xref: /kernel/linux/linux-5.10/drivers/md/dm.c (revision 8c2ecf20)
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-core.h"
9#include "dm-rq.h"
10#include "dm-uevent.h"
11
12#include <linux/init.h>
13#include <linux/module.h>
14#include <linux/mutex.h>
15#include <linux/sched/mm.h>
16#include <linux/sched/signal.h>
17#include <linux/blkpg.h>
18#include <linux/bio.h>
19#include <linux/mempool.h>
20#include <linux/dax.h>
21#include <linux/slab.h>
22#include <linux/idr.h>
23#include <linux/uio.h>
24#include <linux/hdreg.h>
25#include <linux/delay.h>
26#include <linux/wait.h>
27#include <linux/pr.h>
28#include <linux/refcount.h>
29#include <linux/part_stat.h>
30#include <linux/blk-crypto.h>
31
32#define DM_MSG_PREFIX "core"
33
34/*
35 * Cookies are numeric values sent with CHANGE and REMOVE
36 * uevents while resuming, removing or renaming the device.
37 */
38#define DM_COOKIE_ENV_VAR_NAME "DM_COOKIE"
39#define DM_COOKIE_LENGTH 24
40
41static const char *_name = DM_NAME;
42
43static unsigned int major = 0;
44static unsigned int _major = 0;
45
46static DEFINE_IDR(_minor_idr);
47
48static DEFINE_SPINLOCK(_minor_lock);
49
50static void do_deferred_remove(struct work_struct *w);
51
52static DECLARE_WORK(deferred_remove_work, do_deferred_remove);
53
54static struct workqueue_struct *deferred_remove_workqueue;
55
56atomic_t dm_global_event_nr = ATOMIC_INIT(0);
57DECLARE_WAIT_QUEUE_HEAD(dm_global_eventq);
58
59void dm_issue_global_event(void)
60{
61	atomic_inc(&dm_global_event_nr);
62	wake_up(&dm_global_eventq);
63}
64
65/*
66 * One of these is allocated (on-stack) per original bio.
67 */
68struct clone_info {
69	struct dm_table *map;
70	struct bio *bio;
71	struct dm_io *io;
72	sector_t sector;
73	unsigned sector_count;
74};
75
76/*
77 * One of these is allocated per clone bio.
78 */
79#define DM_TIO_MAGIC 7282014
80struct dm_target_io {
81	unsigned magic;
82	struct dm_io *io;
83	struct dm_target *ti;
84	unsigned target_bio_nr;
85	unsigned *len_ptr;
86	bool inside_dm_io;
87	struct bio clone;
88};
89
90/*
91 * One of these is allocated per original bio.
92 * It contains the first clone used for that original.
93 */
94#define DM_IO_MAGIC 5191977
95struct dm_io {
96	unsigned magic;
97	struct mapped_device *md;
98	blk_status_t status;
99	atomic_t io_count;
100	struct bio *orig_bio;
101	unsigned long start_time;
102	spinlock_t endio_lock;
103	struct dm_stats_aux stats_aux;
104	/* last member of dm_target_io is 'struct bio' */
105	struct dm_target_io tio;
106};
107
108void *dm_per_bio_data(struct bio *bio, size_t data_size)
109{
110	struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone);
111	if (!tio->inside_dm_io)
112		return (char *)bio - offsetof(struct dm_target_io, clone) - data_size;
113	return (char *)bio - offsetof(struct dm_target_io, clone) - offsetof(struct dm_io, tio) - data_size;
114}
115EXPORT_SYMBOL_GPL(dm_per_bio_data);
116
117struct bio *dm_bio_from_per_bio_data(void *data, size_t data_size)
118{
119	struct dm_io *io = (struct dm_io *)((char *)data + data_size);
120	if (io->magic == DM_IO_MAGIC)
121		return (struct bio *)((char *)io + offsetof(struct dm_io, tio) + offsetof(struct dm_target_io, clone));
122	BUG_ON(io->magic != DM_TIO_MAGIC);
123	return (struct bio *)((char *)io + offsetof(struct dm_target_io, clone));
124}
125EXPORT_SYMBOL_GPL(dm_bio_from_per_bio_data);
126
127unsigned dm_bio_get_target_bio_nr(const struct bio *bio)
128{
129	return container_of(bio, struct dm_target_io, clone)->target_bio_nr;
130}
131EXPORT_SYMBOL_GPL(dm_bio_get_target_bio_nr);
132
133#define MINOR_ALLOCED ((void *)-1)
134
135/*
136 * Bits for the md->flags field.
137 */
138#define DMF_BLOCK_IO_FOR_SUSPEND 0
139#define DMF_SUSPENDED 1
140#define DMF_FROZEN 2
141#define DMF_FREEING 3
142#define DMF_DELETING 4
143#define DMF_NOFLUSH_SUSPENDING 5
144#define DMF_DEFERRED_REMOVE 6
145#define DMF_SUSPENDED_INTERNALLY 7
146#define DMF_POST_SUSPENDING 8
147
148#define DM_NUMA_NODE NUMA_NO_NODE
149static int dm_numa_node = DM_NUMA_NODE;
150
151#define DEFAULT_SWAP_BIOS	(8 * 1048576 / PAGE_SIZE)
152static int swap_bios = DEFAULT_SWAP_BIOS;
153static int get_swap_bios(void)
154{
155	int latch = READ_ONCE(swap_bios);
156	if (unlikely(latch <= 0))
157		latch = DEFAULT_SWAP_BIOS;
158	return latch;
159}
160
161/*
162 * For mempools pre-allocation at the table loading time.
163 */
164struct dm_md_mempools {
165	struct bio_set bs;
166	struct bio_set io_bs;
167};
168
169struct table_device {
170	struct list_head list;
171	refcount_t count;
172	struct dm_dev dm_dev;
173};
174
175/*
176 * Bio-based DM's mempools' reserved IOs set by the user.
177 */
178#define RESERVED_BIO_BASED_IOS		16
179static unsigned reserved_bio_based_ios = RESERVED_BIO_BASED_IOS;
180
181static int __dm_get_module_param_int(int *module_param, int min, int max)
182{
183	int param = READ_ONCE(*module_param);
184	int modified_param = 0;
185	bool modified = true;
186
187	if (param < min)
188		modified_param = min;
189	else if (param > max)
190		modified_param = max;
191	else
192		modified = false;
193
194	if (modified) {
195		(void)cmpxchg(module_param, param, modified_param);
196		param = modified_param;
197	}
198
199	return param;
200}
201
202unsigned __dm_get_module_param(unsigned *module_param,
203			       unsigned def, unsigned max)
204{
205	unsigned param = READ_ONCE(*module_param);
206	unsigned modified_param = 0;
207
208	if (!param)
209		modified_param = def;
210	else if (param > max)
211		modified_param = max;
212
213	if (modified_param) {
214		(void)cmpxchg(module_param, param, modified_param);
215		param = modified_param;
216	}
217
218	return param;
219}
220
221unsigned dm_get_reserved_bio_based_ios(void)
222{
223	return __dm_get_module_param(&reserved_bio_based_ios,
224				     RESERVED_BIO_BASED_IOS, DM_RESERVED_MAX_IOS);
225}
226EXPORT_SYMBOL_GPL(dm_get_reserved_bio_based_ios);
227
228static unsigned dm_get_numa_node(void)
229{
230	return __dm_get_module_param_int(&dm_numa_node,
231					 DM_NUMA_NODE, num_online_nodes() - 1);
232}
233
234static int __init local_init(void)
235{
236	int r;
237
238	r = dm_uevent_init();
239	if (r)
240		return r;
241
242	deferred_remove_workqueue = alloc_workqueue("kdmremove", WQ_UNBOUND, 1);
243	if (!deferred_remove_workqueue) {
244		r = -ENOMEM;
245		goto out_uevent_exit;
246	}
247
248	_major = major;
249	r = register_blkdev(_major, _name);
250	if (r < 0)
251		goto out_free_workqueue;
252
253	if (!_major)
254		_major = r;
255
256	return 0;
257
258out_free_workqueue:
259	destroy_workqueue(deferred_remove_workqueue);
260out_uevent_exit:
261	dm_uevent_exit();
262
263	return r;
264}
265
266static void local_exit(void)
267{
268	destroy_workqueue(deferred_remove_workqueue);
269
270	unregister_blkdev(_major, _name);
271	dm_uevent_exit();
272
273	_major = 0;
274
275	DMINFO("cleaned up");
276}
277
278static int (*_inits[])(void) __initdata = {
279	local_init,
280	dm_target_init,
281	dm_linear_init,
282	dm_stripe_init,
283	dm_io_init,
284	dm_kcopyd_init,
285	dm_interface_init,
286	dm_statistics_init,
287};
288
289static void (*_exits[])(void) = {
290	local_exit,
291	dm_target_exit,
292	dm_linear_exit,
293	dm_stripe_exit,
294	dm_io_exit,
295	dm_kcopyd_exit,
296	dm_interface_exit,
297	dm_statistics_exit,
298};
299
300static int __init dm_init(void)
301{
302	const int count = ARRAY_SIZE(_inits);
303
304	int r, i;
305
306	for (i = 0; i < count; i++) {
307		r = _inits[i]();
308		if (r)
309			goto bad;
310	}
311
312	return 0;
313
314      bad:
315	while (i--)
316		_exits[i]();
317
318	return r;
319}
320
321static void __exit dm_exit(void)
322{
323	int i = ARRAY_SIZE(_exits);
324
325	while (i--)
326		_exits[i]();
327
328	/*
329	 * Should be empty by this point.
330	 */
331	idr_destroy(&_minor_idr);
332}
333
334/*
335 * Block device functions
336 */
337int dm_deleting_md(struct mapped_device *md)
338{
339	return test_bit(DMF_DELETING, &md->flags);
340}
341
342static int dm_blk_open(struct block_device *bdev, fmode_t mode)
343{
344	struct mapped_device *md;
345
346	spin_lock(&_minor_lock);
347
348	md = bdev->bd_disk->private_data;
349	if (!md)
350		goto out;
351
352	if (test_bit(DMF_FREEING, &md->flags) ||
353	    dm_deleting_md(md)) {
354		md = NULL;
355		goto out;
356	}
357
358	dm_get(md);
359	atomic_inc(&md->open_count);
360out:
361	spin_unlock(&_minor_lock);
362
363	return md ? 0 : -ENXIO;
364}
365
366static void dm_blk_close(struct gendisk *disk, fmode_t mode)
367{
368	struct mapped_device *md;
369
370	spin_lock(&_minor_lock);
371
372	md = disk->private_data;
373	if (WARN_ON(!md))
374		goto out;
375
376	if (atomic_dec_and_test(&md->open_count) &&
377	    (test_bit(DMF_DEFERRED_REMOVE, &md->flags)))
378		queue_work(deferred_remove_workqueue, &deferred_remove_work);
379
380	dm_put(md);
381out:
382	spin_unlock(&_minor_lock);
383}
384
385int dm_open_count(struct mapped_device *md)
386{
387	return atomic_read(&md->open_count);
388}
389
390/*
391 * Guarantees nothing is using the device before it's deleted.
392 */
393int dm_lock_for_deletion(struct mapped_device *md, bool mark_deferred, bool only_deferred)
394{
395	int r = 0;
396
397	spin_lock(&_minor_lock);
398
399	if (dm_open_count(md)) {
400		r = -EBUSY;
401		if (mark_deferred)
402			set_bit(DMF_DEFERRED_REMOVE, &md->flags);
403	} else if (only_deferred && !test_bit(DMF_DEFERRED_REMOVE, &md->flags))
404		r = -EEXIST;
405	else
406		set_bit(DMF_DELETING, &md->flags);
407
408	spin_unlock(&_minor_lock);
409
410	return r;
411}
412
413int dm_cancel_deferred_remove(struct mapped_device *md)
414{
415	int r = 0;
416
417	spin_lock(&_minor_lock);
418
419	if (test_bit(DMF_DELETING, &md->flags))
420		r = -EBUSY;
421	else
422		clear_bit(DMF_DEFERRED_REMOVE, &md->flags);
423
424	spin_unlock(&_minor_lock);
425
426	return r;
427}
428
429static void do_deferred_remove(struct work_struct *w)
430{
431	dm_deferred_remove();
432}
433
434static int dm_blk_getgeo(struct block_device *bdev, struct hd_geometry *geo)
435{
436	struct mapped_device *md = bdev->bd_disk->private_data;
437
438	return dm_get_geometry(md, geo);
439}
440
441#ifdef CONFIG_BLK_DEV_ZONED
442int dm_report_zones_cb(struct blk_zone *zone, unsigned int idx, void *data)
443{
444	struct dm_report_zones_args *args = data;
445	sector_t sector_diff = args->tgt->begin - args->start;
446
447	/*
448	 * Ignore zones beyond the target range.
449	 */
450	if (zone->start >= args->start + args->tgt->len)
451		return 0;
452
453	/*
454	 * Remap the start sector and write pointer position of the zone
455	 * to match its position in the target range.
