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