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);
976 * Flush current->bio_list when the target map method blocks.
977 * This fixes deadlocks in snapshot and possibly in other targets.
980 struct blk_plug plug;
981 struct blk_plug_cb cb;
984 static void flush_current_bio_list(struct blk_plug_cb *cb, bool from_schedule)
986 struct dm_offload *o = container_of(cb, struct dm_offload, cb);
987 struct bio_list list;
990 INIT_LIST_HEAD(&o->cb.list);
992 if (unlikely(!current->bio_list))
995 list = *current->bio_list;
996 bio_list_init(current->bio_list);
998 while ((bio = bio_list_pop(&list))) {
999 struct bio_set *bs = bio->bi_pool;
1000 if (unlikely(!bs) || bs == fs_bio_set) {
1001 bio_list_add(current->bio_list, bio);
1005 spin_lock(&bs->rescue_lock);
1006 bio_list_add(&bs->rescue_list, bio);
1007 queue_work(bs->rescue_workqueue, &bs->rescue_work);
1008 spin_unlock(&bs->rescue_lock);
1012 static void dm_offload_start(struct dm_offload *o)
1014 blk_start_plug(&o->plug);
1015 o->cb.callback = flush_current_bio_list;
1016 list_add(&o->cb.list, ¤t->plug->cb_list);
1019 static void dm_offload_end(struct dm_offload *o)
1021 list_del(&o->cb.list);
1022 blk_finish_plug(&o->plug);
1025 static void __map_bio(struct dm_target_io *tio)
1029 struct dm_offload o;
1030 struct bio *clone = &tio->clone;
1031 struct dm_target *ti = tio->ti;
1033 clone->bi_end_io = clone_endio;
1036 * Map the clone. If r == 0 we don't need to do
1037 * anything, the target has assumed ownership of
1040 atomic_inc(&tio->io->io_count);
1041 sector = clone->bi_iter.bi_sector;
1043 dm_offload_start(&o);
1044 r = ti->type->map(ti, clone);
1047 if (r == DM_MAPIO_REMAPPED) {
1048 /* the bio has been remapped so dispatch it */
1050 trace_block_bio_remap(bdev_get_queue(clone->bi_bdev), clone,
1051 tio->io->bio->bi_bdev->bd_dev, sector);
1053 generic_make_request(clone);
1054 } else if (r < 0 || r == DM_MAPIO_REQUEUE) {
1055 /* error the io and bail out, or requeue it if needed */
1056 dec_pending(tio->io, r);
1058 } else if (r != DM_MAPIO_SUBMITTED) {
1059 DMWARN("unimplemented target map return value: %d", r);
1065 struct mapped_device *md;
1066 struct dm_table *map;
1070 unsigned sector_count;
1073 static void bio_setup_sector(struct bio *bio, sector_t sector, unsigned len)
1075 bio->bi_iter.bi_sector = sector;
1076 bio->bi_iter.bi_size = to_bytes(len);
1080 * Creates a bio that consists of range of complete bvecs.
1082 static int clone_bio(struct dm_target_io *tio, struct bio *bio,
1083 sector_t sector, unsigned len)
1085 struct bio *clone = &tio->clone;
1087 __bio_clone_fast(clone, bio);
1089 if (bio_integrity(bio)) {
1090 int r = bio_integrity_clone(clone, bio, GFP_NOIO);
1095 bio_advance(clone, to_bytes(sector - clone->bi_iter.bi_sector));
1096 clone->bi_iter.bi_size = to_bytes(len);
1098 if (bio_integrity(bio))
1099 bio_integrity_trim(clone, 0, len);
1104 static struct dm_target_io *alloc_tio(struct clone_info *ci,
1105 struct dm_target *ti,
1106 unsigned target_bio_nr)
1108 struct dm_target_io *tio;
1111 clone = bio_alloc_bioset(GFP_NOIO, 0, ci->md->bs);
1112 tio = container_of(clone, struct dm_target_io, clone);
1116 tio->target_bio_nr = target_bio_nr;
1121 static void __clone_and_map_simple_bio(struct clone_info *ci,
1122 struct dm_target *ti,
1123 unsigned target_bio_nr, unsigned *len)
1125 struct dm_target_io *tio = alloc_tio(ci, ti, target_bio_nr);
1126 struct bio *clone = &tio->clone;
1130 __bio_clone_fast(clone, ci->bio);
1132 bio_setup_sector(clone, ci->sector, *len);
1137 static void __send_duplicate_bios(struct clone_info *ci, struct dm_target *ti,
1138 unsigned num_bios, unsigned *len)
1140 unsigned target_bio_nr;
1142 for (target_bio_nr = 0; target_bio_nr < num_bios; target_bio_nr++)
1143 __clone_and_map_simple_bio(ci, ti, target_bio_nr, len);
1146 static int __send_empty_flush(struct clone_info *ci)
1148 unsigned target_nr = 0;
1149 struct dm_target *ti;
1151 BUG_ON(bio_has_data(ci->bio));
1152 while ((ti = dm_table_get_target(ci->map, target_nr++)))
1153 __send_duplicate_bios(ci, ti, ti->num_flush_bios, NULL);
1158 static int __clone_and_map_data_bio(struct clone_info *ci, struct dm_target *ti,
1159 sector_t sector, unsigned *len)
1161 struct bio *bio = ci->bio;
1162 struct dm_target_io *tio;
1163 unsigned target_bio_nr;
1164 unsigned num_target_bios = 1;
1168 * Does the target want to receive duplicate copies of the bio?
