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/sched/signal.h>
16 #include <linux/blkpg.h>
17 #include <linux/bio.h>
18 #include <linux/mempool.h>
19 #include <linux/dax.h>
20 #include <linux/slab.h>
21 #include <linux/idr.h>
22 #include <linux/hdreg.h>
23 #include <linux/delay.h>
24 #include <linux/wait.h>
27 #define DM_MSG_PREFIX "core"
31 * ratelimit state to be used in DMXXX_LIMIT().
33 DEFINE_RATELIMIT_STATE(dm_ratelimit_state,
34 DEFAULT_RATELIMIT_INTERVAL,
35 DEFAULT_RATELIMIT_BURST);
36 EXPORT_SYMBOL(dm_ratelimit_state);
40 * Cookies are numeric values sent with CHANGE and REMOVE
41 * uevents while resuming, removing or renaming the device.
43 #define DM_COOKIE_ENV_VAR_NAME "DM_COOKIE"
44 #define DM_COOKIE_LENGTH 24
46 static const char *_name = DM_NAME;
48 static unsigned int major = 0;
49 static unsigned int _major = 0;
51 static DEFINE_IDR(_minor_idr);
53 static DEFINE_SPINLOCK(_minor_lock);
55 static void do_deferred_remove(struct work_struct *w);
57 static DECLARE_WORK(deferred_remove_work, do_deferred_remove);
59 static struct workqueue_struct *deferred_remove_workqueue;
61 atomic_t dm_global_event_nr = ATOMIC_INIT(0);
62 DECLARE_WAIT_QUEUE_HEAD(dm_global_eventq);
65 * One of these is allocated per bio.
68 struct mapped_device *md;
72 unsigned long start_time;
73 spinlock_t endio_lock;
74 struct dm_stats_aux stats_aux;
77 #define MINOR_ALLOCED ((void *)-1)
80 * Bits for the md->flags field.
82 #define DMF_BLOCK_IO_FOR_SUSPEND 0
83 #define DMF_SUSPENDED 1
86 #define DMF_DELETING 4
87 #define DMF_NOFLUSH_SUSPENDING 5
88 #define DMF_DEFERRED_REMOVE 6
89 #define DMF_SUSPENDED_INTERNALLY 7
91 #define DM_NUMA_NODE NUMA_NO_NODE
92 static int dm_numa_node = DM_NUMA_NODE;
95 * For mempools pre-allocation at the table loading time.
97 struct dm_md_mempools {
102 struct table_device {
103 struct list_head list;
105 struct dm_dev dm_dev;
108 static struct kmem_cache *_io_cache;
109 static struct kmem_cache *_rq_tio_cache;
110 static struct kmem_cache *_rq_cache;
113 * Bio-based DM's mempools' reserved IOs set by the user.
115 #define RESERVED_BIO_BASED_IOS 16
116 static unsigned reserved_bio_based_ios = RESERVED_BIO_BASED_IOS;
118 static int __dm_get_module_param_int(int *module_param, int min, int max)
120 int param = ACCESS_ONCE(*module_param);
121 int modified_param = 0;
122 bool modified = true;
125 modified_param = min;
126 else if (param > max)
127 modified_param = max;
132 (void)cmpxchg(module_param, param, modified_param);
133 param = modified_param;
139 unsigned __dm_get_module_param(unsigned *module_param,
140 unsigned def, unsigned max)
142 unsigned param = ACCESS_ONCE(*module_param);
143 unsigned modified_param = 0;
146 modified_param = def;
147 else if (param > max)
148 modified_param = max;
150 if (modified_param) {
151 (void)cmpxchg(module_param, param, modified_param);
152 param = modified_param;
158 unsigned dm_get_reserved_bio_based_ios(void)
160 return __dm_get_module_param(&reserved_bio_based_ios,
161 RESERVED_BIO_BASED_IOS, DM_RESERVED_MAX_IOS);
163 EXPORT_SYMBOL_GPL(dm_get_reserved_bio_based_ios);
165 static unsigned dm_get_numa_node(void)
167 return __dm_get_module_param_int(&dm_numa_node,
168 DM_NUMA_NODE, num_online_nodes() - 1);
171 static int __init local_init(void)
175 /* allocate a slab for the dm_ios */
176 _io_cache = KMEM_CACHE(dm_io, 0);
180 _rq_tio_cache = KMEM_CACHE(dm_rq_target_io, 0);
182 goto out_free_io_cache;
184 _rq_cache = kmem_cache_create("dm_old_clone_request", sizeof(struct request),
185 __alignof__(struct request), 0, NULL);
187 goto out_free_rq_tio_cache;
189 r = dm_uevent_init();
191 goto out_free_rq_cache;
193 deferred_remove_workqueue = alloc_workqueue("kdmremove", WQ_UNBOUND, 1);
194 if (!deferred_remove_workqueue) {
196 goto out_uevent_exit;
200 r = register_blkdev(_major, _name);
202 goto out_free_workqueue;
210 destroy_workqueue(deferred_remove_workqueue);
214 kmem_cache_destroy(_rq_cache);
215 out_free_rq_tio_cache:
216 kmem_cache_destroy(_rq_tio_cache);
218 kmem_cache_destroy(_io_cache);
223 static void local_exit(void)
225 flush_scheduled_work();
226 destroy_workqueue(deferred_remove_workqueue);
228 kmem_cache_destroy(_rq_cache);
229 kmem_cache_destroy(_rq_tio_cache);
230 kmem_cache_destroy(_io_cache);
231 unregister_blkdev(_major, _name);
236 DMINFO("cleaned up");
239 static int (*_inits[])(void) __initdata = {
250 static void (*_exits[])(void) = {
261 static int __init dm_init(void)
263 const int count = ARRAY_SIZE(_inits);
267 for (i = 0; i < count; i++) {
282 static void __exit dm_exit(void)
284 int i = ARRAY_SIZE(_exits);
290 * Should be empty by this point.
292 idr_destroy(&_minor_idr);
296 * Block device functions
298 int dm_deleting_md(struct mapped_device *md)
300 return test_bit(DMF_DELETING, &md->flags);
303 static int dm_blk_open(struct block_device *bdev, fmode_t mode)
305 struct mapped_device *md;
307 spin_lock(&_minor_lock);
309 md = bdev->bd_disk->private_data;
313 if (test_bit(DMF_FREEING, &md->flags) ||
314 dm_deleting_md(md)) {
320 atomic_inc(&md->open_count);
322 spin_unlock(&_minor_lock);
324 return md ? 0 : -ENXIO;
327 static void dm_blk_close(struct gendisk *disk, fmode_t mode)
329 struct mapped_device *md;
331 spin_lock(&_minor_lock);
333 md = disk->private_data;
337 if (atomic_dec_and_test(&md->open_count) &&
338 (test_bit(DMF_DEFERRED_REMOVE, &md->flags)))
339 queue_work(deferred_remove_workqueue, &deferred_remove_work);
343 spin_unlock(&_minor_lock);
346 int dm_open_count(struct mapped_device *md)
348 return atomic_read(&md->open_count);
352 * Guarantees nothing is using the device before it's deleted.
354 int dm_lock_for_deletion(struct mapped_device *md, bool mark_deferred, bool only_deferred)
358 spin_lock(&_minor_lock);
360 if (dm_open_count(md)) {
363 set_bit(DMF_DEFERRED_REMOVE, &md->flags);
364 } else if (only_deferred && !test_bit(DMF_DEFERRED_REMOVE, &md->flags))
367 set_bit(DMF_DELETING, &md->flags);
369 spin_unlock(&_minor_lock);
374 int dm_cancel_deferred_remove(struct mapped_device *md)
378 spin_lock(&_minor_lock);
380 if (test_bit(DMF_DELETING, &md->flags))
383 clear_bit(DMF_DEFERRED_REMOVE, &md->flags);
385 spin_unlock(&_minor_lock);
390 static void do_deferred_remove(struct work_struct *w)
392 dm_deferred_remove();
395 sector_t dm_get_size(struct mapped_device *md)
397 return get_capacity(md->disk);
400 struct request_queue *dm_get_md_queue(struct mapped_device *md)
405 struct dm_stats *dm_get_stats(struct mapped_device *md)
410 static int dm_blk_getgeo(struct block_device *bdev, struct hd_geometry *geo)
412 struct mapped_device *md = bdev->bd_disk->private_data;
414 return dm_get_geometry(md, geo);
417 static int dm_grab_bdev_for_ioctl(struct mapped_device *md,
418 struct block_device **bdev,
421 struct dm_target *tgt;
422 struct dm_table *map;
427 map = dm_get_live_table(md, &srcu_idx);
428 if (!map || !dm_table_get_size(map))
431 /* We only support devices that have a single target */
432 if (dm_table_get_num_targets(map) != 1)
435 tgt = dm_table_get_target(map, 0);
436 if (!tgt->type->prepare_ioctl)
439 if (dm_suspended_md(md)) {
444 r = tgt->type->prepare_ioctl(tgt, bdev, mode);
449 dm_put_live_table(md, srcu_idx);
453 dm_put_live_table(md, srcu_idx);
454 if (r == -ENOTCONN && !fatal_signal_pending(current)) {
461 static int dm_blk_ioctl(struct block_device *bdev, fmode_t mode,
462 unsigned int cmd, unsigned long arg)
464 struct mapped_device *md = bdev->bd_disk->private_data;
467 r = dm_grab_bdev_for_ioctl(md, &bdev, &mode);
473 * Target determined this ioctl is being issued against a
474 * subset of the parent bdev; require extra privileges.
