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