2 * Copyright (C) 2001, 2002 Sistina Software (UK) Limited.
3 * Copyright (C) 2004-2008 Red Hat, Inc. All rights reserved.
5 * This file is released under the GPL.
10 #include "dm-uevent.h"
12 #include <linux/init.h>
13 #include <linux/module.h>
14 #include <linux/mutex.h>
15 #include <linux/blkpg.h>
16 #include <linux/bio.h>
17 #include <linux/mempool.h>
18 #include <linux/slab.h>
19 #include <linux/idr.h>
20 #include <linux/hdreg.h>
21 #include <linux/delay.h>
22 #include <linux/wait.h>
25 #define DM_MSG_PREFIX "core"
29 * ratelimit state to be used in DMXXX_LIMIT().
31 DEFINE_RATELIMIT_STATE(dm_ratelimit_state,
32 DEFAULT_RATELIMIT_INTERVAL,
33 DEFAULT_RATELIMIT_BURST);
34 EXPORT_SYMBOL(dm_ratelimit_state);
38 * Cookies are numeric values sent with CHANGE and REMOVE
39 * uevents while resuming, removing or renaming the device.
41 #define DM_COOKIE_ENV_VAR_NAME "DM_COOKIE"
42 #define DM_COOKIE_LENGTH 24
44 static const char *_name = DM_NAME;
46 static unsigned int major = 0;
47 static unsigned int _major = 0;
49 static DEFINE_IDR(_minor_idr);
51 static DEFINE_SPINLOCK(_minor_lock);
53 static void do_deferred_remove(struct work_struct *w);
55 static DECLARE_WORK(deferred_remove_work, do_deferred_remove);
57 static struct workqueue_struct *deferred_remove_workqueue;
60 * One of these is allocated per bio.
63 struct mapped_device *md;
67 unsigned long start_time;
68 spinlock_t endio_lock;
69 struct dm_stats_aux stats_aux;
72 #define MINOR_ALLOCED ((void *)-1)
75 * Bits for the md->flags field.
77 #define DMF_BLOCK_IO_FOR_SUSPEND 0
78 #define DMF_SUSPENDED 1
81 #define DMF_DELETING 4
82 #define DMF_NOFLUSH_SUSPENDING 5
83 #define DMF_DEFERRED_REMOVE 6
84 #define DMF_SUSPENDED_INTERNALLY 7
86 #define DM_NUMA_NODE NUMA_NO_NODE
87 static int dm_numa_node = DM_NUMA_NODE;
90 * For mempools pre-allocation at the table loading time.
92 struct dm_md_mempools {
99 struct list_head list;
101 struct dm_dev dm_dev;
104 static struct kmem_cache *_io_cache;
105 static struct kmem_cache *_rq_tio_cache;
106 static struct kmem_cache *_rq_cache;
109 * Bio-based DM's mempools' reserved IOs set by the user.
111 #define RESERVED_BIO_BASED_IOS 16
112 static unsigned reserved_bio_based_ios = RESERVED_BIO_BASED_IOS;
114 static int __dm_get_module_param_int(int *module_param, int min, int max)
116 int param = ACCESS_ONCE(*module_param);
117 int modified_param = 0;
118 bool modified = true;
121 modified_param = min;
122 else if (param > max)
123 modified_param = max;
128 (void)cmpxchg(module_param, param, modified_param);
129 param = modified_param;
135 unsigned __dm_get_module_param(unsigned *module_param,
136 unsigned def, unsigned max)
138 unsigned param = ACCESS_ONCE(*module_param);
139 unsigned modified_param = 0;
142 modified_param = def;
143 else if (param > max)
144 modified_param = max;
146 if (modified_param) {
147 (void)cmpxchg(module_param, param, modified_param);
148 param = modified_param;
154 unsigned dm_get_reserved_bio_based_ios(void)
156 return __dm_get_module_param(&reserved_bio_based_ios,
157 RESERVED_BIO_BASED_IOS, DM_RESERVED_MAX_IOS);
159 EXPORT_SYMBOL_GPL(dm_get_reserved_bio_based_ios);
161 static unsigned dm_get_numa_node(void)
163 return __dm_get_module_param_int(&dm_numa_node,
164 DM_NUMA_NODE, num_online_nodes() - 1);
167 static int __init local_init(void)
171 /* allocate a slab for the dm_ios */
172 _io_cache = KMEM_CACHE(dm_io, 0);
176 _rq_tio_cache = KMEM_CACHE(dm_rq_target_io, 0);
178 goto out_free_io_cache;
180 _rq_cache = kmem_cache_create("dm_old_clone_request", sizeof(struct request),
181 __alignof__(struct request), 0, NULL);
183 goto out_free_rq_tio_cache;
185 r = dm_uevent_init();
187 goto out_free_rq_cache;
189 deferred_remove_workqueue = alloc_workqueue("kdmremove", WQ_UNBOUND, 1);
190 if (!deferred_remove_workqueue) {
192 goto out_uevent_exit;
196 r = register_blkdev(_major, _name);
198 goto out_free_workqueue;
206 destroy_workqueue(deferred_remove_workqueue);
210 kmem_cache_destroy(_rq_cache);
211 out_free_rq_tio_cache:
212 kmem_cache_destroy(_rq_tio_cache);
214 kmem_cache_destroy(_io_cache);
219 static void local_exit(void)
221 flush_scheduled_work();
222 destroy_workqueue(deferred_remove_workqueue);
224 kmem_cache_destroy(_rq_cache);
225 kmem_cache_destroy(_rq_tio_cache);
226 kmem_cache_destroy(_io_cache);
227 unregister_blkdev(_major, _name);
232 DMINFO("cleaned up");
235 static int (*_inits[])(void) __initdata = {
246 static void (*_exits[])(void) = {
257 static int __init dm_init(void)
259 const int count = ARRAY_SIZE(_inits);
263 for (i = 0; i < count; i++) {
278 static void __exit dm_exit(void)
280 int i = ARRAY_SIZE(_exits);
286 * Should be empty by this point.
288 idr_destroy(&_minor_idr);
292 * Block device functions
294 int dm_deleting_md(struct mapped_device *md)
296 return test_bit(DMF_DELETING, &md->flags);
299 static int dm_blk_open(struct block_device *bdev, fmode_t mode)
301 struct mapped_device *md;
303 spin_lock(&_minor_lock);
305 md = bdev->bd_disk->private_data;
309 if (test_bit(DMF_FREEING, &md->flags) ||
310 dm_deleting_md(md)) {
316 atomic_inc(&md->open_count);
318 spin_unlock(&_minor_lock);
320 return md ? 0 : -ENXIO;
323 static void dm_blk_close(struct gendisk *disk, fmode_t mode)
325 struct mapped_device *md;
327 spin_lock(&_minor_lock);
329 md = disk->private_data;
333 if (atomic_dec_and_test(&md->open_count) &&
334 (test_bit(DMF_DEFERRED_REMOVE, &md->flags)))
335 queue_work(deferred_remove_workqueue, &deferred_remove_work);
339 spin_unlock(&_minor_lock);
342 int dm_open_count(struct mapped_device *md)
344 return atomic_read(&md->open_count);
348 * Guarantees nothing is using the device before it's deleted.
350 int dm_lock_for_deletion(struct mapped_device *md, bool mark_deferred, bool only_deferred)
354 spin_lock(&_minor_lock);
356 if (dm_open_count(md)) {
359 set_bit(DMF_DEFERRED_REMOVE, &md->flags);
360 } else if (only_deferred && !test_bit(DMF_DEFERRED_REMOVE, &md->flags))
363 set_bit(DMF_DELETING, &md->flags);
365 spin_unlock(&_minor_lock);
370 int dm_cancel_deferred_remove(struct mapped_device *md)
374 spin_lock(&_minor_lock);
376 if (test_bit(DMF_DELETING, &md->flags))
379 clear_bit(DMF_DEFERRED_REMOVE, &md->flags);
381 spin_unlock(&_minor_lock);
386 static void do_deferred_remove(struct work_struct *w)
388 dm_deferred_remove();
391 sector_t dm_get_size(struct mapped_device *md)
393 return get_capacity(md->disk);
396 struct request_queue *dm_get_md_queue(struct mapped_device *md)
401 struct dm_stats *dm_get_stats(struct mapped_device *md)
406 static int dm_blk_getgeo(struct block_device *bdev, struct hd_geometry *geo)
408 struct mapped_device *md = bdev->bd_disk->private_data;
410 return dm_get_geometry(md, geo);
413 static int dm_grab_bdev_for_ioctl(struct mapped_device *md,
414 struct block_device **bdev,
417 struct dm_target *tgt;
418 struct dm_table *map;
423 map = dm_get_live_table(md, &srcu_idx);
424 if (!map || !dm_table_get_size(map))
427 /* We only support devices that have a single target */
428 if (dm_table_get_num_targets(map) != 1)
431 tgt = dm_table_get_target(map, 0);
432 if (!tgt->type->prepare_ioctl)
435 if (dm_suspended_md(md)) {
440 r = tgt->type->prepare_ioctl(tgt, bdev, mode);
445 dm_put_live_table(md, srcu_idx);
449 dm_put_live_table(md, srcu_idx);
450 if (r == -ENOTCONN && !fatal_signal_pending(current)) {
457 static int dm_blk_ioctl(struct block_device *bdev, fmode_t mode,
458 unsigned int cmd, unsigned long arg)
460 struct mapped_device *md = bdev->bd_disk->private_data;
463 r = dm_grab_bdev_for_ioctl(md, &bdev, &mode);
469 * Target determined this ioctl is being issued against
470 * a logical partition of the parent bdev; so extra
471 * validation is needed.
