Loading...
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#include "dm-ima.h"
12
13#include <linux/init.h>
14#include <linux/module.h>
15#include <linux/mutex.h>
16#include <linux/sched/mm.h>
17#include <linux/sched/signal.h>
18#include <linux/blkpg.h>
19#include <linux/bio.h>
20#include <linux/mempool.h>
21#include <linux/dax.h>
22#include <linux/slab.h>
23#include <linux/idr.h>
24#include <linux/uio.h>
25#include <linux/hdreg.h>
26#include <linux/delay.h>
27#include <linux/wait.h>
28#include <linux/pr.h>
29#include <linux/refcount.h>
30#include <linux/part_stat.h>
31#include <linux/blk-crypto.h>
32#include <linux/blk-crypto-profile.h>
33
34#define DM_MSG_PREFIX "core"
35
36/*
37 * Cookies are numeric values sent with CHANGE and REMOVE
38 * uevents while resuming, removing or renaming the device.
39 */
40#define DM_COOKIE_ENV_VAR_NAME "DM_COOKIE"
41#define DM_COOKIE_LENGTH 24
42
43/*
44 * For REQ_POLLED fs bio, this flag is set if we link mapped underlying
45 * dm_io into one list, and reuse bio->bi_private as the list head. Before
46 * ending this fs bio, we will recover its ->bi_private.
47 */
48#define REQ_DM_POLL_LIST REQ_DRV
49
50static const char *_name = DM_NAME;
51
52static unsigned int major = 0;
53static unsigned int _major = 0;
54
55static DEFINE_IDR(_minor_idr);
56
57static DEFINE_SPINLOCK(_minor_lock);
58
59static void do_deferred_remove(struct work_struct *w);
60
61static DECLARE_WORK(deferred_remove_work, do_deferred_remove);
62
63static struct workqueue_struct *deferred_remove_workqueue;
64
65atomic_t dm_global_event_nr = ATOMIC_INIT(0);
66DECLARE_WAIT_QUEUE_HEAD(dm_global_eventq);
67
68void dm_issue_global_event(void)
69{
70 atomic_inc(&dm_global_event_nr);
71 wake_up(&dm_global_eventq);
72}
73
74DEFINE_STATIC_KEY_FALSE(stats_enabled);
75DEFINE_STATIC_KEY_FALSE(swap_bios_enabled);
76DEFINE_STATIC_KEY_FALSE(zoned_enabled);
77
78/*
79 * One of these is allocated (on-stack) per original bio.
80 */
81struct clone_info {
82 struct dm_table *map;
83 struct bio *bio;
84 struct dm_io *io;
85 sector_t sector;
86 unsigned sector_count;
87 bool is_abnormal_io:1;
88 bool submit_as_polled:1;
89};
90
91static inline struct dm_target_io *clone_to_tio(struct bio *clone)
92{
93 return container_of(clone, struct dm_target_io, clone);
94}
95
96void *dm_per_bio_data(struct bio *bio, size_t data_size)
97{
98 if (!dm_tio_flagged(clone_to_tio(bio), DM_TIO_INSIDE_DM_IO))
99 return (char *)bio - DM_TARGET_IO_BIO_OFFSET - data_size;
100 return (char *)bio - DM_IO_BIO_OFFSET - data_size;
101}
102EXPORT_SYMBOL_GPL(dm_per_bio_data);
103
104struct bio *dm_bio_from_per_bio_data(void *data, size_t data_size)
105{
106 struct dm_io *io = (struct dm_io *)((char *)data + data_size);
107 if (io->magic == DM_IO_MAGIC)
108 return (struct bio *)((char *)io + DM_IO_BIO_OFFSET);
109 BUG_ON(io->magic != DM_TIO_MAGIC);
110 return (struct bio *)((char *)io + DM_TARGET_IO_BIO_OFFSET);
111}
112EXPORT_SYMBOL_GPL(dm_bio_from_per_bio_data);
113
114unsigned dm_bio_get_target_bio_nr(const struct bio *bio)
115{
116 return container_of(bio, struct dm_target_io, clone)->target_bio_nr;
117}
118EXPORT_SYMBOL_GPL(dm_bio_get_target_bio_nr);
119
120#define MINOR_ALLOCED ((void *)-1)
121
122#define DM_NUMA_NODE NUMA_NO_NODE
123static int dm_numa_node = DM_NUMA_NODE;
124
125#define DEFAULT_SWAP_BIOS (8 * 1048576 / PAGE_SIZE)
126static int swap_bios = DEFAULT_SWAP_BIOS;
127static int get_swap_bios(void)
128{
129 int latch = READ_ONCE(swap_bios);
130 if (unlikely(latch <= 0))
131 latch = DEFAULT_SWAP_BIOS;
132 return latch;
133}
134
135struct table_device {
136 struct list_head list;
137 refcount_t count;
138 struct dm_dev dm_dev;
139};
140
141/*
142 * Bio-based DM's mempools' reserved IOs set by the user.
143 */
144#define RESERVED_BIO_BASED_IOS 16
145static unsigned reserved_bio_based_ios = RESERVED_BIO_BASED_IOS;
146
147static int __dm_get_module_param_int(int *module_param, int min, int max)
148{
149 int param = READ_ONCE(*module_param);
150 int modified_param = 0;
151 bool modified = true;
152
153 if (param < min)
154 modified_param = min;
155 else if (param > max)
156 modified_param = max;
157 else
158 modified = false;
159
160 if (modified) {
161 (void)cmpxchg(module_param, param, modified_param);
162 param = modified_param;
163 }
164
165 return param;
166}
167
168unsigned __dm_get_module_param(unsigned *module_param,
169 unsigned def, unsigned max)
170{
171 unsigned param = READ_ONCE(*module_param);
172 unsigned modified_param = 0;
173
174 if (!param)
175 modified_param = def;
176 else if (param > max)
177 modified_param = max;
178
179 if (modified_param) {
180 (void)cmpxchg(module_param, param, modified_param);
181 param = modified_param;
182 }
183
184 return param;
185}
186
187unsigned dm_get_reserved_bio_based_ios(void)
188{
189 return __dm_get_module_param(&reserved_bio_based_ios,
190 RESERVED_BIO_BASED_IOS, DM_RESERVED_MAX_IOS);
191}
192EXPORT_SYMBOL_GPL(dm_get_reserved_bio_based_ios);
193
194static unsigned dm_get_numa_node(void)
195{
196 return __dm_get_module_param_int(&dm_numa_node,
197 DM_NUMA_NODE, num_online_nodes() - 1);
198}
199
200static int __init local_init(void)
201{
202 int r;
203
204 r = dm_uevent_init();
205 if (r)
206 return r;
207
208 deferred_remove_workqueue = alloc_workqueue("kdmremove", WQ_UNBOUND, 1);
209 if (!deferred_remove_workqueue) {
210 r = -ENOMEM;
211 goto out_uevent_exit;
212 }
213
214 _major = major;
215 r = register_blkdev(_major, _name);
216 if (r < 0)
217 goto out_free_workqueue;
218
219 if (!_major)
220 _major = r;
221
222 return 0;
223
224out_free_workqueue:
225 destroy_workqueue(deferred_remove_workqueue);
226out_uevent_exit:
227 dm_uevent_exit();
228
229 return r;
230}
231
232static void local_exit(void)
233{
234 flush_scheduled_work();
235 destroy_workqueue(deferred_remove_workqueue);
236
237 unregister_blkdev(_major, _name);
238 dm_uevent_exit();
239
240 _major = 0;
241
242 DMINFO("cleaned up");
243}
244
245static int (*_inits[])(void) __initdata = {
246 local_init,
247 dm_target_init,
248 dm_linear_init,
249 dm_stripe_init,
250 dm_io_init,
251 dm_kcopyd_init,
252 dm_interface_init,
253 dm_statistics_init,
254};
255
256static void (*_exits[])(void) = {
257 local_exit,
258 dm_target_exit,
259 dm_linear_exit,
260 dm_stripe_exit,
261 dm_io_exit,
262 dm_kcopyd_exit,
263 dm_interface_exit,
264 dm_statistics_exit,
265};
266
267static int __init dm_init(void)
268{
269 const int count = ARRAY_SIZE(_inits);
270 int r, i;
271
272#if (IS_ENABLED(CONFIG_IMA) && !IS_ENABLED(CONFIG_IMA_DISABLE_HTABLE))
273 DMWARN("CONFIG_IMA_DISABLE_HTABLE is disabled."
274 " Duplicate IMA measurements will not be recorded in the IMA log.");
275#endif
276
277 for (i = 0; i < count; i++) {
278 r = _inits[i]();
279 if (r)
280 goto bad;
281 }
282
283 return 0;
284bad:
285 while (i--)
286 _exits[i]();
287
288 return r;
289}
290
291static void __exit dm_exit(void)
292{
293 int i = ARRAY_SIZE(_exits);
294
295 while (i--)
296 _exits[i]();
297
298 /*
299 * Should be empty by this point.
300 */
301 idr_destroy(&_minor_idr);
302}
303
304/*
305 * Block device functions
306 */
307int dm_deleting_md(struct mapped_device *md)
308{
309 return test_bit(DMF_DELETING, &md->flags);
310}
311
312static int dm_blk_open(struct block_device *bdev, fmode_t mode)
313{
314 struct mapped_device *md;
315
316 spin_lock(&_minor_lock);
317
318 md = bdev->bd_disk->private_data;
319 if (!md)
320 goto out;
321
322 if (test_bit(DMF_FREEING, &md->flags) ||
323 dm_deleting_md(md)) {
324 md = NULL;
325 goto out;
326 }
327
328 dm_get(md);
329 atomic_inc(&md->open_count);
330out:
331 spin_unlock(&_minor_lock);
332
333 return md ? 0 : -ENXIO;
334}
335
336static void dm_blk_close(struct gendisk *disk, fmode_t mode)
337{
338 struct mapped_device *md;
339
340 spin_lock(&_minor_lock);
341
342 md = disk->private_data;
343 if (WARN_ON(!md))
344 goto out;
345
346 if (atomic_dec_and_test(&md->open_count) &&
347 (test_bit(DMF_DEFERRED_REMOVE, &md->flags)))
348 queue_work(deferred_remove_workqueue, &deferred_remove_work);
349
350 dm_put(md);
351out:
352 spin_unlock(&_minor_lock);
353}
354
355int dm_open_count(struct mapped_device *md)
356{
357 return atomic_read(&md->open_count);
358}
359
360/*
361 * Guarantees nothing is using the device before it's deleted.
362 */
363int dm_lock_for_deletion(struct mapped_device *md, bool mark_deferred, bool only_deferred)
364{
365 int r = 0;
366
367 spin_lock(&_minor_lock);
368
369 if (dm_open_count(md)) {
370 r = -EBUSY;
371 if (mark_deferred)
372 set_bit(DMF_DEFERRED_REMOVE, &md->flags);
373 } else if (only_deferred && !test_bit(DMF_DEFERRED_REMOVE, &md->flags))
374 r = -EEXIST;
375 else
376 set_bit(DMF_DELETING, &md->flags);
377
378 spin_unlock(&_minor_lock);
379
380 return r;
381}
382
383int dm_cancel_deferred_remove(struct mapped_device *md)
384{
385 int r = 0;
386
387 spin_lock(&_minor_lock);
388
389 if (test_bit(DMF_DELETING, &md->flags))
390 r = -EBUSY;
391 else
392 clear_bit(DMF_DEFERRED_REMOVE, &md->flags);
393
394 spin_unlock(&_minor_lock);
395
396 return r;
397}
398
399static void do_deferred_remove(struct work_struct *w)
400{
401 dm_deferred_remove();
402}
403
404static int dm_blk_getgeo(struct block_device *bdev, struct hd_geometry *geo)
405{
406 struct mapped_device *md = bdev->bd_disk->private_data;
407
408 return dm_get_geometry(md, geo);
409}
410
411static int dm_prepare_ioctl(struct mapped_device *md, int *srcu_idx,
412 struct block_device **bdev)
413{
414 struct dm_target *ti;
415 struct dm_table *map;
416 int r;
417
418retry:
419 r = -ENOTTY;
420 map = dm_get_live_table(md, srcu_idx);
421 if (!map || !dm_table_get_size(map))
422 return r;
423
424 /* We only support devices that have a single target */
425 if (map->num_targets != 1)
426 return r;
427
428 ti = dm_table_get_target(map, 0);
429 if (!ti->type->prepare_ioctl)
430 return r;
431
432 if (dm_suspended_md(md))
433 return -EAGAIN;
434
435 r = ti->type->prepare_ioctl(ti, bdev);
436 if (r == -ENOTCONN && !fatal_signal_pending(current)) {
437 dm_put_live_table(md, *srcu_idx);
438 msleep(10);
439 goto retry;
440 }
441
442 return r;
443}
444
445static void dm_unprepare_ioctl(struct mapped_device *md, int srcu_idx)
446{
447 dm_put_live_table(md, srcu_idx);
448}
449
450static int dm_blk_ioctl(struct block_device *bdev, fmode_t mode,
451 unsigned int cmd, unsigned long arg)
452{
453 struct mapped_device *md = bdev->bd_disk->private_data;
454 int r, srcu_idx;
455
456 r = dm_prepare_ioctl(md, &srcu_idx, &bdev);
457 if (r < 0)
458 goto out;
459
460 if (r > 0) {
461 /*
462 * Target determined this ioctl is being issued against a
463 * subset of the parent bdev; require extra privileges.
464 */
465 if (!capable(CAP_SYS_RAWIO)) {
466 DMDEBUG_LIMIT(
467 "%s: sending ioctl %x to DM device without required privilege.",
468 current->comm, cmd);
469 r = -ENOIOCTLCMD;
470 goto out;
471 }
472 }
473
474 if (!bdev->bd_disk->fops->ioctl)
475 r = -ENOTTY;
476 else
477 r = bdev->bd_disk->fops->ioctl(bdev, mode, cmd, arg);
478out:
479 dm_unprepare_ioctl(md, srcu_idx);
480 return r;
481}
482
483u64 dm_start_time_ns_from_clone(struct bio *bio)
484{
485 return jiffies_to_nsecs(clone_to_tio(bio)->io->start_time);
486}
487EXPORT_SYMBOL_GPL(dm_start_time_ns_from_clone);
488
489static bool bio_is_flush_with_data(struct bio *bio)
490{
491 return ((bio->bi_opf & REQ_PREFLUSH) && bio->bi_iter.bi_size);
492}
493
494static void dm_io_acct(struct dm_io *io, bool end)
495{
496 struct dm_stats_aux *stats_aux = &io->stats_aux;
497 unsigned long start_time = io->start_time;
498 struct mapped_device *md = io->md;
499 struct bio *bio = io->orig_bio;
500 unsigned int sectors;
501
502 /*
503 * If REQ_PREFLUSH set, don't account payload, it will be
504 * submitted (and accounted) after this flush completes.
505 */
506 if (bio_is_flush_with_data(bio))
507 sectors = 0;
508 else if (likely(!(dm_io_flagged(io, DM_IO_WAS_SPLIT))))
509 sectors = bio_sectors(bio);
510 else
511 sectors = io->sectors;
512
513 if (!end)
514 bdev_start_io_acct(bio->bi_bdev, sectors, bio_op(bio),
515 start_time);
516 else
517 bdev_end_io_acct(bio->bi_bdev, bio_op(bio), start_time);
518
519 if (static_branch_unlikely(&stats_enabled) &&
520 unlikely(dm_stats_used(&md->stats))) {
521 sector_t sector;
522
523 if (likely(!dm_io_flagged(io, DM_IO_WAS_SPLIT)))
524 sector = bio->bi_iter.bi_sector;
525 else
526 sector = bio_end_sector(bio) - io->sector_offset;
527
528 dm_stats_account_io(&md->stats, bio_data_dir(bio),
529 sector, sectors,
530 end, start_time, stats_aux);
531 }
532}
533
534static void __dm_start_io_acct(struct dm_io *io)
535{
536 dm_io_acct(io, false);
537}
538
539static void dm_start_io_acct(struct dm_io *io, struct bio *clone)
540{
541 /*
542 * Ensure IO accounting is only ever started once.
543 */
544 if (dm_io_flagged(io, DM_IO_ACCOUNTED))
545 return;
546
547 /* Expect no possibility for race unless DM_TIO_IS_DUPLICATE_BIO. */
548 if (!clone || likely(dm_tio_is_normal(clone_to_tio(clone)))) {
549 dm_io_set_flag(io, DM_IO_ACCOUNTED);
550 } else {
551 unsigned long flags;
552 /* Can afford locking given DM_TIO_IS_DUPLICATE_BIO */
553 spin_lock_irqsave(&io->lock, flags);
554 if (dm_io_flagged(io, DM_IO_ACCOUNTED)) {
555 spin_unlock_irqrestore(&io->lock, flags);
556 return;
557 }
558 dm_io_set_flag(io, DM_IO_ACCOUNTED);
559 spin_unlock_irqrestore(&io->lock, flags);
560 }
561
562 __dm_start_io_acct(io);
563}
564
565static void dm_end_io_acct(struct dm_io *io)
566{
567 dm_io_acct(io, true);
568}
569
570static struct dm_io *alloc_io(struct mapped_device *md, struct bio *bio)
571{
572 struct dm_io *io;
573 struct dm_target_io *tio;
574 struct bio *clone;
575
576 clone = bio_alloc_clone(NULL, bio, GFP_NOIO, &md->mempools->io_bs);
577 tio = clone_to_tio(clone);
578 tio->flags = 0;
579 dm_tio_set_flag(tio, DM_TIO_INSIDE_DM_IO);
580 tio->io = NULL;
581
582 io = container_of(tio, struct dm_io, tio);
583 io->magic = DM_IO_MAGIC;
584 io->status = BLK_STS_OK;
585
586 /* one ref is for submission, the other is for completion */
587 atomic_set(&io->io_count, 2);
588 this_cpu_inc(*md->pending_io);
589 io->orig_bio = bio;
590 io->md = md;
591 spin_lock_init(&io->lock);
592 io->start_time = jiffies;
593 io->flags = 0;
594
595 if (static_branch_unlikely(&stats_enabled))
596 dm_stats_record_start(&md->stats, &io->stats_aux);
597
598 return io;
599}
600
601static void free_io(struct dm_io *io)
602{
603 bio_put(&io->tio.clone);
604}
605
606static struct bio *alloc_tio(struct clone_info *ci, struct dm_target *ti,
607 unsigned target_bio_nr, unsigned *len, gfp_t gfp_mask)
608{
609 struct mapped_device *md = ci->io->md;
610 struct dm_target_io *tio;
611 struct bio *clone;
612
613 if (!ci->io->tio.io) {
614 /* the dm_target_io embedded in ci->io is available */
615 tio = &ci->io->tio;
616 /* alloc_io() already initialized embedded clone */
617 clone = &tio->clone;
618 } else {
619 clone = bio_alloc_clone(NULL, ci->bio, gfp_mask,
620 &md->mempools->bs);
621 if (!clone)
622 return NULL;
623
624 /* REQ_DM_POLL_LIST shouldn't be inherited */
625 clone->bi_opf &= ~REQ_DM_POLL_LIST;
626
627 tio = clone_to_tio(clone);
628 tio->flags = 0; /* also clears DM_TIO_INSIDE_DM_IO */
629 }
630
631 tio->magic = DM_TIO_MAGIC;
632 tio->io = ci->io;
633 tio->ti = ti;
634 tio->target_bio_nr = target_bio_nr;
635 tio->len_ptr = len;
636 tio->old_sector = 0;
637
638 /* Set default bdev, but target must bio_set_dev() before issuing IO */
639 clone->bi_bdev = md->disk->part0;
640 if (unlikely(ti->needs_bio_set_dev))
641 bio_set_dev(clone, md->disk->part0);
642
643 if (len) {
644 clone->bi_iter.bi_size = to_bytes(*len);
645 if (bio_integrity(clone))
646 bio_integrity_trim(clone);
647 }
648
649 return clone;
650}
651
652static void free_tio(struct bio *clone)
653{
654 if (dm_tio_flagged(clone_to_tio(clone), DM_TIO_INSIDE_DM_IO))
655 return;
656 bio_put(clone);
657}
658
659/*
660 * Add the bio to the list of deferred io.
661 */
662static void queue_io(struct mapped_device *md, struct bio *bio)
663{
664 unsigned long flags;
665
666 spin_lock_irqsave(&md->deferred_lock, flags);
667 bio_list_add(&md->deferred, bio);
668 spin_unlock_irqrestore(&md->deferred_lock, flags);
669 queue_work(md->wq, &md->work);
670}
671
672/*
673 * Everyone (including functions in this file), should use this
674 * function to access the md->map field, and make sure they call
675 * dm_put_live_table() when finished.
676 */
677struct dm_table *dm_get_live_table(struct mapped_device *md,
678 int *srcu_idx) __acquires(md->io_barrier)
679{
680 *srcu_idx = srcu_read_lock(&md->io_barrier);
681
682 return srcu_dereference(md->map, &md->io_barrier);
683}
684
685void dm_put_live_table(struct mapped_device *md,
686 int srcu_idx) __releases(md->io_barrier)
687{
688 srcu_read_unlock(&md->io_barrier, srcu_idx);
689}
690
691void dm_sync_table(struct mapped_device *md)
692{
693 synchronize_srcu(&md->io_barrier);
694 synchronize_rcu_expedited();
695}
696
697/*
698 * A fast alternative to dm_get_live_table/dm_put_live_table.
699 * The caller must not block between these two functions.
700 */
701static struct dm_table *dm_get_live_table_fast(struct mapped_device *md) __acquires(RCU)
702{
703 rcu_read_lock();
704 return rcu_dereference(md->map);
705}
706
707static void dm_put_live_table_fast(struct mapped_device *md) __releases(RCU)
708{
709 rcu_read_unlock();
710}
711
712static inline struct dm_table *dm_get_live_table_bio(struct mapped_device *md,
713 int *srcu_idx, blk_opf_t bio_opf)
714{
715 if (bio_opf & REQ_NOWAIT)
716 return dm_get_live_table_fast(md);
717 else
718 return dm_get_live_table(md, srcu_idx);
719}
720
721static inline void dm_put_live_table_bio(struct mapped_device *md, int srcu_idx,
722 blk_opf_t bio_opf)
723{
724 if (bio_opf & REQ_NOWAIT)
725 dm_put_live_table_fast(md);
726 else
727 dm_put_live_table(md, srcu_idx);
728}
729
730static char *_dm_claim_ptr = "I belong to device-mapper";
731
732/*
733 * Open a table device so we can use it as a map destination.
734 */
735static struct table_device *open_table_device(struct mapped_device *md,
736 dev_t dev, fmode_t mode)
737{
738 struct table_device *td;
739 struct block_device *bdev;
740 u64 part_off;
741 int r;
742
743 td = kmalloc_node(sizeof(*td), GFP_KERNEL, md->numa_node_id);
744 if (!td)
745 return ERR_PTR(-ENOMEM);
746 refcount_set(&td->count, 1);
747
748 bdev = blkdev_get_by_dev(dev, mode | FMODE_EXCL, _dm_claim_ptr);
749 if (IS_ERR(bdev)) {
750 r = PTR_ERR(bdev);
751 goto out_free_td;
752 }
753
754 /*
755 * We can be called before the dm disk is added. In that case we can't
756 * register the holder relation here. It will be done once add_disk was
757 * called.
758 */
759 if (md->disk->slave_dir) {
760 r = bd_link_disk_holder(bdev, md->disk);
761 if (r)
762 goto out_blkdev_put;
763 }
764
765 td->dm_dev.mode = mode;
766 td->dm_dev.bdev = bdev;
767 td->dm_dev.dax_dev = fs_dax_get_by_bdev(bdev, &part_off, NULL, NULL);
768 format_dev_t(td->dm_dev.name, dev);
769 list_add(&td->list, &md->table_devices);
770 return td;
771
772out_blkdev_put:
773 blkdev_put(bdev, mode | FMODE_EXCL);
774out_free_td:
775 kfree(td);
776 return ERR_PTR(r);
777}
778
779/*
780 * Close a table device that we've been using.
781 */
782static void close_table_device(struct table_device *td, struct mapped_device *md)
783{
784 if (md->disk->slave_dir)
785 bd_unlink_disk_holder(td->dm_dev.bdev, md->disk);
786 blkdev_put(td->dm_dev.bdev, td->dm_dev.mode | FMODE_EXCL);
787 put_dax(td->dm_dev.dax_dev);
788 list_del(&td->list);
789 kfree(td);
790}
791
792static struct table_device *find_table_device(struct list_head *l, dev_t dev,
793 fmode_t mode)
794{
795 struct table_device *td;
796
797 list_for_each_entry(td, l, list)
798 if (td->dm_dev.bdev->bd_dev == dev && td->dm_dev.mode == mode)
799 return td;
800
801 return NULL;
802}
803
804int dm_get_table_device(struct mapped_device *md, dev_t dev, fmode_t mode,
805 struct dm_dev **result)
806{
807 struct table_device *td;
808
809 mutex_lock(&md->table_devices_lock);
810 td = find_table_device(&md->table_devices, dev, mode);
811 if (!td) {
812 td = open_table_device(md, dev, mode);
813 if (IS_ERR(td)) {
814 mutex_unlock(&md->table_devices_lock);
815 return PTR_ERR(td);
816 }
817 } else {
818 refcount_inc(&td->count);
819 }
820 mutex_unlock(&md->table_devices_lock);
821
822 *result = &td->dm_dev;
823 return 0;
824}
825
826void dm_put_table_device(struct mapped_device *md, struct dm_dev *d)
827{
828 struct table_device *td = container_of(d, struct table_device, dm_dev);
829
830 mutex_lock(&md->table_devices_lock);
831 if (refcount_dec_and_test(&td->count))
832 close_table_device(td, md);
833 mutex_unlock(&md->table_devices_lock);
834}
835
836/*
837 * Get the geometry associated with a dm device
838 */
839int dm_get_geometry(struct mapped_device *md, struct hd_geometry *geo)
840{
841 *geo = md->geometry;
842
843 return 0;
844}
845
846/*
847 * Set the geometry of a device.
848 */
849int dm_set_geometry(struct mapped_device *md, struct hd_geometry *geo)
850{
851 sector_t sz = (sector_t)geo->cylinders * geo->heads * geo->sectors;
852
853 if (geo->start > sz) {
854 DMERR("Start sector is beyond the geometry limits.");
855 return -EINVAL;
856 }
857
858 md->geometry = *geo;
859
860 return 0;
861}
862
863static int __noflush_suspending(struct mapped_device *md)
864{
865 return test_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
866}
867
868static void dm_requeue_add_io(struct dm_io *io, bool first_stage)
869{
870 struct mapped_device *md = io->md;
871
872 if (first_stage) {
873 struct dm_io *next = md->requeue_list;
874
875 md->requeue_list = io;
876 io->next = next;
877 } else {
878 bio_list_add_head(&md->deferred, io->orig_bio);
879 }
880}
881
882static void dm_kick_requeue(struct mapped_device *md, bool first_stage)
883{
884 if (first_stage)
885 queue_work(md->wq, &md->requeue_work);
886 else
887 queue_work(md->wq, &md->work);
888}
889
890/*
891 * Return true if the dm_io's original bio is requeued.
