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 */
8 #include "dm-core.h"
9 #include "dm-rq.h"
10 #include "dm-uevent.h"
12 #include <linux/init.h>
13 #include <linux/module.h>
14 #include <linux/mutex.h>
15 #include <linux/sched/signal.h>
16 #include <linux/blkpg.h>
17 #include <linux/bio.h>
18 #include <linux/mempool.h>
19 #include <linux/dax.h>
20 #include <linux/slab.h>
21 #include <linux/idr.h>
22 #include <linux/uio.h>
23 #include <linux/hdreg.h>
24 #include <linux/delay.h>
25 #include <linux/wait.h>
26 #include <linux/pr.h>
27 #include <linux/refcount.h>
29 #define DM_MSG_PREFIX "core"
31 /*
32 * Cookies are numeric values sent with CHANGE and REMOVE
33 * uevents while resuming, removing or renaming the device.
34 */
35 #define DM_COOKIE_ENV_VAR_NAME "DM_COOKIE"
36 #define DM_COOKIE_LENGTH 24
38 static const char *_name = DM_NAME;
40 static unsigned int major = 0;
41 static unsigned int _major = 0;
43 static DEFINE_IDR(_minor_idr);
45 static DEFINE_SPINLOCK(_minor_lock);
47 static void do_deferred_remove(struct work_struct *w);
49 static DECLARE_WORK(deferred_remove_work, do_deferred_remove);
51 static struct workqueue_struct *deferred_remove_workqueue;
53 atomic_t dm_global_event_nr = ATOMIC_INIT(0);
54 DECLARE_WAIT_QUEUE_HEAD(dm_global_eventq);
56 void dm_issue_global_event(void)
57 {
58 atomic_inc(&dm_global_event_nr);
59 wake_up(&dm_global_eventq);
60 }
62 /*
63 * One of these is allocated (on-stack) per original bio.
64 */
65 struct clone_info {
66 struct dm_table *map;
67 struct bio *bio;
68 struct dm_io *io;
69 sector_t sector;
70 unsigned sector_count;
71 };
73 /*
74 * One of these is allocated per clone bio.
75 */
76 #define DM_TIO_MAGIC 7282014
77 struct dm_target_io {
78 unsigned magic;
79 struct dm_io *io;
80 struct dm_target *ti;
81 unsigned target_bio_nr;
82 unsigned *len_ptr;
83 bool inside_dm_io;
84 struct bio clone;
85 };
87 /*
88 * One of these is allocated per original bio.
89 * It contains the first clone used for that original.
90 */
91 #define DM_IO_MAGIC 5191977
92 struct dm_io {
93 unsigned magic;
94 struct mapped_device *md;
95 blk_status_t status;
96 atomic_t io_count;
97 struct bio *orig_bio;
98 unsigned long start_time;
99 spinlock_t endio_lock;
100 struct dm_stats_aux stats_aux;
101 /* last member of dm_target_io is 'struct bio' */
102 struct dm_target_io tio;
103 };
105 void *dm_per_bio_data(struct bio *bio, size_t data_size)
106 {
107 struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone);
108 if (!tio->inside_dm_io)
109 return (char *)bio - offsetof(struct dm_target_io, clone) - data_size;
110 return (char *)bio - offsetof(struct dm_target_io, clone) - offsetof(struct dm_io, tio) - data_size;
111 }
112 EXPORT_SYMBOL_GPL(dm_per_bio_data);
114 struct bio *dm_bio_from_per_bio_data(void *data, size_t data_size)
115 {
116 struct dm_io *io = (struct dm_io *)((char *)data + data_size);
117 if (io->magic == DM_IO_MAGIC)
118 return (struct bio *)((char *)io + offsetof(struct dm_io, tio) + offsetof(struct dm_target_io, clone));
119 BUG_ON(io->magic != DM_TIO_MAGIC);
120 return (struct bio *)((char *)io + offsetof(struct dm_target_io, clone));
121 }
122 EXPORT_SYMBOL_GPL(dm_bio_from_per_bio_data);
124 unsigned dm_bio_get_target_bio_nr(const struct bio *bio)
125 {
126 return container_of(bio, struct dm_target_io, clone)->target_bio_nr;
127 }
128 EXPORT_SYMBOL_GPL(dm_bio_get_target_bio_nr);
130 #define MINOR_ALLOCED ((void *)-1)
132 /*
133 * Bits for the md->flags field.
134 */
135 #define DMF_BLOCK_IO_FOR_SUSPEND 0
136 #define DMF_SUSPENDED 1
137 #define DMF_FROZEN 2
138 #define DMF_FREEING 3
139 #define DMF_DELETING 4
140 #define DMF_NOFLUSH_SUSPENDING 5
141 #define DMF_DEFERRED_REMOVE 6
142 #define DMF_SUSPENDED_INTERNALLY 7
144 #define DM_NUMA_NODE NUMA_NO_NODE
145 static int dm_numa_node = DM_NUMA_NODE;
147 /*
148 * For mempools pre-allocation at the table loading time.
149 */
150 struct dm_md_mempools {
151 struct bio_set bs;
152 struct bio_set io_bs;
153 };
155 struct table_device {
156 struct list_head list;
157 refcount_t count;
158 struct dm_dev dm_dev;
159 };
161 static struct kmem_cache *_rq_tio_cache;
162 static struct kmem_cache *_rq_cache;
164 /*
165 * Bio-based DM's mempools' reserved IOs set by the user.
166 */
167 #define RESERVED_BIO_BASED_IOS 16
168 static unsigned reserved_bio_based_ios = RESERVED_BIO_BASED_IOS;
170 static int __dm_get_module_param_int(int *module_param, int min, int max)
171 {
172 int param = READ_ONCE(*module_param);
173 int modified_param = 0;
174 bool modified = true;
176 if (param < min)
177 modified_param = min;
178 else if (param > max)
179 modified_param = max;
180 else
181 modified = false;
183 if (modified) {
184 (void)cmpxchg(module_param, param, modified_param);
185 param = modified_param;
186 }
188 return param;
189 }
191 unsigned __dm_get_module_param(unsigned *module_param,
192 unsigned def, unsigned max)
193 {
194 unsigned param = READ_ONCE(*module_param);
195 unsigned modified_param = 0;
197 if (!param)
198 modified_param = def;
199 else if (param > max)
200 modified_param = max;
202 if (modified_param) {
203 (void)cmpxchg(module_param, param, modified_param);
204 param = modified_param;
205 }
207 return param;
208 }
210 unsigned dm_get_reserved_bio_based_ios(void)
211 {
212 return __dm_get_module_param(&reserved_bio_based_ios,
213 RESERVED_BIO_BASED_IOS, DM_RESERVED_MAX_IOS);
214 }
215 EXPORT_SYMBOL_GPL(dm_get_reserved_bio_based_ios);
217 static unsigned dm_get_numa_node(void)
218 {
219 return __dm_get_module_param_int(&dm_numa_node,
220 DM_NUMA_NODE, num_online_nodes() - 1);
221 }
223 static int __init local_init(void)
224 {
225 int r = -ENOMEM;
227 _rq_tio_cache = KMEM_CACHE(dm_rq_target_io, 0);
228 if (!_rq_tio_cache)
229 return r;
231 _rq_cache = kmem_cache_create("dm_old_clone_request", sizeof(struct request),
232 __alignof__(struct request), 0, NULL);
233 if (!_rq_cache)
234 goto out_free_rq_tio_cache;
236 r = dm_uevent_init();
237 if (r)
238 goto out_free_rq_cache;
240 deferred_remove_workqueue = alloc_workqueue("kdmremove", WQ_UNBOUND, 1);
241 if (!deferred_remove_workqueue) {
242 r = -ENOMEM;
243 goto out_uevent_exit;
244 }
246 _major = major;
247 r = register_blkdev(_major, _name);
248 if (r < 0)
249 goto out_free_workqueue;
251 if (!_major)
252 _major = r;
254 return 0;
256 out_free_workqueue:
257 destroy_workqueue(deferred_remove_workqueue);
258 out_uevent_exit:
259 dm_uevent_exit();
260 out_free_rq_cache:
261 kmem_cache_destroy(_rq_cache);
262 out_free_rq_tio_cache:
263 kmem_cache_destroy(_rq_tio_cache);
265 return r;
266 }
268 static void local_exit(void)
269 {
270 flush_scheduled_work();
271 destroy_workqueue(deferred_remove_workqueue);
273 kmem_cache_destroy(_rq_cache);
274 kmem_cache_destroy(_rq_tio_cache);
275 unregister_blkdev(_major, _name);
276 dm_uevent_exit();
278 _major = 0;
280 DMINFO("cleaned up");
281 }
283 static int (*_inits[])(void) __initdata = {
284 local_init,
285 dm_target_init,
286 dm_linear_init,
287 dm_stripe_init,
288 dm_io_init,
289 dm_kcopyd_init,
290 dm_interface_init,
291 dm_statistics_init,
292 };
294 static void (*_exits[])(void) = {
295 local_exit,
296 dm_target_exit,
297 dm_linear_exit,
298 dm_stripe_exit,
299 dm_io_exit,
300 dm_kcopyd_exit,
301 dm_interface_exit,
302 dm_statistics_exit,
303 };
305 static int __init dm_init(void)
306 {
307 const int count = ARRAY_SIZE(_inits);
309 int r, i;
311 for (i = 0; i < count; i++) {
312 r = _inits[i]();
313 if (r)
314 goto bad;
315 }
317 return 0;
319 bad:
320 while (i--)
321 _exits[i]();
323 return r;
324 }
326 static void __exit dm_exit(void)
327 {
328 int i = ARRAY_SIZE(_exits);
330 while (i--)
331 _exits[i]();
333 /*
334 * Should be empty by this point.
335 */
336 idr_destroy(&_minor_idr);
337 }
339 /*
340 * Block device functions
341 */
342 int dm_deleting_md(struct mapped_device *md)
343 {
344 return test_bit(DMF_DELETING, &md->flags);
345 }
347 static int dm_blk_open(struct block_device *bdev, fmode_t mode)
348 {
349 struct mapped_device *md;
351 spin_lock(&_minor_lock);
353 md = bdev->bd_disk->private_data;
354 if (!md)
355 goto out;
357 if (test_bit(DMF_FREEING, &md->flags) ||
358 dm_deleting_md(md)) {
359 md = NULL;
360 goto out;
361 }
363 dm_get(md);
364 atomic_inc(&md->open_count);
365 out:
366 spin_unlock(&_minor_lock);
368 return md ? 0 : -ENXIO;
369 }
371 static void dm_blk_close(struct gendisk *disk, fmode_t mode)
372 {
373 struct mapped_device *md;
375 spin_lock(&_minor_lock);
377 md = disk->private_data;
378 if (WARN_ON(!md))
379 goto out;
381 if (atomic_dec_and_test(&md->open_count) &&
382 (test_bit(DMF_DEFERRED_REMOVE, &md->flags)))
383 queue_work(deferred_remove_workqueue, &deferred_remove_work);
385 dm_put(md);
386 out:
387 spin_unlock(&_minor_lock);
388 }
390 int dm_open_count(struct mapped_device *md)
391 {
392 return atomic_read(&md->open_count);
393 }
395 /*
396 * Guarantees nothing is using the device before it's deleted.
397 */
398 int dm_lock_for_deletion(struct mapped_device *md, bool mark_deferred, bool only_deferred)
399 {
400 int r = 0;
402 spin_lock(&_minor_lock);
404 if (dm_open_count(md)) {
405 r = -EBUSY;
406 if (mark_deferred)
407 set_bit(DMF_DEFERRED_REMOVE, &md->flags);
408 } else if (only_deferred && !test_bit(DMF_DEFERRED_REMOVE, &md->flags))
409 r = -EEXIST;
410 else
411 set_bit(DMF_DELETING, &md->flags);
413 spin_unlock(&_minor_lock);
415 return r;
416 }
418 int dm_cancel_deferred_remove(struct mapped_device *md)
419 {
420 int r = 0;
422 spin_lock(&_minor_lock);
424 if (test_bit(DMF_DELETING, &md->flags))
425 r = -EBUSY;
426 else
427 clear_bit(DMF_DEFERRED_REMOVE, &md->flags);
429 spin_unlock(&_minor_lock);
431 return r;
432 }
434 static void do_deferred_remove(struct work_struct *w)
435 {
436 dm_deferred_remove();
437 }
439 sector_t dm_get_size(struct mapped_device *md)
440 {
441 return get_capacity(md->disk);
442 }
444 struct request_queue *dm_get_md_queue(struct mapped_device *md)
445 {
446 return md->queue;
447 }
449 struct dm_stats *dm_get_stats(struct mapped_device *md)
450 {
451 return &md->stats;
452 }
454 static int dm_blk_getgeo(struct block_device *bdev, struct hd_geometry *geo)
455 {
456 struct mapped_device *md = bdev->bd_disk->private_data;
458 return dm_get_geometry(md, geo);
459 }
461 static int dm_prepare_ioctl(struct mapped_device *md, int *srcu_idx,
462 struct block_device **bdev)
463 __acquires(md->io_barrier)
464 {
465 struct dm_target *tgt;
466 struct dm_table *map;
467 int r;
469 retry:
470 r = -ENOTTY;
471 map = dm_get_live_table(md, srcu_idx);
472 if (!map || !dm_table_get_size(map))
473 return r;
475 /* We only support devices that have a single target */
476 if (dm_table_get_num_targets(map) != 1)
477 return r;
479 tgt = dm_table_get_target(map, 0);
480 if (!tgt->type->prepare_ioctl)
481 return r;
483 if (dm_suspended_md(md))
484 return -EAGAIN;
486 r = tgt->type->prepare_ioctl(tgt, bdev);
487 if (r == -ENOTCONN && !fatal_signal_pending(current)) {
488 dm_put_live_table(md, *srcu_idx);
489 msleep(10);
490 goto retry;
491 }
493 return r;
494 }
496 static void dm_unprepare_ioctl(struct mapped_device *md, int srcu_idx)
497 __releases(md->io_barrier)
498 {
499 dm_put_live_table(md, srcu_idx);
500 }
502 static int dm_blk_ioctl(struct block_device *bdev, fmode_t mode,
503 unsigned int cmd, unsigned long arg)
504 {
505 struct mapped_device *md = bdev->bd_disk->private_data;
506 int r, srcu_idx;
508 r = dm_prepare_ioctl(md, &srcu_idx, &bdev);
509 if (r < 0)
510 goto out;
512 if (r > 0) {
513 /*
514 * Target determined this ioctl is being issued against a
515 * subset of the parent bdev; require extra privileges.
