Loading...
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * Copyright (C) 2001, 2002 Sistina Software (UK) Limited.
4 * Copyright (C) 2004-2008 Red Hat, Inc. All rights reserved.
5 *
6 * This file is released under the GPL.
7 */
8
9#include "dm-core.h"
10#include "dm-rq.h"
11#include "dm-uevent.h"
12#include "dm-ima.h"
13
14#include <linux/init.h>
15#include <linux/module.h>
16#include <linux/mutex.h>
17#include <linux/sched/mm.h>
18#include <linux/sched/signal.h>
19#include <linux/blkpg.h>
20#include <linux/bio.h>
21#include <linux/mempool.h>
22#include <linux/dax.h>
23#include <linux/slab.h>
24#include <linux/idr.h>
25#include <linux/uio.h>
26#include <linux/hdreg.h>
27#include <linux/delay.h>
28#include <linux/wait.h>
29#include <linux/pr.h>
30#include <linux/refcount.h>
31#include <linux/part_stat.h>
32#include <linux/blk-crypto.h>
33#include <linux/blk-crypto-profile.h>
34
35#define DM_MSG_PREFIX "core"
36
37/*
38 * Cookies are numeric values sent with CHANGE and REMOVE
39 * uevents while resuming, removing or renaming the device.
40 */
41#define DM_COOKIE_ENV_VAR_NAME "DM_COOKIE"
42#define DM_COOKIE_LENGTH 24
43
44/*
45 * For REQ_POLLED fs bio, this flag is set if we link mapped underlying
46 * dm_io into one list, and reuse bio->bi_private as the list head. Before
47 * ending this fs bio, we will recover its ->bi_private.
48 */
49#define REQ_DM_POLL_LIST REQ_DRV
50
51static const char *_name = DM_NAME;
52
53static unsigned int major;
54static unsigned int _major;
55
56static DEFINE_IDR(_minor_idr);
57
58static DEFINE_SPINLOCK(_minor_lock);
59
60static void do_deferred_remove(struct work_struct *w);
61
62static DECLARE_WORK(deferred_remove_work, do_deferred_remove);
63
64static struct workqueue_struct *deferred_remove_workqueue;
65
66atomic_t dm_global_event_nr = ATOMIC_INIT(0);
67DECLARE_WAIT_QUEUE_HEAD(dm_global_eventq);
68
69void dm_issue_global_event(void)
70{
71 atomic_inc(&dm_global_event_nr);
72 wake_up(&dm_global_eventq);
73}
74
75DEFINE_STATIC_KEY_FALSE(stats_enabled);
76DEFINE_STATIC_KEY_FALSE(swap_bios_enabled);
77DEFINE_STATIC_KEY_FALSE(zoned_enabled);
78
79/*
80 * One of these is allocated (on-stack) per original bio.
81 */
82struct clone_info {
83 struct dm_table *map;
84 struct bio *bio;
85 struct dm_io *io;
86 sector_t sector;
87 unsigned int sector_count;
88 bool is_abnormal_io:1;
89 bool submit_as_polled:1;
90};
91
92static inline struct dm_target_io *clone_to_tio(struct bio *clone)
93{
94 return container_of(clone, struct dm_target_io, clone);
95}
96
97void *dm_per_bio_data(struct bio *bio, size_t data_size)
98{
99 if (!dm_tio_flagged(clone_to_tio(bio), DM_TIO_INSIDE_DM_IO))
100 return (char *)bio - DM_TARGET_IO_BIO_OFFSET - data_size;
101 return (char *)bio - DM_IO_BIO_OFFSET - data_size;
102}
103EXPORT_SYMBOL_GPL(dm_per_bio_data);
104
105struct bio *dm_bio_from_per_bio_data(void *data, size_t data_size)
106{
107 struct dm_io *io = (struct dm_io *)((char *)data + data_size);
108
109 if (io->magic == DM_IO_MAGIC)
110 return (struct bio *)((char *)io + DM_IO_BIO_OFFSET);
111 BUG_ON(io->magic != DM_TIO_MAGIC);
112 return (struct bio *)((char *)io + DM_TARGET_IO_BIO_OFFSET);
113}
114EXPORT_SYMBOL_GPL(dm_bio_from_per_bio_data);
115
116unsigned int dm_bio_get_target_bio_nr(const struct bio *bio)
117{
118 return container_of(bio, struct dm_target_io, clone)->target_bio_nr;
119}
120EXPORT_SYMBOL_GPL(dm_bio_get_target_bio_nr);
121
122#define MINOR_ALLOCED ((void *)-1)
123
124#define DM_NUMA_NODE NUMA_NO_NODE
125static int dm_numa_node = DM_NUMA_NODE;
126
127#define DEFAULT_SWAP_BIOS (8 * 1048576 / PAGE_SIZE)
128static int swap_bios = DEFAULT_SWAP_BIOS;
129static int get_swap_bios(void)
130{
131 int latch = READ_ONCE(swap_bios);
132
133 if (unlikely(latch <= 0))
134 latch = DEFAULT_SWAP_BIOS;
135 return latch;
136}
137
138struct table_device {
139 struct list_head list;
140 refcount_t count;
141 struct dm_dev dm_dev;
142};
143
144/*
145 * Bio-based DM's mempools' reserved IOs set by the user.
146 */
147#define RESERVED_BIO_BASED_IOS 16
148static unsigned int reserved_bio_based_ios = RESERVED_BIO_BASED_IOS;
149
150static int __dm_get_module_param_int(int *module_param, int min, int max)
151{
152 int param = READ_ONCE(*module_param);
153 int modified_param = 0;
154 bool modified = true;
155
156 if (param < min)
157 modified_param = min;
158 else if (param > max)
159 modified_param = max;
160 else
161 modified = false;
162
163 if (modified) {
164 (void)cmpxchg(module_param, param, modified_param);
165 param = modified_param;
166 }
167
168 return param;
169}
170
171unsigned int __dm_get_module_param(unsigned int *module_param, unsigned int def, unsigned int max)
172{
173 unsigned int param = READ_ONCE(*module_param);
174 unsigned int modified_param = 0;
175
176 if (!param)
177 modified_param = def;
178 else if (param > max)
179 modified_param = max;
180
181 if (modified_param) {
182 (void)cmpxchg(module_param, param, modified_param);
183 param = modified_param;
184 }
185
186 return param;
187}
188
189unsigned int dm_get_reserved_bio_based_ios(void)
190{
191 return __dm_get_module_param(&reserved_bio_based_ios,
192 RESERVED_BIO_BASED_IOS, DM_RESERVED_MAX_IOS);
193}
194EXPORT_SYMBOL_GPL(dm_get_reserved_bio_based_ios);
195
196static unsigned int dm_get_numa_node(void)
197{
198 return __dm_get_module_param_int(&dm_numa_node,
199 DM_NUMA_NODE, num_online_nodes() - 1);
200}
201
202static int __init local_init(void)
203{
204 int r;
205
206 r = dm_uevent_init();
207 if (r)
208 return r;
209
210 deferred_remove_workqueue = alloc_ordered_workqueue("kdmremove", 0);
211 if (!deferred_remove_workqueue) {
212 r = -ENOMEM;
213 goto out_uevent_exit;
214 }
215
216 _major = major;
217 r = register_blkdev(_major, _name);
218 if (r < 0)
219 goto out_free_workqueue;
220
221 if (!_major)
222 _major = r;
223
224 return 0;
225
226out_free_workqueue:
227 destroy_workqueue(deferred_remove_workqueue);
228out_uevent_exit:
229 dm_uevent_exit();
230
231 return r;
232}
233
234static void local_exit(void)
235{
236 destroy_workqueue(deferred_remove_workqueue);
237
238 unregister_blkdev(_major, _name);
239 dm_uevent_exit();
240
241 _major = 0;
242
243 DMINFO("cleaned up");
244}
245
246static int (*_inits[])(void) __initdata = {
247 local_init,
248 dm_target_init,
249 dm_linear_init,
250 dm_stripe_init,
251 dm_io_init,
252 dm_kcopyd_init,
253 dm_interface_init,
254 dm_statistics_init,
255};
256
257static void (*_exits[])(void) = {
258 local_exit,
259 dm_target_exit,
260 dm_linear_exit,
261 dm_stripe_exit,
262 dm_io_exit,
263 dm_kcopyd_exit,
264 dm_interface_exit,
265 dm_statistics_exit,
266};
267
268static int __init dm_init(void)
269{
270 const int count = ARRAY_SIZE(_inits);
271 int r, i;
272
273#if (IS_ENABLED(CONFIG_IMA) && !IS_ENABLED(CONFIG_IMA_DISABLE_HTABLE))
274 DMWARN("CONFIG_IMA_DISABLE_HTABLE is disabled."
275 " Duplicate IMA measurements will not be recorded in the IMA log.");
276#endif
277
278 for (i = 0; i < count; i++) {
279 r = _inits[i]();
280 if (r)
281 goto bad;
282 }
283
284 return 0;
285bad:
286 while (i--)
287 _exits[i]();
288
289 return r;
290}
291
292static void __exit dm_exit(void)
293{
294 int i = ARRAY_SIZE(_exits);
295
296 while (i--)
297 _exits[i]();
298
299 /*
300 * Should be empty by this point.
301 */
302 idr_destroy(&_minor_idr);
303}
304
305/*
306 * Block device functions
307 */
308int dm_deleting_md(struct mapped_device *md)
309{
310 return test_bit(DMF_DELETING, &md->flags);
311}
312
313static int dm_blk_open(struct gendisk *disk, blk_mode_t mode)
314{
315 struct mapped_device *md;
316
317 spin_lock(&_minor_lock);
318
319 md = disk->private_data;
320 if (!md)
321 goto out;
322
323 if (test_bit(DMF_FREEING, &md->flags) ||
324 dm_deleting_md(md)) {
325 md = NULL;
326 goto out;
327 }
328
329 dm_get(md);
330 atomic_inc(&md->open_count);
331out:
332 spin_unlock(&_minor_lock);
333
334 return md ? 0 : -ENXIO;
335}
336
337static void dm_blk_close(struct gendisk *disk)
338{
339 struct mapped_device *md;
340
341 spin_lock(&_minor_lock);
342
343 md = disk->private_data;
344 if (WARN_ON(!md))
345 goto out;
346
347 if (atomic_dec_and_test(&md->open_count) &&
348 (test_bit(DMF_DEFERRED_REMOVE, &md->flags)))
349 queue_work(deferred_remove_workqueue, &deferred_remove_work);
350
351 dm_put(md);
352out:
353 spin_unlock(&_minor_lock);
354}
355
356int dm_open_count(struct mapped_device *md)
357{
358 return atomic_read(&md->open_count);
359}
360
361/*
362 * Guarantees nothing is using the device before it's deleted.
363 */
364int dm_lock_for_deletion(struct mapped_device *md, bool mark_deferred, bool only_deferred)
365{
366 int r = 0;
367
368 spin_lock(&_minor_lock);
369
370 if (dm_open_count(md)) {
371 r = -EBUSY;
372 if (mark_deferred)
373 set_bit(DMF_DEFERRED_REMOVE, &md->flags);
374 } else if (only_deferred && !test_bit(DMF_DEFERRED_REMOVE, &md->flags))
375 r = -EEXIST;
376 else
377 set_bit(DMF_DELETING, &md->flags);
378
379 spin_unlock(&_minor_lock);
380
381 return r;
382}
383
384int dm_cancel_deferred_remove(struct mapped_device *md)
385{
386 int r = 0;
387
388 spin_lock(&_minor_lock);
389
390 if (test_bit(DMF_DELETING, &md->flags))
391 r = -EBUSY;
392 else
393 clear_bit(DMF_DEFERRED_REMOVE, &md->flags);
394
395 spin_unlock(&_minor_lock);
396
397 return r;
398}
399
400static void do_deferred_remove(struct work_struct *w)
401{
402 dm_deferred_remove();
403}
404
405static int dm_blk_getgeo(struct block_device *bdev, struct hd_geometry *geo)
406{
407 struct mapped_device *md = bdev->bd_disk->private_data;
408
409 return dm_get_geometry(md, geo);
410}
411
412static int dm_prepare_ioctl(struct mapped_device *md, int *srcu_idx,
413 struct block_device **bdev)
414{
415 struct dm_target *ti;
416 struct dm_table *map;
417 int r;
418
419retry:
420 r = -ENOTTY;
421 map = dm_get_live_table(md, srcu_idx);
422 if (!map || !dm_table_get_size(map))
423 return r;
424
425 /* We only support devices that have a single target */
426 if (map->num_targets != 1)
427 return r;
428
429 ti = dm_table_get_target(map, 0);
430 if (!ti->type->prepare_ioctl)
431 return r;
432
433 if (dm_suspended_md(md))
434 return -EAGAIN;
435
436 r = ti->type->prepare_ioctl(ti, bdev);
437 if (r == -ENOTCONN && !fatal_signal_pending(current)) {
438 dm_put_live_table(md, *srcu_idx);
439 fsleep(10000);
440 goto retry;
441 }
442
443 return r;
444}
445
446static void dm_unprepare_ioctl(struct mapped_device *md, int srcu_idx)
447{
448 dm_put_live_table(md, srcu_idx);
449}
450
451static int dm_blk_ioctl(struct block_device *bdev, blk_mode_t mode,
452 unsigned int cmd, unsigned long arg)
453{
454 struct mapped_device *md = bdev->bd_disk->private_data;
455 int r, srcu_idx;
456
457 r = dm_prepare_ioctl(md, &srcu_idx, &bdev);
458 if (r < 0)
459 goto out;
460
461 if (r > 0) {
462 /*
463 * Target determined this ioctl is being issued against a
464 * subset of the parent bdev; require extra privileges.
465 */
466 if (!capable(CAP_SYS_RAWIO)) {
467 DMDEBUG_LIMIT(
468 "%s: sending ioctl %x to DM device without required privilege.",
469 current->comm, cmd);
470 r = -ENOIOCTLCMD;
471 goto out;
472 }
473 }
474
475 if (!bdev->bd_disk->fops->ioctl)
476 r = -ENOTTY;
477 else
478 r = bdev->bd_disk->fops->ioctl(bdev, mode, cmd, arg);
479out:
480 dm_unprepare_ioctl(md, srcu_idx);
481 return r;
482}
483
484u64 dm_start_time_ns_from_clone(struct bio *bio)
485{
486 return jiffies_to_nsecs(clone_to_tio(bio)->io->start_time);
487}
488EXPORT_SYMBOL_GPL(dm_start_time_ns_from_clone);
489
490static inline bool bio_is_flush_with_data(struct bio *bio)
491{
492 return ((bio->bi_opf & REQ_PREFLUSH) && bio->bi_iter.bi_size);
493}
494
495static inline unsigned int dm_io_sectors(struct dm_io *io, struct bio *bio)
496{
497 /*
498 * If REQ_PREFLUSH set, don't account payload, it will be
499 * submitted (and accounted) after this flush completes.
