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_rw & REQ_PREFLUSH) && bio->bi_iter.bi_size) {
803 * Preflush done for flush with data, reissue
804 * without REQ_PREFLUSH.
806 bio->bi_rw &= ~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);
909 * A target may call dm_accept_partial_bio only from the map routine. It is
910 * allowed for all bio types except REQ_PREFLUSH.
912 * dm_accept_partial_bio informs the dm that the target only wants to process
913 * additional n_sectors sectors of the bio and the rest of the data should be
914 * sent in a next bio.
916 * A diagram that explains the arithmetics:
917 * +--------------------+---------------+-------+
919 * +--------------------+---------------+-------+
921 * <-------------- *tio->len_ptr --------------->
922 * <------- bi_size ------->
925 * Region 1 was already iterated over with bio_advance or similar function.
926 * (it may be empty if the target doesn't use bio_advance)
927 * Region 2 is the remaining bio size that the target wants to process.
928 * (it may be empty if region 1 is non-empty, although there is no reason
930 * The target requires that region 3 is to be sent in the next bio.
932 * If the target wants to receive multiple copies of the bio (via num_*bios, etc),
933 * the partially processed part (the sum of regions 1+2) must be the same for all
936 void dm_accept_partial_bio(struct bio *bio, unsigned n_sectors)
938 struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone);
939 unsigned bi_size = bio->bi_iter.bi_size >> SECTOR_SHIFT;
940 BUG_ON(bio->bi_rw & REQ_PREFLUSH);
941 BUG_ON(bi_size > *tio->len_ptr);
942 BUG_ON(n_sectors > bi_size);
943 *tio->len_ptr -= bi_size - n_sectors;
944 bio->bi_iter.bi_size = n_sectors << SECTOR_SHIFT;
946 EXPORT_SYMBOL_GPL(dm_accept_partial_bio);
948 static void __map_bio(struct dm_target_io *tio)
952 struct bio *clone = &tio->clone;
953 struct dm_target *ti = tio->ti;
955 clone->bi_end_io = clone_endio;
958 * Map the clone. If r == 0 we don't need to do
959 * anything, the target has assumed ownership of
962 atomic_inc(&tio->io->io_count);
963 sector = clone->bi_iter.bi_sector;
964 r = ti->type->map(ti, clone);
965 if (r == DM_MAPIO_REMAPPED) {
966 /* the bio has been remapped so dispatch it */
968 trace_block_bio_remap(bdev_get_queue(clone->bi_bdev), clone,
969 tio->io->bio->bi_bdev->bd_dev, sector);
971 generic_make_request(clone);
972 } else if (r < 0 || r == DM_MAPIO_REQUEUE) {
973 /* error the io and bail out, or requeue it if needed */
974 dec_pending(tio->io, r);
976 } else if (r != DM_MAPIO_SUBMITTED) {
977 DMWARN("unimplemented target map return value: %d", r);
983 struct mapped_device *md;
984 struct dm_table *map;
988 unsigned sector_count;
991 static void bio_setup_sector(struct bio *bio, sector_t sector, unsigned len)
993 bio->bi_iter.bi_sector = sector;
994 bio->bi_iter.bi_size = to_bytes(len);
998 * Creates a bio that consists of range of complete bvecs.
1000 static int clone_bio(struct dm_target_io *tio, struct bio *bio,
1001 sector_t sector, unsigned len)
1003 struct bio *clone = &tio->clone;
1005 __bio_clone_fast(clone, bio);
1007 if (bio_integrity(bio)) {
1008 int r = bio_integrity_clone(clone, bio, GFP_NOIO);
1013 bio_advance(clone, to_bytes(sector - clone->bi_iter.bi_sector));
1014 clone->bi_iter.bi_size = to_bytes(len);
1016 if (bio_integrity(bio))
1017 bio_integrity_trim(clone, 0, len);
1022 static struct dm_target_io *alloc_tio(struct clone_info *ci,
1023 struct dm_target *ti,
1024 unsigned target_bio_nr)
1026 struct dm_target_io *tio;
1029 clone = bio_alloc_bioset(GFP_NOIO, 0, ci->md->bs);
1030 tio = container_of(clone, struct dm_target_io, clone);
1034 tio->target_bio_nr = target_bio_nr;
1039 static void __clone_and_map_simple_bio(struct clone_info *ci,
1040 struct dm_target *ti,
1041 unsigned target_bio_nr, unsigned *len)
1043 struct dm_target_io *tio = alloc_tio(ci, ti, target_bio_nr);
1044 struct bio *clone = &tio->clone;
1048 __bio_clone_fast(clone, ci->bio);
1050 bio_setup_sector(clone, ci->sector, *len);
1055 static void __send_duplicate_bios(struct clone_info *ci, struct dm_target *ti,
1056 unsigned num_bios, unsigned *len)
1058 unsigned target_bio_nr;
1060 for (target_bio_nr = 0; target_bio_nr < num_bios; target_bio_nr++)
1061 __clone_and_map_simple_bio(ci, ti, target_bio_nr, len);
1064 static int __send_empty_flush(struct clone_info *ci)
1066 unsigned target_nr = 0;
1067 struct dm_target *ti;
1069 BUG_ON(bio_has_data(ci->bio));
1070 while ((ti = dm_table_get_target(ci->map, target_nr++)))
1071 __send_duplicate_bios(ci, ti, ti->num_flush_bios, NULL);
1076 static int __clone_and_map_data_bio(struct clone_info *ci, struct dm_target *ti,
1077 sector_t sector, unsigned *len)
1079 struct bio *bio = ci->bio;
1080 struct dm_target_io *tio;
1081 unsigned target_bio_nr;
1082 unsigned num_target_bios = 1;
1086 * Does the target want to receive duplicate copies of the bio?