892 * io->status is updated with error if requeue disallowed.
893 */
894static bool dm_handle_requeue(struct dm_io *io, bool first_stage)
895{
896 struct bio *bio = io->orig_bio;
897 bool handle_requeue = (io->status == BLK_STS_DM_REQUEUE);
898 bool handle_polled_eagain = ((io->status == BLK_STS_AGAIN) &&
899 (bio->bi_opf & REQ_POLLED));
900 struct mapped_device *md = io->md;
901 bool requeued = false;
902
903 if (handle_requeue || handle_polled_eagain) {
904 unsigned long flags;
905
906 if (bio->bi_opf & REQ_POLLED) {
907 /*
908 * Upper layer won't help us poll split bio
909 * (io->orig_bio may only reflect a subset of the
910 * pre-split original) so clear REQ_POLLED.
911 */
912 bio_clear_polled(bio);
913 }
914
915 /*
916 * Target requested pushing back the I/O or
917 * polled IO hit BLK_STS_AGAIN.
918 */
919 spin_lock_irqsave(&md->deferred_lock, flags);
920 if ((__noflush_suspending(md) &&
921 !WARN_ON_ONCE(dm_is_zone_write(md, bio))) ||
922 handle_polled_eagain || first_stage) {
923 dm_requeue_add_io(io, first_stage);
924 requeued = true;
925 } else {
926 /*
927 * noflush suspend was interrupted or this is
928 * a write to a zoned target.
929 */
930 io->status = BLK_STS_IOERR;
931 }
932 spin_unlock_irqrestore(&md->deferred_lock, flags);
933 }
934
935 if (requeued)
936 dm_kick_requeue(md, first_stage);
937
938 return requeued;
939}
940
941static void __dm_io_complete(struct dm_io *io, bool first_stage)
942{
943 struct bio *bio = io->orig_bio;
944 struct mapped_device *md = io->md;
945 blk_status_t io_error;
946 bool requeued;
947
948 requeued = dm_handle_requeue(io, first_stage);
949 if (requeued && first_stage)
950 return;
951
952 io_error = io->status;
953 if (dm_io_flagged(io, DM_IO_ACCOUNTED))
954 dm_end_io_acct(io);
955 else if (!io_error) {
956 /*
957 * Must handle target that DM_MAPIO_SUBMITTED only to
958 * then bio_endio() rather than dm_submit_bio_remap()
959 */
960 __dm_start_io_acct(io);
961 dm_end_io_acct(io);
962 }
963 free_io(io);
964 smp_wmb();
965 this_cpu_dec(*md->pending_io);
966
967 /* nudge anyone waiting on suspend queue */
968 if (unlikely(wq_has_sleeper(&md->wait)))
969 wake_up(&md->wait);
970
971 /* Return early if the original bio was requeued */
972 if (requeued)
973 return;
974
975 if (bio_is_flush_with_data(bio)) {
976 /*
977 * Preflush done for flush with data, reissue
978 * without REQ_PREFLUSH.
979 */
980 bio->bi_opf &= ~REQ_PREFLUSH;
981 queue_io(md, bio);
982 } else {
983 /* done with normal IO or empty flush */
984 if (io_error)
985 bio->bi_status = io_error;
986 bio_endio(bio);
987 }
988}
989
990static void dm_wq_requeue_work(struct work_struct *work)
991{
992 struct mapped_device *md = container_of(work, struct mapped_device,
993 requeue_work);
994 unsigned long flags;
995 struct dm_io *io;
996
997 /* reuse deferred lock to simplify dm_handle_requeue */
998 spin_lock_irqsave(&md->deferred_lock, flags);
999 io = md->requeue_list;
1000 md->requeue_list = NULL;
1001 spin_unlock_irqrestore(&md->deferred_lock, flags);
1002
1003 while (io) {
1004 struct dm_io *next = io->next;
1005
1006 dm_io_rewind(io, &md->disk->bio_split);
1007
1008 io->next = NULL;
1009 __dm_io_complete(io, false);
1010 io = next;
1011 }
1012}
1013
1014/*
1015 * Two staged requeue:
1016 *
1017 * 1) io->orig_bio points to the real original bio, and the part mapped to
1018 * this io must be requeued, instead of other parts of the original bio.
1019 *
1020 * 2) io->orig_bio points to new cloned bio which matches the requeued dm_io.
1021 */
1022static void dm_io_complete(struct dm_io *io)
1023{
1024 bool first_requeue;
1025
1026 /*
1027 * Only dm_io that has been split needs two stage requeue, otherwise
1028 * we may run into long bio clone chain during suspend and OOM could
1029 * be triggered.
1030 *
1031 * Also flush data dm_io won't be marked as DM_IO_WAS_SPLIT, so they
1032 * also aren't handled via the first stage requeue.
1033 */
1034 if (dm_io_flagged(io, DM_IO_WAS_SPLIT))
1035 first_requeue = true;
1036 else
1037 first_requeue = false;
1038
1039 __dm_io_complete(io, first_requeue);
1040}
1041
1042/*
1043 * Decrements the number of outstanding ios that a bio has been
1044 * cloned into, completing the original io if necc.
1045 */
1046static inline void __dm_io_dec_pending(struct dm_io *io)
1047{
1048 if (atomic_dec_and_test(&io->io_count))
1049 dm_io_complete(io);
1050}
1051
1052static void dm_io_set_error(struct dm_io *io, blk_status_t error)
1053{
1054 unsigned long flags;
1055
1056 /* Push-back supersedes any I/O errors */
1057 spin_lock_irqsave(&io->lock, flags);
1058 if (!(io->status == BLK_STS_DM_REQUEUE &&
1059 __noflush_suspending(io->md))) {
1060 io->status = error;
1061 }
1062 spin_unlock_irqrestore(&io->lock, flags);
1063}
1064
1065static void dm_io_dec_pending(struct dm_io *io, blk_status_t error)
1066{
1067 if (unlikely(error))
1068 dm_io_set_error(io, error);
1069
1070 __dm_io_dec_pending(io);
1071}
1072
1073void disable_discard(struct mapped_device *md)
1074{
1075 struct queue_limits *limits = dm_get_queue_limits(md);
1076
1077 /* device doesn't really support DISCARD, disable it */
1078 limits->max_discard_sectors = 0;
1079}
1080
1081void disable_write_zeroes(struct mapped_device *md)
1082{
1083 struct queue_limits *limits = dm_get_queue_limits(md);
1084
1085 /* device doesn't really support WRITE ZEROES, disable it */
1086 limits->max_write_zeroes_sectors = 0;
1087}
1088
1089static bool swap_bios_limit(struct dm_target *ti, struct bio *bio)
1090{
1091 return unlikely((bio->bi_opf & REQ_SWAP) != 0) && unlikely(ti->limit_swap_bios);
1092}
1093
1094static void clone_endio(struct bio *bio)
1095{
1096 blk_status_t error = bio->bi_status;
1097 struct dm_target_io *tio = clone_to_tio(bio);
1098 struct dm_target *ti = tio->ti;
1099 dm_endio_fn endio = ti->type->end_io;
1100 struct dm_io *io = tio->io;
1101 struct mapped_device *md = io->md;
1102
1103 if (unlikely(error == BLK_STS_TARGET)) {
1104 if (bio_op(bio) == REQ_OP_DISCARD &&
1105 !bdev_max_discard_sectors(bio->bi_bdev))
1106 disable_discard(md);
1107 else if (bio_op(bio) == REQ_OP_WRITE_ZEROES &&
1108 !bdev_write_zeroes_sectors(bio->bi_bdev))
1109 disable_write_zeroes(md);
1110 }
1111
1112 if (static_branch_unlikely(&zoned_enabled) &&
1113 unlikely(bdev_is_zoned(bio->bi_bdev)))
1114 dm_zone_endio(io, bio);
1115
1116 if (endio) {
1117 int r = endio(ti, bio, &error);
1118 switch (r) {
1119 case DM_ENDIO_REQUEUE:
1120 if (static_branch_unlikely(&zoned_enabled)) {
1121 /*
1122 * Requeuing writes to a sequential zone of a zoned
1123 * target will break the sequential write pattern:
1124 * fail such IO.
1125 */
1126 if (WARN_ON_ONCE(dm_is_zone_write(md, bio)))
1127 error = BLK_STS_IOERR;
1128 else
1129 error = BLK_STS_DM_REQUEUE;
1130 } else
1131 error = BLK_STS_DM_REQUEUE;
1132 fallthrough;
1133 case DM_ENDIO_DONE:
1134 break;
1135 case DM_ENDIO_INCOMPLETE:
1136 /* The target will handle the io */
1137 return;
1138 default:
1139 DMCRIT("unimplemented target endio return value: %d", r);
1140 BUG();
1141 }
1142 }
1143
1144 if (static_branch_unlikely(&swap_bios_enabled) &&
1145 unlikely(swap_bios_limit(ti, bio)))
1146 up(&md->swap_bios_semaphore);
1147
1148 free_tio(bio);
1149 dm_io_dec_pending(io, error);
1150}
1151
1152/*
1153 * Return maximum size of I/O possible at the supplied sector up to the current
1154 * target boundary.
1155 */
1156static inline sector_t max_io_len_target_boundary(struct dm_target *ti,
1157 sector_t target_offset)
1158{
1159 return ti->len - target_offset;
1160}
1161
1162static sector_t max_io_len(struct dm_target *ti, sector_t sector)
1163{
1164 sector_t target_offset = dm_target_offset(ti, sector);
1165 sector_t len = max_io_len_target_boundary(ti, target_offset);
1166
1167 /*
1168 * Does the target need to split IO even further?
1169 * - varied (per target) IO splitting is a tenet of DM; this
1170 * explains why stacked chunk_sectors based splitting via
1171 * bio_split_to_limits() isn't possible here.
1172 */
1173 if (!ti->max_io_len)
1174 return len;
1175 return min_t(sector_t, len,
1176 min(queue_max_sectors(ti->table->md->queue),
1177 blk_chunk_sectors_left(target_offset, ti->max_io_len)));
1178}
1179
1180int dm_set_target_max_io_len(struct dm_target *ti, sector_t len)
1181{
1182 if (len > UINT_MAX) {
1183 DMERR("Specified maximum size of target IO (%llu) exceeds limit (%u)",
1184 (unsigned long long)len, UINT_MAX);
1185 ti->error = "Maximum size of target IO is too large";
1186 return -EINVAL;
1187 }
1188
1189 ti->max_io_len = (uint32_t) len;
1190
1191 return 0;
1192}
1193EXPORT_SYMBOL_GPL(dm_set_target_max_io_len);
1194
1195static struct dm_target *dm_dax_get_live_target(struct mapped_device *md,
1196 sector_t sector, int *srcu_idx)
1197 __acquires(md->io_barrier)
1198{
1199 struct dm_table *map;
1200 struct dm_target *ti;
1201
1202 map = dm_get_live_table(md, srcu_idx);
1203 if (!map)
1204 return NULL;
1205
1206 ti = dm_table_find_target(map, sector);
1207 if (!ti)
1208 return NULL;
1209
1210 return ti;
1211}
1212
1213static long dm_dax_direct_access(struct dax_device *dax_dev, pgoff_t pgoff,
1214 long nr_pages, enum dax_access_mode mode, void **kaddr,
1215 pfn_t *pfn)
1216{
1217 struct mapped_device *md = dax_get_private(dax_dev);
1218 sector_t sector = pgoff * PAGE_SECTORS;
1219 struct dm_target *ti;
1220 long len, ret = -EIO;
1221 int srcu_idx;
1222
1223 ti = dm_dax_get_live_target(md, sector, &srcu_idx);
1224
1225 if (!ti)
1226 goto out;
1227 if (!ti->type->direct_access)
1228 goto out;
1229 len = max_io_len(ti, sector) / PAGE_SECTORS;
1230 if (len < 1)
1231 goto out;
1232 nr_pages = min(len, nr_pages);
1233 ret = ti->type->direct_access(ti, pgoff, nr_pages, mode, kaddr, pfn);
1234
1235 out:
1236 dm_put_live_table(md, srcu_idx);
1237
1238 return ret;
1239}
1240
1241static int dm_dax_zero_page_range(struct dax_device *dax_dev, pgoff_t pgoff,
1242 size_t nr_pages)
1243{
1244 struct mapped_device *md = dax_get_private(dax_dev);
1245 sector_t sector = pgoff * PAGE_SECTORS;
1246 struct dm_target *ti;
1247 int ret = -EIO;
1248 int srcu_idx;
1249
1250 ti = dm_dax_get_live_target(md, sector, &srcu_idx);
1251
1252 if (!ti)
1253 goto out;
1254 if (WARN_ON(!ti->type->dax_zero_page_range)) {
1255 /*
1256 * ->zero_page_range() is mandatory dax operation. If we are
1257 * here, something is wrong.
1258 */
1259 goto out;
1260 }
1261 ret = ti->type->dax_zero_page_range(ti, pgoff, nr_pages);
1262 out:
1263 dm_put_live_table(md, srcu_idx);
1264
1265 return ret;
1266}
1267
1268static size_t dm_dax_recovery_write(struct dax_device *dax_dev, pgoff_t pgoff,
1269 void *addr, size_t bytes, struct iov_iter *i)
1270{
1271 struct mapped_device *md = dax_get_private(dax_dev);
1272 sector_t sector = pgoff * PAGE_SECTORS;
1273 struct dm_target *ti;
1274 int srcu_idx;
1275 long ret = 0;
1276
1277 ti = dm_dax_get_live_target(md, sector, &srcu_idx);
1278 if (!ti || !ti->type->dax_recovery_write)
1279 goto out;
1280
1281 ret = ti->type->dax_recovery_write(ti, pgoff, addr, bytes, i);
1282out:
1283 dm_put_live_table(md, srcu_idx);
1284 return ret;
1285}
1286
1287/*
1288 * A target may call dm_accept_partial_bio only from the map routine. It is
1289 * allowed for all bio types except REQ_PREFLUSH, REQ_OP_ZONE_* zone management
1290 * operations, REQ_OP_ZONE_APPEND (zone append writes) and any bio serviced by
1291 * __send_duplicate_bios().
1292 *
1293 * dm_accept_partial_bio informs the dm that the target only wants to process
1294 * additional n_sectors sectors of the bio and the rest of the data should be
1295 * sent in a next bio.
1296 *
1297 * A diagram that explains the arithmetics:
1298 * +--------------------+---------------+-------+
1299 * | 1 | 2 | 3 |
1300 * +--------------------+---------------+-------+
1301 *
1302 * <-------------- *tio->len_ptr --------------->
1303 * <----- bio_sectors ----->
1304 * <-- n_sectors -->
1305 *
1306 * Region 1 was already iterated over with bio_advance or similar function.
1307 * (it may be empty if the target doesn't use bio_advance)
1308 * Region 2 is the remaining bio size that the target wants to process.
1309 * (it may be empty if region 1 is non-empty, although there is no reason
1310 * to make it empty)
1311 * The target requires that region 3 is to be sent in the next bio.
1312 *
1313 * If the target wants to receive multiple copies of the bio (via num_*bios, etc),
1314 * the partially processed part (the sum of regions 1+2) must be the same for all
1315 * copies of the bio.
1316 */
1317void dm_accept_partial_bio(struct bio *bio, unsigned n_sectors)
1318{
1319 struct dm_target_io *tio = clone_to_tio(bio);
1320 struct dm_io *io = tio->io;
1321 unsigned bio_sectors = bio_sectors(bio);
1322
1323 BUG_ON(dm_tio_flagged(tio, DM_TIO_IS_DUPLICATE_BIO));
1324 BUG_ON(op_is_zone_mgmt(bio_op(bio)));
1325 BUG_ON(bio_op(bio) == REQ_OP_ZONE_APPEND);
1326 BUG_ON(bio_sectors > *tio->len_ptr);
1327 BUG_ON(n_sectors > bio_sectors);
1328
1329 *tio->len_ptr -= bio_sectors - n_sectors;
1330 bio->bi_iter.bi_size = n_sectors << SECTOR_SHIFT;
1331
1332 /*
1333 * __split_and_process_bio() may have already saved mapped part
1334 * for accounting but it is being reduced so update accordingly.
1335 */
1336 dm_io_set_flag(io, DM_IO_WAS_SPLIT);
1337 io->sectors = n_sectors;
1338 io->sector_offset = bio_sectors(io->orig_bio);
1339}
1340EXPORT_SYMBOL_GPL(dm_accept_partial_bio);
1341
1342/*
1343 * @clone: clone bio that DM core passed to target's .map function
1344 * @tgt_clone: clone of @clone bio that target needs submitted
1345 *
1346 * Targets should use this interface to submit bios they take
1347 * ownership of when returning DM_MAPIO_SUBMITTED.
1348 *
1349 * Target should also enable ti->accounts_remapped_io
1350 */
1351void dm_submit_bio_remap(struct bio *clone, struct bio *tgt_clone)
1352{
1353 struct dm_target_io *tio = clone_to_tio(clone);
1354 struct dm_io *io = tio->io;
1355
1356 /* establish bio that will get submitted */
1357 if (!tgt_clone)
1358 tgt_clone = clone;
1359
1360 /*
1361 * Account io->origin_bio to DM dev on behalf of target
1362 * that took ownership of IO with DM_MAPIO_SUBMITTED.
1363 */
1364 dm_start_io_acct(io, clone);
1365
1366 trace_block_bio_remap(tgt_clone, disk_devt(io->md->disk),
1367 tio->old_sector);
1368 submit_bio_noacct(tgt_clone);
1369}
1370EXPORT_SYMBOL_GPL(dm_submit_bio_remap);
1371
1372static noinline void __set_swap_bios_limit(struct mapped_device *md, int latch)
1373{
1374 mutex_lock(&md->swap_bios_lock);
1375 while (latch < md->swap_bios) {
1376 cond_resched();
1377 down(&md->swap_bios_semaphore);
1378 md->swap_bios--;
1379 }
1380 while (latch > md->swap_bios) {
1381 cond_resched();
1382 up(&md->swap_bios_semaphore);
1383 md->swap_bios++;
1384 }
1385 mutex_unlock(&md->swap_bios_lock);
1386}
1387
1388static void __map_bio(struct bio *clone)
1389{
1390 struct dm_target_io *tio = clone_to_tio(clone);
1391 struct dm_target *ti = tio->ti;
1392 struct dm_io *io = tio->io;
1393 struct mapped_device *md = io->md;
1394 int r;
1395
1396 clone->bi_end_io = clone_endio;
1397
1398 /*
1399 * Map the clone.
1400 */
1401 tio->old_sector = clone->bi_iter.bi_sector;
1402
1403 if (static_branch_unlikely(&swap_bios_enabled) &&
1404 unlikely(swap_bios_limit(ti, clone))) {
1405 int latch = get_swap_bios();
1406 if (unlikely(latch != md->swap_bios))
1407 __set_swap_bios_limit(md, latch);
1408 down(&md->swap_bios_semaphore);
1409 }
1410
1411 if (static_branch_unlikely(&zoned_enabled)) {
1412 /*
1413 * Check if the IO needs a special mapping due to zone append
1414 * emulation on zoned target. In this case, dm_zone_map_bio()
1415 * calls the target map operation.
1416 */
1417 if (unlikely(dm_emulate_zone_append(md)))
1418 r = dm_zone_map_bio(tio);
1419 else
1420 r = ti->type->map(ti, clone);
1421 } else
1422 r = ti->type->map(ti, clone);
1423
1424 switch (r) {
1425 case DM_MAPIO_SUBMITTED:
1426 /* target has assumed ownership of this io */
1427 if (!ti->accounts_remapped_io)
1428 dm_start_io_acct(io, clone);
1429 break;
1430 case DM_MAPIO_REMAPPED:
1431 dm_submit_bio_remap(clone, NULL);
1432 break;
1433 case DM_MAPIO_KILL:
1434 case DM_MAPIO_REQUEUE:
1435 if (static_branch_unlikely(&swap_bios_enabled) &&
1436 unlikely(swap_bios_limit(ti, clone)))
1437 up(&md->swap_bios_semaphore);
1438 free_tio(clone);
1439 if (r == DM_MAPIO_KILL)
1440 dm_io_dec_pending(io, BLK_STS_IOERR);
1441 else
1442 dm_io_dec_pending(io, BLK_STS_DM_REQUEUE);
1443 break;
1444 default:
1445 DMCRIT("unimplemented target map return value: %d", r);
1446 BUG();
1447 }
1448}
1449
1450static void setup_split_accounting(struct clone_info *ci, unsigned len)
1451{
1452 struct dm_io *io = ci->io;
1453
1454 if (ci->sector_count > len) {
1455 /*
1456 * Split needed, save the mapped part for accounting.
1457 * NOTE: dm_accept_partial_bio() will update accordingly.
1458 */
1459 dm_io_set_flag(io, DM_IO_WAS_SPLIT);
1460 io->sectors = len;
1461 io->sector_offset = bio_sectors(ci->bio);
1462 }
1463}
1464
1465static void alloc_multiple_bios(struct bio_list *blist, struct clone_info *ci,
1466 struct dm_target *ti, unsigned num_bios)
1467{
1468 struct bio *bio;
1469 int try;
1470
1471 for (try = 0; try < 2; try++) {
1472 int bio_nr;
1473
1474 if (try)
1475 mutex_lock(&ci->io->md->table_devices_lock);
1476 for (bio_nr = 0; bio_nr < num_bios; bio_nr++) {
1477 bio = alloc_tio(ci, ti, bio_nr, NULL,
1478 try ? GFP_NOIO : GFP_NOWAIT);
1479 if (!bio)
1480 break;
1481
1482 bio_list_add(blist, bio);
1483 }
1484 if (try)
1485 mutex_unlock(&ci->io->md->table_devices_lock);
1486 if (bio_nr == num_bios)
1487 return;
1488
1489 while ((bio = bio_list_pop(blist)))
1490 free_tio(bio);
1491 }
1492}
1493
1494static int __send_duplicate_bios(struct clone_info *ci, struct dm_target *ti,
1495 unsigned int num_bios, unsigned *len)
1496{
1497 struct bio_list blist = BIO_EMPTY_LIST;
1498 struct bio *clone;
1499 unsigned int ret = 0;
1500
1501 switch (num_bios) {
1502 case 0:
1503 break;
1504 case 1:
1505 if (len)
1506 setup_split_accounting(ci, *len);
1507 clone = alloc_tio(ci, ti, 0, len, GFP_NOIO);
1508 __map_bio(clone);
1509 ret = 1;
1510 break;
1511 default:
1512 /* dm_accept_partial_bio() is not supported with shared tio->len_ptr */
1513 alloc_multiple_bios(&blist, ci, ti, num_bios);
1514 while ((clone = bio_list_pop(&blist))) {
1515 dm_tio_set_flag(clone_to_tio(clone), DM_TIO_IS_DUPLICATE_BIO);
1516 __map_bio(clone);
1517 ret += 1;
1518 }
1519 break;
1520 }
1521
1522 return ret;
1523}
1524
1525static void __send_empty_flush(struct clone_info *ci)
1526{
1527 struct dm_table *t = ci->map;
1528 struct bio flush_bio;
1529
1530 /*
1531 * Use an on-stack bio for this, it's safe since we don't
1532 * need to reference it after submit. It's just used as
1533 * the basis for the clone(s).
1534 */
1535 bio_init(&flush_bio, ci->io->md->disk->part0, NULL, 0,
1536 REQ_OP_WRITE | REQ_PREFLUSH | REQ_SYNC);
1537
1538 ci->bio = &flush_bio;
1539 ci->sector_count = 0;
1540 ci->io->tio.clone.bi_iter.bi_size = 0;
1541
1542 for (unsigned int i = 0; i < t->num_targets; i++) {
1543 unsigned int bios;
1544 struct dm_target *ti = dm_table_get_target(t, i);
1545
1546 atomic_add(ti->num_flush_bios, &ci->io->io_count);
1547 bios = __send_duplicate_bios(ci, ti, ti->num_flush_bios, NULL);
1548 atomic_sub(ti->num_flush_bios - bios, &ci->io->io_count);
1549 }
1550
1551 /*
1552 * alloc_io() takes one extra reference for submission, so the
1553 * reference won't reach 0 without the following subtraction
1554 */
1555 atomic_sub(1, &ci->io->io_count);
1556
1557 bio_uninit(ci->bio);
1558}
1559
1560static void __send_changing_extent_only(struct clone_info *ci, struct dm_target *ti,
1561 unsigned num_bios)
1562{
1563 unsigned len;
1564 unsigned int bios;
1565
1566 len = min_t(sector_t, ci->sector_count,
1567 max_io_len_target_boundary(ti, dm_target_offset(ti, ci->sector)));
1568
1569 atomic_add(num_bios, &ci->io->io_count);
1570 bios = __send_duplicate_bios(ci, ti, num_bios, &len);
1571 /*
1572 * alloc_io() takes one extra reference for submission, so the
1573 * reference won't reach 0 without the following (+1) subtraction
1574 */
1575 atomic_sub(num_bios - bios + 1, &ci->io->io_count);
1576
1577 ci->sector += len;
1578 ci->sector_count -= len;
1579}
1580
1581static bool is_abnormal_io(struct bio *bio)
1582{
1583 enum req_op op = bio_op(bio);
1584
1585 if (op != REQ_OP_READ && op != REQ_OP_WRITE && op != REQ_OP_FLUSH) {
1586 switch (op) {
1587 case REQ_OP_DISCARD:
1588 case REQ_OP_SECURE_ERASE:
1589 case REQ_OP_WRITE_ZEROES:
1590 return true;
1591 default:
1592 break;
1593 }
1594 }
1595
1596 return false;
1597}
1598
1599static blk_status_t __process_abnormal_io(struct clone_info *ci,
1600 struct dm_target *ti)
1601{
1602 unsigned num_bios = 0;
1603
1604 switch (bio_op(ci->bio)) {
1605 case REQ_OP_DISCARD:
1606 num_bios = ti->num_discard_bios;
1607 break;
1608 case REQ_OP_SECURE_ERASE:
1609 num_bios = ti->num_secure_erase_bios;
1610 break;
1611 case REQ_OP_WRITE_ZEROES:
1612 num_bios = ti->num_write_zeroes_bios;
1613 break;
1614 default:
1615 break;
1616 }
1617
1618 /*
1619 * Even though the device advertised support for this type of
1620 * request, that does not mean every target supports it, and
1621 * reconfiguration might also have changed that since the
1622 * check was performed.
1623 */
1624 if (unlikely(!num_bios))
1625 return BLK_STS_NOTSUPP;
1626
1627 __send_changing_extent_only(ci, ti, num_bios);
1628 return BLK_STS_OK;
1629}
1630
1631/*
1632 * Reuse ->bi_private as dm_io list head for storing all dm_io instances
1633 * associated with this bio, and this bio's bi_private needs to be
1634 * stored in dm_io->data before the reuse.