456	 */
457	zone->start += sector_diff;
458	if (zone->type != BLK_ZONE_TYPE_CONVENTIONAL) {
459		if (zone->cond == BLK_ZONE_COND_FULL)
460			zone->wp = zone->start + zone->len;
461		else if (zone->cond == BLK_ZONE_COND_EMPTY)
462			zone->wp = zone->start;
463		else
464			zone->wp += sector_diff;
465	}
466
467	args->next_sector = zone->start + zone->len;
468	return args->orig_cb(zone, args->zone_idx++, args->orig_data);
469}
470EXPORT_SYMBOL_GPL(dm_report_zones_cb);
471
472static int dm_blk_report_zones(struct gendisk *disk, sector_t sector,
473		unsigned int nr_zones, report_zones_cb cb, void *data)
474{
475	struct mapped_device *md = disk->private_data;
476	struct dm_table *map;
477	int srcu_idx, ret;
478	struct dm_report_zones_args args = {
479		.next_sector = sector,
480		.orig_data = data,
481		.orig_cb = cb,
482	};
483
484	if (dm_suspended_md(md))
485		return -EAGAIN;
486
487	map = dm_get_live_table(md, &srcu_idx);
488	if (!map) {
489		ret = -EIO;
490		goto out;
491	}
492
493	do {
494		struct dm_target *tgt;
495
496		tgt = dm_table_find_target(map, args.next_sector);
497		if (WARN_ON_ONCE(!tgt->type->report_zones)) {
498			ret = -EIO;
499			goto out;
500		}
501
502		args.tgt = tgt;
503		ret = tgt->type->report_zones(tgt, &args,
504					      nr_zones - args.zone_idx);
505		if (ret < 0)
506			goto out;
507	} while (args.zone_idx < nr_zones &&
508		 args.next_sector < get_capacity(disk));
509
510	ret = args.zone_idx;
511out:
512	dm_put_live_table(md, srcu_idx);
513	return ret;
514}
515#else
516#define dm_blk_report_zones		NULL
517#endif /* CONFIG_BLK_DEV_ZONED */
518
519static int dm_prepare_ioctl(struct mapped_device *md, int *srcu_idx,
520			    struct block_device **bdev)
521{
522	struct dm_target *tgt;
523	struct dm_table *map;
524	int r;
525
526retry:
527	r = -ENOTTY;
528	map = dm_get_live_table(md, srcu_idx);
529	if (!map || !dm_table_get_size(map))
530		return r;
531
532	/* We only support devices that have a single target */
533	if (dm_table_get_num_targets(map) != 1)
534		return r;
535
536	tgt = dm_table_get_target(map, 0);
537	if (!tgt->type->prepare_ioctl)
538		return r;
539
540	if (dm_suspended_md(md))
541		return -EAGAIN;
542
543	r = tgt->type->prepare_ioctl(tgt, bdev);
544	if (r == -ENOTCONN && !fatal_signal_pending(current)) {
545		dm_put_live_table(md, *srcu_idx);
546		msleep(10);
547		goto retry;
548	}
549
550	return r;
551}
552
553static void dm_unprepare_ioctl(struct mapped_device *md, int srcu_idx)
554{
555	dm_put_live_table(md, srcu_idx);
556}
557
558static int dm_blk_ioctl(struct block_device *bdev, fmode_t mode,
559			unsigned int cmd, unsigned long arg)
560{
561	struct mapped_device *md = bdev->bd_disk->private_data;
562	int r, srcu_idx;
563
564	r = dm_prepare_ioctl(md, &srcu_idx, &bdev);
565	if (r < 0)
566		goto out;
567
568	if (r > 0) {
569		/*
570		 * Target determined this ioctl is being issued against a
571		 * subset of the parent bdev; require extra privileges.
572		 */
573		if (!capable(CAP_SYS_RAWIO)) {
574			DMDEBUG_LIMIT(
575	"%s: sending ioctl %x to DM device without required privilege.",
576				current->comm, cmd);
577			r = -ENOIOCTLCMD;
578			goto out;
579		}
580	}
581
582	r =  __blkdev_driver_ioctl(bdev, mode, cmd, arg);
583out:
584	dm_unprepare_ioctl(md, srcu_idx);
585	return r;
586}
587
588u64 dm_start_time_ns_from_clone(struct bio *bio)
589{
590	struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone);
591	struct dm_io *io = tio->io;
592
593	return jiffies_to_nsecs(io->start_time);
594}
595EXPORT_SYMBOL_GPL(dm_start_time_ns_from_clone);
596
597static void start_io_acct(struct dm_io *io)
598{
599	struct mapped_device *md = io->md;
600	struct bio *bio = io->orig_bio;
601
602	io->start_time = bio_start_io_acct(bio);
603	if (unlikely(dm_stats_used(&md->stats)))
604		dm_stats_account_io(&md->stats, bio_data_dir(bio),
605				    bio->bi_iter.bi_sector, bio_sectors(bio),
606				    false, 0, &io->stats_aux);
607}
608
609static void end_io_acct(struct mapped_device *md, struct bio *bio,
610			unsigned long start_time, struct dm_stats_aux *stats_aux)
611{
612	unsigned long duration = jiffies - start_time;
613
614	if (unlikely(dm_stats_used(&md->stats)))
615		dm_stats_account_io(&md->stats, bio_data_dir(bio),
616				    bio->bi_iter.bi_sector, bio_sectors(bio),
617				    true, duration, stats_aux);
618
619	smp_wmb();
620
621	bio_end_io_acct(bio, start_time);
622
623	/* nudge anyone waiting on suspend queue */
624	if (unlikely(wq_has_sleeper(&md->wait)))
625		wake_up(&md->wait);
626}
627
628static struct dm_io *alloc_io(struct mapped_device *md, struct bio *bio)
629{
630	struct dm_io *io;
631	struct dm_target_io *tio;
632	struct bio *clone;
633
634	clone = bio_alloc_bioset(GFP_NOIO, 0, &md->io_bs);
635	if (!clone)
636		return NULL;
637
638	tio = container_of(clone, struct dm_target_io, clone);
639	tio->inside_dm_io = true;
640	tio->io = NULL;
641
642	io = container_of(tio, struct dm_io, tio);
643	io->magic = DM_IO_MAGIC;
644	io->status = 0;
645	atomic_set(&io->io_count, 1);
646	io->orig_bio = bio;
647	io->md = md;
648	spin_lock_init(&io->endio_lock);
649
650	start_io_acct(io);
651
652	return io;
653}
654
655static void free_io(struct mapped_device *md, struct dm_io *io)
656{
657	bio_put(&io->tio.clone);
658}
659
660static struct dm_target_io *alloc_tio(struct clone_info *ci, struct dm_target *ti,
661				      unsigned target_bio_nr, gfp_t gfp_mask)
662{
663	struct dm_target_io *tio;
664
665	if (!ci->io->tio.io) {
666		/* the dm_target_io embedded in ci->io is available */
667		tio = &ci->io->tio;
668	} else {
669		struct bio *clone = bio_alloc_bioset(gfp_mask, 0, &ci->io->md->bs);
670		if (!clone)
671			return NULL;
672
673		tio = container_of(clone, struct dm_target_io, clone);
674		tio->inside_dm_io = false;
675	}
676
677	tio->magic = DM_TIO_MAGIC;
678	tio->io = ci->io;
679	tio->ti = ti;
680	tio->target_bio_nr = target_bio_nr;
681
682	return tio;
683}
684
685static void free_tio(struct dm_target_io *tio)
686{
687	if (tio->inside_dm_io)
688		return;
689	bio_put(&tio->clone);
690}
691
692/*
693 * Add the bio to the list of deferred io.
694 */
695static void queue_io(struct mapped_device *md, struct bio *bio)
696{
697	unsigned long flags;
698
699	spin_lock_irqsave(&md->deferred_lock, flags);
700	bio_list_add(&md->deferred, bio);
701	spin_unlock_irqrestore(&md->deferred_lock, flags);
702	queue_work(md->wq, &md->work);
703}
704
705/*
706 * Everyone (including functions in this file), should use this
707 * function to access the md->map field, and make sure they call
708 * dm_put_live_table() when finished.
709 */
710struct dm_table *dm_get_live_table(struct mapped_device *md, int *srcu_idx) __acquires(md->io_barrier)
711{
712	*srcu_idx = srcu_read_lock(&md->io_barrier);
713
714	return srcu_dereference(md->map, &md->io_barrier);
715}
716
717void dm_put_live_table(struct mapped_device *md, int srcu_idx) __releases(md->io_barrier)
718{
719	srcu_read_unlock(&md->io_barrier, srcu_idx);
720}
721
722void dm_sync_table(struct mapped_device *md)
723{
724	synchronize_srcu(&md->io_barrier);
725	synchronize_rcu_expedited();
726}
727
728/*
729 * A fast alternative to dm_get_live_table/dm_put_live_table.
730 * The caller must not block between these two functions.
731 */
732static struct dm_table *dm_get_live_table_fast(struct mapped_device *md) __acquires(RCU)
733{
734	rcu_read_lock();
735	return rcu_dereference(md->map);
736}
737
738static void dm_put_live_table_fast(struct mapped_device *md) __releases(RCU)
739{
740	rcu_read_unlock();
741}
742
743static char *_dm_claim_ptr = "I belong to device-mapper";
744
745/*
746 * Open a table device so we can use it as a map destination.
747 */
748static int open_table_device(struct table_device *td, dev_t dev,
749			     struct mapped_device *md)
750{
751	struct block_device *bdev;
752
753	int r;
754
755	BUG_ON(td->dm_dev.bdev);
756
757	bdev = blkdev_get_by_dev(dev, td->dm_dev.mode | FMODE_EXCL, _dm_claim_ptr);
758	if (IS_ERR(bdev))
759		return PTR_ERR(bdev);
760
761	r = bd_link_disk_holder(bdev, dm_disk(md));
762	if (r) {
763		blkdev_put(bdev, td->dm_dev.mode | FMODE_EXCL);
764		return r;
765	}
766
767	td->dm_dev.bdev = bdev;
768	td->dm_dev.dax_dev = dax_get_by_host(bdev->bd_disk->disk_name);
769	return 0;
770}
771
772/*
773 * Close a table device that we've been using.
774 */
775static void close_table_device(struct table_device *td, struct mapped_device *md)
776{
777	if (!td->dm_dev.bdev)
778		return;
779
780	bd_unlink_disk_holder(td->dm_dev.bdev, dm_disk(md));
781	blkdev_put(td->dm_dev.bdev, td->dm_dev.mode | FMODE_EXCL);
782	put_dax(td->dm_dev.dax_dev);
783	td->dm_dev.bdev = NULL;
784	td->dm_dev.dax_dev = NULL;
785}
786
787static struct table_device *find_table_device(struct list_head *l, dev_t dev,
788					      fmode_t mode)
789{
790	struct table_device *td;
791
792	list_for_each_entry(td, l, list)
793		if (td->dm_dev.bdev->bd_dev == dev && td->dm_dev.mode == mode)
794			return td;
795
796	return NULL;
797}
798
799int dm_get_table_device(struct mapped_device *md, dev_t dev, fmode_t mode,
800			struct dm_dev **result)
801{
802	int r;
803	struct table_device *td;
804
805	mutex_lock(&md->table_devices_lock);
806	td = find_table_device(&md->table_devices, dev, mode);
807	if (!td) {
808		td = kmalloc_node(sizeof(*td), GFP_KERNEL, md->numa_node_id);
809		if (!td) {
810			mutex_unlock(&md->table_devices_lock);
811			return -ENOMEM;
812		}
813
814		td->dm_dev.mode = mode;
815		td->dm_dev.bdev = NULL;
816
817		if ((r = open_table_device(td, dev, md))) {
818			mutex_unlock(&md->table_devices_lock);
819			kfree(td);
820			return r;
821		}
822
823		format_dev_t(td->dm_dev.name, dev);
824
825		refcount_set(&td->count, 1);
826		list_add(&td->list, &md->table_devices);
827	} else {
828		refcount_inc(&td->count);
829	}
830	mutex_unlock(&md->table_devices_lock);
831
832	*result = &td->dm_dev;
833	return 0;
834}
835EXPORT_SYMBOL_GPL(dm_get_table_device);
836
837void dm_put_table_device(struct mapped_device *md, struct dm_dev *d)
838{
839	struct table_device *td = container_of(d, struct table_device, dm_dev);
840
841	mutex_lock(&md->table_devices_lock);
842	if (refcount_dec_and_test(&td->count)) {
843		close_table_device(td, md);
844		list_del(&td->list);
845		kfree(td);
846	}
847	mutex_unlock(&md->table_devices_lock);
848}
849EXPORT_SYMBOL(dm_put_table_device);
850
851static void free_table_devices(struct list_head *devices)
852{
853	struct list_head *tmp, *next;
854
855	list_for_each_safe(tmp, next, devices) {
856		struct table_device *td = list_entry(tmp, struct table_device, list);
857
858		DMWARN("dm_destroy: %s still exists with %d references",
859		       td->dm_dev.name, refcount_read(&td->count));
860		kfree(td);
861	}
862}
863
864/*
865 * Get the geometry associated with a dm device
866 */
867int dm_get_geometry(struct mapped_device *md, struct hd_geometry *geo)
868{
869	*geo = md->geometry;
870
871	return 0;
872}
873
874/*
875 * Set the geometry of a device.