1170 if (bio_data_dir(bio) == WRITE && ti->num_write_bios)
1171 num_target_bios = ti->num_write_bios(ti, bio);
1173 for (target_bio_nr = 0; target_bio_nr < num_target_bios; target_bio_nr++) {
1174 tio = alloc_tio(ci, ti, target_bio_nr);
1176 r = clone_bio(tio, bio, sector, *len);
1187 typedef unsigned (*get_num_bios_fn)(struct dm_target *ti);
1189 static unsigned get_num_discard_bios(struct dm_target *ti)
1191 return ti->num_discard_bios;
1194 static unsigned get_num_write_same_bios(struct dm_target *ti)
1196 return ti->num_write_same_bios;
1199 typedef bool (*is_split_required_fn)(struct dm_target *ti);
1201 static bool is_split_required_for_discard(struct dm_target *ti)
1203 return ti->split_discard_bios;
1206 static int __send_changing_extent_only(struct clone_info *ci,
1207 get_num_bios_fn get_num_bios,
1208 is_split_required_fn is_split_required)
1210 struct dm_target *ti;
1215 ti = dm_table_find_target(ci->map, ci->sector);
1216 if (!dm_target_is_valid(ti))
1220 * Even though the device advertised support for this type of
1221 * request, that does not mean every target supports it, and
1222 * reconfiguration might also have changed that since the
1223 * check was performed.
1225 num_bios = get_num_bios ? get_num_bios(ti) : 0;
1229 if (is_split_required && !is_split_required(ti))
1230 len = min((sector_t)ci->sector_count, max_io_len_target_boundary(ci->sector, ti));
1232 len = min((sector_t)ci->sector_count, max_io_len(ci->sector, ti));
1234 __send_duplicate_bios(ci, ti, num_bios, &len);
1237 } while (ci->sector_count -= len);
1242 static int __send_discard(struct clone_info *ci)
1244 return __send_changing_extent_only(ci, get_num_discard_bios,
1245 is_split_required_for_discard);
1248 static int __send_write_same(struct clone_info *ci)
1250 return __send_changing_extent_only(ci, get_num_write_same_bios, NULL);
1254 * Select the correct strategy for processing a non-flush bio.
1256 static int __split_and_process_non_flush(struct clone_info *ci)
1258 struct bio *bio = ci->bio;
1259 struct dm_target *ti;
1263 if (unlikely(bio_op(bio) == REQ_OP_DISCARD))
1264 return __send_discard(ci);
1265 else if (unlikely(bio_op(bio) == REQ_OP_WRITE_SAME))
1266 return __send_write_same(ci);
1268 ti = dm_table_find_target(ci->map, ci->sector);
1269 if (!dm_target_is_valid(ti))
1272 len = min_t(sector_t, max_io_len(ci->sector, ti), ci->sector_count);
1274 r = __clone_and_map_data_bio(ci, ti, ci->sector, &len);
1279 ci->sector_count -= len;
1285 * Entry point to split a bio into clones and submit them to the targets.
1287 static void __split_and_process_bio(struct mapped_device *md,
1288 struct dm_table *map, struct bio *bio)
1290 struct clone_info ci;
1293 if (unlikely(!map)) {
1300 ci.io = alloc_io(md);
1302 atomic_set(&ci.io->io_count, 1);
1305 spin_lock_init(&ci.io->endio_lock);
1306 ci.sector = bio->bi_iter.bi_sector;
1308 start_io_acct(ci.io);
1310 if (bio->bi_opf & REQ_PREFLUSH) {
1311 ci.bio = &ci.md->flush_bio;
1312 ci.sector_count = 0;
1313 error = __send_empty_flush(&ci);
1314 /* dec_pending submits any data associated with flush */
1317 ci.sector_count = bio_sectors(bio);
1318 while (ci.sector_count && !error)
1319 error = __split_and_process_non_flush(&ci);
1322 /* drop the extra reference count */
1323 dec_pending(ci.io, error);
1325 /*-----------------------------------------------------------------
1327 *---------------------------------------------------------------*/
1330 * The request function that just remaps the bio built up by
1333 static blk_qc_t dm_make_request(struct request_queue *q, struct bio *bio)
1335 int rw = bio_data_dir(bio);
1336 struct mapped_device *md = q->queuedata;
1338 struct dm_table *map;
1340 map = dm_get_live_table(md, &srcu_idx);
1342 generic_start_io_acct(rw, bio_sectors(bio), &dm_disk(md)->part0);
1344 /* if we're suspended, we have to queue this io for later */
1345 if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags))) {
1346 dm_put_live_table(md, srcu_idx);
1348 if (!(bio->bi_opf & REQ_RAHEAD))
1352 return BLK_QC_T_NONE;
1355 __split_and_process_bio(md, map, bio);
1356 dm_put_live_table(md, srcu_idx);
1357 return BLK_QC_T_NONE;
1360 static int dm_any_congested(void *congested_data, int bdi_bits)
1363 struct mapped_device *md = congested_data;
1364 struct dm_table *map;
1366 if (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
1367 if (dm_request_based(md)) {
1369 * With request-based DM we only need to check the
1370 * top-level queue for congestion.
1372 r = md->queue->backing_dev_info.wb.state & bdi_bits;
1374 map = dm_get_live_table_fast(md);
1376 r = dm_table_any_congested(map, bdi_bits);
1377 dm_put_live_table_fast(md);
1384 /*-----------------------------------------------------------------
1385 * An IDR is used to keep track of allocated minor numbers.
1386 *---------------------------------------------------------------*/
1387 static void free_minor(int minor)
1389 spin_lock(&_minor_lock);
1390 idr_remove(&_minor_idr, minor);
1391 spin_unlock(&_minor_lock);
1395 * See if the device with a specific minor # is free.
1397 static int specific_minor(int minor)
1401 if (minor >= (1 << MINORBITS))
1404 idr_preload(GFP_KERNEL);
1405 spin_lock(&_minor_lock);
1407 r = idr_alloc(&_minor_idr, MINOR_ALLOCED, minor, minor + 1, GFP_NOWAIT);
1409 spin_unlock(&_minor_lock);
1412 return r == -ENOSPC ? -EBUSY : r;
1416 static int next_free_minor(int *minor)
1420 idr_preload(GFP_KERNEL);
1421 spin_lock(&_minor_lock);
1423 r = idr_alloc(&_minor_idr, MINOR_ALLOCED, 0, 1 << MINORBITS, GFP_NOWAIT);
1425 spin_unlock(&_minor_lock);
1433 static const struct block_device_operations dm_blk_dops;
1435 static void dm_wq_work(struct work_struct *work);
1437 void dm_init_md_queue(struct mapped_device *md)
1440 * Request-based dm devices cannot be stacked on top of bio-based dm
1441 * devices. The type of this dm device may not have been decided yet.
