
1147 
1148 static void __start_queue(struct request_queue *q)
1149 {
1150 	if (blk_queue_stopped(q))
1151 		blk_start_queue(q);
1152 }
1153 
1154 static void start_queue(struct request_queue *q)
1155 {
1156 	unsigned long flags;
1157 
1158 	spin_lock_irqsave(q->queue_lock, flags);
1159 	__start_queue(q);
1160 	spin_unlock_irqrestore(q->queue_lock, flags);
1161 }
1162 
1163 static void dm_done(struct request *clone, int error, bool mapped)
1164 {
1165 	int r = error;
1166 	struct dm_rq_target_io *tio = clone->end_io_data;
1167 	dm_request_endio_fn rq_end_io = NULL;
1168 
1169 	if (tio->ti) {
1170 		rq_end_io = tio->ti->type->rq_end_io;
1171 
1172 		if (mapped && rq_end_io)
1173 			r = rq_end_io(tio->ti, clone, error, &tio->info);
1174 	}
1175 
1176 	if (unlikely(r == -EREMOTEIO && (clone->cmd_flags & REQ_WRITE_SAME) &&
1177 		     !clone->q->limits.max_write_same_sectors))
1178 		disable_write_same(tio->md);
1179 
1180 	if (r <= 0)
1181 		/* The target wants to complete the I/O */
1182 		dm_end_request(clone, r);
1183 	else if (r == DM_ENDIO_INCOMPLETE)
1184 		/* The target will handle the I/O */
1185 		return;
1186 	else if (r == DM_ENDIO_REQUEUE)
1187 		/* The target wants to requeue the I/O */
1188 		dm_requeue_unmapped_request(clone);
1189 	else {
1190 		DMWARN("unimplemented target endio return value: %d", r);
1191 		BUG();
1192 	}
1193 }
1194 
1195 /*
1196  * Request completion handler for request-based dm
1197  */
1198 static void dm_softirq_done(struct request *rq)
1199 {
1200 	bool mapped = true;
1201 	struct dm_rq_target_io *tio = rq->special;
1202 	struct request *clone = tio->clone;
1203 
1204 	if (!clone) {
1205 		blk_end_request_all(rq, tio->error);
1206 		rq_completed(tio->md, rq_data_dir(rq), false);
1207 		free_rq_tio(tio);
1208 		return;
1209 	}
1210 
1211 	if (rq->cmd_flags & REQ_FAILED)
1212 		mapped = false;
1213 
1214 	dm_done(clone, tio->error, mapped);
1215 }
1216 
1217 /*
1218  * Complete the clone and the original request with the error status
1219  * through softirq context.
1220  */
1221 static void dm_complete_request(struct request *rq, int error)
1222 {
1223 	struct dm_rq_target_io *tio = rq->special;
1224 
1225 	tio->error = error;
1226 	blk_complete_request(rq);
1227 }
1228 
1229 /*
1230  * Complete the not-mapped clone and the original request with the error status
1231  * through softirq context.
1232  * Target's rq_end_io() function isn't called.
1233  * This may be used when the target's map_rq() or clone_and_map_rq() functions fail.
1234  */
1235 static void dm_kill_unmapped_request(struct request *rq, int error)
1236 {
1237 	rq->cmd_flags |= REQ_FAILED;
1238 	dm_complete_request(rq, error);
1239 }
1240 
1241 /*
1242  * Called with the clone's queue lock held
1243  */
1244 static void end_clone_request(struct request *clone, int error)
1245 {
1246 	struct dm_rq_target_io *tio = clone->end_io_data;
1247 
1248 	if (!clone->q->mq_ops) {
1249 		/*
1250 		 * For just cleaning up the information of the queue in which
1251 		 * the clone was dispatched.
1252 		 * The clone is *NOT* freed actually here because it is alloced
1253 		 * from dm own mempool (REQ_ALLOCED isn't set).
1254 		 */
1255 		__blk_put_request(clone->q, clone);
1256 	}
1257 
1258 	/*
1259 	 * Actual request completion is done in a softirq context which doesn't
1260 	 * hold the clone's queue lock.  Otherwise, deadlock could occur because:
1261 	 *     - another request may be submitted by the upper level driver
1262 	 *       of the stacking during the completion
1263 	 *     - the submission which requires queue lock may be done
1264 	 *       against this clone's queue
1265 	 */
1266 	dm_complete_request(tio->orig, error);
1267 }
1268 
1269 /*
1270  * Return maximum size of I/O possible at the supplied sector up to the current
1271  * target boundary.
1272  */
1273 static sector_t max_io_len_target_boundary(sector_t sector, struct dm_target *ti)
1274 {
1275 	sector_t target_offset = dm_target_offset(ti, sector);
1276 
1277 	return ti->len - target_offset;
1278 }
1279 
1280 static sector_t max_io_len(sector_t sector, struct dm_target *ti)
1281 {
1282 	sector_t len = max_io_len_target_boundary(sector, ti);
1283 	sector_t offset, max_len;
1284 
1285 	/*
1286 	 * Does the target need to split even further?
1287 	 */
1288 	if (ti->max_io_len) {
1289 		offset = dm_target_offset(ti, sector);
1290 		if (unlikely(ti->max_io_len & (ti->max_io_len - 1)))
1291 			max_len = sector_div(offset, ti->max_io_len);
1292 		else
1293 			max_len = offset & (ti->max_io_len - 1);
1294 		max_len = ti->max_io_len - max_len;
1295 
1296 		if (len > max_len)
1297 			len = max_len;
1298 	}
1299 
1300 	return len;
1301 }
1302 
1303 int dm_set_target_max_io_len(struct dm_target *ti, sector_t len)
1304 {
1305 	if (len > UINT_MAX) {
1306 		DMERR("Specified maximum size of target IO (%llu) exceeds limit (%u)",
1307 		      (unsigned long long)len, UINT_MAX);
1308 		ti->error = "Maximum size of target IO is too large";
1309 		return -EINVAL;
1310 	}
1311 
1312 	ti->max_io_len = (uint32_t) len;
1313 
1314 	return 0;
1315 }
1316 EXPORT_SYMBOL_GPL(dm_set_target_max_io_len);
1317 
1318 /*
1319  * A target may call dm_accept_partial_bio only from the map routine.  It is
1320  * allowed for all bio types except REQ_FLUSH.
1321  *
1322  * dm_accept_partial_bio informs the dm that the target only wants to process
1323  * additional n_sectors sectors of the bio and the rest of the data should be
1324  * sent in a next bio.
1325  *
1326  * A diagram that explains the arithmetics:
1327  * +--------------------+---------------+-------+
1328  * |         1          |       2       |   3   |
1329  * +--------------------+---------------+-------+
1330  *
1331  * <-------------- *tio->len_ptr --------------->
1332  *                      <------- bi_size ------->
1333  *                      <-- n_sectors -->
1334  *
1335  * Region 1 was already iterated over with bio_advance or similar function.
1336  *	(it may be empty if the target doesn't use bio_advance)
1337  * Region 2 is the remaining bio size that the target wants to process.
1338  *	(it may be empty if region 1 is non-empty, although there is no reason
1339  *	 to make it empty)
1340  * The target requires that region 3 is to be sent in the next bio.
1341  *
1342  * If the target wants to receive multiple copies of the bio (via num_*bios, etc),
1343  * the partially processed part (the sum of regions 1+2) must be the same for all
1344  * copies of the bio.
1345  */
1346 void dm_accept_partial_bio(struct bio *bio, unsigned n_sectors)
1347 {
1348 	struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone);
1349 	unsigned bi_size = bio->bi_iter.bi_size >> SECTOR_SHIFT;
1350 	BUG_ON(bio->bi_rw & REQ_FLUSH);
1351 	BUG_ON(bi_size > *tio->len_ptr);
1352 	BUG_ON(n_sectors > bi_size);
1353 	*tio->len_ptr -= bi_size - n_sectors;
1354 	bio->bi_iter.bi_size = n_sectors << SECTOR_SHIFT;
1355 }
1356 EXPORT_SYMBOL_GPL(dm_accept_partial_bio);
1357 
1358 static void __map_bio(struct dm_target_io *tio)
1359 {
1360 	int r;
1361 	sector_t sector;
1362 	struct mapped_device *md;
1363 	struct bio *clone = &tio->clone;
1364 	struct dm_target *ti = tio->ti;
1365 
1366 	clone->bi_end_io = clone_endio;
1367 
1368 	/*
1369 	 * Map the clone.  If r == 0 we don't need to do
1370 	 * anything, the target has assumed ownership of
1371 	 * this io.
1372 	 */
1373 	atomic_inc(&tio->io->io_count);
1374 	sector = clone->bi_iter.bi_sector;
1375 	r = ti->type->map(ti, clone);
1376 	if (r == DM_MAPIO_REMAPPED) {
1377 		/* the bio has been remapped so dispatch it */
1378 
1379 		trace_block_bio_remap(bdev_get_queue(clone->bi_bdev), clone,
1380 				      tio->io->bio->bi_bdev->bd_dev, sector);
1381 
1382 		generic_make_request(clone);
1383 	} else if (r < 0 || r == DM_MAPIO_REQUEUE) {
1384 		/* error the io and bail out, or requeue it if needed */
1385 		md = tio->io->md;
1386 		dec_pending(tio->io, r);
1387 		free_tio(md, tio);
1388 	} else if (r) {
1389 		DMWARN("unimplemented target map return value: %d", r);
1390 		BUG();
1391 	}
1392 }
1393 
1394 struct clone_info {
1395 	struct mapped_device *md;
1396 	struct dm_table *map;
1397 	struct bio *bio;
1398 	struct dm_io *io;
1399 	sector_t sector;
1400 	unsigned sector_count;
1401 };
1402 
1403 static void bio_setup_sector(struct bio *bio, sector_t sector, unsigned len)
1404 {
1405 	bio->bi_iter.bi_sector = sector;
1406 	bio->bi_iter.bi_size = to_bytes(len);
1407 }
1408 
1409 /*
1410  * Creates a bio that consists of range of complete bvecs.