876 */
877int dm_set_geometry(struct mapped_device *md, struct hd_geometry *geo)
878{
879	sector_t sz = (sector_t)geo->cylinders * geo->heads * geo->sectors;
880
881	if (geo->start > sz) {
882		DMWARN("Start sector is beyond the geometry limits.");
883		return -EINVAL;
884	}
885
886	md->geometry = *geo;
887
888	return 0;
889}
890
891static int __noflush_suspending(struct mapped_device *md)
892{
893	return test_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
894}
895
896/*
897 * Decrements the number of outstanding ios that a bio has been
898 * cloned into, completing the original io if necc.
899 */
900static void dec_pending(struct dm_io *io, blk_status_t error)
901{
902	unsigned long flags;
903	blk_status_t io_error;
904	struct bio *bio;
905	struct mapped_device *md = io->md;
906	unsigned long start_time = 0;
907	struct dm_stats_aux stats_aux;
908
909	/* Push-back supersedes any I/O errors */
910	if (unlikely(error)) {
911		spin_lock_irqsave(&io->endio_lock, flags);
912		if (!(io->status == BLK_STS_DM_REQUEUE && __noflush_suspending(md)))
913			io->status = error;
914		spin_unlock_irqrestore(&io->endio_lock, flags);
915	}
916
917	if (atomic_dec_and_test(&io->io_count)) {
918		if (io->status == BLK_STS_DM_REQUEUE) {
919			/*
920			 * Target requested pushing back the I/O.
921			 */
922			spin_lock_irqsave(&md->deferred_lock, flags);
923			if (__noflush_suspending(md))
924				/* NOTE early return due to BLK_STS_DM_REQUEUE below */
925				bio_list_add_head(&md->deferred, io->orig_bio);
926			else
927				/* noflush suspend was interrupted. */
928				io->status = BLK_STS_IOERR;
929			spin_unlock_irqrestore(&md->deferred_lock, flags);
930		}
931
932		io_error = io->status;
933		bio = io->orig_bio;
934		start_time = io->start_time;
935		stats_aux = io->stats_aux;
936		free_io(md, io);
937		end_io_acct(md, bio, start_time, &stats_aux);
938
939		if (io_error == BLK_STS_DM_REQUEUE)
940			return;
941
942		if ((bio->bi_opf & REQ_PREFLUSH) && bio->bi_iter.bi_size) {
943			/*
944			 * Preflush done for flush with data, reissue
945			 * without REQ_PREFLUSH.
946			 */
947			bio->bi_opf &= ~REQ_PREFLUSH;
948			queue_io(md, bio);
949		} else {
950			/* done with normal IO or empty flush */
951			if (io_error)
952				bio->bi_status = io_error;
953			bio_endio(bio);
954		}
955	}
956}
957
958void disable_discard(struct mapped_device *md)
959{
960	struct queue_limits *limits = dm_get_queue_limits(md);
961
962	/* device doesn't really support DISCARD, disable it */
963	limits->max_discard_sectors = 0;
964	blk_queue_flag_clear(QUEUE_FLAG_DISCARD, md->queue);
965}
966
967void disable_write_same(struct mapped_device *md)
968{
969	struct queue_limits *limits = dm_get_queue_limits(md);
970
971	/* device doesn't really support WRITE SAME, disable it */
972	limits->max_write_same_sectors = 0;
973}
974
975void disable_write_zeroes(struct mapped_device *md)
976{
977	struct queue_limits *limits = dm_get_queue_limits(md);
978
979	/* device doesn't really support WRITE ZEROES, disable it */
980	limits->max_write_zeroes_sectors = 0;
981}
982
983static bool swap_bios_limit(struct dm_target *ti, struct bio *bio)
984{
985	return unlikely((bio->bi_opf & REQ_SWAP) != 0) && unlikely(ti->limit_swap_bios);
986}
987
988static void clone_endio(struct bio *bio)
989{
990	blk_status_t error = bio->bi_status;
991	struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone);
992	struct dm_io *io = tio->io;
993	struct mapped_device *md = tio->io->md;
994	dm_endio_fn endio = tio->ti->type->end_io;
995	struct bio *orig_bio = io->orig_bio;
996
997	if (unlikely(error == BLK_STS_TARGET)) {
998		if (bio_op(bio) == REQ_OP_DISCARD &&
999		    !bio->bi_disk->queue->limits.max_discard_sectors)
1000			disable_discard(md);
1001		else if (bio_op(bio) == REQ_OP_WRITE_SAME &&
1002			 !bio->bi_disk->queue->limits.max_write_same_sectors)
1003			disable_write_same(md);
1004		else if (bio_op(bio) == REQ_OP_WRITE_ZEROES &&
1005			 !bio->bi_disk->queue->limits.max_write_zeroes_sectors)
1006			disable_write_zeroes(md);
1007	}
1008
1009	/*
1010	 * For zone-append bios get offset in zone of the written
1011	 * sector and add that to the original bio sector pos.
1012	 */
1013	if (bio_op(orig_bio) == REQ_OP_ZONE_APPEND) {
1014		sector_t written_sector = bio->bi_iter.bi_sector;
1015		struct request_queue *q = orig_bio->bi_disk->queue;
1016		u64 mask = (u64)blk_queue_zone_sectors(q) - 1;
1017
1018		orig_bio->bi_iter.bi_sector += written_sector & mask;
1019	}
1020
1021	if (endio) {
1022		int r = endio(tio->ti, bio, &error);
1023		switch (r) {
1024		case DM_ENDIO_REQUEUE:
1025			error = BLK_STS_DM_REQUEUE;
1026			fallthrough;
1027		case DM_ENDIO_DONE:
1028			break;
1029		case DM_ENDIO_INCOMPLETE:
1030			/* The target will handle the io */
1031			return;
1032		default:
1033			DMWARN("unimplemented target endio return value: %d", r);
1034			BUG();
1035		}
1036	}
1037
1038	if (unlikely(swap_bios_limit(tio->ti, bio))) {
1039		struct mapped_device *md = io->md;
1040		up(&md->swap_bios_semaphore);
1041	}
1042
1043	free_tio(tio);
1044	dec_pending(io, error);
1045}
1046
1047/*
1048 * Return maximum size of I/O possible at the supplied sector up to the current
1049 * target boundary.
1050 */
1051static inline sector_t max_io_len_target_boundary(struct dm_target *ti,
1052						  sector_t target_offset)
1053{
1054	return ti->len - target_offset;
1055}
1056
1057static sector_t max_io_len(struct dm_target *ti, sector_t sector)
1058{
1059	sector_t target_offset = dm_target_offset(ti, sector);
1060	sector_t len = max_io_len_target_boundary(ti, target_offset);
1061	sector_t max_len;
1062
1063	/*
1064	 * Does the target need to split IO even further?
1065	 * - varied (per target) IO splitting is a tenet of DM; this
1066	 *   explains why stacked chunk_sectors based splitting via
1067	 *   blk_max_size_offset() isn't possible here. So pass in
1068	 *   ti->max_io_len to override stacked chunk_sectors.
1069	 */
1070	if (ti->max_io_len) {
1071		max_len = blk_max_size_offset(ti->table->md->queue,
1072					      target_offset, ti->max_io_len);
1073		if (len > max_len)
1074			len = max_len;
1075	}
1076
1077	return len;
1078}
1079
1080int dm_set_target_max_io_len(struct dm_target *ti, sector_t len)
1081{
1082	if (len > UINT_MAX) {
1083		DMERR("Specified maximum size of target IO (%llu) exceeds limit (%u)",
1084		      (unsigned long long)len, UINT_MAX);
1085		ti->error = "Maximum size of target IO is too large";
1086		return -EINVAL;
1087	}
1088
1089	ti->max_io_len = (uint32_t) len;
1090
1091	return 0;
1092}
1093EXPORT_SYMBOL_GPL(dm_set_target_max_io_len);
1094
1095static struct dm_target *dm_dax_get_live_target(struct mapped_device *md,
1096						sector_t sector, int *srcu_idx)
1097	__acquires(md->io_barrier)
1098{
1099	struct dm_table *map;
1100	struct dm_target *ti;
1101
1102	map = dm_get_live_table(md, srcu_idx);
1103	if (!map)
1104		return NULL;
1105
1106	ti = dm_table_find_target(map, sector);
1107	if (!ti)
1108		return NULL;
1109
1110	return ti;
1111}
1112
1113static long dm_dax_direct_access(struct dax_device *dax_dev, pgoff_t pgoff,
1114				 long nr_pages, void **kaddr, pfn_t *pfn)
1115{
1116	struct mapped_device *md = dax_get_private(dax_dev);
1117	sector_t sector = pgoff * PAGE_SECTORS;
1118	struct dm_target *ti;
1119	long len, ret = -EIO;
1120	int srcu_idx;
1121
1122	ti = dm_dax_get_live_target(md, sector, &srcu_idx);
1123
1124	if (!ti)
1125		goto out;
1126	if (!ti->type->direct_access)
1127		goto out;
1128	len = max_io_len(ti, sector) / PAGE_SECTORS;
1129	if (len < 1)
1130		goto out;
1131	nr_pages = min(len, nr_pages);
1132	ret = ti->type->direct_access(ti, pgoff, nr_pages, kaddr, pfn);
1133
1134 out:
1135	dm_put_live_table(md, srcu_idx);
1136
1137	return ret;
1138}
1139
1140static bool dm_dax_supported(struct dax_device *dax_dev, struct block_device *bdev,
1141		int blocksize, sector_t start, sector_t len)
1142{
1143	struct mapped_device *md = dax_get_private(dax_dev);
1144	struct dm_table *map;
1145	bool ret = false;
1146	int srcu_idx;
1147
1148	map = dm_get_live_table(md, &srcu_idx);
1149	if (!map)
1150		goto out;
1151
1152	ret = dm_table_supports_dax(map, device_not_dax_capable, &blocksize);
1153
1154out:
1155	dm_put_live_table(md, srcu_idx);
1156
1157	return ret;
1158}
1159
1160static size_t dm_dax_copy_from_iter(struct dax_device *dax_dev, pgoff_t pgoff,
1161				    void *addr, size_t bytes, struct iov_iter *i)
1162{
1163	struct mapped_device *md = dax_get_private(dax_dev);
1164	sector_t sector = pgoff * PAGE_SECTORS;
1165	struct dm_target *ti;
1166	long ret = 0;
1167	int srcu_idx;
1168
1169	ti = dm_dax_get_live_target(md, sector, &srcu_idx);
1170
1171	if (!ti)
1172		goto out;
1173	if (!ti->type->dax_copy_from_iter) {
1174		ret = copy_from_iter(addr, bytes, i);
1175		goto out;
1176	}
1177	ret = ti->type->dax_copy_from_iter(ti, pgoff, addr, bytes, i);
1178 out:
1179	dm_put_live_table(md, srcu_idx);
1180
1181	return ret;
1182}
1183
1184static size_t dm_dax_copy_to_iter(struct dax_device *dax_dev, pgoff_t pgoff,
1185		void *addr, size_t bytes, struct iov_iter *i)
1186{
1187	struct mapped_device *md = dax_get_private(dax_dev);
1188	sector_t sector = pgoff * PAGE_SECTORS;
1189	struct dm_target *ti;
1190	long ret = 0;
1191	int srcu_idx;
1192
1193	ti = dm_dax_get_live_target(md, sector, &srcu_idx);
1194
1195	if (!ti)
1196		goto out;
1197	if (!ti->type->dax_copy_to_iter) {
1198		ret = copy_to_iter(addr, bytes, i);
1199		goto out;
1200	}
1201	ret = ti->type->dax_copy_to_iter(ti, pgoff, addr, bytes, i);
1202 out:
1203	dm_put_live_table(md, srcu_idx);
1204
1205	return ret;
1206}
1207
1208static int dm_dax_zero_page_range(struct dax_device *dax_dev, pgoff_t pgoff,
1209				  size_t nr_pages)
1210{
1211	struct mapped_device *md = dax_get_private(dax_dev);
1212	sector_t sector = pgoff * PAGE_SECTORS;
1213	struct dm_target *ti;
1214	int ret = -EIO;
1215	int srcu_idx;
1216
1217	ti = dm_dax_get_live_target(md, sector, &srcu_idx);
1218
1219	if (!ti)
1220		goto out;
1221	if (WARN_ON(!ti->type->dax_zero_page_range)) {
1222		/*
1223		 * ->zero_page_range() is mandatory dax operation. If we are
1224		 *  here, something is wrong.
1225		 */
1226		goto out;
1227	}
1228	ret = ti->type->dax_zero_page_range(ti, pgoff, nr_pages);
1229 out:
1230	dm_put_live_table(md, srcu_idx);
1231
1232	return ret;
1233}
1234
1235/*
1236 * A target may call dm_accept_partial_bio only from the map routine.  It is
1237 * allowed for all bio types except REQ_PREFLUSH, REQ_OP_ZONE_* zone management
1238 * operations and REQ_OP_ZONE_APPEND (zone append writes).
1239 *
1240 * dm_accept_partial_bio informs the dm that the target only wants to process
1241 * additional n_sectors sectors of the bio and the rest of the data should be
1242 * sent in a next bio.
1243 *
1244 * A diagram that explains the arithmetics:
1245 * +--------------------+---------------+-------+
1246 * |         1          |       2       |   3   |
1247 * +--------------------+---------------+-------+
1248 *
1249 * <-------------- *tio->len_ptr --------------->
1250 *                      <------- bi_size ------->
1251 *                      <-- n_sectors -->
1252 *
1253 * Region 1 was already iterated over with bio_advance or similar function.