1635 *
1636 * bio->bi_private is owned by fs or upper layer, so block layer won't
1637 * touch it after splitting. Meantime it won't be changed by anyone after
1638 * bio is submitted. So this reuse is safe.
1639 */
1640static inline struct dm_io **dm_poll_list_head(struct bio *bio)
1641{
1642 return (struct dm_io **)&bio->bi_private;
1643}
1644
1645static void dm_queue_poll_io(struct bio *bio, struct dm_io *io)
1646{
1647 struct dm_io **head = dm_poll_list_head(bio);
1648
1649 if (!(bio->bi_opf & REQ_DM_POLL_LIST)) {
1650 bio->bi_opf |= REQ_DM_POLL_LIST;
1651 /*
1652 * Save .bi_private into dm_io, so that we can reuse
1653 * .bi_private as dm_io list head for storing dm_io list
1654 */
1655 io->data = bio->bi_private;
1656
1657 /* tell block layer to poll for completion */
1658 bio->bi_cookie = ~BLK_QC_T_NONE;
1659
1660 io->next = NULL;
1661 } else {
1662 /*
1663 * bio recursed due to split, reuse original poll list,
1664 * and save bio->bi_private too.
1665 */
1666 io->data = (*head)->data;
1667 io->next = *head;
1668 }
1669
1670 *head = io;
1671}
1672
1673/*
1674 * Select the correct strategy for processing a non-flush bio.
1675 */
1676static blk_status_t __split_and_process_bio(struct clone_info *ci)
1677{
1678 struct bio *clone;
1679 struct dm_target *ti;
1680 unsigned len;
1681
1682 ti = dm_table_find_target(ci->map, ci->sector);
1683 if (unlikely(!ti))
1684 return BLK_STS_IOERR;
1685
1686 if (unlikely((ci->bio->bi_opf & REQ_NOWAIT) != 0) &&
1687 unlikely(!dm_target_supports_nowait(ti->type)))
1688 return BLK_STS_NOTSUPP;
1689
1690 if (unlikely(ci->is_abnormal_io))
1691 return __process_abnormal_io(ci, ti);
1692
1693 /*
1694 * Only support bio polling for normal IO, and the target io is
1695 * exactly inside the dm_io instance (verified in dm_poll_dm_io)
1696 */
1697 ci->submit_as_polled = !!(ci->bio->bi_opf & REQ_POLLED);
1698
1699 len = min_t(sector_t, max_io_len(ti, ci->sector), ci->sector_count);
1700 setup_split_accounting(ci, len);
1701 clone = alloc_tio(ci, ti, 0, &len, GFP_NOIO);
1702 __map_bio(clone);
1703
1704 ci->sector += len;
1705 ci->sector_count -= len;
1706
1707 return BLK_STS_OK;
1708}
1709
1710static void init_clone_info(struct clone_info *ci, struct mapped_device *md,
1711 struct dm_table *map, struct bio *bio, bool is_abnormal)
1712{
1713 ci->map = map;
1714 ci->io = alloc_io(md, bio);
1715 ci->bio = bio;
1716 ci->is_abnormal_io = is_abnormal;
1717 ci->submit_as_polled = false;
1718 ci->sector = bio->bi_iter.bi_sector;
1719 ci->sector_count = bio_sectors(bio);
1720
1721 /* Shouldn't happen but sector_count was being set to 0 so... */
1722 if (static_branch_unlikely(&zoned_enabled) &&
1723 WARN_ON_ONCE(op_is_zone_mgmt(bio_op(bio)) && ci->sector_count))
1724 ci->sector_count = 0;
1725}
1726
1727/*
1728 * Entry point to split a bio into clones and submit them to the targets.
1729 */
1730static void dm_split_and_process_bio(struct mapped_device *md,
1731 struct dm_table *map, struct bio *bio)
1732{
1733 struct clone_info ci;
1734 struct dm_io *io;
1735 blk_status_t error = BLK_STS_OK;
1736 bool is_abnormal;
1737
1738 is_abnormal = is_abnormal_io(bio);
1739 if (unlikely(is_abnormal)) {
1740 /*
1741 * Use bio_split_to_limits() for abnormal IO (e.g. discard, etc)
1742 * otherwise associated queue_limits won't be imposed.
1743 */
1744 bio = bio_split_to_limits(bio);
1745 if (!bio)
1746 return;
1747 }
1748
1749 init_clone_info(&ci, md, map, bio, is_abnormal);
1750 io = ci.io;
1751
1752 if (bio->bi_opf & REQ_PREFLUSH) {
1753 __send_empty_flush(&ci);
1754 /* dm_io_complete submits any data associated with flush */
1755 goto out;
1756 }
1757
1758 error = __split_and_process_bio(&ci);
1759 if (error || !ci.sector_count)
1760 goto out;
1761 /*
1762 * Remainder must be passed to submit_bio_noacct() so it gets handled
1763 * *after* bios already submitted have been completely processed.
1764 */
1765 bio_trim(bio, io->sectors, ci.sector_count);
1766 trace_block_split(bio, bio->bi_iter.bi_sector);
1767 bio_inc_remaining(bio);
1768 submit_bio_noacct(bio);
1769out:
1770 /*
1771 * Drop the extra reference count for non-POLLED bio, and hold one
1772 * reference for POLLED bio, which will be released in dm_poll_bio
1773 *
1774 * Add every dm_io instance into the dm_io list head which is stored
1775 * in bio->bi_private, so that dm_poll_bio can poll them all.
1776 */
1777 if (error || !ci.submit_as_polled) {
1778 /*
1779 * In case of submission failure, the extra reference for
1780 * submitting io isn't consumed yet
1781 */
1782 if (error)
1783 atomic_dec(&io->io_count);
1784 dm_io_dec_pending(io, error);
1785 } else
1786 dm_queue_poll_io(bio, io);
1787}
1788
1789static void dm_submit_bio(struct bio *bio)
1790{
1791 struct mapped_device *md = bio->bi_bdev->bd_disk->private_data;
1792 int srcu_idx;
1793 struct dm_table *map;
1794 blk_opf_t bio_opf = bio->bi_opf;
1795
1796 map = dm_get_live_table_bio(md, &srcu_idx, bio_opf);
1797
1798 /* If suspended, or map not yet available, queue this IO for later */
1799 if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) ||
1800 unlikely(!map)) {
1801 if (bio->bi_opf & REQ_NOWAIT)
1802 bio_wouldblock_error(bio);
1803 else if (bio->bi_opf & REQ_RAHEAD)
1804 bio_io_error(bio);
1805 else
1806 queue_io(md, bio);
1807 goto out;
1808 }
1809
1810 dm_split_and_process_bio(md, map, bio);
1811out:
1812 dm_put_live_table_bio(md, srcu_idx, bio_opf);
1813}
1814
1815static bool dm_poll_dm_io(struct dm_io *io, struct io_comp_batch *iob,
1816 unsigned int flags)
1817{
1818 WARN_ON_ONCE(!dm_tio_is_normal(&io->tio));
1819
1820 /* don't poll if the mapped io is done */
1821 if (atomic_read(&io->io_count) > 1)
1822 bio_poll(&io->tio.clone, iob, flags);
1823
1824 /* bio_poll holds the last reference */
1825 return atomic_read(&io->io_count) == 1;
1826}
1827
1828static int dm_poll_bio(struct bio *bio, struct io_comp_batch *iob,
1829 unsigned int flags)
1830{
1831 struct dm_io **head = dm_poll_list_head(bio);
1832 struct dm_io *list = *head;
1833 struct dm_io *tmp = NULL;
1834 struct dm_io *curr, *next;
1835
1836 /* Only poll normal bio which was marked as REQ_DM_POLL_LIST */
1837 if (!(bio->bi_opf & REQ_DM_POLL_LIST))
1838 return 0;
1839
1840 WARN_ON_ONCE(!list);
1841
1842 /*
1843 * Restore .bi_private before possibly completing dm_io.
1844 *
1845 * bio_poll() is only possible once @bio has been completely
1846 * submitted via submit_bio_noacct()'s depth-first submission.
1847 * So there is no dm_queue_poll_io() race associated with
1848 * clearing REQ_DM_POLL_LIST here.
1849 */
1850 bio->bi_opf &= ~REQ_DM_POLL_LIST;
1851 bio->bi_private = list->data;
1852
1853 for (curr = list, next = curr->next; curr; curr = next, next =
1854 curr ? curr->next : NULL) {
1855 if (dm_poll_dm_io(curr, iob, flags)) {
1856 /*
1857 * clone_endio() has already occurred, so no
1858 * error handling is needed here.
1859 */
1860 __dm_io_dec_pending(curr);
1861 } else {
1862 curr->next = tmp;
1863 tmp = curr;
1864 }
1865 }
1866
1867 /* Not done? */
1868 if (tmp) {
1869 bio->bi_opf |= REQ_DM_POLL_LIST;
1870 /* Reset bio->bi_private to dm_io list head */
1871 *head = tmp;
1872 return 0;
1873 }
1874 return 1;
1875}
1876
1877/*-----------------------------------------------------------------
1878 * An IDR is used to keep track of allocated minor numbers.
1879 *---------------------------------------------------------------*/
1880static void free_minor(int minor)
1881{
1882 spin_lock(&_minor_lock);
1883 idr_remove(&_minor_idr, minor);
1884 spin_unlock(&_minor_lock);
1885}
1886
1887/*
1888 * See if the device with a specific minor # is free.
1889 */
1890static int specific_minor(int minor)
1891{
1892 int r;
1893
1894 if (minor >= (1 << MINORBITS))
1895 return -EINVAL;
1896
1897 idr_preload(GFP_KERNEL);
1898 spin_lock(&_minor_lock);
1899
1900 r = idr_alloc(&_minor_idr, MINOR_ALLOCED, minor, minor + 1, GFP_NOWAIT);
1901
1902 spin_unlock(&_minor_lock);
1903 idr_preload_end();
1904 if (r < 0)
1905 return r == -ENOSPC ? -EBUSY : r;
1906 return 0;
1907}
1908
1909static int next_free_minor(int *minor)
1910{
1911 int r;
1912
1913 idr_preload(GFP_KERNEL);
1914 spin_lock(&_minor_lock);
1915
1916 r = idr_alloc(&_minor_idr, MINOR_ALLOCED, 0, 1 << MINORBITS, GFP_NOWAIT);
1917
1918 spin_unlock(&_minor_lock);
1919 idr_preload_end();
1920 if (r < 0)
1921 return r;
1922 *minor = r;
1923 return 0;
1924}
1925
1926static const struct block_device_operations dm_blk_dops;
1927static const struct block_device_operations dm_rq_blk_dops;
1928static const struct dax_operations dm_dax_ops;
1929
1930static void dm_wq_work(struct work_struct *work);
1931
1932#ifdef CONFIG_BLK_INLINE_ENCRYPTION
1933static void dm_queue_destroy_crypto_profile(struct request_queue *q)
1934{
1935 dm_destroy_crypto_profile(q->crypto_profile);
1936}
1937
1938#else /* CONFIG_BLK_INLINE_ENCRYPTION */
1939
1940static inline void dm_queue_destroy_crypto_profile(struct request_queue *q)
1941{
1942}
1943#endif /* !CONFIG_BLK_INLINE_ENCRYPTION */
1944
1945static void cleanup_mapped_device(struct mapped_device *md)
1946{
1947 if (md->wq)
1948 destroy_workqueue(md->wq);
1949 dm_free_md_mempools(md->mempools);
1950
1951 if (md->dax_dev) {
1952 dax_remove_host(md->disk);
1953 kill_dax(md->dax_dev);
1954 put_dax(md->dax_dev);
1955 md->dax_dev = NULL;
1956 }
1957
1958 dm_cleanup_zoned_dev(md);
1959 if (md->disk) {
1960 spin_lock(&_minor_lock);
1961 md->disk->private_data = NULL;
1962 spin_unlock(&_minor_lock);
1963 if (dm_get_md_type(md) != DM_TYPE_NONE) {
1964 struct table_device *td;
1965
1966 dm_sysfs_exit(md);
1967 list_for_each_entry(td, &md->table_devices, list) {
1968 bd_unlink_disk_holder(td->dm_dev.bdev,
1969 md->disk);
1970 }
1971
1972 /*
1973 * Hold lock to make sure del_gendisk() won't concurrent
1974 * with open/close_table_device().
1975 */
1976 mutex_lock(&md->table_devices_lock);
1977 del_gendisk(md->disk);
1978 mutex_unlock(&md->table_devices_lock);
1979 }
1980 dm_queue_destroy_crypto_profile(md->queue);
1981 put_disk(md->disk);
1982 }
1983
1984 if (md->pending_io) {
1985 free_percpu(md->pending_io);
1986 md->pending_io = NULL;
1987 }
1988
1989 cleanup_srcu_struct(&md->io_barrier);
1990
1991 mutex_destroy(&md->suspend_lock);
1992 mutex_destroy(&md->type_lock);
1993 mutex_destroy(&md->table_devices_lock);
1994 mutex_destroy(&md->swap_bios_lock);
1995
1996 dm_mq_cleanup_mapped_device(md);
1997}
1998
1999/*
2000 * Allocate and initialise a blank device with a given minor.
2001 */
2002static struct mapped_device *alloc_dev(int minor)
2003{
2004 int r, numa_node_id = dm_get_numa_node();
2005 struct mapped_device *md;
2006 void *old_md;
2007
2008 md = kvzalloc_node(sizeof(*md), GFP_KERNEL, numa_node_id);
2009 if (!md) {
2010 DMERR("unable to allocate device, out of memory.");
2011 return NULL;
2012 }
2013
2014 if (!try_module_get(THIS_MODULE))
2015 goto bad_module_get;
2016
2017 /* get a minor number for the dev */
2018 if (minor == DM_ANY_MINOR)
2019 r = next_free_minor(&minor);
2020 else
2021 r = specific_minor(minor);
2022 if (r < 0)
2023 goto bad_minor;
2024
2025 r = init_srcu_struct(&md->io_barrier);
2026 if (r < 0)
2027 goto bad_io_barrier;
2028
2029 md->numa_node_id = numa_node_id;
2030 md->init_tio_pdu = false;
2031 md->type = DM_TYPE_NONE;
2032 mutex_init(&md->suspend_lock);
2033 mutex_init(&md->type_lock);
2034 mutex_init(&md->table_devices_lock);
2035 spin_lock_init(&md->deferred_lock);
2036 atomic_set(&md->holders, 1);
2037 atomic_set(&md->open_count, 0);
2038 atomic_set(&md->event_nr, 0);
2039 atomic_set(&md->uevent_seq, 0);
2040 INIT_LIST_HEAD(&md->uevent_list);
2041 INIT_LIST_HEAD(&md->table_devices);
2042 spin_lock_init(&md->uevent_lock);
2043
2044 /*
2045 * default to bio-based until DM table is loaded and md->type
2046 * established. If request-based table is loaded: blk-mq will
2047 * override accordingly.
2048 */
2049 md->disk = blk_alloc_disk(md->numa_node_id);
2050 if (!md->disk)
2051 goto bad;
2052 md->queue = md->disk->queue;
2053
2054 init_waitqueue_head(&md->wait);
2055 INIT_WORK(&md->work, dm_wq_work);
2056 INIT_WORK(&md->requeue_work, dm_wq_requeue_work);
2057 init_waitqueue_head(&md->eventq);
2058 init_completion(&md->kobj_holder.completion);
2059
2060 md->requeue_list = NULL;
2061 md->swap_bios = get_swap_bios();
2062 sema_init(&md->swap_bios_semaphore, md->swap_bios);
2063 mutex_init(&md->swap_bios_lock);
2064
2065 md->disk->major = _major;
2066 md->disk->first_minor = minor;
2067 md->disk->minors = 1;
2068 md->disk->flags |= GENHD_FL_NO_PART;
2069 md->disk->fops = &dm_blk_dops;
2070 md->disk->private_data = md;
2071 sprintf(md->disk->disk_name, "dm-%d", minor);
2072
2073 if (IS_ENABLED(CONFIG_FS_DAX)) {
2074 md->dax_dev = alloc_dax(md, &dm_dax_ops);
2075 if (IS_ERR(md->dax_dev)) {
2076 md->dax_dev = NULL;
2077 goto bad;
2078 }
2079 set_dax_nocache(md->dax_dev);
2080 set_dax_nomc(md->dax_dev);
2081 if (dax_add_host(md->dax_dev, md->disk))
2082 goto bad;
2083 }
2084
2085 format_dev_t(md->name, MKDEV(_major, minor));
2086
2087 md->wq = alloc_workqueue("kdmflush/%s", WQ_MEM_RECLAIM, 0, md->name);
2088 if (!md->wq)
2089 goto bad;
2090
2091 md->pending_io = alloc_percpu(unsigned long);
2092 if (!md->pending_io)
2093 goto bad;
2094
2095 dm_stats_init(&md->stats);
2096
2097 /* Populate the mapping, nobody knows we exist yet */
2098 spin_lock(&_minor_lock);
2099 old_md = idr_replace(&_minor_idr, md, minor);
2100 spin_unlock(&_minor_lock);
2101
2102 BUG_ON(old_md != MINOR_ALLOCED);
2103
2104 return md;
2105
2106bad:
2107 cleanup_mapped_device(md);
2108bad_io_barrier:
2109 free_minor(minor);
2110bad_minor:
2111 module_put(THIS_MODULE);
2112bad_module_get:
2113 kvfree(md);
2114 return NULL;
2115}
2116
2117static void unlock_fs(struct mapped_device *md);
2118
2119static void free_dev(struct mapped_device *md)
2120{
2121 int minor = MINOR(disk_devt(md->disk));
2122
2123 unlock_fs(md);
2124
2125 cleanup_mapped_device(md);
2126
2127 WARN_ON_ONCE(!list_empty(&md->table_devices));
2128 dm_stats_cleanup(&md->stats);
2129 free_minor(minor);
2130
2131 module_put(THIS_MODULE);
2132 kvfree(md);
2133}
2134
2135/*
2136 * Bind a table to the device.
2137 */
2138static void event_callback(void *context)
2139{
2140 unsigned long flags;
2141 LIST_HEAD(uevents);
2142 struct mapped_device *md = (struct mapped_device *) context;
2143
2144 spin_lock_irqsave(&md->uevent_lock, flags);
2145 list_splice_init(&md->uevent_list, &uevents);
2146 spin_unlock_irqrestore(&md->uevent_lock, flags);
2147
2148 dm_send_uevents(&uevents, &disk_to_dev(md->disk)->kobj);
2149
2150 atomic_inc(&md->event_nr);
2151 wake_up(&md->eventq);
2152 dm_issue_global_event();
2153}
2154
2155/*
2156 * Returns old map, which caller must destroy.
2157 */
2158static struct dm_table *__bind(struct mapped_device *md, struct dm_table *t,
2159 struct queue_limits *limits)
2160{
2161 struct dm_table *old_map;
2162 sector_t size;
2163 int ret;
2164
2165 lockdep_assert_held(&md->suspend_lock);
2166
2167 size = dm_table_get_size(t);
2168
2169 /*
2170 * Wipe any geometry if the size of the table changed.
2171 */
2172 if (size != dm_get_size(md))
2173 memset(&md->geometry, 0, sizeof(md->geometry));
2174
2175 if (!get_capacity(md->disk))
2176 set_capacity(md->disk, size);
2177 else
2178 set_capacity_and_notify(md->disk, size);
2179
2180 dm_table_event_callback(t, event_callback, md);
2181
2182 if (dm_table_request_based(t)) {
2183 /*
2184 * Leverage the fact that request-based DM targets are
2185 * immutable singletons - used to optimize dm_mq_queue_rq.
2186 */
2187 md->immutable_target = dm_table_get_immutable_target(t);
2188
2189 /*
2190 * There is no need to reload with request-based dm because the
2191 * size of front_pad doesn't change.
2192 *
2193 * Note for future: If you are to reload bioset, prep-ed
2194 * requests in the queue may refer to bio from the old bioset,
2195 * so you must walk through the queue to unprep.
2196 */
2197 if (!md->mempools) {
2198 md->mempools = t->mempools;
2199 t->mempools = NULL;
2200 }
2201 } else {
2202 /*
2203 * The md may already have mempools that need changing.
2204 * If so, reload bioset because front_pad may have changed
2205 * because a different table was loaded.
2206 */
2207 dm_free_md_mempools(md->mempools);
2208 md->mempools = t->mempools;
2209 t->mempools = NULL;
2210 }
2211
2212 ret = dm_table_set_restrictions(t, md->queue, limits);
2213 if (ret) {
2214 old_map = ERR_PTR(ret);
2215 goto out;
2216 }
2217
2218 old_map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2219 rcu_assign_pointer(md->map, (void *)t);
2220 md->immutable_target_type = dm_table_get_immutable_target_type(t);
2221
2222 if (old_map)
2223 dm_sync_table(md);
2224out:
2225 return old_map;
2226}
2227
2228/*
2229 * Returns unbound table for the caller to free.
2230 */
2231static struct dm_table *__unbind(struct mapped_device *md)
2232{
2233 struct dm_table *map = rcu_dereference_protected(md->map, 1);
2234
2235 if (!map)
2236 return NULL;
2237
2238 dm_table_event_callback(map, NULL, NULL);
2239 RCU_INIT_POINTER(md->map, NULL);
2240 dm_sync_table(md);
2241
2242 return map;
2243}
2244
2245/*
2246 * Constructor for a new device.
2247 */
2248int dm_create(int minor, struct mapped_device **result)
2249{
2250 struct mapped_device *md;
2251
2252 md = alloc_dev(minor);
2253 if (!md)
2254 return -ENXIO;
2255
2256 dm_ima_reset_data(md);
2257
2258 *result = md;
2259 return 0;
2260}
2261
2262/*
2263 * Functions to manage md->type.
2264 * All are required to hold md->type_lock.
2265 */
2266void dm_lock_md_type(struct mapped_device *md)
2267{
2268 mutex_lock(&md->type_lock);
2269}
2270
2271void dm_unlock_md_type(struct mapped_device *md)
2272{
2273 mutex_unlock(&md->type_lock);
2274}
2275
2276void dm_set_md_type(struct mapped_device *md, enum dm_queue_mode type)
2277{
2278 BUG_ON(!mutex_is_locked(&md->type_lock));
2279 md->type = type;
2280}
2281
2282enum dm_queue_mode dm_get_md_type(struct mapped_device *md)
2283{
2284 return md->type;
2285}
2286
2287struct target_type *dm_get_immutable_target_type(struct mapped_device *md)
2288{
2289 return md->immutable_target_type;
2290}
2291
2292/*
2293 * The queue_limits are only valid as long as you have a reference
2294 * count on 'md'.
2295 */
2296struct queue_limits *dm_get_queue_limits(struct mapped_device *md)
2297{
2298 BUG_ON(!atomic_read(&md->holders));
2299 return &md->queue->limits;
2300}
2301EXPORT_SYMBOL_GPL(dm_get_queue_limits);
2302
2303/*
2304 * Setup the DM device's queue based on md's type
2305 */
2306int dm_setup_md_queue(struct mapped_device *md, struct dm_table *t)
2307{
2308 enum dm_queue_mode type = dm_table_get_type(t);
2309 struct queue_limits limits;
2310 struct table_device *td;
2311 int r;
2312
2313 switch (type) {
2314 case DM_TYPE_REQUEST_BASED:
2315 md->disk->fops = &dm_rq_blk_dops;
2316 r = dm_mq_init_request_queue(md, t);
2317 if (r) {
2318 DMERR("Cannot initialize queue for request-based dm mapped device");
2319 return r;
2320 }
2321 break;
2322 case DM_TYPE_BIO_BASED:
2323 case DM_TYPE_DAX_BIO_BASED:
2324 break;
2325 case DM_TYPE_NONE:
2326 WARN_ON_ONCE(true);
2327 break;
2328 }
2329
2330 r = dm_calculate_queue_limits(t, &limits);
2331 if (r) {
2332 DMERR("Cannot calculate initial queue limits");
2333 return r;
2334 }
2335 r = dm_table_set_restrictions(t, md->queue, &limits);
2336 if (r)
2337 return r;
2338
2339 /*
2340 * Hold lock to make sure add_disk() and del_gendisk() won't concurrent
2341 * with open_table_device() and close_table_device().
2342 */
2343 mutex_lock(&md->table_devices_lock);
2344 r = add_disk(md->disk);
2345 mutex_unlock(&md->table_devices_lock);
2346 if (r)
2347 return r;
2348
2349 /*
2350 * Register the holder relationship for devices added before the disk
2351 * was live.
2352 */
2353 list_for_each_entry(td, &md->table_devices, list) {
2354 r = bd_link_disk_holder(td->dm_dev.bdev, md->disk);
2355 if (r)
2356 goto out_undo_holders;
2357 }
2358
2359 r = dm_sysfs_init(md);
2360 if (r)
2361 goto out_undo_holders;
2362
2363 md->type = type;
2364 return 0;
2365
2366out_undo_holders:
2367 list_for_each_entry_continue_reverse(td, &md->table_devices, list)
2368 bd_unlink_disk_holder(td->dm_dev.bdev, md->disk);
2369 mutex_lock(&md->table_devices_lock);
2370 del_gendisk(md->disk);
2371 mutex_unlock(&md->table_devices_lock);
2372 return r;
2373}
2374
2375struct mapped_device *dm_get_md(dev_t dev)
2376{
2377 struct mapped_device *md;
2378 unsigned minor = MINOR(dev);
2379
2380 if (MAJOR(dev) != _major || minor >= (1 << MINORBITS))
2381 return NULL;
2382
2383 spin_lock(&_minor_lock);
2384
2385 md = idr_find(&_minor_idr, minor);
2386 if (!md || md == MINOR_ALLOCED || (MINOR(disk_devt(dm_disk(md))) != minor) ||
2387 test_bit(DMF_FREEING, &md->flags) || dm_deleting_md(md)) {
2388 md = NULL;
2389 goto out;
2390 }
2391 dm_get(md);
2392out:
2393 spin_unlock(&_minor_lock);
2394
2395 return md;
2396}
2397EXPORT_SYMBOL_GPL(dm_get_md);
2398
2399void *dm_get_mdptr(struct mapped_device *md)
2400{
2401 return md->interface_ptr;
2402}
2403
2404void dm_set_mdptr(struct mapped_device *md, void *ptr)
2405{
2406 md->interface_ptr = ptr;
2407}
2408
2409void dm_get(struct mapped_device *md)
2410{
2411 atomic_inc(&md->holders);
2412 BUG_ON(test_bit(DMF_FREEING, &md->flags));
2413}
2414
2415int dm_hold(struct mapped_device *md)
2416{
2417 spin_lock(&_minor_lock);
2418 if (test_bit(DMF_FREEING, &md->flags)) {
2419 spin_unlock(&_minor_lock);
2420 return -EBUSY;
2421 }
2422 dm_get(md);
2423 spin_unlock(&_minor_lock);
2424 return 0;
2425}
2426EXPORT_SYMBOL_GPL(dm_hold);
2427
2428const char *dm_device_name(struct mapped_device *md)
2429{
2430 return md->name;
2431}
2432EXPORT_SYMBOL_GPL(dm_device_name);
2433
2434static void __dm_destroy(struct mapped_device *md, bool wait)
2435{
2436 struct dm_table *map;
2437 int srcu_idx;
2438
2439 might_sleep();
2440
2441 spin_lock(&_minor_lock);
2442 idr_replace(&_minor_idr, MINOR_ALLOCED, MINOR(disk_devt(dm_disk(md))));
2443 set_bit(DMF_FREEING, &md->flags);
2444 spin_unlock(&_minor_lock);
2445
2446 blk_mark_disk_dead(md->disk);
2447
2448 /*
2449 * Take suspend_lock so that presuspend and postsuspend methods
2450 * do not race with internal suspend.