1411  */
1412 static void clone_bio(struct dm_target_io *tio, struct bio *bio,
1413 		      sector_t sector, unsigned len)
1414 {
1415 	struct bio *clone = &tio->clone;
1416 
1417 	__bio_clone_fast(clone, bio);
1418 
1419 	if (bio_integrity(bio))
1420 		bio_integrity_clone(clone, bio, GFP_NOIO);
1421 
1422 	bio_advance(clone, to_bytes(sector - clone->bi_iter.bi_sector));
1423 	clone->bi_iter.bi_size = to_bytes(len);
1424 
1425 	if (bio_integrity(bio))
1426 		bio_integrity_trim(clone, 0, len);
1427 }
1428 
1429 static struct dm_target_io *alloc_tio(struct clone_info *ci,
1430 				      struct dm_target *ti,
1431 				      unsigned target_bio_nr)
1432 {
1433 	struct dm_target_io *tio;
1434 	struct bio *clone;
1435 
1436 	clone = bio_alloc_bioset(GFP_NOIO, 0, ci->md->bs);
1437 	tio = container_of(clone, struct dm_target_io, clone);
1438 
1439 	tio->io = ci->io;
1440 	tio->ti = ti;
1441 	tio->target_bio_nr = target_bio_nr;
1442 
1443 	return tio;
1444 }
1445 
1446 static void __clone_and_map_simple_bio(struct clone_info *ci,
1447 				       struct dm_target *ti,
1448 				       unsigned target_bio_nr, unsigned *len)
1449 {
1450 	struct dm_target_io *tio = alloc_tio(ci, ti, target_bio_nr);
1451 	struct bio *clone = &tio->clone;
1452 
1453 	tio->len_ptr = len;
1454 
1455 	__bio_clone_fast(clone, ci->bio);
1456 	if (len)
1457 		bio_setup_sector(clone, ci->sector, *len);
1458 
1459 	__map_bio(tio);
1460 }
1461 
1462 static void __send_duplicate_bios(struct clone_info *ci, struct dm_target *ti,
1463 				  unsigned num_bios, unsigned *len)
1464 {
1465 	unsigned target_bio_nr;
1466 
1467 	for (target_bio_nr = 0; target_bio_nr < num_bios; target_bio_nr++)
1468 		__clone_and_map_simple_bio(ci, ti, target_bio_nr, len);
1469 }
1470 
1471 static int __send_empty_flush(struct clone_info *ci)
1472 {
1473 	unsigned target_nr = 0;
1474 	struct dm_target *ti;
1475 
1476 	BUG_ON(bio_has_data(ci->bio));
1477 	while ((ti = dm_table_get_target(ci->map, target_nr++)))
1478 		__send_duplicate_bios(ci, ti, ti->num_flush_bios, NULL);
1479 
1480 	return 0;
1481 }
1482 
1483 static void __clone_and_map_data_bio(struct clone_info *ci, struct dm_target *ti,
1484 				     sector_t sector, unsigned *len)
1485 {
1486 	struct bio *bio = ci->bio;
1487 	struct dm_target_io *tio;
1488 	unsigned target_bio_nr;
1489 	unsigned num_target_bios = 1;
1490 
1491 	/*
1492 	 * Does the target want to receive duplicate copies of the bio?
1493 	 */
1494 	if (bio_data_dir(bio) == WRITE && ti->num_write_bios)
1495 		num_target_bios = ti->num_write_bios(ti, bio);
1496 
1497 	for (target_bio_nr = 0; target_bio_nr < num_target_bios; target_bio_nr++) {
1498 		tio = alloc_tio(ci, ti, target_bio_nr);
1499 		tio->len_ptr = len;
1500 		clone_bio(tio, bio, sector, *len);
1501 		__map_bio(tio);
1502 	}
1503 }
1504 
1505 typedef unsigned (*get_num_bios_fn)(struct dm_target *ti);
1506 
1507 static unsigned get_num_discard_bios(struct dm_target *ti)
1508 {
1509 	return ti->num_discard_bios;
1510 }
1511 
1512 static unsigned get_num_write_same_bios(struct dm_target *ti)
1513 {
1514 	return ti->num_write_same_bios;
1515 }
1516 
1517 typedef bool (*is_split_required_fn)(struct dm_target *ti);
1518 
1519 static bool is_split_required_for_discard(struct dm_target *ti)
1520 {
1521 	return ti->split_discard_bios;
1522 }
1523 
1524 static int __send_changing_extent_only(struct clone_info *ci,
1525 				       get_num_bios_fn get_num_bios,
1526 				       is_split_required_fn is_split_required)
1527 {
1528 	struct dm_target *ti;
1529 	unsigned len;
1530 	unsigned num_bios;
1531 
1532 	do {
1533 		ti = dm_table_find_target(ci->map, ci->sector);
1534 		if (!dm_target_is_valid(ti))
1535 			return -EIO;
1536 
1537 		/*
1538 		 * Even though the device advertised support for this type of
1539 		 * request, that does not mean every target supports it, and
1540 		 * reconfiguration might also have changed that since the
1541 		 * check was performed.
1542 		 */
1543 		num_bios = get_num_bios ? get_num_bios(ti) : 0;
1544 		if (!num_bios)
1545 			return -EOPNOTSUPP;
1546 
1547 		if (is_split_required && !is_split_required(ti))
1548 			len = min((sector_t)ci->sector_count, max_io_len_target_boundary(ci->sector, ti));
1549 		else
1550 			len = min((sector_t)ci->sector_count, max_io_len(ci->sector, ti));
1551 
1552 		__send_duplicate_bios(ci, ti, num_bios, &len);
1553 
1554 		ci->sector += len;
1555 	} while (ci->sector_count -= len);
1556 
1557 	return 0;
1558 }
1559 
1560 static int __send_discard(struct clone_info *ci)
1561 {
1562 	return __send_changing_extent_only(ci, get_num_discard_bios,
1563 					   is_split_required_for_discard);
1564 }
1565 
1566 static int __send_write_same(struct clone_info *ci)
1567 {
1568 	return __send_changing_extent_only(ci, get_num_write_same_bios, NULL);
1569 }
1570 
1571 /*
1572  * Select the correct strategy for processing a non-flush bio.
1573  */
1574 static int __split_and_process_non_flush(struct clone_info *ci)
1575 {
1576 	struct bio *bio = ci->bio;
1577 	struct dm_target *ti;
1578 	unsigned len;
1579 
1580 	if (unlikely(bio->bi_rw & REQ_DISCARD))
1581 		return __send_discard(ci);
1582 	else if (unlikely(bio->bi_rw & REQ_WRITE_SAME))
1583 		return __send_write_same(ci);
1584 
1585 	ti = dm_table_find_target(ci->map, ci->sector);
1586 	if (!dm_target_is_valid(ti))
1587 		return -EIO;
1588 
1589 	len = min_t(sector_t, max_io_len(ci->sector, ti), ci->sector_count);
1590 
1591 	__clone_and_map_data_bio(ci, ti, ci->sector, &len);
1592 
1593 	ci->sector += len;
1594 	ci->sector_count -= len;
1595 
1596 	return 0;
1597 }
1598 
1599 /*
1600  * Entry point to split a bio into clones and submit them to the targets.
1601  */
1602 static void __split_and_process_bio(struct mapped_device *md,
1603 				    struct dm_table *map, struct bio *bio)
1604 {
1605 	struct clone_info ci;
1606 	int error = 0;
1607 
1608 	if (unlikely(!map)) {
1609 		bio_io_error(bio);
1610 		return;
1611 	}
1612 
1613 	ci.map = map;
1614 	ci.md = md;
1615 	ci.io = alloc_io(md);
1616 	ci.io->error = 0;
1617 	atomic_set(&ci.io->io_count, 1);
1618 	ci.io->bio = bio;
1619 	ci.io->md = md;
1620 	spin_lock_init(&ci.io->endio_lock);
1621 	ci.sector = bio->bi_iter.bi_sector;
1622 
1623 	start_io_acct(ci.io);
1624 
1625 	if (bio->bi_rw & REQ_FLUSH) {
1626 		ci.bio = &ci.md->flush_bio;
1627 		ci.sector_count = 0;
1628 		error = __send_empty_flush(&ci);
1629 		/* dec_pending submits any data associated with flush */
1630 	} else {
1631 		ci.bio = bio;
1632 		ci.sector_count = bio_sectors(bio);
1633 		while (ci.sector_count && !error)
1634 			error = __split_and_process_non_flush(&ci);
1635 	}
1636 
1637 	/* drop the extra reference count */
1638 	dec_pending(ci.io, error);
1639 }
1640 /*-----------------------------------------------------------------
1641  * CRUD END
1642  *---------------------------------------------------------------*/
1643 
1644 static int dm_merge_bvec(struct request_queue *q,
1645 			 struct bvec_merge_data *bvm,
1646 			 struct bio_vec *biovec)
1647 {
1648 	struct mapped_device *md = q->queuedata;
1649 	struct dm_table *map = dm_get_live_table_fast(md);
1650 	struct dm_target *ti;
1651 	sector_t max_sectors;
1652 	int max_size = 0;
1653 
1654 	if (unlikely(!map))
1655 		goto out;
1656 
1657 	ti = dm_table_find_target(map, bvm->bi_sector);
1658 	if (!dm_target_is_valid(ti))
1659 		goto out;
1660 
1661 	/*
1662 	 * Find maximum amount of I/O that won't need splitting
1663 	 */
1664 	max_sectors = min(max_io_len(bvm->bi_sector, ti),
1665 			  (sector_t) queue_max_sectors(q));
1666 	max_size = (max_sectors << SECTOR_SHIFT) - bvm->bi_size;
1667 	if (unlikely(max_size < 0)) /* this shouldn't _ever_ happen */
1668 		max_size = 0;
1669 
1670 	/*
1671 	 * merge_bvec_fn() returns number of bytes
1672 	 * it can accept at this offset
1673 	 * max is precomputed maximal io size
1674 	 */
1675 	if (max_size && ti->type->merge)
1676 		max_size = ti->type->merge(ti, bvm, biovec, max_size);
1677 	/*
1678 	 * If the target doesn't support merge method and some of the devices
1679 	 * provided their merge_bvec method (we know this by looking for the
1680 	 * max_hw_sectors that dm_set_device_limits may set), then we can't
1681 	 * allow bios with multiple vector entries.  So always set max_size
1682 	 * to 0, and the code below allows just one page.