1442 * The type is decided at the first table loading time.
1443 * To prevent problematic device stacking, clear the queue flag
1444 * for request stacking support until then.
1446 * This queue is new, so no concurrency on the queue_flags.
1448 queue_flag_clear_unlocked(QUEUE_FLAG_STACKABLE, md->queue);
1451 * Initialize data that will only be used by a non-blk-mq DM queue
1452 * - must do so here (in alloc_dev callchain) before queue is used
1454 md->queue->queuedata = md;
1455 md->queue->backing_dev_info.congested_data = md;
1458 void dm_init_normal_md_queue(struct mapped_device *md)
1460 md->use_blk_mq = false;
1461 dm_init_md_queue(md);
1464 * Initialize aspects of queue that aren't relevant for blk-mq
1466 md->queue->backing_dev_info.congested_fn = dm_any_congested;
1467 blk_queue_bounce_limit(md->queue, BLK_BOUNCE_ANY);
1470 static void cleanup_mapped_device(struct mapped_device *md)
1473 destroy_workqueue(md->wq);
1474 if (md->kworker_task)
1475 kthread_stop(md->kworker_task);
1476 mempool_destroy(md->io_pool);
1477 mempool_destroy(md->rq_pool);
1479 bioset_free(md->bs);
1482 spin_lock(&_minor_lock);
1483 md->disk->private_data = NULL;
1484 spin_unlock(&_minor_lock);
1485 del_gendisk(md->disk);
1490 blk_cleanup_queue(md->queue);
1492 cleanup_srcu_struct(&md->io_barrier);
1499 dm_mq_cleanup_mapped_device(md);
1503 * Allocate and initialise a blank device with a given minor.
1505 static struct mapped_device *alloc_dev(int minor)
1507 int r, numa_node_id = dm_get_numa_node();
1508 struct mapped_device *md;
1511 md = kzalloc_node(sizeof(*md), GFP_KERNEL, numa_node_id);
1513 DMWARN("unable to allocate device, out of memory.");
1517 if (!try_module_get(THIS_MODULE))
1518 goto bad_module_get;
1520 /* get a minor number for the dev */
1521 if (minor == DM_ANY_MINOR)
1522 r = next_free_minor(&minor);
1524 r = specific_minor(minor);
1528 r = init_srcu_struct(&md->io_barrier);
1530 goto bad_io_barrier;
1532 md->numa_node_id = numa_node_id;
1533 md->use_blk_mq = dm_use_blk_mq_default();
1534 md->init_tio_pdu = false;
1535 md->type = DM_TYPE_NONE;
1536 mutex_init(&md->suspend_lock);
1537 mutex_init(&md->type_lock);
1538 mutex_init(&md->table_devices_lock);
1539 spin_lock_init(&md->deferred_lock);
1540 atomic_set(&md->holders, 1);
1541 atomic_set(&md->open_count, 0);
1542 atomic_set(&md->event_nr, 0);
1543 atomic_set(&md->uevent_seq, 0);
1544 INIT_LIST_HEAD(&md->uevent_list);
1545 INIT_LIST_HEAD(&md->table_devices);
1546 spin_lock_init(&md->uevent_lock);
1548 md->queue = blk_alloc_queue_node(GFP_KERNEL, numa_node_id);
1552 dm_init_md_queue(md);
1554 md->disk = alloc_disk_node(1, numa_node_id);
1558 atomic_set(&md->pending[0], 0);
1559 atomic_set(&md->pending[1], 0);
1560 init_waitqueue_head(&md->wait);
1561 INIT_WORK(&md->work, dm_wq_work);
1562 init_waitqueue_head(&md->eventq);
1563 init_completion(&md->kobj_holder.completion);
1564 md->kworker_task = NULL;
1566 md->disk->major = _major;
1567 md->disk->first_minor = minor;
1568 md->disk->fops = &dm_blk_dops;
1569 md->disk->queue = md->queue;
1570 md->disk->private_data = md;
1571 sprintf(md->disk->disk_name, "dm-%d", minor);
1573 format_dev_t(md->name, MKDEV(_major, minor));
1575 md->wq = alloc_workqueue("kdmflush", WQ_MEM_RECLAIM, 0);
1579 md->bdev = bdget_disk(md->disk, 0);
1583 bio_init(&md->flush_bio);
1584 md->flush_bio.bi_bdev = md->bdev;
1585 bio_set_op_attrs(&md->flush_bio, REQ_OP_WRITE, WRITE_FLUSH);
1587 dm_stats_init(&md->stats);
1589 /* Populate the mapping, nobody knows we exist yet */
1590 spin_lock(&_minor_lock);
1591 old_md = idr_replace(&_minor_idr, md, minor);
1592 spin_unlock(&_minor_lock);
1594 BUG_ON(old_md != MINOR_ALLOCED);
1599 cleanup_mapped_device(md);
1603 module_put(THIS_MODULE);
1609 static void unlock_fs(struct mapped_device *md);
1611 static void free_dev(struct mapped_device *md)
1613 int minor = MINOR(disk_devt(md->disk));
1617 cleanup_mapped_device(md);
1619 free_table_devices(&md->table_devices);
1620 dm_stats_cleanup(&md->stats);
1623 module_put(THIS_MODULE);
1627 static void __bind_mempools(struct mapped_device *md, struct dm_table *t)
1629 struct dm_md_mempools *p = dm_table_get_md_mempools(t);
1632 /* The md already has necessary mempools. */
1633 if (dm_table_bio_based(t)) {
1635 * Reload bioset because front_pad may have changed
1636 * because a different table was loaded.
1638 bioset_free(md->bs);
1643 * There's no need to reload with request-based dm
1644 * because the size of front_pad doesn't change.