1254 *	(it may be empty if the target doesn't use bio_advance)
1255 * Region 2 is the remaining bio size that the target wants to process.
1256 *	(it may be empty if region 1 is non-empty, although there is no reason
1257 *	 to make it empty)
1258 * The target requires that region 3 is to be sent in the next bio.
1259 *
1260 * If the target wants to receive multiple copies of the bio (via num_*bios, etc),
1261 * the partially processed part (the sum of regions 1+2) must be the same for all
1262 * copies of the bio.
1263 */
1264void dm_accept_partial_bio(struct bio *bio, unsigned n_sectors)
1265{
1266	struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone);
1267	unsigned bi_size = bio->bi_iter.bi_size >> SECTOR_SHIFT;
1268
1269	BUG_ON(bio->bi_opf & REQ_PREFLUSH);
1270	BUG_ON(op_is_zone_mgmt(bio_op(bio)));
1271	BUG_ON(bio_op(bio) == REQ_OP_ZONE_APPEND);
1272	BUG_ON(bi_size > *tio->len_ptr);
1273	BUG_ON(n_sectors > bi_size);
1274
1275	*tio->len_ptr -= bi_size - n_sectors;
1276	bio->bi_iter.bi_size = n_sectors << SECTOR_SHIFT;
1277}
1278EXPORT_SYMBOL_GPL(dm_accept_partial_bio);
1279
1280static noinline void __set_swap_bios_limit(struct mapped_device *md, int latch)
1281{
1282	mutex_lock(&md->swap_bios_lock);
1283	while (latch < md->swap_bios) {
1284		cond_resched();
1285		down(&md->swap_bios_semaphore);
1286		md->swap_bios--;
1287	}
1288	while (latch > md->swap_bios) {
1289		cond_resched();
1290		up(&md->swap_bios_semaphore);
1291		md->swap_bios++;
1292	}
1293	mutex_unlock(&md->swap_bios_lock);
1294}
1295
1296static blk_qc_t __map_bio(struct dm_target_io *tio)
1297{
1298	int r;
1299	sector_t sector;
1300	struct bio *clone = &tio->clone;
1301	struct dm_io *io = tio->io;
1302	struct dm_target *ti = tio->ti;
1303	blk_qc_t ret = BLK_QC_T_NONE;
1304
1305	clone->bi_end_io = clone_endio;
1306
1307	/*
1308	 * Map the clone.  If r == 0 we don't need to do
1309	 * anything, the target has assumed ownership of
1310	 * this io.
1311	 */
1312	atomic_inc(&io->io_count);
1313	sector = clone->bi_iter.bi_sector;
1314
1315	if (unlikely(swap_bios_limit(ti, clone))) {
1316		struct mapped_device *md = io->md;
1317		int latch = get_swap_bios();
1318		if (unlikely(latch != md->swap_bios))
1319			__set_swap_bios_limit(md, latch);
1320		down(&md->swap_bios_semaphore);
1321	}
1322
1323	r = ti->type->map(ti, clone);
1324	switch (r) {
1325	case DM_MAPIO_SUBMITTED:
1326		break;
1327	case DM_MAPIO_REMAPPED:
1328		/* the bio has been remapped so dispatch it */
1329		trace_block_bio_remap(clone->bi_disk->queue, clone,
1330				      bio_dev(io->orig_bio), sector);
1331		ret = submit_bio_noacct(clone);
1332		break;
1333	case DM_MAPIO_KILL:
1334		if (unlikely(swap_bios_limit(ti, clone))) {
1335			struct mapped_device *md = io->md;
1336			up(&md->swap_bios_semaphore);
1337		}
1338		free_tio(tio);
1339		dec_pending(io, BLK_STS_IOERR);
1340		break;
1341	case DM_MAPIO_REQUEUE:
1342		if (unlikely(swap_bios_limit(ti, clone))) {
1343			struct mapped_device *md = io->md;
1344			up(&md->swap_bios_semaphore);
1345		}
1346		free_tio(tio);
1347		dec_pending(io, BLK_STS_DM_REQUEUE);
1348		break;
1349	default:
1350		DMWARN("unimplemented target map return value: %d", r);
1351		BUG();
1352	}
1353
1354	return ret;
1355}
1356
1357static void bio_setup_sector(struct bio *bio, sector_t sector, unsigned len)
1358{
1359	bio->bi_iter.bi_sector = sector;
1360	bio->bi_iter.bi_size = to_bytes(len);
1361}
1362
1363/*
1364 * Creates a bio that consists of range of complete bvecs.
1365 */
1366static int clone_bio(struct dm_target_io *tio, struct bio *bio,
1367		     sector_t sector, unsigned len)
1368{
1369	struct bio *clone = &tio->clone;
1370	int r;
1371
1372	__bio_clone_fast(clone, bio);
1373
1374	r = bio_crypt_clone(clone, bio, GFP_NOIO);
1375	if (r < 0)
1376		return r;
1377
1378	if (bio_integrity(bio)) {
1379		if (unlikely(!dm_target_has_integrity(tio->ti->type) &&
1380			     !dm_target_passes_integrity(tio->ti->type))) {
1381			DMWARN("%s: the target %s doesn't support integrity data.",
1382				dm_device_name(tio->io->md),
1383				tio->ti->type->name);
1384			return -EIO;
1385		}
1386
1387		r = bio_integrity_clone(clone, bio, GFP_NOIO);
1388		if (r < 0)
1389			return r;
1390	}
1391
1392	bio_advance(clone, to_bytes(sector - clone->bi_iter.bi_sector));
1393	clone->bi_iter.bi_size = to_bytes(len);
1394
1395	if (bio_integrity(bio))
1396		bio_integrity_trim(clone);
1397
1398	return 0;
1399}
1400
1401static void alloc_multiple_bios(struct bio_list *blist, struct clone_info *ci,
1402				struct dm_target *ti, unsigned num_bios)
1403{
1404	struct dm_target_io *tio;
1405	int try;
1406
1407	if (!num_bios)
1408		return;
1409
1410	if (num_bios == 1) {
1411		tio = alloc_tio(ci, ti, 0, GFP_NOIO);
1412		bio_list_add(blist, &tio->clone);
1413		return;
1414	}
1415
1416	for (try = 0; try < 2; try++) {
1417		int bio_nr;
1418		struct bio *bio;
1419
1420		if (try)
1421			mutex_lock(&ci->io->md->table_devices_lock);
1422		for (bio_nr = 0; bio_nr < num_bios; bio_nr++) {
1423			tio = alloc_tio(ci, ti, bio_nr, try ? GFP_NOIO : GFP_NOWAIT);
1424			if (!tio)
1425				break;
1426
1427			bio_list_add(blist, &tio->clone);
1428		}
1429		if (try)
1430			mutex_unlock(&ci->io->md->table_devices_lock);
1431		if (bio_nr == num_bios)
1432			return;
1433
1434		while ((bio = bio_list_pop(blist))) {
1435			tio = container_of(bio, struct dm_target_io, clone);
1436			free_tio(tio);
1437		}
1438	}
1439}
1440
1441static blk_qc_t __clone_and_map_simple_bio(struct clone_info *ci,
1442					   struct dm_target_io *tio, unsigned *len)
1443{
1444	struct bio *clone = &tio->clone;
1445
1446	tio->len_ptr = len;
1447
1448	__bio_clone_fast(clone, ci->bio);
1449	if (len)
1450		bio_setup_sector(clone, ci->sector, *len);
1451
1452	return __map_bio(tio);
1453}
1454
1455static void __send_duplicate_bios(struct clone_info *ci, struct dm_target *ti,
1456				  unsigned num_bios, unsigned *len)
1457{
1458	struct bio_list blist = BIO_EMPTY_LIST;
1459	struct bio *bio;
1460	struct dm_target_io *tio;
1461
1462	alloc_multiple_bios(&blist, ci, ti, num_bios);
1463
1464	while ((bio = bio_list_pop(&blist))) {
1465		tio = container_of(bio, struct dm_target_io, clone);
1466		(void) __clone_and_map_simple_bio(ci, tio, len);
1467	}
1468}
1469
1470static int __send_empty_flush(struct clone_info *ci)
1471{
1472	unsigned target_nr = 0;
1473	struct dm_target *ti;
1474	struct bio flush_bio;
1475
1476	/*
1477	 * Use an on-stack bio for this, it's safe since we don't
1478	 * need to reference it after submit. It's just used as
1479	 * the basis for the clone(s).
1480	 */
1481	bio_init(&flush_bio, NULL, 0);
1482	flush_bio.bi_opf = REQ_OP_WRITE | REQ_PREFLUSH | REQ_SYNC;
1483	ci->bio = &flush_bio;
1484	ci->sector_count = 0;
1485
1486	/*
1487	 * Empty flush uses a statically initialized bio, as the base for
1488	 * cloning.  However, blkg association requires that a bdev is
1489	 * associated with a gendisk, which doesn't happen until the bdev is
1490	 * opened.  So, blkg association is done at issue time of the flush
1491	 * rather than when the device is created in alloc_dev().
1492	 */
1493	bio_set_dev(ci->bio, ci->io->md->bdev);
1494
1495	BUG_ON(bio_has_data(ci->bio));
1496	while ((ti = dm_table_get_target(ci->map, target_nr++)))
1497		__send_duplicate_bios(ci, ti, ti->num_flush_bios, NULL);
1498
1499	bio_uninit(ci->bio);
1500	return 0;
1501}
1502
1503static int __clone_and_map_data_bio(struct clone_info *ci, struct dm_target *ti,
1504				    sector_t sector, unsigned *len)
1505{
1506	struct bio *bio = ci->bio;
1507	struct dm_target_io *tio;
1508	int r;
1509
1510	tio = alloc_tio(ci, ti, 0, GFP_NOIO);
1511	tio->len_ptr = len;
1512	r = clone_bio(tio, bio, sector, *len);
1513	if (r < 0) {
1514		free_tio(tio);
1515		return r;
1516	}
1517	(void) __map_bio(tio);
1518
1519	return 0;
1520}
1521
1522static int __send_changing_extent_only(struct clone_info *ci, struct dm_target *ti,
1523				       unsigned num_bios)
1524{
1525	unsigned len;
1526
1527	/*
1528	 * Even though the device advertised support for this type of
1529	 * request, that does not mean every target supports it, and
1530	 * reconfiguration might also have changed that since the
1531	 * check was performed.
1532	 */
1533	if (!num_bios)
1534		return -EOPNOTSUPP;
1535
1536	len = min_t(sector_t, ci->sector_count,
1537		    max_io_len_target_boundary(ti, dm_target_offset(ti, ci->sector)));
1538
1539	__send_duplicate_bios(ci, ti, num_bios, &len);
1540
1541	ci->sector += len;
1542	ci->sector_count -= len;
1543
1544	return 0;
1545}
1546
1547static bool is_abnormal_io(struct bio *bio)
1548{
1549	bool r = false;
1550
1551	switch (bio_op(bio)) {
1552	case REQ_OP_DISCARD:
1553	case REQ_OP_SECURE_ERASE:
1554	case REQ_OP_WRITE_SAME:
1555	case REQ_OP_WRITE_ZEROES:
1556		r = true;
1557		break;
1558	}
1559
1560	return r;
1561}
1562
1563static bool __process_abnormal_io(struct clone_info *ci, struct dm_target *ti,
1564				  int *result)
1565{
1566	struct bio *bio = ci->bio;
1567	unsigned num_bios = 0;
1568
1569	switch (bio_op(bio)) {
1570	case REQ_OP_DISCARD:
1571		num_bios = ti->num_discard_bios;
1572		break;
1573	case REQ_OP_SECURE_ERASE:
1574		num_bios = ti->num_secure_erase_bios;
1575		break;
1576	case REQ_OP_WRITE_SAME:
1577		num_bios = ti->num_write_same_bios;
1578		break;
1579	case REQ_OP_WRITE_ZEROES:
1580		num_bios = ti->num_write_zeroes_bios;
1581		break;
1582	default:
1583		return false;
1584	}
1585
1586	*result = __send_changing_extent_only(ci, ti, num_bios);
1587	return true;
1588}
1589
1590/*
1591 * Select the correct strategy for processing a non-flush bio.
1592 */
1593static int __split_and_process_non_flush(struct clone_info *ci)
1594{
1595	struct dm_target *ti;
1596	unsigned len;
1597	int r;
1598
1599	ti = dm_table_find_target(ci->map, ci->sector);
1600	if (!ti)
1601		return -EIO;
1602
1603	if (__process_abnormal_io(ci, ti, &r))
1604		return r;
1605
1606	len = min_t(sector_t, max_io_len(ti, ci->sector), ci->sector_count);
1607
1608	r = __clone_and_map_data_bio(ci, ti, ci->sector, &len);
1609	if (r < 0)
1610		return r;
1611
1612	ci->sector += len;
1613	ci->sector_count -= len;
1614
1615	return 0;
1616}
1617
1618static void init_clone_info(struct clone_info *ci, struct mapped_device *md,
1619			    struct dm_table *map, struct bio *bio)
1620{
1621	ci->map = map;
1622	ci->io = alloc_io(md, bio);
1623	ci->sector = bio->bi_iter.bi_sector;
1624}
1625
1626#define __dm_part_stat_sub(part, field, subnd)	\
1627	(part_stat_get(part, field) -= (subnd))
1628
1629/*
1630 * Entry point to split a bio into clones and submit them to the targets.