1683 	 */
1684 	else if (queue_max_hw_sectors(q) <= PAGE_SIZE >> 9)
1685 		max_size = 0;
1686 
1687 out:
1688 	dm_put_live_table_fast(md);
1689 	/*
1690 	 * Always allow an entire first page
1691 	 */
1692 	if (max_size <= biovec->bv_len && !(bvm->bi_size >> SECTOR_SHIFT))
1693 		max_size = biovec->bv_len;
1694 
1695 	return max_size;
1696 }
1697 
1698 /*
1699  * The request function that just remaps the bio built up by
1700  * dm_merge_bvec.
1701  */
1702 static void dm_make_request(struct request_queue *q, struct bio *bio)
1703 {
1704 	int rw = bio_data_dir(bio);
1705 	struct mapped_device *md = q->queuedata;
1706 	int srcu_idx;
1707 	struct dm_table *map;
1708 
1709 	map = dm_get_live_table(md, &srcu_idx);
1710 
1711 	generic_start_io_acct(rw, bio_sectors(bio), &dm_disk(md)->part0);
1712 
1713 	/* if we're suspended, we have to queue this io for later */
1714 	if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags))) {
1715 		dm_put_live_table(md, srcu_idx);
1716 
1717 		if (bio_rw(bio) != READA)
1718 			queue_io(md, bio);
1719 		else
1720 			bio_io_error(bio);
1721 		return;
1722 	}
1723 
1724 	__split_and_process_bio(md, map, bio);
1725 	dm_put_live_table(md, srcu_idx);
1726 	return;
1727 }
1728 
1729 int dm_request_based(struct mapped_device *md)
1730 {
1731 	return blk_queue_stackable(md->queue);
1732 }
1733 
1734 static void dm_dispatch_clone_request(struct request *clone, struct request *rq)
1735 {
1736 	int r;
1737 
1738 	if (blk_queue_io_stat(clone->q))
1739 		clone->cmd_flags |= REQ_IO_STAT;
1740 
1741 	clone->start_time = jiffies;
1742 	r = blk_insert_cloned_request(clone->q, clone);
1743 	if (r)
1744 		/* must complete clone in terms of original request */
1745 		dm_complete_request(rq, r);
1746 }
1747 
1748 static int dm_rq_bio_constructor(struct bio *bio, struct bio *bio_orig,
1749 				 void *data)
1750 {
1751 	struct dm_rq_target_io *tio = data;
1752 	struct dm_rq_clone_bio_info *info =
1753 		container_of(bio, struct dm_rq_clone_bio_info, clone);
1754 
1755 	info->orig = bio_orig;
1756 	info->tio = tio;
1757 	bio->bi_end_io = end_clone_bio;
1758 
1759 	return 0;
1760 }
1761 
1762 static int setup_clone(struct request *clone, struct request *rq,
1763 		       struct dm_rq_target_io *tio, gfp_t gfp_mask)
1764 {
1765 	int r;
1766 
1767 	r = blk_rq_prep_clone(clone, rq, tio->md->bs, gfp_mask,
1768 			      dm_rq_bio_constructor, tio);
1769 	if (r)
1770 		return r;
1771 
1772 	clone->cmd = rq->cmd;
1773 	clone->cmd_len = rq->cmd_len;
1774 	clone->sense = rq->sense;
1775 	clone->end_io = end_clone_request;
1776 	clone->end_io_data = tio;
1777 
1778 	tio->clone = clone;
1779 
1780 	return 0;
1781 }
1782 
1783 static struct request *clone_rq(struct request *rq, struct mapped_device *md,
1784 				struct dm_rq_target_io *tio, gfp_t gfp_mask)
1785 {
1786 	struct request *clone = alloc_clone_request(md, gfp_mask);
1787 
1788 	if (!clone)
1789 		return NULL;
1790 
1791 	blk_rq_init(NULL, clone);
1792 	if (setup_clone(clone, rq, tio, gfp_mask)) {
1793 		/* -ENOMEM */
1794 		free_clone_request(md, clone);
1795 		return NULL;
1796 	}
1797 
1798 	return clone;
1799 }
1800 
1801 static void map_tio_request(struct kthread_work *work);
1802 
1803 static struct dm_rq_target_io *prep_tio(struct request *rq,
1804 					struct mapped_device *md, gfp_t gfp_mask)
1805 {
1806 	struct dm_rq_target_io *tio;
1807 	int srcu_idx;
1808 	struct dm_table *table;
1809 
1810 	tio = alloc_rq_tio(md, gfp_mask);
1811 	if (!tio)
1812 		return NULL;
1813 
1814 	tio->md = md;
1815 	tio->ti = NULL;
1816 	tio->clone = NULL;
1817 	tio->orig = rq;
1818 	tio->error = 0;
1819 	memset(&tio->info, 0, sizeof(tio->info));
1820 	init_kthread_work(&tio->work, map_tio_request);
1821 
1822 	table = dm_get_live_table(md, &srcu_idx);
1823 	if (!dm_table_mq_request_based(table)) {
1824 		if (!clone_rq(rq, md, tio, gfp_mask)) {
1825 			dm_put_live_table(md, srcu_idx);
1826 			free_rq_tio(tio);
1827 			return NULL;
1828 		}
1829 	}
1830 	dm_put_live_table(md, srcu_idx);
1831 
1832 	return tio;
1833 }
1834 
1835 /*
1836  * Called with the queue lock held.
1837  */
1838 static int dm_prep_fn(struct request_queue *q, struct request *rq)
1839 {
1840 	struct mapped_device *md = q->queuedata;
1841 	struct dm_rq_target_io *tio;
1842 
1843 	if (unlikely(rq->special)) {
1844 		DMWARN("Already has something in rq->special.");
1845 		return BLKPREP_KILL;
1846 	}
1847 
1848 	tio = prep_tio(rq, md, GFP_ATOMIC);
1849 	if (!tio)
1850 		return BLKPREP_DEFER;
1851 
1852 	rq->special = tio;
1853 	rq->cmd_flags |= REQ_DONTPREP;
1854 
1855 	return BLKPREP_OK;
1856 }
1857 
1858 /*
1859  * Returns:
1860  * 0                : the request has been processed
1861  * DM_MAPIO_REQUEUE : the original request needs to be requeued
1862  * < 0              : the request was completed due to failure
1863  */
1864 static int map_request(struct dm_target *ti, struct request *rq,
1865 		       struct mapped_device *md)
1866 {
1867 	int r;
1868 	struct dm_rq_target_io *tio = rq->special;
1869 	struct request *clone = NULL;
1870 
1871 	if (tio->clone) {
1872 		clone = tio->clone;
1873 		r = ti->type->map_rq(ti, clone, &tio->info);
1874 	} else {
1875 		r = ti->type->clone_and_map_rq(ti, rq, &tio->info, &clone);
1876 		if (r < 0) {
1877 			/* The target wants to complete the I/O */
1878 			dm_kill_unmapped_request(rq, r);
1879 			return r;
1880 		}
1881 		if (IS_ERR(clone))
1882 			return DM_MAPIO_REQUEUE;
1883 		if (setup_clone(clone, rq, tio, GFP_KERNEL)) {
1884 			/* -ENOMEM */
1885 			ti->type->release_clone_rq(clone);
1886 			return DM_MAPIO_REQUEUE;
1887 		}
1888 	}
1889 
1890 	switch (r) {
1891 	case DM_MAPIO_SUBMITTED:
1892 		/* The target has taken the I/O to submit by itself later */
1893 		break;
1894 	case DM_MAPIO_REMAPPED:
1895 		/* The target has remapped the I/O so dispatch it */
1896 		trace_block_rq_remap(clone->q, clone, disk_devt(dm_disk(md)),
1897 				     blk_rq_pos(rq));
1898 		dm_dispatch_clone_request(clone, rq);
1899 		break;
1900 	case DM_MAPIO_REQUEUE:
1901 		/* The target wants to requeue the I/O */
1902 		dm_requeue_unmapped_request(clone);
1903 		break;
1904 	default:
1905 		if (r > 0) {
1906 			DMWARN("unimplemented target map return value: %d", r);
1907 			BUG();
1908 		}
1909 
1910 		/* The target wants to complete the I/O */
1911 		dm_kill_unmapped_request(rq, r);
1912 		return r;
1913 	}
1914 
1915 	return 0;
1916 }
1917 
1918 static void map_tio_request(struct kthread_work *work)
1919 {
1920 	struct dm_rq_target_io *tio = container_of(work, struct dm_rq_target_io, work);
1921 	struct request *rq = tio->orig;
1922 	struct mapped_device *md = tio->md;
1923 
1924 	if (map_request(tio->ti, rq, md) == DM_MAPIO_REQUEUE)
1925 		dm_requeue_unmapped_original_request(md, rq);
1926 }
1927 
1928 static void dm_start_request(struct mapped_device *md, struct request *orig)
1929 {
1930 	blk_start_request(orig);
1931 	atomic_inc(&md->pending[rq_data_dir(orig)]);
1932 
1933 	/*
1934 	 * Hold the md reference here for the in-flight I/O.
1935 	 * We can't rely on the reference count by device opener,
1936 	 * because the device may be closed during the request completion
1937 	 * when all bios are completed.