516 */
517 if (!capable(CAP_SYS_RAWIO)) {
518 DMWARN_LIMIT(
519 "%s: sending ioctl %x to DM device without required privilege.",
520 current->comm, cmd);
521 r = -ENOIOCTLCMD;
522 goto out;
523 }
524 }
526 r = __blkdev_driver_ioctl(bdev, mode, cmd, arg);
527 out:
528 dm_unprepare_ioctl(md, srcu_idx);
529 return r;
530 }
532 static void start_io_acct(struct dm_io *io);
534 static struct dm_io *alloc_io(struct mapped_device *md, struct bio *bio)
535 {
536 struct dm_io *io;
537 struct dm_target_io *tio;
538 struct bio *clone;
540 clone = bio_alloc_bioset(GFP_NOIO, 0, &md->io_bs);
541 if (!clone)
542 return NULL;
544 tio = container_of(clone, struct dm_target_io, clone);
545 tio->inside_dm_io = true;
546 tio->io = NULL;
548 io = container_of(tio, struct dm_io, tio);
549 io->magic = DM_IO_MAGIC;
550 io->status = 0;
551 atomic_set(&io->io_count, 1);
552 io->orig_bio = bio;
553 io->md = md;
554 spin_lock_init(&io->endio_lock);
556 start_io_acct(io);
558 return io;
559 }
561 static void free_io(struct mapped_device *md, struct dm_io *io)
562 {
563 bio_put(&io->tio.clone);
564 }
566 static struct dm_target_io *alloc_tio(struct clone_info *ci, struct dm_target *ti,
567 unsigned target_bio_nr, gfp_t gfp_mask)
568 {
569 struct dm_target_io *tio;
571 if (!ci->io->tio.io) {
572 /* the dm_target_io embedded in ci->io is available */
573 tio = &ci->io->tio;
574 } else {
575 struct bio *clone = bio_alloc_bioset(gfp_mask, 0, &ci->io->md->bs);
576 if (!clone)
577 return NULL;
579 tio = container_of(clone, struct dm_target_io, clone);
580 tio->inside_dm_io = false;
581 }
583 tio->magic = DM_TIO_MAGIC;
584 tio->io = ci->io;
585 tio->ti = ti;
586 tio->target_bio_nr = target_bio_nr;
588 return tio;
589 }
591 static void free_tio(struct dm_target_io *tio)
592 {
593 if (tio->inside_dm_io)
594 return;
595 bio_put(&tio->clone);
596 }
598 int md_in_flight(struct mapped_device *md)
599 {
600 return atomic_read(&md->pending[READ]) +
601 atomic_read(&md->pending[WRITE]);
602 }
604 static void start_io_acct(struct dm_io *io)
605 {
606 struct mapped_device *md = io->md;
607 struct bio *bio = io->orig_bio;
608 int rw = bio_data_dir(bio);
610 io->start_time = jiffies;
612 generic_start_io_acct(md->queue, bio_op(bio), bio_sectors(bio),
613 &dm_disk(md)->part0);
615 atomic_set(&dm_disk(md)->part0.in_flight[rw],
616 atomic_inc_return(&md->pending[rw]));
618 if (unlikely(dm_stats_used(&md->stats)))
619 dm_stats_account_io(&md->stats, bio_data_dir(bio),
620 bio->bi_iter.bi_sector, bio_sectors(bio),
621 false, 0, &io->stats_aux);
622 }
624 static void end_io_acct(struct dm_io *io)
625 {
626 struct mapped_device *md = io->md;
627 struct bio *bio = io->orig_bio;
628 unsigned long duration = jiffies - io->start_time;
629 int pending;
630 int rw = bio_data_dir(bio);
632 generic_end_io_acct(md->queue, bio_op(bio), &dm_disk(md)->part0,
633 io->start_time);
635 if (unlikely(dm_stats_used(&md->stats)))
636 dm_stats_account_io(&md->stats, bio_data_dir(bio),
637 bio->bi_iter.bi_sector, bio_sectors(bio),
638 true, duration, &io->stats_aux);
640 /*
641 * After this is decremented the bio must not be touched if it is
642 * a flush.
643 */
644 pending = atomic_dec_return(&md->pending[rw]);
645 atomic_set(&dm_disk(md)->part0.in_flight[rw], pending);
646 pending += atomic_read(&md->pending[rw^0x1]);
648 /* nudge anyone waiting on suspend queue */
649 if (!pending)
650 wake_up(&md->wait);
651 }
653 /*
654 * Add the bio to the list of deferred io.
655 */
656 static void queue_io(struct mapped_device *md, struct bio *bio)
657 {
658 unsigned long flags;
660 spin_lock_irqsave(&md->deferred_lock, flags);
661 bio_list_add(&md->deferred, bio);
662 spin_unlock_irqrestore(&md->deferred_lock, flags);
663 queue_work(md->wq, &md->work);
664 }
666 /*
667 * Everyone (including functions in this file), should use this
668 * function to access the md->map field, and make sure they call
669 * dm_put_live_table() when finished.
670 */
671 struct dm_table *dm_get_live_table(struct mapped_device *md, int *srcu_idx) __acquires(md->io_barrier)
672 {
673 *srcu_idx = srcu_read_lock(&md->io_barrier);
675 return srcu_dereference(md->map, &md->io_barrier);
676 }
678 void dm_put_live_table(struct mapped_device *md, int srcu_idx) __releases(md->io_barrier)
679 {
680 srcu_read_unlock(&md->io_barrier, srcu_idx);
681 }
683 void dm_sync_table(struct mapped_device *md)
684 {
685 synchronize_srcu(&md->io_barrier);
686 synchronize_rcu_expedited();
687 }
689 /*
690 * A fast alternative to dm_get_live_table/dm_put_live_table.
691 * The caller must not block between these two functions.
692 */
693 static struct dm_table *dm_get_live_table_fast(struct mapped_device *md) __acquires(RCU)
694 {
695 rcu_read_lock();
696 return rcu_dereference(md->map);
697 }
699 static void dm_put_live_table_fast(struct mapped_device *md) __releases(RCU)
700 {
701 rcu_read_unlock();
702 }
704 static char *_dm_claim_ptr = "I belong to device-mapper";
706 /*
707 * Open a table device so we can use it as a map destination.
708 */
709 static int open_table_device(struct table_device *td, dev_t dev,
710 struct mapped_device *md)
711 {
712 struct block_device *bdev;
714 int r;
716 BUG_ON(td->dm_dev.bdev);
718 bdev = blkdev_get_by_dev(dev, td->dm_dev.mode | FMODE_EXCL, _dm_claim_ptr);
719 if (IS_ERR(bdev))
720 return PTR_ERR(bdev);
722 r = bd_link_disk_holder(bdev, dm_disk(md));
723 if (r) {
724 blkdev_put(bdev, td->dm_dev.mode | FMODE_EXCL);
725 return r;
726 }
728 td->dm_dev.bdev = bdev;
729 td->dm_dev.dax_dev = dax_get_by_host(bdev->bd_disk->disk_name);
730 return 0;
731 }
733 /*
734 * Close a table device that we've been using.
735 */
736 static void close_table_device(struct table_device *td, struct mapped_device *md)
737 {
738 if (!td->dm_dev.bdev)
739 return;
741 bd_unlink_disk_holder(td->dm_dev.bdev, dm_disk(md));
742 blkdev_put(td->dm_dev.bdev, td->dm_dev.mode | FMODE_EXCL);
743 put_dax(td->dm_dev.dax_dev);
744 td->dm_dev.bdev = NULL;
745 td->dm_dev.dax_dev = NULL;
746 }
748 static struct table_device *find_table_device(struct list_head *l, dev_t dev,
749 fmode_t mode) {
750 struct table_device *td;
752 list_for_each_entry(td, l, list)
753 if (td->dm_dev.bdev->bd_dev == dev && td->dm_dev.mode == mode)
754 return td;
756 return NULL;
757 }
759 int dm_get_table_device(struct mapped_device *md, dev_t dev, fmode_t mode,
760 struct dm_dev **result) {
761 int r;
762 struct table_device *td;
764 mutex_lock(&md->table_devices_lock);
765 td = find_table_device(&md->table_devices, dev, mode);
766 if (!td) {
767 td = kmalloc_node(sizeof(*td), GFP_KERNEL, md->numa_node_id);
768 if (!td) {
769 mutex_unlock(&md->table_devices_lock);
770 return -ENOMEM;
771 }
773 td->dm_dev.mode = mode;
774 td->dm_dev.bdev = NULL;
776 if ((r = open_table_device(td, dev, md))) {
777 mutex_unlock(&md->table_devices_lock);
778 kfree(td);
779 return r;
780 }
782 format_dev_t(td->dm_dev.name, dev);
784 refcount_set(&td->count, 1);
785 list_add(&td->list, &md->table_devices);
786 } else {
787 refcount_inc(&td->count);
788 }
789 mutex_unlock(&md->table_devices_lock);
791 *result = &td->dm_dev;
792 return 0;
793 }
794 EXPORT_SYMBOL_GPL(dm_get_table_device);
796 void dm_put_table_device(struct mapped_device *md, struct dm_dev *d)
797 {
798 struct table_device *td = container_of(d, struct table_device, dm_dev);
800 mutex_lock(&md->table_devices_lock);
801 if (refcount_dec_and_test(&td->count)) {
802 close_table_device(td, md);
803 list_del(&td->list);
804 kfree(td);
805 }
806 mutex_unlock(&md->table_devices_lock);
807 }
808 EXPORT_SYMBOL(dm_put_table_device);
810 static void free_table_devices(struct list_head *devices)
811 {
812 struct list_head *tmp, *next;
814 list_for_each_safe(tmp, next, devices) {
815 struct table_device *td = list_entry(tmp, struct table_device, list);
817 DMWARN("dm_destroy: %s still exists with %d references",
818 td->dm_dev.name, refcount_read(&td->count));
819 kfree(td);
820 }
821 }
823 /*
824 * Get the geometry associated with a dm device
825 */
826 int dm_get_geometry(struct mapped_device *md, struct hd_geometry *geo)
827 {
828 *geo = md->geometry;
830 return 0;
831 }
833 /*
834 * Set the geometry of a device.
835 */
836 int dm_set_geometry(struct mapped_device *md, struct hd_geometry *geo)
837 {
838 sector_t sz = (sector_t)geo->cylinders * geo->heads * geo->sectors;
840 if (geo->start > sz) {
841 DMWARN("Start sector is beyond the geometry limits.");
842 return -EINVAL;
843 }
845 md->geometry = *geo;
847 return 0;
848 }
850 static int __noflush_suspending(struct mapped_device *md)
851 {
852 return test_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
853 }
855 /*
856 * Decrements the number of outstanding ios that a bio has been
857 * cloned into, completing the original io if necc.
858 */
859 static void dec_pending(struct dm_io *io, blk_status_t error)
860 {
861 unsigned long flags;
862 blk_status_t io_error;
863 struct bio *bio;
864 struct mapped_device *md = io->md;
866 /* Push-back supersedes any I/O errors */
867 if (unlikely(error)) {
868 spin_lock_irqsave(&io->endio_lock, flags);
869 if (!(io->status == BLK_STS_DM_REQUEUE && __noflush_suspending(md)))
870 io->status = error;
871 spin_unlock_irqrestore(&io->endio_lock, flags);
872 }
874 if (atomic_dec_and_test(&io->io_count)) {
875 if (io->status == BLK_STS_DM_REQUEUE) {
876 /*
877 * Target requested pushing back the I/O.