476 if (!capable(CAP_SYS_RAWIO)) {
478 "%s: sending ioctl %x to DM device without required privilege.",
485 r = __blkdev_driver_ioctl(bdev, mode, cmd, arg);
491 static struct dm_io *alloc_io(struct mapped_device *md)
493 return mempool_alloc(md->io_pool, GFP_NOIO);
496 static void free_io(struct mapped_device *md, struct dm_io *io)
498 mempool_free(io, md->io_pool);
501 static void free_tio(struct dm_target_io *tio)
503 bio_put(&tio->clone);
506 int md_in_flight(struct mapped_device *md)
508 return atomic_read(&md->pending[READ]) +
509 atomic_read(&md->pending[WRITE]);
512 static void start_io_acct(struct dm_io *io)
514 struct mapped_device *md = io->md;
515 struct bio *bio = io->bio;
517 int rw = bio_data_dir(bio);
519 io->start_time = jiffies;
521 cpu = part_stat_lock();
522 part_round_stats(cpu, &dm_disk(md)->part0);
524 atomic_set(&dm_disk(md)->part0.in_flight[rw],
525 atomic_inc_return(&md->pending[rw]));
527 if (unlikely(dm_stats_used(&md->stats)))
528 dm_stats_account_io(&md->stats, bio_data_dir(bio),
529 bio->bi_iter.bi_sector, bio_sectors(bio),
530 false, 0, &io->stats_aux);
533 static void end_io_acct(struct dm_io *io)
535 struct mapped_device *md = io->md;
536 struct bio *bio = io->bio;
537 unsigned long duration = jiffies - io->start_time;
539 int rw = bio_data_dir(bio);
541 generic_end_io_acct(rw, &dm_disk(md)->part0, io->start_time);
543 if (unlikely(dm_stats_used(&md->stats)))
544 dm_stats_account_io(&md->stats, bio_data_dir(bio),
545 bio->bi_iter.bi_sector, bio_sectors(bio),
546 true, duration, &io->stats_aux);
549 * After this is decremented the bio must not be touched if it is
552 pending = atomic_dec_return(&md->pending[rw]);
553 atomic_set(&dm_disk(md)->part0.in_flight[rw], pending);
554 pending += atomic_read(&md->pending[rw^0x1]);
556 /* nudge anyone waiting on suspend queue */
562 * Add the bio to the list of deferred io.
564 static void queue_io(struct mapped_device *md, struct bio *bio)
568 spin_lock_irqsave(&md->deferred_lock, flags);
569 bio_list_add(&md->deferred, bio);
570 spin_unlock_irqrestore(&md->deferred_lock, flags);
571 queue_work(md->wq, &md->work);
575 * Everyone (including functions in this file), should use this
576 * function to access the md->map field, and make sure they call
577 * dm_put_live_table() when finished.
579 struct dm_table *dm_get_live_table(struct mapped_device *md, int *srcu_idx) __acquires(md->io_barrier)
581 *srcu_idx = srcu_read_lock(&md->io_barrier);
583 return srcu_dereference(md->map, &md->io_barrier);
586 void dm_put_live_table(struct mapped_device *md, int srcu_idx) __releases(md->io_barrier)
588 srcu_read_unlock(&md->io_barrier, srcu_idx);
591 void dm_sync_table(struct mapped_device *md)
593 synchronize_srcu(&md->io_barrier);
594 synchronize_rcu_expedited();
598 * A fast alternative to dm_get_live_table/dm_put_live_table.
599 * The caller must not block between these two functions.
601 static struct dm_table *dm_get_live_table_fast(struct mapped_device *md) __acquires(RCU)
604 return rcu_dereference(md->map);
607 static void dm_put_live_table_fast(struct mapped_device *md) __releases(RCU)
613 * Open a table device so we can use it as a map destination.
615 static int open_table_device(struct table_device *td, dev_t dev,
616 struct mapped_device *md)
618 static char *_claim_ptr = "I belong to device-mapper";
619 struct block_device *bdev;
623 BUG_ON(td->dm_dev.bdev);
625 bdev = blkdev_get_by_dev(dev, td->dm_dev.mode | FMODE_EXCL, _claim_ptr);
627 return PTR_ERR(bdev);
629 r = bd_link_disk_holder(bdev, dm_disk(md));
631 blkdev_put(bdev, td->dm_dev.mode | FMODE_EXCL);
635 td->dm_dev.bdev = bdev;
636 td->dm_dev.dax_dev = dax_get_by_host(bdev->bd_disk->disk_name);
641 * Close a table device that we've been using.
643 static void close_table_device(struct table_device *td, struct mapped_device *md)
645 if (!td->dm_dev.bdev)
648 bd_unlink_disk_holder(td->dm_dev.bdev, dm_disk(md));
649 blkdev_put(td->dm_dev.bdev, td->dm_dev.mode | FMODE_EXCL);
650 put_dax(td->dm_dev.dax_dev);
651 td->dm_dev.bdev = NULL;
652 td->dm_dev.dax_dev = NULL;
655 static struct table_device *find_table_device(struct list_head *l, dev_t dev,
657 struct table_device *td;
659 list_for_each_entry(td, l, list)
660 if (td->dm_dev.bdev->bd_dev == dev && td->dm_dev.mode == mode)
666 int dm_get_table_device(struct mapped_device *md, dev_t dev, fmode_t mode,
667 struct dm_dev **result) {
669 struct table_device *td;
671 mutex_lock(&md->table_devices_lock);
672 td = find_table_device(&md->table_devices, dev, mode);
674 td = kmalloc_node(sizeof(*td), GFP_KERNEL, md->numa_node_id);
676 mutex_unlock(&md->table_devices_lock);
680 td->dm_dev.mode = mode;
681 td->dm_dev.bdev = NULL;
683 if ((r = open_table_device(td, dev, md))) {
684 mutex_unlock(&md->table_devices_lock);
689 format_dev_t(td->dm_dev.name, dev);
691 atomic_set(&td->count, 0);
692 list_add(&td->list, &md->table_devices);
694 atomic_inc(&td->count);
695 mutex_unlock(&md->table_devices_lock);
697 *result = &td->dm_dev;
700 EXPORT_SYMBOL_GPL(dm_get_table_device);
702 void dm_put_table_device(struct mapped_device *md, struct dm_dev *d)
704 struct table_device *td = container_of(d, struct table_device, dm_dev);
706 mutex_lock(&md->table_devices_lock);
707 if (atomic_dec_and_test(&td->count)) {
708 close_table_device(td, md);
712 mutex_unlock(&md->table_devices_lock);
714 EXPORT_SYMBOL(dm_put_table_device);
716 static void free_table_devices(struct list_head *devices)
718 struct list_head *tmp, *next;
720 list_for_each_safe(tmp, next, devices) {
721 struct table_device *td = list_entry(tmp, struct table_device, list);
723 DMWARN("dm_destroy: %s still exists with %d references",
724 td->dm_dev.name, atomic_read(&td->count));
730 * Get the geometry associated with a dm device
732 int dm_get_geometry(struct mapped_device *md, struct hd_geometry *geo)
740 * Set the geometry of a device.
742 int dm_set_geometry(struct mapped_device *md, struct hd_geometry *geo)
744 sector_t sz = (sector_t)geo->cylinders * geo->heads * geo->sectors;
746 if (geo->start > sz) {
747 DMWARN("Start sector is beyond the geometry limits.");
756 /*-----------------------------------------------------------------
758 * A more elegant soln is in the works that uses the queue
759 * merge fn, unfortunately there are a couple of changes to
760 * the block layer that I want to make for this. So in the
761 * interests of getting something for people to use I give
762 * you this clearly demarcated crap.