1088 if (bio_data_dir(bio) == WRITE && ti->num_write_bios)
1089 num_target_bios = ti->num_write_bios(ti, bio);
1091 for (target_bio_nr = 0; target_bio_nr < num_target_bios; target_bio_nr++) {
1092 tio = alloc_tio(ci, ti, target_bio_nr);
1094 r = clone_bio(tio, bio, sector, *len);
1105 typedef unsigned (*get_num_bios_fn)(struct dm_target *ti);
1107 static unsigned get_num_discard_bios(struct dm_target *ti)
1109 return ti->num_discard_bios;
1112 static unsigned get_num_write_same_bios(struct dm_target *ti)
1114 return ti->num_write_same_bios;
1117 typedef bool (*is_split_required_fn)(struct dm_target *ti);
1119 static bool is_split_required_for_discard(struct dm_target *ti)
1121 return ti->split_discard_bios;
1124 static int __send_changing_extent_only(struct clone_info *ci,
1125 get_num_bios_fn get_num_bios,
1126 is_split_required_fn is_split_required)
1128 struct dm_target *ti;
1133 ti = dm_table_find_target(ci->map, ci->sector);
1134 if (!dm_target_is_valid(ti))
1138 * Even though the device advertised support for this type of
1139 * request, that does not mean every target supports it, and
1140 * reconfiguration might also have changed that since the
1141 * check was performed.
1143 num_bios = get_num_bios ? get_num_bios(ti) : 0;
1147 if (is_split_required && !is_split_required(ti))
1148 len = min((sector_t)ci->sector_count, max_io_len_target_boundary(ci->sector, ti));
1150 len = min((sector_t)ci->sector_count, max_io_len(ci->sector, ti));
1152 __send_duplicate_bios(ci, ti, num_bios, &len);
1155 } while (ci->sector_count -= len);
1160 static int __send_discard(struct clone_info *ci)
1162 return __send_changing_extent_only(ci, get_num_discard_bios,
1163 is_split_required_for_discard);
1166 static int __send_write_same(struct clone_info *ci)
1168 return __send_changing_extent_only(ci, get_num_write_same_bios, NULL);
1172 * Select the correct strategy for processing a non-flush bio.
1174 static int __split_and_process_non_flush(struct clone_info *ci)
1176 struct bio *bio = ci->bio;
1177 struct dm_target *ti;
1181 if (unlikely(bio_op(bio) == REQ_OP_DISCARD))
1182 return __send_discard(ci);
1183 else if (unlikely(bio_op(bio) == REQ_OP_WRITE_SAME))
1184 return __send_write_same(ci);
1186 ti = dm_table_find_target(ci->map, ci->sector);
1187 if (!dm_target_is_valid(ti))
1190 len = min_t(sector_t, max_io_len(ci->sector, ti), ci->sector_count);
1192 r = __clone_and_map_data_bio(ci, ti, ci->sector, &len);
1197 ci->sector_count -= len;
1203 * Entry point to split a bio into clones and submit them to the targets.
1205 static void __split_and_process_bio(struct mapped_device *md,
1206 struct dm_table *map, struct bio *bio)
1208 struct clone_info ci;
1211 if (unlikely(!map)) {
1218 ci.io = alloc_io(md);
1220 atomic_set(&ci.io->io_count, 1);
1223 spin_lock_init(&ci.io->endio_lock);
1224 ci.sector = bio->bi_iter.bi_sector;
1226 start_io_acct(ci.io);
1228 if (bio->bi_rw & REQ_PREFLUSH) {
1229 ci.bio = &ci.md->flush_bio;
1230 ci.sector_count = 0;
1231 error = __send_empty_flush(&ci);
1232 /* dec_pending submits any data associated with flush */
1235 ci.sector_count = bio_sectors(bio);
1236 while (ci.sector_count && !error)
1237 error = __split_and_process_non_flush(&ci);
1240 /* drop the extra reference count */
1241 dec_pending(ci.io, error);
1243 /*-----------------------------------------------------------------
1245 *---------------------------------------------------------------*/
1248 * The request function that just remaps the bio built up by
1251 static blk_qc_t dm_make_request(struct request_queue *q, struct bio *bio)
1253 int rw = bio_data_dir(bio);
1254 struct mapped_device *md = q->queuedata;
1256 struct dm_table *map;
1258 map = dm_get_live_table(md, &srcu_idx);
1260 generic_start_io_acct(rw, bio_sectors(bio), &dm_disk(md)->part0);
1262 /* if we're suspended, we have to queue this io for later */
1263 if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags))) {
1264 dm_put_live_table(md, srcu_idx);
1266 if (bio_rw(bio) != READA)
1270 return BLK_QC_T_NONE;
1273 __split_and_process_bio(md, map, bio);
1274 dm_put_live_table(md, srcu_idx);
1275 return BLK_QC_T_NONE;
1278 static int dm_any_congested(void *congested_data, int bdi_bits)
1281 struct mapped_device *md = congested_data;
1282 struct dm_table *map;
1284 if (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
1285 if (dm_request_based(md)) {
1287 * With request-based DM we only need to check the
1288 * top-level queue for congestion.
1290 r = md->queue->backing_dev_info.wb.state & bdi_bits;
1292 map = dm_get_live_table_fast(md);
1294 r = dm_table_any_congested(map, bdi_bits);
1295 dm_put_live_table_fast(md);
1302 /*-----------------------------------------------------------------
1303 * An IDR is used to keep track of allocated minor numbers.
1304 *---------------------------------------------------------------*/
1305 static void free_minor(int minor)
1307 spin_lock(&_minor_lock);
1308 idr_remove(&_minor_idr, minor);
1309 spin_unlock(&_minor_lock);
1313 * See if the device with a specific minor # is free.