878 */
879 spin_lock_irqsave(&md->deferred_lock, flags);
880 if (__noflush_suspending(md))
881 /* NOTE early return due to BLK_STS_DM_REQUEUE below */
882 bio_list_add_head(&md->deferred, io->orig_bio);
883 else
884 /* noflush suspend was interrupted. */
885 io->status = BLK_STS_IOERR;
886 spin_unlock_irqrestore(&md->deferred_lock, flags);
887 }
889 io_error = io->status;
890 bio = io->orig_bio;
891 end_io_acct(io);
892 free_io(md, io);
894 if (io_error == BLK_STS_DM_REQUEUE)
895 return;
897 if ((bio->bi_opf & REQ_PREFLUSH) && bio->bi_iter.bi_size) {
898 /*
899 * Preflush done for flush with data, reissue
900 * without REQ_PREFLUSH.
901 */
902 bio->bi_opf &= ~REQ_PREFLUSH;
903 queue_io(md, bio);
904 } else {
905 /* done with normal IO or empty flush */
906 if (io_error)
907 bio->bi_status = io_error;
908 bio_endio(bio);
909 }
910 }
911 }
913 void disable_write_same(struct mapped_device *md)
914 {
915 struct queue_limits *limits = dm_get_queue_limits(md);
917 /* device doesn't really support WRITE SAME, disable it */
918 limits->max_write_same_sectors = 0;
919 }
921 void disable_write_zeroes(struct mapped_device *md)
922 {
923 struct queue_limits *limits = dm_get_queue_limits(md);
925 /* device doesn't really support WRITE ZEROES, disable it */
926 limits->max_write_zeroes_sectors = 0;
927 }
929 static void clone_endio(struct bio *bio)
930 {
931 blk_status_t error = bio->bi_status;
932 struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone);
933 struct dm_io *io = tio->io;
934 struct mapped_device *md = tio->io->md;
935 dm_endio_fn endio = tio->ti->type->end_io;
937 if (unlikely(error == BLK_STS_TARGET) && md->type != DM_TYPE_NVME_BIO_BASED) {
938 if (bio_op(bio) == REQ_OP_WRITE_SAME &&
939 !bio->bi_disk->queue->limits.max_write_same_sectors)
940 disable_write_same(md);
941 if (bio_op(bio) == REQ_OP_WRITE_ZEROES &&
942 !bio->bi_disk->queue->limits.max_write_zeroes_sectors)
943 disable_write_zeroes(md);
944 }
946 if (endio) {
947 int r = endio(tio->ti, bio, &error);
948 switch (r) {
949 case DM_ENDIO_REQUEUE:
950 error = BLK_STS_DM_REQUEUE;
951 /*FALLTHRU*/
952 case DM_ENDIO_DONE:
953 break;
954 case DM_ENDIO_INCOMPLETE:
955 /* The target will handle the io */
956 return;
957 default:
958 DMWARN("unimplemented target endio return value: %d", r);
959 BUG();
960 }
961 }
963 free_tio(tio);
964 dec_pending(io, error);
965 }
967 /*
968 * Return maximum size of I/O possible at the supplied sector up to the current
969 * target boundary.
970 */
971 static sector_t max_io_len_target_boundary(sector_t sector, struct dm_target *ti)
972 {
973 sector_t target_offset = dm_target_offset(ti, sector);
975 return ti->len - target_offset;
976 }
978 static sector_t max_io_len(sector_t sector, struct dm_target *ti)
979 {
980 sector_t len = max_io_len_target_boundary(sector, ti);
981 sector_t offset, max_len;
983 /*
984 * Does the target need to split even further?
985 */
986 if (ti->max_io_len) {
987 offset = dm_target_offset(ti, sector);
988 if (unlikely(ti->max_io_len & (ti->max_io_len - 1)))
989 max_len = sector_div(offset, ti->max_io_len);
990 else
991 max_len = offset & (ti->max_io_len - 1);
992 max_len = ti->max_io_len - max_len;
994 if (len > max_len)
995 len = max_len;
996 }
998 return len;
999 }
1001 int dm_set_target_max_io_len(struct dm_target *ti, sector_t len)
1002 {
1003 if (len > UINT_MAX) {
1004 DMERR("Specified maximum size of target IO (%llu) exceeds limit (%u)",
1005 (unsigned long long)len, UINT_MAX);
1006 ti->error = "Maximum size of target IO is too large";
1007 return -EINVAL;
1008 }
1010 /*
1011 * BIO based queue uses its own splitting. When multipage bvecs
1012 * is switched on, size of the incoming bio may be too big to
1013 * be handled in some targets, such as crypt.
1014 *
1015 * When these targets are ready for the big bio, we can remove
1016 * the limit.
1017 */
1018 ti->max_io_len = min_t(uint32_t, len, BIO_MAX_PAGES * PAGE_SIZE);
1020 return 0;
1021 }
1022 EXPORT_SYMBOL_GPL(dm_set_target_max_io_len);
1024 static struct dm_target *dm_dax_get_live_target(struct mapped_device *md,
1025 sector_t sector, int *srcu_idx)
1026 __acquires(md->io_barrier)
1027 {
1028 struct dm_table *map;
1029 struct dm_target *ti;
1031 map = dm_get_live_table(md, srcu_idx);
1032 if (!map)
1033 return NULL;
1035 ti = dm_table_find_target(map, sector);
1036 if (!dm_target_is_valid(ti))
1037 return NULL;
1039 return ti;
1040 }
1042 static long dm_dax_direct_access(struct dax_device *dax_dev, pgoff_t pgoff,
1043 long nr_pages, void **kaddr, pfn_t *pfn)
1044 {
1045 struct mapped_device *md = dax_get_private(dax_dev);
1046 sector_t sector = pgoff * PAGE_SECTORS;
1047 struct dm_target *ti;
1048 long len, ret = -EIO;
1049 int srcu_idx;
1051 ti = dm_dax_get_live_target(md, sector, &srcu_idx);
1053 if (!ti)
1054 goto out;
1055 if (!ti->type->direct_access)
1056 goto out;
1057 len = max_io_len(sector, ti) / PAGE_SECTORS;
1058 if (len < 1)
1059 goto out;
1060 nr_pages = min(len, nr_pages);
1061 ret = ti->type->direct_access(ti, pgoff, nr_pages, kaddr, pfn);
1063 out:
1064 dm_put_live_table(md, srcu_idx);
1066 return ret;
1067 }
1069 static size_t dm_dax_copy_from_iter(struct dax_device *dax_dev, pgoff_t pgoff,
1070 void *addr, size_t bytes, struct iov_iter *i)
1071 {
1072 struct mapped_device *md = dax_get_private(dax_dev);
1073 sector_t sector = pgoff * PAGE_SECTORS;
1074 struct dm_target *ti;
1075 long ret = 0;
1076 int srcu_idx;
1078 ti = dm_dax_get_live_target(md, sector, &srcu_idx);
1080 if (!ti)
1081 goto out;
1082 if (!ti->type->dax_copy_from_iter) {
1083 ret = copy_from_iter(addr, bytes, i);
1084 goto out;
1085 }
1086 ret = ti->type->dax_copy_from_iter(ti, pgoff, addr, bytes, i);
1087 out:
1088 dm_put_live_table(md, srcu_idx);
1090 return ret;
1091 }
1093 static size_t dm_dax_copy_to_iter(struct dax_device *dax_dev, pgoff_t pgoff,
1094 void *addr, size_t bytes, struct iov_iter *i)
1095 {
1096 struct mapped_device *md = dax_get_private(dax_dev);
1097 sector_t sector = pgoff * PAGE_SECTORS;
1098 struct dm_target *ti;
1099 long ret = 0;
1100 int srcu_idx;
1102 ti = dm_dax_get_live_target(md, sector, &srcu_idx);
1104 if (!ti)
1105 goto out;
1106 if (!ti->type->dax_copy_to_iter) {
1107 ret = copy_to_iter(addr, bytes, i);
1108 goto out;
1109 }
1110 ret = ti->type->dax_copy_to_iter(ti, pgoff, addr, bytes, i);
1111 out:
1112 dm_put_live_table(md, srcu_idx);
1114 return ret;
1115 }
1117 /*
1118 * A target may call dm_accept_partial_bio only from the map routine. It is
1119 * allowed for all bio types except REQ_PREFLUSH and REQ_OP_ZONE_RESET.
1120 *
1121 * dm_accept_partial_bio informs the dm that the target only wants to process
1122 * additional n_sectors sectors of the bio and the rest of the data should be
1123 * sent in a next bio.
1124 *
1125 * A diagram that explains the arithmetics:
1126 * +--------------------+---------------+-------+
1127 * | 1 | 2 | 3 |
1128 * +--------------------+---------------+-------+
1129 *
1130 * <-------------- *tio->len_ptr --------------->
1131 * <------- bi_size ------->
1132 * <-- n_sectors -->
1133 *
1134 * Region 1 was already iterated over with bio_advance or similar function.
1135 * (it may be empty if the target doesn't use bio_advance)
1136 * Region 2 is the remaining bio size that the target wants to process.
1137 * (it may be empty if region 1 is non-empty, although there is no reason
1138 * to make it empty)
1139 * The target requires that region 3 is to be sent in the next bio.
1140 *
1141 * If the target wants to receive multiple copies of the bio (via num_*bios, etc),
1142 * the partially processed part (the sum of regions 1+2) must be the same for all
1143 * copies of the bio.
1144 */
1145 void dm_accept_partial_bio(struct bio *bio, unsigned n_sectors)
1146 {
1147 struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone);
1148 unsigned bi_size = bio->bi_iter.bi_size >> SECTOR_SHIFT;
1149 BUG_ON(bio->bi_opf & REQ_PREFLUSH);
1150 BUG_ON(bi_size > *tio->len_ptr);
1151 BUG_ON(n_sectors > bi_size);
1152 *tio->len_ptr -= bi_size - n_sectors;
1153 bio->bi_iter.bi_size = n_sectors << SECTOR_SHIFT;
1154 }
1155 EXPORT_SYMBOL_GPL(dm_accept_partial_bio);
1157 /*
1158 * The zone descriptors obtained with a zone report indicate zone positions
1159 * within the target backing device, regardless of that device is a partition
1160 * and regardless of the target mapping start sector on the device or partition.
1161 * The zone descriptors start sector and write pointer position must be adjusted
1162 * to match their relative position within the dm device.
1163 * A target may call dm_remap_zone_report() after completion of a
1164 * REQ_OP_ZONE_REPORT bio to remap the zone descriptors obtained from the
1165 * backing device.
1166 */
1167 void dm_remap_zone_report(struct dm_target *ti, struct bio *bio, sector_t start)
1168 {
1169 #ifdef CONFIG_BLK_DEV_ZONED
1170 struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone);
1171 struct bio *report_bio = tio->io->orig_bio;
1172 struct blk_zone_report_hdr *hdr = NULL;
1173 struct blk_zone *zone;
1174 unsigned int nr_rep = 0;
1175 unsigned int ofst;
1176 sector_t part_offset;
1177 struct bio_vec bvec;
1178 struct bvec_iter iter;
1179 void *addr;
1181 if (bio->bi_status)
1182 return;
1184 /*
1185 * bio sector was incremented by the request size on completion. Taking
1186 * into account the original request sector, the target start offset on
1187 * the backing device and the target mapping offset (ti->begin), the
1188 * start sector of the backing device. The partition offset is always 0
1189 * if the target uses a whole device.
1190 */
1191 part_offset = bio->bi_iter.bi_sector + ti->begin - (start + bio_end_sector(report_bio));
1193 /*
1194 * Remap the start sector of the reported zones. For sequential zones,
1195 * also remap the write pointer position.
1196 */
1197 bio_for_each_segment(bvec, report_bio, iter) {
1198 addr = kmap_atomic(bvec.bv_page);
1200 /* Remember the report header in the first page */
1201 if (!hdr) {
1202 hdr = addr;
1203 ofst = sizeof(struct blk_zone_report_hdr);
1204 } else
1205 ofst = 0;
1207 /* Set zones start sector */
1208 while (hdr->nr_zones && ofst < bvec.bv_len) {
1209 zone = addr + ofst;
1210 zone->start -= part_offset;
1211 if (zone->start >= start + ti->len) {
1212 hdr->nr_zones = 0;
1213 break;
1214 }
1215 zone->start = zone->start + ti->begin - start;
1216 if (zone->type != BLK_ZONE_TYPE_CONVENTIONAL) {
1217 if (zone->cond == BLK_ZONE_COND_FULL)
1218 zone->wp = zone->start + zone->len;
1219 else if (zone->cond == BLK_ZONE_COND_EMPTY)
1220 zone->wp = zone->start;
1221 else
1222 zone->wp = zone->wp + ti->begin - start - part_offset;
1223 }
1224 ofst += sizeof(struct blk_zone);
1225 hdr->nr_zones--;
1226 nr_rep++;
1227 }
1229 if (addr != hdr)
1230 kunmap_atomic(addr);
1232 if (!hdr->nr_zones)
1233 break;
1234 }
1236 if (hdr) {
1237 hdr->nr_zones = nr_rep;
1238 kunmap_atomic(hdr);
1239 }
1241 bio_advance(report_bio, report_bio->bi_iter.bi_size);
1243 #else /* !CONFIG_BLK_DEV_ZONED */
1244 bio->bi_status = BLK_STS_NOTSUPP;
1245 #endif
1246 }
1247 EXPORT_SYMBOL_GPL(dm_remap_zone_report);
1249 static blk_qc_t __map_bio(struct dm_target_io *tio)
1250 {
1251 int r;
1252 sector_t sector;
1253 struct bio *clone = &tio->clone;
1254 struct dm_io *io = tio->io;
1255 struct mapped_device *md = io->md;
1256 struct dm_target *ti = tio->ti;
1257 blk_qc_t ret = BLK_QC_T_NONE;
1259 clone->bi_end_io = clone_endio;
1261 /*
1262 * Map the clone. If r == 0 we don't need to do
1263 * anything, the target has assumed ownership of
1264 * this io.