1938 	 * See the comment in rq_completed() too.
1939 	 */
1940 	dm_get(md);
1941 }
1942 
1943 /*
1944  * q->request_fn for request-based dm.
1945  * Called with the queue lock held.
1946  */
1947 static void dm_request_fn(struct request_queue *q)
1948 {
1949 	struct mapped_device *md = q->queuedata;
1950 	int srcu_idx;
1951 	struct dm_table *map = dm_get_live_table(md, &srcu_idx);
1952 	struct dm_target *ti;
1953 	struct request *rq;
1954 	struct dm_rq_target_io *tio;
1955 	sector_t pos;
1956 
1957 	/*
1958 	 * For suspend, check blk_queue_stopped() and increment
1959 	 * ->pending within a single queue_lock not to increment the
1960 	 * number of in-flight I/Os after the queue is stopped in
1961 	 * dm_suspend().
1962 	 */
1963 	while (!blk_queue_stopped(q)) {
1964 		rq = blk_peek_request(q);
1965 		if (!rq)
1966 			goto out;
1967 
1968 		/* always use block 0 to find the target for flushes for now */
1969 		pos = 0;
1970 		if (!(rq->cmd_flags & REQ_FLUSH))
1971 			pos = blk_rq_pos(rq);
1972 
1973 		ti = dm_table_find_target(map, pos);
1974 		if (!dm_target_is_valid(ti)) {
1975 			/*
1976 			 * Must perform setup, that rq_completed() requires,
1977 			 * before calling dm_kill_unmapped_request
1978 			 */
1979 			DMERR_LIMIT("request attempted access beyond the end of device");
1980 			dm_start_request(md, rq);
1981 			dm_kill_unmapped_request(rq, -EIO);
1982 			continue;
1983 		}
1984 
1985 		if (ti->type->busy && ti->type->busy(ti))
1986 			goto delay_and_out;
1987 
1988 		dm_start_request(md, rq);
1989 
1990 		tio = rq->special;
1991 		/* Establish tio->ti before queuing work (map_tio_request) */
1992 		tio->ti = ti;
1993 		queue_kthread_work(&md->kworker, &tio->work);
1994 		BUG_ON(!irqs_disabled());
1995 	}
1996 
1997 	goto out;
1998 
1999 delay_and_out:
2000 	blk_delay_queue(q, HZ / 10);
2001 out:
2002 	dm_put_live_table(md, srcu_idx);
2003 }
2004 
2005 static int dm_any_congested(void *congested_data, int bdi_bits)
2006 {
2007 	int r = bdi_bits;
2008 	struct mapped_device *md = congested_data;
2009 	struct dm_table *map;
2010 
2011 	if (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
2012 		map = dm_get_live_table_fast(md);
2013 		if (map) {
2014 			/*
2015 			 * Request-based dm cares about only own queue for
2016 			 * the query about congestion status of request_queue
2017 			 */
2018 			if (dm_request_based(md))
2019 				r = md->queue->backing_dev_info.state &
2020 				    bdi_bits;
2021 			else
2022 				r = dm_table_any_congested(map, bdi_bits);
2023 		}
2024 		dm_put_live_table_fast(md);
2025 	}
2026 
2027 	return r;
2028 }
2029 
2030 /*-----------------------------------------------------------------
2031  * An IDR is used to keep track of allocated minor numbers.
2032  *---------------------------------------------------------------*/
2033 static void free_minor(int minor)
2034 {
2035 	spin_lock(&_minor_lock);
2036 	idr_remove(&_minor_idr, minor);
2037 	spin_unlock(&_minor_lock);
2038 }
2039 
2040 /*
2041  * See if the device with a specific minor # is free.
2042  */
2043 static int specific_minor(int minor)
2044 {
2045 	int r;
2046 
2047 	if (minor >= (1 << MINORBITS))
2048 		return -EINVAL;
2049 
2050 	idr_preload(GFP_KERNEL);
2051 	spin_lock(&_minor_lock);
2052 
2053 	r = idr_alloc(&_minor_idr, MINOR_ALLOCED, minor, minor + 1, GFP_NOWAIT);
2054 
2055 	spin_unlock(&_minor_lock);
2056 	idr_preload_end();
2057 	if (r < 0)
2058 		return r == -ENOSPC ? -EBUSY : r;
2059 	return 0;
2060 }
2061 
2062 static int next_free_minor(int *minor)
2063 {
2064 	int r;
2065 
2066 	idr_preload(GFP_KERNEL);
2067 	spin_lock(&_minor_lock);
2068 
2069 	r = idr_alloc(&_minor_idr, MINOR_ALLOCED, 0, 1 << MINORBITS, GFP_NOWAIT);
2070 
2071 	spin_unlock(&_minor_lock);
2072 	idr_preload_end();
2073 	if (r < 0)
2074 		return r;
2075 	*minor = r;
2076 	return 0;
2077 }
2078 
2079 static const struct block_device_operations dm_blk_dops;
2080 
2081 static void dm_wq_work(struct work_struct *work);
2082 
2083 static void dm_init_md_queue(struct mapped_device *md)
2084 {
2085 	/*
2086 	 * Request-based dm devices cannot be stacked on top of bio-based dm
2087 	 * devices.  The type of this dm device has not been decided yet.
2088 	 * The type is decided at the first table loading time.
2089 	 * To prevent problematic device stacking, clear the queue flag
2090 	 * for request stacking support until then.
2091 	 *
2092 	 * This queue is new, so no concurrency on the queue_flags.
2093 	 */
2094 	queue_flag_clear_unlocked(QUEUE_FLAG_STACKABLE, md->queue);
2095 
2096 	md->queue->queuedata = md;
2097 	md->queue->backing_dev_info.congested_fn = dm_any_congested;
2098 	md->queue->backing_dev_info.congested_data = md;
2099 
2100 	blk_queue_bounce_limit(md->queue, BLK_BOUNCE_ANY);
2101 }
2102 
2103 /*
2104  * Allocate and initialise a blank device with a given minor.
2105  */
2106 static struct mapped_device *alloc_dev(int minor)
2107 {
2108 	int r;
2109 	struct mapped_device *md = kzalloc(sizeof(*md), GFP_KERNEL);
2110 	void *old_md;
2111 
2112 	if (!md) {
2113 		DMWARN("unable to allocate device, out of memory.");
2114 		return NULL;
2115 	}
2116 
2117 	if (!try_module_get(THIS_MODULE))
2118 		goto bad_module_get;
2119 
2120 	/* get a minor number for the dev */
2121 	if (minor == DM_ANY_MINOR)
2122 		r = next_free_minor(&minor);
2123 	else
2124 		r = specific_minor(minor);
2125 	if (r < 0)
2126 		goto bad_minor;
2127 
2128 	r = init_srcu_struct(&md->io_barrier);
2129 	if (r < 0)
2130 		goto bad_io_barrier;
2131 
2132 	md->type = DM_TYPE_NONE;
2133 	mutex_init(&md->suspend_lock);
2134 	mutex_init(&md->type_lock);
2135 	mutex_init(&md->table_devices_lock);
2136 	spin_lock_init(&md->deferred_lock);
2137 	atomic_set(&md->holders, 1);
2138 	atomic_set(&md->open_count, 0);
2139 	atomic_set(&md->event_nr, 0);
2140 	atomic_set(&md->uevent_seq, 0);
2141 	INIT_LIST_HEAD(&md->uevent_list);
2142 	INIT_LIST_HEAD(&md->table_devices);
2143 	spin_lock_init(&md->uevent_lock);
2144 
2145 	md->queue = blk_alloc_queue(GFP_KERNEL);
2146 	if (!md->queue)
2147 		goto bad_queue;
2148 
2149 	dm_init_md_queue(md);
2150 
2151 	md->disk = alloc_disk(1);
2152 	if (!md->disk)
2153 		goto bad_disk;
2154 
2155 	atomic_set(&md->pending[0], 0);
2156 	atomic_set(&md->pending[1], 0);
2157 	init_waitqueue_head(&md->wait);
2158 	INIT_WORK(&md->work, dm_wq_work);
2159 	init_waitqueue_head(&md->eventq);
2160 	init_completion(&md->kobj_holder.completion);
2161 	md->kworker_task = NULL;
2162 
2163 	md->disk->major = _major;
2164 	md->disk->first_minor = minor;
2165 	md->disk->fops = &dm_blk_dops;
2166 	md->disk->queue = md->queue;
2167 	md->disk->private_data = md;
2168 	sprintf(md->disk->disk_name, "dm-%d", minor);
2169 	add_disk(md->disk);
2170 	format_dev_t(md->name, MKDEV(_major, minor));
2171 
2172 	md->wq = alloc_workqueue("kdmflush", WQ_MEM_RECLAIM, 0);
2173 	if (!md->wq)
2174 		goto bad_thread;
2175 
2176 	md->bdev = bdget_disk(md->disk, 0);
2177 	if (!md->bdev)
2178 		goto bad_bdev;
2179 
2180 	bio_init(&md->flush_bio);
2181 	md->flush_bio.bi_bdev = md->bdev;
2182 	md->flush_bio.bi_rw = WRITE_FLUSH;
2183 
2184 	dm_stats_init(&md->stats);
2185 
2186 	/* Populate the mapping, nobody knows we exist yet */
2187 	spin_lock(&_minor_lock);
2188 	old_md = idr_replace(&_minor_idr, md, minor);
2189 	spin_unlock(&_minor_lock);
2190 
2191 	BUG_ON(old_md != MINOR_ALLOCED);
2192 
2193 	return md;
2194 
2195 bad_bdev:
2196 	destroy_workqueue(md->wq);
2197 bad_thread:
2198 	del_gendisk(md->disk);
2199 	put_disk(md->disk);
2200 bad_disk:
2201 	blk_cleanup_queue(md->queue);
2202 bad_queue:
2203 	cleanup_srcu_struct(&md->io_barrier);
2204 bad_io_barrier:
2205 	free_minor(minor);
2206 bad_minor:
2207 	module_put(THIS_MODULE);
2208 bad_module_get:
2209 	kfree(md);
2210 	return NULL;
2211 }
2212 
2213 static void unlock_fs(struct mapped_device *md);
2214 
2215 static void free_dev(struct mapped_device *md)
2216 {
2217 	int minor = MINOR(disk_devt(md->disk));
2218 
2219 	unlock_fs(md);
2220 	destroy_workqueue(md->wq);
2221 
2222 	if (md->kworker_task)
2223 		kthread_stop(md->kworker_task);
2224 	if (md->io_pool)
2225 		mempool_destroy(md->io_pool);
2226 	if (md->rq_pool)
2227 		mempool_destroy(md->rq_pool);
2228 	if (md->bs)
2229 		bioset_free(md->bs);
2230 
2231 	cleanup_srcu_struct(&md->io_barrier);
2232 	free_table_devices(&md->table_devices);
2233 	dm_stats_cleanup(&md->stats);
2234 
2235 	spin_lock(&_minor_lock);
2236 	md->disk->private_data = NULL;
2237 	spin_unlock(&_minor_lock);
2238 	if (blk_get_integrity(md->disk))
2239 		blk_integrity_unregister(md->disk);
2240 	del_gendisk(md->disk);
2241 	put_disk(md->disk);
2242 	blk_cleanup_queue(md->queue);
2243 	bdput(md->bdev);
2244 	free_minor(minor);
2245 
2246 	module_put(THIS_MODULE);
2247 	kfree(md);
2248 }
2249 
2250 static void __bind_mempools(struct mapped_device *md, struct dm_table *t)
2251 {
2252 	struct dm_md_mempools *p = dm_table_get_md_mempools(t);
2253 
2254 	if (md->io_pool && md->bs) {
2255 		/* The md already has necessary mempools. */
2256 		if (dm_table_get_type(t) == DM_TYPE_BIO_BASED) {
2257 			/*
2258 			 * Reload bioset because front_pad may have changed
2259 			 * because a different table was loaded.