473 r = scsi_verify_blk_ioctl(NULL, cmd);
478 r = __blkdev_driver_ioctl(bdev, mode, cmd, arg);
484 static struct dm_io *alloc_io(struct mapped_device *md)
486 return mempool_alloc(md->io_pool, GFP_NOIO);
489 static void free_io(struct mapped_device *md, struct dm_io *io)
491 mempool_free(io, md->io_pool);
494 static void free_tio(struct dm_target_io *tio)
496 bio_put(&tio->clone);
499 int md_in_flight(struct mapped_device *md)
501 return atomic_read(&md->pending[READ]) +
502 atomic_read(&md->pending[WRITE]);
505 static void start_io_acct(struct dm_io *io)
507 struct mapped_device *md = io->md;
508 struct bio *bio = io->bio;
510 int rw = bio_data_dir(bio);
512 io->start_time = jiffies;
514 cpu = part_stat_lock();
515 part_round_stats(cpu, &dm_disk(md)->part0);
517 atomic_set(&dm_disk(md)->part0.in_flight[rw],
518 atomic_inc_return(&md->pending[rw]));
520 if (unlikely(dm_stats_used(&md->stats)))
521 dm_stats_account_io(&md->stats, bio_data_dir(bio),
522 bio->bi_iter.bi_sector, bio_sectors(bio),
523 false, 0, &io->stats_aux);
526 static void end_io_acct(struct dm_io *io)
528 struct mapped_device *md = io->md;
529 struct bio *bio = io->bio;
530 unsigned long duration = jiffies - io->start_time;
532 int rw = bio_data_dir(bio);
534 generic_end_io_acct(rw, &dm_disk(md)->part0, io->start_time);
536 if (unlikely(dm_stats_used(&md->stats)))
537 dm_stats_account_io(&md->stats, bio_data_dir(bio),
538 bio->bi_iter.bi_sector, bio_sectors(bio),
539 true, duration, &io->stats_aux);
542 * After this is decremented the bio must not be touched if it is
545 pending = atomic_dec_return(&md->pending[rw]);
546 atomic_set(&dm_disk(md)->part0.in_flight[rw], pending);
547 pending += atomic_read(&md->pending[rw^0x1]);
549 /* nudge anyone waiting on suspend queue */
555 * Add the bio to the list of deferred io.
557 static void queue_io(struct mapped_device *md, struct bio *bio)
561 spin_lock_irqsave(&md->deferred_lock, flags);
562 bio_list_add(&md->deferred, bio);
563 spin_unlock_irqrestore(&md->deferred_lock, flags);
564 queue_work(md->wq, &md->work);
568 * Everyone (including functions in this file), should use this
569 * function to access the md->map field, and make sure they call
570 * dm_put_live_table() when finished.
572 struct dm_table *dm_get_live_table(struct mapped_device *md, int *srcu_idx) __acquires(md->io_barrier)
574 *srcu_idx = srcu_read_lock(&md->io_barrier);
576 return srcu_dereference(md->map, &md->io_barrier);
579 void dm_put_live_table(struct mapped_device *md, int srcu_idx) __releases(md->io_barrier)
581 srcu_read_unlock(&md->io_barrier, srcu_idx);
584 void dm_sync_table(struct mapped_device *md)
586 synchronize_srcu(&md->io_barrier);
587 synchronize_rcu_expedited();
591 * A fast alternative to dm_get_live_table/dm_put_live_table.
592 * The caller must not block between these two functions.
594 static struct dm_table *dm_get_live_table_fast(struct mapped_device *md) __acquires(RCU)
597 return rcu_dereference(md->map);
600 static void dm_put_live_table_fast(struct mapped_device *md) __releases(RCU)
606 * Open a table device so we can use it as a map destination.
608 static int open_table_device(struct table_device *td, dev_t dev,
609 struct mapped_device *md)
611 static char *_claim_ptr = "I belong to device-mapper";
612 struct block_device *bdev;
616 BUG_ON(td->dm_dev.bdev);
618 bdev = blkdev_get_by_dev(dev, td->dm_dev.mode | FMODE_EXCL, _claim_ptr);
620 return PTR_ERR(bdev);
622 r = bd_link_disk_holder(bdev, dm_disk(md));
624 blkdev_put(bdev, td->dm_dev.mode | FMODE_EXCL);
628 td->dm_dev.bdev = bdev;
633 * Close a table device that we've been using.
635 static void close_table_device(struct table_device *td, struct mapped_device *md)
637 if (!td->dm_dev.bdev)
640 bd_unlink_disk_holder(td->dm_dev.bdev, dm_disk(md));
641 blkdev_put(td->dm_dev.bdev, td->dm_dev.mode | FMODE_EXCL);
642 td->dm_dev.bdev = NULL;
645 static struct table_device *find_table_device(struct list_head *l, dev_t dev,
647 struct table_device *td;
649 list_for_each_entry(td, l, list)
650 if (td->dm_dev.bdev->bd_dev == dev && td->dm_dev.mode == mode)
656 int dm_get_table_device(struct mapped_device *md, dev_t dev, fmode_t mode,
657 struct dm_dev **result) {
659 struct table_device *td;
661 mutex_lock(&md->table_devices_lock);
662 td = find_table_device(&md->table_devices, dev, mode);
664 td = kmalloc_node(sizeof(*td), GFP_KERNEL, md->numa_node_id);
666 mutex_unlock(&md->table_devices_lock);
670 td->dm_dev.mode = mode;
671 td->dm_dev.bdev = NULL;
673 if ((r = open_table_device(td, dev, md))) {
674 mutex_unlock(&md->table_devices_lock);
679 format_dev_t(td->dm_dev.name, dev);
681 atomic_set(&td->count, 0);
682 list_add(&td->list, &md->table_devices);
684 atomic_inc(&td->count);
685 mutex_unlock(&md->table_devices_lock);
687 *result = &td->dm_dev;
690 EXPORT_SYMBOL_GPL(dm_get_table_device);
692 void dm_put_table_device(struct mapped_device *md, struct dm_dev *d)
694 struct table_device *td = container_of(d, struct table_device, dm_dev);
696 mutex_lock(&md->table_devices_lock);
697 if (atomic_dec_and_test(&td->count)) {
698 close_table_device(td, md);
702 mutex_unlock(&md->table_devices_lock);
704 EXPORT_SYMBOL(dm_put_table_device);
706 static void free_table_devices(struct list_head *devices)
708 struct list_head *tmp, *next;
710 list_for_each_safe(tmp, next, devices) {
711 struct table_device *td = list_entry(tmp, struct table_device, list);
713 DMWARN("dm_destroy: %s still exists with %d references",
714 td->dm_dev.name, atomic_read(&td->count));
720 * Get the geometry associated with a dm device
722 int dm_get_geometry(struct mapped_device *md, struct hd_geometry *geo)
730 * Set the geometry of a device.
732 int dm_set_geometry(struct mapped_device *md, struct hd_geometry *geo)
734 sector_t sz = (sector_t)geo->cylinders * geo->heads * geo->sectors;
736 if (geo->start > sz) {
737 DMWARN("Start sector is beyond the geometry limits.");
746 /*-----------------------------------------------------------------
748 * A more elegant soln is in the works that uses the queue
749 * merge fn, unfortunately there are a couple of changes to
750 * the block layer that I want to make for this. So in the
751 * interests of getting something for people to use I give
752 * you this clearly demarcated crap.
753 *---------------------------------------------------------------*/
755 static int __noflush_suspending(struct mapped_device *md)
757 return test_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
761 * Decrements the number of outstanding ios that a bio has been
762 * cloned into, completing the original io if necc.