763 *---------------------------------------------------------------*/
765 static int __noflush_suspending(struct mapped_device *md)
767 return test_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
771 * Decrements the number of outstanding ios that a bio has been
772 * cloned into, completing the original io if necc.
774 static void dec_pending(struct dm_io *io, blk_status_t error)
777 blk_status_t io_error;
779 struct mapped_device *md = io->md;
781 /* Push-back supersedes any I/O errors */
782 if (unlikely(error)) {
783 spin_lock_irqsave(&io->endio_lock, flags);
784 if (!(io->status == BLK_STS_DM_REQUEUE &&
785 __noflush_suspending(md)))
787 spin_unlock_irqrestore(&io->endio_lock, flags);
790 if (atomic_dec_and_test(&io->io_count)) {
791 if (io->status == BLK_STS_DM_REQUEUE) {
793 * Target requested pushing back the I/O.
795 spin_lock_irqsave(&md->deferred_lock, flags);
796 if (__noflush_suspending(md))
797 bio_list_add_head(&md->deferred, io->bio);
799 /* noflush suspend was interrupted. */
800 io->status = BLK_STS_IOERR;
801 spin_unlock_irqrestore(&md->deferred_lock, flags);
804 io_error = io->status;
809 if (io_error == BLK_STS_DM_REQUEUE)
812 if ((bio->bi_opf & REQ_PREFLUSH) && bio->bi_iter.bi_size) {
814 * Preflush done for flush with data, reissue
815 * without REQ_PREFLUSH.
817 bio->bi_opf &= ~REQ_PREFLUSH;
820 /* done with normal IO or empty flush */
821 bio->bi_status = io_error;
827 void disable_write_same(struct mapped_device *md)
829 struct queue_limits *limits = dm_get_queue_limits(md);
831 /* device doesn't really support WRITE SAME, disable it */
832 limits->max_write_same_sectors = 0;
835 void disable_write_zeroes(struct mapped_device *md)
837 struct queue_limits *limits = dm_get_queue_limits(md);
839 /* device doesn't really support WRITE ZEROES, disable it */
840 limits->max_write_zeroes_sectors = 0;
843 static void clone_endio(struct bio *bio)
845 blk_status_t error = bio->bi_status;
846 struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone);
847 struct dm_io *io = tio->io;
848 struct mapped_device *md = tio->io->md;
849 dm_endio_fn endio = tio->ti->type->end_io;
851 if (unlikely(error == BLK_STS_TARGET)) {
852 if (bio_op(bio) == REQ_OP_WRITE_SAME &&
853 !bdev_get_queue(bio->bi_bdev)->limits.max_write_same_sectors)
854 disable_write_same(md);
855 if (bio_op(bio) == REQ_OP_WRITE_ZEROES &&
856 !bdev_get_queue(bio->bi_bdev)->limits.max_write_zeroes_sectors)
857 disable_write_zeroes(md);
861 int r = endio(tio->ti, bio, &error);
863 case DM_ENDIO_REQUEUE:
864 error = BLK_STS_DM_REQUEUE;
868 case DM_ENDIO_INCOMPLETE:
869 /* The target will handle the io */
872 DMWARN("unimplemented target endio return value: %d", r);
878 dec_pending(io, error);
882 * Return maximum size of I/O possible at the supplied sector up to the current
885 static sector_t max_io_len_target_boundary(sector_t sector, struct dm_target *ti)
887 sector_t target_offset = dm_target_offset(ti, sector);
889 return ti->len - target_offset;
892 static sector_t max_io_len(sector_t sector, struct dm_target *ti)
894 sector_t len = max_io_len_target_boundary(sector, ti);
895 sector_t offset, max_len;
898 * Does the target need to split even further?
900 if (ti->max_io_len) {
901 offset = dm_target_offset(ti, sector);
902 if (unlikely(ti->max_io_len & (ti->max_io_len - 1)))
903 max_len = sector_div(offset, ti->max_io_len);
905 max_len = offset & (ti->max_io_len - 1);
906 max_len = ti->max_io_len - max_len;
915 int dm_set_target_max_io_len(struct dm_target *ti, sector_t len)
917 if (len > UINT_MAX) {
918 DMERR("Specified maximum size of target IO (%llu) exceeds limit (%u)",
919 (unsigned long long)len, UINT_MAX);
920 ti->error = "Maximum size of target IO is too large";
924 ti->max_io_len = (uint32_t) len;
928 EXPORT_SYMBOL_GPL(dm_set_target_max_io_len);
930 static struct dm_target *dm_dax_get_live_target(struct mapped_device *md,
931 sector_t sector, int *srcu_idx)
933 struct dm_table *map;
934 struct dm_target *ti;
936 map = dm_get_live_table(md, srcu_idx);
940 ti = dm_table_find_target(map, sector);
941 if (!dm_target_is_valid(ti))
947 static long dm_dax_direct_access(struct dax_device *dax_dev, pgoff_t pgoff,
948 long nr_pages, void **kaddr, pfn_t *pfn)
950 struct mapped_device *md = dax_get_private(dax_dev);
951 sector_t sector = pgoff * PAGE_SECTORS;
952 struct dm_target *ti;
953 long len, ret = -EIO;
956 ti = dm_dax_get_live_target(md, sector, &srcu_idx);
960 if (!ti->type->direct_access)
962 len = max_io_len(sector, ti) / PAGE_SECTORS;
965 nr_pages = min(len, nr_pages);
966 if (ti->type->direct_access)
967 ret = ti->type->direct_access(ti, pgoff, nr_pages, kaddr, pfn);
970 dm_put_live_table(md, srcu_idx);
976 * A target may call dm_accept_partial_bio only from the map routine. It is
977 * allowed for all bio types except REQ_PREFLUSH.
979 * dm_accept_partial_bio informs the dm that the target only wants to process
980 * additional n_sectors sectors of the bio and the rest of the data should be
981 * sent in a next bio.
983 * A diagram that explains the arithmetics:
984 * +--------------------+---------------+-------+
986 * +--------------------+---------------+-------+
988 * <-------------- *tio->len_ptr --------------->
989 * <------- bi_size ------->
992 * Region 1 was already iterated over with bio_advance or similar function.
993 * (it may be empty if the target doesn't use bio_advance)
994 * Region 2 is the remaining bio size that the target wants to process.
995 * (it may be empty if region 1 is non-empty, although there is no reason
997 * The target requires that region 3 is to be sent in the next bio.
999 * If the target wants to receive multiple copies of the bio (via num_*bios, etc),
1000 * the partially processed part (the sum of regions 1+2) must be the same for all
1001 * copies of the bio.
1003 void dm_accept_partial_bio(struct bio *bio, unsigned n_sectors)
1005 struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone);
1006 unsigned bi_size = bio->bi_iter.bi_size >> SECTOR_SHIFT;
1007 BUG_ON(bio->bi_opf & REQ_PREFLUSH);
1008 BUG_ON(bi_size > *tio->len_ptr);
1009 BUG_ON(n_sectors > bi_size);
1010 *tio->len_ptr -= bi_size - n_sectors;
1011 bio->bi_iter.bi_size = n_sectors << SECTOR_SHIFT;
1013 EXPORT_SYMBOL_GPL(dm_accept_partial_bio);
1016 * Flush current->bio_list when the target map method blocks.
1017 * This fixes deadlocks in snapshot and possibly in other targets.