500 */
501 if (bio_is_flush_with_data(bio))
502 return 0;
503 if (unlikely(dm_io_flagged(io, DM_IO_WAS_SPLIT)))
504 return io->sectors;
505 return bio_sectors(bio);
506}
507
508static void dm_io_acct(struct dm_io *io, bool end)
509{
510 struct bio *bio = io->orig_bio;
511
512 if (dm_io_flagged(io, DM_IO_BLK_STAT)) {
513 if (!end)
514 bdev_start_io_acct(bio->bi_bdev, bio_op(bio),
515 io->start_time);
516 else
517 bdev_end_io_acct(bio->bi_bdev, bio_op(bio),
518 dm_io_sectors(io, bio),
519 io->start_time);
520 }
521
522 if (static_branch_unlikely(&stats_enabled) &&
523 unlikely(dm_stats_used(&io->md->stats))) {
524 sector_t sector;
525
526 if (unlikely(dm_io_flagged(io, DM_IO_WAS_SPLIT)))
527 sector = bio_end_sector(bio) - io->sector_offset;
528 else
529 sector = bio->bi_iter.bi_sector;
530
531 dm_stats_account_io(&io->md->stats, bio_data_dir(bio),
532 sector, dm_io_sectors(io, bio),
533 end, io->start_time, &io->stats_aux);
534 }
535}
536
537static void __dm_start_io_acct(struct dm_io *io)
538{
539 dm_io_acct(io, false);
540}
541
542static void dm_start_io_acct(struct dm_io *io, struct bio *clone)
543{
544 /*
545 * Ensure IO accounting is only ever started once.
546 */
547 if (dm_io_flagged(io, DM_IO_ACCOUNTED))
548 return;
549
550 /* Expect no possibility for race unless DM_TIO_IS_DUPLICATE_BIO. */
551 if (!clone || likely(dm_tio_is_normal(clone_to_tio(clone)))) {
552 dm_io_set_flag(io, DM_IO_ACCOUNTED);
553 } else {
554 unsigned long flags;
555 /* Can afford locking given DM_TIO_IS_DUPLICATE_BIO */
556 spin_lock_irqsave(&io->lock, flags);
557 if (dm_io_flagged(io, DM_IO_ACCOUNTED)) {
558 spin_unlock_irqrestore(&io->lock, flags);
559 return;
560 }
561 dm_io_set_flag(io, DM_IO_ACCOUNTED);
562 spin_unlock_irqrestore(&io->lock, flags);
563 }
564
565 __dm_start_io_acct(io);
566}
567
568static void dm_end_io_acct(struct dm_io *io)
569{
570 dm_io_acct(io, true);
571}
572
573static struct dm_io *alloc_io(struct mapped_device *md, struct bio *bio, gfp_t gfp_mask)
574{
575 struct dm_io *io;
576 struct dm_target_io *tio;
577 struct bio *clone;
578
579 clone = bio_alloc_clone(NULL, bio, gfp_mask, &md->mempools->io_bs);
580 if (unlikely(!clone))
581 return NULL;
582 tio = clone_to_tio(clone);
583 tio->flags = 0;
584 dm_tio_set_flag(tio, DM_TIO_INSIDE_DM_IO);
585 tio->io = NULL;
586
587 io = container_of(tio, struct dm_io, tio);
588 io->magic = DM_IO_MAGIC;
589 io->status = BLK_STS_OK;
590
591 /* one ref is for submission, the other is for completion */
592 atomic_set(&io->io_count, 2);
593 this_cpu_inc(*md->pending_io);
594 io->orig_bio = bio;
595 io->md = md;
596 spin_lock_init(&io->lock);
597 io->start_time = jiffies;
598 io->flags = 0;
599 if (blk_queue_io_stat(md->queue))
600 dm_io_set_flag(io, DM_IO_BLK_STAT);
601
602 if (static_branch_unlikely(&stats_enabled) &&
603 unlikely(dm_stats_used(&md->stats)))
604 dm_stats_record_start(&md->stats, &io->stats_aux);
605
606 return io;
607}
608
609static void free_io(struct dm_io *io)
610{
611 bio_put(&io->tio.clone);
612}
613
614static struct bio *alloc_tio(struct clone_info *ci, struct dm_target *ti,
615 unsigned int target_bio_nr, unsigned int *len, gfp_t gfp_mask)
616{
617 struct mapped_device *md = ci->io->md;
618 struct dm_target_io *tio;
619 struct bio *clone;
620
621 if (!ci->io->tio.io) {
622 /* the dm_target_io embedded in ci->io is available */
623 tio = &ci->io->tio;
624 /* alloc_io() already initialized embedded clone */
625 clone = &tio->clone;
626 } else {
627 clone = bio_alloc_clone(NULL, ci->bio, gfp_mask,
628 &md->mempools->bs);
629 if (!clone)
630 return NULL;
631
632 /* REQ_DM_POLL_LIST shouldn't be inherited */
633 clone->bi_opf &= ~REQ_DM_POLL_LIST;
634
635 tio = clone_to_tio(clone);
636 tio->flags = 0; /* also clears DM_TIO_INSIDE_DM_IO */
637 }
638
639 tio->magic = DM_TIO_MAGIC;
640 tio->io = ci->io;
641 tio->ti = ti;
642 tio->target_bio_nr = target_bio_nr;
643 tio->len_ptr = len;
644 tio->old_sector = 0;
645
646 /* Set default bdev, but target must bio_set_dev() before issuing IO */
647 clone->bi_bdev = md->disk->part0;
648 if (unlikely(ti->needs_bio_set_dev))
649 bio_set_dev(clone, md->disk->part0);
650
651 if (len) {
652 clone->bi_iter.bi_size = to_bytes(*len);
653 if (bio_integrity(clone))
654 bio_integrity_trim(clone);
655 }
656
657 return clone;
658}
659
660static void free_tio(struct bio *clone)
661{
662 if (dm_tio_flagged(clone_to_tio(clone), DM_TIO_INSIDE_DM_IO))
663 return;
664 bio_put(clone);
665}
666
667/*
668 * Add the bio to the list of deferred io.
669 */
670static void queue_io(struct mapped_device *md, struct bio *bio)
671{
672 unsigned long flags;
673
674 spin_lock_irqsave(&md->deferred_lock, flags);
675 bio_list_add(&md->deferred, bio);
676 spin_unlock_irqrestore(&md->deferred_lock, flags);
677 queue_work(md->wq, &md->work);
678}
679
680/*
681 * Everyone (including functions in this file), should use this
682 * function to access the md->map field, and make sure they call
683 * dm_put_live_table() when finished.
684 */
685struct dm_table *dm_get_live_table(struct mapped_device *md,
686 int *srcu_idx) __acquires(md->io_barrier)
687{
688 *srcu_idx = srcu_read_lock(&md->io_barrier);
689
690 return srcu_dereference(md->map, &md->io_barrier);
691}
692
693void dm_put_live_table(struct mapped_device *md,
694 int srcu_idx) __releases(md->io_barrier)
695{
696 srcu_read_unlock(&md->io_barrier, srcu_idx);
697}
698
699void dm_sync_table(struct mapped_device *md)
700{
701 synchronize_srcu(&md->io_barrier);
702 synchronize_rcu_expedited();
703}
704
705/*
706 * A fast alternative to dm_get_live_table/dm_put_live_table.
707 * The caller must not block between these two functions.
708 */
709static struct dm_table *dm_get_live_table_fast(struct mapped_device *md) __acquires(RCU)
710{
711 rcu_read_lock();
712 return rcu_dereference(md->map);
713}
714
715static void dm_put_live_table_fast(struct mapped_device *md) __releases(RCU)
716{
717 rcu_read_unlock();
718}
719
720static char *_dm_claim_ptr = "I belong to device-mapper";
721
722/*
723 * Open a table device so we can use it as a map destination.
724 */
725static struct table_device *open_table_device(struct mapped_device *md,
726 dev_t dev, blk_mode_t mode)
727{
728 struct table_device *td;
729 struct bdev_handle *bdev_handle;
730 u64 part_off;
731 int r;
732
733 td = kmalloc_node(sizeof(*td), GFP_KERNEL, md->numa_node_id);
734 if (!td)
735 return ERR_PTR(-ENOMEM);
736 refcount_set(&td->count, 1);
737
738 bdev_handle = bdev_open_by_dev(dev, mode, _dm_claim_ptr, NULL);
739 if (IS_ERR(bdev_handle)) {
740 r = PTR_ERR(bdev_handle);
741 goto out_free_td;
742 }
743
744 /*
745 * We can be called before the dm disk is added. In that case we can't
746 * register the holder relation here. It will be done once add_disk was
747 * called.
748 */
749 if (md->disk->slave_dir) {
750 r = bd_link_disk_holder(bdev_handle->bdev, md->disk);
751 if (r)
752 goto out_blkdev_put;
753 }
754
755 td->dm_dev.mode = mode;
756 td->dm_dev.bdev = bdev_handle->bdev;
757 td->dm_dev.bdev_handle = bdev_handle;
758 td->dm_dev.dax_dev = fs_dax_get_by_bdev(bdev_handle->bdev, &part_off,
759 NULL, NULL);
760 format_dev_t(td->dm_dev.name, dev);
761 list_add(&td->list, &md->table_devices);
762 return td;
763
764out_blkdev_put:
765 bdev_release(bdev_handle);
766out_free_td:
767 kfree(td);
768 return ERR_PTR(r);
769}
770
771/*
772 * Close a table device that we've been using.
773 */
774static void close_table_device(struct table_device *td, struct mapped_device *md)
775{
776 if (md->disk->slave_dir)
777 bd_unlink_disk_holder(td->dm_dev.bdev, md->disk);
778 bdev_release(td->dm_dev.bdev_handle);
779 put_dax(td->dm_dev.dax_dev);
780 list_del(&td->list);
781 kfree(td);
782}
783
784static struct table_device *find_table_device(struct list_head *l, dev_t dev,
785 blk_mode_t mode)
786{
787 struct table_device *td;
788
789 list_for_each_entry(td, l, list)
790 if (td->dm_dev.bdev->bd_dev == dev && td->dm_dev.mode == mode)
791 return td;
792
793 return NULL;
794}
795
796int dm_get_table_device(struct mapped_device *md, dev_t dev, blk_mode_t mode,
797 struct dm_dev **result)
798{
799 struct table_device *td;
800
801 mutex_lock(&md->table_devices_lock);
802 td = find_table_device(&md->table_devices, dev, mode);
803 if (!td) {
804 td = open_table_device(md, dev, mode);
805 if (IS_ERR(td)) {
806 mutex_unlock(&md->table_devices_lock);
807 return PTR_ERR(td);
808 }
809 } else {
810 refcount_inc(&td->count);
811 }
812 mutex_unlock(&md->table_devices_lock);
813
814 *result = &td->dm_dev;
815 return 0;
816}
817
818void dm_put_table_device(struct mapped_device *md, struct dm_dev *d)
819{
820 struct table_device *td = container_of(d, struct table_device, dm_dev);
821
822 mutex_lock(&md->table_devices_lock);
823 if (refcount_dec_and_test(&td->count))
824 close_table_device(td, md);
825 mutex_unlock(&md->table_devices_lock);
826}
827
828/*
829 * Get the geometry associated with a dm device
830 */
831int dm_get_geometry(struct mapped_device *md, struct hd_geometry *geo)
832{
833 *geo = md->geometry;
834
835 return 0;
836}
837
838/*
839 * Set the geometry of a device.
840 */
841int dm_set_geometry(struct mapped_device *md, struct hd_geometry *geo)
842{
843 sector_t sz = (sector_t)geo->cylinders * geo->heads * geo->sectors;
844
845 if (geo->start > sz) {
846 DMERR("Start sector is beyond the geometry limits.");
847 return -EINVAL;
848 }
849
850 md->geometry = *geo;
851
852 return 0;
853}
854
855static int __noflush_suspending(struct mapped_device *md)
856{
857 return test_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
858}
859
860static void dm_requeue_add_io(struct dm_io *io, bool first_stage)
861{
862 struct mapped_device *md = io->md;
863
864 if (first_stage) {
865 struct dm_io *next = md->requeue_list;
866
867 md->requeue_list = io;
868 io->next = next;
869 } else {
870 bio_list_add_head(&md->deferred, io->orig_bio);
871 }
872}
873
874static void dm_kick_requeue(struct mapped_device *md, bool first_stage)
875{
876 if (first_stage)
877 queue_work(md->wq, &md->requeue_work);
878 else
879 queue_work(md->wq, &md->work);
880}
881
882/*
883 * Return true if the dm_io's original bio is requeued.
884 * io->status is updated with error if requeue disallowed.
885 */
886static bool dm_handle_requeue(struct dm_io *io, bool first_stage)
887{
888 struct bio *bio = io->orig_bio;
889 bool handle_requeue = (io->status == BLK_STS_DM_REQUEUE);
890 bool handle_polled_eagain = ((io->status == BLK_STS_AGAIN) &&
891 (bio->bi_opf & REQ_POLLED));
892 struct mapped_device *md = io->md;
893 bool requeued = false;
894
895 if (handle_requeue || handle_polled_eagain) {
896 unsigned long flags;
897
898 if (bio->bi_opf & REQ_POLLED) {
899 /*
900 * Upper layer won't help us poll split bio
901 * (io->orig_bio may only reflect a subset of the
902 * pre-split original) so clear REQ_POLLED.
903 */
904 bio_clear_polled(bio);
905 }
906
907 /*
908 * Target requested pushing back the I/O or
909 * polled IO hit BLK_STS_AGAIN.
910 */
911 spin_lock_irqsave(&md->deferred_lock, flags);
912 if ((__noflush_suspending(md) &&
913 !WARN_ON_ONCE(dm_is_zone_write(md, bio))) ||
914 handle_polled_eagain || first_stage) {
915 dm_requeue_add_io(io, first_stage);
916 requeued = true;
917 } else {
918 /*
919 * noflush suspend was interrupted or this is
920 * a write to a zoned target.
921 */
922 io->status = BLK_STS_IOERR;
923 }
924 spin_unlock_irqrestore(&md->deferred_lock, flags);
925 }
926
927 if (requeued)
928 dm_kick_requeue(md, first_stage);
929
930 return requeued;
931}
932
933static void __dm_io_complete(struct dm_io *io, bool first_stage)
934{
935 struct bio *bio = io->orig_bio;
936 struct mapped_device *md = io->md;
937 blk_status_t io_error;
938 bool requeued;
939
940 requeued = dm_handle_requeue(io, first_stage);
941 if (requeued && first_stage)
942 return;
943
944 io_error = io->status;
945 if (dm_io_flagged(io, DM_IO_ACCOUNTED))
946 dm_end_io_acct(io);
947 else if (!io_error) {
948 /*
949 * Must handle target that DM_MAPIO_SUBMITTED only to
950 * then bio_endio() rather than dm_submit_bio_remap()
951 */
952 __dm_start_io_acct(io);
953 dm_end_io_acct(io);
954 }
955 free_io(io);
956 smp_wmb();
957 this_cpu_dec(*md->pending_io);
958
959 /* nudge anyone waiting on suspend queue */
960 if (unlikely(wq_has_sleeper(&md->wait)))
961 wake_up(&md->wait);
962
963 /* Return early if the original bio was requeued */
964 if (requeued)
965 return;
966
967 if (bio_is_flush_with_data(bio)) {
968 /*
969 * Preflush done for flush with data, reissue
970 * without REQ_PREFLUSH.