2451 */
2452 mutex_lock(&md->suspend_lock);
2453 map = dm_get_live_table(md, &srcu_idx);
2454 if (!dm_suspended_md(md)) {
2455 dm_table_presuspend_targets(map);
2456 set_bit(DMF_SUSPENDED, &md->flags);
2457 set_bit(DMF_POST_SUSPENDING, &md->flags);
2458 dm_table_postsuspend_targets(map);
2459 }
2460 /* dm_put_live_table must be before msleep, otherwise deadlock is possible */
2461 dm_put_live_table(md, srcu_idx);
2462 mutex_unlock(&md->suspend_lock);
2463
2464 /*
2465 * Rare, but there may be I/O requests still going to complete,
2466 * for example. Wait for all references to disappear.
2467 * No one should increment the reference count of the mapped_device,
2468 * after the mapped_device state becomes DMF_FREEING.
2469 */
2470 if (wait)
2471 while (atomic_read(&md->holders))
2472 msleep(1);
2473 else if (atomic_read(&md->holders))
2474 DMWARN("%s: Forcibly removing mapped_device still in use! (%d users)",
2475 dm_device_name(md), atomic_read(&md->holders));
2476
2477 dm_table_destroy(__unbind(md));
2478 free_dev(md);
2479}
2480
2481void dm_destroy(struct mapped_device *md)
2482{
2483 __dm_destroy(md, true);
2484}
2485
2486void dm_destroy_immediate(struct mapped_device *md)
2487{
2488 __dm_destroy(md, false);
2489}
2490
2491void dm_put(struct mapped_device *md)
2492{
2493 atomic_dec(&md->holders);
2494}
2495EXPORT_SYMBOL_GPL(dm_put);
2496
2497static bool dm_in_flight_bios(struct mapped_device *md)
2498{
2499 int cpu;
2500 unsigned long sum = 0;
2501
2502 for_each_possible_cpu(cpu)
2503 sum += *per_cpu_ptr(md->pending_io, cpu);
2504
2505 return sum != 0;
2506}
2507
2508static int dm_wait_for_bios_completion(struct mapped_device *md, unsigned int task_state)
2509{
2510 int r = 0;
2511 DEFINE_WAIT(wait);
2512
2513 while (true) {
2514 prepare_to_wait(&md->wait, &wait, task_state);
2515
2516 if (!dm_in_flight_bios(md))
2517 break;
2518
2519 if (signal_pending_state(task_state, current)) {
2520 r = -EINTR;
2521 break;
2522 }
2523
2524 io_schedule();
2525 }
2526 finish_wait(&md->wait, &wait);
2527
2528 smp_rmb();
2529
2530 return r;
2531}
2532
2533static int dm_wait_for_completion(struct mapped_device *md, unsigned int task_state)
2534{
2535 int r = 0;
2536
2537 if (!queue_is_mq(md->queue))
2538 return dm_wait_for_bios_completion(md, task_state);
2539
2540 while (true) {
2541 if (!blk_mq_queue_inflight(md->queue))
2542 break;
2543
2544 if (signal_pending_state(task_state, current)) {
2545 r = -EINTR;
2546 break;
2547 }
2548
2549 msleep(5);
2550 }
2551
2552 return r;
2553}
2554
2555/*
2556 * Process the deferred bios
2557 */
2558static void dm_wq_work(struct work_struct *work)
2559{
2560 struct mapped_device *md = container_of(work, struct mapped_device, work);
2561 struct bio *bio;
2562
2563 while (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
2564 spin_lock_irq(&md->deferred_lock);
2565 bio = bio_list_pop(&md->deferred);
2566 spin_unlock_irq(&md->deferred_lock);
2567
2568 if (!bio)
2569 break;
2570
2571 submit_bio_noacct(bio);
2572 }
2573}
2574
2575static void dm_queue_flush(struct mapped_device *md)
2576{
2577 clear_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2578 smp_mb__after_atomic();
2579 queue_work(md->wq, &md->work);
2580}
2581
2582/*
2583 * Swap in a new table, returning the old one for the caller to destroy.
2584 */
2585struct dm_table *dm_swap_table(struct mapped_device *md, struct dm_table *table)
2586{
2587 struct dm_table *live_map = NULL, *map = ERR_PTR(-EINVAL);
2588 struct queue_limits limits;
2589 int r;
2590
2591 mutex_lock(&md->suspend_lock);
2592
2593 /* device must be suspended */
2594 if (!dm_suspended_md(md))
2595 goto out;
2596
2597 /*
2598 * If the new table has no data devices, retain the existing limits.
2599 * This helps multipath with queue_if_no_path if all paths disappear,
2600 * then new I/O is queued based on these limits, and then some paths
2601 * reappear.
2602 */
2603 if (dm_table_has_no_data_devices(table)) {
2604 live_map = dm_get_live_table_fast(md);
2605 if (live_map)
2606 limits = md->queue->limits;
2607 dm_put_live_table_fast(md);
2608 }
2609
2610 if (!live_map) {
2611 r = dm_calculate_queue_limits(table, &limits);
2612 if (r) {
2613 map = ERR_PTR(r);
2614 goto out;
2615 }
2616 }
2617
2618 map = __bind(md, table, &limits);
2619 dm_issue_global_event();
2620
2621out:
2622 mutex_unlock(&md->suspend_lock);
2623 return map;
2624}
2625
2626/*
2627 * Functions to lock and unlock any filesystem running on the
2628 * device.
2629 */
2630static int lock_fs(struct mapped_device *md)
2631{
2632 int r;
2633
2634 WARN_ON(test_bit(DMF_FROZEN, &md->flags));
2635
2636 r = freeze_bdev(md->disk->part0);
2637 if (!r)
2638 set_bit(DMF_FROZEN, &md->flags);
2639 return r;
2640}
2641
2642static void unlock_fs(struct mapped_device *md)
2643{
2644 if (!test_bit(DMF_FROZEN, &md->flags))
2645 return;
2646 thaw_bdev(md->disk->part0);
2647 clear_bit(DMF_FROZEN, &md->flags);
2648}
2649
2650/*
2651 * @suspend_flags: DM_SUSPEND_LOCKFS_FLAG and/or DM_SUSPEND_NOFLUSH_FLAG
2652 * @task_state: e.g. TASK_INTERRUPTIBLE or TASK_UNINTERRUPTIBLE
2653 * @dmf_suspended_flag: DMF_SUSPENDED or DMF_SUSPENDED_INTERNALLY
2654 *
2655 * If __dm_suspend returns 0, the device is completely quiescent
2656 * now. There is no request-processing activity. All new requests
2657 * are being added to md->deferred list.
2658 */
2659static int __dm_suspend(struct mapped_device *md, struct dm_table *map,
2660 unsigned suspend_flags, unsigned int task_state,
2661 int dmf_suspended_flag)
2662{
2663 bool do_lockfs = suspend_flags & DM_SUSPEND_LOCKFS_FLAG;
2664 bool noflush = suspend_flags & DM_SUSPEND_NOFLUSH_FLAG;
2665 int r;
2666
2667 lockdep_assert_held(&md->suspend_lock);
2668
2669 /*
2670 * DMF_NOFLUSH_SUSPENDING must be set before presuspend.
2671 * This flag is cleared before dm_suspend returns.
2672 */
2673 if (noflush)
2674 set_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
2675 else
2676 DMDEBUG("%s: suspending with flush", dm_device_name(md));
2677
2678 /*
2679 * This gets reverted if there's an error later and the targets
2680 * provide the .presuspend_undo hook.
2681 */
2682 dm_table_presuspend_targets(map);
2683
2684 /*
2685 * Flush I/O to the device.
2686 * Any I/O submitted after lock_fs() may not be flushed.
2687 * noflush takes precedence over do_lockfs.
2688 * (lock_fs() flushes I/Os and waits for them to complete.)
2689 */
2690 if (!noflush && do_lockfs) {
2691 r = lock_fs(md);
2692 if (r) {
2693 dm_table_presuspend_undo_targets(map);
2694 return r;
2695 }
2696 }
2697
2698 /*
2699 * Here we must make sure that no processes are submitting requests
2700 * to target drivers i.e. no one may be executing
2701 * dm_split_and_process_bio from dm_submit_bio.
2702 *
2703 * To get all processes out of dm_split_and_process_bio in dm_submit_bio,
2704 * we take the write lock. To prevent any process from reentering
2705 * dm_split_and_process_bio from dm_submit_bio and quiesce the thread
2706 * (dm_wq_work), we set DMF_BLOCK_IO_FOR_SUSPEND and call
2707 * flush_workqueue(md->wq).
2708 */
2709 set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2710 if (map)
2711 synchronize_srcu(&md->io_barrier);
2712
2713 /*
2714 * Stop md->queue before flushing md->wq in case request-based
2715 * dm defers requests to md->wq from md->queue.
2716 */
2717 if (dm_request_based(md))
2718 dm_stop_queue(md->queue);
2719
2720 flush_workqueue(md->wq);
2721
2722 /*
2723 * At this point no more requests are entering target request routines.
2724 * We call dm_wait_for_completion to wait for all existing requests
2725 * to finish.
2726 */
2727 r = dm_wait_for_completion(md, task_state);
2728 if (!r)
2729 set_bit(dmf_suspended_flag, &md->flags);
2730
2731 if (noflush)
2732 clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
2733 if (map)
2734 synchronize_srcu(&md->io_barrier);
2735
2736 /* were we interrupted ? */
2737 if (r < 0) {
2738 dm_queue_flush(md);
2739
2740 if (dm_request_based(md))
2741 dm_start_queue(md->queue);
2742
2743 unlock_fs(md);
2744 dm_table_presuspend_undo_targets(map);
2745 /* pushback list is already flushed, so skip flush */
2746 }
2747
2748 return r;
2749}
2750
2751/*
2752 * We need to be able to change a mapping table under a mounted
2753 * filesystem. For example we might want to move some data in
2754 * the background. Before the table can be swapped with
2755 * dm_bind_table, dm_suspend must be called to flush any in
2756 * flight bios and ensure that any further io gets deferred.
2757 */
2758/*
2759 * Suspend mechanism in request-based dm.
2760 *
2761 * 1. Flush all I/Os by lock_fs() if needed.
2762 * 2. Stop dispatching any I/O by stopping the request_queue.
2763 * 3. Wait for all in-flight I/Os to be completed or requeued.
2764 *
2765 * To abort suspend, start the request_queue.
2766 */
2767int dm_suspend(struct mapped_device *md, unsigned suspend_flags)
2768{
2769 struct dm_table *map = NULL;
2770 int r = 0;
2771
2772retry:
2773 mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING);
2774
2775 if (dm_suspended_md(md)) {
2776 r = -EINVAL;
2777 goto out_unlock;
2778 }
2779
2780 if (dm_suspended_internally_md(md)) {
2781 /* already internally suspended, wait for internal resume */
2782 mutex_unlock(&md->suspend_lock);
2783 r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE);
2784 if (r)
2785 return r;
2786 goto retry;
2787 }
2788
2789 map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2790
2791 r = __dm_suspend(md, map, suspend_flags, TASK_INTERRUPTIBLE, DMF_SUSPENDED);
2792 if (r)
2793 goto out_unlock;
2794
2795 set_bit(DMF_POST_SUSPENDING, &md->flags);
2796 dm_table_postsuspend_targets(map);
2797 clear_bit(DMF_POST_SUSPENDING, &md->flags);
2798
2799out_unlock:
2800 mutex_unlock(&md->suspend_lock);
2801 return r;
2802}
2803
2804static int __dm_resume(struct mapped_device *md, struct dm_table *map)
2805{
2806 if (map) {
2807 int r = dm_table_resume_targets(map);
2808 if (r)
2809 return r;
2810 }
2811
2812 dm_queue_flush(md);
2813
2814 /*
2815 * Flushing deferred I/Os must be done after targets are resumed
2816 * so that mapping of targets can work correctly.
2817 * Request-based dm is queueing the deferred I/Os in its request_queue.
2818 */
2819 if (dm_request_based(md))
2820 dm_start_queue(md->queue);
2821
2822 unlock_fs(md);
2823
2824 return 0;
2825}
2826
2827int dm_resume(struct mapped_device *md)
2828{
2829 int r;
2830 struct dm_table *map = NULL;
2831
2832retry:
2833 r = -EINVAL;
2834 mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING);
2835
2836 if (!dm_suspended_md(md))
2837 goto out;
2838
2839 if (dm_suspended_internally_md(md)) {
2840 /* already internally suspended, wait for internal resume */
2841 mutex_unlock(&md->suspend_lock);
2842 r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE);
2843 if (r)
2844 return r;
2845 goto retry;
2846 }
2847
2848 map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2849 if (!map || !dm_table_get_size(map))
2850 goto out;
2851
2852 r = __dm_resume(md, map);
2853 if (r)
2854 goto out;
2855
2856 clear_bit(DMF_SUSPENDED, &md->flags);
2857out:
2858 mutex_unlock(&md->suspend_lock);
2859
2860 return r;
2861}
2862
2863/*
2864 * Internal suspend/resume works like userspace-driven suspend. It waits
2865 * until all bios finish and prevents issuing new bios to the target drivers.
2866 * It may be used only from the kernel.
2867 */
2868
2869static void __dm_internal_suspend(struct mapped_device *md, unsigned suspend_flags)
2870{
2871 struct dm_table *map = NULL;
2872
2873 lockdep_assert_held(&md->suspend_lock);
2874
2875 if (md->internal_suspend_count++)
2876 return; /* nested internal suspend */
2877
2878 if (dm_suspended_md(md)) {
2879 set_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
2880 return; /* nest suspend */
2881 }
2882
2883 map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2884
2885 /*
2886 * Using TASK_UNINTERRUPTIBLE because only NOFLUSH internal suspend is
2887 * supported. Properly supporting a TASK_INTERRUPTIBLE internal suspend
2888 * would require changing .presuspend to return an error -- avoid this
2889 * until there is a need for more elaborate variants of internal suspend.
2890 */
2891 (void) __dm_suspend(md, map, suspend_flags, TASK_UNINTERRUPTIBLE,
2892 DMF_SUSPENDED_INTERNALLY);
2893
2894 set_bit(DMF_POST_SUSPENDING, &md->flags);
2895 dm_table_postsuspend_targets(map);
2896 clear_bit(DMF_POST_SUSPENDING, &md->flags);
2897}
2898
2899static void __dm_internal_resume(struct mapped_device *md)
2900{
2901 BUG_ON(!md->internal_suspend_count);
2902
2903 if (--md->internal_suspend_count)
2904 return; /* resume from nested internal suspend */
2905
2906 if (dm_suspended_md(md))
2907 goto done; /* resume from nested suspend */
2908
2909 /*
2910 * NOTE: existing callers don't need to call dm_table_resume_targets
2911 * (which may fail -- so best to avoid it for now by passing NULL map)
2912 */
2913 (void) __dm_resume(md, NULL);
2914
2915done:
2916 clear_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
2917 smp_mb__after_atomic();
2918 wake_up_bit(&md->flags, DMF_SUSPENDED_INTERNALLY);
2919}
2920
2921void dm_internal_suspend_noflush(struct mapped_device *md)
2922{
2923 mutex_lock(&md->suspend_lock);
2924 __dm_internal_suspend(md, DM_SUSPEND_NOFLUSH_FLAG);
2925 mutex_unlock(&md->suspend_lock);
2926}
2927EXPORT_SYMBOL_GPL(dm_internal_suspend_noflush);
2928
2929void dm_internal_resume(struct mapped_device *md)
2930{
2931 mutex_lock(&md->suspend_lock);
2932 __dm_internal_resume(md);
2933 mutex_unlock(&md->suspend_lock);
2934}
2935EXPORT_SYMBOL_GPL(dm_internal_resume);
2936
2937/*
2938 * Fast variants of internal suspend/resume hold md->suspend_lock,
2939 * which prevents interaction with userspace-driven suspend.
2940 */
2941
2942void dm_internal_suspend_fast(struct mapped_device *md)
2943{
2944 mutex_lock(&md->suspend_lock);
2945 if (dm_suspended_md(md) || dm_suspended_internally_md(md))
2946 return;
2947
2948 set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2949 synchronize_srcu(&md->io_barrier);
2950 flush_workqueue(md->wq);
2951 dm_wait_for_completion(md, TASK_UNINTERRUPTIBLE);
2952}
2953EXPORT_SYMBOL_GPL(dm_internal_suspend_fast);
2954
2955void dm_internal_resume_fast(struct mapped_device *md)
2956{
2957 if (dm_suspended_md(md) || dm_suspended_internally_md(md))
2958 goto done;
2959
2960 dm_queue_flush(md);
2961
2962done:
2963 mutex_unlock(&md->suspend_lock);
2964}
2965EXPORT_SYMBOL_GPL(dm_internal_resume_fast);
2966
2967/*-----------------------------------------------------------------
2968 * Event notification.
2969 *---------------------------------------------------------------*/
2970int dm_kobject_uevent(struct mapped_device *md, enum kobject_action action,
2971 unsigned cookie)
2972{
2973 int r;
2974 unsigned noio_flag;
2975 char udev_cookie[DM_COOKIE_LENGTH];
2976 char *envp[] = { udev_cookie, NULL };
2977
2978 noio_flag = memalloc_noio_save();
2979
2980 if (!cookie)
2981 r = kobject_uevent(&disk_to_dev(md->disk)->kobj, action);
2982 else {
2983 snprintf(udev_cookie, DM_COOKIE_LENGTH, "%s=%u",
2984 DM_COOKIE_ENV_VAR_NAME, cookie);
2985 r = kobject_uevent_env(&disk_to_dev(md->disk)->kobj,
2986 action, envp);
2987 }
2988
2989 memalloc_noio_restore(noio_flag);
2990
2991 return r;
2992}
2993
2994uint32_t dm_next_uevent_seq(struct mapped_device *md)
2995{
2996 return atomic_add_return(1, &md->uevent_seq);
2997}
2998
2999uint32_t dm_get_event_nr(struct mapped_device *md)
3000{
3001 return atomic_read(&md->event_nr);
3002}
3003
3004int dm_wait_event(struct mapped_device *md, int event_nr)
3005{
3006 return wait_event_interruptible(md->eventq,
3007 (event_nr != atomic_read(&md->event_nr)));
3008}
3009
3010void dm_uevent_add(struct mapped_device *md, struct list_head *elist)
3011{
3012 unsigned long flags;
3013
3014 spin_lock_irqsave(&md->uevent_lock, flags);
3015 list_add(elist, &md->uevent_list);
3016 spin_unlock_irqrestore(&md->uevent_lock, flags);
3017}
3018
3019/*
3020 * The gendisk is only valid as long as you have a reference
3021 * count on 'md'.
3022 */
3023struct gendisk *dm_disk(struct mapped_device *md)
3024{
3025 return md->disk;
3026}
3027EXPORT_SYMBOL_GPL(dm_disk);
3028
3029struct kobject *dm_kobject(struct mapped_device *md)
3030{
3031 return &md->kobj_holder.kobj;
3032}
3033
3034struct mapped_device *dm_get_from_kobject(struct kobject *kobj)
3035{
3036 struct mapped_device *md;
3037
3038 md = container_of(kobj, struct mapped_device, kobj_holder.kobj);
3039
3040 spin_lock(&_minor_lock);
3041 if (test_bit(DMF_FREEING, &md->flags) || dm_deleting_md(md)) {
3042 md = NULL;
3043 goto out;
3044 }
3045 dm_get(md);
3046out:
3047 spin_unlock(&_minor_lock);
3048
3049 return md;
3050}
3051
3052int dm_suspended_md(struct mapped_device *md)
3053{
3054 return test_bit(DMF_SUSPENDED, &md->flags);
3055}
3056
3057static int dm_post_suspending_md(struct mapped_device *md)
3058{
3059 return test_bit(DMF_POST_SUSPENDING, &md->flags);
3060}
3061
3062int dm_suspended_internally_md(struct mapped_device *md)
3063{
3064 return test_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
3065}
3066
3067int dm_test_deferred_remove_flag(struct mapped_device *md)
3068{
3069 return test_bit(DMF_DEFERRED_REMOVE, &md->flags);
3070}
3071
3072int dm_suspended(struct dm_target *ti)
3073{
3074 return dm_suspended_md(ti->table->md);
3075}
3076EXPORT_SYMBOL_GPL(dm_suspended);
3077
3078int dm_post_suspending(struct dm_target *ti)
3079{
3080 return dm_post_suspending_md(ti->table->md);
3081}
3082EXPORT_SYMBOL_GPL(dm_post_suspending);
3083
3084int dm_noflush_suspending(struct dm_target *ti)
3085{
3086 return __noflush_suspending(ti->table->md);
3087}
3088EXPORT_SYMBOL_GPL(dm_noflush_suspending);
3089
3090void dm_free_md_mempools(struct dm_md_mempools *pools)
3091{
3092 if (!pools)
3093 return;
3094
3095 bioset_exit(&pools->bs);
3096 bioset_exit(&pools->io_bs);
3097
3098 kfree(pools);
3099}
3100
3101struct dm_pr {
3102 u64 old_key;
3103 u64 new_key;
3104 u32 flags;
3105 bool abort;
3106 bool fail_early;
3107 int ret;
3108 enum pr_type type;
3109};
3110
3111static int dm_call_pr(struct block_device *bdev, iterate_devices_callout_fn fn,
3112 struct dm_pr *pr)
3113{
3114 struct mapped_device *md = bdev->bd_disk->private_data;
3115 struct dm_table *table;
3116 struct dm_target *ti;
3117 int ret = -ENOTTY, srcu_idx;
3118
3119 table = dm_get_live_table(md, &srcu_idx);
3120 if (!table || !dm_table_get_size(table))
3121 goto out;
3122
3123 /* We only support devices that have a single target */
3124 if (table->num_targets != 1)
3125 goto out;
3126 ti = dm_table_get_target(table, 0);
3127
3128 if (dm_suspended_md(md)) {
3129 ret = -EAGAIN;
3130 goto out;
3131 }
3132
3133 ret = -EINVAL;
3134 if (!ti->type->iterate_devices)
3135 goto out;
3136
3137 ti->type->iterate_devices(ti, fn, pr);
3138 ret = 0;
3139out:
3140 dm_put_live_table(md, srcu_idx);
3141 return ret;
3142}
3143
3144/*
3145 * For register / unregister we need to manually call out to every path.
3146 */
3147static int __dm_pr_register(struct dm_target *ti, struct dm_dev *dev,
3148 sector_t start, sector_t len, void *data)
3149{
3150 struct dm_pr *pr = data;
3151 const struct pr_ops *ops = dev->bdev->bd_disk->fops->pr_ops;
3152 int ret;
3153
3154 if (!ops || !ops->pr_register) {
3155 pr->ret = -EOPNOTSUPP;
3156 return -1;
3157 }
3158
3159 ret = ops->pr_register(dev->bdev, pr->old_key, pr->new_key, pr->flags);
3160 if (!ret)
3161 return 0;
3162
3163 if (!pr->ret)
3164 pr->ret = ret;
3165
3166 if (pr->fail_early)
3167 return -1;
3168
3169 return 0;
3170}
3171
3172static int dm_pr_register(struct block_device *bdev, u64 old_key, u64 new_key,
3173 u32 flags)
3174{
3175 struct dm_pr pr = {
3176 .old_key = old_key,
3177 .new_key = new_key,
3178 .flags = flags,
3179 .fail_early = true,
3180 .ret = 0,
3181 };
3182 int ret;
3183
3184 ret = dm_call_pr(bdev, __dm_pr_register, &pr);
3185 if (ret) {
3186 /* Didn't even get to register a path */
3187 return ret;
3188 }
3189
3190 if (!pr.ret)
3191 return 0;
3192 ret = pr.ret;
3193
3194 if (!new_key)
3195 return ret;
3196
3197 /* unregister all paths if we failed to register any path */
3198 pr.old_key = new_key;
3199 pr.new_key = 0;
3200 pr.flags = 0;
3201 pr.fail_early = false;
3202 (void) dm_call_pr(bdev, __dm_pr_register, &pr);
3203 return ret;
3204}
3205
3206
3207static int __dm_pr_reserve(struct dm_target *ti, struct dm_dev *dev,
3208 sector_t start, sector_t len, void *data)
3209{
3210 struct dm_pr *pr = data;
3211 const struct pr_ops *ops = dev->bdev->bd_disk->fops->pr_ops;
3212
3213 if (!ops || !ops->pr_reserve) {
3214 pr->ret = -EOPNOTSUPP;
3215 return -1;
3216 }
3217
3218 pr->ret = ops->pr_reserve(dev->bdev, pr->old_key, pr->type, pr->flags);
3219 if (!pr->ret)
3220 return -1;
3221
3222 return 0;
3223}
3224
3225static int dm_pr_reserve(struct block_device *bdev, u64 key, enum pr_type type,
3226 u32 flags)
3227{
3228 struct dm_pr pr = {
3229 .old_key = key,
3230 .flags = flags,
3231 .type = type,
3232 .fail_early = false,
3233 .ret = 0,
3234 };
3235 int ret;
3236
3237 ret = dm_call_pr(bdev, __dm_pr_reserve, &pr);
3238 if (ret)
3239 return ret;
3240
3241 return pr.ret;
3242}
3243
3244/*
3245 * If there is a non-All Registrants type of reservation, the release must be
3246 * sent down the holding path. For the cases where there is no reservation or
3247 * the path is not the holder the device will also return success, so we must
3248 * try each path to make sure we got the correct path.