2260 			 */
2261 			bioset_free(md->bs);
2262 			md->bs = p->bs;
2263 			p->bs = NULL;
2264 		}
2265 		/*
2266 		 * There's no need to reload with request-based dm
2267 		 * because the size of front_pad doesn't change.
2268 		 * Note for future: If you are to reload bioset,
2269 		 * prep-ed requests in the queue may refer
2270 		 * to bio from the old bioset, so you must walk
2271 		 * through the queue to unprep.
2272 		 */
2273 		goto out;
2274 	}
2275 
2276 	BUG_ON(!p || md->io_pool || md->rq_pool || md->bs);
2277 
2278 	md->io_pool = p->io_pool;
2279 	p->io_pool = NULL;
2280 	md->rq_pool = p->rq_pool;
2281 	p->rq_pool = NULL;
2282 	md->bs = p->bs;
2283 	p->bs = NULL;
2284 
2285 out:
2286 	/* mempool bind completed, now no need any mempools in the table */
2287 	dm_table_free_md_mempools(t);
2288 }
2289 
2290 /*
2291  * Bind a table to the device.
2292  */
2293 static void event_callback(void *context)
2294 {
2295 	unsigned long flags;
2296 	LIST_HEAD(uevents);
2297 	struct mapped_device *md = (struct mapped_device *) context;
2298 
2299 	spin_lock_irqsave(&md->uevent_lock, flags);
2300 	list_splice_init(&md->uevent_list, &uevents);
2301 	spin_unlock_irqrestore(&md->uevent_lock, flags);
2302 
2303 	dm_send_uevents(&uevents, &disk_to_dev(md->disk)->kobj);
2304 
2305 	atomic_inc(&md->event_nr);
2306 	wake_up(&md->eventq);
2307 }
2308 
2309 /*
2310  * Protected by md->suspend_lock obtained by dm_swap_table().
2311  */
2312 static void __set_size(struct mapped_device *md, sector_t size)
2313 {
2314 	set_capacity(md->disk, size);
2315 
2316 	i_size_write(md->bdev->bd_inode, (loff_t)size << SECTOR_SHIFT);
2317 }
2318 
2319 /*
2320  * Return 1 if the queue has a compulsory merge_bvec_fn function.
2321  *
2322  * If this function returns 0, then the device is either a non-dm
2323  * device without a merge_bvec_fn, or it is a dm device that is
2324  * able to split any bios it receives that are too big.
2325  */
2326 int dm_queue_merge_is_compulsory(struct request_queue *q)
2327 {
2328 	struct mapped_device *dev_md;
2329 
2330 	if (!q->merge_bvec_fn)
2331 		return 0;
2332 
2333 	if (q->make_request_fn == dm_make_request) {
2334 		dev_md = q->queuedata;
2335 		if (test_bit(DMF_MERGE_IS_OPTIONAL, &dev_md->flags))
2336 			return 0;
2337 	}
2338 
2339 	return 1;
2340 }
2341 
2342 static int dm_device_merge_is_compulsory(struct dm_target *ti,
2343 					 struct dm_dev *dev, sector_t start,
2344 					 sector_t len, void *data)
2345 {
2346 	struct block_device *bdev = dev->bdev;
2347 	struct request_queue *q = bdev_get_queue(bdev);
2348 
2349 	return dm_queue_merge_is_compulsory(q);
2350 }
2351 
2352 /*
2353  * Return 1 if it is acceptable to ignore merge_bvec_fn based
2354  * on the properties of the underlying devices.
2355  */
2356 static int dm_table_merge_is_optional(struct dm_table *table)
2357 {
2358 	unsigned i = 0;
2359 	struct dm_target *ti;
2360 
2361 	while (i < dm_table_get_num_targets(table)) {
2362 		ti = dm_table_get_target(table, i++);
2363 
2364 		if (ti->type->iterate_devices &&
2365 		    ti->type->iterate_devices(ti, dm_device_merge_is_compulsory, NULL))
2366 			return 0;
2367 	}
2368 
2369 	return 1;
2370 }
2371 
2372 /*
2373  * Returns old map, which caller must destroy.
2374  */
2375 static struct dm_table *__bind(struct mapped_device *md, struct dm_table *t,
2376 			       struct queue_limits *limits)
2377 {
2378 	struct dm_table *old_map;
2379 	struct request_queue *q = md->queue;
2380 	sector_t size;
2381 	int merge_is_optional;
2382 
2383 	size = dm_table_get_size(t);
2384 
2385 	/*
2386 	 * Wipe any geometry if the size of the table changed.
2387 	 */
2388 	if (size != dm_get_size(md))
2389 		memset(&md->geometry, 0, sizeof(md->geometry));
2390 
2391 	__set_size(md, size);
2392 
2393 	dm_table_event_callback(t, event_callback, md);
2394 
2395 	/*
2396 	 * The queue hasn't been stopped yet, if the old table type wasn't
2397 	 * for request-based during suspension.  So stop it to prevent
2398 	 * I/O mapping before resume.
2399 	 * This must be done before setting the queue restrictions,
2400 	 * because request-based dm may be run just after the setting.
2401 	 */
2402 	if (dm_table_request_based(t) && !blk_queue_stopped(q))
2403 		stop_queue(q);
2404 
2405 	__bind_mempools(md, t);
2406 
2407 	merge_is_optional = dm_table_merge_is_optional(t);
2408 
2409 	old_map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2410 	rcu_assign_pointer(md->map, t);
2411 	md->immutable_target_type = dm_table_get_immutable_target_type(t);
2412 
2413 	dm_table_set_restrictions(t, q, limits);
2414 	if (merge_is_optional)
2415 		set_bit(DMF_MERGE_IS_OPTIONAL, &md->flags);
2416 	else
2417 		clear_bit(DMF_MERGE_IS_OPTIONAL, &md->flags);
2418 	if (old_map)
2419 		dm_sync_table(md);
2420 
2421 	return old_map;
2422 }
2423 
2424 /*
2425  * Returns unbound table for the caller to free.
2426  */
2427 static struct dm_table *__unbind(struct mapped_device *md)
2428 {
2429 	struct dm_table *map = rcu_dereference_protected(md->map, 1);
2430 
2431 	if (!map)
2432 		return NULL;
2433 
2434 	dm_table_event_callback(map, NULL, NULL);
2435 	RCU_INIT_POINTER(md->map, NULL);
2436 	dm_sync_table(md);
2437 
2438 	return map;
2439 }
2440 
2441 /*
2442  * Constructor for a new device.
2443  */
2444 int dm_create(int minor, struct mapped_device **result)
2445 {
2446 	struct mapped_device *md;
2447 
2448 	md = alloc_dev(minor);
2449 	if (!md)
2450 		return -ENXIO;
2451 
2452 	dm_sysfs_init(md);
2453 
2454 	*result = md;
2455 	return 0;
2456 }
2457 
2458 /*
2459  * Functions to manage md->type.
2460  * All are required to hold md->type_lock.
2461  */
2462 void dm_lock_md_type(struct mapped_device *md)
2463 {
2464 	mutex_lock(&md->type_lock);
2465 }
2466 
2467 void dm_unlock_md_type(struct mapped_device *md)
2468 {
2469 	mutex_unlock(&md->type_lock);
2470 }
2471 
2472 void dm_set_md_type(struct mapped_device *md, unsigned type)
2473 {
2474 	BUG_ON(!mutex_is_locked(&md->type_lock));
2475 	md->type = type;
2476 }
2477 
2478 unsigned dm_get_md_type(struct mapped_device *md)
2479 {
2480 	BUG_ON(!mutex_is_locked(&md->type_lock));
2481 	return md->type;
2482 }
2483 
2484 static bool dm_md_type_request_based(struct mapped_device *md)
2485 {
2486 	unsigned table_type = dm_get_md_type(md);
2487 
2488 	return (table_type == DM_TYPE_REQUEST_BASED ||
2489 		table_type == DM_TYPE_MQ_REQUEST_BASED);
2490 }
2491 
2492 struct target_type *dm_get_immutable_target_type(struct mapped_device *md)
2493 {
2494 	return md->immutable_target_type;
2495 }
2496 
2497 /*
2498  * The queue_limits are only valid as long as you have a reference
2499  * count on 'md'.