971 */
972 bio->bi_opf &= ~REQ_PREFLUSH;
973 queue_io(md, bio);
974 } else {
975 /* done with normal IO or empty flush */
976 if (io_error)
977 bio->bi_status = io_error;
978 bio_endio(bio);
979 }
980}
981
982static void dm_wq_requeue_work(struct work_struct *work)
983{
984 struct mapped_device *md = container_of(work, struct mapped_device,
985 requeue_work);
986 unsigned long flags;
987 struct dm_io *io;
988
989 /* reuse deferred lock to simplify dm_handle_requeue */
990 spin_lock_irqsave(&md->deferred_lock, flags);
991 io = md->requeue_list;
992 md->requeue_list = NULL;
993 spin_unlock_irqrestore(&md->deferred_lock, flags);
994
995 while (io) {
996 struct dm_io *next = io->next;
997
998 dm_io_rewind(io, &md->disk->bio_split);
999
1000 io->next = NULL;
1001 __dm_io_complete(io, false);
1002 io = next;
1003 cond_resched();
1004 }
1005}
1006
1007/*
1008 * Two staged requeue:
1009 *
1010 * 1) io->orig_bio points to the real original bio, and the part mapped to
1011 * this io must be requeued, instead of other parts of the original bio.
1012 *
1013 * 2) io->orig_bio points to new cloned bio which matches the requeued dm_io.
1014 */
1015static void dm_io_complete(struct dm_io *io)
1016{
1017 bool first_requeue;
1018
1019 /*
1020 * Only dm_io that has been split needs two stage requeue, otherwise
1021 * we may run into long bio clone chain during suspend and OOM could
1022 * be triggered.
1023 *
1024 * Also flush data dm_io won't be marked as DM_IO_WAS_SPLIT, so they
1025 * also aren't handled via the first stage requeue.
1026 */
1027 if (dm_io_flagged(io, DM_IO_WAS_SPLIT))
1028 first_requeue = true;
1029 else
1030 first_requeue = false;
1031
1032 __dm_io_complete(io, first_requeue);
1033}
1034
1035/*
1036 * Decrements the number of outstanding ios that a bio has been
1037 * cloned into, completing the original io if necc.
1038 */
1039static inline void __dm_io_dec_pending(struct dm_io *io)
1040{
1041 if (atomic_dec_and_test(&io->io_count))
1042 dm_io_complete(io);
1043}
1044
1045static void dm_io_set_error(struct dm_io *io, blk_status_t error)
1046{
1047 unsigned long flags;
1048
1049 /* Push-back supersedes any I/O errors */
1050 spin_lock_irqsave(&io->lock, flags);
1051 if (!(io->status == BLK_STS_DM_REQUEUE &&
1052 __noflush_suspending(io->md))) {
1053 io->status = error;
1054 }
1055 spin_unlock_irqrestore(&io->lock, flags);
1056}
1057
1058static void dm_io_dec_pending(struct dm_io *io, blk_status_t error)
1059{
1060 if (unlikely(error))
1061 dm_io_set_error(io, error);
1062
1063 __dm_io_dec_pending(io);
1064}
1065
1066/*
1067 * The queue_limits are only valid as long as you have a reference
1068 * count on 'md'. But _not_ imposing verification to avoid atomic_read(),
1069 */
1070static inline struct queue_limits *dm_get_queue_limits(struct mapped_device *md)
1071{
1072 return &md->queue->limits;
1073}
1074
1075void disable_discard(struct mapped_device *md)
1076{
1077 struct queue_limits *limits = dm_get_queue_limits(md);
1078
1079 /* device doesn't really support DISCARD, disable it */
1080 limits->max_discard_sectors = 0;
1081}
1082
1083void disable_write_zeroes(struct mapped_device *md)
1084{
1085 struct queue_limits *limits = dm_get_queue_limits(md);
1086
1087 /* device doesn't really support WRITE ZEROES, disable it */
1088 limits->max_write_zeroes_sectors = 0;
1089}
1090
1091static bool swap_bios_limit(struct dm_target *ti, struct bio *bio)
1092{
1093 return unlikely((bio->bi_opf & REQ_SWAP) != 0) && unlikely(ti->limit_swap_bios);
1094}
1095
1096static void clone_endio(struct bio *bio)
1097{
1098 blk_status_t error = bio->bi_status;
1099 struct dm_target_io *tio = clone_to_tio(bio);
1100 struct dm_target *ti = tio->ti;
1101 dm_endio_fn endio = ti->type->end_io;
1102 struct dm_io *io = tio->io;
1103 struct mapped_device *md = io->md;
1104
1105 if (unlikely(error == BLK_STS_TARGET)) {
1106 if (bio_op(bio) == REQ_OP_DISCARD &&
1107 !bdev_max_discard_sectors(bio->bi_bdev))
1108 disable_discard(md);
1109 else if (bio_op(bio) == REQ_OP_WRITE_ZEROES &&
1110 !bdev_write_zeroes_sectors(bio->bi_bdev))
1111 disable_write_zeroes(md);
1112 }
1113
1114 if (static_branch_unlikely(&zoned_enabled) &&
1115 unlikely(bdev_is_zoned(bio->bi_bdev)))
1116 dm_zone_endio(io, bio);
1117
1118 if (endio) {
1119 int r = endio(ti, bio, &error);
1120
1121 switch (r) {
1122 case DM_ENDIO_REQUEUE:
1123 if (static_branch_unlikely(&zoned_enabled)) {
1124 /*
1125 * Requeuing writes to a sequential zone of a zoned
1126 * target will break the sequential write pattern:
1127 * fail such IO.
1128 */
1129 if (WARN_ON_ONCE(dm_is_zone_write(md, bio)))
1130 error = BLK_STS_IOERR;
1131 else
1132 error = BLK_STS_DM_REQUEUE;
1133 } else
1134 error = BLK_STS_DM_REQUEUE;
1135 fallthrough;
1136 case DM_ENDIO_DONE:
1137 break;
1138 case DM_ENDIO_INCOMPLETE:
1139 /* The target will handle the io */
1140 return;
1141 default:
1142 DMCRIT("unimplemented target endio return value: %d", r);
1143 BUG();
1144 }
1145 }
1146
1147 if (static_branch_unlikely(&swap_bios_enabled) &&
1148 unlikely(swap_bios_limit(ti, bio)))
1149 up(&md->swap_bios_semaphore);
1150
1151 free_tio(bio);
1152 dm_io_dec_pending(io, error);
1153}
1154
1155/*
1156 * Return maximum size of I/O possible at the supplied sector up to the current
1157 * target boundary.
1158 */
1159static inline sector_t max_io_len_target_boundary(struct dm_target *ti,
1160 sector_t target_offset)
1161{
1162 return ti->len - target_offset;
1163}
1164
1165static sector_t __max_io_len(struct dm_target *ti, sector_t sector,
1166 unsigned int max_granularity,
1167 unsigned int max_sectors)
1168{
1169 sector_t target_offset = dm_target_offset(ti, sector);
1170 sector_t len = max_io_len_target_boundary(ti, target_offset);
1171
1172 /*
1173 * Does the target need to split IO even further?
1174 * - varied (per target) IO splitting is a tenet of DM; this
1175 * explains why stacked chunk_sectors based splitting via
1176 * bio_split_to_limits() isn't possible here.
1177 */
1178 if (!max_granularity)
1179 return len;
1180 return min_t(sector_t, len,
1181 min(max_sectors ? : queue_max_sectors(ti->table->md->queue),
1182 blk_chunk_sectors_left(target_offset, max_granularity)));
1183}
1184
1185static inline sector_t max_io_len(struct dm_target *ti, sector_t sector)
1186{
1187 return __max_io_len(ti, sector, ti->max_io_len, 0);
1188}
1189
1190int dm_set_target_max_io_len(struct dm_target *ti, sector_t len)
1191{
1192 if (len > UINT_MAX) {
1193 DMERR("Specified maximum size of target IO (%llu) exceeds limit (%u)",
1194 (unsigned long long)len, UINT_MAX);
1195 ti->error = "Maximum size of target IO is too large";
1196 return -EINVAL;
1197 }
1198
1199 ti->max_io_len = (uint32_t) len;
1200
1201 return 0;
1202}
1203EXPORT_SYMBOL_GPL(dm_set_target_max_io_len);
1204
1205static struct dm_target *dm_dax_get_live_target(struct mapped_device *md,
1206 sector_t sector, int *srcu_idx)
1207 __acquires(md->io_barrier)
1208{
1209 struct dm_table *map;
1210 struct dm_target *ti;
1211
1212 map = dm_get_live_table(md, srcu_idx);
1213 if (!map)
1214 return NULL;
1215
1216 ti = dm_table_find_target(map, sector);
1217 if (!ti)
1218 return NULL;
1219
1220 return ti;
1221}
1222
1223static long dm_dax_direct_access(struct dax_device *dax_dev, pgoff_t pgoff,
1224 long nr_pages, enum dax_access_mode mode, void **kaddr,
1225 pfn_t *pfn)
1226{
1227 struct mapped_device *md = dax_get_private(dax_dev);
1228 sector_t sector = pgoff * PAGE_SECTORS;
1229 struct dm_target *ti;
1230 long len, ret = -EIO;
1231 int srcu_idx;
1232
1233 ti = dm_dax_get_live_target(md, sector, &srcu_idx);
1234
1235 if (!ti)
1236 goto out;
1237 if (!ti->type->direct_access)
1238 goto out;
1239 len = max_io_len(ti, sector) / PAGE_SECTORS;
1240 if (len < 1)
1241 goto out;
1242 nr_pages = min(len, nr_pages);
1243 ret = ti->type->direct_access(ti, pgoff, nr_pages, mode, kaddr, pfn);
1244
1245 out:
1246 dm_put_live_table(md, srcu_idx);
1247
1248 return ret;
1249}
1250
1251static int dm_dax_zero_page_range(struct dax_device *dax_dev, pgoff_t pgoff,
1252 size_t nr_pages)
1253{
1254 struct mapped_device *md = dax_get_private(dax_dev);
1255 sector_t sector = pgoff * PAGE_SECTORS;
1256 struct dm_target *ti;
1257 int ret = -EIO;
1258 int srcu_idx;
1259
1260 ti = dm_dax_get_live_target(md, sector, &srcu_idx);
1261
1262 if (!ti)
1263 goto out;
1264 if (WARN_ON(!ti->type->dax_zero_page_range)) {
1265 /*
1266 * ->zero_page_range() is mandatory dax operation. If we are
1267 * here, something is wrong.
1268 */
1269 goto out;
1270 }
1271 ret = ti->type->dax_zero_page_range(ti, pgoff, nr_pages);
1272 out:
1273 dm_put_live_table(md, srcu_idx);
1274
1275 return ret;
1276}
1277
1278static size_t dm_dax_recovery_write(struct dax_device *dax_dev, pgoff_t pgoff,
1279 void *addr, size_t bytes, struct iov_iter *i)
1280{
1281 struct mapped_device *md = dax_get_private(dax_dev);
1282 sector_t sector = pgoff * PAGE_SECTORS;
1283 struct dm_target *ti;
1284 int srcu_idx;
1285 long ret = 0;
1286
1287 ti = dm_dax_get_live_target(md, sector, &srcu_idx);
1288 if (!ti || !ti->type->dax_recovery_write)
1289 goto out;
1290
1291 ret = ti->type->dax_recovery_write(ti, pgoff, addr, bytes, i);
1292out:
1293 dm_put_live_table(md, srcu_idx);
1294 return ret;
1295}
1296
1297/*
1298 * A target may call dm_accept_partial_bio only from the map routine. It is
1299 * allowed for all bio types except REQ_PREFLUSH, REQ_OP_ZONE_* zone management
1300 * operations, REQ_OP_ZONE_APPEND (zone append writes) and any bio serviced by
1301 * __send_duplicate_bios().
1302 *
1303 * dm_accept_partial_bio informs the dm that the target only wants to process
1304 * additional n_sectors sectors of the bio and the rest of the data should be
1305 * sent in a next bio.
1306 *
1307 * A diagram that explains the arithmetics:
1308 * +--------------------+---------------+-------+
1309 * | 1 | 2 | 3 |
1310 * +--------------------+---------------+-------+
1311 *
1312 * <-------------- *tio->len_ptr --------------->
1313 * <----- bio_sectors ----->
1314 * <-- n_sectors -->
1315 *
1316 * Region 1 was already iterated over with bio_advance or similar function.
1317 * (it may be empty if the target doesn't use bio_advance)
1318 * Region 2 is the remaining bio size that the target wants to process.
1319 * (it may be empty if region 1 is non-empty, although there is no reason
1320 * to make it empty)
1321 * The target requires that region 3 is to be sent in the next bio.
1322 *
1323 * If the target wants to receive multiple copies of the bio (via num_*bios, etc),
1324 * the partially processed part (the sum of regions 1+2) must be the same for all
1325 * copies of the bio.
1326 */
1327void dm_accept_partial_bio(struct bio *bio, unsigned int n_sectors)
1328{
1329 struct dm_target_io *tio = clone_to_tio(bio);
1330 struct dm_io *io = tio->io;
1331 unsigned int bio_sectors = bio_sectors(bio);
1332
1333 BUG_ON(dm_tio_flagged(tio, DM_TIO_IS_DUPLICATE_BIO));
1334 BUG_ON(op_is_zone_mgmt(bio_op(bio)));
1335 BUG_ON(bio_op(bio) == REQ_OP_ZONE_APPEND);
1336 BUG_ON(bio_sectors > *tio->len_ptr);
1337 BUG_ON(n_sectors > bio_sectors);
1338
1339 *tio->len_ptr -= bio_sectors - n_sectors;
1340 bio->bi_iter.bi_size = n_sectors << SECTOR_SHIFT;
1341
1342 /*
1343 * __split_and_process_bio() may have already saved mapped part
1344 * for accounting but it is being reduced so update accordingly.
1345 */
1346 dm_io_set_flag(io, DM_IO_WAS_SPLIT);
1347 io->sectors = n_sectors;
1348 io->sector_offset = bio_sectors(io->orig_bio);
1349}
1350EXPORT_SYMBOL_GPL(dm_accept_partial_bio);
1351
1352/*
1353 * @clone: clone bio that DM core passed to target's .map function
1354 * @tgt_clone: clone of @clone bio that target needs submitted
1355 *
1356 * Targets should use this interface to submit bios they take
1357 * ownership of when returning DM_MAPIO_SUBMITTED.