1265 */
1266 atomic_inc(&io->io_count);
1267 sector = clone->bi_iter.bi_sector;
1269 r = ti->type->map(ti, clone);
1270 switch (r) {
1271 case DM_MAPIO_SUBMITTED:
1272 break;
1273 case DM_MAPIO_REMAPPED:
1274 /* the bio has been remapped so dispatch it */
1275 trace_block_bio_remap(clone->bi_disk->queue, clone,
1276 bio_dev(io->orig_bio), sector);
1277 if (md->type == DM_TYPE_NVME_BIO_BASED)
1278 ret = direct_make_request(clone);
1279 else
1280 ret = generic_make_request(clone);
1281 break;
1282 case DM_MAPIO_KILL:
1283 free_tio(tio);
1284 dec_pending(io, BLK_STS_IOERR);
1285 break;
1286 case DM_MAPIO_REQUEUE:
1287 free_tio(tio);
1288 dec_pending(io, BLK_STS_DM_REQUEUE);
1289 break;
1290 default:
1291 DMWARN("unimplemented target map return value: %d", r);
1292 BUG();
1293 }
1295 return ret;
1296 }
1298 static void bio_setup_sector(struct bio *bio, sector_t sector, unsigned len)
1299 {
1300 bio->bi_iter.bi_sector = sector;
1301 bio->bi_iter.bi_size = to_bytes(len);
1302 }
1304 /*
1305 * Creates a bio that consists of range of complete bvecs.
1306 */
1307 static int clone_bio(struct dm_target_io *tio, struct bio *bio,
1308 sector_t sector, unsigned len)
1309 {
1310 struct bio *clone = &tio->clone;
1312 __bio_clone_fast(clone, bio);
1314 if (unlikely(bio_integrity(bio) != NULL)) {
1315 int r;
1317 if (unlikely(!dm_target_has_integrity(tio->ti->type) &&
1318 !dm_target_passes_integrity(tio->ti->type))) {
1319 DMWARN("%s: the target %s doesn't support integrity data.",
1320 dm_device_name(tio->io->md),
1321 tio->ti->type->name);
1322 return -EIO;
1323 }
1325 r = bio_integrity_clone(clone, bio, GFP_NOIO);
1326 if (r < 0)
1327 return r;
1328 }
1330 if (bio_op(bio) != REQ_OP_ZONE_REPORT)
1331 bio_advance(clone, to_bytes(sector - clone->bi_iter.bi_sector));
1332 clone->bi_iter.bi_size = to_bytes(len);
1334 if (unlikely(bio_integrity(bio) != NULL))
1335 bio_integrity_trim(clone);
1337 return 0;
1338 }
1340 static void alloc_multiple_bios(struct bio_list *blist, struct clone_info *ci,
1341 struct dm_target *ti, unsigned num_bios)
1342 {
1343 struct dm_target_io *tio;
1344 int try;
1346 if (!num_bios)
1347 return;
1349 if (num_bios == 1) {
1350 tio = alloc_tio(ci, ti, 0, GFP_NOIO);
1351 bio_list_add(blist, &tio->clone);
1352 return;
1353 }
1355 for (try = 0; try < 2; try++) {
1356 int bio_nr;
1357 struct bio *bio;
1359 if (try)
1360 mutex_lock(&ci->io->md->table_devices_lock);
1361 for (bio_nr = 0; bio_nr < num_bios; bio_nr++) {
1362 tio = alloc_tio(ci, ti, bio_nr, try ? GFP_NOIO : GFP_NOWAIT);
1363 if (!tio)
1364 break;
1366 bio_list_add(blist, &tio->clone);
1367 }
1368 if (try)
1369 mutex_unlock(&ci->io->md->table_devices_lock);
1370 if (bio_nr == num_bios)
1371 return;
1373 while ((bio = bio_list_pop(blist))) {
1374 tio = container_of(bio, struct dm_target_io, clone);
1375 free_tio(tio);
1376 }
1377 }
1378 }
1380 static blk_qc_t __clone_and_map_simple_bio(struct clone_info *ci,
1381 struct dm_target_io *tio, unsigned *len)
1382 {
1383 struct bio *clone = &tio->clone;
1385 tio->len_ptr = len;
1387 __bio_clone_fast(clone, ci->bio);
1388 if (len)
1389 bio_setup_sector(clone, ci->sector, *len);
1391 return __map_bio(tio);
1392 }
1394 static void __send_duplicate_bios(struct clone_info *ci, struct dm_target *ti,
1395 unsigned num_bios, unsigned *len)
1396 {
1397 struct bio_list blist = BIO_EMPTY_LIST;
1398 struct bio *bio;
1399 struct dm_target_io *tio;
1401 alloc_multiple_bios(&blist, ci, ti, num_bios);
1403 while ((bio = bio_list_pop(&blist))) {
1404 tio = container_of(bio, struct dm_target_io, clone);
1405 (void) __clone_and_map_simple_bio(ci, tio, len);
1406 }
1407 }
1409 static int __send_empty_flush(struct clone_info *ci)
1410 {
1411 unsigned target_nr = 0;
1412 struct dm_target *ti;
1414 BUG_ON(bio_has_data(ci->bio));
1415 while ((ti = dm_table_get_target(ci->map, target_nr++)))
1416 __send_duplicate_bios(ci, ti, ti->num_flush_bios, NULL);
1418 return 0;
1419 }
1421 static int __clone_and_map_data_bio(struct clone_info *ci, struct dm_target *ti,
1422 sector_t sector, unsigned *len)
1423 {
1424 struct bio *bio = ci->bio;
1425 struct dm_target_io *tio;
1426 int r;
1428 tio = alloc_tio(ci, ti, 0, GFP_NOIO);
1429 tio->len_ptr = len;
1430 r = clone_bio(tio, bio, sector, *len);
1431 if (r < 0) {
1432 free_tio(tio);
1433 return r;
1434 }
1435 (void) __map_bio(tio);
1437 return 0;
1438 }
1440 typedef unsigned (*get_num_bios_fn)(struct dm_target *ti);
1442 static unsigned get_num_discard_bios(struct dm_target *ti)
1443 {
1444 return ti->num_discard_bios;
1445 }
1447 static unsigned get_num_secure_erase_bios(struct dm_target *ti)
1448 {
1449 return ti->num_secure_erase_bios;
1450 }
1452 static unsigned get_num_write_same_bios(struct dm_target *ti)
1453 {
1454 return ti->num_write_same_bios;
1455 }
1457 static unsigned get_num_write_zeroes_bios(struct dm_target *ti)
1458 {
1459 return ti->num_write_zeroes_bios;
1460 }
1462 typedef bool (*is_split_required_fn)(struct dm_target *ti);
1464 static bool is_split_required_for_discard(struct dm_target *ti)
1465 {
1466 return ti->split_discard_bios;
1467 }
1469 static int __send_changing_extent_only(struct clone_info *ci, struct dm_target *ti,
1470 get_num_bios_fn get_num_bios,
1471 is_split_required_fn is_split_required)
1472 {
1473 unsigned len;
1474 unsigned num_bios;
1476 /*
1477 * Even though the device advertised support for this type of
1478 * request, that does not mean every target supports it, and
1479 * reconfiguration might also have changed that since the
1480 * check was performed.
1481 */
1482 num_bios = get_num_bios ? get_num_bios(ti) : 0;
1483 if (!num_bios)
1484 return -EOPNOTSUPP;
1486 if (is_split_required && !is_split_required(ti))
1487 len = min((sector_t)ci->sector_count, max_io_len_target_boundary(ci->sector, ti));
1488 else
1489 len = min((sector_t)ci->sector_count, max_io_len(ci->sector, ti));
1491 __send_duplicate_bios(ci, ti, num_bios, &len);
1493 ci->sector += len;
1494 ci->sector_count -= len;
1496 return 0;
1497 }
1499 static int __send_discard(struct clone_info *ci, struct dm_target *ti)
1500 {
1501 return __send_changing_extent_only(ci, ti, get_num_discard_bios,
1502 is_split_required_for_discard);
1503 }
1505 static int __send_secure_erase(struct clone_info *ci, struct dm_target *ti)
1506 {
1507 return __send_changing_extent_only(ci, ti, get_num_secure_erase_bios, NULL);
1508 }
1510 static int __send_write_same(struct clone_info *ci, struct dm_target *ti)
1511 {
1512 return __send_changing_extent_only(ci, ti, get_num_write_same_bios, NULL);
1513 }
1515 static int __send_write_zeroes(struct clone_info *ci, struct dm_target *ti)
1516 {
1517 return __send_changing_extent_only(ci, ti, get_num_write_zeroes_bios, NULL);
1518 }
1520 static bool __process_abnormal_io(struct clone_info *ci, struct dm_target *ti,
1521 int *result)
1522 {
1523 struct bio *bio = ci->bio;
1525 if (bio_op(bio) == REQ_OP_DISCARD)
1526 *result = __send_discard(ci, ti);
1527 else if (bio_op(bio) == REQ_OP_SECURE_ERASE)
1528 *result = __send_secure_erase(ci, ti);
1529 else if (bio_op(bio) == REQ_OP_WRITE_SAME)
1530 *result = __send_write_same(ci, ti);
1531 else if (bio_op(bio) == REQ_OP_WRITE_ZEROES)
1532 *result = __send_write_zeroes(ci, ti);
1533 else
1534 return false;
1536 return true;
1537 }
1539 /*
1540 * Select the correct strategy for processing a non-flush bio.
1541 */
1542 static int __split_and_process_non_flush(struct clone_info *ci)
1543 {
1544 struct bio *bio = ci->bio;
1545 struct dm_target *ti;
1546 unsigned len;
1547 int r;
1549 ti = dm_table_find_target(ci->map, ci->sector);
1550 if (!dm_target_is_valid(ti))
1551 return -EIO;
1553 if (unlikely(__process_abnormal_io(ci, ti, &r)))
1554 return r;
1556 if (bio_op(bio) == REQ_OP_ZONE_REPORT)
1557 len = ci->sector_count;
1558 else
1559 len = min_t(sector_t, max_io_len(ci->sector, ti),
1560 ci->sector_count);
1562 r = __clone_and_map_data_bio(ci, ti, ci->sector, &len);
1563 if (r < 0)
1564 return r;
1566 ci->sector += len;
1567 ci->sector_count -= len;
1569 return 0;
1570 }
1572 static void init_clone_info(struct clone_info *ci, struct mapped_device *md,
1573 struct dm_table *map, struct bio *bio)
1574 {
1575 ci->map = map;
1576 ci->io = alloc_io(md, bio);
1577 ci->sector = bio->bi_iter.bi_sector;
1578 }
1580 /*
1581 * Entry point to split a bio into clones and submit them to the targets.
1582 */
1583 static blk_qc_t __split_and_process_bio(struct mapped_device *md,
1584 struct dm_table *map, struct bio *bio)
1585 {
1586 struct clone_info ci;
1587 blk_qc_t ret = BLK_QC_T_NONE;
1588 int error = 0;
1590 if (unlikely(!map)) {
1591 bio_io_error(bio);
1592 return ret;
1593 }
1595 init_clone_info(&ci, md, map, bio);
1597 if (bio->bi_opf & REQ_PREFLUSH) {
1598 ci.bio = &ci.io->md->flush_bio;
1599 ci.sector_count = 0;
1600 error = __send_empty_flush(&ci);
1601 /* dec_pending submits any data associated with flush */
1602 } else if (bio_op(bio) == REQ_OP_ZONE_RESET) {
1603 ci.bio = bio;
1604 ci.sector_count = 0;
1605 error = __split_and_process_non_flush(&ci);
1606 } else {
1607 ci.bio = bio;
1608 ci.sector_count = bio_sectors(bio);
1609 while (ci.sector_count && !error) {
1610 error = __split_and_process_non_flush(&ci);
1611 if (current->bio_list && ci.sector_count && !error) {
1612 /*
1613 * Remainder must be passed to generic_make_request()
1614 * so that it gets handled *after* bios already submitted
1615 * have been completely processed.