1315 static int specific_minor(int minor)
1319 if (minor >= (1 << MINORBITS))
1322 idr_preload(GFP_KERNEL);
1323 spin_lock(&_minor_lock);
1325 r = idr_alloc(&_minor_idr, MINOR_ALLOCED, minor, minor + 1, GFP_NOWAIT);
1327 spin_unlock(&_minor_lock);
1330 return r == -ENOSPC ? -EBUSY : r;
1334 static int next_free_minor(int *minor)
1338 idr_preload(GFP_KERNEL);
1339 spin_lock(&_minor_lock);
1341 r = idr_alloc(&_minor_idr, MINOR_ALLOCED, 0, 1 << MINORBITS, GFP_NOWAIT);
1343 spin_unlock(&_minor_lock);
1351 static const struct block_device_operations dm_blk_dops;
1353 static void dm_wq_work(struct work_struct *work);
1355 void dm_init_md_queue(struct mapped_device *md)
1358 * Request-based dm devices cannot be stacked on top of bio-based dm
1359 * devices. The type of this dm device may not have been decided yet.
1360 * The type is decided at the first table loading time.
1361 * To prevent problematic device stacking, clear the queue flag
1362 * for request stacking support until then.
1364 * This queue is new, so no concurrency on the queue_flags.
1366 queue_flag_clear_unlocked(QUEUE_FLAG_STACKABLE, md->queue);
1369 * Initialize data that will only be used by a non-blk-mq DM queue
1370 * - must do so here (in alloc_dev callchain) before queue is used
1372 md->queue->queuedata = md;
1373 md->queue->backing_dev_info.congested_data = md;
1376 void dm_init_normal_md_queue(struct mapped_device *md)
1378 md->use_blk_mq = false;
1379 dm_init_md_queue(md);
1382 * Initialize aspects of queue that aren't relevant for blk-mq
1384 md->queue->backing_dev_info.congested_fn = dm_any_congested;
1385 blk_queue_bounce_limit(md->queue, BLK_BOUNCE_ANY);
1388 static void cleanup_mapped_device(struct mapped_device *md)
1391 destroy_workqueue(md->wq);
1392 if (md->kworker_task)
1393 kthread_stop(md->kworker_task);
1394 mempool_destroy(md->io_pool);
1395 mempool_destroy(md->rq_pool);
1397 bioset_free(md->bs);
1399 cleanup_srcu_struct(&md->io_barrier);
1402 spin_lock(&_minor_lock);
1403 md->disk->private_data = NULL;
1404 spin_unlock(&_minor_lock);
1405 del_gendisk(md->disk);
1410 blk_cleanup_queue(md->queue);
1417 dm_mq_cleanup_mapped_device(md);
1421 * Allocate and initialise a blank device with a given minor.
1423 static struct mapped_device *alloc_dev(int minor)
1425 int r, numa_node_id = dm_get_numa_node();
1426 struct mapped_device *md;
1429 md = kzalloc_node(sizeof(*md), GFP_KERNEL, numa_node_id);
1431 DMWARN("unable to allocate device, out of memory.");
1435 if (!try_module_get(THIS_MODULE))
1436 goto bad_module_get;
1438 /* get a minor number for the dev */
1439 if (minor == DM_ANY_MINOR)
1440 r = next_free_minor(&minor);
1442 r = specific_minor(minor);
1446 r = init_srcu_struct(&md->io_barrier);
1448 goto bad_io_barrier;
1450 md->numa_node_id = numa_node_id;
1451 md->use_blk_mq = dm_use_blk_mq_default();
1452 md->init_tio_pdu = false;
1453 md->type = DM_TYPE_NONE;
1454 mutex_init(&md->suspend_lock);
1455 mutex_init(&md->type_lock);
1456 mutex_init(&md->table_devices_lock);
1457 spin_lock_init(&md->deferred_lock);
1458 atomic_set(&md->holders, 1);
1459 atomic_set(&md->open_count, 0);
1460 atomic_set(&md->event_nr, 0);
1461 atomic_set(&md->uevent_seq, 0);
1462 INIT_LIST_HEAD(&md->uevent_list);
1463 INIT_LIST_HEAD(&md->table_devices);
1464 spin_lock_init(&md->uevent_lock);
1466 md->queue = blk_alloc_queue_node(GFP_KERNEL, numa_node_id);
1470 dm_init_md_queue(md);
1472 md->disk = alloc_disk_node(1, numa_node_id);
1476 atomic_set(&md->pending[0], 0);
1477 atomic_set(&md->pending[1], 0);
1478 init_waitqueue_head(&md->wait);
1479 INIT_WORK(&md->work, dm_wq_work);
1480 init_waitqueue_head(&md->eventq);
1481 init_completion(&md->kobj_holder.completion);
1482 md->kworker_task = NULL;
1484 md->disk->major = _major;
1485 md->disk->first_minor = minor;
1486 md->disk->fops = &dm_blk_dops;
1487 md->disk->queue = md->queue;
1488 md->disk->private_data = md;
1489 sprintf(md->disk->disk_name, "dm-%d", minor);
1491 format_dev_t(md->name, MKDEV(_major, minor));
1493 md->wq = alloc_workqueue("kdmflush", WQ_MEM_RECLAIM, 0);
1497 md->bdev = bdget_disk(md->disk, 0);
1501 bio_init(&md->flush_bio);
1502 md->flush_bio.bi_bdev = md->bdev;
1503 bio_set_op_attrs(&md->flush_bio, REQ_OP_WRITE, WRITE_FLUSH);
1505 dm_stats_init(&md->stats);
1507 /* Populate the mapping, nobody knows we exist yet */
1508 spin_lock(&_minor_lock);
1509 old_md = idr_replace(&_minor_idr, md, minor);
1510 spin_unlock(&_minor_lock);
1512 BUG_ON(old_md != MINOR_ALLOCED);
1517 cleanup_mapped_device(md);
1521 module_put(THIS_MODULE);
1527 static void unlock_fs(struct mapped_device *md);
1529 static void free_dev(struct mapped_device *md)
1531 int minor = MINOR(disk_devt(md->disk));
1535 cleanup_mapped_device(md);
1537 free_table_devices(&md->table_devices);
1538 dm_stats_cleanup(&md->stats);
1541 module_put(THIS_MODULE);
1545 static void __bind_mempools(struct mapped_device *md, struct dm_table *t)
1547 struct dm_md_mempools *p = dm_table_get_md_mempools(t);
1550 /* The md already has necessary mempools. */
1551 if (dm_table_get_type(t) == DM_TYPE_BIO_BASED) {
1553 * Reload bioset because front_pad may have changed
1554 * because a different table was loaded.