1358 *
1359 * Target should also enable ti->accounts_remapped_io
1360 */
1361void dm_submit_bio_remap(struct bio *clone, struct bio *tgt_clone)
1362{
1363 struct dm_target_io *tio = clone_to_tio(clone);
1364 struct dm_io *io = tio->io;
1365
1366 /* establish bio that will get submitted */
1367 if (!tgt_clone)
1368 tgt_clone = clone;
1369
1370 /*
1371 * Account io->origin_bio to DM dev on behalf of target
1372 * that took ownership of IO with DM_MAPIO_SUBMITTED.
1373 */
1374 dm_start_io_acct(io, clone);
1375
1376 trace_block_bio_remap(tgt_clone, disk_devt(io->md->disk),
1377 tio->old_sector);
1378 submit_bio_noacct(tgt_clone);
1379}
1380EXPORT_SYMBOL_GPL(dm_submit_bio_remap);
1381
1382static noinline void __set_swap_bios_limit(struct mapped_device *md, int latch)
1383{
1384 mutex_lock(&md->swap_bios_lock);
1385 while (latch < md->swap_bios) {
1386 cond_resched();
1387 down(&md->swap_bios_semaphore);
1388 md->swap_bios--;
1389 }
1390 while (latch > md->swap_bios) {
1391 cond_resched();
1392 up(&md->swap_bios_semaphore);
1393 md->swap_bios++;
1394 }
1395 mutex_unlock(&md->swap_bios_lock);
1396}
1397
1398static void __map_bio(struct bio *clone)
1399{
1400 struct dm_target_io *tio = clone_to_tio(clone);
1401 struct dm_target *ti = tio->ti;
1402 struct dm_io *io = tio->io;
1403 struct mapped_device *md = io->md;
1404 int r;
1405
1406 clone->bi_end_io = clone_endio;
1407
1408 /*
1409 * Map the clone.
1410 */
1411 tio->old_sector = clone->bi_iter.bi_sector;
1412
1413 if (static_branch_unlikely(&swap_bios_enabled) &&
1414 unlikely(swap_bios_limit(ti, clone))) {
1415 int latch = get_swap_bios();
1416
1417 if (unlikely(latch != md->swap_bios))
1418 __set_swap_bios_limit(md, latch);
1419 down(&md->swap_bios_semaphore);
1420 }
1421
1422 if (static_branch_unlikely(&zoned_enabled)) {
1423 /*
1424 * Check if the IO needs a special mapping due to zone append
1425 * emulation on zoned target. In this case, dm_zone_map_bio()
1426 * calls the target map operation.
1427 */
1428 if (unlikely(dm_emulate_zone_append(md)))
1429 r = dm_zone_map_bio(tio);
1430 else
1431 goto do_map;
1432 } else {
1433do_map:
1434 if (likely(ti->type->map == linear_map))
1435 r = linear_map(ti, clone);
1436 else if (ti->type->map == stripe_map)
1437 r = stripe_map(ti, clone);
1438 else
1439 r = ti->type->map(ti, clone);
1440 }
1441
1442 switch (r) {
1443 case DM_MAPIO_SUBMITTED:
1444 /* target has assumed ownership of this io */
1445 if (!ti->accounts_remapped_io)
1446 dm_start_io_acct(io, clone);
1447 break;
1448 case DM_MAPIO_REMAPPED:
1449 dm_submit_bio_remap(clone, NULL);
1450 break;
1451 case DM_MAPIO_KILL:
1452 case DM_MAPIO_REQUEUE:
1453 if (static_branch_unlikely(&swap_bios_enabled) &&
1454 unlikely(swap_bios_limit(ti, clone)))
1455 up(&md->swap_bios_semaphore);
1456 free_tio(clone);
1457 if (r == DM_MAPIO_KILL)
1458 dm_io_dec_pending(io, BLK_STS_IOERR);
1459 else
1460 dm_io_dec_pending(io, BLK_STS_DM_REQUEUE);
1461 break;
1462 default:
1463 DMCRIT("unimplemented target map return value: %d", r);
1464 BUG();
1465 }
1466}
1467
1468static void setup_split_accounting(struct clone_info *ci, unsigned int len)
1469{
1470 struct dm_io *io = ci->io;
1471
1472 if (ci->sector_count > len) {
1473 /*
1474 * Split needed, save the mapped part for accounting.
1475 * NOTE: dm_accept_partial_bio() will update accordingly.
1476 */
1477 dm_io_set_flag(io, DM_IO_WAS_SPLIT);
1478 io->sectors = len;
1479 io->sector_offset = bio_sectors(ci->bio);
1480 }
1481}
1482
1483static void alloc_multiple_bios(struct bio_list *blist, struct clone_info *ci,
1484 struct dm_target *ti, unsigned int num_bios,
1485 unsigned *len, gfp_t gfp_flag)
1486{
1487 struct bio *bio;
1488 int try = (gfp_flag & GFP_NOWAIT) ? 0 : 1;
1489
1490 for (; try < 2; try++) {
1491 int bio_nr;
1492
1493 if (try && num_bios > 1)
1494 mutex_lock(&ci->io->md->table_devices_lock);
1495 for (bio_nr = 0; bio_nr < num_bios; bio_nr++) {
1496 bio = alloc_tio(ci, ti, bio_nr, len,
1497 try ? GFP_NOIO : GFP_NOWAIT);
1498 if (!bio)
1499 break;
1500
1501 bio_list_add(blist, bio);
1502 }
1503 if (try && num_bios > 1)
1504 mutex_unlock(&ci->io->md->table_devices_lock);
1505 if (bio_nr == num_bios)
1506 return;
1507
1508 while ((bio = bio_list_pop(blist)))
1509 free_tio(bio);
1510 }
1511}
1512
1513static unsigned int __send_duplicate_bios(struct clone_info *ci, struct dm_target *ti,
1514 unsigned int num_bios, unsigned int *len,
1515 gfp_t gfp_flag)
1516{
1517 struct bio_list blist = BIO_EMPTY_LIST;
1518 struct bio *clone;
1519 unsigned int ret = 0;
1520
1521 if (WARN_ON_ONCE(num_bios == 0)) /* num_bios = 0 is a bug in caller */
1522 return 0;
1523
1524 /* dm_accept_partial_bio() is not supported with shared tio->len_ptr */
1525 if (len)
1526 setup_split_accounting(ci, *len);
1527
1528 /*
1529 * Using alloc_multiple_bios(), even if num_bios is 1, to consistently
1530 * support allocating using GFP_NOWAIT with GFP_NOIO fallback.
1531 */
1532 alloc_multiple_bios(&blist, ci, ti, num_bios, len, gfp_flag);
1533 while ((clone = bio_list_pop(&blist))) {
1534 if (num_bios > 1)
1535 dm_tio_set_flag(clone_to_tio(clone), DM_TIO_IS_DUPLICATE_BIO);
1536 __map_bio(clone);
1537 ret += 1;
1538 }
1539
1540 return ret;
1541}
1542
1543static void __send_empty_flush(struct clone_info *ci)
1544{
1545 struct dm_table *t = ci->map;
1546 struct bio flush_bio;
1547
1548 /*
1549 * Use an on-stack bio for this, it's safe since we don't
1550 * need to reference it after submit. It's just used as
1551 * the basis for the clone(s).
1552 */
1553 bio_init(&flush_bio, ci->io->md->disk->part0, NULL, 0,
1554 REQ_OP_WRITE | REQ_PREFLUSH | REQ_SYNC);
1555
1556 ci->bio = &flush_bio;
1557 ci->sector_count = 0;
1558 ci->io->tio.clone.bi_iter.bi_size = 0;
1559
1560 for (unsigned int i = 0; i < t->num_targets; i++) {
1561 unsigned int bios;
1562 struct dm_target *ti = dm_table_get_target(t, i);
1563
1564 if (unlikely(ti->num_flush_bios == 0))
1565 continue;
1566
1567 atomic_add(ti->num_flush_bios, &ci->io->io_count);
1568 bios = __send_duplicate_bios(ci, ti, ti->num_flush_bios,
1569 NULL, GFP_NOWAIT);
1570 atomic_sub(ti->num_flush_bios - bios, &ci->io->io_count);
1571 }
1572
1573 /*
1574 * alloc_io() takes one extra reference for submission, so the
1575 * reference won't reach 0 without the following subtraction
1576 */
1577 atomic_sub(1, &ci->io->io_count);
1578
1579 bio_uninit(ci->bio);
1580}
1581
1582static void __send_abnormal_io(struct clone_info *ci, struct dm_target *ti,
1583 unsigned int num_bios, unsigned int max_granularity,
1584 unsigned int max_sectors)
1585{
1586 unsigned int len, bios;
1587
1588 len = min_t(sector_t, ci->sector_count,
1589 __max_io_len(ti, ci->sector, max_granularity, max_sectors));
1590
1591 atomic_add(num_bios, &ci->io->io_count);
1592 bios = __send_duplicate_bios(ci, ti, num_bios, &len, GFP_NOIO);
1593 /*
1594 * alloc_io() takes one extra reference for submission, so the
1595 * reference won't reach 0 without the following (+1) subtraction
1596 */
1597 atomic_sub(num_bios - bios + 1, &ci->io->io_count);
1598
1599 ci->sector += len;
1600 ci->sector_count -= len;
1601}
1602
1603static bool is_abnormal_io(struct bio *bio)
1604{
1605 enum req_op op = bio_op(bio);
1606
1607 if (op != REQ_OP_READ && op != REQ_OP_WRITE && op != REQ_OP_FLUSH) {
1608 switch (op) {
1609 case REQ_OP_DISCARD:
1610 case REQ_OP_SECURE_ERASE:
1611 case REQ_OP_WRITE_ZEROES:
1612 return true;
1613 default:
1614 break;
1615 }
1616 }
1617
1618 return false;
1619}
1620
1621static blk_status_t __process_abnormal_io(struct clone_info *ci,
1622 struct dm_target *ti)
1623{
1624 unsigned int num_bios = 0;
1625 unsigned int max_granularity = 0;
1626 unsigned int max_sectors = 0;
1627 struct queue_limits *limits = dm_get_queue_limits(ti->table->md);
1628
1629 switch (bio_op(ci->bio)) {
1630 case REQ_OP_DISCARD:
1631 num_bios = ti->num_discard_bios;
1632 max_sectors = limits->max_discard_sectors;
1633 if (ti->max_discard_granularity)
1634 max_granularity = max_sectors;
1635 break;
1636 case REQ_OP_SECURE_ERASE:
1637 num_bios = ti->num_secure_erase_bios;
1638 max_sectors = limits->max_secure_erase_sectors;
1639 if (ti->max_secure_erase_granularity)
1640 max_granularity = max_sectors;
1641 break;
1642 case REQ_OP_WRITE_ZEROES:
1643 num_bios = ti->num_write_zeroes_bios;
1644 max_sectors = limits->max_write_zeroes_sectors;
1645 if (ti->max_write_zeroes_granularity)
1646 max_granularity = max_sectors;
1647 break;
1648 default:
1649 break;
1650 }
1651
1652 /*
1653 * Even though the device advertised support for this type of
1654 * request, that does not mean every target supports it, and
1655 * reconfiguration might also have changed that since the
1656 * check was performed.
1657 */
1658 if (unlikely(!num_bios))
1659 return BLK_STS_NOTSUPP;
1660
1661 __send_abnormal_io(ci, ti, num_bios, max_granularity, max_sectors);
1662
1663 return BLK_STS_OK;
1664}
1665
1666/*
1667 * Reuse ->bi_private as dm_io list head for storing all dm_io instances
1668 * associated with this bio, and this bio's bi_private needs to be
1669 * stored in dm_io->data before the reuse.
1670 *
1671 * bio->bi_private is owned by fs or upper layer, so block layer won't
1672 * touch it after splitting. Meantime it won't be changed by anyone after
1673 * bio is submitted. So this reuse is safe.
1674 */
1675static inline struct dm_io **dm_poll_list_head(struct bio *bio)
1676{
1677 return (struct dm_io **)&bio->bi_private;
1678}
1679
1680static void dm_queue_poll_io(struct bio *bio, struct dm_io *io)
1681{
1682 struct dm_io **head = dm_poll_list_head(bio);
1683
1684 if (!(bio->bi_opf & REQ_DM_POLL_LIST)) {
1685 bio->bi_opf |= REQ_DM_POLL_LIST;
1686 /*
1687 * Save .bi_private into dm_io, so that we can reuse
1688 * .bi_private as dm_io list head for storing dm_io list
1689 */
1690 io->data = bio->bi_private;
1691
1692 /* tell block layer to poll for completion */
1693 bio->bi_cookie = ~BLK_QC_T_NONE;
1694
1695 io->next = NULL;
1696 } else {
1697 /*
1698 * bio recursed due to split, reuse original poll list,
1699 * and save bio->bi_private too.
1700 */
1701 io->data = (*head)->data;
1702 io->next = *head;
1703 }
1704
1705 *head = io;
1706}
1707
1708/*
1709 * Select the correct strategy for processing a non-flush bio.
1710 */
1711static blk_status_t __split_and_process_bio(struct clone_info *ci)
1712{
1713 struct bio *clone;
1714 struct dm_target *ti;
1715 unsigned int len;
1716
1717 ti = dm_table_find_target(ci->map, ci->sector);
1718 if (unlikely(!ti))
1719 return BLK_STS_IOERR;
1720
1721 if (unlikely(ci->is_abnormal_io))
1722 return __process_abnormal_io(ci, ti);
1723
1724 /*
1725 * Only support bio polling for normal IO, and the target io is
1726 * exactly inside the dm_io instance (verified in dm_poll_dm_io)
1727 */
1728 ci->submit_as_polled = !!(ci->bio->bi_opf & REQ_POLLED);
1729
1730 len = min_t(sector_t, max_io_len(ti, ci->sector), ci->sector_count);
1731 setup_split_accounting(ci, len);
1732
1733 if (unlikely(ci->bio->bi_opf & REQ_NOWAIT)) {
1734 if (unlikely(!dm_target_supports_nowait(ti->type)))
1735 return BLK_STS_NOTSUPP;
1736
1737 clone = alloc_tio(ci, ti, 0, &len, GFP_NOWAIT);
1738 if (unlikely(!clone))
1739 return BLK_STS_AGAIN;
1740 } else {
1741 clone = alloc_tio(ci, ti, 0, &len, GFP_NOIO);
1742 }
1743 __map_bio(clone);
1744
1745 ci->sector += len;
1746 ci->sector_count -= len;
1747
1748 return BLK_STS_OK;
1749}
1750
1751static void init_clone_info(struct clone_info *ci, struct dm_io *io,
1752 struct dm_table *map, struct bio *bio, bool is_abnormal)
1753{
1754 ci->map = map;
1755 ci->io = io;
1756 ci->bio = bio;
1757 ci->is_abnormal_io = is_abnormal;
1758 ci->submit_as_polled = false;
1759 ci->sector = bio->bi_iter.bi_sector;
1760 ci->sector_count = bio_sectors(bio);
1761
1762 /* Shouldn't happen but sector_count was being set to 0 so... */
1763 if (static_branch_unlikely(&zoned_enabled) &&
1764 WARN_ON_ONCE(op_is_zone_mgmt(bio_op(bio)) && ci->sector_count))
1765 ci->sector_count = 0;
1766}
1767
1768/*
1769 * Entry point to split a bio into clones and submit them to the targets.