1631 */
1632static blk_qc_t __split_and_process_bio(struct mapped_device *md,
1633					struct dm_table *map, struct bio *bio)
1634{
1635	struct clone_info ci;
1636	blk_qc_t ret = BLK_QC_T_NONE;
1637	int error = 0;
1638
1639	init_clone_info(&ci, md, map, bio);
1640
1641	if (bio->bi_opf & REQ_PREFLUSH) {
1642		error = __send_empty_flush(&ci);
1643		/* dec_pending submits any data associated with flush */
1644	} else if (op_is_zone_mgmt(bio_op(bio))) {
1645		ci.bio = bio;
1646		ci.sector_count = 0;
1647		error = __split_and_process_non_flush(&ci);
1648	} else {
1649		ci.bio = bio;
1650		ci.sector_count = bio_sectors(bio);
1651		while (ci.sector_count && !error) {
1652			error = __split_and_process_non_flush(&ci);
1653			if (current->bio_list && ci.sector_count && !error) {
1654				/*
1655				 * Remainder must be passed to submit_bio_noacct()
1656				 * so that it gets handled *after* bios already submitted
1657				 * have been completely processed.
1658				 * We take a clone of the original to store in
1659				 * ci.io->orig_bio to be used by end_io_acct() and
1660				 * for dec_pending to use for completion handling.
1661				 */
1662				struct bio *b = bio_split(bio, bio_sectors(bio) - ci.sector_count,
1663							  GFP_NOIO, &md->queue->bio_split);
1664				ci.io->orig_bio = b;
1665
1666				/*
1667				 * Adjust IO stats for each split, otherwise upon queue
1668				 * reentry there will be redundant IO accounting.
1669				 * NOTE: this is a stop-gap fix, a proper fix involves
1670				 * significant refactoring of DM core's bio splitting
1671				 * (by eliminating DM's splitting and just using bio_split)
1672				 */
1673				part_stat_lock();
1674				__dm_part_stat_sub(&dm_disk(md)->part0,
1675						   sectors[op_stat_group(bio_op(bio))], ci.sector_count);
1676				part_stat_unlock();
1677
1678				bio_chain(b, bio);
1679				trace_block_split(md->queue, b, bio->bi_iter.bi_sector);
1680				ret = submit_bio_noacct(bio);
1681				break;
1682			}
1683		}
1684	}
1685
1686	/* drop the extra reference count */
1687	dec_pending(ci.io, errno_to_blk_status(error));
1688	return ret;
1689}
1690
1691static blk_qc_t dm_submit_bio(struct bio *bio)
1692{
1693	struct mapped_device *md = bio->bi_disk->private_data;
1694	blk_qc_t ret = BLK_QC_T_NONE;
1695	int srcu_idx;
1696	struct dm_table *map;
1697
1698	map = dm_get_live_table(md, &srcu_idx);
1699
1700	/* If suspended, or map not yet available, queue this IO for later */
1701	if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) ||
1702	    unlikely(!map)) {
1703		if (bio->bi_opf & REQ_NOWAIT)
1704			bio_wouldblock_error(bio);
1705		else if (bio->bi_opf & REQ_RAHEAD)
1706			bio_io_error(bio);
1707		else
1708			queue_io(md, bio);
1709		goto out;
1710	}
1711
1712	/*
1713	 * Use blk_queue_split() for abnormal IO (e.g. discard, writesame, etc)
1714	 * otherwise associated queue_limits won't be imposed.
1715	 */
1716	if (is_abnormal_io(bio))
1717		blk_queue_split(&bio);
1718
1719	ret = __split_and_process_bio(md, map, bio);
1720out:
1721	dm_put_live_table(md, srcu_idx);
1722	return ret;
1723}
1724
1725/*-----------------------------------------------------------------
1726 * An IDR is used to keep track of allocated minor numbers.
1727 *---------------------------------------------------------------*/
1728static void free_minor(int minor)
1729{
1730	spin_lock(&_minor_lock);
1731	idr_remove(&_minor_idr, minor);
1732	spin_unlock(&_minor_lock);
1733}
1734
1735/*
1736 * See if the device with a specific minor # is free.
1737 */
1738static int specific_minor(int minor)
1739{
1740	int r;
1741
1742	if (minor >= (1 << MINORBITS))
1743		return -EINVAL;
1744
1745	idr_preload(GFP_KERNEL);
1746	spin_lock(&_minor_lock);
1747
1748	r = idr_alloc(&_minor_idr, MINOR_ALLOCED, minor, minor + 1, GFP_NOWAIT);
1749
1750	spin_unlock(&_minor_lock);
1751	idr_preload_end();
1752	if (r < 0)
1753		return r == -ENOSPC ? -EBUSY : r;
1754	return 0;
1755}
1756
1757static int next_free_minor(int *minor)
1758{
1759	int r;
1760
1761	idr_preload(GFP_KERNEL);
1762	spin_lock(&_minor_lock);
1763
1764	r = idr_alloc(&_minor_idr, MINOR_ALLOCED, 0, 1 << MINORBITS, GFP_NOWAIT);
1765
1766	spin_unlock(&_minor_lock);
1767	idr_preload_end();
1768	if (r < 0)
1769		return r;
1770	*minor = r;
1771	return 0;
1772}
1773
1774static const struct block_device_operations dm_blk_dops;
1775static const struct block_device_operations dm_rq_blk_dops;
1776static const struct dax_operations dm_dax_ops;
1777
1778static void dm_wq_work(struct work_struct *work);
1779
1780static void cleanup_mapped_device(struct mapped_device *md)
1781{
1782	if (md->wq)
1783		destroy_workqueue(md->wq);
1784	bioset_exit(&md->bs);
1785	bioset_exit(&md->io_bs);
1786
1787	if (md->dax_dev) {
1788		kill_dax(md->dax_dev);
1789		put_dax(md->dax_dev);
1790		md->dax_dev = NULL;
1791	}
1792
1793	if (md->disk) {
1794		spin_lock(&_minor_lock);
1795		md->disk->private_data = NULL;
1796		spin_unlock(&_minor_lock);
1797		del_gendisk(md->disk);
1798		put_disk(md->disk);
1799	}
1800
1801	if (md->queue)
1802		blk_cleanup_queue(md->queue);
1803
1804	cleanup_srcu_struct(&md->io_barrier);
1805
1806	if (md->bdev) {
1807		bdput(md->bdev);
1808		md->bdev = NULL;
1809	}
1810
1811	mutex_destroy(&md->suspend_lock);
1812	mutex_destroy(&md->type_lock);
1813	mutex_destroy(&md->table_devices_lock);
1814	mutex_destroy(&md->swap_bios_lock);
1815
1816	dm_mq_cleanup_mapped_device(md);
1817}
1818
1819/*
1820 * Allocate and initialise a blank device with a given minor.
1821 */
1822static struct mapped_device *alloc_dev(int minor)
1823{
1824	int r, numa_node_id = dm_get_numa_node();
1825	struct mapped_device *md;
1826	void *old_md;
1827
1828	md = kvzalloc_node(sizeof(*md), GFP_KERNEL, numa_node_id);
1829	if (!md) {
1830		DMWARN("unable to allocate device, out of memory.");
1831		return NULL;
1832	}
1833
1834	if (!try_module_get(THIS_MODULE))
1835		goto bad_module_get;
1836
1837	/* get a minor number for the dev */
1838	if (minor == DM_ANY_MINOR)
1839		r = next_free_minor(&minor);
1840	else
1841		r = specific_minor(minor);
1842	if (r < 0)
1843		goto bad_minor;
1844
1845	r = init_srcu_struct(&md->io_barrier);
1846	if (r < 0)
1847		goto bad_io_barrier;
1848
1849	md->numa_node_id = numa_node_id;
1850	md->init_tio_pdu = false;
1851	md->type = DM_TYPE_NONE;
1852	mutex_init(&md->suspend_lock);
1853	mutex_init(&md->type_lock);
1854	mutex_init(&md->table_devices_lock);
1855	spin_lock_init(&md->deferred_lock);
1856	atomic_set(&md->holders, 1);
1857	atomic_set(&md->open_count, 0);
1858	atomic_set(&md->event_nr, 0);
1859	atomic_set(&md->uevent_seq, 0);
1860	INIT_LIST_HEAD(&md->uevent_list);
1861	INIT_LIST_HEAD(&md->table_devices);
1862	spin_lock_init(&md->uevent_lock);
1863
1864	/*
1865	 * default to bio-based until DM table is loaded and md->type
1866	 * established. If request-based table is loaded: blk-mq will
1867	 * override accordingly.
1868	 */
1869	md->queue = blk_alloc_queue(numa_node_id);
1870	if (!md->queue)
1871		goto bad;
1872
1873	md->disk = alloc_disk_node(1, md->numa_node_id);
1874	if (!md->disk)
1875		goto bad;
1876
1877	init_waitqueue_head(&md->wait);
1878	INIT_WORK(&md->work, dm_wq_work);
1879	init_waitqueue_head(&md->eventq);
1880	init_completion(&md->kobj_holder.completion);
1881
1882	md->swap_bios = get_swap_bios();
1883	sema_init(&md->swap_bios_semaphore, md->swap_bios);
1884	mutex_init(&md->swap_bios_lock);
1885
1886	md->disk->major = _major;
1887	md->disk->first_minor = minor;
1888	md->disk->fops = &dm_blk_dops;
1889	md->disk->queue = md->queue;
1890	md->disk->private_data = md;
1891	sprintf(md->disk->disk_name, "dm-%d", minor);
1892
1893	if (IS_ENABLED(CONFIG_DAX_DRIVER)) {
1894		md->dax_dev = alloc_dax(md, md->disk->disk_name,
1895					&dm_dax_ops, 0);
1896		if (IS_ERR(md->dax_dev)) {
1897			md->dax_dev = NULL;
1898			goto bad;
1899		}
1900	}
1901
1902	add_disk_no_queue_reg(md->disk);
1903	format_dev_t(md->name, MKDEV(_major, minor));
1904
1905	md->wq = alloc_workqueue("kdmflush", WQ_MEM_RECLAIM, 0);
1906	if (!md->wq)
1907		goto bad;
1908
1909	md->bdev = bdget_disk(md->disk, 0);
1910	if (!md->bdev)
1911		goto bad;
1912
1913	r = dm_stats_init(&md->stats);
1914	if (r < 0)
1915		goto bad;
1916
1917	/* Populate the mapping, nobody knows we exist yet */
1918	spin_lock(&_minor_lock);
1919	old_md = idr_replace(&_minor_idr, md, minor);
1920	spin_unlock(&_minor_lock);
1921
1922	BUG_ON(old_md != MINOR_ALLOCED);
1923
1924	return md;
1925
1926bad:
1927	cleanup_mapped_device(md);
1928bad_io_barrier:
1929	free_minor(minor);
1930bad_minor:
1931	module_put(THIS_MODULE);
1932bad_module_get:
1933	kvfree(md);
1934	return NULL;
1935}
1936
1937static void unlock_fs(struct mapped_device *md);
1938
1939static void free_dev(struct mapped_device *md)
1940{
1941	int minor = MINOR(disk_devt(md->disk));
1942
1943	unlock_fs(md);
1944
1945	cleanup_mapped_device(md);
1946
1947	free_table_devices(&md->table_devices);
1948	dm_stats_cleanup(&md->stats);
1949	free_minor(minor);
1950
1951	module_put(THIS_MODULE);
1952	kvfree(md);
1953}
1954
1955static int __bind_mempools(struct mapped_device *md, struct dm_table *t)
1956{
1957	struct dm_md_mempools *p = dm_table_get_md_mempools(t);
1958	int ret = 0;
1959
1960	if (dm_table_bio_based(t)) {
1961		/*
1962		 * The md may already have mempools that need changing.
1963		 * If so, reload bioset because front_pad may have changed
1964		 * because a different table was loaded.
1965		 */
1966		bioset_exit(&md->bs);
1967		bioset_exit(&md->io_bs);
1968
1969	} else if (bioset_initialized(&md->bs)) {
1970		/*
1971		 * There's no need to reload with request-based dm
1972		 * because the size of front_pad doesn't change.
1973		 * Note for future: If you are to reload bioset,
1974		 * prep-ed requests in the queue may refer
1975		 * to bio from the old bioset, so you must walk
1976		 * through the queue to unprep.
1977		 */
1978		goto out;
1979	}
1980
1981	BUG_ON(!p ||
1982	       bioset_initialized(&md->bs) ||
1983	       bioset_initialized(&md->io_bs));
1984
1985	ret = bioset_init_from_src(&md->bs, &p->bs);
1986	if (ret)
1987		goto out;
1988	ret = bioset_init_from_src(&md->io_bs, &p->io_bs);
1989	if (ret)
1990		bioset_exit(&md->bs);
1991out:
1992	/* mempool bind completed, no longer need any mempools in the table */
1993	dm_table_free_md_mempools(t);
1994	return ret;
1995}
1996
1997/*
1998 * Bind a table to the device.