3249 */
3250static int __dm_pr_release(struct dm_target *ti, struct dm_dev *dev,
3251 sector_t start, sector_t len, void *data)
3252{
3253 struct dm_pr *pr = data;
3254 const struct pr_ops *ops = dev->bdev->bd_disk->fops->pr_ops;
3255
3256 if (!ops || !ops->pr_release) {
3257 pr->ret = -EOPNOTSUPP;
3258 return -1;
3259 }
3260
3261 pr->ret = ops->pr_release(dev->bdev, pr->old_key, pr->type);
3262 if (pr->ret)
3263 return -1;
3264
3265 return 0;
3266}
3267
3268static int dm_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
3269{
3270 struct dm_pr pr = {
3271 .old_key = key,
3272 .type = type,
3273 .fail_early = false,
3274 };
3275 int ret;
3276
3277 ret = dm_call_pr(bdev, __dm_pr_release, &pr);
3278 if (ret)
3279 return ret;
3280
3281 return pr.ret;
3282}
3283
3284static int __dm_pr_preempt(struct dm_target *ti, struct dm_dev *dev,
3285 sector_t start, sector_t len, void *data)
3286{
3287 struct dm_pr *pr = data;
3288 const struct pr_ops *ops = dev->bdev->bd_disk->fops->pr_ops;
3289
3290 if (!ops || !ops->pr_preempt) {
3291 pr->ret = -EOPNOTSUPP;
3292 return -1;
3293 }
3294
3295 pr->ret = ops->pr_preempt(dev->bdev, pr->old_key, pr->new_key, pr->type,
3296 pr->abort);
3297 if (!pr->ret)
3298 return -1;
3299
3300 return 0;
3301}
3302
3303static int dm_pr_preempt(struct block_device *bdev, u64 old_key, u64 new_key,
3304 enum pr_type type, bool abort)
3305{
3306 struct dm_pr pr = {
3307 .new_key = new_key,
3308 .old_key = old_key,
3309 .type = type,
3310 .fail_early = false,
3311 };
3312 int ret;
3313
3314 ret = dm_call_pr(bdev, __dm_pr_preempt, &pr);
3315 if (ret)
3316 return ret;
3317
3318 return pr.ret;
3319}
3320
3321static int dm_pr_clear(struct block_device *bdev, u64 key)
3322{
3323 struct mapped_device *md = bdev->bd_disk->private_data;
3324 const struct pr_ops *ops;
3325 int r, srcu_idx;
3326
3327 r = dm_prepare_ioctl(md, &srcu_idx, &bdev);
3328 if (r < 0)
3329 goto out;
3330
3331 ops = bdev->bd_disk->fops->pr_ops;
3332 if (ops && ops->pr_clear)
3333 r = ops->pr_clear(bdev, key);
3334 else
3335 r = -EOPNOTSUPP;
3336out:
3337 dm_unprepare_ioctl(md, srcu_idx);
3338 return r;
3339}
3340
3341static const struct pr_ops dm_pr_ops = {
3342 .pr_register = dm_pr_register,
3343 .pr_reserve = dm_pr_reserve,
3344 .pr_release = dm_pr_release,
3345 .pr_preempt = dm_pr_preempt,
3346 .pr_clear = dm_pr_clear,
3347};
3348
3349static const struct block_device_operations dm_blk_dops = {
3350 .submit_bio = dm_submit_bio,
3351 .poll_bio = dm_poll_bio,
3352 .open = dm_blk_open,
3353 .release = dm_blk_close,
3354 .ioctl = dm_blk_ioctl,
3355 .getgeo = dm_blk_getgeo,
3356 .report_zones = dm_blk_report_zones,
3357 .pr_ops = &dm_pr_ops,
3358 .owner = THIS_MODULE
3359};
3360
3361static const struct block_device_operations dm_rq_blk_dops = {
3362 .open = dm_blk_open,
3363 .release = dm_blk_close,
3364 .ioctl = dm_blk_ioctl,
3365 .getgeo = dm_blk_getgeo,
3366 .pr_ops = &dm_pr_ops,
3367 .owner = THIS_MODULE
3368};
3369
3370static const struct dax_operations dm_dax_ops = {
3371 .direct_access = dm_dax_direct_access,
3372 .zero_page_range = dm_dax_zero_page_range,
3373 .recovery_write = dm_dax_recovery_write,
3374};
3375
3376/*
3377 * module hooks
3378 */
3379module_init(dm_init);
3380module_exit(dm_exit);
3381
3382module_param(major, uint, 0);
3383MODULE_PARM_DESC(major, "The major number of the device mapper");
3384
3385module_param(reserved_bio_based_ios, uint, S_IRUGO | S_IWUSR);
3386MODULE_PARM_DESC(reserved_bio_based_ios, "Reserved IOs in bio-based mempools");
3387
3388module_param(dm_numa_node, int, S_IRUGO | S_IWUSR);
3389MODULE_PARM_DESC(dm_numa_node, "NUMA node for DM device memory allocations");
3390
3391module_param(swap_bios, int, S_IRUGO | S_IWUSR);
3392MODULE_PARM_DESC(swap_bios, "Maximum allowed inflight swap IOs");
3393
3394MODULE_DESCRIPTION(DM_NAME " driver");
3395MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
3396MODULE_LICENSE("GPL");
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/signal.h>
16#include <linux/blkpg.h>
17#include <linux/bio.h>
18#include <linux/mempool.h>
19#include <linux/dax.h>
20#include <linux/slab.h>
21#include <linux/idr.h>
22#include <linux/uio.h>
23#include <linux/hdreg.h>
24#include <linux/delay.h>
25#include <linux/wait.h>
26#include <linux/pr.h>
27#include <linux/refcount.h>
28
29#define DM_MSG_PREFIX "core"
30
31/*
32 * Cookies are numeric values sent with CHANGE and REMOVE
33 * uevents while resuming, removing or renaming the device.
34 */
35#define DM_COOKIE_ENV_VAR_NAME "DM_COOKIE"
36#define DM_COOKIE_LENGTH 24
37
38static const char *_name = DM_NAME;
39
40static unsigned int major = 0;
41static unsigned int _major = 0;
42
43static DEFINE_IDR(_minor_idr);
44
45static DEFINE_SPINLOCK(_minor_lock);
46
47static void do_deferred_remove(struct work_struct *w);
48
49static DECLARE_WORK(deferred_remove_work, do_deferred_remove);
50
51static struct workqueue_struct *deferred_remove_workqueue;
52
53atomic_t dm_global_event_nr = ATOMIC_INIT(0);
54DECLARE_WAIT_QUEUE_HEAD(dm_global_eventq);
55
56void dm_issue_global_event(void)
57{
58 atomic_inc(&dm_global_event_nr);
59 wake_up(&dm_global_eventq);
60}
61
62/*
63 * One of these is allocated (on-stack) per original bio.
64 */
65struct clone_info {
66 struct dm_table *map;
67 struct bio *bio;
68 struct dm_io *io;
69 sector_t sector;
70 unsigned sector_count;
71};
72
73/*
74 * One of these is allocated per clone bio.
75 */
76#define DM_TIO_MAGIC 7282014
77struct dm_target_io {
78 unsigned magic;
79 struct dm_io *io;
80 struct dm_target *ti;
81 unsigned target_bio_nr;
82 unsigned *len_ptr;
83 bool inside_dm_io;
84 struct bio clone;
85};
86
87/*
88 * One of these is allocated per original bio.
89 * It contains the first clone used for that original.
90 */
91#define DM_IO_MAGIC 5191977
92struct dm_io {
93 unsigned magic;
94 struct mapped_device *md;
95 blk_status_t status;
96 atomic_t io_count;
97 struct bio *orig_bio;
98 unsigned long start_time;
99 spinlock_t endio_lock;
100 struct dm_stats_aux stats_aux;
101 /* last member of dm_target_io is 'struct bio' */
102 struct dm_target_io tio;
103};
104
105void *dm_per_bio_data(struct bio *bio, size_t data_size)
106{
107 struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone);
108 if (!tio->inside_dm_io)
109 return (char *)bio - offsetof(struct dm_target_io, clone) - data_size;
110 return (char *)bio - offsetof(struct dm_target_io, clone) - offsetof(struct dm_io, tio) - data_size;
111}
112EXPORT_SYMBOL_GPL(dm_per_bio_data);
113
114struct bio *dm_bio_from_per_bio_data(void *data, size_t data_size)
115{
116 struct dm_io *io = (struct dm_io *)((char *)data + data_size);
117 if (io->magic == DM_IO_MAGIC)
118 return (struct bio *)((char *)io + offsetof(struct dm_io, tio) + offsetof(struct dm_target_io, clone));
119 BUG_ON(io->magic != DM_TIO_MAGIC);
120 return (struct bio *)((char *)io + offsetof(struct dm_target_io, clone));
121}
122EXPORT_SYMBOL_GPL(dm_bio_from_per_bio_data);
123
124unsigned dm_bio_get_target_bio_nr(const struct bio *bio)
125{
126 return container_of(bio, struct dm_target_io, clone)->target_bio_nr;
127}
128EXPORT_SYMBOL_GPL(dm_bio_get_target_bio_nr);
129
130#define MINOR_ALLOCED ((void *)-1)
131
132/*
133 * Bits for the md->flags field.
134 */
135#define DMF_BLOCK_IO_FOR_SUSPEND 0
136#define DMF_SUSPENDED 1
137#define DMF_FROZEN 2
138#define DMF_FREEING 3
139#define DMF_DELETING 4
140#define DMF_NOFLUSH_SUSPENDING 5
141#define DMF_DEFERRED_REMOVE 6
142#define DMF_SUSPENDED_INTERNALLY 7
143
144#define DM_NUMA_NODE NUMA_NO_NODE
145static int dm_numa_node = DM_NUMA_NODE;
146
147/*
148 * For mempools pre-allocation at the table loading time.
149 */
150struct dm_md_mempools {
151 struct bio_set *bs;
152 struct bio_set *io_bs;
153};
154
155struct table_device {
156 struct list_head list;
157 refcount_t count;
158 struct dm_dev dm_dev;
159};
160
161static struct kmem_cache *_rq_tio_cache;
162static struct kmem_cache *_rq_cache;
163
164/*
165 * Bio-based DM's mempools' reserved IOs set by the user.
166 */
167#define RESERVED_BIO_BASED_IOS 16
168static unsigned reserved_bio_based_ios = RESERVED_BIO_BASED_IOS;
169
170static int __dm_get_module_param_int(int *module_param, int min, int max)
171{
172 int param = READ_ONCE(*module_param);
173 int modified_param = 0;
174 bool modified = true;
175
176 if (param < min)
177 modified_param = min;
178 else if (param > max)
179 modified_param = max;
180 else
181 modified = false;
182
183 if (modified) {
184 (void)cmpxchg(module_param, param, modified_param);
185 param = modified_param;
186 }
187
188 return param;
189}
190
191unsigned __dm_get_module_param(unsigned *module_param,
192 unsigned def, unsigned max)
193{
194 unsigned param = READ_ONCE(*module_param);
195 unsigned modified_param = 0;
196
197 if (!param)
198 modified_param = def;
199 else if (param > max)
200 modified_param = max;
201
202 if (modified_param) {
203 (void)cmpxchg(module_param, param, modified_param);
204 param = modified_param;
205 }
206
207 return param;
208}
209
210unsigned dm_get_reserved_bio_based_ios(void)
211{
212 return __dm_get_module_param(&reserved_bio_based_ios,
213 RESERVED_BIO_BASED_IOS, DM_RESERVED_MAX_IOS);
214}
215EXPORT_SYMBOL_GPL(dm_get_reserved_bio_based_ios);
216
217static unsigned dm_get_numa_node(void)
218{
219 return __dm_get_module_param_int(&dm_numa_node,
220 DM_NUMA_NODE, num_online_nodes() - 1);
221}
222
223static int __init local_init(void)
224{
225 int r = -ENOMEM;
226
227 _rq_tio_cache = KMEM_CACHE(dm_rq_target_io, 0);
228 if (!_rq_tio_cache)
229 return r;
230
231 _rq_cache = kmem_cache_create("dm_old_clone_request", sizeof(struct request),
232 __alignof__(struct request), 0, NULL);
233 if (!_rq_cache)
234 goto out_free_rq_tio_cache;
235
236 r = dm_uevent_init();
237 if (r)
238 goto out_free_rq_cache;
239
240 deferred_remove_workqueue = alloc_workqueue("kdmremove", WQ_UNBOUND, 1);
241 if (!deferred_remove_workqueue) {
242 r = -ENOMEM;
243 goto out_uevent_exit;
244 }
245
246 _major = major;
247 r = register_blkdev(_major, _name);
248 if (r < 0)
249 goto out_free_workqueue;
250
251 if (!_major)
252 _major = r;
253
254 return 0;
255
256out_free_workqueue:
257 destroy_workqueue(deferred_remove_workqueue);
258out_uevent_exit:
259 dm_uevent_exit();
260out_free_rq_cache:
261 kmem_cache_destroy(_rq_cache);
262out_free_rq_tio_cache:
263 kmem_cache_destroy(_rq_tio_cache);
264
265 return r;
266}
267
268static void local_exit(void)
269{
270 flush_scheduled_work();
271 destroy_workqueue(deferred_remove_workqueue);
272
273 kmem_cache_destroy(_rq_cache);
274 kmem_cache_destroy(_rq_tio_cache);
275 unregister_blkdev(_major, _name);
276 dm_uevent_exit();
277
278 _major = 0;
279
280 DMINFO("cleaned up");
281}
282
283static int (*_inits[])(void) __initdata = {
284 local_init,
285 dm_target_init,
286 dm_linear_init,
287 dm_stripe_init,
288 dm_io_init,
289 dm_kcopyd_init,
290 dm_interface_init,
291 dm_statistics_init,
292};
293
294static void (*_exits[])(void) = {
295 local_exit,
296 dm_target_exit,
297 dm_linear_exit,
298 dm_stripe_exit,
299 dm_io_exit,
300 dm_kcopyd_exit,
301 dm_interface_exit,
302 dm_statistics_exit,
303};
304
305static int __init dm_init(void)
306{
307 const int count = ARRAY_SIZE(_inits);
308
309 int r, i;
310
311 for (i = 0; i < count; i++) {
312 r = _inits[i]();
313 if (r)
314 goto bad;
315 }
316
317 return 0;
318
319 bad:
320 while (i--)
321 _exits[i]();
322
323 return r;
324}
325
326static void __exit dm_exit(void)
327{
328 int i = ARRAY_SIZE(_exits);
329
330 while (i--)
331 _exits[i]();
332
333 /*
334 * Should be empty by this point.
335 */
336 idr_destroy(&_minor_idr);
337}
338
339/*
340 * Block device functions
341 */
342int dm_deleting_md(struct mapped_device *md)
343{
344 return test_bit(DMF_DELETING, &md->flags);
345}
346
347static int dm_blk_open(struct block_device *bdev, fmode_t mode)
348{
349 struct mapped_device *md;
350
351 spin_lock(&_minor_lock);
352
353 md = bdev->bd_disk->private_data;
354 if (!md)
355 goto out;
356
357 if (test_bit(DMF_FREEING, &md->flags) ||
358 dm_deleting_md(md)) {
359 md = NULL;
360 goto out;
361 }
362
363 dm_get(md);
364 atomic_inc(&md->open_count);
365out:
366 spin_unlock(&_minor_lock);
367
368 return md ? 0 : -ENXIO;
369}
370
371static void dm_blk_close(struct gendisk *disk, fmode_t mode)
372{
373 struct mapped_device *md;
374
375 spin_lock(&_minor_lock);
376
377 md = disk->private_data;
378 if (WARN_ON(!md))
379 goto out;
380
381 if (atomic_dec_and_test(&md->open_count) &&
382 (test_bit(DMF_DEFERRED_REMOVE, &md->flags)))
383 queue_work(deferred_remove_workqueue, &deferred_remove_work);
384
385 dm_put(md);
386out:
387 spin_unlock(&_minor_lock);
388}
389
390int dm_open_count(struct mapped_device *md)
391{
392 return atomic_read(&md->open_count);
393}
394
395/*
396 * Guarantees nothing is using the device before it's deleted.
397 */
398int dm_lock_for_deletion(struct mapped_device *md, bool mark_deferred, bool only_deferred)
399{
400 int r = 0;
401
402 spin_lock(&_minor_lock);
403
404 if (dm_open_count(md)) {
405 r = -EBUSY;
406 if (mark_deferred)
407 set_bit(DMF_DEFERRED_REMOVE, &md->flags);
408 } else if (only_deferred && !test_bit(DMF_DEFERRED_REMOVE, &md->flags))
409 r = -EEXIST;
410 else
411 set_bit(DMF_DELETING, &md->flags);
412
413 spin_unlock(&_minor_lock);
414
415 return r;
416}
417
418int dm_cancel_deferred_remove(struct mapped_device *md)
419{
420 int r = 0;
421
422 spin_lock(&_minor_lock);
423
424 if (test_bit(DMF_DELETING, &md->flags))
425 r = -EBUSY;
426 else
427 clear_bit(DMF_DEFERRED_REMOVE, &md->flags);
428
429 spin_unlock(&_minor_lock);
430
431 return r;
432}
433
434static void do_deferred_remove(struct work_struct *w)
435{
436 dm_deferred_remove();
437}
438
439sector_t dm_get_size(struct mapped_device *md)
440{
441 return get_capacity(md->disk);
442}
443
444struct request_queue *dm_get_md_queue(struct mapped_device *md)
445{
446 return md->queue;
447}
448
449struct dm_stats *dm_get_stats(struct mapped_device *md)
450{
451 return &md->stats;
452}
453
454static int dm_blk_getgeo(struct block_device *bdev, struct hd_geometry *geo)
455{
456 struct mapped_device *md = bdev->bd_disk->private_data;
457
458 return dm_get_geometry(md, geo);
459}
460
461static int dm_prepare_ioctl(struct mapped_device *md, int *srcu_idx,
462 struct block_device **bdev)
463 __acquires(md->io_barrier)
464{
465 struct dm_target *tgt;
466 struct dm_table *map;
467 int r;
468
469retry:
470 r = -ENOTTY;
471 map = dm_get_live_table(md, srcu_idx);
472 if (!map || !dm_table_get_size(map))
473 return r;
474
475 /* We only support devices that have a single target */
476 if (dm_table_get_num_targets(map) != 1)
477 return r;
478
479 tgt = dm_table_get_target(map, 0);
480 if (!tgt->type->prepare_ioctl)
481 return r;
482
483 if (dm_suspended_md(md))
484 return -EAGAIN;
485
486 r = tgt->type->prepare_ioctl(tgt, bdev);
487 if (r == -ENOTCONN && !fatal_signal_pending(current)) {
488 dm_put_live_table(md, *srcu_idx);
489 msleep(10);
490 goto retry;
491 }
492
493 return r;
494}
495
496static void dm_unprepare_ioctl(struct mapped_device *md, int srcu_idx)
497 __releases(md->io_barrier)
498{
499 dm_put_live_table(md, srcu_idx);
500}
501
502static int dm_blk_ioctl(struct block_device *bdev, fmode_t mode,
503 unsigned int cmd, unsigned long arg)
504{
505 struct mapped_device *md = bdev->bd_disk->private_data;
506 int r, srcu_idx;
507
508 r = dm_prepare_ioctl(md, &srcu_idx, &bdev);
509 if (r < 0)
510 goto out;
511
512 if (r > 0) {
513 /*
514 * Target determined this ioctl is being issued against a
515 * subset of the parent bdev; require extra privileges.
516 */
517 if (!capable(CAP_SYS_RAWIO)) {
518 DMWARN_LIMIT(
519 "%s: sending ioctl %x to DM device without required privilege.",
520 current->comm, cmd);
521 r = -ENOIOCTLCMD;
522 goto out;
523 }
524 }
525
526 r = __blkdev_driver_ioctl(bdev, mode, cmd, arg);
527out:
528 dm_unprepare_ioctl(md, srcu_idx);
529 return r;
530}
531
532static void start_io_acct(struct dm_io *io);
533
534static struct dm_io *alloc_io(struct mapped_device *md, struct bio *bio)
535{
536 struct dm_io *io;
537 struct dm_target_io *tio;
538 struct bio *clone;
539
540 clone = bio_alloc_bioset(GFP_NOIO, 0, md->io_bs);
541 if (!clone)
542 return NULL;
543
544 tio = container_of(clone, struct dm_target_io, clone);
545 tio->inside_dm_io = true;
546 tio->io = NULL;
547
548 io = container_of(tio, struct dm_io, tio);
549 io->magic = DM_IO_MAGIC;
550 io->status = 0;
551 atomic_set(&io->io_count, 1);
552 io->orig_bio = bio;
553 io->md = md;
554 spin_lock_init(&io->endio_lock);
555
556 start_io_acct(io);
557
558 return io;
559}
560
561static void free_io(struct mapped_device *md, struct dm_io *io)
562{
563 bio_put(&io->tio.clone);
564}
565
566static struct dm_target_io *alloc_tio(struct clone_info *ci, struct dm_target *ti,
567 unsigned target_bio_nr, gfp_t gfp_mask)
568{
569 struct dm_target_io *tio;
570
571 if (!ci->io->tio.io) {
572 /* the dm_target_io embedded in ci->io is available */
573 tio = &ci->io->tio;
574 } else {
575 struct bio *clone = bio_alloc_bioset(gfp_mask, 0, ci->io->md->bs);
576 if (!clone)
577 return NULL;
578
579 tio = container_of(clone, struct dm_target_io, clone);
580 tio->inside_dm_io = false;
581 }
582
583 tio->magic = DM_TIO_MAGIC;
584 tio->io = ci->io;
585 tio->ti = ti;
586 tio->target_bio_nr = target_bio_nr;
587
588 return tio;
589}
590
591static void free_tio(struct dm_target_io *tio)
592{
593 if (tio->inside_dm_io)
594 return;
595 bio_put(&tio->clone);
596}
597
598int md_in_flight(struct mapped_device *md)
599{
600 return atomic_read(&md->pending[READ]) +
601 atomic_read(&md->pending[WRITE]);
602}
603
604static void start_io_acct(struct dm_io *io)
605{
606 struct mapped_device *md = io->md;
607 struct bio *bio = io->orig_bio;
608 int rw = bio_data_dir(bio);
609
610 io->start_time = jiffies;
611
612 generic_start_io_acct(md->queue, rw, bio_sectors(bio), &dm_disk(md)->part0);
613
614 atomic_set(&dm_disk(md)->part0.in_flight[rw],
615 atomic_inc_return(&md->pending[rw]));
616
617 if (unlikely(dm_stats_used(&md->stats)))
618 dm_stats_account_io(&md->stats, bio_data_dir(bio),
619 bio->bi_iter.bi_sector, bio_sectors(bio),
620 false, 0, &io->stats_aux);
621}
622
623static void end_io_acct(struct dm_io *io)
624{
625 struct mapped_device *md = io->md;
626 struct bio *bio = io->orig_bio;
627 unsigned long duration = jiffies - io->start_time;
628 int pending;
629 int rw = bio_data_dir(bio);
630
631 generic_end_io_acct(md->queue, rw, &dm_disk(md)->part0, io->start_time);
632
633 if (unlikely(dm_stats_used(&md->stats)))
634 dm_stats_account_io(&md->stats, bio_data_dir(bio),
635 bio->bi_iter.bi_sector, bio_sectors(bio),
636 true, duration, &io->stats_aux);
637
638 /*
639 * After this is decremented the bio must not be touched if it is
640 * a flush.
641 */
642 pending = atomic_dec_return(&md->pending[rw]);
643 atomic_set(&dm_disk(md)->part0.in_flight[rw], pending);
644 pending += atomic_read(&md->pending[rw^0x1]);
645
646 /* nudge anyone waiting on suspend queue */
647 if (!pending)
648 wake_up(&md->wait);
649}
650
651/*
652 * Add the bio to the list of deferred io.
653 */
654static void queue_io(struct mapped_device *md, struct bio *bio)
655{
656 unsigned long flags;
657
658 spin_lock_irqsave(&md->deferred_lock, flags);
659 bio_list_add(&md->deferred, bio);
660 spin_unlock_irqrestore(&md->deferred_lock, flags);
661 queue_work(md->wq, &md->work);
662}
663
664/*
665 * Everyone (including functions in this file), should use this
666 * function to access the md->map field, and make sure they call
667 * dm_put_live_table() when finished.
668 */
669struct dm_table *dm_get_live_table(struct mapped_device *md, int *srcu_idx) __acquires(md->io_barrier)
670{
671 *srcu_idx = srcu_read_lock(&md->io_barrier);
672
673 return srcu_dereference(md->map, &md->io_barrier);
674}
675
676void dm_put_live_table(struct mapped_device *md, int srcu_idx) __releases(md->io_barrier)
677{
678 srcu_read_unlock(&md->io_barrier, srcu_idx);
679}
680
681void dm_sync_table(struct mapped_device *md)
682{
683 synchronize_srcu(&md->io_barrier);
684 synchronize_rcu_expedited();
685}
686
687/*
688 * A fast alternative to dm_get_live_table/dm_put_live_table.
689 * The caller must not block between these two functions.
690 */
691static struct dm_table *dm_get_live_table_fast(struct mapped_device *md) __acquires(RCU)
692{
693 rcu_read_lock();
694 return rcu_dereference(md->map);
695}
696
697static void dm_put_live_table_fast(struct mapped_device *md) __releases(RCU)
698{
699 rcu_read_unlock();
700}
701
702static char *_dm_claim_ptr = "I belong to device-mapper";
703
704/*
705 * Open a table device so we can use it as a map destination.
706 */
707static int open_table_device(struct table_device *td, dev_t dev,
708 struct mapped_device *md)
709{
710 struct block_device *bdev;
711
712 int r;
713
714 BUG_ON(td->dm_dev.bdev);
715
716 bdev = blkdev_get_by_dev(dev, td->dm_dev.mode | FMODE_EXCL, _dm_claim_ptr);
717 if (IS_ERR(bdev))
718 return PTR_ERR(bdev);
719
720 r = bd_link_disk_holder(bdev, dm_disk(md));
721 if (r) {
722 blkdev_put(bdev, td->dm_dev.mode | FMODE_EXCL);
723 return r;
724 }
725
726 td->dm_dev.bdev = bdev;
727 td->dm_dev.dax_dev = dax_get_by_host(bdev->bd_disk->disk_name);
728 return 0;
729}
730
731/*
732 * Close a table device that we've been using.