1556 bioset_free(md->bs);
1561 * There's no need to reload with request-based dm
1562 * because the size of front_pad doesn't change.
1563 * Note for future: If you are to reload bioset,
1564 * prep-ed requests in the queue may refer
1565 * to bio from the old bioset, so you must walk
1566 * through the queue to unprep.
1571 BUG_ON(!p || md->io_pool || md->rq_pool || md->bs);
1573 md->io_pool = p->io_pool;
1575 md->rq_pool = p->rq_pool;
1581 /* mempool bind completed, no longer need any mempools in the table */
1582 dm_table_free_md_mempools(t);
1586 * Bind a table to the device.
1588 static void event_callback(void *context)
1590 unsigned long flags;
1592 struct mapped_device *md = (struct mapped_device *) context;
1594 spin_lock_irqsave(&md->uevent_lock, flags);
1595 list_splice_init(&md->uevent_list, &uevents);
1596 spin_unlock_irqrestore(&md->uevent_lock, flags);
1598 dm_send_uevents(&uevents, &disk_to_dev(md->disk)->kobj);
1600 atomic_inc(&md->event_nr);
1601 wake_up(&md->eventq);
1605 * Protected by md->suspend_lock obtained by dm_swap_table().
1607 static void __set_size(struct mapped_device *md, sector_t size)
1609 set_capacity(md->disk, size);
1611 i_size_write(md->bdev->bd_inode, (loff_t)size << SECTOR_SHIFT);
1615 * Returns old map, which caller must destroy.
1617 static struct dm_table *__bind(struct mapped_device *md, struct dm_table *t,
1618 struct queue_limits *limits)
1620 struct dm_table *old_map;
1621 struct request_queue *q = md->queue;
1624 size = dm_table_get_size(t);
1627 * Wipe any geometry if the size of the table changed.
1629 if (size != dm_get_size(md))
1630 memset(&md->geometry, 0, sizeof(md->geometry));
1632 __set_size(md, size);
1634 dm_table_event_callback(t, event_callback, md);
1637 * The queue hasn't been stopped yet, if the old table type wasn't
1638 * for request-based during suspension. So stop it to prevent
1639 * I/O mapping before resume.
1640 * This must be done before setting the queue restrictions,
1641 * because request-based dm may be run just after the setting.
1643 if (dm_table_request_based(t)) {
1646 * Leverage the fact that request-based DM targets are
1647 * immutable singletons and establish md->immutable_target
1648 * - used to optimize both dm_request_fn and dm_mq_queue_rq
1650 md->immutable_target = dm_table_get_immutable_target(t);
1653 __bind_mempools(md, t);
1655 old_map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
1656 rcu_assign_pointer(md->map, (void *)t);
1657 md->immutable_target_type = dm_table_get_immutable_target_type(t);
1659 dm_table_set_restrictions(t, q, limits);
1667 * Returns unbound table for the caller to free.
1669 static struct dm_table *__unbind(struct mapped_device *md)
1671 struct dm_table *map = rcu_dereference_protected(md->map, 1);
1676 dm_table_event_callback(map, NULL, NULL);
1677 RCU_INIT_POINTER(md->map, NULL);
1684 * Constructor for a new device.
1686 int dm_create(int minor, struct mapped_device **result)
1688 struct mapped_device *md;
1690 md = alloc_dev(minor);
1701 * Functions to manage md->type.
1702 * All are required to hold md->type_lock.
1704 void dm_lock_md_type(struct mapped_device *md)
1706 mutex_lock(&md->type_lock);
1709 void dm_unlock_md_type(struct mapped_device *md)
1711 mutex_unlock(&md->type_lock);
1714 void dm_set_md_type(struct mapped_device *md, unsigned type)
1716 BUG_ON(!mutex_is_locked(&md->type_lock));
1720 unsigned dm_get_md_type(struct mapped_device *md)
1725 struct target_type *dm_get_immutable_target_type(struct mapped_device *md)
1727 return md->immutable_target_type;
1731 * The queue_limits are only valid as long as you have a reference
1734 struct queue_limits *dm_get_queue_limits(struct mapped_device *md)
1736 BUG_ON(!atomic_read(&md->holders));
1737 return &md->queue->limits;
1739 EXPORT_SYMBOL_GPL(dm_get_queue_limits);
1742 * Setup the DM device's queue based on md's type
1744 int dm_setup_md_queue(struct mapped_device *md, struct dm_table *t)
1748 switch (dm_get_md_type(md)) {
1749 case DM_TYPE_REQUEST_BASED:
1750 r = dm_old_init_request_queue(md);
1752 DMERR("Cannot initialize queue for request-based mapped device");
1756 case DM_TYPE_MQ_REQUEST_BASED:
1757 r = dm_mq_init_request_queue(md, t);
1759 DMERR("Cannot initialize queue for request-based dm-mq mapped device");
1763 case DM_TYPE_BIO_BASED:
1764 dm_init_normal_md_queue(md);
1765 blk_queue_make_request(md->queue, dm_make_request);
1767 * DM handles splitting bios as needed. Free the bio_split bioset
1768 * since it won't be used (saves 1 process per bio-based DM device).