1616 * We take a clone of the original to store in
1617 * ci.io->orig_bio to be used by end_io_acct() and
1618 * for dec_pending to use for completion handling.
1619 * As this path is not used for REQ_OP_ZONE_REPORT,
1620 * the usage of io->orig_bio in dm_remap_zone_report()
1621 * won't be affected by this reassignment.
1622 */
1623 struct bio *b = bio_split(bio, bio_sectors(bio) - ci.sector_count,
1624 GFP_NOIO, &md->queue->bio_split);
1625 ci.io->orig_bio = b;
1626 bio_chain(b, bio);
1627 ret = generic_make_request(bio);
1628 break;
1629 }
1630 }
1631 }
1633 /* drop the extra reference count */
1634 dec_pending(ci.io, errno_to_blk_status(error));
1635 return ret;
1636 }
1638 /*
1639 * Optimized variant of __split_and_process_bio that leverages the
1640 * fact that targets that use it do _not_ have a need to split bios.
1641 */
1642 static blk_qc_t __process_bio(struct mapped_device *md,
1643 struct dm_table *map, struct bio *bio)
1644 {
1645 struct clone_info ci;
1646 blk_qc_t ret = BLK_QC_T_NONE;
1647 int error = 0;
1649 if (unlikely(!map)) {
1650 bio_io_error(bio);
1651 return ret;
1652 }
1654 init_clone_info(&ci, md, map, bio);
1656 if (bio->bi_opf & REQ_PREFLUSH) {
1657 ci.bio = &ci.io->md->flush_bio;
1658 ci.sector_count = 0;
1659 error = __send_empty_flush(&ci);
1660 /* dec_pending submits any data associated with flush */
1661 } else {
1662 struct dm_target *ti = md->immutable_target;
1663 struct dm_target_io *tio;
1665 /*
1666 * Defend against IO still getting in during teardown
1667 * - as was seen for a time with nvme-fcloop
1668 */
1669 if (unlikely(WARN_ON_ONCE(!ti || !dm_target_is_valid(ti)))) {
1670 error = -EIO;
1671 goto out;
1672 }
1674 ci.bio = bio;
1675 ci.sector_count = bio_sectors(bio);
1676 if (unlikely(__process_abnormal_io(&ci, ti, &error)))
1677 goto out;
1679 tio = alloc_tio(&ci, ti, 0, GFP_NOIO);
1680 ret = __clone_and_map_simple_bio(&ci, tio, NULL);
1681 }
1682 out:
1683 /* drop the extra reference count */
1684 dec_pending(ci.io, errno_to_blk_status(error));
1685 return ret;
1686 }
1688 typedef blk_qc_t (process_bio_fn)(struct mapped_device *, struct dm_table *, struct bio *);
1690 static blk_qc_t __dm_make_request(struct request_queue *q, struct bio *bio,
1691 process_bio_fn process_bio)
1692 {
1693 struct mapped_device *md = q->queuedata;
1694 blk_qc_t ret = BLK_QC_T_NONE;
1695 int srcu_idx;
1696 struct dm_table *map;
1698 map = dm_get_live_table(md, &srcu_idx);
1700 /* if we're suspended, we have to queue this io for later */
1701 if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags))) {
1702 dm_put_live_table(md, srcu_idx);
1704 if (!(bio->bi_opf & REQ_RAHEAD))
1705 queue_io(md, bio);
1706 else
1707 bio_io_error(bio);
1708 return ret;
1709 }
1711 ret = process_bio(md, map, bio);
1713 dm_put_live_table(md, srcu_idx);
1714 return ret;
1715 }
1717 /*
1718 * The request function that remaps the bio to one target and
1719 * splits off any remainder.
1720 */
1721 static blk_qc_t dm_make_request(struct request_queue *q, struct bio *bio)
1722 {
1723 return __dm_make_request(q, bio, __split_and_process_bio);
1724 }
1726 static blk_qc_t dm_make_request_nvme(struct request_queue *q, struct bio *bio)
1727 {
1728 return __dm_make_request(q, bio, __process_bio);
1729 }
1731 static int dm_any_congested(void *congested_data, int bdi_bits)
1732 {
1733 int r = bdi_bits;
1734 struct mapped_device *md = congested_data;
1735 struct dm_table *map;
1737 if (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
1738 if (dm_request_based(md)) {
1739 /*
1740 * With request-based DM we only need to check the
1741 * top-level queue for congestion.
1742 */
1743 r = md->queue->backing_dev_info->wb.state & bdi_bits;
1744 } else {
1745 map = dm_get_live_table_fast(md);
1746 if (map)
1747 r = dm_table_any_congested(map, bdi_bits);
1748 dm_put_live_table_fast(md);
1749 }
1750 }
1752 return r;
1753 }
1755 /*-----------------------------------------------------------------
1756 * An IDR is used to keep track of allocated minor numbers.
1757 *---------------------------------------------------------------*/
1758 static void free_minor(int minor)
1759 {
1760 spin_lock(&_minor_lock);
1761 idr_remove(&_minor_idr, minor);
1762 spin_unlock(&_minor_lock);
1763 }
1765 /*
1766 * See if the device with a specific minor # is free.
1767 */
1768 static int specific_minor(int minor)
1769 {
1770 int r;
1772 if (minor >= (1 << MINORBITS))
1773 return -EINVAL;
1775 idr_preload(GFP_KERNEL);
1776 spin_lock(&_minor_lock);
1778 r = idr_alloc(&_minor_idr, MINOR_ALLOCED, minor, minor + 1, GFP_NOWAIT);
1780 spin_unlock(&_minor_lock);
1781 idr_preload_end();
1782 if (r < 0)
1783 return r == -ENOSPC ? -EBUSY : r;
1784 return 0;
1785 }
1787 static int next_free_minor(int *minor)
1788 {
1789 int r;
1791 idr_preload(GFP_KERNEL);
1792 spin_lock(&_minor_lock);
1794 r = idr_alloc(&_minor_idr, MINOR_ALLOCED, 0, 1 << MINORBITS, GFP_NOWAIT);
1796 spin_unlock(&_minor_lock);
1797 idr_preload_end();
1798 if (r < 0)
1799 return r;
1800 *minor = r;
1801 return 0;
1802 }
1804 static const struct block_device_operations dm_blk_dops;
1805 static const struct dax_operations dm_dax_ops;
1807 static void dm_wq_work(struct work_struct *work);
1809 static void dm_init_normal_md_queue(struct mapped_device *md)
1810 {
1811 md->use_blk_mq = false;
1813 /*
1814 * Initialize aspects of queue that aren't relevant for blk-mq
1815 */
1816 md->queue->backing_dev_info->congested_fn = dm_any_congested;
1817 }
1819 static void cleanup_mapped_device(struct mapped_device *md)
1820 {
1821 if (md->wq)
1822 destroy_workqueue(md->wq);
1823 if (md->kworker_task)
1824 kthread_stop(md->kworker_task);
1825 bioset_exit(&md->bs);
1826 bioset_exit(&md->io_bs);
1828 if (md->dax_dev) {
1829 kill_dax(md->dax_dev);
1830 put_dax(md->dax_dev);
1831 md->dax_dev = NULL;
1832 }
1834 if (md->disk) {
1835 spin_lock(&_minor_lock);
1836 md->disk->private_data = NULL;
1837 spin_unlock(&_minor_lock);
1838 del_gendisk(md->disk);
1839 put_disk(md->disk);
1840 }
1842 if (md->queue)
1843 blk_cleanup_queue(md->queue);
1845 cleanup_srcu_struct(&md->io_barrier);
1847 if (md->bdev) {
1848 bdput(md->bdev);
1849 md->bdev = NULL;
1850 }
1852 mutex_destroy(&md->suspend_lock);
1853 mutex_destroy(&md->type_lock);
1854 mutex_destroy(&md->table_devices_lock);
1856 dm_mq_cleanup_mapped_device(md);
1857 }
1859 /*
1860 * Allocate and initialise a blank device with a given minor.
1861 */
1862 static struct mapped_device *alloc_dev(int minor)
1863 {
1864 int r, numa_node_id = dm_get_numa_node();
1865 struct dax_device *dax_dev = NULL;
1866 struct mapped_device *md;
1867 void *old_md;
1869 md = kvzalloc_node(sizeof(*md), GFP_KERNEL, numa_node_id);
1870 if (!md) {
1871 DMWARN("unable to allocate device, out of memory.");
1872 return NULL;
1873 }
1875 if (!try_module_get(THIS_MODULE))
1876 goto bad_module_get;
1878 /* get a minor number for the dev */
1879 if (minor == DM_ANY_MINOR)
1880 r = next_free_minor(&minor);
1881 else
1882 r = specific_minor(minor);
1883 if (r < 0)
1884 goto bad_minor;
1886 r = init_srcu_struct(&md->io_barrier);
1887 if (r < 0)
1888 goto bad_io_barrier;
1890 md->numa_node_id = numa_node_id;
1891 md->use_blk_mq = dm_use_blk_mq_default();
1892 md->init_tio_pdu = false;
1893 md->type = DM_TYPE_NONE;
1894 mutex_init(&md->suspend_lock);
1895 mutex_init(&md->type_lock);
1896 mutex_init(&md->table_devices_lock);
1897 spin_lock_init(&md->deferred_lock);
1898 atomic_set(&md->holders, 1);
1899 atomic_set(&md->open_count, 0);
1900 atomic_set(&md->event_nr, 0);
1901 atomic_set(&md->uevent_seq, 0);
1902 INIT_LIST_HEAD(&md->uevent_list);
1903 INIT_LIST_HEAD(&md->table_devices);
1904 spin_lock_init(&md->uevent_lock);
1906 md->queue = blk_alloc_queue_node(GFP_KERNEL, numa_node_id, NULL);
1907 if (!md->queue)
1908 goto bad;
1909 md->queue->queuedata = md;
1910 md->queue->backing_dev_info->congested_data = md;
1912 md->disk = alloc_disk_node(1, md->numa_node_id);
1913 if (!md->disk)
1914 goto bad;
1916 atomic_set(&md->pending[0], 0);
1917 atomic_set(&md->pending[1], 0);
1918 init_waitqueue_head(&md->wait);
1919 INIT_WORK(&md->work, dm_wq_work);
1920 init_waitqueue_head(&md->eventq);
1921 init_completion(&md->kobj_holder.completion);
1922 md->kworker_task = NULL;
1924 md->disk->major = _major;
1925 md->disk->first_minor = minor;
1926 md->disk->fops = &dm_blk_dops;
1927 md->disk->queue = md->queue;
1928 md->disk->private_data = md;
1929 sprintf(md->disk->disk_name, "dm-%d", minor);
1931 if (IS_ENABLED(CONFIG_DAX_DRIVER)) {
1932 dax_dev = alloc_dax(md, md->disk->disk_name, &dm_dax_ops);
1933 if (!dax_dev)
1934 goto bad;
1935 }
1936 md->dax_dev = dax_dev;
1938 add_disk_no_queue_reg(md->disk);
1939 format_dev_t(md->name, MKDEV(_major, minor));
1941 md->wq = alloc_workqueue("kdmflush", WQ_MEM_RECLAIM, 0);
1942 if (!md->wq)
1943 goto bad;
1945 md->bdev = bdget_disk(md->disk, 0);
1946 if (!md->bdev)
1947 goto bad;
1949 bio_init(&md->flush_bio, NULL, 0);
1950 bio_set_dev(&md->flush_bio, md->bdev);
1951 md->flush_bio.bi_opf = REQ_OP_WRITE | REQ_PREFLUSH | REQ_SYNC;
1953 dm_stats_init(&md->stats);
1955 /* Populate the mapping, nobody knows we exist yet */
1956 spin_lock(&_minor_lock);
1957 old_md = idr_replace(&_minor_idr, md, minor);
1958 spin_unlock(&_minor_lock);
1960 BUG_ON(old_md != MINOR_ALLOCED);
1962 return md;
1964 bad:
1965 cleanup_mapped_device(md);
1966 bad_io_barrier:
1967 free_minor(minor);
1968 bad_minor:
1969 module_put(THIS_MODULE);
1970 bad_module_get:
1971 kvfree(md);
1972 return NULL;
1973 }
1975 static void unlock_fs(struct mapped_device *md);
1977 static void free_dev(struct mapped_device *md)
1978 {
1979 int minor = MINOR(disk_devt(md->disk));
1981 unlock_fs(md);
1983 cleanup_mapped_device(md);
1985 free_table_devices(&md->table_devices);
1986 dm_stats_cleanup(&md->stats);
1987 free_minor(minor);
1989 module_put(THIS_MODULE);
1990 kvfree(md);
1991 }
1993 static int __bind_mempools(struct mapped_device *md, struct dm_table *t)
1994 {
1995 struct dm_md_mempools *p = dm_table_get_md_mempools(t);
1996 int ret = 0;
1998 if (dm_table_bio_based(t)) {
1999 /*
2000 * The md may already have mempools that need changing.