733 */
734static void close_table_device(struct table_device *td, struct mapped_device *md)
735{
736 if (!td->dm_dev.bdev)
737 return;
738
739 bd_unlink_disk_holder(td->dm_dev.bdev, dm_disk(md));
740 blkdev_put(td->dm_dev.bdev, td->dm_dev.mode | FMODE_EXCL);
741 put_dax(td->dm_dev.dax_dev);
742 td->dm_dev.bdev = NULL;
743 td->dm_dev.dax_dev = NULL;
744}
745
746static struct table_device *find_table_device(struct list_head *l, dev_t dev,
747 fmode_t mode) {
748 struct table_device *td;
749
750 list_for_each_entry(td, l, list)
751 if (td->dm_dev.bdev->bd_dev == dev && td->dm_dev.mode == mode)
752 return td;
753
754 return NULL;
755}
756
757int dm_get_table_device(struct mapped_device *md, dev_t dev, fmode_t mode,
758 struct dm_dev **result) {
759 int r;
760 struct table_device *td;
761
762 mutex_lock(&md->table_devices_lock);
763 td = find_table_device(&md->table_devices, dev, mode);
764 if (!td) {
765 td = kmalloc_node(sizeof(*td), GFP_KERNEL, md->numa_node_id);
766 if (!td) {
767 mutex_unlock(&md->table_devices_lock);
768 return -ENOMEM;
769 }
770
771 td->dm_dev.mode = mode;
772 td->dm_dev.bdev = NULL;
773
774 if ((r = open_table_device(td, dev, md))) {
775 mutex_unlock(&md->table_devices_lock);
776 kfree(td);
777 return r;
778 }
779
780 format_dev_t(td->dm_dev.name, dev);
781
782 refcount_set(&td->count, 1);
783 list_add(&td->list, &md->table_devices);
784 } else {
785 refcount_inc(&td->count);
786 }
787 mutex_unlock(&md->table_devices_lock);
788
789 *result = &td->dm_dev;
790 return 0;
791}
792EXPORT_SYMBOL_GPL(dm_get_table_device);
793
794void dm_put_table_device(struct mapped_device *md, struct dm_dev *d)
795{
796 struct table_device *td = container_of(d, struct table_device, dm_dev);
797
798 mutex_lock(&md->table_devices_lock);
799 if (refcount_dec_and_test(&td->count)) {
800 close_table_device(td, md);
801 list_del(&td->list);
802 kfree(td);
803 }
804 mutex_unlock(&md->table_devices_lock);
805}
806EXPORT_SYMBOL(dm_put_table_device);
807
808static void free_table_devices(struct list_head *devices)
809{
810 struct list_head *tmp, *next;
811
812 list_for_each_safe(tmp, next, devices) {
813 struct table_device *td = list_entry(tmp, struct table_device, list);
814
815 DMWARN("dm_destroy: %s still exists with %d references",
816 td->dm_dev.name, refcount_read(&td->count));
817 kfree(td);
818 }
819}
820
821/*
822 * Get the geometry associated with a dm device
823 */
824int dm_get_geometry(struct mapped_device *md, struct hd_geometry *geo)
825{
826 *geo = md->geometry;
827
828 return 0;
829}
830
831/*
832 * Set the geometry of a device.
833 */
834int dm_set_geometry(struct mapped_device *md, struct hd_geometry *geo)
835{
836 sector_t sz = (sector_t)geo->cylinders * geo->heads * geo->sectors;
837
838 if (geo->start > sz) {
839 DMWARN("Start sector is beyond the geometry limits.");
840 return -EINVAL;
841 }
842
843 md->geometry = *geo;
844
845 return 0;
846}
847
848static int __noflush_suspending(struct mapped_device *md)
849{
850 return test_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
851}
852
853/*
854 * Decrements the number of outstanding ios that a bio has been
855 * cloned into, completing the original io if necc.
856 */
857static void dec_pending(struct dm_io *io, blk_status_t error)
858{
859 unsigned long flags;
860 blk_status_t io_error;
861 struct bio *bio;
862 struct mapped_device *md = io->md;
863
864 /* Push-back supersedes any I/O errors */
865 if (unlikely(error)) {
866 spin_lock_irqsave(&io->endio_lock, flags);
867 if (!(io->status == BLK_STS_DM_REQUEUE && __noflush_suspending(md)))
868 io->status = error;
869 spin_unlock_irqrestore(&io->endio_lock, flags);
870 }
871
872 if (atomic_dec_and_test(&io->io_count)) {
873 if (io->status == BLK_STS_DM_REQUEUE) {
874 /*
875 * Target requested pushing back the I/O.
876 */
877 spin_lock_irqsave(&md->deferred_lock, flags);
878 if (__noflush_suspending(md))
879 /* NOTE early return due to BLK_STS_DM_REQUEUE below */
880 bio_list_add_head(&md->deferred, io->orig_bio);
881 else
882 /* noflush suspend was interrupted. */
883 io->status = BLK_STS_IOERR;
884 spin_unlock_irqrestore(&md->deferred_lock, flags);
885 }
886
887 io_error = io->status;
888 bio = io->orig_bio;
889 end_io_acct(io);
890 free_io(md, io);
891
892 if (io_error == BLK_STS_DM_REQUEUE)
893 return;
894
895 if ((bio->bi_opf & REQ_PREFLUSH) && bio->bi_iter.bi_size) {
896 /*
897 * Preflush done for flush with data, reissue
898 * without REQ_PREFLUSH.
899 */
900 bio->bi_opf &= ~REQ_PREFLUSH;
901 queue_io(md, bio);
902 } else {
903 /* done with normal IO or empty flush */
904 if (io_error)
905 bio->bi_status = io_error;
906 bio_endio(bio);
907 }
908 }
909}
910
911void disable_write_same(struct mapped_device *md)
912{
913 struct queue_limits *limits = dm_get_queue_limits(md);
914
915 /* device doesn't really support WRITE SAME, disable it */
916 limits->max_write_same_sectors = 0;
917}
918
919void disable_write_zeroes(struct mapped_device *md)
920{
921 struct queue_limits *limits = dm_get_queue_limits(md);
922
923 /* device doesn't really support WRITE ZEROES, disable it */
924 limits->max_write_zeroes_sectors = 0;
925}
926
927static void clone_endio(struct bio *bio)
928{
929 blk_status_t error = bio->bi_status;
930 struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone);
931 struct dm_io *io = tio->io;
932 struct mapped_device *md = tio->io->md;
933 dm_endio_fn endio = tio->ti->type->end_io;
934
935 if (unlikely(error == BLK_STS_TARGET) && md->type != DM_TYPE_NVME_BIO_BASED) {
936 if (bio_op(bio) == REQ_OP_WRITE_SAME &&
937 !bio->bi_disk->queue->limits.max_write_same_sectors)
938 disable_write_same(md);
939 if (bio_op(bio) == REQ_OP_WRITE_ZEROES &&
940 !bio->bi_disk->queue->limits.max_write_zeroes_sectors)
941 disable_write_zeroes(md);
942 }
943
944 if (endio) {
945 int r = endio(tio->ti, bio, &error);
946 switch (r) {
947 case DM_ENDIO_REQUEUE:
948 error = BLK_STS_DM_REQUEUE;
949 /*FALLTHRU*/
950 case DM_ENDIO_DONE:
951 break;
952 case DM_ENDIO_INCOMPLETE:
953 /* The target will handle the io */
954 return;
955 default:
956 DMWARN("unimplemented target endio return value: %d", r);
957 BUG();
958 }
959 }
960
961 free_tio(tio);
962 dec_pending(io, error);
963}
964
965/*
966 * Return maximum size of I/O possible at the supplied sector up to the current
967 * target boundary.
968 */
969static sector_t max_io_len_target_boundary(sector_t sector, struct dm_target *ti)
970{
971 sector_t target_offset = dm_target_offset(ti, sector);
972
973 return ti->len - target_offset;
974}
975
976static sector_t max_io_len(sector_t sector, struct dm_target *ti)
977{
978 sector_t len = max_io_len_target_boundary(sector, ti);
979 sector_t offset, max_len;
980
981 /*
982 * Does the target need to split even further?
983 */
984 if (ti->max_io_len) {
985 offset = dm_target_offset(ti, sector);
986 if (unlikely(ti->max_io_len & (ti->max_io_len - 1)))
987 max_len = sector_div(offset, ti->max_io_len);
988 else
989 max_len = offset & (ti->max_io_len - 1);
990 max_len = ti->max_io_len - max_len;
991
992 if (len > max_len)
993 len = max_len;
994 }
995
996 return len;
997}
998
999int dm_set_target_max_io_len(struct dm_target *ti, sector_t len)
1000{
1001 if (len > UINT_MAX) {
1002 DMERR("Specified maximum size of target IO (%llu) exceeds limit (%u)",
1003 (unsigned long long)len, UINT_MAX);
1004 ti->error = "Maximum size of target IO is too large";
1005 return -EINVAL;
1006 }
1007
1008 /*
1009 * BIO based queue uses its own splitting. When multipage bvecs
1010 * is switched on, size of the incoming bio may be too big to
1011 * be handled in some targets, such as crypt.
1012 *
1013 * When these targets are ready for the big bio, we can remove
1014 * the limit.
1015 */
1016 ti->max_io_len = min_t(uint32_t, len, BIO_MAX_PAGES * PAGE_SIZE);
1017
1018 return 0;
1019}
1020EXPORT_SYMBOL_GPL(dm_set_target_max_io_len);
1021
1022static struct dm_target *dm_dax_get_live_target(struct mapped_device *md,
1023 sector_t sector, int *srcu_idx)
1024 __acquires(md->io_barrier)
1025{
1026 struct dm_table *map;
1027 struct dm_target *ti;
1028
1029 map = dm_get_live_table(md, srcu_idx);
1030 if (!map)
1031 return NULL;
1032
1033 ti = dm_table_find_target(map, sector);
1034 if (!dm_target_is_valid(ti))
1035 return NULL;
1036
1037 return ti;
1038}
1039
1040static long dm_dax_direct_access(struct dax_device *dax_dev, pgoff_t pgoff,
1041 long nr_pages, void **kaddr, pfn_t *pfn)
1042{
1043 struct mapped_device *md = dax_get_private(dax_dev);
1044 sector_t sector = pgoff * PAGE_SECTORS;
1045 struct dm_target *ti;
1046 long len, ret = -EIO;
1047 int srcu_idx;
1048
1049 ti = dm_dax_get_live_target(md, sector, &srcu_idx);
1050
1051 if (!ti)
1052 goto out;
1053 if (!ti->type->direct_access)
1054 goto out;
1055 len = max_io_len(sector, ti) / PAGE_SECTORS;
1056 if (len < 1)
1057 goto out;
1058 nr_pages = min(len, nr_pages);
1059 if (ti->type->direct_access)
1060 ret = ti->type->direct_access(ti, pgoff, nr_pages, kaddr, pfn);
1061
1062 out:
1063 dm_put_live_table(md, srcu_idx);
1064
1065 return ret;
1066}
1067
1068static size_t dm_dax_copy_from_iter(struct dax_device *dax_dev, pgoff_t pgoff,
1069 void *addr, size_t bytes, struct iov_iter *i)
1070{
1071 struct mapped_device *md = dax_get_private(dax_dev);
1072 sector_t sector = pgoff * PAGE_SECTORS;
1073 struct dm_target *ti;
1074 long ret = 0;
1075 int srcu_idx;
1076
1077 ti = dm_dax_get_live_target(md, sector, &srcu_idx);
1078
1079 if (!ti)
1080 goto out;
1081 if (!ti->type->dax_copy_from_iter) {
1082 ret = copy_from_iter(addr, bytes, i);
1083 goto out;
1084 }
1085 ret = ti->type->dax_copy_from_iter(ti, pgoff, addr, bytes, i);
1086 out:
1087 dm_put_live_table(md, srcu_idx);
1088
1089 return ret;
1090}
1091
1092/*
1093 * A target may call dm_accept_partial_bio only from the map routine. It is
1094 * allowed for all bio types except REQ_PREFLUSH and REQ_OP_ZONE_RESET.
1095 *
1096 * dm_accept_partial_bio informs the dm that the target only wants to process
1097 * additional n_sectors sectors of the bio and the rest of the data should be
1098 * sent in a next bio.
1099 *
1100 * A diagram that explains the arithmetics:
1101 * +--------------------+---------------+-------+
1102 * | 1 | 2 | 3 |
1103 * +--------------------+---------------+-------+
1104 *
1105 * <-------------- *tio->len_ptr --------------->
1106 * <------- bi_size ------->
1107 * <-- n_sectors -->
1108 *
1109 * Region 1 was already iterated over with bio_advance or similar function.
1110 * (it may be empty if the target doesn't use bio_advance)
1111 * Region 2 is the remaining bio size that the target wants to process.
1112 * (it may be empty if region 1 is non-empty, although there is no reason
1113 * to make it empty)
1114 * The target requires that region 3 is to be sent in the next bio.
1115 *
1116 * If the target wants to receive multiple copies of the bio (via num_*bios, etc),
1117 * the partially processed part (the sum of regions 1+2) must be the same for all
1118 * copies of the bio.
1119 */
1120void dm_accept_partial_bio(struct bio *bio, unsigned n_sectors)
1121{
1122 struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone);
1123 unsigned bi_size = bio->bi_iter.bi_size >> SECTOR_SHIFT;
1124 BUG_ON(bio->bi_opf & REQ_PREFLUSH);
1125 BUG_ON(bi_size > *tio->len_ptr);
1126 BUG_ON(n_sectors > bi_size);
1127 *tio->len_ptr -= bi_size - n_sectors;
1128 bio->bi_iter.bi_size = n_sectors << SECTOR_SHIFT;
1129}
1130EXPORT_SYMBOL_GPL(dm_accept_partial_bio);
1131
1132/*
1133 * The zone descriptors obtained with a zone report indicate
1134 * zone positions within the target device. The zone descriptors
1135 * must be remapped to match their position within the dm device.
1136 * A target may call dm_remap_zone_report after completion of a
1137 * REQ_OP_ZONE_REPORT bio to remap the zone descriptors obtained
1138 * from the target device mapping to the dm device.
1139 */
1140void dm_remap_zone_report(struct dm_target *ti, struct bio *bio, sector_t start)
1141{
1142#ifdef CONFIG_BLK_DEV_ZONED
1143 struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone);
1144 struct bio *report_bio = tio->io->orig_bio;
1145 struct blk_zone_report_hdr *hdr = NULL;
1146 struct blk_zone *zone;
1147 unsigned int nr_rep = 0;
1148 unsigned int ofst;
1149 struct bio_vec bvec;
1150 struct bvec_iter iter;
1151 void *addr;
1152
1153 if (bio->bi_status)
1154 return;
1155
1156 /*
1157 * Remap the start sector of the reported zones. For sequential zones,
1158 * also remap the write pointer position.
1159 */
1160 bio_for_each_segment(bvec, report_bio, iter) {
1161 addr = kmap_atomic(bvec.bv_page);
1162
1163 /* Remember the report header in the first page */
1164 if (!hdr) {
1165 hdr = addr;
1166 ofst = sizeof(struct blk_zone_report_hdr);
1167 } else
1168 ofst = 0;
1169
1170 /* Set zones start sector */
1171 while (hdr->nr_zones && ofst < bvec.bv_len) {
1172 zone = addr + ofst;
1173 if (zone->start >= start + ti->len) {
1174 hdr->nr_zones = 0;
1175 break;
1176 }
1177 zone->start = zone->start + ti->begin - start;
1178 if (zone->type != BLK_ZONE_TYPE_CONVENTIONAL) {
1179 if (zone->cond == BLK_ZONE_COND_FULL)
1180 zone->wp = zone->start + zone->len;
1181 else if (zone->cond == BLK_ZONE_COND_EMPTY)
1182 zone->wp = zone->start;
1183 else
1184 zone->wp = zone->wp + ti->begin - start;
1185 }
1186 ofst += sizeof(struct blk_zone);
1187 hdr->nr_zones--;
1188 nr_rep++;
1189 }
1190
1191 if (addr != hdr)
1192 kunmap_atomic(addr);
1193
1194 if (!hdr->nr_zones)
1195 break;
1196 }
1197
1198 if (hdr) {
1199 hdr->nr_zones = nr_rep;
1200 kunmap_atomic(hdr);
1201 }
1202
1203 bio_advance(report_bio, report_bio->bi_iter.bi_size);
1204
1205#else /* !CONFIG_BLK_DEV_ZONED */
1206 bio->bi_status = BLK_STS_NOTSUPP;
1207#endif
1208}
1209EXPORT_SYMBOL_GPL(dm_remap_zone_report);
1210
1211static blk_qc_t __map_bio(struct dm_target_io *tio)
1212{
1213 int r;
1214 sector_t sector;
1215 struct bio *clone = &tio->clone;
1216 struct dm_io *io = tio->io;
1217 struct mapped_device *md = io->md;
1218 struct dm_target *ti = tio->ti;
1219 blk_qc_t ret = BLK_QC_T_NONE;
1220
1221 clone->bi_end_io = clone_endio;
1222
1223 /*
1224 * Map the clone. If r == 0 we don't need to do
1225 * anything, the target has assumed ownership of
1226 * this io.
1227 */
1228 atomic_inc(&io->io_count);
1229 sector = clone->bi_iter.bi_sector;
1230
1231 r = ti->type->map(ti, clone);
1232 switch (r) {
1233 case DM_MAPIO_SUBMITTED:
1234 break;
1235 case DM_MAPIO_REMAPPED:
1236 /* the bio has been remapped so dispatch it */
1237 trace_block_bio_remap(clone->bi_disk->queue, clone,
1238 bio_dev(io->orig_bio), sector);
1239 if (md->type == DM_TYPE_NVME_BIO_BASED)
1240 ret = direct_make_request(clone);
1241 else
1242 ret = generic_make_request(clone);
1243 break;
1244 case DM_MAPIO_KILL:
1245 free_tio(tio);
1246 dec_pending(io, BLK_STS_IOERR);
1247 break;
1248 case DM_MAPIO_REQUEUE:
1249 free_tio(tio);
1250 dec_pending(io, BLK_STS_DM_REQUEUE);
1251 break;
1252 default:
1253 DMWARN("unimplemented target map return value: %d", r);
1254 BUG();
1255 }
1256
1257 return ret;
1258}
1259
1260static void bio_setup_sector(struct bio *bio, sector_t sector, unsigned len)
1261{
1262 bio->bi_iter.bi_sector = sector;
1263 bio->bi_iter.bi_size = to_bytes(len);
1264}
1265
1266/*
1267 * Creates a bio that consists of range of complete bvecs.
1268 */
1269static int clone_bio(struct dm_target_io *tio, struct bio *bio,
1270 sector_t sector, unsigned len)
1271{
1272 struct bio *clone = &tio->clone;
1273
1274 __bio_clone_fast(clone, bio);
1275
1276 if (unlikely(bio_integrity(bio) != NULL)) {
1277 int r;
1278
1279 if (unlikely(!dm_target_has_integrity(tio->ti->type) &&
1280 !dm_target_passes_integrity(tio->ti->type))) {
1281 DMWARN("%s: the target %s doesn't support integrity data.",
1282 dm_device_name(tio->io->md),
1283 tio->ti->type->name);
1284 return -EIO;
1285 }
1286
1287 r = bio_integrity_clone(clone, bio, GFP_NOIO);
1288 if (r < 0)
1289 return r;
1290 }
1291
1292 if (bio_op(bio) != REQ_OP_ZONE_REPORT)
1293 bio_advance(clone, to_bytes(sector - clone->bi_iter.bi_sector));
1294 clone->bi_iter.bi_size = to_bytes(len);
1295
1296 if (unlikely(bio_integrity(bio) != NULL))
1297 bio_integrity_trim(clone);
1298
1299 return 0;
1300}
1301
1302static void alloc_multiple_bios(struct bio_list *blist, struct clone_info *ci,
1303 struct dm_target *ti, unsigned num_bios)
1304{
1305 struct dm_target_io *tio;
1306 int try;
1307
1308 if (!num_bios)
1309 return;
1310
1311 if (num_bios == 1) {
1312 tio = alloc_tio(ci, ti, 0, GFP_NOIO);
1313 bio_list_add(blist, &tio->clone);
1314 return;
1315 }
1316
1317 for (try = 0; try < 2; try++) {
1318 int bio_nr;
1319 struct bio *bio;
1320
1321 if (try)
1322 mutex_lock(&ci->io->md->table_devices_lock);
1323 for (bio_nr = 0; bio_nr < num_bios; bio_nr++) {
1324 tio = alloc_tio(ci, ti, bio_nr, try ? GFP_NOIO : GFP_NOWAIT);
1325 if (!tio)
1326 break;
1327
1328 bio_list_add(blist, &tio->clone);
1329 }
1330 if (try)
1331 mutex_unlock(&ci->io->md->table_devices_lock);
1332 if (bio_nr == num_bios)
1333 return;
1334
1335 while ((bio = bio_list_pop(blist))) {
1336 tio = container_of(bio, struct dm_target_io, clone);
1337 free_tio(tio);
1338 }
1339 }
1340}
1341
1342static blk_qc_t __clone_and_map_simple_bio(struct clone_info *ci,
1343 struct dm_target_io *tio, unsigned *len)
1344{
1345 struct bio *clone = &tio->clone;
1346
1347 tio->len_ptr = len;
1348
1349 __bio_clone_fast(clone, ci->bio);
1350 if (len)
1351 bio_setup_sector(clone, ci->sector, *len);
1352
1353 return __map_bio(tio);
1354}
1355
1356static void __send_duplicate_bios(struct clone_info *ci, struct dm_target *ti,
1357 unsigned num_bios, unsigned *len)
1358{
1359 struct bio_list blist = BIO_EMPTY_LIST;
1360 struct bio *bio;
1361 struct dm_target_io *tio;
1362
1363 alloc_multiple_bios(&blist, ci, ti, num_bios);
1364
1365 while ((bio = bio_list_pop(&blist))) {
1366 tio = container_of(bio, struct dm_target_io, clone);
1367 (void) __clone_and_map_simple_bio(ci, tio, len);
1368 }
1369}
1370
1371static int __send_empty_flush(struct clone_info *ci)
1372{
1373 unsigned target_nr = 0;
1374 struct dm_target *ti;
1375
1376 BUG_ON(bio_has_data(ci->bio));
1377 while ((ti = dm_table_get_target(ci->map, target_nr++)))
1378 __send_duplicate_bios(ci, ti, ti->num_flush_bios, NULL);
1379
1380 return 0;
1381}
1382
1383static int __clone_and_map_data_bio(struct clone_info *ci, struct dm_target *ti,
1384 sector_t sector, unsigned *len)
1385{
1386 struct bio *bio = ci->bio;
1387 struct dm_target_io *tio;
1388 int r;
1389
1390 tio = alloc_tio(ci, ti, 0, GFP_NOIO);
1391 tio->len_ptr = len;
1392 r = clone_bio(tio, bio, sector, *len);
1393 if (r < 0) {
1394 free_tio(tio);
1395 return r;
1396 }
1397 (void) __map_bio(tio);
1398
1399 return 0;
1400}
1401
1402typedef unsigned (*get_num_bios_fn)(struct dm_target *ti);
1403
1404static unsigned get_num_discard_bios(struct dm_target *ti)
1405{
1406 return ti->num_discard_bios;
1407}
1408
1409static unsigned get_num_secure_erase_bios(struct dm_target *ti)
1410{
1411 return ti->num_secure_erase_bios;
1412}
1413
1414static unsigned get_num_write_same_bios(struct dm_target *ti)
1415{
1416 return ti->num_write_same_bios;
1417}
1418
1419static unsigned get_num_write_zeroes_bios(struct dm_target *ti)
1420{
1421 return ti->num_write_zeroes_bios;
1422}
1423
1424typedef bool (*is_split_required_fn)(struct dm_target *ti);
1425
1426static bool is_split_required_for_discard(struct dm_target *ti)
1427{
1428 return ti->split_discard_bios;
1429}
1430
1431static int __send_changing_extent_only(struct clone_info *ci, struct dm_target *ti,
1432 get_num_bios_fn get_num_bios,
1433 is_split_required_fn is_split_required)
1434{
1435 unsigned len;
1436 unsigned num_bios;
1437
1438 /*
1439 * Even though the device advertised support for this type of
1440 * request, that does not mean every target supports it, and
1441 * reconfiguration might also have changed that since the
1442 * check was performed.
1443 */
1444 num_bios = get_num_bios ? get_num_bios(ti) : 0;
1445 if (!num_bios)
1446 return -EOPNOTSUPP;
1447
1448 if (is_split_required && !is_split_required(ti))
1449 len = min((sector_t)ci->sector_count, max_io_len_target_boundary(ci->sector, ti));
1450 else
1451 len = min((sector_t)ci->sector_count, max_io_len(ci->sector, ti));
1452
1453 __send_duplicate_bios(ci, ti, num_bios, &len);
1454
1455 ci->sector += len;
1456 ci->sector_count -= len;
1457
1458 return 0;
1459}
1460
1461static int __send_discard(struct clone_info *ci, struct dm_target *ti)
1462{
1463 return __send_changing_extent_only(ci, ti, get_num_discard_bios,
1464 is_split_required_for_discard);
1465}
1466
1467static int __send_secure_erase(struct clone_info *ci, struct dm_target *ti)
1468{
1469 return __send_changing_extent_only(ci, ti, get_num_secure_erase_bios, NULL);
1470}
1471
1472static int __send_write_same(struct clone_info *ci, struct dm_target *ti)
1473{
1474 return __send_changing_extent_only(ci, ti, get_num_write_same_bios, NULL);
1475}
1476
1477static int __send_write_zeroes(struct clone_info *ci, struct dm_target *ti)
1478{
1479 return __send_changing_extent_only(ci, ti, get_num_write_zeroes_bios, NULL);
1480}
1481
1482static bool __process_abnormal_io(struct clone_info *ci, struct dm_target *ti,
1483 int *result)
1484{
1485 struct bio *bio = ci->bio;
1486
1487 if (bio_op(bio) == REQ_OP_DISCARD)
1488 *result = __send_discard(ci, ti);
1489 else if (bio_op(bio) == REQ_OP_SECURE_ERASE)
1490 *result = __send_secure_erase(ci, ti);
1491 else if (bio_op(bio) == REQ_OP_WRITE_SAME)
1492 *result = __send_write_same(ci, ti);
1493 else if (bio_op(bio) == REQ_OP_WRITE_ZEROES)
1494 *result = __send_write_zeroes(ci, ti);
1495 else
1496 return false;
1497
1498 return true;
1499}
1500
1501/*
1502 * Select the correct strategy for processing a non-flush bio.
1503 */
1504static int __split_and_process_non_flush(struct clone_info *ci)
1505{
1506 struct bio *bio = ci->bio;
1507 struct dm_target *ti;
1508 unsigned len;
1509 int r;
1510
1511 ti = dm_table_find_target(ci->map, ci->sector);
1512 if (!dm_target_is_valid(ti))
1513 return -EIO;
1514
1515 if (unlikely(__process_abnormal_io(ci, ti, &r)))
1516 return r;
1517
1518 if (bio_op(bio) == REQ_OP_ZONE_REPORT)
1519 len = ci->sector_count;
1520 else
1521 len = min_t(sector_t, max_io_len(ci->sector, ti),
1522 ci->sector_count);
1523
1524 r = __clone_and_map_data_bio(ci, ti, ci->sector, &len);
1525 if (r < 0)
1526 return r;
1527
1528 ci->sector += len;
1529 ci->sector_count -= len;
1530
1531 return 0;
1532}
1533
1534static void init_clone_info(struct clone_info *ci, struct mapped_device *md,
1535 struct dm_table *map, struct bio *bio)
1536{
1537 ci->map = map;
1538 ci->io = alloc_io(md, bio);
1539 ci->sector = bio->bi_iter.bi_sector;
1540}
1541
1542/*
1543 * Entry point to split a bio into clones and submit them to the targets.