1770 bioset_free(md->queue->bio_split);
1771 md->queue->bio_split = NULL;
1778 struct mapped_device *dm_get_md(dev_t dev)
1780 struct mapped_device *md;
1781 unsigned minor = MINOR(dev);
1783 if (MAJOR(dev) != _major || minor >= (1 << MINORBITS))
1786 spin_lock(&_minor_lock);
1788 md = idr_find(&_minor_idr, minor);
1790 if ((md == MINOR_ALLOCED ||
1791 (MINOR(disk_devt(dm_disk(md))) != minor) ||
1792 dm_deleting_md(md) ||
1793 test_bit(DMF_FREEING, &md->flags))) {
1801 spin_unlock(&_minor_lock);
1805 EXPORT_SYMBOL_GPL(dm_get_md);
1807 void *dm_get_mdptr(struct mapped_device *md)
1809 return md->interface_ptr;
1812 void dm_set_mdptr(struct mapped_device *md, void *ptr)
1814 md->interface_ptr = ptr;
1817 void dm_get(struct mapped_device *md)
1819 atomic_inc(&md->holders);
1820 BUG_ON(test_bit(DMF_FREEING, &md->flags));
1823 int dm_hold(struct mapped_device *md)
1825 spin_lock(&_minor_lock);
1826 if (test_bit(DMF_FREEING, &md->flags)) {
1827 spin_unlock(&_minor_lock);
1831 spin_unlock(&_minor_lock);
1834 EXPORT_SYMBOL_GPL(dm_hold);
1836 const char *dm_device_name(struct mapped_device *md)
1840 EXPORT_SYMBOL_GPL(dm_device_name);
1842 static void __dm_destroy(struct mapped_device *md, bool wait)
1844 struct dm_table *map;
1849 spin_lock(&_minor_lock);
1850 idr_replace(&_minor_idr, MINOR_ALLOCED, MINOR(disk_devt(dm_disk(md))));
1851 set_bit(DMF_FREEING, &md->flags);
1852 spin_unlock(&_minor_lock);
1854 if (dm_request_based(md) && md->kworker_task)
1855 flush_kthread_worker(&md->kworker);
1858 * Take suspend_lock so that presuspend and postsuspend methods
1859 * do not race with internal suspend.
1861 mutex_lock(&md->suspend_lock);
1862 map = dm_get_live_table(md, &srcu_idx);
1863 if (!dm_suspended_md(md)) {
1864 dm_table_presuspend_targets(map);
1865 dm_table_postsuspend_targets(map);
1867 /* dm_put_live_table must be before msleep, otherwise deadlock is possible */
1868 dm_put_live_table(md, srcu_idx);
1869 mutex_unlock(&md->suspend_lock);
1872 * Rare, but there may be I/O requests still going to complete,
1873 * for example. Wait for all references to disappear.
1874 * No one should increment the reference count of the mapped_device,
1875 * after the mapped_device state becomes DMF_FREEING.
1878 while (atomic_read(&md->holders))
1880 else if (atomic_read(&md->holders))
1881 DMWARN("%s: Forcibly removing mapped_device still in use! (%d users)",
1882 dm_device_name(md), atomic_read(&md->holders));
1885 dm_table_destroy(__unbind(md));
1889 void dm_destroy(struct mapped_device *md)
1891 __dm_destroy(md, true);
1894 void dm_destroy_immediate(struct mapped_device *md)
1896 __dm_destroy(md, false);
1899 void dm_put(struct mapped_device *md)
1901 atomic_dec(&md->holders);
1903 EXPORT_SYMBOL_GPL(dm_put);
1905 static int dm_wait_for_completion(struct mapped_device *md, int interruptible)
1908 DECLARE_WAITQUEUE(wait, current);
1910 add_wait_queue(&md->wait, &wait);
1913 set_current_state(interruptible);
1915 if (!md_in_flight(md))
1918 if (interruptible == TASK_INTERRUPTIBLE &&
1919 signal_pending(current)) {
1926 set_current_state(TASK_RUNNING);
1928 remove_wait_queue(&md->wait, &wait);
1934 * Process the deferred bios
1936 static void dm_wq_work(struct work_struct *work)
1938 struct mapped_device *md = container_of(work, struct mapped_device,
1942 struct dm_table *map;
1944 map = dm_get_live_table(md, &srcu_idx);
1946 while (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
1947 spin_lock_irq(&md->deferred_lock);
1948 c = bio_list_pop(&md->deferred);
1949 spin_unlock_irq(&md->deferred_lock);
1954 if (dm_request_based(md))
1955 generic_make_request(c);
1957 __split_and_process_bio(md, map, c);
1960 dm_put_live_table(md, srcu_idx);
1963 static void dm_queue_flush(struct mapped_device *md)
1965 clear_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
1966 smp_mb__after_atomic();
1967 queue_work(md->wq, &md->work);
1971 * Swap in a new table, returning the old one for the caller to destroy.
1973 struct dm_table *dm_swap_table(struct mapped_device *md, struct dm_table *table)
1975 struct dm_table *live_map = NULL, *map = ERR_PTR(-EINVAL);
1976 struct queue_limits limits;
1979 mutex_lock(&md->suspend_lock);
1981 /* device must be suspended */
1982 if (!dm_suspended_md(md))
1986 * If the new table has no data devices, retain the existing limits.