1770 */
1771static void dm_split_and_process_bio(struct mapped_device *md,
1772 struct dm_table *map, struct bio *bio)
1773{
1774 struct clone_info ci;
1775 struct dm_io *io;
1776 blk_status_t error = BLK_STS_OK;
1777 bool is_abnormal;
1778
1779 is_abnormal = is_abnormal_io(bio);
1780 if (unlikely(is_abnormal)) {
1781 /*
1782 * Use bio_split_to_limits() for abnormal IO (e.g. discard, etc)
1783 * otherwise associated queue_limits won't be imposed.
1784 */
1785 bio = bio_split_to_limits(bio);
1786 if (!bio)
1787 return;
1788 }
1789
1790 /* Only support nowait for normal IO */
1791 if (unlikely(bio->bi_opf & REQ_NOWAIT) && !is_abnormal) {
1792 io = alloc_io(md, bio, GFP_NOWAIT);
1793 if (unlikely(!io)) {
1794 /* Unable to do anything without dm_io. */
1795 bio_wouldblock_error(bio);
1796 return;
1797 }
1798 } else {
1799 io = alloc_io(md, bio, GFP_NOIO);
1800 }
1801 init_clone_info(&ci, io, map, bio, is_abnormal);
1802
1803 if (bio->bi_opf & REQ_PREFLUSH) {
1804 __send_empty_flush(&ci);
1805 /* dm_io_complete submits any data associated with flush */
1806 goto out;
1807 }
1808
1809 error = __split_and_process_bio(&ci);
1810 if (error || !ci.sector_count)
1811 goto out;
1812 /*
1813 * Remainder must be passed to submit_bio_noacct() so it gets handled
1814 * *after* bios already submitted have been completely processed.
1815 */
1816 bio_trim(bio, io->sectors, ci.sector_count);
1817 trace_block_split(bio, bio->bi_iter.bi_sector);
1818 bio_inc_remaining(bio);
1819 submit_bio_noacct(bio);
1820out:
1821 /*
1822 * Drop the extra reference count for non-POLLED bio, and hold one
1823 * reference for POLLED bio, which will be released in dm_poll_bio
1824 *
1825 * Add every dm_io instance into the dm_io list head which is stored
1826 * in bio->bi_private, so that dm_poll_bio can poll them all.
1827 */
1828 if (error || !ci.submit_as_polled) {
1829 /*
1830 * In case of submission failure, the extra reference for
1831 * submitting io isn't consumed yet
1832 */
1833 if (error)
1834 atomic_dec(&io->io_count);
1835 dm_io_dec_pending(io, error);
1836 } else
1837 dm_queue_poll_io(bio, io);
1838}
1839
1840static void dm_submit_bio(struct bio *bio)
1841{
1842 struct mapped_device *md = bio->bi_bdev->bd_disk->private_data;
1843 int srcu_idx;
1844 struct dm_table *map;
1845
1846 map = dm_get_live_table(md, &srcu_idx);
1847
1848 /* If suspended, or map not yet available, queue this IO for later */
1849 if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) ||
1850 unlikely(!map)) {
1851 if (bio->bi_opf & REQ_NOWAIT)
1852 bio_wouldblock_error(bio);
1853 else if (bio->bi_opf & REQ_RAHEAD)
1854 bio_io_error(bio);
1855 else
1856 queue_io(md, bio);
1857 goto out;
1858 }
1859
1860 dm_split_and_process_bio(md, map, bio);
1861out:
1862 dm_put_live_table(md, srcu_idx);
1863}
1864
1865static bool dm_poll_dm_io(struct dm_io *io, struct io_comp_batch *iob,
1866 unsigned int flags)
1867{
1868 WARN_ON_ONCE(!dm_tio_is_normal(&io->tio));
1869
1870 /* don't poll if the mapped io is done */
1871 if (atomic_read(&io->io_count) > 1)
1872 bio_poll(&io->tio.clone, iob, flags);
1873
1874 /* bio_poll holds the last reference */
1875 return atomic_read(&io->io_count) == 1;
1876}
1877
1878static int dm_poll_bio(struct bio *bio, struct io_comp_batch *iob,
1879 unsigned int flags)
1880{
1881 struct dm_io **head = dm_poll_list_head(bio);
1882 struct dm_io *list = *head;
1883 struct dm_io *tmp = NULL;
1884 struct dm_io *curr, *next;
1885
1886 /* Only poll normal bio which was marked as REQ_DM_POLL_LIST */
1887 if (!(bio->bi_opf & REQ_DM_POLL_LIST))
1888 return 0;
1889
1890 WARN_ON_ONCE(!list);
1891
1892 /*
1893 * Restore .bi_private before possibly completing dm_io.
1894 *
1895 * bio_poll() is only possible once @bio has been completely
1896 * submitted via submit_bio_noacct()'s depth-first submission.
1897 * So there is no dm_queue_poll_io() race associated with
1898 * clearing REQ_DM_POLL_LIST here.
1899 */
1900 bio->bi_opf &= ~REQ_DM_POLL_LIST;
1901 bio->bi_private = list->data;
1902
1903 for (curr = list, next = curr->next; curr; curr = next, next =
1904 curr ? curr->next : NULL) {
1905 if (dm_poll_dm_io(curr, iob, flags)) {
1906 /*
1907 * clone_endio() has already occurred, so no
1908 * error handling is needed here.
1909 */
1910 __dm_io_dec_pending(curr);
1911 } else {
1912 curr->next = tmp;
1913 tmp = curr;
1914 }
1915 }
1916
1917 /* Not done? */
1918 if (tmp) {
1919 bio->bi_opf |= REQ_DM_POLL_LIST;
1920 /* Reset bio->bi_private to dm_io list head */
1921 *head = tmp;
1922 return 0;
1923 }
1924 return 1;
1925}
1926
1927/*
1928 *---------------------------------------------------------------
1929 * An IDR is used to keep track of allocated minor numbers.
1930 *---------------------------------------------------------------
1931 */
1932static void free_minor(int minor)
1933{
1934 spin_lock(&_minor_lock);
1935 idr_remove(&_minor_idr, minor);
1936 spin_unlock(&_minor_lock);
1937}
1938
1939/*
1940 * See if the device with a specific minor # is free.
1941 */
1942static int specific_minor(int minor)
1943{
1944 int r;
1945
1946 if (minor >= (1 << MINORBITS))
1947 return -EINVAL;
1948
1949 idr_preload(GFP_KERNEL);
1950 spin_lock(&_minor_lock);
1951
1952 r = idr_alloc(&_minor_idr, MINOR_ALLOCED, minor, minor + 1, GFP_NOWAIT);
1953
1954 spin_unlock(&_minor_lock);
1955 idr_preload_end();
1956 if (r < 0)
1957 return r == -ENOSPC ? -EBUSY : r;
1958 return 0;
1959}
1960
1961static int next_free_minor(int *minor)
1962{
1963 int r;
1964
1965 idr_preload(GFP_KERNEL);
1966 spin_lock(&_minor_lock);
1967
1968 r = idr_alloc(&_minor_idr, MINOR_ALLOCED, 0, 1 << MINORBITS, GFP_NOWAIT);
1969
1970 spin_unlock(&_minor_lock);
1971 idr_preload_end();
1972 if (r < 0)
1973 return r;
1974 *minor = r;
1975 return 0;
1976}
1977
1978static const struct block_device_operations dm_blk_dops;
1979static const struct block_device_operations dm_rq_blk_dops;
1980static const struct dax_operations dm_dax_ops;
1981
1982static void dm_wq_work(struct work_struct *work);
1983
1984#ifdef CONFIG_BLK_INLINE_ENCRYPTION
1985static void dm_queue_destroy_crypto_profile(struct request_queue *q)
1986{
1987 dm_destroy_crypto_profile(q->crypto_profile);
1988}
1989
1990#else /* CONFIG_BLK_INLINE_ENCRYPTION */
1991
1992static inline void dm_queue_destroy_crypto_profile(struct request_queue *q)
1993{
1994}
1995#endif /* !CONFIG_BLK_INLINE_ENCRYPTION */
1996
1997static void cleanup_mapped_device(struct mapped_device *md)
1998{
1999 if (md->wq)
2000 destroy_workqueue(md->wq);
2001 dm_free_md_mempools(md->mempools);
2002
2003 if (md->dax_dev) {
2004 dax_remove_host(md->disk);
2005 kill_dax(md->dax_dev);
2006 put_dax(md->dax_dev);
2007 md->dax_dev = NULL;
2008 }
2009
2010 dm_cleanup_zoned_dev(md);
2011 if (md->disk) {
2012 spin_lock(&_minor_lock);
2013 md->disk->private_data = NULL;
2014 spin_unlock(&_minor_lock);
2015 if (dm_get_md_type(md) != DM_TYPE_NONE) {
2016 struct table_device *td;
2017
2018 dm_sysfs_exit(md);
2019 list_for_each_entry(td, &md->table_devices, list) {
2020 bd_unlink_disk_holder(td->dm_dev.bdev,
2021 md->disk);
2022 }
2023
2024 /*
2025 * Hold lock to make sure del_gendisk() won't concurrent
2026 * with open/close_table_device().
2027 */
2028 mutex_lock(&md->table_devices_lock);
2029 del_gendisk(md->disk);
2030 mutex_unlock(&md->table_devices_lock);
2031 }
2032 dm_queue_destroy_crypto_profile(md->queue);
2033 put_disk(md->disk);
2034 }
2035
2036 if (md->pending_io) {
2037 free_percpu(md->pending_io);
2038 md->pending_io = NULL;
2039 }
2040
2041 cleanup_srcu_struct(&md->io_barrier);
2042
2043 mutex_destroy(&md->suspend_lock);
2044 mutex_destroy(&md->type_lock);
2045 mutex_destroy(&md->table_devices_lock);
2046 mutex_destroy(&md->swap_bios_lock);
2047
2048 dm_mq_cleanup_mapped_device(md);
2049}
2050
2051/*
2052 * Allocate and initialise a blank device with a given minor.
2053 */
2054static struct mapped_device *alloc_dev(int minor)
2055{
2056 int r, numa_node_id = dm_get_numa_node();
2057 struct mapped_device *md;
2058 void *old_md;
2059
2060 md = kvzalloc_node(sizeof(*md), GFP_KERNEL, numa_node_id);
2061 if (!md) {
2062 DMERR("unable to allocate device, out of memory.");
2063 return NULL;
2064 }
2065
2066 if (!try_module_get(THIS_MODULE))
2067 goto bad_module_get;
2068
2069 /* get a minor number for the dev */
2070 if (minor == DM_ANY_MINOR)
2071 r = next_free_minor(&minor);
2072 else
2073 r = specific_minor(minor);
2074 if (r < 0)
2075 goto bad_minor;
2076
2077 r = init_srcu_struct(&md->io_barrier);
2078 if (r < 0)
2079 goto bad_io_barrier;
2080
2081 md->numa_node_id = numa_node_id;
2082 md->init_tio_pdu = false;
2083 md->type = DM_TYPE_NONE;
2084 mutex_init(&md->suspend_lock);
2085 mutex_init(&md->type_lock);
2086 mutex_init(&md->table_devices_lock);
2087 spin_lock_init(&md->deferred_lock);
2088 atomic_set(&md->holders, 1);
2089 atomic_set(&md->open_count, 0);
2090 atomic_set(&md->event_nr, 0);
2091 atomic_set(&md->uevent_seq, 0);
2092 INIT_LIST_HEAD(&md->uevent_list);
2093 INIT_LIST_HEAD(&md->table_devices);
2094 spin_lock_init(&md->uevent_lock);
2095
2096 /*
2097 * default to bio-based until DM table is loaded and md->type
2098 * established. If request-based table is loaded: blk-mq will
2099 * override accordingly.
2100 */
2101 md->disk = blk_alloc_disk(md->numa_node_id);
2102 if (!md->disk)
2103 goto bad;
2104 md->queue = md->disk->queue;
2105
2106 init_waitqueue_head(&md->wait);
2107 INIT_WORK(&md->work, dm_wq_work);
2108 INIT_WORK(&md->requeue_work, dm_wq_requeue_work);
2109 init_waitqueue_head(&md->eventq);
2110 init_completion(&md->kobj_holder.completion);
2111
2112 md->requeue_list = NULL;
2113 md->swap_bios = get_swap_bios();
2114 sema_init(&md->swap_bios_semaphore, md->swap_bios);
2115 mutex_init(&md->swap_bios_lock);
2116
2117 md->disk->major = _major;
2118 md->disk->first_minor = minor;
2119 md->disk->minors = 1;
2120 md->disk->flags |= GENHD_FL_NO_PART;
2121 md->disk->fops = &dm_blk_dops;
2122 md->disk->private_data = md;
2123 sprintf(md->disk->disk_name, "dm-%d", minor);
2124
2125 if (IS_ENABLED(CONFIG_FS_DAX)) {
2126 md->dax_dev = alloc_dax(md, &dm_dax_ops);
2127 if (IS_ERR(md->dax_dev)) {
2128 md->dax_dev = NULL;
2129 goto bad;
2130 }
2131 set_dax_nocache(md->dax_dev);
2132 set_dax_nomc(md->dax_dev);
2133 if (dax_add_host(md->dax_dev, md->disk))
2134 goto bad;
2135 }
2136
2137 format_dev_t(md->name, MKDEV(_major, minor));
2138
2139 md->wq = alloc_workqueue("kdmflush/%s", WQ_MEM_RECLAIM, 0, md->name);
2140 if (!md->wq)
2141 goto bad;
2142
2143 md->pending_io = alloc_percpu(unsigned long);
2144 if (!md->pending_io)
2145 goto bad;
2146
2147 r = dm_stats_init(&md->stats);
2148 if (r < 0)
2149 goto bad;
2150
2151 /* Populate the mapping, nobody knows we exist yet */
2152 spin_lock(&_minor_lock);
2153 old_md = idr_replace(&_minor_idr, md, minor);
2154 spin_unlock(&_minor_lock);
2155
2156 BUG_ON(old_md != MINOR_ALLOCED);
2157
2158 return md;
2159
2160bad:
2161 cleanup_mapped_device(md);
2162bad_io_barrier:
2163 free_minor(minor);
2164bad_minor:
2165 module_put(THIS_MODULE);
2166bad_module_get:
2167 kvfree(md);
2168 return NULL;
2169}
2170
2171static void unlock_fs(struct mapped_device *md);
2172
2173static void free_dev(struct mapped_device *md)
2174{
2175 int minor = MINOR(disk_devt(md->disk));
2176
2177 unlock_fs(md);
2178
2179 cleanup_mapped_device(md);
2180
2181 WARN_ON_ONCE(!list_empty(&md->table_devices));
2182 dm_stats_cleanup(&md->stats);
2183 free_minor(minor);
2184
2185 module_put(THIS_MODULE);
2186 kvfree(md);
2187}
2188
2189/*
2190 * Bind a table to the device.