1544 */
1545static blk_qc_t __split_and_process_bio(struct mapped_device *md,
1546 struct dm_table *map, struct bio *bio)
1547{
1548 struct clone_info ci;
1549 blk_qc_t ret = BLK_QC_T_NONE;
1550 int error = 0;
1551
1552 if (unlikely(!map)) {
1553 bio_io_error(bio);
1554 return ret;
1555 }
1556
1557 init_clone_info(&ci, md, map, bio);
1558
1559 if (bio->bi_opf & REQ_PREFLUSH) {
1560 ci.bio = &ci.io->md->flush_bio;
1561 ci.sector_count = 0;
1562 error = __send_empty_flush(&ci);
1563 /* dec_pending submits any data associated with flush */
1564 } else if (bio_op(bio) == REQ_OP_ZONE_RESET) {
1565 ci.bio = bio;
1566 ci.sector_count = 0;
1567 error = __split_and_process_non_flush(&ci);
1568 } else {
1569 ci.bio = bio;
1570 ci.sector_count = bio_sectors(bio);
1571 while (ci.sector_count && !error) {
1572 error = __split_and_process_non_flush(&ci);
1573 if (current->bio_list && ci.sector_count && !error) {
1574 /*
1575 * Remainder must be passed to generic_make_request()
1576 * so that it gets handled *after* bios already submitted
1577 * have been completely processed.
1578 * We take a clone of the original to store in
1579 * ci.io->orig_bio to be used by end_io_acct() and
1580 * for dec_pending to use for completion handling.
1581 * As this path is not used for REQ_OP_ZONE_REPORT,
1582 * the usage of io->orig_bio in dm_remap_zone_report()
1583 * won't be affected by this reassignment.
1584 */
1585 struct bio *b = bio_clone_bioset(bio, GFP_NOIO,
1586 md->queue->bio_split);
1587 ci.io->orig_bio = b;
1588 bio_advance(bio, (bio_sectors(bio) - ci.sector_count) << 9);
1589 bio_chain(b, bio);
1590 ret = generic_make_request(bio);
1591 break;
1592 }
1593 }
1594 }
1595
1596 /* drop the extra reference count */
1597 dec_pending(ci.io, errno_to_blk_status(error));
1598 return ret;
1599}
1600
1601/*
1602 * Optimized variant of __split_and_process_bio that leverages the
1603 * fact that targets that use it do _not_ have a need to split bios.
1604 */
1605static blk_qc_t __process_bio(struct mapped_device *md,
1606 struct dm_table *map, struct bio *bio)
1607{
1608 struct clone_info ci;
1609 blk_qc_t ret = BLK_QC_T_NONE;
1610 int error = 0;
1611
1612 if (unlikely(!map)) {
1613 bio_io_error(bio);
1614 return ret;
1615 }
1616
1617 init_clone_info(&ci, md, map, bio);
1618
1619 if (bio->bi_opf & REQ_PREFLUSH) {
1620 ci.bio = &ci.io->md->flush_bio;
1621 ci.sector_count = 0;
1622 error = __send_empty_flush(&ci);
1623 /* dec_pending submits any data associated with flush */
1624 } else {
1625 struct dm_target *ti = md->immutable_target;
1626 struct dm_target_io *tio;
1627
1628 /*
1629 * Defend against IO still getting in during teardown
1630 * - as was seen for a time with nvme-fcloop
1631 */
1632 if (unlikely(WARN_ON_ONCE(!ti || !dm_target_is_valid(ti)))) {
1633 error = -EIO;
1634 goto out;
1635 }
1636
1637 ci.bio = bio;
1638 ci.sector_count = bio_sectors(bio);
1639 if (unlikely(__process_abnormal_io(&ci, ti, &error)))
1640 goto out;
1641
1642 tio = alloc_tio(&ci, ti, 0, GFP_NOIO);
1643 ret = __clone_and_map_simple_bio(&ci, tio, NULL);
1644 }
1645out:
1646 /* drop the extra reference count */
1647 dec_pending(ci.io, errno_to_blk_status(error));
1648 return ret;
1649}
1650
1651typedef blk_qc_t (process_bio_fn)(struct mapped_device *, struct dm_table *, struct bio *);
1652
1653static blk_qc_t __dm_make_request(struct request_queue *q, struct bio *bio,
1654 process_bio_fn process_bio)
1655{
1656 struct mapped_device *md = q->queuedata;
1657 blk_qc_t ret = BLK_QC_T_NONE;
1658 int srcu_idx;
1659 struct dm_table *map;
1660
1661 map = dm_get_live_table(md, &srcu_idx);
1662
1663 /* if we're suspended, we have to queue this io for later */
1664 if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags))) {
1665 dm_put_live_table(md, srcu_idx);
1666
1667 if (!(bio->bi_opf & REQ_RAHEAD))
1668 queue_io(md, bio);
1669 else
1670 bio_io_error(bio);
1671 return ret;
1672 }
1673
1674 ret = process_bio(md, map, bio);
1675
1676 dm_put_live_table(md, srcu_idx);
1677 return ret;
1678}
1679
1680/*
1681 * The request function that remaps the bio to one target and
1682 * splits off any remainder.
1683 */
1684static blk_qc_t dm_make_request(struct request_queue *q, struct bio *bio)
1685{
1686 return __dm_make_request(q, bio, __split_and_process_bio);
1687}
1688
1689static blk_qc_t dm_make_request_nvme(struct request_queue *q, struct bio *bio)
1690{
1691 return __dm_make_request(q, bio, __process_bio);
1692}
1693
1694static int dm_any_congested(void *congested_data, int bdi_bits)
1695{
1696 int r = bdi_bits;
1697 struct mapped_device *md = congested_data;
1698 struct dm_table *map;
1699
1700 if (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
1701 if (dm_request_based(md)) {
1702 /*
1703 * With request-based DM we only need to check the
1704 * top-level queue for congestion.
1705 */
1706 r = md->queue->backing_dev_info->wb.state & bdi_bits;
1707 } else {
1708 map = dm_get_live_table_fast(md);
1709 if (map)
1710 r = dm_table_any_congested(map, bdi_bits);
1711 dm_put_live_table_fast(md);
1712 }
1713 }
1714
1715 return r;
1716}
1717
1718/*-----------------------------------------------------------------
1719 * An IDR is used to keep track of allocated minor numbers.
1720 *---------------------------------------------------------------*/
1721static void free_minor(int minor)
1722{
1723 spin_lock(&_minor_lock);
1724 idr_remove(&_minor_idr, minor);
1725 spin_unlock(&_minor_lock);
1726}
1727
1728/*
1729 * See if the device with a specific minor # is free.
1730 */
1731static int specific_minor(int minor)
1732{
1733 int r;
1734
1735 if (minor >= (1 << MINORBITS))
1736 return -EINVAL;
1737
1738 idr_preload(GFP_KERNEL);
1739 spin_lock(&_minor_lock);
1740
1741 r = idr_alloc(&_minor_idr, MINOR_ALLOCED, minor, minor + 1, GFP_NOWAIT);
1742
1743 spin_unlock(&_minor_lock);
1744 idr_preload_end();
1745 if (r < 0)
1746 return r == -ENOSPC ? -EBUSY : r;
1747 return 0;
1748}
1749
1750static int next_free_minor(int *minor)
1751{
1752 int r;
1753
1754 idr_preload(GFP_KERNEL);
1755 spin_lock(&_minor_lock);
1756
1757 r = idr_alloc(&_minor_idr, MINOR_ALLOCED, 0, 1 << MINORBITS, GFP_NOWAIT);
1758
1759 spin_unlock(&_minor_lock);
1760 idr_preload_end();
1761 if (r < 0)
1762 return r;
1763 *minor = r;
1764 return 0;
1765}
1766
1767static const struct block_device_operations dm_blk_dops;
1768static const struct dax_operations dm_dax_ops;
1769
1770static void dm_wq_work(struct work_struct *work);
1771
1772static void dm_init_normal_md_queue(struct mapped_device *md)
1773{
1774 md->use_blk_mq = false;
1775
1776 /*
1777 * Initialize aspects of queue that aren't relevant for blk-mq
1778 */
1779 md->queue->backing_dev_info->congested_fn = dm_any_congested;
1780}
1781
1782static void cleanup_mapped_device(struct mapped_device *md)
1783{
1784 if (md->wq)
1785 destroy_workqueue(md->wq);
1786 if (md->kworker_task)
1787 kthread_stop(md->kworker_task);
1788 if (md->bs)
1789 bioset_free(md->bs);
1790 if (md->io_bs)
1791 bioset_free(md->io_bs);
1792
1793 if (md->dax_dev) {
1794 kill_dax(md->dax_dev);
1795 put_dax(md->dax_dev);
1796 md->dax_dev = NULL;
1797 }
1798
1799 if (md->disk) {
1800 spin_lock(&_minor_lock);
1801 md->disk->private_data = NULL;
1802 spin_unlock(&_minor_lock);
1803 del_gendisk(md->disk);
1804 put_disk(md->disk);
1805 }
1806
1807 if (md->queue)
1808 blk_cleanup_queue(md->queue);
1809
1810 cleanup_srcu_struct(&md->io_barrier);
1811
1812 if (md->bdev) {
1813 bdput(md->bdev);
1814 md->bdev = NULL;
1815 }
1816
1817 mutex_destroy(&md->suspend_lock);
1818 mutex_destroy(&md->type_lock);
1819 mutex_destroy(&md->table_devices_lock);
1820
1821 dm_mq_cleanup_mapped_device(md);
1822}
1823
1824/*
1825 * Allocate and initialise a blank device with a given minor.
1826 */
1827static struct mapped_device *alloc_dev(int minor)
1828{
1829 int r, numa_node_id = dm_get_numa_node();
1830 struct dax_device *dax_dev = NULL;
1831 struct mapped_device *md;
1832 void *old_md;
1833
1834 md = kvzalloc_node(sizeof(*md), GFP_KERNEL, numa_node_id);
1835 if (!md) {
1836 DMWARN("unable to allocate device, out of memory.");
1837 return NULL;
1838 }
1839
1840 if (!try_module_get(THIS_MODULE))
1841 goto bad_module_get;
1842
1843 /* get a minor number for the dev */
1844 if (minor == DM_ANY_MINOR)
1845 r = next_free_minor(&minor);
1846 else
1847 r = specific_minor(minor);
1848 if (r < 0)
1849 goto bad_minor;
1850
1851 r = init_srcu_struct(&md->io_barrier);
1852 if (r < 0)
1853 goto bad_io_barrier;
1854
1855 md->numa_node_id = numa_node_id;
1856 md->use_blk_mq = dm_use_blk_mq_default();
1857 md->init_tio_pdu = false;
1858 md->type = DM_TYPE_NONE;
1859 mutex_init(&md->suspend_lock);
1860 mutex_init(&md->type_lock);
1861 mutex_init(&md->table_devices_lock);
1862 spin_lock_init(&md->deferred_lock);
1863 atomic_set(&md->holders, 1);
1864 atomic_set(&md->open_count, 0);
1865 atomic_set(&md->event_nr, 0);
1866 atomic_set(&md->uevent_seq, 0);
1867 INIT_LIST_HEAD(&md->uevent_list);
1868 INIT_LIST_HEAD(&md->table_devices);
1869 spin_lock_init(&md->uevent_lock);
1870
1871 md->queue = blk_alloc_queue_node(GFP_KERNEL, numa_node_id, NULL);
1872 if (!md->queue)
1873 goto bad;
1874 md->queue->queuedata = md;
1875 md->queue->backing_dev_info->congested_data = md;
1876
1877 md->disk = alloc_disk_node(1, md->numa_node_id);
1878 if (!md->disk)
1879 goto bad;
1880
1881 atomic_set(&md->pending[0], 0);
1882 atomic_set(&md->pending[1], 0);
1883 init_waitqueue_head(&md->wait);
1884 INIT_WORK(&md->work, dm_wq_work);
1885 init_waitqueue_head(&md->eventq);
1886 init_completion(&md->kobj_holder.completion);
1887 md->kworker_task = NULL;
1888
1889 md->disk->major = _major;
1890 md->disk->first_minor = minor;
1891 md->disk->fops = &dm_blk_dops;
1892 md->disk->queue = md->queue;
1893 md->disk->private_data = md;
1894 sprintf(md->disk->disk_name, "dm-%d", minor);
1895
1896 if (IS_ENABLED(CONFIG_DAX_DRIVER)) {
1897 dax_dev = alloc_dax(md, md->disk->disk_name, &dm_dax_ops);
1898 if (!dax_dev)
1899 goto bad;
1900 }
1901 md->dax_dev = dax_dev;
1902
1903 add_disk_no_queue_reg(md->disk);
1904 format_dev_t(md->name, MKDEV(_major, minor));
1905
1906 md->wq = alloc_workqueue("kdmflush", WQ_MEM_RECLAIM, 0);
1907 if (!md->wq)
1908 goto bad;
1909
1910 md->bdev = bdget_disk(md->disk, 0);
1911 if (!md->bdev)
1912 goto bad;
1913
1914 bio_init(&md->flush_bio, NULL, 0);
1915 bio_set_dev(&md->flush_bio, md->bdev);
1916 md->flush_bio.bi_opf = REQ_OP_WRITE | REQ_PREFLUSH | REQ_SYNC;
1917
1918 dm_stats_init(&md->stats);
1919
1920 /* Populate the mapping, nobody knows we exist yet */
1921 spin_lock(&_minor_lock);
1922 old_md = idr_replace(&_minor_idr, md, minor);
1923 spin_unlock(&_minor_lock);
1924
1925 BUG_ON(old_md != MINOR_ALLOCED);
1926
1927 return md;
1928
1929bad:
1930 cleanup_mapped_device(md);
1931bad_io_barrier:
1932 free_minor(minor);
1933bad_minor:
1934 module_put(THIS_MODULE);
1935bad_module_get:
1936 kvfree(md);
1937 return NULL;
1938}
1939
1940static void unlock_fs(struct mapped_device *md);
1941
1942static void free_dev(struct mapped_device *md)
1943{
1944 int minor = MINOR(disk_devt(md->disk));
1945
1946 unlock_fs(md);
1947
1948 cleanup_mapped_device(md);
1949
1950 free_table_devices(&md->table_devices);
1951 dm_stats_cleanup(&md->stats);
1952 free_minor(minor);
1953
1954 module_put(THIS_MODULE);
1955 kvfree(md);
1956}
1957
1958static void __bind_mempools(struct mapped_device *md, struct dm_table *t)
1959{
1960 struct dm_md_mempools *p = dm_table_get_md_mempools(t);
1961
1962 if (dm_table_bio_based(t)) {
1963 /*
1964 * The md may already have mempools that need changing.
1965 * If so, reload bioset because front_pad may have changed
1966 * because a different table was loaded.
1967 */
1968 if (md->bs) {
1969 bioset_free(md->bs);
1970 md->bs = NULL;
1971 }
1972 if (md->io_bs) {
1973 bioset_free(md->io_bs);
1974 md->io_bs = NULL;
1975 }
1976
1977 } else if (md->bs) {
1978 /*
1979 * There's no need to reload with request-based dm
1980 * because the size of front_pad doesn't change.
1981 * Note for future: If you are to reload bioset,
1982 * prep-ed requests in the queue may refer
1983 * to bio from the old bioset, so you must walk
1984 * through the queue to unprep.
1985 */
1986 goto out;
1987 }
1988
1989 BUG_ON(!p || md->bs || md->io_bs);
1990
1991 md->bs = p->bs;
1992 p->bs = NULL;
1993 md->io_bs = p->io_bs;
1994 p->io_bs = NULL;
1995out:
1996 /* mempool bind completed, no longer need any mempools in the table */
1997 dm_table_free_md_mempools(t);
1998}
1999
2000/*
2001 * Bind a table to the device.
2002 */
2003static void event_callback(void *context)
2004{
2005 unsigned long flags;
2006 LIST_HEAD(uevents);
2007 struct mapped_device *md = (struct mapped_device *) context;
2008
2009 spin_lock_irqsave(&md->uevent_lock, flags);
2010 list_splice_init(&md->uevent_list, &uevents);
2011 spin_unlock_irqrestore(&md->uevent_lock, flags);
2012
2013 dm_send_uevents(&uevents, &disk_to_dev(md->disk)->kobj);
2014
2015 atomic_inc(&md->event_nr);
2016 wake_up(&md->eventq);
2017 dm_issue_global_event();
2018}
2019
2020/*
2021 * Protected by md->suspend_lock obtained by dm_swap_table().
2022 */
2023static void __set_size(struct mapped_device *md, sector_t size)
2024{
2025 lockdep_assert_held(&md->suspend_lock);
2026
2027 set_capacity(md->disk, size);
2028
2029 i_size_write(md->bdev->bd_inode, (loff_t)size << SECTOR_SHIFT);
2030}
2031
2032/*
2033 * Returns old map, which caller must destroy.
2034 */
2035static struct dm_table *__bind(struct mapped_device *md, struct dm_table *t,
2036 struct queue_limits *limits)
2037{
2038 struct dm_table *old_map;
2039 struct request_queue *q = md->queue;
2040 bool request_based = dm_table_request_based(t);
2041 sector_t size;
2042
2043 lockdep_assert_held(&md->suspend_lock);
2044
2045 size = dm_table_get_size(t);
2046
2047 /*
2048 * Wipe any geometry if the size of the table changed.
2049 */
2050 if (size != dm_get_size(md))
2051 memset(&md->geometry, 0, sizeof(md->geometry));
2052
2053 __set_size(md, size);
2054
2055 dm_table_event_callback(t, event_callback, md);
2056
2057 /*
2058 * The queue hasn't been stopped yet, if the old table type wasn't
2059 * for request-based during suspension. So stop it to prevent
2060 * I/O mapping before resume.
2061 * This must be done before setting the queue restrictions,
2062 * because request-based dm may be run just after the setting.
2063 */
2064 if (request_based)
2065 dm_stop_queue(q);
2066
2067 if (request_based || md->type == DM_TYPE_NVME_BIO_BASED) {
2068 /*
2069 * Leverage the fact that request-based DM targets and
2070 * NVMe bio based targets are immutable singletons
2071 * - used to optimize both dm_request_fn and dm_mq_queue_rq;
2072 * and __process_bio.
2073 */
2074 md->immutable_target = dm_table_get_immutable_target(t);
2075 }
2076
2077 __bind_mempools(md, t);
2078
2079 old_map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2080 rcu_assign_pointer(md->map, (void *)t);
2081 md->immutable_target_type = dm_table_get_immutable_target_type(t);
2082
2083 dm_table_set_restrictions(t, q, limits);
2084 if (old_map)
2085 dm_sync_table(md);
2086
2087 return old_map;
2088}
2089
2090/*
2091 * Returns unbound table for the caller to free.
2092 */
2093static struct dm_table *__unbind(struct mapped_device *md)
2094{
2095 struct dm_table *map = rcu_dereference_protected(md->map, 1);
2096
2097 if (!map)
2098 return NULL;
2099
2100 dm_table_event_callback(map, NULL, NULL);
2101 RCU_INIT_POINTER(md->map, NULL);
2102 dm_sync_table(md);
2103
2104 return map;
2105}
2106
2107/*
2108 * Constructor for a new device.
2109 */
2110int dm_create(int minor, struct mapped_device **result)
2111{
2112 int r;
2113 struct mapped_device *md;
2114
2115 md = alloc_dev(minor);
2116 if (!md)
2117 return -ENXIO;
2118
2119 r = dm_sysfs_init(md);
2120 if (r) {
2121 free_dev(md);
2122 return r;
2123 }
2124
2125 *result = md;
2126 return 0;
2127}
2128
2129/*
2130 * Functions to manage md->type.
2131 * All are required to hold md->type_lock.
2132 */
2133void dm_lock_md_type(struct mapped_device *md)
2134{
2135 mutex_lock(&md->type_lock);
2136}
2137
2138void dm_unlock_md_type(struct mapped_device *md)
2139{
2140 mutex_unlock(&md->type_lock);
2141}
2142
2143void dm_set_md_type(struct mapped_device *md, enum dm_queue_mode type)
2144{
2145 BUG_ON(!mutex_is_locked(&md->type_lock));
2146 md->type = type;
2147}
2148
2149enum dm_queue_mode dm_get_md_type(struct mapped_device *md)
2150{
2151 return md->type;
2152}
2153
2154struct target_type *dm_get_immutable_target_type(struct mapped_device *md)
2155{
2156 return md->immutable_target_type;
2157}
2158
2159/*
2160 * The queue_limits are only valid as long as you have a reference
2161 * count on 'md'.
2162 */
2163struct queue_limits *dm_get_queue_limits(struct mapped_device *md)
2164{
2165 BUG_ON(!atomic_read(&md->holders));
2166 return &md->queue->limits;
2167}
2168EXPORT_SYMBOL_GPL(dm_get_queue_limits);
2169
2170/*
2171 * Setup the DM device's queue based on md's type
2172 */
2173int dm_setup_md_queue(struct mapped_device *md, struct dm_table *t)
2174{
2175 int r;
2176 struct queue_limits limits;
2177 enum dm_queue_mode type = dm_get_md_type(md);
2178
2179 switch (type) {
2180 case DM_TYPE_REQUEST_BASED:
2181 dm_init_normal_md_queue(md);
2182 r = dm_old_init_request_queue(md, t);
2183 if (r) {
2184 DMERR("Cannot initialize queue for request-based mapped device");
2185 return r;
2186 }
2187 break;
2188 case DM_TYPE_MQ_REQUEST_BASED:
2189 r = dm_mq_init_request_queue(md, t);
2190 if (r) {
2191 DMERR("Cannot initialize queue for request-based dm-mq mapped device");
2192 return r;
2193 }
2194 break;
2195 case DM_TYPE_BIO_BASED:
2196 case DM_TYPE_DAX_BIO_BASED:
2197 dm_init_normal_md_queue(md);
2198 blk_queue_make_request(md->queue, dm_make_request);
2199 break;
2200 case DM_TYPE_NVME_BIO_BASED:
2201 dm_init_normal_md_queue(md);
2202 blk_queue_make_request(md->queue, dm_make_request_nvme);
2203 break;
2204 case DM_TYPE_NONE:
2205 WARN_ON_ONCE(true);
2206 break;
2207 }
2208
2209 r = dm_calculate_queue_limits(t, &limits);
2210 if (r) {
2211 DMERR("Cannot calculate initial queue limits");
2212 return r;
2213 }
2214 dm_table_set_restrictions(t, md->queue, &limits);
2215 blk_register_queue(md->disk);
2216
2217 return 0;
2218}
2219
2220struct mapped_device *dm_get_md(dev_t dev)
2221{
2222 struct mapped_device *md;
2223 unsigned minor = MINOR(dev);
2224
2225 if (MAJOR(dev) != _major || minor >= (1 << MINORBITS))
2226 return NULL;
2227
2228 spin_lock(&_minor_lock);
2229
2230 md = idr_find(&_minor_idr, minor);
2231 if (!md || md == MINOR_ALLOCED || (MINOR(disk_devt(dm_disk(md))) != minor) ||
2232 test_bit(DMF_FREEING, &md->flags) || dm_deleting_md(md)) {
2233 md = NULL;
2234 goto out;
2235 }
2236 dm_get(md);
2237out:
2238 spin_unlock(&_minor_lock);
2239
2240 return md;
2241}
2242EXPORT_SYMBOL_GPL(dm_get_md);
2243
2244void *dm_get_mdptr(struct mapped_device *md)
2245{
2246 return md->interface_ptr;
2247}
2248
2249void dm_set_mdptr(struct mapped_device *md, void *ptr)
2250{
2251 md->interface_ptr = ptr;
2252}
2253
2254void dm_get(struct mapped_device *md)
2255{
2256 atomic_inc(&md->holders);
2257 BUG_ON(test_bit(DMF_FREEING, &md->flags));
2258}
2259
2260int dm_hold(struct mapped_device *md)
2261{
2262 spin_lock(&_minor_lock);
2263 if (test_bit(DMF_FREEING, &md->flags)) {
2264 spin_unlock(&_minor_lock);
2265 return -EBUSY;
2266 }
2267 dm_get(md);
2268 spin_unlock(&_minor_lock);
2269 return 0;
2270}
2271EXPORT_SYMBOL_GPL(dm_hold);
2272
2273const char *dm_device_name(struct mapped_device *md)
2274{
2275 return md->name;
2276}
2277EXPORT_SYMBOL_GPL(dm_device_name);
2278
2279static void __dm_destroy(struct mapped_device *md, bool wait)
2280{
2281 struct dm_table *map;
2282 int srcu_idx;
2283
2284 might_sleep();
2285
2286 spin_lock(&_minor_lock);
2287 idr_replace(&_minor_idr, MINOR_ALLOCED, MINOR(disk_devt(dm_disk(md))));
2288 set_bit(DMF_FREEING, &md->flags);
2289 spin_unlock(&_minor_lock);
2290
2291 blk_set_queue_dying(md->queue);
2292
2293 if (dm_request_based(md) && md->kworker_task)
2294 kthread_flush_worker(&md->kworker);
2295
2296 /*
2297 * Take suspend_lock so that presuspend and postsuspend methods
2298 * do not race with internal suspend.
2299 */
2300 mutex_lock(&md->suspend_lock);
2301 map = dm_get_live_table(md, &srcu_idx);
2302 if (!dm_suspended_md(md)) {
2303 dm_table_presuspend_targets(map);
2304 dm_table_postsuspend_targets(map);
2305 }
2306 /* dm_put_live_table must be before msleep, otherwise deadlock is possible */
2307 dm_put_live_table(md, srcu_idx);
2308 mutex_unlock(&md->suspend_lock);
2309
2310 /*
2311 * Rare, but there may be I/O requests still going to complete,
2312 * for example. Wait for all references to disappear.
2313 * No one should increment the reference count of the mapped_device,
2314 * after the mapped_device state becomes DMF_FREEING.