1987 * This helps multipath with queue_if_no_path if all paths disappear,
1988 * then new I/O is queued based on these limits, and then some paths
1991 if (dm_table_has_no_data_devices(table)) {
1992 live_map = dm_get_live_table_fast(md);
1994 limits = md->queue->limits;
1995 dm_put_live_table_fast(md);
1999 r = dm_calculate_queue_limits(table, &limits);
2006 map = __bind(md, table, &limits);
2009 mutex_unlock(&md->suspend_lock);
2014 * Functions to lock and unlock any filesystem running on the
2017 static int lock_fs(struct mapped_device *md)
2021 WARN_ON(md->frozen_sb);
2023 md->frozen_sb = freeze_bdev(md->bdev);
2024 if (IS_ERR(md->frozen_sb)) {
2025 r = PTR_ERR(md->frozen_sb);
2026 md->frozen_sb = NULL;
2030 set_bit(DMF_FROZEN, &md->flags);
2035 static void unlock_fs(struct mapped_device *md)
2037 if (!test_bit(DMF_FROZEN, &md->flags))
2040 thaw_bdev(md->bdev, md->frozen_sb);
2041 md->frozen_sb = NULL;
2042 clear_bit(DMF_FROZEN, &md->flags);
2046 * If __dm_suspend returns 0, the device is completely quiescent
2047 * now. There is no request-processing activity. All new requests
2048 * are being added to md->deferred list.
2050 * Caller must hold md->suspend_lock
2052 static int __dm_suspend(struct mapped_device *md, struct dm_table *map,
2053 unsigned suspend_flags, int interruptible)
2055 bool do_lockfs = suspend_flags & DM_SUSPEND_LOCKFS_FLAG;
2056 bool noflush = suspend_flags & DM_SUSPEND_NOFLUSH_FLAG;
2060 * DMF_NOFLUSH_SUSPENDING must be set before presuspend.
2061 * This flag is cleared before dm_suspend returns.
2064 set_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
2067 * This gets reverted if there's an error later and the targets
2068 * provide the .presuspend_undo hook.
2070 dm_table_presuspend_targets(map);
2073 * Flush I/O to the device.
2074 * Any I/O submitted after lock_fs() may not be flushed.
2075 * noflush takes precedence over do_lockfs.
2076 * (lock_fs() flushes I/Os and waits for them to complete.)
2078 if (!noflush && do_lockfs) {
2081 dm_table_presuspend_undo_targets(map);
2087 * Here we must make sure that no processes are submitting requests
2088 * to target drivers i.e. no one may be executing
2089 * __split_and_process_bio. This is called from dm_request and
2092 * To get all processes out of __split_and_process_bio in dm_request,
2093 * we take the write lock. To prevent any process from reentering
2094 * __split_and_process_bio from dm_request and quiesce the thread
2095 * (dm_wq_work), we set BMF_BLOCK_IO_FOR_SUSPEND and call
2096 * flush_workqueue(md->wq).
2098 set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2100 synchronize_srcu(&md->io_barrier);
2103 * Stop md->queue before flushing md->wq in case request-based
2104 * dm defers requests to md->wq from md->queue.
2106 if (dm_request_based(md)) {
2107 dm_stop_queue(md->queue);
2108 if (md->kworker_task)
2109 flush_kthread_worker(&md->kworker);
2112 flush_workqueue(md->wq);
2115 * At this point no more requests are entering target request routines.
2116 * We call dm_wait_for_completion to wait for all existing requests
2119 r = dm_wait_for_completion(md, interruptible);
2122 clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
2124 synchronize_srcu(&md->io_barrier);
2126 /* were we interrupted ? */
2130 if (dm_request_based(md))
2131 dm_start_queue(md->queue);
2134 dm_table_presuspend_undo_targets(map);
2135 /* pushback list is already flushed, so skip flush */
2142 * We need to be able to change a mapping table under a mounted
2143 * filesystem. For example we might want to move some data in
2144 * the background. Before the table can be swapped with
2145 * dm_bind_table, dm_suspend must be called to flush any in
2146 * flight bios and ensure that any further io gets deferred.
2149 * Suspend mechanism in request-based dm.
2151 * 1. Flush all I/Os by lock_fs() if needed.
2152 * 2. Stop dispatching any I/O by stopping the request_queue.
2153 * 3. Wait for all in-flight I/Os to be completed or requeued.
2155 * To abort suspend, start the request_queue.
2157 int dm_suspend(struct mapped_device *md, unsigned suspend_flags)
2159 struct dm_table *map = NULL;
2163 mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING);
2165 if (dm_suspended_md(md)) {
2170 if (dm_suspended_internally_md(md)) {
2171 /* already internally suspended, wait for internal resume */
2172 mutex_unlock(&md->suspend_lock);
2173 r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE);
2179 map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2181 r = __dm_suspend(md, map, suspend_flags, TASK_INTERRUPTIBLE);
2185 set_bit(DMF_SUSPENDED, &md->flags);
2187 dm_table_postsuspend_targets(map);
2190 mutex_unlock(&md->suspend_lock);
2194 static int __dm_resume(struct mapped_device *md, struct dm_table *map)
2197 int r = dm_table_resume_targets(map);
2205 * Flushing deferred I/Os must be done after targets are resumed
2206 * so that mapping of targets can work correctly.
2207 * Request-based dm is queueing the deferred I/Os in its request_queue.
2209 if (dm_request_based(md))
2210 dm_start_queue(md->queue);
2217 int dm_resume(struct mapped_device *md)
2220 struct dm_table *map = NULL;
2223 mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING);
2225 if (!dm_suspended_md(md))
2228 if (dm_suspended_internally_md(md)) {
2229 /* already internally suspended, wait for internal resume */
2230 mutex_unlock(&md->suspend_lock);
2231 r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE);
2237 map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2238 if (!map || !dm_table_get_size(map))
2241 r = __dm_resume(md, map);
2245 clear_bit(DMF_SUSPENDED, &md->flags);
2249 mutex_unlock(&md->suspend_lock);
2255 * Internal suspend/resume works like userspace-driven suspend. It waits
2256 * until all bios finish and prevents issuing new bios to the target drivers.