1999 */
2000static void event_callback(void *context)
2001{
2002	unsigned long flags;
2003	LIST_HEAD(uevents);
2004	struct mapped_device *md = (struct mapped_device *) context;
2005
2006	spin_lock_irqsave(&md->uevent_lock, flags);
2007	list_splice_init(&md->uevent_list, &uevents);
2008	spin_unlock_irqrestore(&md->uevent_lock, flags);
2009
2010	dm_send_uevents(&uevents, &disk_to_dev(md->disk)->kobj);
2011
2012	atomic_inc(&md->event_nr);
2013	wake_up(&md->eventq);
2014	dm_issue_global_event();
2015}
2016
2017/*
2018 * Returns old map, which caller must destroy.
2019 */
2020static struct dm_table *__bind(struct mapped_device *md, struct dm_table *t,
2021			       struct queue_limits *limits)
2022{
2023	struct dm_table *old_map;
2024	struct request_queue *q = md->queue;
2025	bool request_based = dm_table_request_based(t);
2026	sector_t size;
2027	int ret;
2028
2029	lockdep_assert_held(&md->suspend_lock);
2030
2031	size = dm_table_get_size(t);
2032
2033	/*
2034	 * Wipe any geometry if the size of the table changed.
2035	 */
2036	if (size != dm_get_size(md))
2037		memset(&md->geometry, 0, sizeof(md->geometry));
2038
2039	set_capacity(md->disk, size);
2040	bd_set_nr_sectors(md->bdev, size);
2041
2042	dm_table_event_callback(t, event_callback, md);
2043
2044	if (request_based) {
2045		/*
2046		 * Leverage the fact that request-based DM targets are
2047		 * immutable singletons - used to optimize dm_mq_queue_rq.
2048		 */
2049		md->immutable_target = dm_table_get_immutable_target(t);
2050	}
2051
2052	ret = __bind_mempools(md, t);
2053	if (ret) {
2054		old_map = ERR_PTR(ret);
2055		goto out;
2056	}
2057
2058	old_map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2059	rcu_assign_pointer(md->map, (void *)t);
2060	md->immutable_target_type = dm_table_get_immutable_target_type(t);
2061
2062	dm_table_set_restrictions(t, q, limits);
2063	if (old_map)
2064		dm_sync_table(md);
2065
2066out:
2067	return old_map;
2068}
2069
2070/*
2071 * Returns unbound table for the caller to free.
2072 */
2073static struct dm_table *__unbind(struct mapped_device *md)
2074{
2075	struct dm_table *map = rcu_dereference_protected(md->map, 1);
2076
2077	if (!map)
2078		return NULL;
2079
2080	dm_table_event_callback(map, NULL, NULL);
2081	RCU_INIT_POINTER(md->map, NULL);
2082	dm_sync_table(md);
2083
2084	return map;
2085}
2086
2087/*
2088 * Constructor for a new device.
2089 */
2090int dm_create(int minor, struct mapped_device **result)
2091{
2092	int r;
2093	struct mapped_device *md;
2094
2095	md = alloc_dev(minor);
2096	if (!md)
2097		return -ENXIO;
2098
2099	r = dm_sysfs_init(md);
2100	if (r) {
2101		free_dev(md);
2102		return r;
2103	}
2104
2105	*result = md;
2106	return 0;
2107}
2108
2109/*
2110 * Functions to manage md->type.
2111 * All are required to hold md->type_lock.
2112 */
2113void dm_lock_md_type(struct mapped_device *md)
2114{
2115	mutex_lock(&md->type_lock);
2116}
2117
2118void dm_unlock_md_type(struct mapped_device *md)
2119{
2120	mutex_unlock(&md->type_lock);
2121}
2122
2123void dm_set_md_type(struct mapped_device *md, enum dm_queue_mode type)
2124{
2125	BUG_ON(!mutex_is_locked(&md->type_lock));
2126	md->type = type;
2127}
2128
2129enum dm_queue_mode dm_get_md_type(struct mapped_device *md)
2130{
2131	return md->type;
2132}
2133
2134struct target_type *dm_get_immutable_target_type(struct mapped_device *md)
2135{
2136	return md->immutable_target_type;
2137}
2138
2139/*
2140 * The queue_limits are only valid as long as you have a reference
2141 * count on 'md'.
2142 */
2143struct queue_limits *dm_get_queue_limits(struct mapped_device *md)
2144{
2145	BUG_ON(!atomic_read(&md->holders));
2146	return &md->queue->limits;
2147}
2148EXPORT_SYMBOL_GPL(dm_get_queue_limits);
2149
2150/*
2151 * Setup the DM device's queue based on md's type
2152 */
2153int dm_setup_md_queue(struct mapped_device *md, struct dm_table *t)
2154{
2155	int r;
2156	struct queue_limits limits;
2157	enum dm_queue_mode type = dm_get_md_type(md);
2158
2159	switch (type) {
2160	case DM_TYPE_REQUEST_BASED:
2161		md->disk->fops = &dm_rq_blk_dops;
2162		r = dm_mq_init_request_queue(md, t);
2163		if (r) {
2164			DMERR("Cannot initialize queue for request-based dm mapped device");
2165			return r;
2166		}
2167		break;
2168	case DM_TYPE_BIO_BASED:
2169	case DM_TYPE_DAX_BIO_BASED:
2170		break;
2171	case DM_TYPE_NONE:
2172		WARN_ON_ONCE(true);
2173		break;
2174	}
2175
2176	r = dm_calculate_queue_limits(t, &limits);
2177	if (r) {
2178		DMERR("Cannot calculate initial queue limits");
2179		return r;
2180	}
2181	dm_table_set_restrictions(t, md->queue, &limits);
2182	blk_register_queue(md->disk);
2183
2184	return 0;
2185}
2186
2187struct mapped_device *dm_get_md(dev_t dev)
2188{
2189	struct mapped_device *md;
2190	unsigned minor = MINOR(dev);
2191
2192	if (MAJOR(dev) != _major || minor >= (1 << MINORBITS))
2193		return NULL;
2194
2195	spin_lock(&_minor_lock);
2196
2197	md = idr_find(&_minor_idr, minor);
2198	if (!md || md == MINOR_ALLOCED || (MINOR(disk_devt(dm_disk(md))) != minor) ||
2199	    test_bit(DMF_FREEING, &md->flags) || dm_deleting_md(md)) {
2200		md = NULL;
2201		goto out;
2202	}
2203	dm_get(md);
2204out:
2205	spin_unlock(&_minor_lock);
2206
2207	return md;
2208}
2209EXPORT_SYMBOL_GPL(dm_get_md);
2210
2211void *dm_get_mdptr(struct mapped_device *md)
2212{
2213	return md->interface_ptr;
2214}
2215
2216void dm_set_mdptr(struct mapped_device *md, void *ptr)
2217{
2218	md->interface_ptr = ptr;
2219}
2220
2221void dm_get(struct mapped_device *md)
2222{
2223	atomic_inc(&md->holders);
2224	BUG_ON(test_bit(DMF_FREEING, &md->flags));
2225}
2226
2227int dm_hold(struct mapped_device *md)
2228{
2229	spin_lock(&_minor_lock);
2230	if (test_bit(DMF_FREEING, &md->flags)) {
2231		spin_unlock(&_minor_lock);
2232		return -EBUSY;
2233	}
2234	dm_get(md);
2235	spin_unlock(&_minor_lock);
2236	return 0;
2237}
2238EXPORT_SYMBOL_GPL(dm_hold);
2239
2240const char *dm_device_name(struct mapped_device *md)
2241{
2242	return md->name;
2243}
2244EXPORT_SYMBOL_GPL(dm_device_name);
2245
2246static void __dm_destroy(struct mapped_device *md, bool wait)
2247{
2248	struct dm_table *map;
2249	int srcu_idx;
2250
2251	might_sleep();
2252
2253	spin_lock(&_minor_lock);
2254	idr_replace(&_minor_idr, MINOR_ALLOCED, MINOR(disk_devt(dm_disk(md))));
2255	set_bit(DMF_FREEING, &md->flags);
2256	spin_unlock(&_minor_lock);
2257
2258	blk_set_queue_dying(md->queue);
2259
2260	/*
2261	 * Take suspend_lock so that presuspend and postsuspend methods
2262	 * do not race with internal suspend.
2263	 */
2264	mutex_lock(&md->suspend_lock);
2265	map = dm_get_live_table(md, &srcu_idx);
2266	if (!dm_suspended_md(md)) {
2267		dm_table_presuspend_targets(map);
2268		set_bit(DMF_SUSPENDED, &md->flags);
2269		set_bit(DMF_POST_SUSPENDING, &md->flags);
2270		dm_table_postsuspend_targets(map);
2271	}
2272	/* dm_put_live_table must be before msleep, otherwise deadlock is possible */
2273	dm_put_live_table(md, srcu_idx);
2274	mutex_unlock(&md->suspend_lock);
2275
2276	/*
2277	 * Rare, but there may be I/O requests still going to complete,
2278	 * for example.  Wait for all references to disappear.
2279	 * No one should increment the reference count of the mapped_device,
2280	 * after the mapped_device state becomes DMF_FREEING.
2281	 */
2282	if (wait)
2283		while (atomic_read(&md->holders))
2284			msleep(1);
2285	else if (atomic_read(&md->holders))
2286		DMWARN("%s: Forcibly removing mapped_device still in use! (%d users)",
2287		       dm_device_name(md), atomic_read(&md->holders));
2288
2289	dm_sysfs_exit(md);
2290	dm_table_destroy(__unbind(md));
2291	free_dev(md);
2292}
2293
2294void dm_destroy(struct mapped_device *md)
2295{
2296	__dm_destroy(md, true);
2297}
2298
2299void dm_destroy_immediate(struct mapped_device *md)
2300{
2301	__dm_destroy(md, false);
2302}
2303
2304void dm_put(struct mapped_device *md)
2305{
2306	atomic_dec(&md->holders);
2307}
2308EXPORT_SYMBOL_GPL(dm_put);
2309
2310static bool md_in_flight_bios(struct mapped_device *md)
2311{
2312	int cpu;
2313	struct hd_struct *part = &dm_disk(md)->part0;
2314	long sum = 0;
2315
2316	for_each_possible_cpu(cpu) {
2317		sum += part_stat_local_read_cpu(part, in_flight[0], cpu);
2318		sum += part_stat_local_read_cpu(part, in_flight[1], cpu);
2319	}
2320
2321	return sum != 0;
2322}
2323
2324static int dm_wait_for_bios_completion(struct mapped_device *md, long task_state)
2325{
2326	int r = 0;
2327	DEFINE_WAIT(wait);
2328
2329	while (true) {
2330		prepare_to_wait(&md->wait, &wait, task_state);
2331
2332		if (!md_in_flight_bios(md))
2333			break;
2334
2335		if (signal_pending_state(task_state, current)) {
2336			r = -EINTR;
2337			break;
2338		}
2339
2340		io_schedule();
2341	}
2342	finish_wait(&md->wait, &wait);
2343
2344	smp_rmb();
2345
2346	return r;
2347}
2348
2349static int dm_wait_for_completion(struct mapped_device *md, long task_state)
2350{
2351	int r = 0;
2352
2353	if (!queue_is_mq(md->queue))
2354		return dm_wait_for_bios_completion(md, task_state);
2355
2356	while (true) {
2357		if (!blk_mq_queue_inflight(md->queue))
2358			break;
2359
2360		if (signal_pending_state(task_state, current)) {
2361			r = -EINTR;
2362			break;
2363		}
2364
2365		msleep(5);
2366	}
2367
2368	return r;
2369}
2370
2371/*
2372 * Process the deferred bios
2373 */
2374static void dm_wq_work(struct work_struct *work)
2375{
2376	struct mapped_device *md = container_of(work, struct mapped_device, work);
2377	struct bio *bio;
2378
2379	while (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
2380		spin_lock_irq(&md->deferred_lock);
2381		bio = bio_list_pop(&md->deferred);
2382		spin_unlock_irq(&md->deferred_lock);
2383
2384		if (!bio)
2385			break;
2386
2387		submit_bio_noacct(bio);
2388		cond_resched();
2389	}
2390}
2391
2392static void dm_queue_flush(struct mapped_device *md)
2393{
2394	clear_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2395	smp_mb__after_atomic();
2396	queue_work(md->wq, &md->work);
2397}
2398
2399/*
2400 * Swap in a new table, returning the old one for the caller to destroy.
2401 */
2402struct dm_table *dm_swap_table(struct mapped_device *md, struct dm_table *table)
2403{
2404	struct dm_table *live_map = NULL, *map = ERR_PTR(-EINVAL);
2405	struct queue_limits limits;
2406	int r;
2407
2408	mutex_lock(&md->suspend_lock);
2409
2410	/* device must be suspended */
2411	if (!dm_suspended_md(md))
2412		goto out;
2413
2414	/*
2415	 * If the new table has no data devices, retain the existing limits.
2416	 * This helps multipath with queue_if_no_path if all paths disappear,
2417	 * then new I/O is queued based on these limits, and then some paths
2418	 * reappear.