2191 */
2192static void event_callback(void *context)
2193{
2194 unsigned long flags;
2195 LIST_HEAD(uevents);
2196 struct mapped_device *md = context;
2197
2198 spin_lock_irqsave(&md->uevent_lock, flags);
2199 list_splice_init(&md->uevent_list, &uevents);
2200 spin_unlock_irqrestore(&md->uevent_lock, flags);
2201
2202 dm_send_uevents(&uevents, &disk_to_dev(md->disk)->kobj);
2203
2204 atomic_inc(&md->event_nr);
2205 wake_up(&md->eventq);
2206 dm_issue_global_event();
2207}
2208
2209/*
2210 * Returns old map, which caller must destroy.
2211 */
2212static struct dm_table *__bind(struct mapped_device *md, struct dm_table *t,
2213 struct queue_limits *limits)
2214{
2215 struct dm_table *old_map;
2216 sector_t size;
2217 int ret;
2218
2219 lockdep_assert_held(&md->suspend_lock);
2220
2221 size = dm_table_get_size(t);
2222
2223 /*
2224 * Wipe any geometry if the size of the table changed.
2225 */
2226 if (size != dm_get_size(md))
2227 memset(&md->geometry, 0, sizeof(md->geometry));
2228
2229 set_capacity(md->disk, size);
2230
2231 dm_table_event_callback(t, event_callback, md);
2232
2233 if (dm_table_request_based(t)) {
2234 /*
2235 * Leverage the fact that request-based DM targets are
2236 * immutable singletons - used to optimize dm_mq_queue_rq.
2237 */
2238 md->immutable_target = dm_table_get_immutable_target(t);
2239
2240 /*
2241 * There is no need to reload with request-based dm because the
2242 * size of front_pad doesn't change.
2243 *
2244 * Note for future: If you are to reload bioset, prep-ed
2245 * requests in the queue may refer to bio from the old bioset,
2246 * so you must walk through the queue to unprep.
2247 */
2248 if (!md->mempools) {
2249 md->mempools = t->mempools;
2250 t->mempools = NULL;
2251 }
2252 } else {
2253 /*
2254 * The md may already have mempools that need changing.
2255 * If so, reload bioset because front_pad may have changed
2256 * because a different table was loaded.
2257 */
2258 dm_free_md_mempools(md->mempools);
2259 md->mempools = t->mempools;
2260 t->mempools = NULL;
2261 }
2262
2263 ret = dm_table_set_restrictions(t, md->queue, limits);
2264 if (ret) {
2265 old_map = ERR_PTR(ret);
2266 goto out;
2267 }
2268
2269 old_map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2270 rcu_assign_pointer(md->map, (void *)t);
2271 md->immutable_target_type = dm_table_get_immutable_target_type(t);
2272
2273 if (old_map)
2274 dm_sync_table(md);
2275out:
2276 return old_map;
2277}
2278
2279/*
2280 * Returns unbound table for the caller to free.
2281 */
2282static struct dm_table *__unbind(struct mapped_device *md)
2283{
2284 struct dm_table *map = rcu_dereference_protected(md->map, 1);
2285
2286 if (!map)
2287 return NULL;
2288
2289 dm_table_event_callback(map, NULL, NULL);
2290 RCU_INIT_POINTER(md->map, NULL);
2291 dm_sync_table(md);
2292
2293 return map;
2294}
2295
2296/*
2297 * Constructor for a new device.
2298 */
2299int dm_create(int minor, struct mapped_device **result)
2300{
2301 struct mapped_device *md;
2302
2303 md = alloc_dev(minor);
2304 if (!md)
2305 return -ENXIO;
2306
2307 dm_ima_reset_data(md);
2308
2309 *result = md;
2310 return 0;
2311}
2312
2313/*
2314 * Functions to manage md->type.
2315 * All are required to hold md->type_lock.
2316 */
2317void dm_lock_md_type(struct mapped_device *md)
2318{
2319 mutex_lock(&md->type_lock);
2320}
2321
2322void dm_unlock_md_type(struct mapped_device *md)
2323{
2324 mutex_unlock(&md->type_lock);
2325}
2326
2327void dm_set_md_type(struct mapped_device *md, enum dm_queue_mode type)
2328{
2329 BUG_ON(!mutex_is_locked(&md->type_lock));
2330 md->type = type;
2331}
2332
2333enum dm_queue_mode dm_get_md_type(struct mapped_device *md)
2334{
2335 return md->type;
2336}
2337
2338struct target_type *dm_get_immutable_target_type(struct mapped_device *md)
2339{
2340 return md->immutable_target_type;
2341}
2342
2343/*
2344 * Setup the DM device's queue based on md's type
2345 */
2346int dm_setup_md_queue(struct mapped_device *md, struct dm_table *t)
2347{
2348 enum dm_queue_mode type = dm_table_get_type(t);
2349 struct queue_limits limits;
2350 struct table_device *td;
2351 int r;
2352
2353 switch (type) {
2354 case DM_TYPE_REQUEST_BASED:
2355 md->disk->fops = &dm_rq_blk_dops;
2356 r = dm_mq_init_request_queue(md, t);
2357 if (r) {
2358 DMERR("Cannot initialize queue for request-based dm mapped device");
2359 return r;
2360 }
2361 break;
2362 case DM_TYPE_BIO_BASED:
2363 case DM_TYPE_DAX_BIO_BASED:
2364 blk_queue_flag_set(QUEUE_FLAG_IO_STAT, md->queue);
2365 break;
2366 case DM_TYPE_NONE:
2367 WARN_ON_ONCE(true);
2368 break;
2369 }
2370
2371 r = dm_calculate_queue_limits(t, &limits);
2372 if (r) {
2373 DMERR("Cannot calculate initial queue limits");
2374 return r;
2375 }
2376 r = dm_table_set_restrictions(t, md->queue, &limits);
2377 if (r)
2378 return r;
2379
2380 /*
2381 * Hold lock to make sure add_disk() and del_gendisk() won't concurrent
2382 * with open_table_device() and close_table_device().
2383 */
2384 mutex_lock(&md->table_devices_lock);
2385 r = add_disk(md->disk);
2386 mutex_unlock(&md->table_devices_lock);
2387 if (r)
2388 return r;
2389
2390 /*
2391 * Register the holder relationship for devices added before the disk
2392 * was live.
2393 */
2394 list_for_each_entry(td, &md->table_devices, list) {
2395 r = bd_link_disk_holder(td->dm_dev.bdev, md->disk);
2396 if (r)
2397 goto out_undo_holders;
2398 }
2399
2400 r = dm_sysfs_init(md);
2401 if (r)
2402 goto out_undo_holders;
2403
2404 md->type = type;
2405 return 0;
2406
2407out_undo_holders:
2408 list_for_each_entry_continue_reverse(td, &md->table_devices, list)
2409 bd_unlink_disk_holder(td->dm_dev.bdev, md->disk);
2410 mutex_lock(&md->table_devices_lock);
2411 del_gendisk(md->disk);
2412 mutex_unlock(&md->table_devices_lock);
2413 return r;
2414}
2415
2416struct mapped_device *dm_get_md(dev_t dev)
2417{
2418 struct mapped_device *md;
2419 unsigned int minor = MINOR(dev);
2420
2421 if (MAJOR(dev) != _major || minor >= (1 << MINORBITS))
2422 return NULL;
2423
2424 spin_lock(&_minor_lock);
2425
2426 md = idr_find(&_minor_idr, minor);
2427 if (!md || md == MINOR_ALLOCED || (MINOR(disk_devt(dm_disk(md))) != minor) ||
2428 test_bit(DMF_FREEING, &md->flags) || dm_deleting_md(md)) {
2429 md = NULL;
2430 goto out;
2431 }
2432 dm_get(md);
2433out:
2434 spin_unlock(&_minor_lock);
2435
2436 return md;
2437}
2438EXPORT_SYMBOL_GPL(dm_get_md);
2439
2440void *dm_get_mdptr(struct mapped_device *md)
2441{
2442 return md->interface_ptr;
2443}
2444
2445void dm_set_mdptr(struct mapped_device *md, void *ptr)
2446{
2447 md->interface_ptr = ptr;
2448}
2449
2450void dm_get(struct mapped_device *md)
2451{
2452 atomic_inc(&md->holders);
2453 BUG_ON(test_bit(DMF_FREEING, &md->flags));
2454}
2455
2456int dm_hold(struct mapped_device *md)
2457{
2458 spin_lock(&_minor_lock);
2459 if (test_bit(DMF_FREEING, &md->flags)) {
2460 spin_unlock(&_minor_lock);
2461 return -EBUSY;
2462 }
2463 dm_get(md);
2464 spin_unlock(&_minor_lock);
2465 return 0;
2466}
2467EXPORT_SYMBOL_GPL(dm_hold);
2468
2469const char *dm_device_name(struct mapped_device *md)
2470{
2471 return md->name;
2472}
2473EXPORT_SYMBOL_GPL(dm_device_name);
2474
2475static void __dm_destroy(struct mapped_device *md, bool wait)
2476{
2477 struct dm_table *map;
2478 int srcu_idx;
2479
2480 might_sleep();
2481
2482 spin_lock(&_minor_lock);
2483 idr_replace(&_minor_idr, MINOR_ALLOCED, MINOR(disk_devt(dm_disk(md))));
2484 set_bit(DMF_FREEING, &md->flags);
2485 spin_unlock(&_minor_lock);
2486
2487 blk_mark_disk_dead(md->disk);
2488
2489 /*
2490 * Take suspend_lock so that presuspend and postsuspend methods
2491 * do not race with internal suspend.
2492 */
2493 mutex_lock(&md->suspend_lock);
2494 map = dm_get_live_table(md, &srcu_idx);
2495 if (!dm_suspended_md(md)) {
2496 dm_table_presuspend_targets(map);
2497 set_bit(DMF_SUSPENDED, &md->flags);
2498 set_bit(DMF_POST_SUSPENDING, &md->flags);
2499 dm_table_postsuspend_targets(map);
2500 }
2501 /* dm_put_live_table must be before fsleep, otherwise deadlock is possible */
2502 dm_put_live_table(md, srcu_idx);
2503 mutex_unlock(&md->suspend_lock);
2504
2505 /*
2506 * Rare, but there may be I/O requests still going to complete,
2507 * for example. Wait for all references to disappear.
2508 * No one should increment the reference count of the mapped_device,
2509 * after the mapped_device state becomes DMF_FREEING.
2510 */
2511 if (wait)
2512 while (atomic_read(&md->holders))
2513 fsleep(1000);
2514 else if (atomic_read(&md->holders))
2515 DMWARN("%s: Forcibly removing mapped_device still in use! (%d users)",
2516 dm_device_name(md), atomic_read(&md->holders));
2517
2518 dm_table_destroy(__unbind(md));
2519 free_dev(md);
2520}
2521
2522void dm_destroy(struct mapped_device *md)
2523{
2524 __dm_destroy(md, true);
2525}
2526
2527void dm_destroy_immediate(struct mapped_device *md)
2528{
2529 __dm_destroy(md, false);
2530}
2531
2532void dm_put(struct mapped_device *md)
2533{
2534 atomic_dec(&md->holders);
2535}
2536EXPORT_SYMBOL_GPL(dm_put);
2537
2538static bool dm_in_flight_bios(struct mapped_device *md)
2539{
2540 int cpu;
2541 unsigned long sum = 0;
2542
2543 for_each_possible_cpu(cpu)
2544 sum += *per_cpu_ptr(md->pending_io, cpu);
2545
2546 return sum != 0;
2547}
2548
2549static int dm_wait_for_bios_completion(struct mapped_device *md, unsigned int task_state)
2550{
2551 int r = 0;
2552 DEFINE_WAIT(wait);
2553
2554 while (true) {
2555 prepare_to_wait(&md->wait, &wait, task_state);
2556
2557 if (!dm_in_flight_bios(md))
2558 break;
2559
2560 if (signal_pending_state(task_state, current)) {
2561 r = -EINTR;
2562 break;
2563 }
2564
2565 io_schedule();
2566 }
2567 finish_wait(&md->wait, &wait);
2568
2569 smp_rmb();
2570
2571 return r;
2572}
2573
2574static int dm_wait_for_completion(struct mapped_device *md, unsigned int task_state)
2575{
2576 int r = 0;
2577
2578 if (!queue_is_mq(md->queue))
2579 return dm_wait_for_bios_completion(md, task_state);
2580
2581 while (true) {
2582 if (!blk_mq_queue_inflight(md->queue))
2583 break;
2584
2585 if (signal_pending_state(task_state, current)) {
2586 r = -EINTR;
2587 break;
2588 }
2589
2590 fsleep(5000);
2591 }
2592
2593 return r;
2594}
2595
2596/*
2597 * Process the deferred bios
2598 */
2599static void dm_wq_work(struct work_struct *work)
2600{
2601 struct mapped_device *md = container_of(work, struct mapped_device, work);
2602 struct bio *bio;
2603
2604 while (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
2605 spin_lock_irq(&md->deferred_lock);
2606 bio = bio_list_pop(&md->deferred);
2607 spin_unlock_irq(&md->deferred_lock);
2608
2609 if (!bio)
2610 break;
2611
2612 submit_bio_noacct(bio);
2613 cond_resched();
2614 }
2615}
2616
2617static void dm_queue_flush(struct mapped_device *md)
2618{
2619 clear_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2620 smp_mb__after_atomic();
2621 queue_work(md->wq, &md->work);
2622}
2623
2624/*
2625 * Swap in a new table, returning the old one for the caller to destroy.
2626 */
2627struct dm_table *dm_swap_table(struct mapped_device *md, struct dm_table *table)
2628{
2629 struct dm_table *live_map = NULL, *map = ERR_PTR(-EINVAL);
2630 struct queue_limits limits;
2631 int r;
2632
2633 mutex_lock(&md->suspend_lock);
2634
2635 /* device must be suspended */
2636 if (!dm_suspended_md(md))
2637 goto out;
2638
2639 /*
2640 * If the new table has no data devices, retain the existing limits.
2641 * This helps multipath with queue_if_no_path if all paths disappear,
2642 * then new I/O is queued based on these limits, and then some paths
2643 * reappear.