1645 * Note for future: If you are to reload bioset,
1646 * prep-ed requests in the queue may refer
1647 * to bio from the old bioset, so you must walk
1648 * through the queue to unprep.
1653 BUG_ON(!p || md->io_pool || md->rq_pool || md->bs);
1655 md->io_pool = p->io_pool;
1657 md->rq_pool = p->rq_pool;
1663 /* mempool bind completed, no longer need any mempools in the table */
1664 dm_table_free_md_mempools(t);
1668 * Bind a table to the device.
1670 static void event_callback(void *context)
1672 unsigned long flags;
1674 struct mapped_device *md = (struct mapped_device *) context;
1676 spin_lock_irqsave(&md->uevent_lock, flags);
1677 list_splice_init(&md->uevent_list, &uevents);
1678 spin_unlock_irqrestore(&md->uevent_lock, flags);
1680 dm_send_uevents(&uevents, &disk_to_dev(md->disk)->kobj);
1682 atomic_inc(&md->event_nr);
1683 wake_up(&md->eventq);
1687 * Protected by md->suspend_lock obtained by dm_swap_table().
1689 static void __set_size(struct mapped_device *md, sector_t size)
1691 set_capacity(md->disk, size);
1693 i_size_write(md->bdev->bd_inode, (loff_t)size << SECTOR_SHIFT);
1697 * Returns old map, which caller must destroy.
1699 static struct dm_table *__bind(struct mapped_device *md, struct dm_table *t,
1700 struct queue_limits *limits)
1702 struct dm_table *old_map;
1703 struct request_queue *q = md->queue;
1706 lockdep_assert_held(&md->suspend_lock);
1708 size = dm_table_get_size(t);
1711 * Wipe any geometry if the size of the table changed.
1713 if (size != dm_get_size(md))
1714 memset(&md->geometry, 0, sizeof(md->geometry));
1716 __set_size(md, size);
1718 dm_table_event_callback(t, event_callback, md);
1721 * The queue hasn't been stopped yet, if the old table type wasn't
1722 * for request-based during suspension. So stop it to prevent
1723 * I/O mapping before resume.
1724 * This must be done before setting the queue restrictions,
1725 * because request-based dm may be run just after the setting.
1727 if (dm_table_request_based(t)) {
1730 * Leverage the fact that request-based DM targets are
1731 * immutable singletons and establish md->immutable_target
1732 * - used to optimize both dm_request_fn and dm_mq_queue_rq
1734 md->immutable_target = dm_table_get_immutable_target(t);
1737 __bind_mempools(md, t);
1739 old_map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
1740 rcu_assign_pointer(md->map, (void *)t);
1741 md->immutable_target_type = dm_table_get_immutable_target_type(t);
1743 dm_table_set_restrictions(t, q, limits);
1751 * Returns unbound table for the caller to free.
1753 static struct dm_table *__unbind(struct mapped_device *md)
1755 struct dm_table *map = rcu_dereference_protected(md->map, 1);
1760 dm_table_event_callback(map, NULL, NULL);
1761 RCU_INIT_POINTER(md->map, NULL);
1768 * Constructor for a new device.
1770 int dm_create(int minor, struct mapped_device **result)
1772 struct mapped_device *md;
1774 md = alloc_dev(minor);
1785 * Functions to manage md->type.
1786 * All are required to hold md->type_lock.
1788 void dm_lock_md_type(struct mapped_device *md)
1790 mutex_lock(&md->type_lock);
1793 void dm_unlock_md_type(struct mapped_device *md)
1795 mutex_unlock(&md->type_lock);
1798 void dm_set_md_type(struct mapped_device *md, unsigned type)
1800 BUG_ON(!mutex_is_locked(&md->type_lock));
1804 unsigned dm_get_md_type(struct mapped_device *md)
1809 struct target_type *dm_get_immutable_target_type(struct mapped_device *md)
1811 return md->immutable_target_type;
1815 * The queue_limits are only valid as long as you have a reference
1818 struct queue_limits *dm_get_queue_limits(struct mapped_device *md)
1820 BUG_ON(!atomic_read(&md->holders));
1821 return &md->queue->limits;
1823 EXPORT_SYMBOL_GPL(dm_get_queue_limits);
1826 * Setup the DM device's queue based on md's type
1828 int dm_setup_md_queue(struct mapped_device *md, struct dm_table *t)
1831 unsigned type = dm_get_md_type(md);
1834 case DM_TYPE_REQUEST_BASED:
1835 r = dm_old_init_request_queue(md);
1837 DMERR("Cannot initialize queue for request-based mapped device");
1841 case DM_TYPE_MQ_REQUEST_BASED:
1842 r = dm_mq_init_request_queue(md, t);
1844 DMERR("Cannot initialize queue for request-based dm-mq mapped device");
1848 case DM_TYPE_BIO_BASED:
1849 case DM_TYPE_DAX_BIO_BASED:
1850 dm_init_normal_md_queue(md);
1851 blk_queue_make_request(md->queue, dm_make_request);
1853 * DM handles splitting bios as needed. Free the bio_split bioset
1854 * since it won't be used (saves 1 process per bio-based DM device).