764 static void dec_pending(struct dm_io *io, int error)
769 struct mapped_device *md = io->md;
771 /* Push-back supersedes any I/O errors */
772 if (unlikely(error)) {
773 spin_lock_irqsave(&io->endio_lock, flags);
774 if (!(io->error > 0 && __noflush_suspending(md)))
776 spin_unlock_irqrestore(&io->endio_lock, flags);
779 if (atomic_dec_and_test(&io->io_count)) {
780 if (io->error == DM_ENDIO_REQUEUE) {
782 * Target requested pushing back the I/O.
784 spin_lock_irqsave(&md->deferred_lock, flags);
785 if (__noflush_suspending(md))
786 bio_list_add_head(&md->deferred, io->bio);
788 /* noflush suspend was interrupted. */
790 spin_unlock_irqrestore(&md->deferred_lock, flags);
793 io_error = io->error;
798 if (io_error == DM_ENDIO_REQUEUE)
801 if ((bio->bi_opf & REQ_PREFLUSH) && bio->bi_iter.bi_size) {
803 * Preflush done for flush with data, reissue
804 * without REQ_PREFLUSH.
806 bio->bi_opf &= ~REQ_PREFLUSH;
809 /* done with normal IO or empty flush */
810 trace_block_bio_complete(md->queue, bio, io_error);
811 bio->bi_error = io_error;
817 void disable_write_same(struct mapped_device *md)
819 struct queue_limits *limits = dm_get_queue_limits(md);
821 /* device doesn't really support WRITE SAME, disable it */
822 limits->max_write_same_sectors = 0;
825 static void clone_endio(struct bio *bio)
827 int error = bio->bi_error;
829 struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone);
830 struct dm_io *io = tio->io;
831 struct mapped_device *md = tio->io->md;
832 dm_endio_fn endio = tio->ti->type->end_io;
835 r = endio(tio->ti, bio, error);
836 if (r < 0 || r == DM_ENDIO_REQUEUE)
838 * error and requeue request are handled
842 else if (r == DM_ENDIO_INCOMPLETE)
843 /* The target will handle the io */
846 DMWARN("unimplemented target endio return value: %d", r);
851 if (unlikely(r == -EREMOTEIO && (bio_op(bio) == REQ_OP_WRITE_SAME) &&
852 !bdev_get_queue(bio->bi_bdev)->limits.max_write_same_sectors))
853 disable_write_same(md);
856 dec_pending(io, error);
860 * Return maximum size of I/O possible at the supplied sector up to the current
863 static sector_t max_io_len_target_boundary(sector_t sector, struct dm_target *ti)
865 sector_t target_offset = dm_target_offset(ti, sector);
867 return ti->len - target_offset;
870 static sector_t max_io_len(sector_t sector, struct dm_target *ti)
872 sector_t len = max_io_len_target_boundary(sector, ti);
873 sector_t offset, max_len;
876 * Does the target need to split even further?
878 if (ti->max_io_len) {
879 offset = dm_target_offset(ti, sector);
880 if (unlikely(ti->max_io_len & (ti->max_io_len - 1)))
881 max_len = sector_div(offset, ti->max_io_len);
883 max_len = offset & (ti->max_io_len - 1);
884 max_len = ti->max_io_len - max_len;
893 int dm_set_target_max_io_len(struct dm_target *ti, sector_t len)
895 if (len > UINT_MAX) {
896 DMERR("Specified maximum size of target IO (%llu) exceeds limit (%u)",
897 (unsigned long long)len, UINT_MAX);
898 ti->error = "Maximum size of target IO is too large";
902 ti->max_io_len = (uint32_t) len;
906 EXPORT_SYMBOL_GPL(dm_set_target_max_io_len);
908 static long dm_blk_direct_access(struct block_device *bdev, sector_t sector,
909 void **kaddr, pfn_t *pfn, long size)
911 struct mapped_device *md = bdev->bd_disk->private_data;
912 struct dm_table *map;
913 struct dm_target *ti;
915 long len, ret = -EIO;
917 map = dm_get_live_table(md, &srcu_idx);
921 ti = dm_table_find_target(map, sector);
922 if (!dm_target_is_valid(ti))
925 len = max_io_len(sector, ti) << SECTOR_SHIFT;
926 size = min(len, size);
928 if (ti->type->direct_access)
929 ret = ti->type->direct_access(ti, sector, kaddr, pfn, size);
931 dm_put_live_table(md, srcu_idx);
932 return min(ret, size);
936 * A target may call dm_accept_partial_bio only from the map routine. It is
937 * allowed for all bio types except REQ_PREFLUSH.
939 * dm_accept_partial_bio informs the dm that the target only wants to process
940 * additional n_sectors sectors of the bio and the rest of the data should be
941 * sent in a next bio.
943 * A diagram that explains the arithmetics:
944 * +--------------------+---------------+-------+
946 * +--------------------+---------------+-------+
948 * <-------------- *tio->len_ptr --------------->
949 * <------- bi_size ------->
952 * Region 1 was already iterated over with bio_advance or similar function.
953 * (it may be empty if the target doesn't use bio_advance)
954 * Region 2 is the remaining bio size that the target wants to process.
955 * (it may be empty if region 1 is non-empty, although there is no reason
957 * The target requires that region 3 is to be sent in the next bio.
959 * If the target wants to receive multiple copies of the bio (via num_*bios, etc),
960 * the partially processed part (the sum of regions 1+2) must be the same for all
963 void dm_accept_partial_bio(struct bio *bio, unsigned n_sectors)
965 struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone);
966 unsigned bi_size = bio->bi_iter.bi_size >> SECTOR_SHIFT;
967 BUG_ON(bio->bi_opf & REQ_PREFLUSH);
968 BUG_ON(bi_size > *tio->len_ptr);
969 BUG_ON(n_sectors > bi_size);
970 *tio->len_ptr -= bi_size - n_sectors;
971 bio->bi_iter.bi_size = n_sectors << SECTOR_SHIFT;
973 EXPORT_SYMBOL_GPL(dm_accept_partial_bio);
975 static void __map_bio(struct dm_target_io *tio)
979 struct bio *clone = &tio->clone;
980 struct dm_target *ti = tio->ti;
982 clone->bi_end_io = clone_endio;
985 * Map the clone. If r == 0 we don't need to do
986 * anything, the target has assumed ownership of
989 atomic_inc(&tio->io->io_count);
990 sector = clone->bi_iter.bi_sector;
991 r = ti->type->map(ti, clone);
992 if (r == DM_MAPIO_REMAPPED) {
993 /* the bio has been remapped so dispatch it */
995 trace_block_bio_remap(bdev_get_queue(clone->bi_bdev), clone,
996 tio->io->bio->bi_bdev->bd_dev, sector);
998 generic_make_request(clone);
999 } else if (r < 0 || r == DM_MAPIO_REQUEUE) {
1000 /* error the io and bail out, or requeue it if needed */
1001 dec_pending(tio->io, r);
1003 } else if (r != DM_MAPIO_SUBMITTED) {
1004 DMWARN("unimplemented target map return value: %d", r);
1010 struct mapped_device *md;
1011 struct dm_table *map;
1015 unsigned sector_count;
1018 static void bio_setup_sector(struct bio *bio, sector_t sector, unsigned len)
1020 bio->bi_iter.bi_sector = sector;
1021 bio->bi_iter.bi_size = to_bytes(len);
1025 * Creates a bio that consists of range of complete bvecs.