2500  */
2501 struct queue_limits *dm_get_queue_limits(struct mapped_device *md)
2502 {
2503 	BUG_ON(!atomic_read(&md->holders));
2504 	return &md->queue->limits;
2505 }
2506 EXPORT_SYMBOL_GPL(dm_get_queue_limits);
2507 
2508 /*
2509  * Fully initialize a request-based queue (->elevator, ->request_fn, etc).
2510  */
2511 static int dm_init_request_based_queue(struct mapped_device *md)
2512 {
2513 	struct request_queue *q = NULL;
2514 
2515 	if (md->queue->elevator)
2516 		return 1;
2517 
2518 	/* Fully initialize the queue */
2519 	q = blk_init_allocated_queue(md->queue, dm_request_fn, NULL);
2520 	if (!q)
2521 		return 0;
2522 
2523 	md->queue = q;
2524 	dm_init_md_queue(md);
2525 	blk_queue_softirq_done(md->queue, dm_softirq_done);
2526 	blk_queue_prep_rq(md->queue, dm_prep_fn);
2527 
2528 	/* Also initialize the request-based DM worker thread */
2529 	init_kthread_worker(&md->kworker);
2530 	md->kworker_task = kthread_run(kthread_worker_fn, &md->kworker,
2531 				       "kdmwork-%s", dm_device_name(md));
2532 
2533 	elv_register_queue(md->queue);
2534 
2535 	return 1;
2536 }
2537 
2538 /*
2539  * Setup the DM device's queue based on md's type
2540  */
2541 int dm_setup_md_queue(struct mapped_device *md)
2542 {
2543 	if (dm_md_type_request_based(md)) {
2544 		if (!dm_init_request_based_queue(md)) {
2545 			DMWARN("Cannot initialize queue for request-based mapped device");
2546 			return -EINVAL;
2547 		}
2548 	} else {
2549 		/* bio-based specific initialization */
2550 		blk_queue_make_request(md->queue, dm_make_request);
2551 		blk_queue_merge_bvec(md->queue, dm_merge_bvec);
2552 	}
2553 
2554 	return 0;
2555 }
2556 
2557 struct mapped_device *dm_get_md(dev_t dev)
2558 {
2559 	struct mapped_device *md;
2560 	unsigned minor = MINOR(dev);
2561 
2562 	if (MAJOR(dev) != _major || minor >= (1 << MINORBITS))
2563 		return NULL;
2564 
2565 	spin_lock(&_minor_lock);
2566 
2567 	md = idr_find(&_minor_idr, minor);
2568 	if (md) {
2569 		if ((md == MINOR_ALLOCED ||
2570 		     (MINOR(disk_devt(dm_disk(md))) != minor) ||
2571 		     dm_deleting_md(md) ||
2572 		     test_bit(DMF_FREEING, &md->flags))) {
2573 			md = NULL;
2574 			goto out;
2575 		}
2576 		dm_get(md);
2577 	}
2578 
2579 out:
2580 	spin_unlock(&_minor_lock);
2581 
2582 	return md;
2583 }
2584 EXPORT_SYMBOL_GPL(dm_get_md);
2585 
2586 void *dm_get_mdptr(struct mapped_device *md)
2587 {
2588 	return md->interface_ptr;
2589 }
2590 
2591 void dm_set_mdptr(struct mapped_device *md, void *ptr)
2592 {
2593 	md->interface_ptr = ptr;
2594 }
2595 
2596 void dm_get(struct mapped_device *md)
2597 {
2598 	atomic_inc(&md->holders);
2599 	BUG_ON(test_bit(DMF_FREEING, &md->flags));
2600 }
2601 
2602 int dm_hold(struct mapped_device *md)
2603 {
2604 	spin_lock(&_minor_lock);
2605 	if (test_bit(DMF_FREEING, &md->flags)) {
2606 		spin_unlock(&_minor_lock);
2607 		return -EBUSY;
2608 	}
2609 	dm_get(md);
2610 	spin_unlock(&_minor_lock);
2611 	return 0;
2612 }
2613 EXPORT_SYMBOL_GPL(dm_hold);
2614 
2615 const char *dm_device_name(struct mapped_device *md)
2616 {
2617 	return md->name;
2618 }
2619 EXPORT_SYMBOL_GPL(dm_device_name);
2620 
2621 static void __dm_destroy(struct mapped_device *md, bool wait)
2622 {
2623 	struct dm_table *map;
2624 	int srcu_idx;
2625 
2626 	might_sleep();
2627 
2628 	map = dm_get_live_table(md, &srcu_idx);
2629 
2630 	spin_lock(&_minor_lock);
2631 	idr_replace(&_minor_idr, MINOR_ALLOCED, MINOR(disk_devt(dm_disk(md))));
2632 	set_bit(DMF_FREEING, &md->flags);
2633 	spin_unlock(&_minor_lock);
2634 
2635 	if (dm_request_based(md))
2636 		flush_kthread_worker(&md->kworker);
2637 
2638 	/*
2639 	 * Take suspend_lock so that presuspend and postsuspend methods
2640 	 * do not race with internal suspend.
2641 	 */
2642 	mutex_lock(&md->suspend_lock);
2643 	if (!dm_suspended_md(md)) {
2644 		dm_table_presuspend_targets(map);
2645 		dm_table_postsuspend_targets(map);
2646 	}
2647 	mutex_unlock(&md->suspend_lock);
2648 
2649 	/* dm_put_live_table must be before msleep, otherwise deadlock is possible */
2650 	dm_put_live_table(md, srcu_idx);
2651 
2652 	/*
2653 	 * Rare, but there may be I/O requests still going to complete,
2654 	 * for example.  Wait for all references to disappear.
2655 	 * No one should increment the reference count of the mapped_device,
2656 	 * after the mapped_device state becomes DMF_FREEING.
2657 	 */
2658 	if (wait)
2659 		while (atomic_read(&md->holders))
2660 			msleep(1);
2661 	else if (atomic_read(&md->holders))
2662 		DMWARN("%s: Forcibly removing mapped_device still in use! (%d users)",
2663 		       dm_device_name(md), atomic_read(&md->holders));
2664 
2665 	dm_sysfs_exit(md);
2666 	dm_table_destroy(__unbind(md));
2667 	free_dev(md);
2668 }
2669 
2670 void dm_destroy(struct mapped_device *md)
2671 {
2672 	__dm_destroy(md, true);
2673 }
2674 
2675 void dm_destroy_immediate(struct mapped_device *md)
2676 {
2677 	__dm_destroy(md, false);
2678 }
2679 
2680 void dm_put(struct mapped_device *md)
2681 {
2682 	atomic_dec(&md->holders);
2683 }
2684 EXPORT_SYMBOL_GPL(dm_put);
2685 
2686 static int dm_wait_for_completion(struct mapped_device *md, int interruptible)
2687 {
2688 	int r = 0;
2689 	DECLARE_WAITQUEUE(wait, current);
2690 
2691 	add_wait_queue(&md->wait, &wait);
2692 
2693 	while (1) {
2694 		set_current_state(interruptible);
2695 
2696 		if (!md_in_flight(md))
2697 			break;
2698 
2699 		if (interruptible == TASK_INTERRUPTIBLE &&
2700 		    signal_pending(current)) {
2701 			r = -EINTR;
2702 			break;
2703 		}
2704 
2705 		io_schedule();
2706 	}
2707 	set_current_state(TASK_RUNNING);
2708 
2709 	remove_wait_queue(&md->wait, &wait);
2710 
2711 	return r;
2712 }
2713 
2714 /*
2715  * Process the deferred bios
2716  */
2717 static void dm_wq_work(struct work_struct *work)
2718 {
2719 	struct mapped_device *md = container_of(work, struct mapped_device,
2720 						work);
2721 	struct bio *c;
2722 	int srcu_idx;
2723 	struct dm_table *map;
2724 
2725 	map = dm_get_live_table(md, &srcu_idx);
2726 
2727 	while (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
2728 		spin_lock_irq(&md->deferred_lock);
2729 		c = bio_list_pop(&md->deferred);
2730 		spin_unlock_irq(&md->deferred_lock);
2731 
2732 		if (!c)
2733 			break;
2734 
2735 		if (dm_request_based(md))
2736 			generic_make_request(c);
2737 		else
2738 			__split_and_process_bio(md, map, c);
2739 	}
2740 
2741 	dm_put_live_table(md, srcu_idx);
2742 }
2743 
2744 static void dm_queue_flush(struct mapped_device *md)
2745 {
2746 	clear_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2747 	smp_mb__after_atomic();
2748 	queue_work(md->wq, &md->work);
2749 }
2750 
2751 /*
2752  * Swap in a new table, returning the old one for the caller to destroy.
2753  */
2754 struct dm_table *dm_swap_table(struct mapped_device *md, struct dm_table *table)
2755 {
2756 	struct dm_table *live_map = NULL, *map = ERR_PTR(-EINVAL);
2757 	struct queue_limits limits;
2758 	int r;
2759 
2760 	mutex_lock(&md->suspend_lock);
2761 
2762 	/* device must be suspended */
2763 	if (!dm_suspended_md(md))
2764 		goto out;
2765 
2766 	/*
2767 	 * If the new table has no data devices, retain the existing limits.
2768 	 * This helps multipath with queue_if_no_path if all paths disappear,
2769 	 * then new I/O is queued based on these limits, and then some paths
2770 	 * reappear.