1856 bioset_free(md->queue->bio_split);
1857 md->queue->bio_split = NULL;
1859 if (type == DM_TYPE_DAX_BIO_BASED)
1860 queue_flag_set_unlocked(QUEUE_FLAG_DAX, md->queue);
1867 struct mapped_device *dm_get_md(dev_t dev)
1869 struct mapped_device *md;
1870 unsigned minor = MINOR(dev);
1872 if (MAJOR(dev) != _major || minor >= (1 << MINORBITS))
1875 spin_lock(&_minor_lock);
1877 md = idr_find(&_minor_idr, minor);
1879 if ((md == MINOR_ALLOCED ||
1880 (MINOR(disk_devt(dm_disk(md))) != minor) ||
1881 dm_deleting_md(md) ||
1882 test_bit(DMF_FREEING, &md->flags))) {
1890 spin_unlock(&_minor_lock);
1894 EXPORT_SYMBOL_GPL(dm_get_md);
1896 void *dm_get_mdptr(struct mapped_device *md)
1898 return md->interface_ptr;
1901 void dm_set_mdptr(struct mapped_device *md, void *ptr)
1903 md->interface_ptr = ptr;
1906 void dm_get(struct mapped_device *md)
1908 atomic_inc(&md->holders);
1909 BUG_ON(test_bit(DMF_FREEING, &md->flags));
1912 int dm_hold(struct mapped_device *md)
1914 spin_lock(&_minor_lock);
1915 if (test_bit(DMF_FREEING, &md->flags)) {
1916 spin_unlock(&_minor_lock);
1920 spin_unlock(&_minor_lock);
1923 EXPORT_SYMBOL_GPL(dm_hold);
1925 const char *dm_device_name(struct mapped_device *md)
1929 EXPORT_SYMBOL_GPL(dm_device_name);
1931 static void __dm_destroy(struct mapped_device *md, bool wait)
1933 struct request_queue *q = dm_get_md_queue(md);
1934 struct dm_table *map;
1939 spin_lock(&_minor_lock);
1940 idr_replace(&_minor_idr, MINOR_ALLOCED, MINOR(disk_devt(dm_disk(md))));
1941 set_bit(DMF_FREEING, &md->flags);
1942 spin_unlock(&_minor_lock);
1944 spin_lock_irq(q->queue_lock);
1945 queue_flag_set(QUEUE_FLAG_DYING, q);
1946 spin_unlock_irq(q->queue_lock);
1948 if (dm_request_based(md) && md->kworker_task)
1949 kthread_flush_worker(&md->kworker);
1952 * Take suspend_lock so that presuspend and postsuspend methods
1953 * do not race with internal suspend.
1955 mutex_lock(&md->suspend_lock);
1956 map = dm_get_live_table(md, &srcu_idx);
1957 if (!dm_suspended_md(md)) {
1958 dm_table_presuspend_targets(map);
1959 dm_table_postsuspend_targets(map);
1961 /* dm_put_live_table must be before msleep, otherwise deadlock is possible */
1962 dm_put_live_table(md, srcu_idx);
1963 mutex_unlock(&md->suspend_lock);
1966 * Rare, but there may be I/O requests still going to complete,
1967 * for example. Wait for all references to disappear.
1968 * No one should increment the reference count of the mapped_device,
1969 * after the mapped_device state becomes DMF_FREEING.
1972 while (atomic_read(&md->holders))
1974 else if (atomic_read(&md->holders))
1975 DMWARN("%s: Forcibly removing mapped_device still in use! (%d users)",
1976 dm_device_name(md), atomic_read(&md->holders));
1979 dm_table_destroy(__unbind(md));
1983 void dm_destroy(struct mapped_device *md)
1985 __dm_destroy(md, true);
1988 void dm_destroy_immediate(struct mapped_device *md)
1990 __dm_destroy(md, false);
1993 void dm_put(struct mapped_device *md)
1995 atomic_dec(&md->holders);
1997 EXPORT_SYMBOL_GPL(dm_put);
1999 static int dm_wait_for_completion(struct mapped_device *md, long task_state)
2005 prepare_to_wait(&md->wait, &wait, task_state);
2007 if (!md_in_flight(md))
2010 if (signal_pending_state(task_state, current)) {
2017 finish_wait(&md->wait, &wait);
2023 * Process the deferred bios
2025 static void dm_wq_work(struct work_struct *work)
2027 struct mapped_device *md = container_of(work, struct mapped_device,
2031 struct dm_table *map;
2033 map = dm_get_live_table(md, &srcu_idx);
2035 while (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
2036 spin_lock_irq(&md->deferred_lock);
2037 c = bio_list_pop(&md->deferred);
2038 spin_unlock_irq(&md->deferred_lock);
2043 if (dm_request_based(md))
2044 generic_make_request(c);
2046 __split_and_process_bio(md, map, c);
2049 dm_put_live_table(md, srcu_idx);
2052 static void dm_queue_flush(struct mapped_device *md)
2054 clear_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2055 smp_mb__after_atomic();
2056 queue_work(md->wq, &md->work);
2060 * Swap in a new table, returning the old one for the caller to destroy.
2062 struct dm_table *dm_swap_table(struct mapped_device *md, struct dm_table *table)
2064 struct dm_table *live_map = NULL, *map = ERR_PTR(-EINVAL);
2065 struct queue_limits limits;
2068 mutex_lock(&md->suspend_lock);
2070 /* device must be suspended */
2071 if (!dm_suspended_md(md))
2075 * If the new table has no data devices, retain the existing limits.
2076 * This helps multipath with queue_if_no_path if all paths disappear,
2077 * then new I/O is queued based on these limits, and then some paths
2080 if (dm_table_has_no_data_devices(table)) {
2081 live_map = dm_get_live_table_fast(md);
2083 limits = md->queue->limits;
2084 dm_put_live_table_fast(md);
2088 r = dm_calculate_queue_limits(table, &limits);
2095 map = __bind(md, table, &limits);
2098 mutex_unlock(&md->suspend_lock);
2103 * Functions to lock and unlock any filesystem running on the
2106 static int lock_fs(struct mapped_device *md)
2110 WARN_ON(md->frozen_sb);
2112 md->frozen_sb = freeze_bdev(md->bdev);
2113 if (IS_ERR(md->frozen_sb)) {
2114 r = PTR_ERR(md->frozen_sb);
2115 md->frozen_sb = NULL;
2119 set_bit(DMF_FROZEN, &md->flags);
2124 static void unlock_fs(struct mapped_device *md)
2126 if (!test_bit(DMF_FROZEN, &md->flags))
2129 thaw_bdev(md->bdev, md->frozen_sb);
2130 md->frozen_sb = NULL;
2131 clear_bit(DMF_FROZEN, &md->flags);
2135 * @suspend_flags: DM_SUSPEND_LOCKFS_FLAG and/or DM_SUSPEND_NOFLUSH_FLAG
2136 * @task_state: e.g. TASK_INTERRUPTIBLE or TASK_UNINTERRUPTIBLE
2137 * @dmf_suspended_flag: DMF_SUSPENDED or DMF_SUSPENDED_INTERNALLY
2139 * If __dm_suspend returns 0, the device is completely quiescent
2140 * now. There is no request-processing activity. All new requests
2141 * are being added to md->deferred list.