1020 struct blk_plug plug;
1021 struct blk_plug_cb cb;
1024 static void flush_current_bio_list(struct blk_plug_cb *cb, bool from_schedule)
1026 struct dm_offload *o = container_of(cb, struct dm_offload, cb);
1027 struct bio_list list;
1031 INIT_LIST_HEAD(&o->cb.list);
1033 if (unlikely(!current->bio_list))
1036 for (i = 0; i < 2; i++) {
1037 list = current->bio_list[i];
1038 bio_list_init(¤t->bio_list[i]);
1040 while ((bio = bio_list_pop(&list))) {
1041 struct bio_set *bs = bio->bi_pool;
1042 if (unlikely(!bs) || bs == fs_bio_set ||
1043 !bs->rescue_workqueue) {
1044 bio_list_add(¤t->bio_list[i], bio);
1048 spin_lock(&bs->rescue_lock);
1049 bio_list_add(&bs->rescue_list, bio);
1050 queue_work(bs->rescue_workqueue, &bs->rescue_work);
1051 spin_unlock(&bs->rescue_lock);
1056 static void dm_offload_start(struct dm_offload *o)
1058 blk_start_plug(&o->plug);
1059 o->cb.callback = flush_current_bio_list;
1060 list_add(&o->cb.list, ¤t->plug->cb_list);
1063 static void dm_offload_end(struct dm_offload *o)
1065 list_del(&o->cb.list);
1066 blk_finish_plug(&o->plug);
1069 static void __map_bio(struct dm_target_io *tio)
1073 struct dm_offload o;
1074 struct bio *clone = &tio->clone;
1075 struct dm_target *ti = tio->ti;
1077 clone->bi_end_io = clone_endio;
1080 * Map the clone. If r == 0 we don't need to do
1081 * anything, the target has assumed ownership of
1084 atomic_inc(&tio->io->io_count);
1085 sector = clone->bi_iter.bi_sector;
1087 dm_offload_start(&o);
1088 r = ti->type->map(ti, clone);
1092 case DM_MAPIO_SUBMITTED:
1094 case DM_MAPIO_REMAPPED:
1095 /* the bio has been remapped so dispatch it */
1096 trace_block_bio_remap(bdev_get_queue(clone->bi_bdev), clone,
1097 tio->io->bio->bi_bdev->bd_dev, sector);
1098 generic_make_request(clone);
1101 dec_pending(tio->io, BLK_STS_IOERR);
1104 case DM_MAPIO_REQUEUE:
1105 dec_pending(tio->io, BLK_STS_DM_REQUEUE);
1109 DMWARN("unimplemented target map return value: %d", r);
1115 struct mapped_device *md;
1116 struct dm_table *map;
1120 unsigned sector_count;
1123 static void bio_setup_sector(struct bio *bio, sector_t sector, unsigned len)
1125 bio->bi_iter.bi_sector = sector;
1126 bio->bi_iter.bi_size = to_bytes(len);
1130 * Creates a bio that consists of range of complete bvecs.
1132 static int clone_bio(struct dm_target_io *tio, struct bio *bio,
1133 sector_t sector, unsigned len)
1135 struct bio *clone = &tio->clone;
1137 __bio_clone_fast(clone, bio);
1139 if (unlikely(bio_integrity(bio) != NULL)) {
1142 if (unlikely(!dm_target_has_integrity(tio->ti->type) &&
1143 !dm_target_passes_integrity(tio->ti->type))) {
1144 DMWARN("%s: the target %s doesn't support integrity data.",
1145 dm_device_name(tio->io->md),
1146 tio->ti->type->name);
1150 r = bio_integrity_clone(clone, bio, GFP_NOIO);
1155 bio_advance(clone, to_bytes(sector - clone->bi_iter.bi_sector));
1156 clone->bi_iter.bi_size = to_bytes(len);
1158 if (unlikely(bio_integrity(bio) != NULL))
1159 bio_integrity_trim(clone, 0, len);
1164 static struct dm_target_io *alloc_tio(struct clone_info *ci,
1165 struct dm_target *ti,
1166 unsigned target_bio_nr)
1168 struct dm_target_io *tio;
1171 clone = bio_alloc_bioset(GFP_NOIO, 0, ci->md->bs);
1172 tio = container_of(clone, struct dm_target_io, clone);
1176 tio->target_bio_nr = target_bio_nr;
1181 static void __clone_and_map_simple_bio(struct clone_info *ci,
1182 struct dm_target *ti,
1183 unsigned target_bio_nr, unsigned *len)
1185 struct dm_target_io *tio = alloc_tio(ci, ti, target_bio_nr);
1186 struct bio *clone = &tio->clone;
1190 __bio_clone_fast(clone, ci->bio);
1192 bio_setup_sector(clone, ci->sector, *len);
1197 static void __send_duplicate_bios(struct clone_info *ci, struct dm_target *ti,
1198 unsigned num_bios, unsigned *len)
1200 unsigned target_bio_nr;
1202 for (target_bio_nr = 0; target_bio_nr < num_bios; target_bio_nr++)
1203 __clone_and_map_simple_bio(ci, ti, target_bio_nr, len);
1206 static int __send_empty_flush(struct clone_info *ci)
1208 unsigned target_nr = 0;
1209 struct dm_target *ti;
1211 BUG_ON(bio_has_data(ci->bio));
1212 while ((ti = dm_table_get_target(ci->map, target_nr++)))
1213 __send_duplicate_bios(ci, ti, ti->num_flush_bios, NULL);
1218 static int __clone_and_map_data_bio(struct clone_info *ci, struct dm_target *ti,
1219 sector_t sector, unsigned *len)
1221 struct bio *bio = ci->bio;
1222 struct dm_target_io *tio;
1223 unsigned target_bio_nr;
1224 unsigned num_target_bios = 1;
1228 * Does the target want to receive duplicate copies of the bio?
1230 if (bio_data_dir(bio) == WRITE && ti->num_write_bios)
1231 num_target_bios = ti->num_write_bios(ti, bio);
1233 for (target_bio_nr = 0; target_bio_nr < num_target_bios; target_bio_nr++) {
1234 tio = alloc_tio(ci, ti, target_bio_nr);
1236 r = clone_bio(tio, bio, sector, *len);
1247 typedef unsigned (*get_num_bios_fn)(struct dm_target *ti);
1249 static unsigned get_num_discard_bios(struct dm_target *ti)
1251 return ti->num_discard_bios;
1254 static unsigned get_num_write_same_bios(struct dm_target *ti)
1256 return ti->num_write_same_bios;
1259 static unsigned get_num_write_zeroes_bios(struct dm_target *ti)
1261 return ti->num_write_zeroes_bios;
1264 typedef bool (*is_split_required_fn)(struct dm_target *ti);
1266 static bool is_split_required_for_discard(struct dm_target *ti)
1268 return ti->split_discard_bios;
1271 static int __send_changing_extent_only(struct clone_info *ci,
1272 get_num_bios_fn get_num_bios,
1273 is_split_required_fn is_split_required)
1275 struct dm_target *ti;
1280 ti = dm_table_find_target(ci->map, ci->sector);
1281 if (!dm_target_is_valid(ti))
1285 * Even though the device advertised support for this type of
1286 * request, that does not mean every target supports it, and
1287 * reconfiguration might also have changed that since the
1288 * check was performed.
1290 num_bios = get_num_bios ? get_num_bios(ti) : 0;
1294 if (is_split_required && !is_split_required(ti))
1295 len = min((sector_t)ci->sector_count, max_io_len_target_boundary(ci->sector, ti));
1297 len = min((sector_t)ci->sector_count, max_io_len(ci->sector, ti));
1299 __send_duplicate_bios(ci, ti, num_bios, &len);
1302 } while (ci->sector_count -= len);
1307 static int __send_discard(struct clone_info *ci)
1309 return __send_changing_extent_only(ci, get_num_discard_bios,
1310 is_split_required_for_discard);
1313 static int __send_write_same(struct clone_info *ci)
1315 return __send_changing_extent_only(ci, get_num_write_same_bios, NULL);
1318 static int __send_write_zeroes(struct clone_info *ci)
1320 return __send_changing_extent_only(ci, get_num_write_zeroes_bios, NULL);
1324 * Select the correct strategy for processing a non-flush bio.
1326 static int __split_and_process_non_flush(struct clone_info *ci)
1328 struct bio *bio = ci->bio;
1329 struct dm_target *ti;
1333 if (unlikely(bio_op(bio) == REQ_OP_DISCARD))
1334 return __send_discard(ci);
1335 else if (unlikely(bio_op(bio) == REQ_OP_WRITE_SAME))
1336 return __send_write_same(ci);
1337 else if (unlikely(bio_op(bio) == REQ_OP_WRITE_ZEROES))
1338 return __send_write_zeroes(ci);
1340 ti = dm_table_find_target(ci->map, ci->sector);
1341 if (!dm_target_is_valid(ti))
1344 len = min_t(sector_t, max_io_len(ci->sector, ti), ci->sector_count);
1346 r = __clone_and_map_data_bio(ci, ti, ci->sector, &len);
1351 ci->sector_count -= len;
1357 * Entry point to split a bio into clones and submit them to the targets.