2315 */
2316 if (wait)
2317 while (atomic_read(&md->holders))
2318 msleep(1);
2319 else if (atomic_read(&md->holders))
2320 DMWARN("%s: Forcibly removing mapped_device still in use! (%d users)",
2321 dm_device_name(md), atomic_read(&md->holders));
2322
2323 dm_sysfs_exit(md);
2324 dm_table_destroy(__unbind(md));
2325 free_dev(md);
2326}
2327
2328void dm_destroy(struct mapped_device *md)
2329{
2330 __dm_destroy(md, true);
2331}
2332
2333void dm_destroy_immediate(struct mapped_device *md)
2334{
2335 __dm_destroy(md, false);
2336}
2337
2338void dm_put(struct mapped_device *md)
2339{
2340 atomic_dec(&md->holders);
2341}
2342EXPORT_SYMBOL_GPL(dm_put);
2343
2344static int dm_wait_for_completion(struct mapped_device *md, long task_state)
2345{
2346 int r = 0;
2347 DEFINE_WAIT(wait);
2348
2349 while (1) {
2350 prepare_to_wait(&md->wait, &wait, task_state);
2351
2352 if (!md_in_flight(md))
2353 break;
2354
2355 if (signal_pending_state(task_state, current)) {
2356 r = -EINTR;
2357 break;
2358 }
2359
2360 io_schedule();
2361 }
2362 finish_wait(&md->wait, &wait);
2363
2364 return r;
2365}
2366
2367/*
2368 * Process the deferred bios
2369 */
2370static void dm_wq_work(struct work_struct *work)
2371{
2372 struct mapped_device *md = container_of(work, struct mapped_device,
2373 work);
2374 struct bio *c;
2375 int srcu_idx;
2376 struct dm_table *map;
2377
2378 map = dm_get_live_table(md, &srcu_idx);
2379
2380 while (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
2381 spin_lock_irq(&md->deferred_lock);
2382 c = bio_list_pop(&md->deferred);
2383 spin_unlock_irq(&md->deferred_lock);
2384
2385 if (!c)
2386 break;
2387
2388 if (dm_request_based(md))
2389 generic_make_request(c);
2390 else
2391 __split_and_process_bio(md, map, c);
2392 }
2393
2394 dm_put_live_table(md, srcu_idx);
2395}
2396
2397static void dm_queue_flush(struct mapped_device *md)
2398{
2399 clear_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2400 smp_mb__after_atomic();
2401 queue_work(md->wq, &md->work);
2402}
2403
2404/*
2405 * Swap in a new table, returning the old one for the caller to destroy.
2406 */
2407struct dm_table *dm_swap_table(struct mapped_device *md, struct dm_table *table)
2408{
2409 struct dm_table *live_map = NULL, *map = ERR_PTR(-EINVAL);
2410 struct queue_limits limits;
2411 int r;
2412
2413 mutex_lock(&md->suspend_lock);
2414
2415 /* device must be suspended */
2416 if (!dm_suspended_md(md))
2417 goto out;
2418
2419 /*
2420 * If the new table has no data devices, retain the existing limits.
2421 * This helps multipath with queue_if_no_path if all paths disappear,
2422 * then new I/O is queued based on these limits, and then some paths
2423 * reappear.
2424 */
2425 if (dm_table_has_no_data_devices(table)) {
2426 live_map = dm_get_live_table_fast(md);
2427 if (live_map)
2428 limits = md->queue->limits;
2429 dm_put_live_table_fast(md);
2430 }
2431
2432 if (!live_map) {
2433 r = dm_calculate_queue_limits(table, &limits);
2434 if (r) {
2435 map = ERR_PTR(r);
2436 goto out;
2437 }
2438 }
2439
2440 map = __bind(md, table, &limits);
2441 dm_issue_global_event();
2442
2443out:
2444 mutex_unlock(&md->suspend_lock);
2445 return map;
2446}
2447
2448/*
2449 * Functions to lock and unlock any filesystem running on the
2450 * device.
2451 */
2452static int lock_fs(struct mapped_device *md)
2453{
2454 int r;
2455
2456 WARN_ON(md->frozen_sb);
2457
2458 md->frozen_sb = freeze_bdev(md->bdev);
2459 if (IS_ERR(md->frozen_sb)) {
2460 r = PTR_ERR(md->frozen_sb);
2461 md->frozen_sb = NULL;
2462 return r;
2463 }
2464
2465 set_bit(DMF_FROZEN, &md->flags);
2466
2467 return 0;
2468}
2469
2470static void unlock_fs(struct mapped_device *md)
2471{
2472 if (!test_bit(DMF_FROZEN, &md->flags))
2473 return;
2474
2475 thaw_bdev(md->bdev, md->frozen_sb);
2476 md->frozen_sb = NULL;
2477 clear_bit(DMF_FROZEN, &md->flags);
2478}
2479
2480/*
2481 * @suspend_flags: DM_SUSPEND_LOCKFS_FLAG and/or DM_SUSPEND_NOFLUSH_FLAG
2482 * @task_state: e.g. TASK_INTERRUPTIBLE or TASK_UNINTERRUPTIBLE
2483 * @dmf_suspended_flag: DMF_SUSPENDED or DMF_SUSPENDED_INTERNALLY
2484 *
2485 * If __dm_suspend returns 0, the device is completely quiescent
2486 * now. There is no request-processing activity. All new requests
2487 * are being added to md->deferred list.
2488 */
2489static int __dm_suspend(struct mapped_device *md, struct dm_table *map,
2490 unsigned suspend_flags, long task_state,
2491 int dmf_suspended_flag)
2492{
2493 bool do_lockfs = suspend_flags & DM_SUSPEND_LOCKFS_FLAG;
2494 bool noflush = suspend_flags & DM_SUSPEND_NOFLUSH_FLAG;
2495 int r;
2496
2497 lockdep_assert_held(&md->suspend_lock);
2498
2499 /*
2500 * DMF_NOFLUSH_SUSPENDING must be set before presuspend.
2501 * This flag is cleared before dm_suspend returns.
2502 */
2503 if (noflush)
2504 set_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
2505 else
2506 pr_debug("%s: suspending with flush\n", dm_device_name(md));
2507
2508 /*
2509 * This gets reverted if there's an error later and the targets
2510 * provide the .presuspend_undo hook.
2511 */
2512 dm_table_presuspend_targets(map);
2513
2514 /*
2515 * Flush I/O to the device.
2516 * Any I/O submitted after lock_fs() may not be flushed.
2517 * noflush takes precedence over do_lockfs.
2518 * (lock_fs() flushes I/Os and waits for them to complete.)
2519 */
2520 if (!noflush && do_lockfs) {
2521 r = lock_fs(md);
2522 if (r) {
2523 dm_table_presuspend_undo_targets(map);
2524 return r;
2525 }
2526 }
2527
2528 /*
2529 * Here we must make sure that no processes are submitting requests
2530 * to target drivers i.e. no one may be executing
2531 * __split_and_process_bio. This is called from dm_request and
2532 * dm_wq_work.
2533 *
2534 * To get all processes out of __split_and_process_bio in dm_request,
2535 * we take the write lock. To prevent any process from reentering
2536 * __split_and_process_bio from dm_request and quiesce the thread
2537 * (dm_wq_work), we set BMF_BLOCK_IO_FOR_SUSPEND and call
2538 * flush_workqueue(md->wq).
2539 */
2540 set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2541 if (map)
2542 synchronize_srcu(&md->io_barrier);
2543
2544 /*
2545 * Stop md->queue before flushing md->wq in case request-based
2546 * dm defers requests to md->wq from md->queue.
2547 */
2548 if (dm_request_based(md)) {
2549 dm_stop_queue(md->queue);
2550 if (md->kworker_task)
2551 kthread_flush_worker(&md->kworker);
2552 }
2553
2554 flush_workqueue(md->wq);
2555
2556 /*
2557 * At this point no more requests are entering target request routines.
2558 * We call dm_wait_for_completion to wait for all existing requests
2559 * to finish.
2560 */
2561 r = dm_wait_for_completion(md, task_state);
2562 if (!r)
2563 set_bit(dmf_suspended_flag, &md->flags);
2564
2565 if (noflush)
2566 clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
2567 if (map)
2568 synchronize_srcu(&md->io_barrier);
2569
2570 /* were we interrupted ? */
2571 if (r < 0) {
2572 dm_queue_flush(md);
2573
2574 if (dm_request_based(md))
2575 dm_start_queue(md->queue);
2576
2577 unlock_fs(md);
2578 dm_table_presuspend_undo_targets(map);
2579 /* pushback list is already flushed, so skip flush */
2580 }
2581
2582 return r;
2583}
2584
2585/*
2586 * We need to be able to change a mapping table under a mounted
2587 * filesystem. For example we might want to move some data in
2588 * the background. Before the table can be swapped with
2589 * dm_bind_table, dm_suspend must be called to flush any in
2590 * flight bios and ensure that any further io gets deferred.
2591 */
2592/*
2593 * Suspend mechanism in request-based dm.
2594 *
2595 * 1. Flush all I/Os by lock_fs() if needed.
2596 * 2. Stop dispatching any I/O by stopping the request_queue.
2597 * 3. Wait for all in-flight I/Os to be completed or requeued.
2598 *
2599 * To abort suspend, start the request_queue.
2600 */
2601int dm_suspend(struct mapped_device *md, unsigned suspend_flags)
2602{
2603 struct dm_table *map = NULL;
2604 int r = 0;
2605
2606retry:
2607 mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING);
2608
2609 if (dm_suspended_md(md)) {
2610 r = -EINVAL;
2611 goto out_unlock;
2612 }
2613
2614 if (dm_suspended_internally_md(md)) {
2615 /* already internally suspended, wait for internal resume */
2616 mutex_unlock(&md->suspend_lock);
2617 r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE);
2618 if (r)
2619 return r;
2620 goto retry;
2621 }
2622
2623 map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2624
2625 r = __dm_suspend(md, map, suspend_flags, TASK_INTERRUPTIBLE, DMF_SUSPENDED);
2626 if (r)
2627 goto out_unlock;
2628
2629 dm_table_postsuspend_targets(map);
2630
2631out_unlock:
2632 mutex_unlock(&md->suspend_lock);
2633 return r;
2634}
2635
2636static int __dm_resume(struct mapped_device *md, struct dm_table *map)
2637{
2638 if (map) {
2639 int r = dm_table_resume_targets(map);
2640 if (r)
2641 return r;
2642 }
2643
2644 dm_queue_flush(md);
2645
2646 /*
2647 * Flushing deferred I/Os must be done after targets are resumed
2648 * so that mapping of targets can work correctly.
2649 * Request-based dm is queueing the deferred I/Os in its request_queue.
2650 */
2651 if (dm_request_based(md))
2652 dm_start_queue(md->queue);
2653
2654 unlock_fs(md);
2655
2656 return 0;
2657}
2658
2659int dm_resume(struct mapped_device *md)
2660{
2661 int r;
2662 struct dm_table *map = NULL;
2663
2664retry:
2665 r = -EINVAL;
2666 mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING);
2667
2668 if (!dm_suspended_md(md))
2669 goto out;
2670
2671 if (dm_suspended_internally_md(md)) {
2672 /* already internally suspended, wait for internal resume */
2673 mutex_unlock(&md->suspend_lock);
2674 r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE);
2675 if (r)
2676 return r;
2677 goto retry;
2678 }
2679
2680 map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2681 if (!map || !dm_table_get_size(map))
2682 goto out;
2683
2684 r = __dm_resume(md, map);
2685 if (r)
2686 goto out;
2687
2688 clear_bit(DMF_SUSPENDED, &md->flags);
2689out:
2690 mutex_unlock(&md->suspend_lock);
2691
2692 return r;
2693}
2694
2695/*
2696 * Internal suspend/resume works like userspace-driven suspend. It waits
2697 * until all bios finish and prevents issuing new bios to the target drivers.
2698 * It may be used only from the kernel.
2699 */
2700
2701static void __dm_internal_suspend(struct mapped_device *md, unsigned suspend_flags)
2702{
2703 struct dm_table *map = NULL;
2704
2705 lockdep_assert_held(&md->suspend_lock);
2706
2707 if (md->internal_suspend_count++)
2708 return; /* nested internal suspend */
2709
2710 if (dm_suspended_md(md)) {
2711 set_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
2712 return; /* nest suspend */
2713 }
2714
2715 map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2716
2717 /*
2718 * Using TASK_UNINTERRUPTIBLE because only NOFLUSH internal suspend is
2719 * supported. Properly supporting a TASK_INTERRUPTIBLE internal suspend
2720 * would require changing .presuspend to return an error -- avoid this
2721 * until there is a need for more elaborate variants of internal suspend.
2722 */
2723 (void) __dm_suspend(md, map, suspend_flags, TASK_UNINTERRUPTIBLE,
2724 DMF_SUSPENDED_INTERNALLY);
2725
2726 dm_table_postsuspend_targets(map);
2727}
2728
2729static void __dm_internal_resume(struct mapped_device *md)
2730{
2731 BUG_ON(!md->internal_suspend_count);
2732
2733 if (--md->internal_suspend_count)
2734 return; /* resume from nested internal suspend */
2735
2736 if (dm_suspended_md(md))
2737 goto done; /* resume from nested suspend */
2738
2739 /*
2740 * NOTE: existing callers don't need to call dm_table_resume_targets
2741 * (which may fail -- so best to avoid it for now by passing NULL map)
2742 */
2743 (void) __dm_resume(md, NULL);
2744
2745done:
2746 clear_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
2747 smp_mb__after_atomic();
2748 wake_up_bit(&md->flags, DMF_SUSPENDED_INTERNALLY);
2749}
2750
2751void dm_internal_suspend_noflush(struct mapped_device *md)
2752{
2753 mutex_lock(&md->suspend_lock);
2754 __dm_internal_suspend(md, DM_SUSPEND_NOFLUSH_FLAG);
2755 mutex_unlock(&md->suspend_lock);
2756}
2757EXPORT_SYMBOL_GPL(dm_internal_suspend_noflush);
2758
2759void dm_internal_resume(struct mapped_device *md)
2760{
2761 mutex_lock(&md->suspend_lock);
2762 __dm_internal_resume(md);
2763 mutex_unlock(&md->suspend_lock);
2764}
2765EXPORT_SYMBOL_GPL(dm_internal_resume);
2766
2767/*
2768 * Fast variants of internal suspend/resume hold md->suspend_lock,
2769 * which prevents interaction with userspace-driven suspend.
2770 */
2771
2772void dm_internal_suspend_fast(struct mapped_device *md)
2773{
2774 mutex_lock(&md->suspend_lock);
2775 if (dm_suspended_md(md) || dm_suspended_internally_md(md))
2776 return;
2777
2778 set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2779 synchronize_srcu(&md->io_barrier);
2780 flush_workqueue(md->wq);
2781 dm_wait_for_completion(md, TASK_UNINTERRUPTIBLE);
2782}
2783EXPORT_SYMBOL_GPL(dm_internal_suspend_fast);
2784
2785void dm_internal_resume_fast(struct mapped_device *md)
2786{
2787 if (dm_suspended_md(md) || dm_suspended_internally_md(md))
2788 goto done;
2789
2790 dm_queue_flush(md);
2791
2792done:
2793 mutex_unlock(&md->suspend_lock);
2794}
2795EXPORT_SYMBOL_GPL(dm_internal_resume_fast);
2796
2797/*-----------------------------------------------------------------
2798 * Event notification.
2799 *---------------------------------------------------------------*/
2800int dm_kobject_uevent(struct mapped_device *md, enum kobject_action action,
2801 unsigned cookie)
2802{
2803 char udev_cookie[DM_COOKIE_LENGTH];
2804 char *envp[] = { udev_cookie, NULL };
2805
2806 if (!cookie)
2807 return kobject_uevent(&disk_to_dev(md->disk)->kobj, action);
2808 else {
2809 snprintf(udev_cookie, DM_COOKIE_LENGTH, "%s=%u",
2810 DM_COOKIE_ENV_VAR_NAME, cookie);
2811 return kobject_uevent_env(&disk_to_dev(md->disk)->kobj,
2812 action, envp);
2813 }
2814}
2815
2816uint32_t dm_next_uevent_seq(struct mapped_device *md)
2817{
2818 return atomic_add_return(1, &md->uevent_seq);
2819}
2820
2821uint32_t dm_get_event_nr(struct mapped_device *md)
2822{
2823 return atomic_read(&md->event_nr);
2824}
2825
2826int dm_wait_event(struct mapped_device *md, int event_nr)
2827{
2828 return wait_event_interruptible(md->eventq,
2829 (event_nr != atomic_read(&md->event_nr)));
2830}
2831
2832void dm_uevent_add(struct mapped_device *md, struct list_head *elist)
2833{
2834 unsigned long flags;
2835
2836 spin_lock_irqsave(&md->uevent_lock, flags);
2837 list_add(elist, &md->uevent_list);
2838 spin_unlock_irqrestore(&md->uevent_lock, flags);
2839}
2840
2841/*
2842 * The gendisk is only valid as long as you have a reference
2843 * count on 'md'.
2844 */
2845struct gendisk *dm_disk(struct mapped_device *md)
2846{
2847 return md->disk;
2848}
2849EXPORT_SYMBOL_GPL(dm_disk);
2850
2851struct kobject *dm_kobject(struct mapped_device *md)
2852{
2853 return &md->kobj_holder.kobj;
2854}
2855
2856struct mapped_device *dm_get_from_kobject(struct kobject *kobj)
2857{
2858 struct mapped_device *md;
2859
2860 md = container_of(kobj, struct mapped_device, kobj_holder.kobj);
2861
2862 spin_lock(&_minor_lock);
2863 if (test_bit(DMF_FREEING, &md->flags) || dm_deleting_md(md)) {
2864 md = NULL;
2865 goto out;
2866 }
2867 dm_get(md);
2868out:
2869 spin_unlock(&_minor_lock);
2870
2871 return md;
2872}
2873
2874int dm_suspended_md(struct mapped_device *md)
2875{
2876 return test_bit(DMF_SUSPENDED, &md->flags);
2877}
2878
2879int dm_suspended_internally_md(struct mapped_device *md)
2880{
2881 return test_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
2882}
2883
2884int dm_test_deferred_remove_flag(struct mapped_device *md)
2885{
2886 return test_bit(DMF_DEFERRED_REMOVE, &md->flags);
2887}
2888
2889int dm_suspended(struct dm_target *ti)
2890{
2891 return dm_suspended_md(dm_table_get_md(ti->table));
2892}
2893EXPORT_SYMBOL_GPL(dm_suspended);
2894
2895int dm_noflush_suspending(struct dm_target *ti)
2896{
2897 return __noflush_suspending(dm_table_get_md(ti->table));
2898}
2899EXPORT_SYMBOL_GPL(dm_noflush_suspending);
2900
2901struct dm_md_mempools *dm_alloc_md_mempools(struct mapped_device *md, enum dm_queue_mode type,
2902 unsigned integrity, unsigned per_io_data_size,
2903 unsigned min_pool_size)
2904{
2905 struct dm_md_mempools *pools = kzalloc_node(sizeof(*pools), GFP_KERNEL, md->numa_node_id);
2906 unsigned int pool_size = 0;
2907 unsigned int front_pad, io_front_pad;
2908
2909 if (!pools)
2910 return NULL;
2911
2912 switch (type) {
2913 case DM_TYPE_BIO_BASED:
2914 case DM_TYPE_DAX_BIO_BASED:
2915 case DM_TYPE_NVME_BIO_BASED:
2916 pool_size = max(dm_get_reserved_bio_based_ios(), min_pool_size);
2917 front_pad = roundup(per_io_data_size, __alignof__(struct dm_target_io)) + offsetof(struct dm_target_io, clone);
2918 io_front_pad = roundup(front_pad, __alignof__(struct dm_io)) + offsetof(struct dm_io, tio);
2919 pools->io_bs = bioset_create(pool_size, io_front_pad, 0);
2920 if (!pools->io_bs)
2921 goto out;
2922 if (integrity && bioset_integrity_create(pools->io_bs, pool_size))
2923 goto out;
2924 break;
2925 case DM_TYPE_REQUEST_BASED:
2926 case DM_TYPE_MQ_REQUEST_BASED:
2927 pool_size = max(dm_get_reserved_rq_based_ios(), min_pool_size);
2928 front_pad = offsetof(struct dm_rq_clone_bio_info, clone);
2929 /* per_io_data_size is used for blk-mq pdu at queue allocation */
2930 break;
2931 default:
2932 BUG();
2933 }
2934
2935 pools->bs = bioset_create(pool_size, front_pad, 0);
2936 if (!pools->bs)
2937 goto out;
2938
2939 if (integrity && bioset_integrity_create(pools->bs, pool_size))
2940 goto out;
2941
2942 return pools;
2943
2944out:
2945 dm_free_md_mempools(pools);
2946
2947 return NULL;
2948}
2949
2950void dm_free_md_mempools(struct dm_md_mempools *pools)
2951{
2952 if (!pools)
2953 return;
2954
2955 if (pools->bs)
2956 bioset_free(pools->bs);
2957 if (pools->io_bs)
2958 bioset_free(pools->io_bs);
2959
2960 kfree(pools);
2961}
2962
2963struct dm_pr {
2964 u64 old_key;
2965 u64 new_key;
2966 u32 flags;
2967 bool fail_early;
2968};
2969
2970static int dm_call_pr(struct block_device *bdev, iterate_devices_callout_fn fn,
2971 void *data)
2972{
2973 struct mapped_device *md = bdev->bd_disk->private_data;
2974 struct dm_table *table;
2975 struct dm_target *ti;
2976 int ret = -ENOTTY, srcu_idx;
2977
2978 table = dm_get_live_table(md, &srcu_idx);
2979 if (!table || !dm_table_get_size(table))
2980 goto out;
2981
2982 /* We only support devices that have a single target */
2983 if (dm_table_get_num_targets(table) != 1)
2984 goto out;
2985 ti = dm_table_get_target(table, 0);
2986
2987 ret = -EINVAL;
2988 if (!ti->type->iterate_devices)
2989 goto out;
2990
2991 ret = ti->type->iterate_devices(ti, fn, data);
2992out:
2993 dm_put_live_table(md, srcu_idx);
2994 return ret;
2995}
2996
2997/*
2998 * For register / unregister we need to manually call out to every path.
2999 */
3000static int __dm_pr_register(struct dm_target *ti, struct dm_dev *dev,
3001 sector_t start, sector_t len, void *data)
3002{
3003 struct dm_pr *pr = data;
3004 const struct pr_ops *ops = dev->bdev->bd_disk->fops->pr_ops;
3005
3006 if (!ops || !ops->pr_register)
3007 return -EOPNOTSUPP;
3008 return ops->pr_register(dev->bdev, pr->old_key, pr->new_key, pr->flags);
3009}
3010
3011static int dm_pr_register(struct block_device *bdev, u64 old_key, u64 new_key,
3012 u32 flags)
3013{
3014 struct dm_pr pr = {
3015 .old_key = old_key,
3016 .new_key = new_key,
3017 .flags = flags,
3018 .fail_early = true,
3019 };
3020 int ret;
3021
3022 ret = dm_call_pr(bdev, __dm_pr_register, &pr);
3023 if (ret && new_key) {
3024 /* unregister all paths if we failed to register any path */
3025 pr.old_key = new_key;
3026 pr.new_key = 0;
3027 pr.flags = 0;
3028 pr.fail_early = false;
3029 dm_call_pr(bdev, __dm_pr_register, &pr);
3030 }
3031
3032 return ret;
3033}
3034
3035static int dm_pr_reserve(struct block_device *bdev, u64 key, enum pr_type type,
3036 u32 flags)
3037{
3038 struct mapped_device *md = bdev->bd_disk->private_data;
3039 const struct pr_ops *ops;
3040 int r, srcu_idx;
3041
3042 r = dm_prepare_ioctl(md, &srcu_idx, &bdev);
3043 if (r < 0)
3044 goto out;
3045
3046 ops = bdev->bd_disk->fops->pr_ops;
3047 if (ops && ops->pr_reserve)
3048 r = ops->pr_reserve(bdev, key, type, flags);
3049 else
3050 r = -EOPNOTSUPP;
3051out:
3052 dm_unprepare_ioctl(md, srcu_idx);
3053 return r;
3054}
3055
3056static int dm_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
3057{
3058 struct mapped_device *md = bdev->bd_disk->private_data;
3059 const struct pr_ops *ops;
3060 int r, srcu_idx;
3061
3062 r = dm_prepare_ioctl(md, &srcu_idx, &bdev);
3063 if (r < 0)
3064 goto out;
3065
3066 ops = bdev->bd_disk->fops->pr_ops;
3067 if (ops && ops->pr_release)
3068 r = ops->pr_release(bdev, key, type);
3069 else
3070 r = -EOPNOTSUPP;
3071out:
3072 dm_unprepare_ioctl(md, srcu_idx);
3073 return r;
3074}
3075
3076static int dm_pr_preempt(struct block_device *bdev, u64 old_key, u64 new_key,
3077 enum pr_type type, bool abort)
3078{
3079 struct mapped_device *md = bdev->bd_disk->private_data;
3080 const struct pr_ops *ops;
3081 int r, srcu_idx;
3082
3083 r = dm_prepare_ioctl(md, &srcu_idx, &bdev);
3084 if (r < 0)
3085 goto out;
3086
3087 ops = bdev->bd_disk->fops->pr_ops;
3088 if (ops && ops->pr_preempt)
3089 r = ops->pr_preempt(bdev, old_key, new_key, type, abort);
3090 else
3091 r = -EOPNOTSUPP;
3092out:
3093 dm_unprepare_ioctl(md, srcu_idx);
3094 return r;
3095}
3096
3097static int dm_pr_clear(struct block_device *bdev, u64 key)
3098{
3099 struct mapped_device *md = bdev->bd_disk->private_data;
3100 const struct pr_ops *ops;
3101 int r, srcu_idx;
3102
3103 r = dm_prepare_ioctl(md, &srcu_idx, &bdev);
3104 if (r < 0)
3105 goto out;
3106
3107 ops = bdev->bd_disk->fops->pr_ops;
3108 if (ops && ops->pr_clear)
3109 r = ops->pr_clear(bdev, key);
3110 else
3111 r = -EOPNOTSUPP;
3112out:
3113 dm_unprepare_ioctl(md, srcu_idx);
3114 return r;
3115}
3116
3117static const struct pr_ops dm_pr_ops = {
3118 .pr_register = dm_pr_register,
3119 .pr_reserve = dm_pr_reserve,
3120 .pr_release = dm_pr_release,
3121 .pr_preempt = dm_pr_preempt,
3122 .pr_clear = dm_pr_clear,
3123};
3124
3125static const struct block_device_operations dm_blk_dops = {
3126 .open = dm_blk_open,
3127 .release = dm_blk_close,
3128 .ioctl = dm_blk_ioctl,
3129 .getgeo = dm_blk_getgeo,
3130 .pr_ops = &dm_pr_ops,
3131 .owner = THIS_MODULE
3132};
3133
3134static const struct dax_operations dm_dax_ops = {
3135 .direct_access = dm_dax_direct_access,
3136 .copy_from_iter = dm_dax_copy_from_iter,
3137};
3138
3139/*
3140 * module hooks
3141 */
3142module_init(dm_init);
3143module_exit(dm_exit);
3144
3145module_param(major, uint, 0);
3146MODULE_PARM_DESC(major, "The major number of the device mapper");
3147
3148module_param(reserved_bio_based_ios, uint, S_IRUGO | S_IWUSR);
3149MODULE_PARM_DESC(reserved_bio_based_ios, "Reserved IOs in bio-based mempools");
3150
3151module_param(dm_numa_node, int, S_IRUGO | S_IWUSR);
3152MODULE_PARM_DESC(dm_numa_node, "NUMA node for DM device memory allocations");
3153
3154MODULE_DESCRIPTION(DM_NAME " driver");
3155MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
3156MODULE_LICENSE("GPL");