2143 * Caller must hold md->suspend_lock
2145 static int __dm_suspend(struct mapped_device *md, struct dm_table *map,
2146 unsigned suspend_flags, long task_state,
2147 int dmf_suspended_flag)
2149 bool do_lockfs = suspend_flags & DM_SUSPEND_LOCKFS_FLAG;
2150 bool noflush = suspend_flags & DM_SUSPEND_NOFLUSH_FLAG;
2153 lockdep_assert_held(&md->suspend_lock);
2156 * DMF_NOFLUSH_SUSPENDING must be set before presuspend.
2157 * This flag is cleared before dm_suspend returns.
2160 set_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
2163 * This gets reverted if there's an error later and the targets
2164 * provide the .presuspend_undo hook.
2166 dm_table_presuspend_targets(map);
2169 * Flush I/O to the device.
2170 * Any I/O submitted after lock_fs() may not be flushed.
2171 * noflush takes precedence over do_lockfs.
2172 * (lock_fs() flushes I/Os and waits for them to complete.)
2174 if (!noflush && do_lockfs) {
2177 dm_table_presuspend_undo_targets(map);
2183 * Here we must make sure that no processes are submitting requests
2184 * to target drivers i.e. no one may be executing
2185 * __split_and_process_bio. This is called from dm_request and
2188 * To get all processes out of __split_and_process_bio in dm_request,
2189 * we take the write lock. To prevent any process from reentering
2190 * __split_and_process_bio from dm_request and quiesce the thread
2191 * (dm_wq_work), we set BMF_BLOCK_IO_FOR_SUSPEND and call
2192 * flush_workqueue(md->wq).
2194 set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2196 synchronize_srcu(&md->io_barrier);
2199 * Stop md->queue before flushing md->wq in case request-based
2200 * dm defers requests to md->wq from md->queue.
2202 if (dm_request_based(md)) {
2203 dm_stop_queue(md->queue);
2204 if (md->kworker_task)
2205 kthread_flush_worker(&md->kworker);
2208 flush_workqueue(md->wq);
2211 * At this point no more requests are entering target request routines.
2212 * We call dm_wait_for_completion to wait for all existing requests
2215 r = dm_wait_for_completion(md, task_state);
2217 set_bit(dmf_suspended_flag, &md->flags);
2220 clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
2222 synchronize_srcu(&md->io_barrier);
2224 /* were we interrupted ? */
2228 if (dm_request_based(md))
2229 dm_start_queue(md->queue);
2232 dm_table_presuspend_undo_targets(map);
2233 /* pushback list is already flushed, so skip flush */
2240 * We need to be able to change a mapping table under a mounted
2241 * filesystem. For example we might want to move some data in
2242 * the background. Before the table can be swapped with
2243 * dm_bind_table, dm_suspend must be called to flush any in
2244 * flight bios and ensure that any further io gets deferred.
2247 * Suspend mechanism in request-based dm.
2249 * 1. Flush all I/Os by lock_fs() if needed.
2250 * 2. Stop dispatching any I/O by stopping the request_queue.
2251 * 3. Wait for all in-flight I/Os to be completed or requeued.
2253 * To abort suspend, start the request_queue.
2255 int dm_suspend(struct mapped_device *md, unsigned suspend_flags)
2257 struct dm_table *map = NULL;
2261 mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING);
2263 if (dm_suspended_md(md)) {
2268 if (dm_suspended_internally_md(md)) {
2269 /* already internally suspended, wait for internal resume */
2270 mutex_unlock(&md->suspend_lock);
2271 r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE);
2277 map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2279 r = __dm_suspend(md, map, suspend_flags, TASK_INTERRUPTIBLE, DMF_SUSPENDED);
2283 dm_table_postsuspend_targets(map);
2286 mutex_unlock(&md->suspend_lock);
2290 static int __dm_resume(struct mapped_device *md, struct dm_table *map)
2293 int r = dm_table_resume_targets(map);
2301 * Flushing deferred I/Os must be done after targets are resumed
2302 * so that mapping of targets can work correctly.
2303 * Request-based dm is queueing the deferred I/Os in its request_queue.
2305 if (dm_request_based(md))
2306 dm_start_queue(md->queue);
2313 int dm_resume(struct mapped_device *md)
2316 struct dm_table *map = NULL;
2320 mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING);
2322 if (!dm_suspended_md(md))
2325 if (dm_suspended_internally_md(md)) {
2326 /* already internally suspended, wait for internal resume */
2327 mutex_unlock(&md->suspend_lock);
2328 r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE);
2334 map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2335 if (!map || !dm_table_get_size(map))
2338 r = __dm_resume(md, map);
2342 clear_bit(DMF_SUSPENDED, &md->flags);
2344 mutex_unlock(&md->suspend_lock);
2350 * Internal suspend/resume works like userspace-driven suspend. It waits
2351 * until all bios finish and prevents issuing new bios to the target drivers.
2352 * It may be used only from the kernel.
2355 static void __dm_internal_suspend(struct mapped_device *md, unsigned suspend_flags)
2357 struct dm_table *map = NULL;
2359 if (md->internal_suspend_count++)
2360 return; /* nested internal suspend */
2362 if (dm_suspended_md(md)) {
2363 set_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
2364 return; /* nest suspend */
2367 map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2370 * Using TASK_UNINTERRUPTIBLE because only NOFLUSH internal suspend is
2371 * supported. Properly supporting a TASK_INTERRUPTIBLE internal suspend
2372 * would require changing .presuspend to return an error -- avoid this
2373 * until there is a need for more elaborate variants of internal suspend.