1027 static int clone_bio(struct dm_target_io *tio, struct bio *bio,
1028 sector_t sector, unsigned len)
1030 struct bio *clone = &tio->clone;
1032 __bio_clone_fast(clone, bio);
1034 if (bio_integrity(bio)) {
1035 int r = bio_integrity_clone(clone, bio, GFP_NOIO);
1040 bio_advance(clone, to_bytes(sector - clone->bi_iter.bi_sector));
1041 clone->bi_iter.bi_size = to_bytes(len);
1043 if (bio_integrity(bio))
1044 bio_integrity_trim(clone, 0, len);
1049 static struct dm_target_io *alloc_tio(struct clone_info *ci,
1050 struct dm_target *ti,
1051 unsigned target_bio_nr)
1053 struct dm_target_io *tio;
1056 clone = bio_alloc_bioset(GFP_NOIO, 0, ci->md->bs);
1057 tio = container_of(clone, struct dm_target_io, clone);
1061 tio->target_bio_nr = target_bio_nr;
1066 static void __clone_and_map_simple_bio(struct clone_info *ci,
1067 struct dm_target *ti,
1068 unsigned target_bio_nr, unsigned *len)
1070 struct dm_target_io *tio = alloc_tio(ci, ti, target_bio_nr);
1071 struct bio *clone = &tio->clone;
1075 __bio_clone_fast(clone, ci->bio);
1077 bio_setup_sector(clone, ci->sector, *len);
1082 static void __send_duplicate_bios(struct clone_info *ci, struct dm_target *ti,
1083 unsigned num_bios, unsigned *len)
1085 unsigned target_bio_nr;
1087 for (target_bio_nr = 0; target_bio_nr < num_bios; target_bio_nr++)
1088 __clone_and_map_simple_bio(ci, ti, target_bio_nr, len);
1091 static int __send_empty_flush(struct clone_info *ci)
1093 unsigned target_nr = 0;
1094 struct dm_target *ti;
1096 BUG_ON(bio_has_data(ci->bio));
1097 while ((ti = dm_table_get_target(ci->map, target_nr++)))
1098 __send_duplicate_bios(ci, ti, ti->num_flush_bios, NULL);
1103 static int __clone_and_map_data_bio(struct clone_info *ci, struct dm_target *ti,
1104 sector_t sector, unsigned *len)
1106 struct bio *bio = ci->bio;
1107 struct dm_target_io *tio;
1108 unsigned target_bio_nr;
1109 unsigned num_target_bios = 1;
1113 * Does the target want to receive duplicate copies of the bio?
1115 if (bio_data_dir(bio) == WRITE && ti->num_write_bios)
1116 num_target_bios = ti->num_write_bios(ti, bio);
1118 for (target_bio_nr = 0; target_bio_nr < num_target_bios; target_bio_nr++) {
1119 tio = alloc_tio(ci, ti, target_bio_nr);
1121 r = clone_bio(tio, bio, sector, *len);
1132 typedef unsigned (*get_num_bios_fn)(struct dm_target *ti);
1134 static unsigned get_num_discard_bios(struct dm_target *ti)
1136 return ti->num_discard_bios;
1139 static unsigned get_num_write_same_bios(struct dm_target *ti)
1141 return ti->num_write_same_bios;
1144 typedef bool (*is_split_required_fn)(struct dm_target *ti);
1146 static bool is_split_required_for_discard(struct dm_target *ti)
1148 return ti->split_discard_bios;
1151 static int __send_changing_extent_only(struct clone_info *ci,
1152 get_num_bios_fn get_num_bios,
1153 is_split_required_fn is_split_required)
1155 struct dm_target *ti;
1160 ti = dm_table_find_target(ci->map, ci->sector);
1161 if (!dm_target_is_valid(ti))
1165 * Even though the device advertised support for this type of
1166 * request, that does not mean every target supports it, and
1167 * reconfiguration might also have changed that since the
1168 * check was performed.
1170 num_bios = get_num_bios ? get_num_bios(ti) : 0;
1174 if (is_split_required && !is_split_required(ti))
1175 len = min((sector_t)ci->sector_count, max_io_len_target_boundary(ci->sector, ti));
1177 len = min((sector_t)ci->sector_count, max_io_len(ci->sector, ti));
1179 __send_duplicate_bios(ci, ti, num_bios, &len);
1182 } while (ci->sector_count -= len);
1187 static int __send_discard(struct clone_info *ci)
1189 return __send_changing_extent_only(ci, get_num_discard_bios,
1190 is_split_required_for_discard);
1193 static int __send_write_same(struct clone_info *ci)
1195 return __send_changing_extent_only(ci, get_num_write_same_bios, NULL);
1199 * Select the correct strategy for processing a non-flush bio.
1201 static int __split_and_process_non_flush(struct clone_info *ci)
1203 struct bio *bio = ci->bio;
1204 struct dm_target *ti;
1208 if (unlikely(bio_op(bio) == REQ_OP_DISCARD))
1209 return __send_discard(ci);
1210 else if (unlikely(bio_op(bio) == REQ_OP_WRITE_SAME))
1211 return __send_write_same(ci);
1213 ti = dm_table_find_target(ci->map, ci->sector);
1214 if (!dm_target_is_valid(ti))
1217 len = min_t(sector_t, max_io_len(ci->sector, ti), ci->sector_count);
1219 r = __clone_and_map_data_bio(ci, ti, ci->sector, &len);
1224 ci->sector_count -= len;
1230 * Entry point to split a bio into clones and submit them to the targets.
1232 static void __split_and_process_bio(struct mapped_device *md,
1233 struct dm_table *map, struct bio *bio)
1235 struct clone_info ci;
1238 if (unlikely(!map)) {
1245 ci.io = alloc_io(md);
1247 atomic_set(&ci.io->io_count, 1);
1250 spin_lock_init(&ci.io->endio_lock);
1251 ci.sector = bio->bi_iter.bi_sector;
1253 start_io_acct(ci.io);
1255 if (bio->bi_opf & REQ_PREFLUSH) {
1256 ci.bio = &ci.md->flush_bio;
1257 ci.sector_count = 0;
1258 error = __send_empty_flush(&ci);
1259 /* dec_pending submits any data associated with flush */
1262 ci.sector_count = bio_sectors(bio);
1263 while (ci.sector_count && !error)
1264 error = __split_and_process_non_flush(&ci);
1267 /* drop the extra reference count */
1268 dec_pending(ci.io, error);
1270 /*-----------------------------------------------------------------
1272 *---------------------------------------------------------------*/
1275 * The request function that just remaps the bio built up by
1278 static blk_qc_t dm_make_request(struct request_queue *q, struct bio *bio)
1280 int rw = bio_data_dir(bio);
1281 struct mapped_device *md = q->queuedata;
1283 struct dm_table *map;
1285 map = dm_get_live_table(md, &srcu_idx);
1287 generic_start_io_acct(rw, bio_sectors(bio), &dm_disk(md)->part0);
1289 /* if we're suspended, we have to queue this io for later */
1290 if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags))) {
1291 dm_put_live_table(md, srcu_idx);
1293 if (!(bio->bi_opf & REQ_RAHEAD))
1297 return BLK_QC_T_NONE;
1300 __split_and_process_bio(md, map, bio);
1301 dm_put_live_table(md, srcu_idx);
1302 return BLK_QC_T_NONE;
1305 static int dm_any_congested(void *congested_data, int bdi_bits)
1308 struct mapped_device *md = congested_data;
1309 struct dm_table *map;
1311 if (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
1312 if (dm_request_based(md)) {
1314 * With request-based DM we only need to check the
1315 * top-level queue for congestion.
1317 r = md->queue->backing_dev_info.wb.state & bdi_bits;
1319 map = dm_get_live_table_fast(md);
1321 r = dm_table_any_congested(map, bdi_bits);
1322 dm_put_live_table_fast(md);
1329 /*-----------------------------------------------------------------
1330 * An IDR is used to keep track of allocated minor numbers.
1331 *---------------------------------------------------------------*/
1332 static void free_minor(int minor)
1334 spin_lock(&_minor_lock);
1335 idr_remove(&_minor_idr, minor);
1336 spin_unlock(&_minor_lock);
1340 * See if the device with a specific minor # is free.
1342 static int specific_minor(int minor)
1346 if (minor >= (1 << MINORBITS))
1349 idr_preload(GFP_KERNEL);
1350 spin_lock(&_minor_lock);
1352 r = idr_alloc(&_minor_idr, MINOR_ALLOCED, minor, minor + 1, GFP_NOWAIT);
1354 spin_unlock(&_minor_lock);
1357 return r == -ENOSPC ? -EBUSY : r;
1361 static int next_free_minor(int *minor)
1365 idr_preload(GFP_KERNEL);
1366 spin_lock(&_minor_lock);
1368 r = idr_alloc(&_minor_idr, MINOR_ALLOCED, 0, 1 << MINORBITS, GFP_NOWAIT);
1370 spin_unlock(&_minor_lock);
1378 static const struct block_device_operations dm_blk_dops;
1380 static void dm_wq_work(struct work_struct *work);
1382 void dm_init_md_queue(struct mapped_device *md)
1385 * Request-based dm devices cannot be stacked on top of bio-based dm
1386 * devices. The type of this dm device may not have been decided yet.
1387 * The type is decided at the first table loading time.
1388 * To prevent problematic device stacking, clear the queue flag
1389 * for request stacking support until then.
1391 * This queue is new, so no concurrency on the queue_flags.