2419	 */
2420	if (dm_table_has_no_data_devices(table)) {
2421		live_map = dm_get_live_table_fast(md);
2422		if (live_map)
2423			limits = md->queue->limits;
2424		dm_put_live_table_fast(md);
2425	}
2426
2427	if (!live_map) {
2428		r = dm_calculate_queue_limits(table, &limits);
2429		if (r) {
2430			map = ERR_PTR(r);
2431			goto out;
2432		}
2433	}
2434
2435	map = __bind(md, table, &limits);
2436	dm_issue_global_event();
2437
2438out:
2439	mutex_unlock(&md->suspend_lock);
2440	return map;
2441}
2442
2443/*
2444 * Functions to lock and unlock any filesystem running on the
2445 * device.
2446 */
2447static int lock_fs(struct mapped_device *md)
2448{
2449	int r;
2450
2451	WARN_ON(md->frozen_sb);
2452
2453	md->frozen_sb = freeze_bdev(md->bdev);
2454	if (IS_ERR(md->frozen_sb)) {
2455		r = PTR_ERR(md->frozen_sb);
2456		md->frozen_sb = NULL;
2457		return r;
2458	}
2459
2460	set_bit(DMF_FROZEN, &md->flags);
2461
2462	return 0;
2463}
2464
2465static void unlock_fs(struct mapped_device *md)
2466{
2467	if (!test_bit(DMF_FROZEN, &md->flags))
2468		return;
2469
2470	thaw_bdev(md->bdev, md->frozen_sb);
2471	md->frozen_sb = NULL;
2472	clear_bit(DMF_FROZEN, &md->flags);
2473}
2474
2475/*
2476 * @suspend_flags: DM_SUSPEND_LOCKFS_FLAG and/or DM_SUSPEND_NOFLUSH_FLAG
2477 * @task_state: e.g. TASK_INTERRUPTIBLE or TASK_UNINTERRUPTIBLE
2478 * @dmf_suspended_flag: DMF_SUSPENDED or DMF_SUSPENDED_INTERNALLY
2479 *
2480 * If __dm_suspend returns 0, the device is completely quiescent
2481 * now. There is no request-processing activity. All new requests
2482 * are being added to md->deferred list.
2483 */
2484static int __dm_suspend(struct mapped_device *md, struct dm_table *map,
2485			unsigned suspend_flags, long task_state,
2486			int dmf_suspended_flag)
2487{
2488	bool do_lockfs = suspend_flags & DM_SUSPEND_LOCKFS_FLAG;
2489	bool noflush = suspend_flags & DM_SUSPEND_NOFLUSH_FLAG;
2490	int r;
2491
2492	lockdep_assert_held(&md->suspend_lock);
2493
2494	/*
2495	 * DMF_NOFLUSH_SUSPENDING must be set before presuspend.
2496	 * This flag is cleared before dm_suspend returns.
2497	 */
2498	if (noflush)
2499		set_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
2500	else
2501		DMDEBUG("%s: suspending with flush", dm_device_name(md));
2502
2503	/*
2504	 * This gets reverted if there's an error later and the targets
2505	 * provide the .presuspend_undo hook.
2506	 */
2507	dm_table_presuspend_targets(map);
2508
2509	/*
2510	 * Flush I/O to the device.
2511	 * Any I/O submitted after lock_fs() may not be flushed.
2512	 * noflush takes precedence over do_lockfs.
2513	 * (lock_fs() flushes I/Os and waits for them to complete.)
2514	 */
2515	if (!noflush && do_lockfs) {
2516		r = lock_fs(md);
2517		if (r) {
2518			dm_table_presuspend_undo_targets(map);
2519			return r;
2520		}
2521	}
2522
2523	/*
2524	 * Here we must make sure that no processes are submitting requests
2525	 * to target drivers i.e. no one may be executing
2526	 * __split_and_process_bio from dm_submit_bio.
2527	 *
2528	 * To get all processes out of __split_and_process_bio in dm_submit_bio,
2529	 * we take the write lock. To prevent any process from reentering
2530	 * __split_and_process_bio from dm_submit_bio and quiesce the thread
2531	 * (dm_wq_work), we set DMF_BLOCK_IO_FOR_SUSPEND and call
2532	 * flush_workqueue(md->wq).
2533	 */
2534	set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2535	if (map)
2536		synchronize_srcu(&md->io_barrier);
2537
2538	/*
2539	 * Stop md->queue before flushing md->wq in case request-based
2540	 * dm defers requests to md->wq from md->queue.
2541	 */
2542	if (dm_request_based(md))
2543		dm_stop_queue(md->queue);
2544
2545	flush_workqueue(md->wq);
2546
2547	/*
2548	 * At this point no more requests are entering target request routines.
2549	 * We call dm_wait_for_completion to wait for all existing requests
2550	 * to finish.
2551	 */
2552	r = dm_wait_for_completion(md, task_state);
2553	if (!r)
2554		set_bit(dmf_suspended_flag, &md->flags);
2555
2556	if (noflush)
2557		clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
2558	if (map)
2559		synchronize_srcu(&md->io_barrier);
2560
2561	/* were we interrupted ? */
2562	if (r < 0) {
2563		dm_queue_flush(md);
2564
2565		if (dm_request_based(md))
2566			dm_start_queue(md->queue);
2567
2568		unlock_fs(md);
2569		dm_table_presuspend_undo_targets(map);
2570		/* pushback list is already flushed, so skip flush */
2571	}
2572
2573	return r;
2574}
2575
2576/*
2577 * We need to be able to change a mapping table under a mounted
2578 * filesystem.  For example we might want to move some data in
2579 * the background.  Before the table can be swapped with
2580 * dm_bind_table, dm_suspend must be called to flush any in
2581 * flight bios and ensure that any further io gets deferred.
2582 */
2583/*
2584 * Suspend mechanism in request-based dm.
2585 *
2586 * 1. Flush all I/Os by lock_fs() if needed.
2587 * 2. Stop dispatching any I/O by stopping the request_queue.
2588 * 3. Wait for all in-flight I/Os to be completed or requeued.
2589 *
2590 * To abort suspend, start the request_queue.
2591 */
2592int dm_suspend(struct mapped_device *md, unsigned suspend_flags)
2593{
2594	struct dm_table *map = NULL;
2595	int r = 0;
2596
2597retry:
2598	mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING);
2599
2600	if (dm_suspended_md(md)) {
2601		r = -EINVAL;
2602		goto out_unlock;
2603	}
2604
2605	if (dm_suspended_internally_md(md)) {
2606		/* already internally suspended, wait for internal resume */
2607		mutex_unlock(&md->suspend_lock);
2608		r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE);
2609		if (r)
2610			return r;
2611		goto retry;
2612	}
2613
2614	map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2615
2616	r = __dm_suspend(md, map, suspend_flags, TASK_INTERRUPTIBLE, DMF_SUSPENDED);
2617	if (r)
2618		goto out_unlock;
2619
2620	set_bit(DMF_POST_SUSPENDING, &md->flags);
2621	dm_table_postsuspend_targets(map);
2622	clear_bit(DMF_POST_SUSPENDING, &md->flags);
2623
2624out_unlock:
2625	mutex_unlock(&md->suspend_lock);
2626	return r;
2627}
2628
2629static int __dm_resume(struct mapped_device *md, struct dm_table *map)
2630{
2631	if (map) {
2632		int r = dm_table_resume_targets(map);
2633		if (r)
2634			return r;
2635	}
2636
2637	dm_queue_flush(md);
2638
2639	/*
2640	 * Flushing deferred I/Os must be done after targets are resumed
2641	 * so that mapping of targets can work correctly.
2642	 * Request-based dm is queueing the deferred I/Os in its request_queue.
2643	 */
2644	if (dm_request_based(md))
2645		dm_start_queue(md->queue);
2646
2647	unlock_fs(md);
2648
2649	return 0;
2650}
2651
2652int dm_resume(struct mapped_device *md)
2653{
2654	int r;
2655	struct dm_table *map = NULL;
2656
2657retry:
2658	r = -EINVAL;
2659	mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING);
2660
2661	if (!dm_suspended_md(md))
2662		goto out;
2663
2664	if (dm_suspended_internally_md(md)) {
2665		/* already internally suspended, wait for internal resume */
2666		mutex_unlock(&md->suspend_lock);
2667		r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE);
2668		if (r)
2669			return r;
2670		goto retry;
2671	}
2672
2673	map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2674	if (!map || !dm_table_get_size(map))
2675		goto out;
2676
2677	r = __dm_resume(md, map);
2678	if (r)
2679		goto out;
2680
2681	clear_bit(DMF_SUSPENDED, &md->flags);
2682out:
2683	mutex_unlock(&md->suspend_lock);
2684
2685	return r;
2686}
2687
2688/*
2689 * Internal suspend/resume works like userspace-driven suspend. It waits
2690 * until all bios finish and prevents issuing new bios to the target drivers.
2691 * It may be used only from the kernel.
2692 */
2693
2694static void __dm_internal_suspend(struct mapped_device *md, unsigned suspend_flags)
2695{
2696	struct dm_table *map = NULL;
2697
2698	lockdep_assert_held(&md->suspend_lock);
2699
2700	if (md->internal_suspend_count++)
2701		return; /* nested internal suspend */
2702
2703	if (dm_suspended_md(md)) {
2704		set_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
2705		return; /* nest suspend */
2706	}
2707
2708	map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2709
2710	/*
2711	 * Using TASK_UNINTERRUPTIBLE because only NOFLUSH internal suspend is
2712	 * supported.  Properly supporting a TASK_INTERRUPTIBLE internal suspend
2713	 * would require changing .presuspend to return an error -- avoid this
2714	 * until there is a need for more elaborate variants of internal suspend.
2715	 */
2716	(void) __dm_suspend(md, map, suspend_flags, TASK_UNINTERRUPTIBLE,
2717			    DMF_SUSPENDED_INTERNALLY);
2718
2719	set_bit(DMF_POST_SUSPENDING, &md->flags);
2720	dm_table_postsuspend_targets(map);
2721	clear_bit(DMF_POST_SUSPENDING, &md->flags);
2722}
2723
2724static void __dm_internal_resume(struct mapped_device *md)
2725{
2726	BUG_ON(!md->internal_suspend_count);
2727
2728	if (--md->internal_suspend_count)
2729		return; /* resume from nested internal suspend */
2730
2731	if (dm_suspended_md(md))
2732		goto done; /* resume from nested suspend */
2733
2734	/*
2735	 * NOTE: existing callers don't need to call dm_table_resume_targets
2736	 * (which may fail -- so best to avoid it for now by passing NULL map)
2737	 */
2738	(void) __dm_resume(md, NULL);
2739
2740done:
2741	clear_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
2742	smp_mb__after_atomic();
2743	wake_up_bit(&md->flags, DMF_SUSPENDED_INTERNALLY);
2744}
2745
2746void dm_internal_suspend_noflush(struct mapped_device *md)
2747{
2748	mutex_lock(&md->suspend_lock);
2749	__dm_internal_suspend(md, DM_SUSPEND_NOFLUSH_FLAG);
2750	mutex_unlock(&md->suspend_lock);
2751}
2752EXPORT_SYMBOL_GPL(dm_internal_suspend_noflush);
2753
2754void dm_internal_resume(struct mapped_device *md)
2755{
2756	mutex_lock(&md->suspend_lock);
2757	__dm_internal_resume(md);
2758	mutex_unlock(&md->suspend_lock);
2759}
2760EXPORT_SYMBOL_GPL(dm_internal_resume);
2761
2762/*
2763 * Fast variants of internal suspend/resume hold md->suspend_lock,
2764 * which prevents interaction with userspace-driven suspend.
2765 */
2766
2767void dm_internal_suspend_fast(struct mapped_device *md)
2768{
2769	mutex_lock(&md->suspend_lock);
2770	if (dm_suspended_md(md) || dm_suspended_internally_md(md))
2771		return;
2772
2773	set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2774	synchronize_srcu(&md->io_barrier);
2775	flush_workqueue(md->wq);
2776	dm_wait_for_completion(md, TASK_UNINTERRUPTIBLE);
2777}
2778EXPORT_SYMBOL_GPL(dm_internal_suspend_fast);
2779
2780void dm_internal_resume_fast(struct mapped_device *md)
2781{
2782	if (dm_suspended_md(md) || dm_suspended_internally_md(md))
2783		goto done;
2784
2785	dm_queue_flush(md);
2786
2787done:
2788	mutex_unlock(&md->suspend_lock);
2789}
2790EXPORT_SYMBOL_GPL(dm_internal_resume_fast);
2791
2792/*-----------------------------------------------------------------
2793 * Event notification.