1359 static void __split_and_process_bio(struct mapped_device *md,
1360 struct dm_table *map, struct bio *bio)
1362 struct clone_info ci;
1365 if (unlikely(!map)) {
1372 ci.io = alloc_io(md);
1374 atomic_set(&ci.io->io_count, 1);
1377 spin_lock_init(&ci.io->endio_lock);
1378 ci.sector = bio->bi_iter.bi_sector;
1380 start_io_acct(ci.io);
1382 if (bio->bi_opf & REQ_PREFLUSH) {
1383 ci.bio = &ci.md->flush_bio;
1384 ci.sector_count = 0;
1385 error = __send_empty_flush(&ci);
1386 /* dec_pending submits any data associated with flush */
1389 ci.sector_count = bio_sectors(bio);
1390 while (ci.sector_count && !error)
1391 error = __split_and_process_non_flush(&ci);
1394 /* drop the extra reference count */
1395 dec_pending(ci.io, error);
1397 /*-----------------------------------------------------------------
1399 *---------------------------------------------------------------*/
1402 * The request function that just remaps the bio built up by
1405 static blk_qc_t dm_make_request(struct request_queue *q, struct bio *bio)
1407 int rw = bio_data_dir(bio);
1408 struct mapped_device *md = q->queuedata;
1410 struct dm_table *map;
1412 map = dm_get_live_table(md, &srcu_idx);
1414 generic_start_io_acct(rw, bio_sectors(bio), &dm_disk(md)->part0);
1416 /* if we're suspended, we have to queue this io for later */
1417 if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags))) {
1418 dm_put_live_table(md, srcu_idx);
1420 if (!(bio->bi_opf & REQ_RAHEAD))
1424 return BLK_QC_T_NONE;
1427 __split_and_process_bio(md, map, bio);
1428 dm_put_live_table(md, srcu_idx);
1429 return BLK_QC_T_NONE;
1432 static int dm_any_congested(void *congested_data, int bdi_bits)
1435 struct mapped_device *md = congested_data;
1436 struct dm_table *map;
1438 if (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
1439 if (dm_request_based(md)) {
1441 * With request-based DM we only need to check the
1442 * top-level queue for congestion.
1444 r = md->queue->backing_dev_info->wb.state & bdi_bits;
1446 map = dm_get_live_table_fast(md);
1448 r = dm_table_any_congested(map, bdi_bits);
1449 dm_put_live_table_fast(md);
1456 /*-----------------------------------------------------------------
1457 * An IDR is used to keep track of allocated minor numbers.
1458 *---------------------------------------------------------------*/
1459 static void free_minor(int minor)
1461 spin_lock(&_minor_lock);
1462 idr_remove(&_minor_idr, minor);
1463 spin_unlock(&_minor_lock);
1467 * See if the device with a specific minor # is free.
1469 static int specific_minor(int minor)
1473 if (minor >= (1 << MINORBITS))
1476 idr_preload(GFP_KERNEL);
1477 spin_lock(&_minor_lock);
1479 r = idr_alloc(&_minor_idr, MINOR_ALLOCED, minor, minor + 1, GFP_NOWAIT);
1481 spin_unlock(&_minor_lock);
1484 return r == -ENOSPC ? -EBUSY : r;
1488 static int next_free_minor(int *minor)
1492 idr_preload(GFP_KERNEL);
1493 spin_lock(&_minor_lock);
1495 r = idr_alloc(&_minor_idr, MINOR_ALLOCED, 0, 1 << MINORBITS, GFP_NOWAIT);
1497 spin_unlock(&_minor_lock);
1505 static const struct block_device_operations dm_blk_dops;
1506 static const struct dax_operations dm_dax_ops;
1508 static void dm_wq_work(struct work_struct *work);
1510 void dm_init_md_queue(struct mapped_device *md)
1513 * Request-based dm devices cannot be stacked on top of bio-based dm
1514 * devices. The type of this dm device may not have been decided yet.
1515 * The type is decided at the first table loading time.
1516 * To prevent problematic device stacking, clear the queue flag
1517 * for request stacking support until then.
1519 * This queue is new, so no concurrency on the queue_flags.
1521 queue_flag_clear_unlocked(QUEUE_FLAG_STACKABLE, md->queue);
1524 * Initialize data that will only be used by a non-blk-mq DM queue
1525 * - must do so here (in alloc_dev callchain) before queue is used
1527 md->queue->queuedata = md;
1528 md->queue->backing_dev_info->congested_data = md;
1531 void dm_init_normal_md_queue(struct mapped_device *md)
1533 md->use_blk_mq = false;
1534 dm_init_md_queue(md);
1537 * Initialize aspects of queue that aren't relevant for blk-mq
1539 md->queue->backing_dev_info->congested_fn = dm_any_congested;
1540 blk_queue_bounce_limit(md->queue, BLK_BOUNCE_ANY);
1543 static void cleanup_mapped_device(struct mapped_device *md)
1546 destroy_workqueue(md->wq);
1547 if (md->kworker_task)
1548 kthread_stop(md->kworker_task);
1549 mempool_destroy(md->io_pool);
1551 bioset_free(md->bs);
1554 kill_dax(md->dax_dev);
1555 put_dax(md->dax_dev);
1560 spin_lock(&_minor_lock);
1561 md->disk->private_data = NULL;
1562 spin_unlock(&_minor_lock);
1563 del_gendisk(md->disk);
1568 blk_cleanup_queue(md->queue);
1570 cleanup_srcu_struct(&md->io_barrier);
1577 dm_mq_cleanup_mapped_device(md);
1581 * Allocate and initialise a blank device with a given minor.
1583 static struct mapped_device *alloc_dev(int minor)
1585 int r, numa_node_id = dm_get_numa_node();
1586 struct dax_device *dax_dev;
1587 struct mapped_device *md;
1590 md = kzalloc_node(sizeof(*md), GFP_KERNEL, numa_node_id);
1592 DMWARN("unable to allocate device, out of memory.");
1596 if (!try_module_get(THIS_MODULE))
1597 goto bad_module_get;
1599 /* get a minor number for the dev */
1600 if (minor == DM_ANY_MINOR)
1601 r = next_free_minor(&minor);
1603 r = specific_minor(minor);
1607 r = init_srcu_struct(&md->io_barrier);
1609 goto bad_io_barrier;
1611 md->numa_node_id = numa_node_id;
1612 md->use_blk_mq = dm_use_blk_mq_default();
1613 md->init_tio_pdu = false;
1614 md->type = DM_TYPE_NONE;
1615 mutex_init(&md->suspend_lock);
1616 mutex_init(&md->type_lock);
1617 mutex_init(&md->table_devices_lock);
1618 spin_lock_init(&md->deferred_lock);
1619 atomic_set(&md->holders, 1);
1620 atomic_set(&md->open_count, 0);
1621 atomic_set(&md->event_nr, 0);
1622 atomic_set(&md->uevent_seq, 0);
1623 INIT_LIST_HEAD(&md->uevent_list);
1624 INIT_LIST_HEAD(&md->table_devices);
1625 spin_lock_init(&md->uevent_lock);
1627 md->queue = blk_alloc_queue_node(GFP_KERNEL, numa_node_id);
1631 dm_init_md_queue(md);
1633 md->disk = alloc_disk_node(1, numa_node_id);
1637 atomic_set(&md->pending[0], 0);
1638 atomic_set(&md->pending[1], 0);
1639 init_waitqueue_head(&md->wait);
1640 INIT_WORK(&md->work, dm_wq_work);
1641 init_waitqueue_head(&md->eventq);
1642 init_completion(&md->kobj_holder.completion);
1643 md->kworker_task = NULL;
1645 md->disk->major = _major;
1646 md->disk->first_minor = minor;
1647 md->disk->fops = &dm_blk_dops;
1648 md->disk->queue = md->queue;
1649 md->disk->private_data = md;
1650 sprintf(md->disk->disk_name, "dm-%d", minor);
1652 dax_dev = alloc_dax(md, md->disk->disk_name, &dm_dax_ops);
1655 md->dax_dev = dax_dev;
1658 format_dev_t(md->name, MKDEV(_major, minor));
1660 md->wq = alloc_workqueue("kdmflush", WQ_MEM_RECLAIM, 0);
1664 md->bdev = bdget_disk(md->disk, 0);
1668 bio_init(&md->flush_bio, NULL, 0);
1669 md->flush_bio.bi_bdev = md->bdev;
1670 md->flush_bio.bi_opf = REQ_OP_WRITE | REQ_PREFLUSH | REQ_SYNC;
1672 dm_stats_init(&md->stats);
1674 /* Populate the mapping, nobody knows we exist yet */
1675 spin_lock(&_minor_lock);
1676 old_md = idr_replace(&_minor_idr, md, minor);
1677 spin_unlock(&_minor_lock);
1679 BUG_ON(old_md != MINOR_ALLOCED);
1684 cleanup_mapped_device(md);
1688 module_put(THIS_MODULE);
1694 static void unlock_fs(struct mapped_device *md);
1696 static void free_dev(struct mapped_device *md)
1698 int minor = MINOR(disk_devt(md->disk));
1702 cleanup_mapped_device(md);
1704 free_table_devices(&md->table_devices);
1705 dm_stats_cleanup(&md->stats);
1708 module_put(THIS_MODULE);
1712 static void __bind_mempools(struct mapped_device *md, struct dm_table *t)
1714 struct dm_md_mempools *p = dm_table_get_md_mempools(t);
1717 /* The md already has necessary mempools. */
1718 if (dm_table_bio_based(t)) {
1720 * Reload bioset because front_pad may have changed
1721 * because a different table was loaded.