2257 * It may be used only from the kernel.
2260 static void __dm_internal_suspend(struct mapped_device *md, unsigned suspend_flags)
2262 struct dm_table *map = NULL;
2264 if (md->internal_suspend_count++)
2265 return; /* nested internal suspend */
2267 if (dm_suspended_md(md)) {
2268 set_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
2269 return; /* nest suspend */
2272 map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2275 * Using TASK_UNINTERRUPTIBLE because only NOFLUSH internal suspend is
2276 * supported. Properly supporting a TASK_INTERRUPTIBLE internal suspend
2277 * would require changing .presuspend to return an error -- avoid this
2278 * until there is a need for more elaborate variants of internal suspend.
2280 (void) __dm_suspend(md, map, suspend_flags, TASK_UNINTERRUPTIBLE);
2282 set_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
2284 dm_table_postsuspend_targets(map);
2287 static void __dm_internal_resume(struct mapped_device *md)
2289 BUG_ON(!md->internal_suspend_count);
2291 if (--md->internal_suspend_count)
2292 return; /* resume from nested internal suspend */
2294 if (dm_suspended_md(md))
2295 goto done; /* resume from nested suspend */
2298 * NOTE: existing callers don't need to call dm_table_resume_targets
2299 * (which may fail -- so best to avoid it for now by passing NULL map)
2301 (void) __dm_resume(md, NULL);
2304 clear_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
2305 smp_mb__after_atomic();
2306 wake_up_bit(&md->flags, DMF_SUSPENDED_INTERNALLY);
2309 void dm_internal_suspend_noflush(struct mapped_device *md)
2311 mutex_lock(&md->suspend_lock);
2312 __dm_internal_suspend(md, DM_SUSPEND_NOFLUSH_FLAG);
2313 mutex_unlock(&md->suspend_lock);
2315 EXPORT_SYMBOL_GPL(dm_internal_suspend_noflush);
2317 void dm_internal_resume(struct mapped_device *md)
2319 mutex_lock(&md->suspend_lock);
2320 __dm_internal_resume(md);
2321 mutex_unlock(&md->suspend_lock);
2323 EXPORT_SYMBOL_GPL(dm_internal_resume);
2326 * Fast variants of internal suspend/resume hold md->suspend_lock,
2327 * which prevents interaction with userspace-driven suspend.
2330 void dm_internal_suspend_fast(struct mapped_device *md)
2332 mutex_lock(&md->suspend_lock);
2333 if (dm_suspended_md(md) || dm_suspended_internally_md(md))
2336 set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2337 synchronize_srcu(&md->io_barrier);
2338 flush_workqueue(md->wq);
2339 dm_wait_for_completion(md, TASK_UNINTERRUPTIBLE);
2341 EXPORT_SYMBOL_GPL(dm_internal_suspend_fast);
2343 void dm_internal_resume_fast(struct mapped_device *md)
2345 if (dm_suspended_md(md) || dm_suspended_internally_md(md))
2351 mutex_unlock(&md->suspend_lock);
2353 EXPORT_SYMBOL_GPL(dm_internal_resume_fast);
2355 /*-----------------------------------------------------------------
2356 * Event notification.
2357 *---------------------------------------------------------------*/
2358 int dm_kobject_uevent(struct mapped_device *md, enum kobject_action action,
2361 char udev_cookie[DM_COOKIE_LENGTH];
2362 char *envp[] = { udev_cookie, NULL };
2365 return kobject_uevent(&disk_to_dev(md->disk)->kobj, action);
2367 snprintf(udev_cookie, DM_COOKIE_LENGTH, "%s=%u",
2368 DM_COOKIE_ENV_VAR_NAME, cookie);
2369 return kobject_uevent_env(&disk_to_dev(md->disk)->kobj,
2374 uint32_t dm_next_uevent_seq(struct mapped_device *md)
2376 return atomic_add_return(1, &md->uevent_seq);
2379 uint32_t dm_get_event_nr(struct mapped_device *md)
2381 return atomic_read(&md->event_nr);
2384 int dm_wait_event(struct mapped_device *md, int event_nr)
2386 return wait_event_interruptible(md->eventq,
2387 (event_nr != atomic_read(&md->event_nr)));
2390 void dm_uevent_add(struct mapped_device *md, struct list_head *elist)
2392 unsigned long flags;
2394 spin_lock_irqsave(&md->uevent_lock, flags);
2395 list_add(elist, &md->uevent_list);
2396 spin_unlock_irqrestore(&md->uevent_lock, flags);
2400 * The gendisk is only valid as long as you have a reference
2403 struct gendisk *dm_disk(struct mapped_device *md)
2407 EXPORT_SYMBOL_GPL(dm_disk);
2409 struct kobject *dm_kobject(struct mapped_device *md)
2411 return &md->kobj_holder.kobj;
2414 struct mapped_device *dm_get_from_kobject(struct kobject *kobj)
2416 struct mapped_device *md;
2418 md = container_of(kobj, struct mapped_device, kobj_holder.kobj);
2420 if (test_bit(DMF_FREEING, &md->flags) ||
2428 int dm_suspended_md(struct mapped_device *md)
2430 return test_bit(DMF_SUSPENDED, &md->flags);
2433 int dm_suspended_internally_md(struct mapped_device *md)
2435 return test_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