2644 */
2645 if (dm_table_has_no_data_devices(table)) {
2646 live_map = dm_get_live_table_fast(md);
2647 if (live_map)
2648 limits = md->queue->limits;
2649 dm_put_live_table_fast(md);
2650 }
2651
2652 if (!live_map) {
2653 r = dm_calculate_queue_limits(table, &limits);
2654 if (r) {
2655 map = ERR_PTR(r);
2656 goto out;
2657 }
2658 }
2659
2660 map = __bind(md, table, &limits);
2661 dm_issue_global_event();
2662
2663out:
2664 mutex_unlock(&md->suspend_lock);
2665 return map;
2666}
2667
2668/*
2669 * Functions to lock and unlock any filesystem running on the
2670 * device.
2671 */
2672static int lock_fs(struct mapped_device *md)
2673{
2674 int r;
2675
2676 WARN_ON(test_bit(DMF_FROZEN, &md->flags));
2677
2678 r = bdev_freeze(md->disk->part0);
2679 if (!r)
2680 set_bit(DMF_FROZEN, &md->flags);
2681 return r;
2682}
2683
2684static void unlock_fs(struct mapped_device *md)
2685{
2686 if (!test_bit(DMF_FROZEN, &md->flags))
2687 return;
2688 bdev_thaw(md->disk->part0);
2689 clear_bit(DMF_FROZEN, &md->flags);
2690}
2691
2692/*
2693 * @suspend_flags: DM_SUSPEND_LOCKFS_FLAG and/or DM_SUSPEND_NOFLUSH_FLAG
2694 * @task_state: e.g. TASK_INTERRUPTIBLE or TASK_UNINTERRUPTIBLE
2695 * @dmf_suspended_flag: DMF_SUSPENDED or DMF_SUSPENDED_INTERNALLY
2696 *
2697 * If __dm_suspend returns 0, the device is completely quiescent
2698 * now. There is no request-processing activity. All new requests
2699 * are being added to md->deferred list.
2700 */
2701static int __dm_suspend(struct mapped_device *md, struct dm_table *map,
2702 unsigned int suspend_flags, unsigned int task_state,
2703 int dmf_suspended_flag)
2704{
2705 bool do_lockfs = suspend_flags & DM_SUSPEND_LOCKFS_FLAG;
2706 bool noflush = suspend_flags & DM_SUSPEND_NOFLUSH_FLAG;
2707 int r;
2708
2709 lockdep_assert_held(&md->suspend_lock);
2710
2711 /*
2712 * DMF_NOFLUSH_SUSPENDING must be set before presuspend.
2713 * This flag is cleared before dm_suspend returns.
2714 */
2715 if (noflush)
2716 set_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
2717 else
2718 DMDEBUG("%s: suspending with flush", dm_device_name(md));
2719
2720 /*
2721 * This gets reverted if there's an error later and the targets
2722 * provide the .presuspend_undo hook.
2723 */
2724 dm_table_presuspend_targets(map);
2725
2726 /*
2727 * Flush I/O to the device.
2728 * Any I/O submitted after lock_fs() may not be flushed.
2729 * noflush takes precedence over do_lockfs.
2730 * (lock_fs() flushes I/Os and waits for them to complete.)
2731 */
2732 if (!noflush && do_lockfs) {
2733 r = lock_fs(md);
2734 if (r) {
2735 dm_table_presuspend_undo_targets(map);
2736 return r;
2737 }
2738 }
2739
2740 /*
2741 * Here we must make sure that no processes are submitting requests
2742 * to target drivers i.e. no one may be executing
2743 * dm_split_and_process_bio from dm_submit_bio.
2744 *
2745 * To get all processes out of dm_split_and_process_bio in dm_submit_bio,
2746 * we take the write lock. To prevent any process from reentering
2747 * dm_split_and_process_bio from dm_submit_bio and quiesce the thread
2748 * (dm_wq_work), we set DMF_BLOCK_IO_FOR_SUSPEND and call
2749 * flush_workqueue(md->wq).
2750 */
2751 set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2752 if (map)
2753 synchronize_srcu(&md->io_barrier);
2754
2755 /*
2756 * Stop md->queue before flushing md->wq in case request-based
2757 * dm defers requests to md->wq from md->queue.
2758 */
2759 if (dm_request_based(md))
2760 dm_stop_queue(md->queue);
2761
2762 flush_workqueue(md->wq);
2763
2764 /*
2765 * At this point no more requests are entering target request routines.
2766 * We call dm_wait_for_completion to wait for all existing requests
2767 * to finish.
2768 */
2769 r = dm_wait_for_completion(md, task_state);
2770 if (!r)
2771 set_bit(dmf_suspended_flag, &md->flags);
2772
2773 if (noflush)
2774 clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
2775 if (map)
2776 synchronize_srcu(&md->io_barrier);
2777
2778 /* were we interrupted ? */
2779 if (r < 0) {
2780 dm_queue_flush(md);
2781
2782 if (dm_request_based(md))
2783 dm_start_queue(md->queue);
2784
2785 unlock_fs(md);
2786 dm_table_presuspend_undo_targets(map);
2787 /* pushback list is already flushed, so skip flush */
2788 }
2789
2790 return r;
2791}
2792
2793/*
2794 * We need to be able to change a mapping table under a mounted
2795 * filesystem. For example we might want to move some data in
2796 * the background. Before the table can be swapped with
2797 * dm_bind_table, dm_suspend must be called to flush any in
2798 * flight bios and ensure that any further io gets deferred.
2799 */
2800/*
2801 * Suspend mechanism in request-based dm.
2802 *
2803 * 1. Flush all I/Os by lock_fs() if needed.
2804 * 2. Stop dispatching any I/O by stopping the request_queue.
2805 * 3. Wait for all in-flight I/Os to be completed or requeued.
2806 *
2807 * To abort suspend, start the request_queue.
2808 */
2809int dm_suspend(struct mapped_device *md, unsigned int suspend_flags)
2810{
2811 struct dm_table *map = NULL;
2812 int r = 0;
2813
2814retry:
2815 mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING);
2816
2817 if (dm_suspended_md(md)) {
2818 r = -EINVAL;
2819 goto out_unlock;
2820 }
2821
2822 if (dm_suspended_internally_md(md)) {
2823 /* already internally suspended, wait for internal resume */
2824 mutex_unlock(&md->suspend_lock);
2825 r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE);
2826 if (r)
2827 return r;
2828 goto retry;
2829 }
2830
2831 map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2832 if (!map) {
2833 /* avoid deadlock with fs/namespace.c:do_mount() */
2834 suspend_flags &= ~DM_SUSPEND_LOCKFS_FLAG;
2835 }
2836
2837 r = __dm_suspend(md, map, suspend_flags, TASK_INTERRUPTIBLE, DMF_SUSPENDED);
2838 if (r)
2839 goto out_unlock;
2840
2841 set_bit(DMF_POST_SUSPENDING, &md->flags);
2842 dm_table_postsuspend_targets(map);
2843 clear_bit(DMF_POST_SUSPENDING, &md->flags);
2844
2845out_unlock:
2846 mutex_unlock(&md->suspend_lock);
2847 return r;
2848}
2849
2850static int __dm_resume(struct mapped_device *md, struct dm_table *map)
2851{
2852 if (map) {
2853 int r = dm_table_resume_targets(map);
2854
2855 if (r)
2856 return r;
2857 }
2858
2859 dm_queue_flush(md);
2860
2861 /*
2862 * Flushing deferred I/Os must be done after targets are resumed
2863 * so that mapping of targets can work correctly.
2864 * Request-based dm is queueing the deferred I/Os in its request_queue.
2865 */
2866 if (dm_request_based(md))
2867 dm_start_queue(md->queue);
2868
2869 unlock_fs(md);
2870
2871 return 0;
2872}
2873
2874int dm_resume(struct mapped_device *md)
2875{
2876 int r;
2877 struct dm_table *map = NULL;
2878
2879retry:
2880 r = -EINVAL;
2881 mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING);
2882
2883 if (!dm_suspended_md(md))
2884 goto out;
2885
2886 if (dm_suspended_internally_md(md)) {
2887 /* already internally suspended, wait for internal resume */
2888 mutex_unlock(&md->suspend_lock);
2889 r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE);
2890 if (r)
2891 return r;
2892 goto retry;
2893 }
2894
2895 map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2896 if (!map || !dm_table_get_size(map))
2897 goto out;
2898
2899 r = __dm_resume(md, map);
2900 if (r)
2901 goto out;
2902
2903 clear_bit(DMF_SUSPENDED, &md->flags);
2904out:
2905 mutex_unlock(&md->suspend_lock);
2906
2907 return r;
2908}
2909
2910/*
2911 * Internal suspend/resume works like userspace-driven suspend. It waits
2912 * until all bios finish and prevents issuing new bios to the target drivers.
2913 * It may be used only from the kernel.
2914 */
2915
2916static void __dm_internal_suspend(struct mapped_device *md, unsigned int suspend_flags)
2917{
2918 struct dm_table *map = NULL;
2919
2920 lockdep_assert_held(&md->suspend_lock);
2921
2922 if (md->internal_suspend_count++)
2923 return; /* nested internal suspend */
2924
2925 if (dm_suspended_md(md)) {
2926 set_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
2927 return; /* nest suspend */
2928 }
2929
2930 map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2931
2932 /*
2933 * Using TASK_UNINTERRUPTIBLE because only NOFLUSH internal suspend is
2934 * supported. Properly supporting a TASK_INTERRUPTIBLE internal suspend
2935 * would require changing .presuspend to return an error -- avoid this
2936 * until there is a need for more elaborate variants of internal suspend.
2937 */
2938 (void) __dm_suspend(md, map, suspend_flags, TASK_UNINTERRUPTIBLE,
2939 DMF_SUSPENDED_INTERNALLY);
2940
2941 set_bit(DMF_POST_SUSPENDING, &md->flags);
2942 dm_table_postsuspend_targets(map);
2943 clear_bit(DMF_POST_SUSPENDING, &md->flags);
2944}
2945
2946static void __dm_internal_resume(struct mapped_device *md)
2947{
2948 BUG_ON(!md->internal_suspend_count);
2949
2950 if (--md->internal_suspend_count)
2951 return; /* resume from nested internal suspend */
2952
2953 if (dm_suspended_md(md))
2954 goto done; /* resume from nested suspend */
2955
2956 /*
2957 * NOTE: existing callers don't need to call dm_table_resume_targets
2958 * (which may fail -- so best to avoid it for now by passing NULL map)
2959 */
2960 (void) __dm_resume(md, NULL);
2961
2962done:
2963 clear_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
2964 smp_mb__after_atomic();
2965 wake_up_bit(&md->flags, DMF_SUSPENDED_INTERNALLY);
2966}
2967
2968void dm_internal_suspend_noflush(struct mapped_device *md)
2969{
2970 mutex_lock(&md->suspend_lock);
2971 __dm_internal_suspend(md, DM_SUSPEND_NOFLUSH_FLAG);
2972 mutex_unlock(&md->suspend_lock);
2973}
2974EXPORT_SYMBOL_GPL(dm_internal_suspend_noflush);
2975
2976void dm_internal_resume(struct mapped_device *md)
2977{
2978 mutex_lock(&md->suspend_lock);
2979 __dm_internal_resume(md);
2980 mutex_unlock(&md->suspend_lock);
2981}
2982EXPORT_SYMBOL_GPL(dm_internal_resume);
2983
2984/*
2985 * Fast variants of internal suspend/resume hold md->suspend_lock,
2986 * which prevents interaction with userspace-driven suspend.
2987 */
2988
2989void dm_internal_suspend_fast(struct mapped_device *md)
2990{
2991 mutex_lock(&md->suspend_lock);
2992 if (dm_suspended_md(md) || dm_suspended_internally_md(md))
2993 return;
2994
2995 set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2996 synchronize_srcu(&md->io_barrier);
2997 flush_workqueue(md->wq);
2998 dm_wait_for_completion(md, TASK_UNINTERRUPTIBLE);
2999}
3000EXPORT_SYMBOL_GPL(dm_internal_suspend_fast);
3001
3002void dm_internal_resume_fast(struct mapped_device *md)
3003{
3004 if (dm_suspended_md(md) || dm_suspended_internally_md(md))
3005 goto done;
3006
3007 dm_queue_flush(md);
3008
3009done:
3010 mutex_unlock(&md->suspend_lock);
3011}
3012EXPORT_SYMBOL_GPL(dm_internal_resume_fast);
3013
3014/*
3015 *---------------------------------------------------------------
3016 * Event notification.
3017 *---------------------------------------------------------------
3018 */
3019int dm_kobject_uevent(struct mapped_device *md, enum kobject_action action,
3020 unsigned int cookie, bool need_resize_uevent)
3021{
3022 int r;
3023 unsigned int noio_flag;
3024 char udev_cookie[DM_COOKIE_LENGTH];
3025 char *envp[3] = { NULL, NULL, NULL };
3026 char **envpp = envp;
3027 if (cookie) {
3028 snprintf(udev_cookie, DM_COOKIE_LENGTH, "%s=%u",
3029 DM_COOKIE_ENV_VAR_NAME, cookie);
3030 *envpp++ = udev_cookie;
3031 }
3032 if (need_resize_uevent) {
3033 *envpp++ = "RESIZE=1";
3034 }
3035
3036 noio_flag = memalloc_noio_save();
3037
3038 r = kobject_uevent_env(&disk_to_dev(md->disk)->kobj, action, envp);
3039
3040 memalloc_noio_restore(noio_flag);
3041
3042 return r;
3043}
3044
3045uint32_t dm_next_uevent_seq(struct mapped_device *md)
3046{
3047 return atomic_add_return(1, &md->uevent_seq);
3048}
3049
3050uint32_t dm_get_event_nr(struct mapped_device *md)
3051{
3052 return atomic_read(&md->event_nr);
3053}
3054
3055int dm_wait_event(struct mapped_device *md, int event_nr)
3056{
3057 return wait_event_interruptible(md->eventq,
3058 (event_nr != atomic_read(&md->event_nr)));
3059}
3060
3061void dm_uevent_add(struct mapped_device *md, struct list_head *elist)
3062{
3063 unsigned long flags;
3064
3065 spin_lock_irqsave(&md->uevent_lock, flags);
3066 list_add(elist, &md->uevent_list);
3067 spin_unlock_irqrestore(&md->uevent_lock, flags);
3068}
3069
3070/*
3071 * The gendisk is only valid as long as you have a reference
3072 * count on 'md'.