1723 bioset_free(md->bs);
1728 * There's no need to reload with request-based dm
1729 * because the size of front_pad doesn't change.
1730 * Note for future: If you are to reload bioset,
1731 * prep-ed requests in the queue may refer
1732 * to bio from the old bioset, so you must walk
1733 * through the queue to unprep.
1738 BUG_ON(!p || md->io_pool || md->bs);
1740 md->io_pool = p->io_pool;
1746 /* mempool bind completed, no longer need any mempools in the table */
1747 dm_table_free_md_mempools(t);
1751 * Bind a table to the device.
1753 static void event_callback(void *context)
1755 unsigned long flags;
1757 struct mapped_device *md = (struct mapped_device *) context;
1759 spin_lock_irqsave(&md->uevent_lock, flags);
1760 list_splice_init(&md->uevent_list, &uevents);
1761 spin_unlock_irqrestore(&md->uevent_lock, flags);
1763 dm_send_uevents(&uevents, &disk_to_dev(md->disk)->kobj);
1765 atomic_inc(&md->event_nr);
1766 atomic_inc(&dm_global_event_nr);
1767 wake_up(&md->eventq);
1768 wake_up(&dm_global_eventq);
1772 * Protected by md->suspend_lock obtained by dm_swap_table().
1774 static void __set_size(struct mapped_device *md, sector_t size)
1776 lockdep_assert_held(&md->suspend_lock);
1778 set_capacity(md->disk, size);
1780 i_size_write(md->bdev->bd_inode, (loff_t)size << SECTOR_SHIFT);
1784 * Returns old map, which caller must destroy.
1786 static struct dm_table *__bind(struct mapped_device *md, struct dm_table *t,
1787 struct queue_limits *limits)
1789 struct dm_table *old_map;
1790 struct request_queue *q = md->queue;
1793 lockdep_assert_held(&md->suspend_lock);
1795 size = dm_table_get_size(t);
1798 * Wipe any geometry if the size of the table changed.
1800 if (size != dm_get_size(md))
1801 memset(&md->geometry, 0, sizeof(md->geometry));
1803 __set_size(md, size);
1805 dm_table_event_callback(t, event_callback, md);
1808 * The queue hasn't been stopped yet, if the old table type wasn't
1809 * for request-based during suspension. So stop it to prevent
1810 * I/O mapping before resume.
1811 * This must be done before setting the queue restrictions,
1812 * because request-based dm may be run just after the setting.
1814 if (dm_table_request_based(t)) {
1817 * Leverage the fact that request-based DM targets are
1818 * immutable singletons and establish md->immutable_target
1819 * - used to optimize both dm_request_fn and dm_mq_queue_rq
1821 md->immutable_target = dm_table_get_immutable_target(t);
1824 __bind_mempools(md, t);
1826 old_map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
1827 rcu_assign_pointer(md->map, (void *)t);
1828 md->immutable_target_type = dm_table_get_immutable_target_type(t);
1830 dm_table_set_restrictions(t, q, limits);
1838 * Returns unbound table for the caller to free.
1840 static struct dm_table *__unbind(struct mapped_device *md)
1842 struct dm_table *map = rcu_dereference_protected(md->map, 1);
1847 dm_table_event_callback(map, NULL, NULL);
1848 RCU_INIT_POINTER(md->map, NULL);
1855 * Constructor for a new device.
1857 int dm_create(int minor, struct mapped_device **result)
1859 struct mapped_device *md;
1861 md = alloc_dev(minor);
1872 * Functions to manage md->type.
1873 * All are required to hold md->type_lock.
1875 void dm_lock_md_type(struct mapped_device *md)
1877 mutex_lock(&md->type_lock);
1880 void dm_unlock_md_type(struct mapped_device *md)
1882 mutex_unlock(&md->type_lock);
1885 void dm_set_md_type(struct mapped_device *md, enum dm_queue_mode type)
1887 BUG_ON(!mutex_is_locked(&md->type_lock));
1891 enum dm_queue_mode dm_get_md_type(struct mapped_device *md)
1896 struct target_type *dm_get_immutable_target_type(struct mapped_device *md)
1898 return md->immutable_target_type;
1902 * The queue_limits are only valid as long as you have a reference
1905 struct queue_limits *dm_get_queue_limits(struct mapped_device *md)
1907 BUG_ON(!atomic_read(&md->holders));
1908 return &md->queue->limits;
1910 EXPORT_SYMBOL_GPL(dm_get_queue_limits);
1913 * Setup the DM device's queue based on md's type
1915 int dm_setup_md_queue(struct mapped_device *md, struct dm_table *t)
1918 enum dm_queue_mode type = dm_get_md_type(md);
1921 case DM_TYPE_REQUEST_BASED:
1922 r = dm_old_init_request_queue(md, t);
1924 DMERR("Cannot initialize queue for request-based mapped device");
1928 case DM_TYPE_MQ_REQUEST_BASED:
1929 r = dm_mq_init_request_queue(md, t);
1931 DMERR("Cannot initialize queue for request-based dm-mq mapped device");
1935 case DM_TYPE_BIO_BASED:
1936 case DM_TYPE_DAX_BIO_BASED:
1937 dm_init_normal_md_queue(md);
1938 blk_queue_make_request(md->queue, dm_make_request);
1940 * DM handles splitting bios as needed. Free the bio_split bioset
1941 * since it won't be used (saves 1 process per bio-based DM device).
1943 bioset_free(md->queue->bio_split);
1944 md->queue->bio_split = NULL;
1946 if (type == DM_TYPE_DAX_BIO_BASED)
1947 queue_flag_set_unlocked(QUEUE_FLAG_DAX, md->queue);
1957 struct mapped_device *dm_get_md(dev_t dev)
1959 struct mapped_device *md;
1960 unsigned minor = MINOR(dev);
1962 if (MAJOR(dev) != _major || minor >= (1 << MINORBITS))
1965 spin_lock(&_minor_lock);
1967 md = idr_find(&_minor_idr, minor);
1969 if ((md == MINOR_ALLOCED ||
1970 (MINOR(disk_devt(dm_disk(md))) != minor) ||
1971 dm_deleting_md(md) ||
1972 test_bit(DMF_FREEING, &md->flags))) {
1980 spin_unlock(&_minor_lock);
1984 EXPORT_SYMBOL_GPL(dm_get_md);
1986 void *dm_get_mdptr(struct mapped_device *md)
1988 return md->interface_ptr;
1991 void dm_set_mdptr(struct mapped_device *md, void *ptr)
1993 md->interface_ptr = ptr;
1996 void dm_get(struct mapped_device *md)
1998 atomic_inc(&md->holders);
1999 BUG_ON(test_bit(DMF_FREEING, &md->flags));
2002 int dm_hold(struct mapped_device *md)
2004 spin_lock(&_minor_lock);
2005 if (test_bit(DMF_FREEING, &md->flags)) {
2006 spin_unlock(&_minor_lock);
2010 spin_unlock(&_minor_lock);
2013 EXPORT_SYMBOL_GPL(dm_hold);
2015 const char *dm_device_name(struct mapped_device *md)
2019 EXPORT_SYMBOL_GPL(dm_device_name);
2021 static void __dm_destroy(struct mapped_device *md, bool wait)
2023 struct request_queue *q = dm_get_md_queue(md);
2024 struct dm_table *map;
2029 spin_lock(&_minor_lock);
2030 idr_replace(&_minor_idr, MINOR_ALLOCED, MINOR(disk_devt(dm_disk(md))));
2031 set_bit(DMF_FREEING, &md->flags);
2032 spin_unlock(&_minor_lock);
2034 blk_set_queue_dying(q);
2036 if (dm_request_based(md) && md->kworker_task)
2037 kthread_flush_worker(&md->kworker);
2040 * Take suspend_lock so that presuspend and postsuspend methods
2041 * do not race with internal suspend.