1393 queue_flag_clear_unlocked(QUEUE_FLAG_STACKABLE, md->queue);
1396 * Initialize data that will only be used by a non-blk-mq DM queue
1397 * - must do so here (in alloc_dev callchain) before queue is used
1399 md->queue->queuedata = md;
1400 md->queue->backing_dev_info.congested_data = md;
1403 void dm_init_normal_md_queue(struct mapped_device *md)
1405 md->use_blk_mq = false;
1406 dm_init_md_queue(md);
1409 * Initialize aspects of queue that aren't relevant for blk-mq
1411 md->queue->backing_dev_info.congested_fn = dm_any_congested;
1412 blk_queue_bounce_limit(md->queue, BLK_BOUNCE_ANY);
1415 static void cleanup_mapped_device(struct mapped_device *md)
1418 destroy_workqueue(md->wq);
1419 if (md->kworker_task)
1420 kthread_stop(md->kworker_task);
1421 mempool_destroy(md->io_pool);
1422 mempool_destroy(md->rq_pool);
1424 bioset_free(md->bs);
1426 cleanup_srcu_struct(&md->io_barrier);
1429 spin_lock(&_minor_lock);
1430 md->disk->private_data = NULL;
1431 spin_unlock(&_minor_lock);
1432 del_gendisk(md->disk);
1437 blk_cleanup_queue(md->queue);
1444 dm_mq_cleanup_mapped_device(md);
1448 * Allocate and initialise a blank device with a given minor.
1450 static struct mapped_device *alloc_dev(int minor)
1452 int r, numa_node_id = dm_get_numa_node();
1453 struct mapped_device *md;
1456 md = kzalloc_node(sizeof(*md), GFP_KERNEL, numa_node_id);
1458 DMWARN("unable to allocate device, out of memory.");
1462 if (!try_module_get(THIS_MODULE))
1463 goto bad_module_get;
1465 /* get a minor number for the dev */
1466 if (minor == DM_ANY_MINOR)
1467 r = next_free_minor(&minor);
1469 r = specific_minor(minor);
1473 r = init_srcu_struct(&md->io_barrier);
1475 goto bad_io_barrier;
1477 md->numa_node_id = numa_node_id;
1478 md->use_blk_mq = dm_use_blk_mq_default();
1479 md->init_tio_pdu = false;
1480 md->type = DM_TYPE_NONE;
1481 mutex_init(&md->suspend_lock);
1482 mutex_init(&md->type_lock);
1483 mutex_init(&md->table_devices_lock);
1484 spin_lock_init(&md->deferred_lock);
1485 atomic_set(&md->holders, 1);
1486 atomic_set(&md->open_count, 0);
1487 atomic_set(&md->event_nr, 0);
1488 atomic_set(&md->uevent_seq, 0);
1489 INIT_LIST_HEAD(&md->uevent_list);
1490 INIT_LIST_HEAD(&md->table_devices);
1491 spin_lock_init(&md->uevent_lock);
1493 md->queue = blk_alloc_queue_node(GFP_KERNEL, numa_node_id);
1497 dm_init_md_queue(md);
1499 md->disk = alloc_disk_node(1, numa_node_id);
1503 atomic_set(&md->pending[0], 0);
1504 atomic_set(&md->pending[1], 0);
1505 init_waitqueue_head(&md->wait);
1506 INIT_WORK(&md->work, dm_wq_work);
1507 init_waitqueue_head(&md->eventq);
1508 init_completion(&md->kobj_holder.completion);
1509 md->kworker_task = NULL;
1511 md->disk->major = _major;
1512 md->disk->first_minor = minor;
1513 md->disk->fops = &dm_blk_dops;
1514 md->disk->queue = md->queue;
1515 md->disk->private_data = md;
1516 sprintf(md->disk->disk_name, "dm-%d", minor);
1518 format_dev_t(md->name, MKDEV(_major, minor));
1520 md->wq = alloc_workqueue("kdmflush", WQ_MEM_RECLAIM, 0);
1524 md->bdev = bdget_disk(md->disk, 0);
1528 bio_init(&md->flush_bio);
1529 md->flush_bio.bi_bdev = md->bdev;
1530 bio_set_op_attrs(&md->flush_bio, REQ_OP_WRITE, WRITE_FLUSH);
1532 dm_stats_init(&md->stats);
1534 /* Populate the mapping, nobody knows we exist yet */
1535 spin_lock(&_minor_lock);
1536 old_md = idr_replace(&_minor_idr, md, minor);
1537 spin_unlock(&_minor_lock);
1539 BUG_ON(old_md != MINOR_ALLOCED);
1544 cleanup_mapped_device(md);
1548 module_put(THIS_MODULE);
1554 static void unlock_fs(struct mapped_device *md);
1556 static void free_dev(struct mapped_device *md)
1558 int minor = MINOR(disk_devt(md->disk));
1562 cleanup_mapped_device(md);
1564 free_table_devices(&md->table_devices);
1565 dm_stats_cleanup(&md->stats);
1568 module_put(THIS_MODULE);
1572 static void __bind_mempools(struct mapped_device *md, struct dm_table *t)
1574 struct dm_md_mempools *p = dm_table_get_md_mempools(t);
1577 /* The md already has necessary mempools. */
1578 if (dm_table_bio_based(t)) {
1580 * Reload bioset because front_pad may have changed
1581 * because a different table was loaded.
1583 bioset_free(md->bs);
1588 * There's no need to reload with request-based dm
1589 * because the size of front_pad doesn't change.
1590 * Note for future: If you are to reload bioset,
1591 * prep-ed requests in the queue may refer
1592 * to bio from the old bioset, so you must walk
1593 * through the queue to unprep.
1598 BUG_ON(!p || md->io_pool || md->rq_pool || md->bs);
1600 md->io_pool = p->io_pool;
1602 md->rq_pool = p->rq_pool;
1608 /* mempool bind completed, no longer need any mempools in the table */
1609 dm_table_free_md_mempools(t);
1613 * Bind a table to the device.
1615 static void event_callback(void *context)
1617 unsigned long flags;
1619 struct mapped_device *md = (struct mapped_device *) context;
1621 spin_lock_irqsave(&md->uevent_lock, flags);
1622 list_splice_init(&md->uevent_list, &uevents);
1623 spin_unlock_irqrestore(&md->uevent_lock, flags);
1625 dm_send_uevents(&uevents, &disk_to_dev(md->disk)->kobj);
1627 atomic_inc(&md->event_nr);
1628 wake_up(&md->eventq);
1632 * Protected by md->suspend_lock obtained by dm_swap_table().
1634 static void __set_size(struct mapped_device *md, sector_t size)
1636 set_capacity(md->disk, size);
1638 i_size_write(md->bdev->bd_inode, (loff_t)size << SECTOR_SHIFT);
1642 * Returns old map, which caller must destroy.
1644 static struct dm_table *__bind(struct mapped_device *md, struct dm_table *t,
1645 struct queue_limits *limits)
1647 struct dm_table *old_map;
1648 struct request_queue *q = md->queue;
1651 size = dm_table_get_size(t);
1654 * Wipe any geometry if the size of the table changed.
1656 if (size != dm_get_size(md))
1657 memset(&md->geometry, 0, sizeof(md->geometry));
1659 __set_size(md, size);
1661 dm_table_event_callback(t, event_callback, md);
1664 * The queue hasn't been stopped yet, if the old table type wasn't
1665 * for request-based during suspension. So stop it to prevent
1666 * I/O mapping before resume.
1667 * This must be done before setting the queue restrictions,
1668 * because request-based dm may be run just after the setting.
1670 if (dm_table_request_based(t)) {
1673 * Leverage the fact that request-based DM targets are
1674 * immutable singletons and establish md->immutable_target
1675 * - used to optimize both dm_request_fn and dm_mq_queue_rq
1677 md->immutable_target = dm_table_get_immutable_target(t);
1680 __bind_mempools(md, t);
1682 old_map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
1683 rcu_assign_pointer(md->map, (void *)t);
1684 md->immutable_target_type = dm_table_get_immutable_target_type(t);
1686 dm_table_set_restrictions(t, q, limits);
1694 * Returns unbound table for the caller to free.
1696 static struct dm_table *__unbind(struct mapped_device *md)
1698 struct dm_table *map = rcu_dereference_protected(md->map, 1);
1703 dm_table_event_callback(map, NULL, NULL);
1704 RCU_INIT_POINTER(md->map, NULL);
1711 * Constructor for a new device.
1713 int dm_create(int minor, struct mapped_device **result)
1715 struct mapped_device *md;
1717 md = alloc_dev(minor);
1728 * Functions to manage md->type.
1729 * All are required to hold md->type_lock.