2771 	 */
2772 	if (dm_table_has_no_data_devices(table)) {
2773 		live_map = dm_get_live_table_fast(md);
2774 		if (live_map)
2775 			limits = md->queue->limits;
2776 		dm_put_live_table_fast(md);
2777 	}
2778 
2779 	if (!live_map) {
2780 		r = dm_calculate_queue_limits(table, &limits);
2781 		if (r) {
2782 			map = ERR_PTR(r);
2783 			goto out;
2784 		}
2785 	}
2786 
2787 	map = __bind(md, table, &limits);
2788 
2789 out:
2790 	mutex_unlock(&md->suspend_lock);
2791 	return map;
2792 }
2793 
2794 /*
2795  * Functions to lock and unlock any filesystem running on the
2796  * device.
2797  */
2798 static int lock_fs(struct mapped_device *md)
2799 {
2800 	int r;
2801 
2802 	WARN_ON(md->frozen_sb);
2803 
2804 	md->frozen_sb = freeze_bdev(md->bdev);
2805 	if (IS_ERR(md->frozen_sb)) {
2806 		r = PTR_ERR(md->frozen_sb);
2807 		md->frozen_sb = NULL;
2808 		return r;
2809 	}
2810 
2811 	set_bit(DMF_FROZEN, &md->flags);
2812 
2813 	return 0;
2814 }
2815 
2816 static void unlock_fs(struct mapped_device *md)
2817 {
2818 	if (!test_bit(DMF_FROZEN, &md->flags))
2819 		return;
2820 
2821 	thaw_bdev(md->bdev, md->frozen_sb);
2822 	md->frozen_sb = NULL;
2823 	clear_bit(DMF_FROZEN, &md->flags);
2824 }
2825 
2826 /*
2827  * If __dm_suspend returns 0, the device is completely quiescent
2828  * now. There is no request-processing activity. All new requests
2829  * are being added to md->deferred list.
2830  *
2831  * Caller must hold md->suspend_lock
2832  */
2833 static int __dm_suspend(struct mapped_device *md, struct dm_table *map,
2834 			unsigned suspend_flags, int interruptible)
2835 {
2836 	bool do_lockfs = suspend_flags & DM_SUSPEND_LOCKFS_FLAG;
2837 	bool noflush = suspend_flags & DM_SUSPEND_NOFLUSH_FLAG;
2838 	int r;
2839 
2840 	/*
2841 	 * DMF_NOFLUSH_SUSPENDING must be set before presuspend.
2842 	 * This flag is cleared before dm_suspend returns.
2843 	 */
2844 	if (noflush)
2845 		set_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
2846 
2847 	/*
2848 	 * This gets reverted if there's an error later and the targets
2849 	 * provide the .presuspend_undo hook.
2850 	 */
2851 	dm_table_presuspend_targets(map);
2852 
2853 	/*
2854 	 * Flush I/O to the device.
2855 	 * Any I/O submitted after lock_fs() may not be flushed.
2856 	 * noflush takes precedence over do_lockfs.
2857 	 * (lock_fs() flushes I/Os and waits for them to complete.)
2858 	 */
2859 	if (!noflush && do_lockfs) {
2860 		r = lock_fs(md);
2861 		if (r) {
2862 			dm_table_presuspend_undo_targets(map);
2863 			return r;
2864 		}
2865 	}
2866 
2867 	/*
2868 	 * Here we must make sure that no processes are submitting requests
2869 	 * to target drivers i.e. no one may be executing
2870 	 * __split_and_process_bio. This is called from dm_request and
2871 	 * dm_wq_work.
2872 	 *
2873 	 * To get all processes out of __split_and_process_bio in dm_request,
2874 	 * we take the write lock. To prevent any process from reentering
2875 	 * __split_and_process_bio from dm_request and quiesce the thread
2876 	 * (dm_wq_work), we set BMF_BLOCK_IO_FOR_SUSPEND and call
2877 	 * flush_workqueue(md->wq).
2878 	 */
2879 	set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2880 	if (map)
2881 		synchronize_srcu(&md->io_barrier);
2882 
2883 	/*
2884 	 * Stop md->queue before flushing md->wq in case request-based
2885 	 * dm defers requests to md->wq from md->queue.
2886 	 */
2887 	if (dm_request_based(md)) {
2888 		stop_queue(md->queue);
2889 		flush_kthread_worker(&md->kworker);
2890 	}
2891 
2892 	flush_workqueue(md->wq);
2893 
2894 	/*
2895 	 * At this point no more requests are entering target request routines.
2896 	 * We call dm_wait_for_completion to wait for all existing requests
2897 	 * to finish.
2898 	 */
2899 	r = dm_wait_for_completion(md, interruptible);
2900 
2901 	if (noflush)
2902 		clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
2903 	if (map)
2904 		synchronize_srcu(&md->io_barrier);
2905 
2906 	/* were we interrupted ? */
2907 	if (r < 0) {
2908 		dm_queue_flush(md);
2909 
2910 		if (dm_request_based(md))
2911 			start_queue(md->queue);
2912 
2913 		unlock_fs(md);
2914 		dm_table_presuspend_undo_targets(map);
2915 		/* pushback list is already flushed, so skip flush */
2916 	}
2917 
2918 	return r;
2919 }
2920 
2921 /*
2922  * We need to be able to change a mapping table under a mounted
2923  * filesystem.  For example we might want to move some data in
2924  * the background.  Before the table can be swapped with
2925  * dm_bind_table, dm_suspend must be called to flush any in
2926  * flight bios and ensure that any further io gets deferred.
2927  */
2928 /*
2929  * Suspend mechanism in request-based dm.
2930  *
2931  * 1. Flush all I/Os by lock_fs() if needed.
2932  * 2. Stop dispatching any I/O by stopping the request_queue.
2933  * 3. Wait for all in-flight I/Os to be completed or requeued.
2934  *
2935  * To abort suspend, start the request_queue.
2936  */
2937 int dm_suspend(struct mapped_device *md, unsigned suspend_flags)
2938 {
2939 	struct dm_table *map = NULL;
2940 	int r = 0;
2941 
2942 retry:
2943 	mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING);
2944 
2945 	if (dm_suspended_md(md)) {
2946 		r = -EINVAL;
2947 		goto out_unlock;
2948 	}
2949 
2950 	if (dm_suspended_internally_md(md)) {
2951 		/* already internally suspended, wait for internal resume */
2952 		mutex_unlock(&md->suspend_lock);
2953 		r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE);
2954 		if (r)
2955 			return r;
2956 		goto retry;
2957 	}
2958 
2959 	map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2960 
2961 	r = __dm_suspend(md, map, suspend_flags, TASK_INTERRUPTIBLE);
2962 	if (r)
2963 		goto out_unlock;
2964 
2965 	set_bit(DMF_SUSPENDED, &md->flags);
2966 
2967 	dm_table_postsuspend_targets(map);
2968 
2969 out_unlock:
2970 	mutex_unlock(&md->suspend_lock);
2971 	return r;
2972 }
2973 
2974 static int __dm_resume(struct mapped_device *md, struct dm_table *map)
2975 {
2976 	if (map) {
2977 		int r = dm_table_resume_targets(map);
2978 		if (r)
2979 			return r;
2980 	}
2981 
2982 	dm_queue_flush(md);
2983 
2984 	/*
2985 	 * Flushing deferred I/Os must be done after targets are resumed
2986 	 * so that mapping of targets can work correctly.
2987 	 * Request-based dm is queueing the deferred I/Os in its request_queue.
2988 	 */
2989 	if (dm_request_based(md))
2990 		start_queue(md->queue);
2991 
2992 	unlock_fs(md);
2993 
2994 	return 0;
2995 }
2996 
2997 int dm_resume(struct mapped_device *md)
2998 {
2999 	int r = -EINVAL;
3000 	struct dm_table *map = NULL;
3001 
3002 retry:
3003 	mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING);
3004 
3005 	if (!dm_suspended_md(md))
3006 		goto out;
3007 
3008 	if (dm_suspended_internally_md(md)) {
3009 		/* already internally suspended, wait for internal resume */
3010 		mutex_unlock(&md->suspend_lock);
3011 		r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE);
3012 		if (r)
3013 			return r;
3014 		goto retry;
3015 	}
3016 
3017 	map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
3018 	if (!map || !dm_table_get_size(map))
3019 		goto out;
3020 
3021 	r = __dm_resume(md, map);
3022 	if (r)
3023 		goto out;
3024 
3025 	clear_bit(DMF_SUSPENDED, &md->flags);
3026 
3027 	r = 0;
3028 out:
3029 	mutex_unlock(&md->suspend_lock);
3030 
3031 	return r;
3032 }
3033 
3034 /*
3035  * Internal suspend/resume works like userspace-driven suspend. It waits
3036  * until all bios finish and prevents issuing new bios to the target drivers.
3037  * It may be used only from the kernel.
3038  */
3039 
3040 static void __dm_internal_suspend(struct mapped_device *md, unsigned suspend_flags)
3041 {
3042 	struct dm_table *map = NULL;
3043 
3044 	if (md->internal_suspend_count++)
3045 		return; /* nested internal suspend */
3046 
3047 	if (dm_suspended_md(md)) {
3048 		set_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
3049 		return; /* nest suspend */
3050 	}
3051 
3052 	map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
3053 
3054 	/*
3055 	 * Using TASK_UNINTERRUPTIBLE because only NOFLUSH internal suspend is
3056 	 * supported.  Properly supporting a TASK_INTERRUPTIBLE internal suspend
3057 	 * would require changing .presuspend to return an error -- avoid this
3058 	 * until there is a need for more elaborate variants of internal suspend.