2043 mutex_lock(&md->suspend_lock);
2044 map = dm_get_live_table(md, &srcu_idx);
2045 if (!dm_suspended_md(md)) {
2046 dm_table_presuspend_targets(map);
2047 dm_table_postsuspend_targets(map);
2049 /* dm_put_live_table must be before msleep, otherwise deadlock is possible */
2050 dm_put_live_table(md, srcu_idx);
2051 mutex_unlock(&md->suspend_lock);
2054 * Rare, but there may be I/O requests still going to complete,
2055 * for example. Wait for all references to disappear.
2056 * No one should increment the reference count of the mapped_device,
2057 * after the mapped_device state becomes DMF_FREEING.
2060 while (atomic_read(&md->holders))
2062 else if (atomic_read(&md->holders))
2063 DMWARN("%s: Forcibly removing mapped_device still in use! (%d users)",
2064 dm_device_name(md), atomic_read(&md->holders));
2067 dm_table_destroy(__unbind(md));
2071 void dm_destroy(struct mapped_device *md)
2073 __dm_destroy(md, true);
2076 void dm_destroy_immediate(struct mapped_device *md)
2078 __dm_destroy(md, false);
2081 void dm_put(struct mapped_device *md)
2083 atomic_dec(&md->holders);
2085 EXPORT_SYMBOL_GPL(dm_put);
2087 static int dm_wait_for_completion(struct mapped_device *md, long task_state)
2093 prepare_to_wait(&md->wait, &wait, task_state);
2095 if (!md_in_flight(md))
2098 if (signal_pending_state(task_state, current)) {
2105 finish_wait(&md->wait, &wait);
2111 * Process the deferred bios
2113 static void dm_wq_work(struct work_struct *work)
2115 struct mapped_device *md = container_of(work, struct mapped_device,
2119 struct dm_table *map;
2121 map = dm_get_live_table(md, &srcu_idx);
2123 while (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
2124 spin_lock_irq(&md->deferred_lock);
2125 c = bio_list_pop(&md->deferred);
2126 spin_unlock_irq(&md->deferred_lock);
2131 if (dm_request_based(md))
2132 generic_make_request(c);
2134 __split_and_process_bio(md, map, c);
2137 dm_put_live_table(md, srcu_idx);
2140 static void dm_queue_flush(struct mapped_device *md)
2142 clear_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2143 smp_mb__after_atomic();
2144 queue_work(md->wq, &md->work);
2148 * Swap in a new table, returning the old one for the caller to destroy.
2150 struct dm_table *dm_swap_table(struct mapped_device *md, struct dm_table *table)
2152 struct dm_table *live_map = NULL, *map = ERR_PTR(-EINVAL);
2153 struct queue_limits limits;
2156 mutex_lock(&md->suspend_lock);
2158 /* device must be suspended */
2159 if (!dm_suspended_md(md))
2163 * If the new table has no data devices, retain the existing limits.
2164 * This helps multipath with queue_if_no_path if all paths disappear,
2165 * then new I/O is queued based on these limits, and then some paths
2168 if (dm_table_has_no_data_devices(table)) {
2169 live_map = dm_get_live_table_fast(md);
2171 limits = md->queue->limits;
2172 dm_put_live_table_fast(md);
2176 r = dm_calculate_queue_limits(table, &limits);
2183 map = __bind(md, table, &limits);
2186 mutex_unlock(&md->suspend_lock);
2191 * Functions to lock and unlock any filesystem running on the
2194 static int lock_fs(struct mapped_device *md)
2198 WARN_ON(md->frozen_sb);
2200 md->frozen_sb = freeze_bdev(md->bdev);
2201 if (IS_ERR(md->frozen_sb)) {
2202 r = PTR_ERR(md->frozen_sb);
2203 md->frozen_sb = NULL;
2207 set_bit(DMF_FROZEN, &md->flags);
2212 static void unlock_fs(struct mapped_device *md)
2214 if (!test_bit(DMF_FROZEN, &md->flags))
2217 thaw_bdev(md->bdev, md->frozen_sb);
2218 md->frozen_sb = NULL;
2219 clear_bit(DMF_FROZEN, &md->flags);
2223 * @suspend_flags: DM_SUSPEND_LOCKFS_FLAG and/or DM_SUSPEND_NOFLUSH_FLAG
2224 * @task_state: e.g. TASK_INTERRUPTIBLE or TASK_UNINTERRUPTIBLE
2225 * @dmf_suspended_flag: DMF_SUSPENDED or DMF_SUSPENDED_INTERNALLY
2227 * If __dm_suspend returns 0, the device is completely quiescent
2228 * now. There is no request-processing activity. All new requests
2229 * are being added to md->deferred list.
2231 static int __dm_suspend(struct mapped_device *md, struct dm_table *map,
2232 unsigned suspend_flags, long task_state,
2233 int dmf_suspended_flag)
2235 bool do_lockfs = suspend_flags & DM_SUSPEND_LOCKFS_FLAG;
2236 bool noflush = suspend_flags & DM_SUSPEND_NOFLUSH_FLAG;
2239 lockdep_assert_held(&md->suspend_lock);
2242 * DMF_NOFLUSH_SUSPENDING must be set before presuspend.
2243 * This flag is cleared before dm_suspend returns.
2246 set_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
2248 pr_debug("%s: suspending with flush\n", dm_device_name(md));
2251 * This gets reverted if there's an error later and the targets
2252 * provide the .presuspend_undo hook.
2254 dm_table_presuspend_targets(map);
2257 * Flush I/O to the device.
2258 * Any I/O submitted after lock_fs() may not be flushed.
2259 * noflush takes precedence over do_lockfs.
2260 * (lock_fs() flushes I/Os and waits for them to complete.)
2262 if (!noflush && do_lockfs) {
2265 dm_table_presuspend_undo_targets(map);
2271 * Here we must make sure that no processes are submitting requests
2272 * to target drivers i.e. no one may be executing
2273 * __split_and_process_bio. This is called from dm_request and
2276 * To get all processes out of __split_and_process_bio in dm_request,
2277 * we take the write lock. To prevent any process from reentering
2278 * __split_and_process_bio from dm_request and quiesce the thread
2279 * (dm_wq_work), we set BMF_BLOCK_IO_FOR_SUSPEND and call
2280 * flush_workqueue(md->wq).
2282 set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2284 synchronize_srcu(&md->io_barrier);
2287 * Stop md->queue before flushing md->wq in case request-based
2288 * dm defers requests to md->wq from md->queue.
2290 if (dm_request_based(md)) {
2291 dm_stop_queue(md->queue);
2292 if (md->kworker_task)
2293 kthread_flush_worker(&md->kworker);
2296 flush_workqueue(md->wq);
2299 * At this point no more requests are entering target request routines.
2300 * We call dm_wait_for_completion to wait for all existing requests
2303 r = dm_wait_for_completion(md, task_state);
2305 set_bit(dmf_suspended_flag, &md->flags);
2308 clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
2310 synchronize_srcu(&md->io_barrier);
2312 /* were we interrupted ? */
2316 if (dm_request_based(md))
2317 dm_start_queue(md->queue);
2320 dm_table_presuspend_undo_targets(map);
2321 /* pushback list is already flushed, so skip flush */
2328 * We need to be able to change a mapping table under a mounted
2329 * filesystem. For example we might want to move some data in
2330 * the background. Before the table can be swapped with
2331 * dm_bind_table, dm_suspend must be called to flush any in
2332 * flight bios and ensure that any further io gets deferred.
2335 * Suspend mechanism in request-based dm.
2337 * 1. Flush all I/Os by lock_fs() if needed.
2338 * 2. Stop dispatching any I/O by stopping the request_queue.
2339 * 3. Wait for all in-flight I/Os to be completed or requeued.
2341 * To abort suspend, start the request_queue.
2343 int dm_suspend(struct mapped_device *md, unsigned suspend_flags)
2345 struct dm_table *map = NULL;
2349 mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING);
2351 if (dm_suspended_md(md)) {
2356 if (dm_suspended_internally_md(md)) {
2357 /* already internally suspended, wait for internal resume */
2358 mutex_unlock(&md->suspend_lock);
2359 r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE);
2365 map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2367 r = __dm_suspend(md, map, suspend_flags, TASK_INTERRUPTIBLE, DMF_SUSPENDED);
2371 dm_table_postsuspend_targets(map);
2374 mutex_unlock(&md->suspend_lock);
2378 static int __dm_resume(struct mapped_device *md, struct dm_table *map)
2381 int r = dm_table_resume_targets(map);
2389 * Flushing deferred I/Os must be done after targets are resumed
2390 * so that mapping of targets can work correctly.