2375 (void) __dm_suspend(md, map, suspend_flags, TASK_UNINTERRUPTIBLE,
2376 DMF_SUSPENDED_INTERNALLY);
2378 dm_table_postsuspend_targets(map);
2381 static void __dm_internal_resume(struct mapped_device *md)
2383 BUG_ON(!md->internal_suspend_count);
2385 if (--md->internal_suspend_count)
2386 return; /* resume from nested internal suspend */
2388 if (dm_suspended_md(md))
2389 goto done; /* resume from nested suspend */
2392 * NOTE: existing callers don't need to call dm_table_resume_targets
2393 * (which may fail -- so best to avoid it for now by passing NULL map)
2395 (void) __dm_resume(md, NULL);
2398 clear_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
2399 smp_mb__after_atomic();
2400 wake_up_bit(&md->flags, DMF_SUSPENDED_INTERNALLY);
2403 void dm_internal_suspend_noflush(struct mapped_device *md)
2405 mutex_lock(&md->suspend_lock);
2406 __dm_internal_suspend(md, DM_SUSPEND_NOFLUSH_FLAG);
2407 mutex_unlock(&md->suspend_lock);
2409 EXPORT_SYMBOL_GPL(dm_internal_suspend_noflush);
2411 void dm_internal_resume(struct mapped_device *md)
2413 mutex_lock(&md->suspend_lock);
2414 __dm_internal_resume(md);
2415 mutex_unlock(&md->suspend_lock);
2417 EXPORT_SYMBOL_GPL(dm_internal_resume);
2420 * Fast variants of internal suspend/resume hold md->suspend_lock,
2421 * which prevents interaction with userspace-driven suspend.
2424 void dm_internal_suspend_fast(struct mapped_device *md)
2426 mutex_lock(&md->suspend_lock);
2427 if (dm_suspended_md(md) || dm_suspended_internally_md(md))
2430 set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2431 synchronize_srcu(&md->io_barrier);
2432 flush_workqueue(md->wq);
2433 dm_wait_for_completion(md, TASK_UNINTERRUPTIBLE);
2435 EXPORT_SYMBOL_GPL(dm_internal_suspend_fast);
2437 void dm_internal_resume_fast(struct mapped_device *md)
2439 if (dm_suspended_md(md) || dm_suspended_internally_md(md))
2445 mutex_unlock(&md->suspend_lock);
2447 EXPORT_SYMBOL_GPL(dm_internal_resume_fast);
2449 /*-----------------------------------------------------------------
2450 * Event notification.
2451 *---------------------------------------------------------------*/
2452 int dm_kobject_uevent(struct mapped_device *md, enum kobject_action action,
2455 char udev_cookie[DM_COOKIE_LENGTH];
2456 char *envp[] = { udev_cookie, NULL };
2459 return kobject_uevent(&disk_to_dev(md->disk)->kobj, action);
2461 snprintf(udev_cookie, DM_COOKIE_LENGTH, "%s=%u",
2462 DM_COOKIE_ENV_VAR_NAME, cookie);
2463 return kobject_uevent_env(&disk_to_dev(md->disk)->kobj,
2468 uint32_t dm_next_uevent_seq(struct mapped_device *md)
2470 return atomic_add_return(1, &md->uevent_seq);
2473 uint32_t dm_get_event_nr(struct mapped_device *md)
2475 return atomic_read(&md->event_nr);
2478 int dm_wait_event(struct mapped_device *md, int event_nr)
2480 return wait_event_interruptible(md->eventq,
2481 (event_nr != atomic_read(&md->event_nr)));
2484 void dm_uevent_add(struct mapped_device *md, struct list_head *elist)
2486 unsigned long flags;
2488 spin_lock_irqsave(&md->uevent_lock, flags);
2489 list_add(elist, &md->uevent_list);
2490 spin_unlock_irqrestore(&md->uevent_lock, flags);
2494 * The gendisk is only valid as long as you have a reference
2497 struct gendisk *dm_disk(struct mapped_device *md)
2501 EXPORT_SYMBOL_GPL(dm_disk);
2503 struct kobject *dm_kobject(struct mapped_device *md)
2505 return &md->kobj_holder.kobj;
2508 struct mapped_device *dm_get_from_kobject(struct kobject *kobj)
2510 struct mapped_device *md;
2512 md = container_of(kobj, struct mapped_device, kobj_holder.kobj);
2514 if (test_bit(DMF_FREEING, &md->flags) ||
2522 int dm_suspended_md(struct mapped_device *md)
2524 return test_bit(DMF_SUSPENDED, &md->flags);
2527 int dm_suspended_internally_md(struct mapped_device *md)
2529 return test_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
2532 int dm_test_deferred_remove_flag(struct mapped_device *md)
2534 return test_bit(DMF_DEFERRED_REMOVE, &md->flags);
2537 int dm_suspended(struct dm_target *ti)
2539 return dm_suspended_md(dm_table_get_md(ti->table));
2541 EXPORT_SYMBOL_GPL(dm_suspended);
2543 int dm_noflush_suspending(struct dm_target *ti)
2545 return __noflush_suspending(dm_table_get_md(ti->table));
2547 EXPORT_SYMBOL_GPL(dm_noflush_suspending);
2549 struct dm_md_mempools *dm_alloc_md_mempools(struct mapped_device *md, unsigned type,
2550 unsigned integrity, unsigned per_io_data_size)
2552 struct dm_md_mempools *pools = kzalloc_node(sizeof(*pools), GFP_KERNEL, md->numa_node_id);
2553 struct kmem_cache *cachep = NULL;
2554 unsigned int pool_size = 0;
2555 unsigned int front_pad;
2561 case DM_TYPE_BIO_BASED:
2562 case DM_TYPE_DAX_BIO_BASED:
2564 pool_size = dm_get_reserved_bio_based_ios();
2565 front_pad = roundup(per_io_data_size, __alignof__(struct dm_target_io)) + offsetof(struct dm_target_io, clone);