1731 void dm_lock_md_type(struct mapped_device *md)
1733 mutex_lock(&md->type_lock);
1736 void dm_unlock_md_type(struct mapped_device *md)
1738 mutex_unlock(&md->type_lock);
1741 void dm_set_md_type(struct mapped_device *md, unsigned type)
1743 BUG_ON(!mutex_is_locked(&md->type_lock));
1747 unsigned dm_get_md_type(struct mapped_device *md)
1752 struct target_type *dm_get_immutable_target_type(struct mapped_device *md)
1754 return md->immutable_target_type;
1758 * The queue_limits are only valid as long as you have a reference
1761 struct queue_limits *dm_get_queue_limits(struct mapped_device *md)
1763 BUG_ON(!atomic_read(&md->holders));
1764 return &md->queue->limits;
1766 EXPORT_SYMBOL_GPL(dm_get_queue_limits);
1769 * Setup the DM device's queue based on md's type
1771 int dm_setup_md_queue(struct mapped_device *md, struct dm_table *t)
1774 unsigned type = dm_get_md_type(md);
1777 case DM_TYPE_REQUEST_BASED:
1778 r = dm_old_init_request_queue(md);
1780 DMERR("Cannot initialize queue for request-based mapped device");
1784 case DM_TYPE_MQ_REQUEST_BASED:
1785 r = dm_mq_init_request_queue(md, t);
1787 DMERR("Cannot initialize queue for request-based dm-mq mapped device");
1791 case DM_TYPE_BIO_BASED:
1792 case DM_TYPE_DAX_BIO_BASED:
1793 dm_init_normal_md_queue(md);
1794 blk_queue_make_request(md->queue, dm_make_request);
1796 * DM handles splitting bios as needed. Free the bio_split bioset
1797 * since it won't be used (saves 1 process per bio-based DM device).
1799 bioset_free(md->queue->bio_split);
1800 md->queue->bio_split = NULL;
1802 if (type == DM_TYPE_DAX_BIO_BASED)
1803 queue_flag_set_unlocked(QUEUE_FLAG_DAX, md->queue);
1810 struct mapped_device *dm_get_md(dev_t dev)
1812 struct mapped_device *md;
1813 unsigned minor = MINOR(dev);
1815 if (MAJOR(dev) != _major || minor >= (1 << MINORBITS))
1818 spin_lock(&_minor_lock);
1820 md = idr_find(&_minor_idr, minor);
1822 if ((md == MINOR_ALLOCED ||
1823 (MINOR(disk_devt(dm_disk(md))) != minor) ||
1824 dm_deleting_md(md) ||
1825 test_bit(DMF_FREEING, &md->flags))) {
1833 spin_unlock(&_minor_lock);
1837 EXPORT_SYMBOL_GPL(dm_get_md);
1839 void *dm_get_mdptr(struct mapped_device *md)
1841 return md->interface_ptr;
1844 void dm_set_mdptr(struct mapped_device *md, void *ptr)
1846 md->interface_ptr = ptr;
1849 void dm_get(struct mapped_device *md)
1851 atomic_inc(&md->holders);
1852 BUG_ON(test_bit(DMF_FREEING, &md->flags));
1855 int dm_hold(struct mapped_device *md)
1857 spin_lock(&_minor_lock);
1858 if (test_bit(DMF_FREEING, &md->flags)) {
1859 spin_unlock(&_minor_lock);
1863 spin_unlock(&_minor_lock);
1866 EXPORT_SYMBOL_GPL(dm_hold);
1868 const char *dm_device_name(struct mapped_device *md)
1872 EXPORT_SYMBOL_GPL(dm_device_name);
1874 static void __dm_destroy(struct mapped_device *md, bool wait)
1876 struct dm_table *map;
1881 spin_lock(&_minor_lock);
1882 idr_replace(&_minor_idr, MINOR_ALLOCED, MINOR(disk_devt(dm_disk(md))));
1883 set_bit(DMF_FREEING, &md->flags);
1884 spin_unlock(&_minor_lock);
1886 if (dm_request_based(md) && md->kworker_task)
1887 flush_kthread_worker(&md->kworker);
1890 * Take suspend_lock so that presuspend and postsuspend methods
1891 * do not race with internal suspend.
1893 mutex_lock(&md->suspend_lock);
1894 map = dm_get_live_table(md, &srcu_idx);
1895 if (!dm_suspended_md(md)) {
1896 dm_table_presuspend_targets(map);
1897 dm_table_postsuspend_targets(map);
1899 /* dm_put_live_table must be before msleep, otherwise deadlock is possible */
1900 dm_put_live_table(md, srcu_idx);
1901 mutex_unlock(&md->suspend_lock);
1904 * Rare, but there may be I/O requests still going to complete,
1905 * for example. Wait for all references to disappear.
1906 * No one should increment the reference count of the mapped_device,
1907 * after the mapped_device state becomes DMF_FREEING.
1910 while (atomic_read(&md->holders))
1912 else if (atomic_read(&md->holders))
1913 DMWARN("%s: Forcibly removing mapped_device still in use! (%d users)",
1914 dm_device_name(md), atomic_read(&md->holders));
1917 dm_table_destroy(__unbind(md));
1921 void dm_destroy(struct mapped_device *md)
1923 __dm_destroy(md, true);
1926 void dm_destroy_immediate(struct mapped_device *md)
1928 __dm_destroy(md, false);
1931 void dm_put(struct mapped_device *md)
1933 atomic_dec(&md->holders);
1935 EXPORT_SYMBOL_GPL(dm_put);
1937 static int dm_wait_for_completion(struct mapped_device *md, int interruptible)
1940 DECLARE_WAITQUEUE(wait, current);
1942 add_wait_queue(&md->wait, &wait);
1945 set_current_state(interruptible);
1947 if (!md_in_flight(md))
1950 if (interruptible == TASK_INTERRUPTIBLE &&
1951 signal_pending(current)) {
1958 set_current_state(TASK_RUNNING);
1960 remove_wait_queue(&md->wait, &wait);
1966 * Process the deferred bios
1968 static void dm_wq_work(struct work_struct *work)
1970 struct mapped_device *md = container_of(work, struct mapped_device,
1974 struct dm_table *map;
1976 map = dm_get_live_table(md, &srcu_idx);
1978 while (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
1979 spin_lock_irq(&md->deferred_lock);
1980 c = bio_list_pop(&md->deferred);
1981 spin_unlock_irq(&md->deferred_lock);
1986 if (dm_request_based(md))
1987 generic_make_request(c);
1989 __split_and_process_bio(md, map, c);
1992 dm_put_live_table(md, srcu_idx);
1995 static void dm_queue_flush(struct mapped_device *md)
1997 clear_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
1998 smp_mb__after_atomic();
1999 queue_work(md->wq, &md->work);
2003 * Swap in a new table, returning the old one for the caller to destroy.
2005 struct dm_table *dm_swap_table(struct mapped_device *md, struct dm_table *table)
2007 struct dm_table *live_map = NULL, *map = ERR_PTR(-EINVAL);
2008 struct queue_limits limits;
2011 mutex_lock(&md->suspend_lock);
2013 /* device must be suspended */
2014 if (!dm_suspended_md(md))
2018 * If the new table has no data devices, retain the existing limits.
2019 * This helps multipath with queue_if_no_path if all paths disappear,
2020 * then new I/O is queued based on these limits, and then some paths
2023 if (dm_table_has_no_data_devices(table)) {
2024 live_map = dm_get_live_table_fast(md);
2026 limits = md->queue->limits;
2027 dm_put_live_table_fast(md);
2031 r = dm_calculate_queue_limits(table, &limits);
2038 map = __bind(md, table, &limits);
2041 mutex_unlock(&md->suspend_lock);
2046 * Functions to lock and unlock any filesystem running on the
2049 static int lock_fs(struct mapped_device *md)
2053 WARN_ON(md->frozen_sb);
2055 md->frozen_sb = freeze_bdev(md->bdev);
2056 if (IS_ERR(md->frozen_sb)) {
2057 r = PTR_ERR(md->frozen_sb);
2058 md->frozen_sb = NULL;
2062 set_bit(DMF_FROZEN, &md->flags);
2067 static void unlock_fs(struct mapped_device *md)
2069 if (!test_bit(DMF_FROZEN, &md->flags))
2072 thaw_bdev(md->bdev, md->frozen_sb);
2073 md->frozen_sb = NULL;
2074 clear_bit(DMF_FROZEN, &md->flags);
2078 * If __dm_suspend returns 0, the device is completely quiescent
2079 * now. There is no request-processing activity. All new requests
2080 * are being added to md->deferred list.