3059 	 */
3060 	(void) __dm_suspend(md, map, suspend_flags, TASK_UNINTERRUPTIBLE);
3061 
3062 	set_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
3063 
3064 	dm_table_postsuspend_targets(map);
3065 }
3066 
3067 static void __dm_internal_resume(struct mapped_device *md)
3068 {
3069 	BUG_ON(!md->internal_suspend_count);
3070 
3071 	if (--md->internal_suspend_count)
3072 		return; /* resume from nested internal suspend */
3073 
3074 	if (dm_suspended_md(md))
3075 		goto done; /* resume from nested suspend */
3076 
3077 	/*
3078 	 * NOTE: existing callers don't need to call dm_table_resume_targets
3079 	 * (which may fail -- so best to avoid it for now by passing NULL map)
3080 	 */
3081 	(void) __dm_resume(md, NULL);
3082 
3083 done:
3084 	clear_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
3085 	smp_mb__after_atomic();
3086 	wake_up_bit(&md->flags, DMF_SUSPENDED_INTERNALLY);
3087 }
3088 
3089 void dm_internal_suspend_noflush(struct mapped_device *md)
3090 {
3091 	mutex_lock(&md->suspend_lock);
3092 	__dm_internal_suspend(md, DM_SUSPEND_NOFLUSH_FLAG);
3093 	mutex_unlock(&md->suspend_lock);
3094 }
3095 EXPORT_SYMBOL_GPL(dm_internal_suspend_noflush);
3096 
3097 void dm_internal_resume(struct mapped_device *md)
3098 {
3099 	mutex_lock(&md->suspend_lock);
3100 	__dm_internal_resume(md);
3101 	mutex_unlock(&md->suspend_lock);
3102 }
3103 EXPORT_SYMBOL_GPL(dm_internal_resume);
3104 
3105 /*
3106  * Fast variants of internal suspend/resume hold md->suspend_lock,
3107  * which prevents interaction with userspace-driven suspend.
3108  */
3109 
3110 void dm_internal_suspend_fast(struct mapped_device *md)
3111 {
3112 	mutex_lock(&md->suspend_lock);
3113 	if (dm_suspended_md(md) || dm_suspended_internally_md(md))
3114 		return;
3115 
3116 	set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
3117 	synchronize_srcu(&md->io_barrier);
3118 	flush_workqueue(md->wq);
3119 	dm_wait_for_completion(md, TASK_UNINTERRUPTIBLE);
3120 }
3121 EXPORT_SYMBOL_GPL(dm_internal_suspend_fast);
3122 
3123 void dm_internal_resume_fast(struct mapped_device *md)
3124 {
3125 	if (dm_suspended_md(md) || dm_suspended_internally_md(md))
3126 		goto done;
3127 
3128 	dm_queue_flush(md);
3129 
3130 done:
3131 	mutex_unlock(&md->suspend_lock);
3132 }
3133 EXPORT_SYMBOL_GPL(dm_internal_resume_fast);
3134 
3135 /*-----------------------------------------------------------------
3136  * Event notification.
3137  *---------------------------------------------------------------*/
3138 int dm_kobject_uevent(struct mapped_device *md, enum kobject_action action,
3139 		       unsigned cookie)
3140 {
3141 	char udev_cookie[DM_COOKIE_LENGTH];
3142 	char *envp[] = { udev_cookie, NULL };
3143 
3144 	if (!cookie)
3145 		return kobject_uevent(&disk_to_dev(md->disk)->kobj, action);
3146 	else {
3147 		snprintf(udev_cookie, DM_COOKIE_LENGTH, "%s=%u",
3148 			 DM_COOKIE_ENV_VAR_NAME, cookie);
3149 		return kobject_uevent_env(&disk_to_dev(md->disk)->kobj,
3150 					  action, envp);
3151 	}
3152 }
3153 
3154 uint32_t dm_next_uevent_seq(struct mapped_device *md)
3155 {
3156 	return atomic_add_return(1, &md->uevent_seq);
3157 }
3158 
3159 uint32_t dm_get_event_nr(struct mapped_device *md)
3160 {
3161 	return atomic_read(&md->event_nr);
3162 }
3163 
3164 int dm_wait_event(struct mapped_device *md, int event_nr)
3165 {
3166 	return wait_event_interruptible(md->eventq,
3167 			(event_nr != atomic_read(&md->event_nr)));
3168 }
3169 
3170 void dm_uevent_add(struct mapped_device *md, struct list_head *elist)
3171 {
3172 	unsigned long flags;
3173 
3174 	spin_lock_irqsave(&md->uevent_lock, flags);
3175 	list_add(elist, &md->uevent_list);
3176 	spin_unlock_irqrestore(&md->uevent_lock, flags);
3177 }
3178 
3179 /*
3180  * The gendisk is only valid as long as you have a reference
3181  * count on 'md'.
3182  */
3183 struct gendisk *dm_disk(struct mapped_device *md)
3184 {
3185 	return md->disk;
3186 }
3187 
3188 struct kobject *dm_kobject(struct mapped_device *md)
3189 {
3190 	return &md->kobj_holder.kobj;
3191 }
3192 
3193 struct mapped_device *dm_get_from_kobject(struct kobject *kobj)
3194 {
3195 	struct mapped_device *md;
3196 
3197 	md = container_of(kobj, struct mapped_device, kobj_holder.kobj);
3198 
3199 	if (test_bit(DMF_FREEING, &md->flags) ||
3200 	    dm_deleting_md(md))
3201 		return NULL;
3202 
3203 	dm_get(md);
3204 	return md;
3205 }
3206 
3207 int dm_suspended_md(struct mapped_device *md)
3208 {
3209 	return test_bit(DMF_SUSPENDED, &md->flags);
3210 }
3211 
3212 int dm_suspended_internally_md(struct mapped_device *md)
3213 {
3214 	return test_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
3215 }
3216 
3217 int dm_test_deferred_remove_flag(struct mapped_device *md)
3218 {
3219 	return test_bit(DMF_DEFERRED_REMOVE, &md->flags);
3220 }
3221 
3222 int dm_suspended(struct dm_target *ti)
3223 {
3224 	return dm_suspended_md(dm_table_get_md(ti->table));
3225 }
3226 EXPORT_SYMBOL_GPL(dm_suspended);
3227 
3228 int dm_noflush_suspending(struct dm_target *ti)
3229 {
3230 	return __noflush_suspending(dm_table_get_md(ti->table));
3231 }
3232 EXPORT_SYMBOL_GPL(dm_noflush_suspending);
3233 
3234 struct dm_md_mempools *dm_alloc_md_mempools(unsigned type, unsigned integrity, unsigned per_bio_data_size)
3235 {
3236 	struct dm_md_mempools *pools = kzalloc(sizeof(*pools), GFP_KERNEL);
3237 	struct kmem_cache *cachep;
3238 	unsigned int pool_size = 0;
3239 	unsigned int front_pad;
3240 
3241 	if (!pools)
3242 		return NULL;
3243 
3244 	switch (type) {
3245 	case DM_TYPE_BIO_BASED:
3246 		cachep = _io_cache;
3247 		pool_size = dm_get_reserved_bio_based_ios();
3248 		front_pad = roundup(per_bio_data_size, __alignof__(struct dm_target_io)) + offsetof(struct dm_target_io, clone);
3249 		break;
3250 	case DM_TYPE_REQUEST_BASED:
3251 		pool_size = dm_get_reserved_rq_based_ios();
3252 		pools->rq_pool = mempool_create_slab_pool(pool_size, _rq_cache);
3253 		if (!pools->rq_pool)
3254 			goto out;
3255 		/* fall through to setup remaining rq-based pools */
3256 	case DM_TYPE_MQ_REQUEST_BASED:
3257 		cachep = _rq_tio_cache;
3258 		if (!pool_size)
3259 			pool_size = dm_get_reserved_rq_based_ios();
3260 		front_pad = offsetof(struct dm_rq_clone_bio_info, clone);
3261 		/* per_bio_data_size is not used. See __bind_mempools(). */
3262 		WARN_ON(per_bio_data_size != 0);
3263 		break;
3264 	default:
3265 		goto out;
3266 	}
3267 
3268 	pools->io_pool = mempool_create_slab_pool(pool_size, cachep);
3269 	if (!pools->io_pool)
3270 		goto out;
3271 
3272 	pools->bs = bioset_create_nobvec(pool_size, front_pad);
3273 	if (!pools->bs)
3274 		goto out;
3275 
3276 	if (integrity && bioset_integrity_create(pools->bs, pool_size))
3277 		goto out;
3278 
3279 	return pools;
3280 
3281 out:
3282 	dm_free_md_mempools(pools);
3283 
3284 	return NULL;
3285 }
3286 
3287 void dm_free_md_mempools(struct dm_md_mempools *pools)
3288 {
3289 	if (!pools)
3290 		return;
3291 
3292 	if (pools->io_pool)
3293 		mempool_destroy(pools->io_pool);
3294 
3295 	if (pools->rq_pool)
3296 		mempool_destroy(pools->rq_pool);
3297 
3298 	if (pools->bs)
3299 		bioset_free(pools->bs);
3300 
3301 	kfree(pools);
3302 }
3303 
3304 static const struct block_device_operations dm_blk_dops = {
3305 	.open = dm_blk_open,
3306 	.release = dm_blk_close,
3307 	.ioctl = dm_blk_ioctl,
3308 	.getgeo = dm_blk_getgeo,
3309 	.owner = THIS_MODULE
3310 };
3311 
3312 /*
3313  * module hooks
3314  */
3315 module_init(dm_init);
3316 module_exit(dm_exit);
3317 
3318 module_param(major, uint, 0);
3319 MODULE_PARM_DESC(major, "The major number of the device mapper");
3320 
3321 module_param(reserved_bio_based_ios, uint, S_IRUGO | S_IWUSR);
3322 MODULE_PARM_DESC(reserved_bio_based_ios, "Reserved IOs in bio-based mempools");
3323 
3324 module_param(reserved_rq_based_ios, uint, S_IRUGO | S_IWUSR);
3325 MODULE_PARM_DESC(reserved_rq_based_ios, "Reserved IOs in request-based mempools");
3326 
3327 MODULE_DESCRIPTION(DM_NAME " driver");
3328 MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
3329 MODULE_LICENSE("GPL");
3330