2001 * If so, reload bioset because front_pad may have changed
2002 * because a different table was loaded.
2003 */
2004 bioset_exit(&md->bs);
2005 bioset_exit(&md->io_bs);
2007 } else if (bioset_initialized(&md->bs)) {
2008 /*
2009 * There's no need to reload with request-based dm
2010 * because the size of front_pad doesn't change.
2011 * Note for future: If you are to reload bioset,
2012 * prep-ed requests in the queue may refer
2013 * to bio from the old bioset, so you must walk
2014 * through the queue to unprep.
2015 */
2016 goto out;
2017 }
2019 BUG_ON(!p ||
2020 bioset_initialized(&md->bs) ||
2021 bioset_initialized(&md->io_bs));
2023 ret = bioset_init_from_src(&md->bs, &p->bs);
2024 if (ret)
2025 goto out;
2026 ret = bioset_init_from_src(&md->io_bs, &p->io_bs);
2027 if (ret)
2028 bioset_exit(&md->bs);
2029 out:
2030 /* mempool bind completed, no longer need any mempools in the table */
2031 dm_table_free_md_mempools(t);
2032 return ret;
2033 }
2035 /*
2036 * Bind a table to the device.
2037 */
2038 static void event_callback(void *context)
2039 {
2040 unsigned long flags;
2041 LIST_HEAD(uevents);
2042 struct mapped_device *md = (struct mapped_device *) context;
2044 spin_lock_irqsave(&md->uevent_lock, flags);
2045 list_splice_init(&md->uevent_list, &uevents);
2046 spin_unlock_irqrestore(&md->uevent_lock, flags);
2048 dm_send_uevents(&uevents, &disk_to_dev(md->disk)->kobj);
2050 atomic_inc(&md->event_nr);
2051 wake_up(&md->eventq);
2052 dm_issue_global_event();
2053 }
2055 /*
2056 * Protected by md->suspend_lock obtained by dm_swap_table().
2057 */
2058 static void __set_size(struct mapped_device *md, sector_t size)
2059 {
2060 lockdep_assert_held(&md->suspend_lock);
2062 set_capacity(md->disk, size);
2064 i_size_write(md->bdev->bd_inode, (loff_t)size << SECTOR_SHIFT);
2065 }
2067 /*
2068 * Returns old map, which caller must destroy.
2069 */
2070 static struct dm_table *__bind(struct mapped_device *md, struct dm_table *t,
2071 struct queue_limits *limits)
2072 {
2073 struct dm_table *old_map;
2074 struct request_queue *q = md->queue;
2075 bool request_based = dm_table_request_based(t);
2076 sector_t size;
2077 int ret;
2079 lockdep_assert_held(&md->suspend_lock);
2081 size = dm_table_get_size(t);
2083 /*
2084 * Wipe any geometry if the size of the table changed.
2085 */
2086 if (size != dm_get_size(md))
2087 memset(&md->geometry, 0, sizeof(md->geometry));
2089 __set_size(md, size);
2091 dm_table_event_callback(t, event_callback, md);
2093 /*
2094 * The queue hasn't been stopped yet, if the old table type wasn't
2095 * for request-based during suspension. So stop it to prevent
2096 * I/O mapping before resume.
2097 * This must be done before setting the queue restrictions,
2098 * because request-based dm may be run just after the setting.
2099 */
2100 if (request_based)
2101 dm_stop_queue(q);
2103 if (request_based || md->type == DM_TYPE_NVME_BIO_BASED) {
2104 /*
2105 * Leverage the fact that request-based DM targets and
2106 * NVMe bio based targets are immutable singletons
2107 * - used to optimize both dm_request_fn and dm_mq_queue_rq;
2108 * and __process_bio.
2109 */
2110 md->immutable_target = dm_table_get_immutable_target(t);
2111 }
2113 ret = __bind_mempools(md, t);
2114 if (ret) {
2115 old_map = ERR_PTR(ret);
2116 goto out;
2117 }
2119 old_map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2120 rcu_assign_pointer(md->map, (void *)t);
2121 md->immutable_target_type = dm_table_get_immutable_target_type(t);
2123 dm_table_set_restrictions(t, q, limits);
2124 if (old_map)
2125 dm_sync_table(md);
2127 out:
2128 return old_map;
2129 }
2131 /*
2132 * Returns unbound table for the caller to free.
2133 */
2134 static struct dm_table *__unbind(struct mapped_device *md)
2135 {
2136 struct dm_table *map = rcu_dereference_protected(md->map, 1);
2138 if (!map)
2139 return NULL;
2141 dm_table_event_callback(map, NULL, NULL);
2142 RCU_INIT_POINTER(md->map, NULL);
2143 dm_sync_table(md);
2145 return map;
2146 }
2148 /*
2149 * Constructor for a new device.
2150 */
2151 int dm_create(int minor, struct mapped_device **result)
2152 {
2153 int r;
2154 struct mapped_device *md;
2156 md = alloc_dev(minor);
2157 if (!md)
2158 return -ENXIO;
2160 r = dm_sysfs_init(md);
2161 if (r) {
2162 free_dev(md);
2163 return r;
2164 }
2166 *result = md;
2167 return 0;
2168 }
2170 /*
2171 * Functions to manage md->type.
2172 * All are required to hold md->type_lock.
2173 */
2174 void dm_lock_md_type(struct mapped_device *md)
2175 {
2176 mutex_lock(&md->type_lock);
2177 }
2179 void dm_unlock_md_type(struct mapped_device *md)
2180 {
2181 mutex_unlock(&md->type_lock);
2182 }
2184 void dm_set_md_type(struct mapped_device *md, enum dm_queue_mode type)
2185 {
2186 BUG_ON(!mutex_is_locked(&md->type_lock));
2187 md->type = type;
2188 }
2190 enum dm_queue_mode dm_get_md_type(struct mapped_device *md)
2191 {
2192 return md->type;
2193 }
2195 struct target_type *dm_get_immutable_target_type(struct mapped_device *md)
2196 {
2197 return md->immutable_target_type;
2198 }
2200 /*
2201 * The queue_limits are only valid as long as you have a reference
2202 * count on 'md'.
2203 */
2204 struct queue_limits *dm_get_queue_limits(struct mapped_device *md)
2205 {
2206 BUG_ON(!atomic_read(&md->holders));
2207 return &md->queue->limits;
2208 }
2209 EXPORT_SYMBOL_GPL(dm_get_queue_limits);
2211 /*
2212 * Setup the DM device's queue based on md's type
2213 */
2214 int dm_setup_md_queue(struct mapped_device *md, struct dm_table *t)
2215 {
2216 int r;
2217 struct queue_limits limits;
2218 enum dm_queue_mode type = dm_get_md_type(md);
2220 switch (type) {
2221 case DM_TYPE_REQUEST_BASED:
2222 dm_init_normal_md_queue(md);
2223 r = dm_old_init_request_queue(md, t);
2224 if (r) {
2225 DMERR("Cannot initialize queue for request-based mapped device");
2226 return r;
2227 }
2228 break;
2229 case DM_TYPE_MQ_REQUEST_BASED:
2230 r = dm_mq_init_request_queue(md, t);
2231 if (r) {
2232 DMERR("Cannot initialize queue for request-based dm-mq mapped device");
2233 return r;
2234 }
2235 break;
2236 case DM_TYPE_BIO_BASED:
2237 case DM_TYPE_DAX_BIO_BASED:
2238 dm_init_normal_md_queue(md);
2239 blk_queue_make_request(md->queue, dm_make_request);
2240 break;
2241 case DM_TYPE_NVME_BIO_BASED:
2242 dm_init_normal_md_queue(md);
2243 blk_queue_make_request(md->queue, dm_make_request_nvme);
2244 break;
2245 case DM_TYPE_NONE:
2246 WARN_ON_ONCE(true);
2247 break;
2248 }
2250 r = dm_calculate_queue_limits(t, &limits);
2251 if (r) {
2252 DMERR("Cannot calculate initial queue limits");
2253 return r;
2254 }
2255 dm_table_set_restrictions(t, md->queue, &limits);
2256 blk_register_queue(md->disk);
2258 return 0;
2259 }
2261 struct mapped_device *dm_get_md(dev_t dev)
2262 {
2263 struct mapped_device *md;
2264 unsigned minor = MINOR(dev);
2266 if (MAJOR(dev) != _major || minor >= (1 << MINORBITS))
2267 return NULL;
2269 spin_lock(&_minor_lock);
2271 md = idr_find(&_minor_idr, minor);
2272 if (!md || md == MINOR_ALLOCED || (MINOR(disk_devt(dm_disk(md))) != minor) ||
2273 test_bit(DMF_FREEING, &md->flags) || dm_deleting_md(md)) {
2274 md = NULL;
2275 goto out;
2276 }
2277 dm_get(md);
2278 out:
2279 spin_unlock(&_minor_lock);
2281 return md;
2282 }
2283 EXPORT_SYMBOL_GPL(dm_get_md);
2285 void *dm_get_mdptr(struct mapped_device *md)
2286 {
2287 return md->interface_ptr;
2288 }
2290 void dm_set_mdptr(struct mapped_device *md, void *ptr)
2291 {
2292 md->interface_ptr = ptr;
2293 }
2295 void dm_get(struct mapped_device *md)
2296 {
2297 atomic_inc(&md->holders);
2298 BUG_ON(test_bit(DMF_FREEING, &md->flags));
2299 }
2301 int dm_hold(struct mapped_device *md)
2302 {
2303 spin_lock(&_minor_lock);
2304 if (test_bit(DMF_FREEING, &md->flags)) {
2305 spin_unlock(&_minor_lock);
2306 return -EBUSY;
2307 }
2308 dm_get(md);
2309 spin_unlock(&_minor_lock);
2310 return 0;
2311 }
2312 EXPORT_SYMBOL_GPL(dm_hold);
2314 const char *dm_device_name(struct mapped_device *md)
2315 {
2316 return md->name;
2317 }
2318 EXPORT_SYMBOL_GPL(dm_device_name);
2320 static void __dm_destroy(struct mapped_device *md, bool wait)
2321 {
2322 struct dm_table *map;
2323 int srcu_idx;
2325 might_sleep();
2327 spin_lock(&_minor_lock);
2328 idr_replace(&_minor_idr, MINOR_ALLOCED, MINOR(disk_devt(dm_disk(md))));
2329 set_bit(DMF_FREEING, &md->flags);
2330 spin_unlock(&_minor_lock);
2332 blk_set_queue_dying(md->queue);
2334 if (dm_request_based(md) && md->kworker_task)
2335 kthread_flush_worker(&md->kworker);
2337 /*
2338 * Take suspend_lock so that presuspend and postsuspend methods
2339 * do not race with internal suspend.
2340 */
2341 mutex_lock(&md->suspend_lock);
2342 map = dm_get_live_table(md, &srcu_idx);
2343 if (!dm_suspended_md(md)) {
2344 dm_table_presuspend_targets(map);
2345 dm_table_postsuspend_targets(map);
2346 }
2347 /* dm_put_live_table must be before msleep, otherwise deadlock is possible */
2348 dm_put_live_table(md, srcu_idx);
2349 mutex_unlock(&md->suspend_lock);
2351 /*
2352 * Rare, but there may be I/O requests still going to complete,
2353 * for example. Wait for all references to disappear.
2354 * No one should increment the reference count of the mapped_device,
2355 * after the mapped_device state becomes DMF_FREEING.
2356 */
2357 if (wait)
2358 while (atomic_read(&md->holders))
2359 msleep(1);
2360 else if (atomic_read(&md->holders))
2361 DMWARN("%s: Forcibly removing mapped_device still in use! (%d users)",
2362 dm_device_name(md), atomic_read(&md->holders));
2364 dm_sysfs_exit(md);
2365 dm_table_destroy(__unbind(md));
2366 free_dev(md);
2367 }
2369 void dm_destroy(struct mapped_device *md)
2370 {
2371 __dm_destroy(md, true);
2372 }
2374 void dm_destroy_immediate(struct mapped_device *md)
2375 {
2376 __dm_destroy(md, false);
2377 }
2379 void dm_put(struct mapped_device *md)
2380 {
2381 atomic_dec(&md->holders);
2382 }
2383 EXPORT_SYMBOL_GPL(dm_put);
2385 static int dm_wait_for_completion(struct mapped_device *md, long task_state)
2386 {
2387 int r = 0;
2388 DEFINE_WAIT(wait);
2390 while (1) {
2391 prepare_to_wait(&md->wait, &wait, task_state);
2393 if (!md_in_flight(md))
2394 break;
2396 if (signal_pending_state(task_state, current)) {
2397 r = -EINTR;
2398 break;
2399 }
2401 io_schedule();
2402 }
2403 finish_wait(&md->wait, &wait);
2405 return r;
2406 }
2408 /*
2409 * Process the deferred bios
2410 */
2411 static void dm_wq_work(struct work_struct *work)
2412 {
2413 struct mapped_device *md = container_of(work, struct mapped_device,
2414 work);
2415 struct bio *c;
2416 int srcu_idx;
2417 struct dm_table *map;
2419 map = dm_get_live_table(md, &srcu_idx);
2421 while (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
2422 spin_lock_irq(&md->deferred_lock);
2423 c = bio_list_pop(&md->deferred);
2424 spin_unlock_irq(&md->deferred_lock);
2426 if (!c)
2427 break;
2429 if (dm_request_based(md))
2430 generic_make_request(c);
2431 else
2432 __split_and_process_bio(md, map, c);
2433 }
2435 dm_put_live_table(md, srcu_idx);
2436 }
2438 static void dm_queue_flush(struct mapped_device *md)
2439 {
2440 clear_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2441 smp_mb__after_atomic();
2442 queue_work(md->wq, &md->work);
2443 }
2445 /*
2446 * Swap in a new table, returning the old one for the caller to destroy.