2082 * Caller must hold md->suspend_lock
2084 static int __dm_suspend(struct mapped_device *md, struct dm_table *map,
2085 unsigned suspend_flags, int interruptible,
2086 int dmf_suspended_flag)
2088 bool do_lockfs = suspend_flags & DM_SUSPEND_LOCKFS_FLAG;
2089 bool noflush = suspend_flags & DM_SUSPEND_NOFLUSH_FLAG;
2093 * DMF_NOFLUSH_SUSPENDING must be set before presuspend.
2094 * This flag is cleared before dm_suspend returns.
2097 set_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
2100 * This gets reverted if there's an error later and the targets
2101 * provide the .presuspend_undo hook.
2103 dm_table_presuspend_targets(map);
2106 * Flush I/O to the device.
2107 * Any I/O submitted after lock_fs() may not be flushed.
2108 * noflush takes precedence over do_lockfs.
2109 * (lock_fs() flushes I/Os and waits for them to complete.)
2111 if (!noflush && do_lockfs) {
2114 dm_table_presuspend_undo_targets(map);
2120 * Here we must make sure that no processes are submitting requests
2121 * to target drivers i.e. no one may be executing
2122 * __split_and_process_bio. This is called from dm_request and
2125 * To get all processes out of __split_and_process_bio in dm_request,
2126 * we take the write lock. To prevent any process from reentering
2127 * __split_and_process_bio from dm_request and quiesce the thread
2128 * (dm_wq_work), we set BMF_BLOCK_IO_FOR_SUSPEND and call
2129 * flush_workqueue(md->wq).
2131 set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2133 synchronize_srcu(&md->io_barrier);
2136 * Stop md->queue before flushing md->wq in case request-based
2137 * dm defers requests to md->wq from md->queue.
2139 if (dm_request_based(md)) {
2140 dm_stop_queue(md->queue);
2141 if (md->kworker_task)
2142 flush_kthread_worker(&md->kworker);
2145 flush_workqueue(md->wq);
2148 * At this point no more requests are entering target request routines.
2149 * We call dm_wait_for_completion to wait for all existing requests
2152 r = dm_wait_for_completion(md, interruptible);
2154 set_bit(dmf_suspended_flag, &md->flags);
2157 clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
2159 synchronize_srcu(&md->io_barrier);
2161 /* were we interrupted ? */
2165 if (dm_request_based(md))
2166 dm_start_queue(md->queue);
2169 dm_table_presuspend_undo_targets(map);
2170 /* pushback list is already flushed, so skip flush */
2177 * We need to be able to change a mapping table under a mounted
2178 * filesystem. For example we might want to move some data in
2179 * the background. Before the table can be swapped with
2180 * dm_bind_table, dm_suspend must be called to flush any in
2181 * flight bios and ensure that any further io gets deferred.
2184 * Suspend mechanism in request-based dm.
2186 * 1. Flush all I/Os by lock_fs() if needed.
2187 * 2. Stop dispatching any I/O by stopping the request_queue.
2188 * 3. Wait for all in-flight I/Os to be completed or requeued.
2190 * To abort suspend, start the request_queue.
2192 int dm_suspend(struct mapped_device *md, unsigned suspend_flags)
2194 struct dm_table *map = NULL;
2198 mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING);
2200 if (dm_suspended_md(md)) {
2205 if (dm_suspended_internally_md(md)) {
2206 /* already internally suspended, wait for internal resume */
2207 mutex_unlock(&md->suspend_lock);
2208 r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE);
2214 map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2216 r = __dm_suspend(md, map, suspend_flags, TASK_INTERRUPTIBLE, DMF_SUSPENDED);
2220 dm_table_postsuspend_targets(map);
2223 mutex_unlock(&md->suspend_lock);
2227 static int __dm_resume(struct mapped_device *md, struct dm_table *map)
2230 int r = dm_table_resume_targets(map);
2238 * Flushing deferred I/Os must be done after targets are resumed
2239 * so that mapping of targets can work correctly.
2240 * Request-based dm is queueing the deferred I/Os in its request_queue.
2242 if (dm_request_based(md))
2243 dm_start_queue(md->queue);
2250 int dm_resume(struct mapped_device *md)
2253 struct dm_table *map = NULL;
2256 mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING);
2258 if (!dm_suspended_md(md))
2261 if (dm_suspended_internally_md(md)) {
2262 /* already internally suspended, wait for internal resume */
2263 mutex_unlock(&md->suspend_lock);
2264 r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE);
2270 map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2271 if (!map || !dm_table_get_size(map))
2274 r = __dm_resume(md, map);
2278 clear_bit(DMF_SUSPENDED, &md->flags);
2282 mutex_unlock(&md->suspend_lock);
2288 * Internal suspend/resume works like userspace-driven suspend. It waits
2289 * until all bios finish and prevents issuing new bios to the target drivers.
2290 * It may be used only from the kernel.
2293 static void __dm_internal_suspend(struct mapped_device *md, unsigned suspend_flags)
2295 struct dm_table *map = NULL;
2297 if (md->internal_suspend_count++)
2298 return; /* nested internal suspend */
2300 if (dm_suspended_md(md)) {
2301 set_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
2302 return; /* nest suspend */
2305 map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2308 * Using TASK_UNINTERRUPTIBLE because only NOFLUSH internal suspend is
2309 * supported. Properly supporting a TASK_INTERRUPTIBLE internal suspend
2310 * would require changing .presuspend to return an error -- avoid this
2311 * until there is a need for more elaborate variants of internal suspend.
2313 (void) __dm_suspend(md, map, suspend_flags, TASK_UNINTERRUPTIBLE,
2314 DMF_SUSPENDED_INTERNALLY);
2316 dm_table_postsuspend_targets(map);
2319 static void __dm_internal_resume(struct mapped_device *md)
2321 BUG_ON(!md->internal_suspend_count);
2323 if (--md->internal_suspend_count)
2324 return; /* resume from nested internal suspend */
2326 if (dm_suspended_md(md))
2327 goto done; /* resume from nested suspend */
2330 * NOTE: existing callers don't need to call dm_table_resume_targets
2331 * (which may fail -- so best to avoid it for now by passing NULL map)
2333 (void) __dm_resume(md, NULL);
2336 clear_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
2337 smp_mb__after_atomic();
2338 wake_up_bit(&md->flags, DMF_SUSPENDED_INTERNALLY);
2341 void dm_internal_suspend_noflush(struct mapped_device *md)
2343 mutex_lock(&md->suspend_lock);
2344 __dm_internal_suspend(md, DM_SUSPEND_NOFLUSH_FLAG);
2345 mutex_unlock(&md->suspend_lock);
2347 EXPORT_SYMBOL_GPL(dm_internal_suspend_noflush);
2349 void dm_internal_resume(struct mapped_device *md)
2351 mutex_lock(&md->suspend_lock);
2352 __dm_internal_resume(md);
2353 mutex_unlock(&md->suspend_lock);
2355 EXPORT_SYMBOL_GPL(dm_internal_resume);
2358 * Fast variants of internal suspend/resume hold md->suspend_lock,
2359 * which prevents interaction with userspace-driven suspend.
2362 void dm_internal_suspend_fast(struct mapped_device *md)
2364 mutex_lock(&md->suspend_lock);
2365 if (dm_suspended_md(md) || dm_suspended_internally_md(md))
2368 set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2369 synchronize_srcu(&md->io_barrier);
2370 flush_workqueue(md->wq);
2371 dm_wait_for_completion(md, TASK_UNINTERRUPTIBLE);
2373 EXPORT_SYMBOL_GPL(dm_internal_suspend_fast);
2375 void dm_internal_resume_fast(struct mapped_device *md)
2377 if (dm_suspended_md(md) || dm_suspended_internally_md(md))
2383 mutex_unlock(&md->suspend_lock);
2385 EXPORT_SYMBOL_GPL(dm_internal_resume_fast);
2387 /*-----------------------------------------------------------------
2388 * Event notification.