3073 */
3074struct gendisk *dm_disk(struct mapped_device *md)
3075{
3076 return md->disk;
3077}
3078EXPORT_SYMBOL_GPL(dm_disk);
3079
3080struct kobject *dm_kobject(struct mapped_device *md)
3081{
3082 return &md->kobj_holder.kobj;
3083}
3084
3085struct mapped_device *dm_get_from_kobject(struct kobject *kobj)
3086{
3087 struct mapped_device *md;
3088
3089 md = container_of(kobj, struct mapped_device, kobj_holder.kobj);
3090
3091 spin_lock(&_minor_lock);
3092 if (test_bit(DMF_FREEING, &md->flags) || dm_deleting_md(md)) {
3093 md = NULL;
3094 goto out;
3095 }
3096 dm_get(md);
3097out:
3098 spin_unlock(&_minor_lock);
3099
3100 return md;
3101}
3102
3103int dm_suspended_md(struct mapped_device *md)
3104{
3105 return test_bit(DMF_SUSPENDED, &md->flags);
3106}
3107
3108static int dm_post_suspending_md(struct mapped_device *md)
3109{
3110 return test_bit(DMF_POST_SUSPENDING, &md->flags);
3111}
3112
3113int dm_suspended_internally_md(struct mapped_device *md)
3114{
3115 return test_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
3116}
3117
3118int dm_test_deferred_remove_flag(struct mapped_device *md)
3119{
3120 return test_bit(DMF_DEFERRED_REMOVE, &md->flags);
3121}
3122
3123int dm_suspended(struct dm_target *ti)
3124{
3125 return dm_suspended_md(ti->table->md);
3126}
3127EXPORT_SYMBOL_GPL(dm_suspended);
3128
3129int dm_post_suspending(struct dm_target *ti)
3130{
3131 return dm_post_suspending_md(ti->table->md);
3132}
3133EXPORT_SYMBOL_GPL(dm_post_suspending);
3134
3135int dm_noflush_suspending(struct dm_target *ti)
3136{
3137 return __noflush_suspending(ti->table->md);
3138}
3139EXPORT_SYMBOL_GPL(dm_noflush_suspending);
3140
3141void dm_free_md_mempools(struct dm_md_mempools *pools)
3142{
3143 if (!pools)
3144 return;
3145
3146 bioset_exit(&pools->bs);
3147 bioset_exit(&pools->io_bs);
3148
3149 kfree(pools);
3150}
3151
3152struct dm_pr {
3153 u64 old_key;
3154 u64 new_key;
3155 u32 flags;
3156 bool abort;
3157 bool fail_early;
3158 int ret;
3159 enum pr_type type;
3160 struct pr_keys *read_keys;
3161 struct pr_held_reservation *rsv;
3162};
3163
3164static int dm_call_pr(struct block_device *bdev, iterate_devices_callout_fn fn,
3165 struct dm_pr *pr)
3166{
3167 struct mapped_device *md = bdev->bd_disk->private_data;
3168 struct dm_table *table;
3169 struct dm_target *ti;
3170 int ret = -ENOTTY, srcu_idx;
3171
3172 table = dm_get_live_table(md, &srcu_idx);
3173 if (!table || !dm_table_get_size(table))
3174 goto out;
3175
3176 /* We only support devices that have a single target */
3177 if (table->num_targets != 1)
3178 goto out;
3179 ti = dm_table_get_target(table, 0);
3180
3181 if (dm_suspended_md(md)) {
3182 ret = -EAGAIN;
3183 goto out;
3184 }
3185
3186 ret = -EINVAL;
3187 if (!ti->type->iterate_devices)
3188 goto out;
3189
3190 ti->type->iterate_devices(ti, fn, pr);
3191 ret = 0;
3192out:
3193 dm_put_live_table(md, srcu_idx);
3194 return ret;
3195}
3196
3197/*
3198 * For register / unregister we need to manually call out to every path.
3199 */
3200static int __dm_pr_register(struct dm_target *ti, struct dm_dev *dev,
3201 sector_t start, sector_t len, void *data)
3202{
3203 struct dm_pr *pr = data;
3204 const struct pr_ops *ops = dev->bdev->bd_disk->fops->pr_ops;
3205 int ret;
3206
3207 if (!ops || !ops->pr_register) {
3208 pr->ret = -EOPNOTSUPP;
3209 return -1;
3210 }
3211
3212 ret = ops->pr_register(dev->bdev, pr->old_key, pr->new_key, pr->flags);
3213 if (!ret)
3214 return 0;
3215
3216 if (!pr->ret)
3217 pr->ret = ret;
3218
3219 if (pr->fail_early)
3220 return -1;
3221
3222 return 0;
3223}
3224
3225static int dm_pr_register(struct block_device *bdev, u64 old_key, u64 new_key,
3226 u32 flags)
3227{
3228 struct dm_pr pr = {
3229 .old_key = old_key,
3230 .new_key = new_key,
3231 .flags = flags,
3232 .fail_early = true,
3233 .ret = 0,
3234 };
3235 int ret;
3236
3237 ret = dm_call_pr(bdev, __dm_pr_register, &pr);
3238 if (ret) {
3239 /* Didn't even get to register a path */
3240 return ret;
3241 }
3242
3243 if (!pr.ret)
3244 return 0;
3245 ret = pr.ret;
3246
3247 if (!new_key)
3248 return ret;
3249
3250 /* unregister all paths if we failed to register any path */
3251 pr.old_key = new_key;
3252 pr.new_key = 0;
3253 pr.flags = 0;
3254 pr.fail_early = false;
3255 (void) dm_call_pr(bdev, __dm_pr_register, &pr);
3256 return ret;
3257}
3258
3259
3260static int __dm_pr_reserve(struct dm_target *ti, struct dm_dev *dev,
3261 sector_t start, sector_t len, void *data)
3262{
3263 struct dm_pr *pr = data;
3264 const struct pr_ops *ops = dev->bdev->bd_disk->fops->pr_ops;
3265
3266 if (!ops || !ops->pr_reserve) {
3267 pr->ret = -EOPNOTSUPP;
3268 return -1;
3269 }
3270
3271 pr->ret = ops->pr_reserve(dev->bdev, pr->old_key, pr->type, pr->flags);
3272 if (!pr->ret)
3273 return -1;
3274
3275 return 0;
3276}
3277
3278static int dm_pr_reserve(struct block_device *bdev, u64 key, enum pr_type type,
3279 u32 flags)
3280{
3281 struct dm_pr pr = {
3282 .old_key = key,
3283 .flags = flags,
3284 .type = type,
3285 .fail_early = false,
3286 .ret = 0,
3287 };
3288 int ret;
3289
3290 ret = dm_call_pr(bdev, __dm_pr_reserve, &pr);
3291 if (ret)
3292 return ret;
3293
3294 return pr.ret;
3295}
3296
3297/*
3298 * If there is a non-All Registrants type of reservation, the release must be
3299 * sent down the holding path. For the cases where there is no reservation or
3300 * the path is not the holder the device will also return success, so we must
3301 * try each path to make sure we got the correct path.
3302 */
3303static int __dm_pr_release(struct dm_target *ti, struct dm_dev *dev,
3304 sector_t start, sector_t len, void *data)
3305{
3306 struct dm_pr *pr = data;
3307 const struct pr_ops *ops = dev->bdev->bd_disk->fops->pr_ops;
3308
3309 if (!ops || !ops->pr_release) {
3310 pr->ret = -EOPNOTSUPP;
3311 return -1;
3312 }
3313
3314 pr->ret = ops->pr_release(dev->bdev, pr->old_key, pr->type);
3315 if (pr->ret)
3316 return -1;
3317
3318 return 0;
3319}
3320
3321static int dm_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
3322{
3323 struct dm_pr pr = {
3324 .old_key = key,
3325 .type = type,
3326 .fail_early = false,
3327 };
3328 int ret;
3329
3330 ret = dm_call_pr(bdev, __dm_pr_release, &pr);
3331 if (ret)
3332 return ret;
3333
3334 return pr.ret;
3335}
3336
3337static int __dm_pr_preempt(struct dm_target *ti, struct dm_dev *dev,
3338 sector_t start, sector_t len, void *data)
3339{
3340 struct dm_pr *pr = data;
3341 const struct pr_ops *ops = dev->bdev->bd_disk->fops->pr_ops;
3342
3343 if (!ops || !ops->pr_preempt) {
3344 pr->ret = -EOPNOTSUPP;
3345 return -1;
3346 }
3347
3348 pr->ret = ops->pr_preempt(dev->bdev, pr->old_key, pr->new_key, pr->type,
3349 pr->abort);
3350 if (!pr->ret)
3351 return -1;
3352
3353 return 0;
3354}
3355
3356static int dm_pr_preempt(struct block_device *bdev, u64 old_key, u64 new_key,
3357 enum pr_type type, bool abort)
3358{
3359 struct dm_pr pr = {
3360 .new_key = new_key,
3361 .old_key = old_key,
3362 .type = type,
3363 .fail_early = false,
3364 };
3365 int ret;
3366
3367 ret = dm_call_pr(bdev, __dm_pr_preempt, &pr);
3368 if (ret)
3369 return ret;
3370
3371 return pr.ret;
3372}
3373
3374static int dm_pr_clear(struct block_device *bdev, u64 key)
3375{
3376 struct mapped_device *md = bdev->bd_disk->private_data;
3377 const struct pr_ops *ops;
3378 int r, srcu_idx;
3379
3380 r = dm_prepare_ioctl(md, &srcu_idx, &bdev);
3381 if (r < 0)
3382 goto out;
3383
3384 ops = bdev->bd_disk->fops->pr_ops;
3385 if (ops && ops->pr_clear)
3386 r = ops->pr_clear(bdev, key);
3387 else
3388 r = -EOPNOTSUPP;
3389out:
3390 dm_unprepare_ioctl(md, srcu_idx);
3391 return r;
3392}
3393
3394static int __dm_pr_read_keys(struct dm_target *ti, struct dm_dev *dev,
3395 sector_t start, sector_t len, void *data)
3396{
3397 struct dm_pr *pr = data;
3398 const struct pr_ops *ops = dev->bdev->bd_disk->fops->pr_ops;
3399
3400 if (!ops || !ops->pr_read_keys) {
3401 pr->ret = -EOPNOTSUPP;
3402 return -1;
3403 }
3404
3405 pr->ret = ops->pr_read_keys(dev->bdev, pr->read_keys);
3406 if (!pr->ret)
3407 return -1;
3408
3409 return 0;
3410}
3411
3412static int dm_pr_read_keys(struct block_device *bdev, struct pr_keys *keys)
3413{
3414 struct dm_pr pr = {
3415 .read_keys = keys,
3416 };
3417 int ret;
3418
3419 ret = dm_call_pr(bdev, __dm_pr_read_keys, &pr);
3420 if (ret)
3421 return ret;
3422
3423 return pr.ret;
3424}
3425
3426static int __dm_pr_read_reservation(struct dm_target *ti, struct dm_dev *dev,
3427 sector_t start, sector_t len, void *data)
3428{
3429 struct dm_pr *pr = data;
3430 const struct pr_ops *ops = dev->bdev->bd_disk->fops->pr_ops;
3431
3432 if (!ops || !ops->pr_read_reservation) {
3433 pr->ret = -EOPNOTSUPP;
3434 return -1;
3435 }
3436
3437 pr->ret = ops->pr_read_reservation(dev->bdev, pr->rsv);
3438 if (!pr->ret)
3439 return -1;
3440
3441 return 0;
3442}
3443
3444static int dm_pr_read_reservation(struct block_device *bdev,
3445 struct pr_held_reservation *rsv)
3446{
3447 struct dm_pr pr = {
3448 .rsv = rsv,
3449 };
3450 int ret;
3451
3452 ret = dm_call_pr(bdev, __dm_pr_read_reservation, &pr);
3453 if (ret)
3454 return ret;
3455
3456 return pr.ret;
3457}
3458
3459static const struct pr_ops dm_pr_ops = {
3460 .pr_register = dm_pr_register,
3461 .pr_reserve = dm_pr_reserve,
3462 .pr_release = dm_pr_release,
3463 .pr_preempt = dm_pr_preempt,
3464 .pr_clear = dm_pr_clear,
3465 .pr_read_keys = dm_pr_read_keys,
3466 .pr_read_reservation = dm_pr_read_reservation,
3467};
3468
3469static const struct block_device_operations dm_blk_dops = {
3470 .submit_bio = dm_submit_bio,
3471 .poll_bio = dm_poll_bio,
3472 .open = dm_blk_open,
3473 .release = dm_blk_close,
3474 .ioctl = dm_blk_ioctl,
3475 .getgeo = dm_blk_getgeo,
3476 .report_zones = dm_blk_report_zones,
3477 .pr_ops = &dm_pr_ops,
3478 .owner = THIS_MODULE
3479};
3480
3481static const struct block_device_operations dm_rq_blk_dops = {
3482 .open = dm_blk_open,
3483 .release = dm_blk_close,
3484 .ioctl = dm_blk_ioctl,
3485 .getgeo = dm_blk_getgeo,
3486 .pr_ops = &dm_pr_ops,
3487 .owner = THIS_MODULE
3488};
3489
3490static const struct dax_operations dm_dax_ops = {
3491 .direct_access = dm_dax_direct_access,
3492 .zero_page_range = dm_dax_zero_page_range,
3493 .recovery_write = dm_dax_recovery_write,
3494};
3495
3496/*
3497 * module hooks
3498 */
3499module_init(dm_init);
3500module_exit(dm_exit);
3501
3502module_param(major, uint, 0);
3503MODULE_PARM_DESC(major, "The major number of the device mapper");
3504
3505module_param(reserved_bio_based_ios, uint, 0644);
3506MODULE_PARM_DESC(reserved_bio_based_ios, "Reserved IOs in bio-based mempools");
3507
3508module_param(dm_numa_node, int, 0644);
3509MODULE_PARM_DESC(dm_numa_node, "NUMA node for DM device memory allocations");
3510
3511module_param(swap_bios, int, 0644);
3512MODULE_PARM_DESC(swap_bios, "Maximum allowed inflight swap IOs");
3513
3514MODULE_DESCRIPTION(DM_NAME " driver");
3515MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
3516MODULE_LICENSE("GPL");
1/*
2 * Copyright (C) 2001, 2002 Sistina Software (UK) Limited.
3 * Copyright (C) 2004-2008 Red Hat, Inc. All rights reserved.
4 *
5 * This file is released under the GPL.
6 */
7
8#include "dm-core.h"
9#include "dm-rq.h"
10#include "dm-uevent.h"
11
12#include <linux/init.h>
13#include <linux/module.h>
14#include <linux/mutex.h>
15#include <linux/sched/mm.h>
16#include <linux/sched/signal.h>
17#include <linux/blkpg.h>
18#include <linux/bio.h>
19#include <linux/mempool.h>
20#include <linux/dax.h>
21#include <linux/slab.h>
22#include <linux/idr.h>
23#include <linux/uio.h>
24#include <linux/hdreg.h>
25#include <linux/delay.h>
26#include <linux/wait.h>
27#include <linux/pr.h>
28#include <linux/refcount.h>
29#include <linux/part_stat.h>
30#include <linux/blk-crypto.h>
31
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");