2447 */
2448 struct dm_table *dm_swap_table(struct mapped_device *md, struct dm_table *table)
2449 {
2450 struct dm_table *live_map = NULL, *map = ERR_PTR(-EINVAL);
2451 struct queue_limits limits;
2452 int r;
2454 mutex_lock(&md->suspend_lock);
2456 /* device must be suspended */
2457 if (!dm_suspended_md(md))
2458 goto out;
2460 /*
2461 * If the new table has no data devices, retain the existing limits.
2462 * This helps multipath with queue_if_no_path if all paths disappear,
2463 * then new I/O is queued based on these limits, and then some paths
2464 * reappear.
2465 */
2466 if (dm_table_has_no_data_devices(table)) {
2467 live_map = dm_get_live_table_fast(md);
2468 if (live_map)
2469 limits = md->queue->limits;
2470 dm_put_live_table_fast(md);
2471 }
2473 if (!live_map) {
2474 r = dm_calculate_queue_limits(table, &limits);
2475 if (r) {
2476 map = ERR_PTR(r);
2477 goto out;
2478 }
2479 }
2481 map = __bind(md, table, &limits);
2482 dm_issue_global_event();
2484 out:
2485 mutex_unlock(&md->suspend_lock);
2486 return map;
2487 }
2489 /*
2490 * Functions to lock and unlock any filesystem running on the
2491 * device.
2492 */
2493 static int lock_fs(struct mapped_device *md)
2494 {
2495 int r;
2497 WARN_ON(md->frozen_sb);
2499 md->frozen_sb = freeze_bdev(md->bdev);
2500 if (IS_ERR(md->frozen_sb)) {
2501 r = PTR_ERR(md->frozen_sb);
2502 md->frozen_sb = NULL;
2503 return r;
2504 }
2506 set_bit(DMF_FROZEN, &md->flags);
2508 return 0;
2509 }
2511 static void unlock_fs(struct mapped_device *md)
2512 {
2513 if (!test_bit(DMF_FROZEN, &md->flags))
2514 return;
2516 thaw_bdev(md->bdev, md->frozen_sb);
2517 md->frozen_sb = NULL;
2518 clear_bit(DMF_FROZEN, &md->flags);
2519 }
2521 /*
2522 * @suspend_flags: DM_SUSPEND_LOCKFS_FLAG and/or DM_SUSPEND_NOFLUSH_FLAG
2523 * @task_state: e.g. TASK_INTERRUPTIBLE or TASK_UNINTERRUPTIBLE
2524 * @dmf_suspended_flag: DMF_SUSPENDED or DMF_SUSPENDED_INTERNALLY
2525 *
2526 * If __dm_suspend returns 0, the device is completely quiescent
2527 * now. There is no request-processing activity. All new requests
2528 * are being added to md->deferred list.
2529 */
2530 static int __dm_suspend(struct mapped_device *md, struct dm_table *map,
2531 unsigned suspend_flags, long task_state,
2532 int dmf_suspended_flag)
2533 {
2534 bool do_lockfs = suspend_flags & DM_SUSPEND_LOCKFS_FLAG;
2535 bool noflush = suspend_flags & DM_SUSPEND_NOFLUSH_FLAG;
2536 int r;
2538 lockdep_assert_held(&md->suspend_lock);
2540 /*
2541 * DMF_NOFLUSH_SUSPENDING must be set before presuspend.
2542 * This flag is cleared before dm_suspend returns.
2543 */
2544 if (noflush)
2545 set_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
2546 else
2547 pr_debug("%s: suspending with flush\n", dm_device_name(md));
2549 /*
2550 * This gets reverted if there's an error later and the targets
2551 * provide the .presuspend_undo hook.
2552 */
2553 dm_table_presuspend_targets(map);
2555 /*
2556 * Flush I/O to the device.
2557 * Any I/O submitted after lock_fs() may not be flushed.
2558 * noflush takes precedence over do_lockfs.
2559 * (lock_fs() flushes I/Os and waits for them to complete.)
2560 */
2561 if (!noflush && do_lockfs) {
2562 r = lock_fs(md);
2563 if (r) {
2564 dm_table_presuspend_undo_targets(map);
2565 return r;
2566 }
2567 }
2569 /*
2570 * Here we must make sure that no processes are submitting requests
2571 * to target drivers i.e. no one may be executing
2572 * __split_and_process_bio. This is called from dm_request and
2573 * dm_wq_work.
2574 *
2575 * To get all processes out of __split_and_process_bio in dm_request,
2576 * we take the write lock. To prevent any process from reentering
2577 * __split_and_process_bio from dm_request and quiesce the thread
2578 * (dm_wq_work), we set BMF_BLOCK_IO_FOR_SUSPEND and call
2579 * flush_workqueue(md->wq).
2580 */
2581 set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2582 if (map)
2583 synchronize_srcu(&md->io_barrier);
2585 /*
2586 * Stop md->queue before flushing md->wq in case request-based
2587 * dm defers requests to md->wq from md->queue.
2588 */
2589 if (dm_request_based(md)) {
2590 dm_stop_queue(md->queue);
2591 if (md->kworker_task)
2592 kthread_flush_worker(&md->kworker);
2593 }
2595 flush_workqueue(md->wq);
2597 /*
2598 * At this point no more requests are entering target request routines.
2599 * We call dm_wait_for_completion to wait for all existing requests
2600 * to finish.
2601 */
2602 r = dm_wait_for_completion(md, task_state);
2603 if (!r)
2604 set_bit(dmf_suspended_flag, &md->flags);
2606 if (noflush)
2607 clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
2608 if (map)
2609 synchronize_srcu(&md->io_barrier);
2611 /* were we interrupted ? */
2612 if (r < 0) {
2613 dm_queue_flush(md);
2615 if (dm_request_based(md))
2616 dm_start_queue(md->queue);
2618 unlock_fs(md);
2619 dm_table_presuspend_undo_targets(map);
2620 /* pushback list is already flushed, so skip flush */
2621 }
2623 return r;
2624 }
2626 /*
2627 * We need to be able to change a mapping table under a mounted
2628 * filesystem. For example we might want to move some data in
2629 * the background. Before the table can be swapped with
2630 * dm_bind_table, dm_suspend must be called to flush any in
2631 * flight bios and ensure that any further io gets deferred.
2632 */
2633 /*
2634 * Suspend mechanism in request-based dm.
2635 *
2636 * 1. Flush all I/Os by lock_fs() if needed.
2637 * 2. Stop dispatching any I/O by stopping the request_queue.
2638 * 3. Wait for all in-flight I/Os to be completed or requeued.
2639 *
2640 * To abort suspend, start the request_queue.
2641 */
2642 int dm_suspend(struct mapped_device *md, unsigned suspend_flags)
2643 {
2644 struct dm_table *map = NULL;
2645 int r = 0;
2647 retry:
2648 mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING);
2650 if (dm_suspended_md(md)) {
2651 r = -EINVAL;
2652 goto out_unlock;
2653 }
2655 if (dm_suspended_internally_md(md)) {
2656 /* already internally suspended, wait for internal resume */
2657 mutex_unlock(&md->suspend_lock);
2658 r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE);
2659 if (r)
2660 return r;
2661 goto retry;
2662 }
2664 map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2666 r = __dm_suspend(md, map, suspend_flags, TASK_INTERRUPTIBLE, DMF_SUSPENDED);
2667 if (r)
2668 goto out_unlock;
2670 dm_table_postsuspend_targets(map);
2672 out_unlock:
2673 mutex_unlock(&md->suspend_lock);
2674 return r;
2675 }
2677 static int __dm_resume(struct mapped_device *md, struct dm_table *map)
2678 {
2679 if (map) {
2680 int r = dm_table_resume_targets(map);
2681 if (r)
2682 return r;
2683 }
2685 dm_queue_flush(md);
2687 /*
2688 * Flushing deferred I/Os must be done after targets are resumed
2689 * so that mapping of targets can work correctly.
2690 * Request-based dm is queueing the deferred I/Os in its request_queue.
2691 */
2692 if (dm_request_based(md))
2693 dm_start_queue(md->queue);
2695 unlock_fs(md);
2697 return 0;
2698 }
2700 int dm_resume(struct mapped_device *md)
2701 {
2702 int r;
2703 struct dm_table *map = NULL;
2705 retry:
2706 r = -EINVAL;
2707 mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING);
2709 if (!dm_suspended_md(md))
2710 goto out;
2712 if (dm_suspended_internally_md(md)) {
2713 /* already internally suspended, wait for internal resume */
2714 mutex_unlock(&md->suspend_lock);
2715 r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE);
2716 if (r)
2717 return r;
2718 goto retry;
2719 }
2721 map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2722 if (!map || !dm_table_get_size(map))
2723 goto out;
2725 r = __dm_resume(md, map);
2726 if (r)
2727 goto out;
2729 clear_bit(DMF_SUSPENDED, &md->flags);
2730 out:
2731 mutex_unlock(&md->suspend_lock);
2733 return r;
2734 }
2736 /*
2737 * Internal suspend/resume works like userspace-driven suspend. It waits
2738 * until all bios finish and prevents issuing new bios to the target drivers.
2739 * It may be used only from the kernel.
2740 */
2742 static void __dm_internal_suspend(struct mapped_device *md, unsigned suspend_flags)
2743 {
2744 struct dm_table *map = NULL;
2746 lockdep_assert_held(&md->suspend_lock);
2748 if (md->internal_suspend_count++)
2749 return; /* nested internal suspend */
2751 if (dm_suspended_md(md)) {
2752 set_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
2753 return; /* nest suspend */
2754 }
2756 map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2758 /*
2759 * Using TASK_UNINTERRUPTIBLE because only NOFLUSH internal suspend is
2760 * supported. Properly supporting a TASK_INTERRUPTIBLE internal suspend
2761 * would require changing .presuspend to return an error -- avoid this
2762 * until there is a need for more elaborate variants of internal suspend.
2763 */
2764 (void) __dm_suspend(md, map, suspend_flags, TASK_UNINTERRUPTIBLE,
2765 DMF_SUSPENDED_INTERNALLY);
2767 dm_table_postsuspend_targets(map);
2768 }
2770 static void __dm_internal_resume(struct mapped_device *md)
2771 {
2772 BUG_ON(!md->internal_suspend_count);
2774 if (--md->internal_suspend_count)
2775 return; /* resume from nested internal suspend */
2777 if (dm_suspended_md(md))
2778 goto done; /* resume from nested suspend */
2780 /*
2781 * NOTE: existing callers don't need to call dm_table_resume_targets
2782 * (which may fail -- so best to avoid it for now by passing NULL map)
2783 */
2784 (void) __dm_resume(md, NULL);
2786 done:
2787 clear_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
2788 smp_mb__after_atomic();
2789 wake_up_bit(&md->flags, DMF_SUSPENDED_INTERNALLY);
2790 }
2792 void dm_internal_suspend_noflush(struct mapped_device *md)
2793 {
2794 mutex_lock(&md->suspend_lock);
2795 __dm_internal_suspend(md, DM_SUSPEND_NOFLUSH_FLAG);
2796 mutex_unlock(&md->suspend_lock);
2797 }
2798 EXPORT_SYMBOL_GPL(dm_internal_suspend_noflush);
2800 void dm_internal_resume(struct mapped_device *md)
2801 {
2802 mutex_lock(&md->suspend_lock);
2803 __dm_internal_resume(md);
2804 mutex_unlock(&md->suspend_lock);
2805 }
2806 EXPORT_SYMBOL_GPL(dm_internal_resume);
2808 /*
2809 * Fast variants of internal suspend/resume hold md->suspend_lock,
2810 * which prevents interaction with userspace-driven suspend.
2811 */
2813 void dm_internal_suspend_fast(struct mapped_device *md)
2814 {
2815 mutex_lock(&md->suspend_lock);
2816 if (dm_suspended_md(md) || dm_suspended_internally_md(md))
2817 return;
2819 set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2820 synchronize_srcu(&md->io_barrier);
2821 flush_workqueue(md->wq);
2822 dm_wait_for_completion(md, TASK_UNINTERRUPTIBLE);
2823 }
2824 EXPORT_SYMBOL_GPL(dm_internal_suspend_fast);
2826 void dm_internal_resume_fast(struct mapped_device *md)
2827 {
2828 if (dm_suspended_md(md) || dm_suspended_internally_md(md))
2829 goto done;
2831 dm_queue_flush(md);
2833 done:
2834 mutex_unlock(&md->suspend_lock);
2835 }
2836 EXPORT_SYMBOL_GPL(dm_internal_resume_fast);
2838 /*-----------------------------------------------------------------
2839 * Event notification.