2389 *---------------------------------------------------------------*/
2390 int dm_kobject_uevent(struct mapped_device *md, enum kobject_action action,
2393 char udev_cookie[DM_COOKIE_LENGTH];
2394 char *envp[] = { udev_cookie, NULL };
2397 return kobject_uevent(&disk_to_dev(md->disk)->kobj, action);
2399 snprintf(udev_cookie, DM_COOKIE_LENGTH, "%s=%u",
2400 DM_COOKIE_ENV_VAR_NAME, cookie);
2401 return kobject_uevent_env(&disk_to_dev(md->disk)->kobj,
2406 uint32_t dm_next_uevent_seq(struct mapped_device *md)
2408 return atomic_add_return(1, &md->uevent_seq);
2411 uint32_t dm_get_event_nr(struct mapped_device *md)
2413 return atomic_read(&md->event_nr);
2416 int dm_wait_event(struct mapped_device *md, int event_nr)
2418 return wait_event_interruptible(md->eventq,
2419 (event_nr != atomic_read(&md->event_nr)));
2422 void dm_uevent_add(struct mapped_device *md, struct list_head *elist)
2424 unsigned long flags;
2426 spin_lock_irqsave(&md->uevent_lock, flags);
2427 list_add(elist, &md->uevent_list);
2428 spin_unlock_irqrestore(&md->uevent_lock, flags);
2432 * The gendisk is only valid as long as you have a reference
2435 struct gendisk *dm_disk(struct mapped_device *md)
2439 EXPORT_SYMBOL_GPL(dm_disk);
2441 struct kobject *dm_kobject(struct mapped_device *md)
2443 return &md->kobj_holder.kobj;
2446 struct mapped_device *dm_get_from_kobject(struct kobject *kobj)
2448 struct mapped_device *md;
2450 md = container_of(kobj, struct mapped_device, kobj_holder.kobj);
2452 if (test_bit(DMF_FREEING, &md->flags) ||
2460 int dm_suspended_md(struct mapped_device *md)
2462 return test_bit(DMF_SUSPENDED, &md->flags);
2465 int dm_suspended_internally_md(struct mapped_device *md)
2467 return test_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
2470 int dm_test_deferred_remove_flag(struct mapped_device *md)
2472 return test_bit(DMF_DEFERRED_REMOVE, &md->flags);
2475 int dm_suspended(struct dm_target *ti)
2477 return dm_suspended_md(dm_table_get_md(ti->table));
2479 EXPORT_SYMBOL_GPL(dm_suspended);
2481 int dm_noflush_suspending(struct dm_target *ti)
2483 return __noflush_suspending(dm_table_get_md(ti->table));
2485 EXPORT_SYMBOL_GPL(dm_noflush_suspending);
2487 struct dm_md_mempools *dm_alloc_md_mempools(struct mapped_device *md, unsigned type,
2488 unsigned integrity, unsigned per_io_data_size)
2490 struct dm_md_mempools *pools = kzalloc_node(sizeof(*pools), GFP_KERNEL, md->numa_node_id);
2491 struct kmem_cache *cachep = NULL;
2492 unsigned int pool_size = 0;
2493 unsigned int front_pad;
2499 case DM_TYPE_BIO_BASED:
2500 case DM_TYPE_DAX_BIO_BASED:
2502 pool_size = dm_get_reserved_bio_based_ios();
2503 front_pad = roundup(per_io_data_size, __alignof__(struct dm_target_io)) + offsetof(struct dm_target_io, clone);
2505 case DM_TYPE_REQUEST_BASED:
2506 cachep = _rq_tio_cache;
2507 pool_size = dm_get_reserved_rq_based_ios();
2508 pools->rq_pool = mempool_create_slab_pool(pool_size, _rq_cache);
2509 if (!pools->rq_pool)
2511 /* fall through to setup remaining rq-based pools */
2512 case DM_TYPE_MQ_REQUEST_BASED:
2514 pool_size = dm_get_reserved_rq_based_ios();
2515 front_pad = offsetof(struct dm_rq_clone_bio_info, clone);
2516 /* per_io_data_size is used for blk-mq pdu at queue allocation */
2523 pools->io_pool = mempool_create_slab_pool(pool_size, cachep);
2524 if (!pools->io_pool)
2528 pools->bs = bioset_create_nobvec(pool_size, front_pad);
2532 if (integrity && bioset_integrity_create(pools->bs, pool_size))
2538 dm_free_md_mempools(pools);
2543 void dm_free_md_mempools(struct dm_md_mempools *pools)
2548 mempool_destroy(pools->io_pool);
2549 mempool_destroy(pools->rq_pool);
2552 bioset_free(pools->bs);
2564 static int dm_call_pr(struct block_device *bdev, iterate_devices_callout_fn fn,
2567 struct mapped_device *md = bdev->bd_disk->private_data;
2568 struct dm_table *table;
2569 struct dm_target *ti;
2570 int ret = -ENOTTY, srcu_idx;
2572 table = dm_get_live_table(md, &srcu_idx);
2573 if (!table || !dm_table_get_size(table))
2576 /* We only support devices that have a single target */
2577 if (dm_table_get_num_targets(table) != 1)
2579 ti = dm_table_get_target(table, 0);
2582 if (!ti->type->iterate_devices)
2585 ret = ti->type->iterate_devices(ti, fn, data);
2587 dm_put_live_table(md, srcu_idx);
2592 * For register / unregister we need to manually call out to every path.
2594 static int __dm_pr_register(struct dm_target *ti, struct dm_dev *dev,
2595 sector_t start, sector_t len, void *data)
2597 struct dm_pr *pr = data;
2598 const struct pr_ops *ops = dev->bdev->bd_disk->fops->pr_ops;
2600 if (!ops || !ops->pr_register)
2602 return ops->pr_register(dev->bdev, pr->old_key, pr->new_key, pr->flags);
2605 static int dm_pr_register(struct block_device *bdev, u64 old_key, u64 new_key,
2616 ret = dm_call_pr(bdev, __dm_pr_register, &pr);
2617 if (ret && new_key) {
2618 /* unregister all paths if we failed to register any path */
2619 pr.old_key = new_key;
2622 pr.fail_early = false;
2623 dm_call_pr(bdev, __dm_pr_register, &pr);
2629 static int dm_pr_reserve(struct block_device *bdev, u64 key, enum pr_type type,
2632 struct mapped_device *md = bdev->bd_disk->private_data;
2633 const struct pr_ops *ops;
2637 r = dm_grab_bdev_for_ioctl(md, &bdev, &mode);
2641 ops = bdev->bd_disk->fops->pr_ops;
2642 if (ops && ops->pr_reserve)
2643 r = ops->pr_reserve(bdev, key, type, flags);
2651 static int dm_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
2653 struct mapped_device *md = bdev->bd_disk->private_data;
2654 const struct pr_ops *ops;
2658 r = dm_grab_bdev_for_ioctl(md, &bdev, &mode);
2662 ops = bdev->bd_disk->fops->pr_ops;
2663 if (ops && ops->pr_release)
2664 r = ops->pr_release(bdev, key, type);
2672 static int dm_pr_preempt(struct block_device *bdev, u64 old_key, u64 new_key,
2673 enum pr_type type, bool abort)
2675 struct mapped_device *md = bdev->bd_disk->private_data;
2676 const struct pr_ops *ops;
2680 r = dm_grab_bdev_for_ioctl(md, &bdev, &mode);
2684 ops = bdev->bd_disk->fops->pr_ops;
2685 if (ops && ops->pr_preempt)
2686 r = ops->pr_preempt(bdev, old_key, new_key, type, abort);
2694 static int dm_pr_clear(struct block_device *bdev, u64 key)
2696 struct mapped_device *md = bdev->bd_disk->private_data;
2697 const struct pr_ops *ops;
2701 r = dm_grab_bdev_for_ioctl(md, &bdev, &mode);
2705 ops = bdev->bd_disk->fops->pr_ops;
2706 if (ops && ops->pr_clear)
2707 r = ops->pr_clear(bdev, key);
2715 static const struct pr_ops dm_pr_ops = {
2716 .pr_register = dm_pr_register,
2717 .pr_reserve = dm_pr_reserve,
2718 .pr_release = dm_pr_release,
2719 .pr_preempt = dm_pr_preempt,
2720 .pr_clear = dm_pr_clear,
2723 static const struct block_device_operations dm_blk_dops = {
2724 .open = dm_blk_open,
2725 .release = dm_blk_close,
2726 .ioctl = dm_blk_ioctl,
2727 .direct_access = dm_blk_direct_access,
2728 .getgeo = dm_blk_getgeo,
2729 .pr_ops = &dm_pr_ops,
2730 .owner = THIS_MODULE
2736 module_init(dm_init);
2737 module_exit(dm_exit);
2739 module_param(major, uint, 0);
2740 MODULE_PARM_DESC(major, "The major number of the device mapper");
2742 module_param(reserved_bio_based_ios, uint, S_IRUGO | S_IWUSR);
2743 MODULE_PARM_DESC(reserved_bio_based_ios, "Reserved IOs in bio-based mempools");
2745 module_param(dm_numa_node, int, S_IRUGO | S_IWUSR);
2746 MODULE_PARM_DESC(dm_numa_node, "NUMA node for DM device memory allocations");
2748 MODULE_DESCRIPTION(DM_NAME " driver");
2749 MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
2750 MODULE_LICENSE("GPL");