2567 case DM_TYPE_REQUEST_BASED:
2568 cachep = _rq_tio_cache;
2569 pool_size = dm_get_reserved_rq_based_ios();
2570 pools->rq_pool = mempool_create_slab_pool(pool_size, _rq_cache);
2571 if (!pools->rq_pool)
2573 /* fall through to setup remaining rq-based pools */
2574 case DM_TYPE_MQ_REQUEST_BASED:
2576 pool_size = dm_get_reserved_rq_based_ios();
2577 front_pad = offsetof(struct dm_rq_clone_bio_info, clone);
2578 /* per_io_data_size is used for blk-mq pdu at queue allocation */
2585 pools->io_pool = mempool_create_slab_pool(pool_size, cachep);
2586 if (!pools->io_pool)
2590 pools->bs = bioset_create_nobvec(pool_size, front_pad);
2594 if (integrity && bioset_integrity_create(pools->bs, pool_size))
2600 dm_free_md_mempools(pools);
2605 void dm_free_md_mempools(struct dm_md_mempools *pools)
2610 mempool_destroy(pools->io_pool);
2611 mempool_destroy(pools->rq_pool);
2614 bioset_free(pools->bs);
2626 static int dm_call_pr(struct block_device *bdev, iterate_devices_callout_fn fn,
2629 struct mapped_device *md = bdev->bd_disk->private_data;
2630 struct dm_table *table;
2631 struct dm_target *ti;
2632 int ret = -ENOTTY, srcu_idx;
2634 table = dm_get_live_table(md, &srcu_idx);
2635 if (!table || !dm_table_get_size(table))
2638 /* We only support devices that have a single target */
2639 if (dm_table_get_num_targets(table) != 1)
2641 ti = dm_table_get_target(table, 0);
2644 if (!ti->type->iterate_devices)
2647 ret = ti->type->iterate_devices(ti, fn, data);
2649 dm_put_live_table(md, srcu_idx);
2654 * For register / unregister we need to manually call out to every path.
2656 static int __dm_pr_register(struct dm_target *ti, struct dm_dev *dev,
2657 sector_t start, sector_t len, void *data)
2659 struct dm_pr *pr = data;
2660 const struct pr_ops *ops = dev->bdev->bd_disk->fops->pr_ops;
2662 if (!ops || !ops->pr_register)
2664 return ops->pr_register(dev->bdev, pr->old_key, pr->new_key, pr->flags);
2667 static int dm_pr_register(struct block_device *bdev, u64 old_key, u64 new_key,
2678 ret = dm_call_pr(bdev, __dm_pr_register, &pr);
2679 if (ret && new_key) {
2680 /* unregister all paths if we failed to register any path */
2681 pr.old_key = new_key;
2684 pr.fail_early = false;
2685 dm_call_pr(bdev, __dm_pr_register, &pr);
2691 static int dm_pr_reserve(struct block_device *bdev, u64 key, enum pr_type type,
2694 struct mapped_device *md = bdev->bd_disk->private_data;
2695 const struct pr_ops *ops;
2699 r = dm_grab_bdev_for_ioctl(md, &bdev, &mode);
2703 ops = bdev->bd_disk->fops->pr_ops;
2704 if (ops && ops->pr_reserve)
2705 r = ops->pr_reserve(bdev, key, type, flags);
2713 static int dm_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
2715 struct mapped_device *md = bdev->bd_disk->private_data;
2716 const struct pr_ops *ops;
2720 r = dm_grab_bdev_for_ioctl(md, &bdev, &mode);
2724 ops = bdev->bd_disk->fops->pr_ops;
2725 if (ops && ops->pr_release)
2726 r = ops->pr_release(bdev, key, type);
2734 static int dm_pr_preempt(struct block_device *bdev, u64 old_key, u64 new_key,
2735 enum pr_type type, bool abort)
2737 struct mapped_device *md = bdev->bd_disk->private_data;
2738 const struct pr_ops *ops;
2742 r = dm_grab_bdev_for_ioctl(md, &bdev, &mode);
2746 ops = bdev->bd_disk->fops->pr_ops;
2747 if (ops && ops->pr_preempt)
2748 r = ops->pr_preempt(bdev, old_key, new_key, type, abort);
2756 static int dm_pr_clear(struct block_device *bdev, u64 key)
2758 struct mapped_device *md = bdev->bd_disk->private_data;
2759 const struct pr_ops *ops;
2763 r = dm_grab_bdev_for_ioctl(md, &bdev, &mode);
2767 ops = bdev->bd_disk->fops->pr_ops;
2768 if (ops && ops->pr_clear)
2769 r = ops->pr_clear(bdev, key);
2777 static const struct pr_ops dm_pr_ops = {
2778 .pr_register = dm_pr_register,
2779 .pr_reserve = dm_pr_reserve,
2780 .pr_release = dm_pr_release,
2781 .pr_preempt = dm_pr_preempt,
2782 .pr_clear = dm_pr_clear,
2785 static const struct block_device_operations dm_blk_dops = {
2786 .open = dm_blk_open,
2787 .release = dm_blk_close,
2788 .ioctl = dm_blk_ioctl,
2789 .direct_access = dm_blk_direct_access,
2790 .getgeo = dm_blk_getgeo,
2791 .pr_ops = &dm_pr_ops,
2792 .owner = THIS_MODULE
2798 module_init(dm_init);
2799 module_exit(dm_exit);
2801 module_param(major, uint, 0);
2802 MODULE_PARM_DESC(major, "The major number of the device mapper");
2804 module_param(reserved_bio_based_ios, uint, S_IRUGO | S_IWUSR);
2805 MODULE_PARM_DESC(reserved_bio_based_ios, "Reserved IOs in bio-based mempools");
2807 module_param(dm_numa_node, int, S_IRUGO | S_IWUSR);
2808 MODULE_PARM_DESC(dm_numa_node, "NUMA node for DM device memory allocations");
2810 MODULE_DESCRIPTION(DM_NAME " driver");
2811 MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
2812 MODULE_LICENSE("GPL");