2438 int dm_test_deferred_remove_flag(struct mapped_device *md)
2440 return test_bit(DMF_DEFERRED_REMOVE, &md->flags);
2443 int dm_suspended(struct dm_target *ti)
2445 return dm_suspended_md(dm_table_get_md(ti->table));
2447 EXPORT_SYMBOL_GPL(dm_suspended);
2449 int dm_noflush_suspending(struct dm_target *ti)
2451 return __noflush_suspending(dm_table_get_md(ti->table));
2453 EXPORT_SYMBOL_GPL(dm_noflush_suspending);
2455 struct dm_md_mempools *dm_alloc_md_mempools(struct mapped_device *md, unsigned type,
2456 unsigned integrity, unsigned per_io_data_size)
2458 struct dm_md_mempools *pools = kzalloc_node(sizeof(*pools), GFP_KERNEL, md->numa_node_id);
2459 struct kmem_cache *cachep = NULL;
2460 unsigned int pool_size = 0;
2461 unsigned int front_pad;
2467 case DM_TYPE_BIO_BASED:
2469 pool_size = dm_get_reserved_bio_based_ios();
2470 front_pad = roundup(per_io_data_size, __alignof__(struct dm_target_io)) + offsetof(struct dm_target_io, clone);
2472 case DM_TYPE_REQUEST_BASED:
2473 cachep = _rq_tio_cache;
2474 pool_size = dm_get_reserved_rq_based_ios();
2475 pools->rq_pool = mempool_create_slab_pool(pool_size, _rq_cache);
2476 if (!pools->rq_pool)
2478 /* fall through to setup remaining rq-based pools */
2479 case DM_TYPE_MQ_REQUEST_BASED:
2481 pool_size = dm_get_reserved_rq_based_ios();
2482 front_pad = offsetof(struct dm_rq_clone_bio_info, clone);
2483 /* per_io_data_size is used for blk-mq pdu at queue allocation */
2490 pools->io_pool = mempool_create_slab_pool(pool_size, cachep);
2491 if (!pools->io_pool)
2495 pools->bs = bioset_create_nobvec(pool_size, front_pad);
2499 if (integrity && bioset_integrity_create(pools->bs, pool_size))
2505 dm_free_md_mempools(pools);
2510 void dm_free_md_mempools(struct dm_md_mempools *pools)
2515 mempool_destroy(pools->io_pool);
2516 mempool_destroy(pools->rq_pool);
2519 bioset_free(pools->bs);
2524 static int dm_pr_register(struct block_device *bdev, u64 old_key, u64 new_key,
2527 struct mapped_device *md = bdev->bd_disk->private_data;
2528 const struct pr_ops *ops;
2532 r = dm_grab_bdev_for_ioctl(md, &bdev, &mode);
2536 ops = bdev->bd_disk->fops->pr_ops;
2537 if (ops && ops->pr_register)
2538 r = ops->pr_register(bdev, old_key, new_key, flags);
2546 static int dm_pr_reserve(struct block_device *bdev, u64 key, enum pr_type type,
2549 struct mapped_device *md = bdev->bd_disk->private_data;
2550 const struct pr_ops *ops;
2554 r = dm_grab_bdev_for_ioctl(md, &bdev, &mode);
2558 ops = bdev->bd_disk->fops->pr_ops;
2559 if (ops && ops->pr_reserve)
2560 r = ops->pr_reserve(bdev, key, type, flags);
2568 static int dm_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
2570 struct mapped_device *md = bdev->bd_disk->private_data;
2571 const struct pr_ops *ops;
2575 r = dm_grab_bdev_for_ioctl(md, &bdev, &mode);
2579 ops = bdev->bd_disk->fops->pr_ops;
2580 if (ops && ops->pr_release)
2581 r = ops->pr_release(bdev, key, type);
2589 static int dm_pr_preempt(struct block_device *bdev, u64 old_key, u64 new_key,
2590 enum pr_type type, bool abort)
2592 struct mapped_device *md = bdev->bd_disk->private_data;
2593 const struct pr_ops *ops;
2597 r = dm_grab_bdev_for_ioctl(md, &bdev, &mode);
2601 ops = bdev->bd_disk->fops->pr_ops;
2602 if (ops && ops->pr_preempt)
2603 r = ops->pr_preempt(bdev, old_key, new_key, type, abort);
2611 static int dm_pr_clear(struct block_device *bdev, u64 key)
2613 struct mapped_device *md = bdev->bd_disk->private_data;
2614 const struct pr_ops *ops;
2618 r = dm_grab_bdev_for_ioctl(md, &bdev, &mode);
2622 ops = bdev->bd_disk->fops->pr_ops;
2623 if (ops && ops->pr_clear)
2624 r = ops->pr_clear(bdev, key);
2632 static const struct pr_ops dm_pr_ops = {
2633 .pr_register = dm_pr_register,
2634 .pr_reserve = dm_pr_reserve,
2635 .pr_release = dm_pr_release,
2636 .pr_preempt = dm_pr_preempt,
2637 .pr_clear = dm_pr_clear,
2640 static const struct block_device_operations dm_blk_dops = {
2641 .open = dm_blk_open,
2642 .release = dm_blk_close,
2643 .ioctl = dm_blk_ioctl,
2644 .getgeo = dm_blk_getgeo,
2645 .pr_ops = &dm_pr_ops,
2646 .owner = THIS_MODULE
2652 module_init(dm_init);
2653 module_exit(dm_exit);
2655 module_param(major, uint, 0);
2656 MODULE_PARM_DESC(major, "The major number of the device mapper");
2658 module_param(reserved_bio_based_ios, uint, S_IRUGO | S_IWUSR);
2659 MODULE_PARM_DESC(reserved_bio_based_ios, "Reserved IOs in bio-based mempools");
2661 module_param(dm_numa_node, int, S_IRUGO | S_IWUSR);
2662 MODULE_PARM_DESC(dm_numa_node, "NUMA node for DM device memory allocations");
2664 MODULE_DESCRIPTION(DM_NAME " driver");
2665 MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
2666 MODULE_LICENSE("GPL");