2794 *---------------------------------------------------------------*/
2795int dm_kobject_uevent(struct mapped_device *md, enum kobject_action action,
2796		       unsigned cookie)
2797{
2798	int r;
2799	unsigned noio_flag;
2800	char udev_cookie[DM_COOKIE_LENGTH];
2801	char *envp[] = { udev_cookie, NULL };
2802
2803	noio_flag = memalloc_noio_save();
2804
2805	if (!cookie)
2806		r = kobject_uevent(&disk_to_dev(md->disk)->kobj, action);
2807	else {
2808		snprintf(udev_cookie, DM_COOKIE_LENGTH, "%s=%u",
2809			 DM_COOKIE_ENV_VAR_NAME, cookie);
2810		r = kobject_uevent_env(&disk_to_dev(md->disk)->kobj,
2811				       action, envp);
2812	}
2813
2814	memalloc_noio_restore(noio_flag);
2815
2816	return r;
2817}
2818
2819uint32_t dm_next_uevent_seq(struct mapped_device *md)
2820{
2821	return atomic_add_return(1, &md->uevent_seq);
2822}
2823
2824uint32_t dm_get_event_nr(struct mapped_device *md)
2825{
2826	return atomic_read(&md->event_nr);
2827}
2828
2829int dm_wait_event(struct mapped_device *md, int event_nr)
2830{
2831	return wait_event_interruptible(md->eventq,
2832			(event_nr != atomic_read(&md->event_nr)));
2833}
2834
2835void dm_uevent_add(struct mapped_device *md, struct list_head *elist)
2836{
2837	unsigned long flags;
2838
2839	spin_lock_irqsave(&md->uevent_lock, flags);
2840	list_add(elist, &md->uevent_list);
2841	spin_unlock_irqrestore(&md->uevent_lock, flags);
2842}
2843
2844/*
2845 * The gendisk is only valid as long as you have a reference
2846 * count on 'md'.
2847 */
2848struct gendisk *dm_disk(struct mapped_device *md)
2849{
2850	return md->disk;
2851}
2852EXPORT_SYMBOL_GPL(dm_disk);
2853
2854struct kobject *dm_kobject(struct mapped_device *md)
2855{
2856	return &md->kobj_holder.kobj;
2857}
2858
2859struct mapped_device *dm_get_from_kobject(struct kobject *kobj)
2860{
2861	struct mapped_device *md;
2862
2863	md = container_of(kobj, struct mapped_device, kobj_holder.kobj);
2864
2865	spin_lock(&_minor_lock);
2866	if (test_bit(DMF_FREEING, &md->flags) || dm_deleting_md(md)) {
2867		md = NULL;
2868		goto out;
2869	}
2870	dm_get(md);
2871out:
2872	spin_unlock(&_minor_lock);
2873
2874	return md;
2875}
2876
2877int dm_suspended_md(struct mapped_device *md)
2878{
2879	return test_bit(DMF_SUSPENDED, &md->flags);
2880}
2881
2882static int dm_post_suspending_md(struct mapped_device *md)
2883{
2884	return test_bit(DMF_POST_SUSPENDING, &md->flags);
2885}
2886
2887int dm_suspended_internally_md(struct mapped_device *md)
2888{
2889	return test_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
2890}
2891
2892int dm_test_deferred_remove_flag(struct mapped_device *md)
2893{
2894	return test_bit(DMF_DEFERRED_REMOVE, &md->flags);
2895}
2896
2897int dm_suspended(struct dm_target *ti)
2898{
2899	return dm_suspended_md(ti->table->md);
2900}
2901EXPORT_SYMBOL_GPL(dm_suspended);
2902
2903int dm_post_suspending(struct dm_target *ti)
2904{
2905	return dm_post_suspending_md(ti->table->md);
2906}
2907EXPORT_SYMBOL_GPL(dm_post_suspending);
2908
2909int dm_noflush_suspending(struct dm_target *ti)
2910{
2911	return __noflush_suspending(ti->table->md);
2912}
2913EXPORT_SYMBOL_GPL(dm_noflush_suspending);
2914
2915struct dm_md_mempools *dm_alloc_md_mempools(struct mapped_device *md, enum dm_queue_mode type,
2916					    unsigned integrity, unsigned per_io_data_size,
2917					    unsigned min_pool_size)
2918{
2919	struct dm_md_mempools *pools = kzalloc_node(sizeof(*pools), GFP_KERNEL, md->numa_node_id);
2920	unsigned int pool_size = 0;
2921	unsigned int front_pad, io_front_pad;
2922	int ret;
2923
2924	if (!pools)
2925		return NULL;
2926
2927	switch (type) {
2928	case DM_TYPE_BIO_BASED:
2929	case DM_TYPE_DAX_BIO_BASED:
2930		pool_size = max(dm_get_reserved_bio_based_ios(), min_pool_size);
2931		front_pad = roundup(per_io_data_size, __alignof__(struct dm_target_io)) + offsetof(struct dm_target_io, clone);
2932		io_front_pad = roundup(front_pad,  __alignof__(struct dm_io)) + offsetof(struct dm_io, tio);
2933		ret = bioset_init(&pools->io_bs, pool_size, io_front_pad, 0);
2934		if (ret)
2935			goto out;
2936		if (integrity && bioset_integrity_create(&pools->io_bs, pool_size))
2937			goto out;
2938		break;
2939	case DM_TYPE_REQUEST_BASED:
2940		pool_size = max(dm_get_reserved_rq_based_ios(), min_pool_size);
2941		front_pad = offsetof(struct dm_rq_clone_bio_info, clone);
2942		/* per_io_data_size is used for blk-mq pdu at queue allocation */
2943		break;
2944	default:
2945		BUG();
2946	}
2947
2948	ret = bioset_init(&pools->bs, pool_size, front_pad, 0);
2949	if (ret)
2950		goto out;
2951
2952	if (integrity && bioset_integrity_create(&pools->bs, pool_size))
2953		goto out;
2954
2955	return pools;
2956
2957out:
2958	dm_free_md_mempools(pools);
2959
2960	return NULL;
2961}
2962
2963void dm_free_md_mempools(struct dm_md_mempools *pools)
2964{
2965	if (!pools)
2966		return;
2967
2968	bioset_exit(&pools->bs);
2969	bioset_exit(&pools->io_bs);
2970
2971	kfree(pools);
2972}
2973
2974struct dm_pr {
2975	u64	old_key;
2976	u64	new_key;
2977	u32	flags;
2978	bool	fail_early;
2979};
2980
2981static int dm_call_pr(struct block_device *bdev, iterate_devices_callout_fn fn,
2982		      void *data)
2983{
2984	struct mapped_device *md = bdev->bd_disk->private_data;
2985	struct dm_table *table;
2986	struct dm_target *ti;
2987	int ret = -ENOTTY, srcu_idx;
2988
2989	table = dm_get_live_table(md, &srcu_idx);
2990	if (!table || !dm_table_get_size(table))
2991		goto out;
2992
2993	/* We only support devices that have a single target */
2994	if (dm_table_get_num_targets(table) != 1)
2995		goto out;
2996	ti = dm_table_get_target(table, 0);
2997
2998	if (dm_suspended_md(md)) {
2999		ret = -EAGAIN;
3000		goto out;
3001	}
3002
3003	ret = -EINVAL;
3004	if (!ti->type->iterate_devices)
3005		goto out;
3006
3007	ret = ti->type->iterate_devices(ti, fn, data);
3008out:
3009	dm_put_live_table(md, srcu_idx);
3010	return ret;
3011}
3012
3013/*
3014 * For register / unregister we need to manually call out to every path.
3015 */
3016static int __dm_pr_register(struct dm_target *ti, struct dm_dev *dev,
3017			    sector_t start, sector_t len, void *data)
3018{
3019	struct dm_pr *pr = data;
3020	const struct pr_ops *ops = dev->bdev->bd_disk->fops->pr_ops;
3021
3022	if (!ops || !ops->pr_register)
3023		return -EOPNOTSUPP;
3024	return ops->pr_register(dev->bdev, pr->old_key, pr->new_key, pr->flags);
3025}
3026
3027static int dm_pr_register(struct block_device *bdev, u64 old_key, u64 new_key,
3028			  u32 flags)
3029{
3030	struct dm_pr pr = {
3031		.old_key	= old_key,
3032		.new_key	= new_key,
3033		.flags		= flags,
3034		.fail_early	= true,
3035	};
3036	int ret;
3037
3038	ret = dm_call_pr(bdev, __dm_pr_register, &pr);
3039	if (ret && new_key) {
3040		/* unregister all paths if we failed to register any path */
3041		pr.old_key = new_key;
3042		pr.new_key = 0;
3043		pr.flags = 0;
3044		pr.fail_early = false;
3045		dm_call_pr(bdev, __dm_pr_register, &pr);
3046	}
3047
3048	return ret;
3049}
3050
3051static int dm_pr_reserve(struct block_device *bdev, u64 key, enum pr_type type,
3052			 u32 flags)
3053{
3054	struct mapped_device *md = bdev->bd_disk->private_data;
3055	const struct pr_ops *ops;
3056	int r, srcu_idx;
3057
3058	r = dm_prepare_ioctl(md, &srcu_idx, &bdev);
3059	if (r < 0)
3060		goto out;
3061
3062	ops = bdev->bd_disk->fops->pr_ops;
3063	if (ops && ops->pr_reserve)
3064		r = ops->pr_reserve(bdev, key, type, flags);
3065	else
3066		r = -EOPNOTSUPP;
3067out:
3068	dm_unprepare_ioctl(md, srcu_idx);
3069	return r;
3070}
3071
3072static int dm_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
3073{
3074	struct mapped_device *md = bdev->bd_disk->private_data;
3075	const struct pr_ops *ops;
3076	int r, srcu_idx;
3077
3078	r = dm_prepare_ioctl(md, &srcu_idx, &bdev);
3079	if (r < 0)
3080		goto out;
3081
3082	ops = bdev->bd_disk->fops->pr_ops;
3083	if (ops && ops->pr_release)
3084		r = ops->pr_release(bdev, key, type);
3085	else
3086		r = -EOPNOTSUPP;
3087out:
3088	dm_unprepare_ioctl(md, srcu_idx);
3089	return r;
3090}
3091
3092static int dm_pr_preempt(struct block_device *bdev, u64 old_key, u64 new_key,
3093			 enum pr_type type, bool abort)
3094{
3095	struct mapped_device *md = bdev->bd_disk->private_data;
3096	const struct pr_ops *ops;
3097	int r, srcu_idx;
3098
3099	r = dm_prepare_ioctl(md, &srcu_idx, &bdev);
3100	if (r < 0)
3101		goto out;
3102
3103	ops = bdev->bd_disk->fops->pr_ops;
3104	if (ops && ops->pr_preempt)
3105		r = ops->pr_preempt(bdev, old_key, new_key, type, abort);
3106	else
3107		r = -EOPNOTSUPP;
3108out:
3109	dm_unprepare_ioctl(md, srcu_idx);
3110	return r;
3111}
3112
3113static int dm_pr_clear(struct block_device *bdev, u64 key)
3114{
3115	struct mapped_device *md = bdev->bd_disk->private_data;
3116	const struct pr_ops *ops;
3117	int r, srcu_idx;
3118
3119	r = dm_prepare_ioctl(md, &srcu_idx, &bdev);
3120	if (r < 0)
3121		goto out;
3122
3123	ops = bdev->bd_disk->fops->pr_ops;
3124	if (ops && ops->pr_clear)
3125		r = ops->pr_clear(bdev, key);
3126	else
3127		r = -EOPNOTSUPP;
3128out:
3129	dm_unprepare_ioctl(md, srcu_idx);
3130	return r;
3131}
3132
3133static const struct pr_ops dm_pr_ops = {
3134	.pr_register	= dm_pr_register,
3135	.pr_reserve	= dm_pr_reserve,
3136	.pr_release	= dm_pr_release,
3137	.pr_preempt	= dm_pr_preempt,
3138	.pr_clear	= dm_pr_clear,
3139};
3140
3141static const struct block_device_operations dm_blk_dops = {
3142	.submit_bio = dm_submit_bio,
3143	.open = dm_blk_open,
3144	.release = dm_blk_close,
3145	.ioctl = dm_blk_ioctl,
3146	.getgeo = dm_blk_getgeo,
3147	.report_zones = dm_blk_report_zones,
3148	.pr_ops = &dm_pr_ops,
3149	.owner = THIS_MODULE
3150};
3151
3152static const struct block_device_operations dm_rq_blk_dops = {
3153	.open = dm_blk_open,
3154	.release = dm_blk_close,
3155	.ioctl = dm_blk_ioctl,
3156	.getgeo = dm_blk_getgeo,
3157	.pr_ops = &dm_pr_ops,
3158	.owner = THIS_MODULE
3159};
3160
3161static const struct dax_operations dm_dax_ops = {
3162	.direct_access = dm_dax_direct_access,
3163	.dax_supported = dm_dax_supported,
3164	.copy_from_iter = dm_dax_copy_from_iter,
3165	.copy_to_iter = dm_dax_copy_to_iter,
3166	.zero_page_range = dm_dax_zero_page_range,
3167};
3168
3169/*
3170 * module hooks
3171 */
3172module_init(dm_init);
3173module_exit(dm_exit);
3174
3175module_param(major, uint, 0);
3176MODULE_PARM_DESC(major, "The major number of the device mapper");
3177
3178module_param(reserved_bio_based_ios, uint, S_IRUGO | S_IWUSR);
3179MODULE_PARM_DESC(reserved_bio_based_ios, "Reserved IOs in bio-based mempools");
3180
3181module_param(dm_numa_node, int, S_IRUGO | S_IWUSR);
3182MODULE_PARM_DESC(dm_numa_node, "NUMA node for DM device memory allocations");
3183
3184module_param(swap_bios, int, S_IRUGO | S_IWUSR);
3185MODULE_PARM_DESC(swap_bios, "Maximum allowed inflight swap IOs");
3186
3187MODULE_DESCRIPTION(DM_NAME " driver");
3188MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
3189MODULE_LICENSE("GPL");
3190