2840 *---------------------------------------------------------------*/
2841 int dm_kobject_uevent(struct mapped_device *md, enum kobject_action action,
2842 unsigned cookie)
2843 {
2844 char udev_cookie[DM_COOKIE_LENGTH];
2845 char *envp[] = { udev_cookie, NULL };
2847 if (!cookie)
2848 return kobject_uevent(&disk_to_dev(md->disk)->kobj, action);
2849 else {
2850 snprintf(udev_cookie, DM_COOKIE_LENGTH, "%s=%u",
2851 DM_COOKIE_ENV_VAR_NAME, cookie);
2852 return kobject_uevent_env(&disk_to_dev(md->disk)->kobj,
2853 action, envp);
2854 }
2855 }
2857 uint32_t dm_next_uevent_seq(struct mapped_device *md)
2858 {
2859 return atomic_add_return(1, &md->uevent_seq);
2860 }
2862 uint32_t dm_get_event_nr(struct mapped_device *md)
2863 {
2864 return atomic_read(&md->event_nr);
2865 }
2867 int dm_wait_event(struct mapped_device *md, int event_nr)
2868 {
2869 return wait_event_interruptible(md->eventq,
2870 (event_nr != atomic_read(&md->event_nr)));
2871 }
2873 void dm_uevent_add(struct mapped_device *md, struct list_head *elist)
2874 {
2875 unsigned long flags;
2877 spin_lock_irqsave(&md->uevent_lock, flags);
2878 list_add(elist, &md->uevent_list);
2879 spin_unlock_irqrestore(&md->uevent_lock, flags);
2880 }
2882 /*
2883 * The gendisk is only valid as long as you have a reference
2884 * count on 'md'.
2885 */
2886 struct gendisk *dm_disk(struct mapped_device *md)
2887 {
2888 return md->disk;
2889 }
2890 EXPORT_SYMBOL_GPL(dm_disk);
2892 struct kobject *dm_kobject(struct mapped_device *md)
2893 {
2894 return &md->kobj_holder.kobj;
2895 }
2897 struct mapped_device *dm_get_from_kobject(struct kobject *kobj)
2898 {
2899 struct mapped_device *md;
2901 md = container_of(kobj, struct mapped_device, kobj_holder.kobj);
2903 spin_lock(&_minor_lock);
2904 if (test_bit(DMF_FREEING, &md->flags) || dm_deleting_md(md)) {
2905 md = NULL;
2906 goto out;
2907 }
2908 dm_get(md);
2909 out:
2910 spin_unlock(&_minor_lock);
2912 return md;
2913 }
2915 int dm_suspended_md(struct mapped_device *md)
2916 {
2917 return test_bit(DMF_SUSPENDED, &md->flags);
2918 }
2920 int dm_suspended_internally_md(struct mapped_device *md)
2921 {
2922 return test_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
2923 }
2925 int dm_test_deferred_remove_flag(struct mapped_device *md)
2926 {
2927 return test_bit(DMF_DEFERRED_REMOVE, &md->flags);
2928 }
2930 int dm_suspended(struct dm_target *ti)
2931 {
2932 return dm_suspended_md(dm_table_get_md(ti->table));
2933 }
2934 EXPORT_SYMBOL_GPL(dm_suspended);
2936 int dm_noflush_suspending(struct dm_target *ti)
2937 {
2938 return __noflush_suspending(dm_table_get_md(ti->table));
2939 }
2940 EXPORT_SYMBOL_GPL(dm_noflush_suspending);
2942 struct dm_md_mempools *dm_alloc_md_mempools(struct mapped_device *md, enum dm_queue_mode type,
2943 unsigned integrity, unsigned per_io_data_size,
2944 unsigned min_pool_size)
2945 {
2946 struct dm_md_mempools *pools = kzalloc_node(sizeof(*pools), GFP_KERNEL, md->numa_node_id);
2947 unsigned int pool_size = 0;
2948 unsigned int front_pad, io_front_pad;
2949 int ret;
2951 if (!pools)
2952 return NULL;
2954 switch (type) {
2955 case DM_TYPE_BIO_BASED:
2956 case DM_TYPE_DAX_BIO_BASED:
2957 case DM_TYPE_NVME_BIO_BASED:
2958 pool_size = max(dm_get_reserved_bio_based_ios(), min_pool_size);
2959 front_pad = roundup(per_io_data_size, __alignof__(struct dm_target_io)) + offsetof(struct dm_target_io, clone);
2960 io_front_pad = roundup(front_pad, __alignof__(struct dm_io)) + offsetof(struct dm_io, tio);
2961 ret = bioset_init(&pools->io_bs, pool_size, io_front_pad, 0);
2962 if (ret)
2963 goto out;
2964 if (integrity && bioset_integrity_create(&pools->io_bs, pool_size))
2965 goto out;
2966 break;
2967 case DM_TYPE_REQUEST_BASED:
2968 case DM_TYPE_MQ_REQUEST_BASED:
2969 pool_size = max(dm_get_reserved_rq_based_ios(), min_pool_size);
2970 front_pad = offsetof(struct dm_rq_clone_bio_info, clone);
2971 /* per_io_data_size is used for blk-mq pdu at queue allocation */
2972 break;
2973 default:
2974 BUG();
2975 }
2977 ret = bioset_init(&pools->bs, pool_size, front_pad, 0);
2978 if (ret)
2979 goto out;
2981 if (integrity && bioset_integrity_create(&pools->bs, pool_size))
2982 goto out;
2984 return pools;
2986 out:
2987 dm_free_md_mempools(pools);
2989 return NULL;
2990 }
2992 void dm_free_md_mempools(struct dm_md_mempools *pools)
2993 {
2994 if (!pools)
2995 return;
2997 bioset_exit(&pools->bs);
2998 bioset_exit(&pools->io_bs);
3000 kfree(pools);
3001 }
3003 struct dm_pr {
3004 u64 old_key;
3005 u64 new_key;
3006 u32 flags;
3007 bool fail_early;
3008 };
3010 static int dm_call_pr(struct block_device *bdev, iterate_devices_callout_fn fn,
3011 void *data)
3012 {
3013 struct mapped_device *md = bdev->bd_disk->private_data;
3014 struct dm_table *table;
3015 struct dm_target *ti;
3016 int ret = -ENOTTY, srcu_idx;
3018 table = dm_get_live_table(md, &srcu_idx);
3019 if (!table || !dm_table_get_size(table))
3020 goto out;
3022 /* We only support devices that have a single target */
3023 if (dm_table_get_num_targets(table) != 1)
3024 goto out;
3025 ti = dm_table_get_target(table, 0);
3027 ret = -EINVAL;
3028 if (!ti->type->iterate_devices)
3029 goto out;
3031 ret = ti->type->iterate_devices(ti, fn, data);
3032 out:
3033 dm_put_live_table(md, srcu_idx);
3034 return ret;
3035 }
3037 /*
3038 * For register / unregister we need to manually call out to every path.
3039 */
3040 static int __dm_pr_register(struct dm_target *ti, struct dm_dev *dev,
3041 sector_t start, sector_t len, void *data)
3042 {
3043 struct dm_pr *pr = data;
3044 const struct pr_ops *ops = dev->bdev->bd_disk->fops->pr_ops;
3046 if (!ops || !ops->pr_register)
3047 return -EOPNOTSUPP;
3048 return ops->pr_register(dev->bdev, pr->old_key, pr->new_key, pr->flags);
3049 }
3051 static int dm_pr_register(struct block_device *bdev, u64 old_key, u64 new_key,
3052 u32 flags)
3053 {
3054 struct dm_pr pr = {
3055 .old_key = old_key,
3056 .new_key = new_key,
3057 .flags = flags,
3058 .fail_early = true,
3059 };
3060 int ret;
3062 ret = dm_call_pr(bdev, __dm_pr_register, &pr);
3063 if (ret && new_key) {
3064 /* unregister all paths if we failed to register any path */
3065 pr.old_key = new_key;
3066 pr.new_key = 0;
3067 pr.flags = 0;
3068 pr.fail_early = false;
3069 dm_call_pr(bdev, __dm_pr_register, &pr);
3070 }
3072 return ret;
3073 }
3075 static int dm_pr_reserve(struct block_device *bdev, u64 key, enum pr_type type,
3076 u32 flags)
3077 {
3078 struct mapped_device *md = bdev->bd_disk->private_data;
3079 const struct pr_ops *ops;
3080 int r, srcu_idx;
3082 r = dm_prepare_ioctl(md, &srcu_idx, &bdev);
3083 if (r < 0)
3084 goto out;
3086 ops = bdev->bd_disk->fops->pr_ops;
3087 if (ops && ops->pr_reserve)
3088 r = ops->pr_reserve(bdev, key, type, flags);
3089 else
3090 r = -EOPNOTSUPP;
3091 out:
3092 dm_unprepare_ioctl(md, srcu_idx);
3093 return r;
3094 }
3096 static int dm_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
3097 {
3098 struct mapped_device *md = bdev->bd_disk->private_data;
3099 const struct pr_ops *ops;
3100 int r, srcu_idx;
3102 r = dm_prepare_ioctl(md, &srcu_idx, &bdev);
3103 if (r < 0)
3104 goto out;
3106 ops = bdev->bd_disk->fops->pr_ops;
3107 if (ops && ops->pr_release)
3108 r = ops->pr_release(bdev, key, type);
3109 else
3110 r = -EOPNOTSUPP;
3111 out:
3112 dm_unprepare_ioctl(md, srcu_idx);
3113 return r;
3114 }
3116 static int dm_pr_preempt(struct block_device *bdev, u64 old_key, u64 new_key,
3117 enum pr_type type, bool abort)
3118 {
3119 struct mapped_device *md = bdev->bd_disk->private_data;
3120 const struct pr_ops *ops;
3121 int r, srcu_idx;
3123 r = dm_prepare_ioctl(md, &srcu_idx, &bdev);
3124 if (r < 0)
3125 goto out;
3127 ops = bdev->bd_disk->fops->pr_ops;
3128 if (ops && ops->pr_preempt)
3129 r = ops->pr_preempt(bdev, old_key, new_key, type, abort);
3130 else
3131 r = -EOPNOTSUPP;
3132 out:
3133 dm_unprepare_ioctl(md, srcu_idx);
3134 return r;
3135 }
3137 static int dm_pr_clear(struct block_device *bdev, u64 key)
3138 {
3139 struct mapped_device *md = bdev->bd_disk->private_data;
3140 const struct pr_ops *ops;
3141 int r, srcu_idx;
3143 r = dm_prepare_ioctl(md, &srcu_idx, &bdev);
3144 if (r < 0)
3145 goto out;
3147 ops = bdev->bd_disk->fops->pr_ops;
3148 if (ops && ops->pr_clear)
3149 r = ops->pr_clear(bdev, key);
3150 else
3151 r = -EOPNOTSUPP;
3152 out:
3153 dm_unprepare_ioctl(md, srcu_idx);
3154 return r;
3155 }
3157 static const struct pr_ops dm_pr_ops = {
3158 .pr_register = dm_pr_register,
3159 .pr_reserve = dm_pr_reserve,
3160 .pr_release = dm_pr_release,
3161 .pr_preempt = dm_pr_preempt,
3162 .pr_clear = dm_pr_clear,
3163 };
3165 static const struct block_device_operations dm_blk_dops = {
3166 .open = dm_blk_open,
3167 .release = dm_blk_close,
3168 .ioctl = dm_blk_ioctl,
3169 .getgeo = dm_blk_getgeo,
3170 .pr_ops = &dm_pr_ops,
3171 .owner = THIS_MODULE
3172 };
3174 static const struct dax_operations dm_dax_ops = {
3175 .direct_access = dm_dax_direct_access,
3176 .copy_from_iter = dm_dax_copy_from_iter,
3177 .copy_to_iter = dm_dax_copy_to_iter,
3178 };
3180 /*
3181 * module hooks
3182 */
3183 module_init(dm_init);
3184 module_exit(dm_exit);
3186 module_param(major, uint, 0);
3187 MODULE_PARM_DESC(major, "The major number of the device mapper");
3189 module_param(reserved_bio_based_ios, uint, S_IRUGO | S_IWUSR);
3190 MODULE_PARM_DESC(reserved_bio_based_ios, "Reserved IOs in bio-based mempools");
3192 module_param(dm_numa_node, int, S_IRUGO | S_IWUSR);
3193 MODULE_PARM_DESC(dm_numa_node, "NUMA node for DM device memory allocations");
3195 MODULE_DESCRIPTION(DM_NAME " driver");
3196 MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
3197 MODULE_LICENSE("GPL");