2391 * Request-based dm is queueing the deferred I/Os in its request_queue.
2393 if (dm_request_based(md))
2394 dm_start_queue(md->queue);
2401 int dm_resume(struct mapped_device *md)
2404 struct dm_table *map = NULL;
2408 mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING);
2410 if (!dm_suspended_md(md))
2413 if (dm_suspended_internally_md(md)) {
2414 /* already internally suspended, wait for internal resume */
2415 mutex_unlock(&md->suspend_lock);
2416 r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE);
2422 map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2423 if (!map || !dm_table_get_size(map))
2426 r = __dm_resume(md, map);
2430 clear_bit(DMF_SUSPENDED, &md->flags);
2432 mutex_unlock(&md->suspend_lock);
2438 * Internal suspend/resume works like userspace-driven suspend. It waits
2439 * until all bios finish and prevents issuing new bios to the target drivers.
2440 * It may be used only from the kernel.
2443 static void __dm_internal_suspend(struct mapped_device *md, unsigned suspend_flags)
2445 struct dm_table *map = NULL;
2447 lockdep_assert_held(&md->suspend_lock);
2449 if (md->internal_suspend_count++)
2450 return; /* nested internal suspend */
2452 if (dm_suspended_md(md)) {
2453 set_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
2454 return; /* nest suspend */
2457 map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2460 * Using TASK_UNINTERRUPTIBLE because only NOFLUSH internal suspend is
2461 * supported. Properly supporting a TASK_INTERRUPTIBLE internal suspend
2462 * would require changing .presuspend to return an error -- avoid this
2463 * until there is a need for more elaborate variants of internal suspend.
2465 (void) __dm_suspend(md, map, suspend_flags, TASK_UNINTERRUPTIBLE,
2466 DMF_SUSPENDED_INTERNALLY);
2468 dm_table_postsuspend_targets(map);
2471 static void __dm_internal_resume(struct mapped_device *md)
2473 BUG_ON(!md->internal_suspend_count);
2475 if (--md->internal_suspend_count)
2476 return; /* resume from nested internal suspend */
2478 if (dm_suspended_md(md))
2479 goto done; /* resume from nested suspend */
2482 * NOTE: existing callers don't need to call dm_table_resume_targets
2483 * (which may fail -- so best to avoid it for now by passing NULL map)
2485 (void) __dm_resume(md, NULL);
2488 clear_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
2489 smp_mb__after_atomic();
2490 wake_up_bit(&md->flags, DMF_SUSPENDED_INTERNALLY);
2493 void dm_internal_suspend_noflush(struct mapped_device *md)
2495 mutex_lock(&md->suspend_lock);
2496 __dm_internal_suspend(md, DM_SUSPEND_NOFLUSH_FLAG);
2497 mutex_unlock(&md->suspend_lock);
2499 EXPORT_SYMBOL_GPL(dm_internal_suspend_noflush);
2501 void dm_internal_resume(struct mapped_device *md)
2503 mutex_lock(&md->suspend_lock);
2504 __dm_internal_resume(md);
2505 mutex_unlock(&md->suspend_lock);
2507 EXPORT_SYMBOL_GPL(dm_internal_resume);
2510 * Fast variants of internal suspend/resume hold md->suspend_lock,
2511 * which prevents interaction with userspace-driven suspend.
2514 void dm_internal_suspend_fast(struct mapped_device *md)
2516 mutex_lock(&md->suspend_lock);
2517 if (dm_suspended_md(md) || dm_suspended_internally_md(md))
2520 set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2521 synchronize_srcu(&md->io_barrier);
2522 flush_workqueue(md->wq);
2523 dm_wait_for_completion(md, TASK_UNINTERRUPTIBLE);
2525 EXPORT_SYMBOL_GPL(dm_internal_suspend_fast);
2527 void dm_internal_resume_fast(struct mapped_device *md)
2529 if (dm_suspended_md(md) || dm_suspended_internally_md(md))
2535 mutex_unlock(&md->suspend_lock);
2537 EXPORT_SYMBOL_GPL(dm_internal_resume_fast);
2539 /*-----------------------------------------------------------------
2540 * Event notification.
2541 *---------------------------------------------------------------*/
2542 int dm_kobject_uevent(struct mapped_device *md, enum kobject_action action,
2545 char udev_cookie[DM_COOKIE_LENGTH];
2546 char *envp[] = { udev_cookie, NULL };
2549 return kobject_uevent(&disk_to_dev(md->disk)->kobj, action);
2551 snprintf(udev_cookie, DM_COOKIE_LENGTH, "%s=%u",
2552 DM_COOKIE_ENV_VAR_NAME, cookie);
2553 return kobject_uevent_env(&disk_to_dev(md->disk)->kobj,
2558 uint32_t dm_next_uevent_seq(struct mapped_device *md)
2560 return atomic_add_return(1, &md->uevent_seq);
2563 uint32_t dm_get_event_nr(struct mapped_device *md)
2565 return atomic_read(&md->event_nr);
2568 int dm_wait_event(struct mapped_device *md, int event_nr)
2570 return wait_event_interruptible(md->eventq,
2571 (event_nr != atomic_read(&md->event_nr)));
2574 void dm_uevent_add(struct mapped_device *md, struct list_head *elist)
2576 unsigned long flags;
2578 spin_lock_irqsave(&md->uevent_lock, flags);
2579 list_add(elist, &md->uevent_list);
2580 spin_unlock_irqrestore(&md->uevent_lock, flags);
2584 * The gendisk is only valid as long as you have a reference
2587 struct gendisk *dm_disk(struct mapped_device *md)
2591 EXPORT_SYMBOL_GPL(dm_disk);
2593 struct kobject *dm_kobject(struct mapped_device *md)
2595 return &md->kobj_holder.kobj;
2598 struct mapped_device *dm_get_from_kobject(struct kobject *kobj)
2600 struct mapped_device *md;
2602 md = container_of(kobj, struct mapped_device, kobj_holder.kobj);
2604 if (test_bit(DMF_FREEING, &md->flags) ||
2612 int dm_suspended_md(struct mapped_device *md)
2614 return test_bit(DMF_SUSPENDED, &md->flags);
2617 int dm_suspended_internally_md(struct mapped_device *md)
2619 return test_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
2622 int dm_test_deferred_remove_flag(struct mapped_device *md)
2624 return test_bit(DMF_DEFERRED_REMOVE, &md->flags);
2627 int dm_suspended(struct dm_target *ti)
2629 return dm_suspended_md(dm_table_get_md(ti->table));
2631 EXPORT_SYMBOL_GPL(dm_suspended);
2633 int dm_noflush_suspending(struct dm_target *ti)
2635 return __noflush_suspending(dm_table_get_md(ti->table));
2637 EXPORT_SYMBOL_GPL(dm_noflush_suspending);
2639 struct dm_md_mempools *dm_alloc_md_mempools(struct mapped_device *md, enum dm_queue_mode type,
2640 unsigned integrity, unsigned per_io_data_size)
2642 struct dm_md_mempools *pools = kzalloc_node(sizeof(*pools), GFP_KERNEL, md->numa_node_id);
2643 unsigned int pool_size = 0;
2644 unsigned int front_pad;
2650 case DM_TYPE_BIO_BASED:
2651 case DM_TYPE_DAX_BIO_BASED:
2652 pool_size = dm_get_reserved_bio_based_ios();
2653 front_pad = roundup(per_io_data_size, __alignof__(struct dm_target_io)) + offsetof(struct dm_target_io, clone);
2655 pools->io_pool = mempool_create_slab_pool(pool_size, _io_cache);
2656 if (!pools->io_pool)
2659 case DM_TYPE_REQUEST_BASED:
2660 case DM_TYPE_MQ_REQUEST_BASED:
2661 pool_size = dm_get_reserved_rq_based_ios();
2662 front_pad = offsetof(struct dm_rq_clone_bio_info, clone);
2663 /* per_io_data_size is used for blk-mq pdu at queue allocation */
2669 pools->bs = bioset_create(pool_size, front_pad, BIOSET_NEED_RESCUER);
2673 if (integrity && bioset_integrity_create(pools->bs, pool_size))
2679 dm_free_md_mempools(pools);
2684 void dm_free_md_mempools(struct dm_md_mempools *pools)
2689 mempool_destroy(pools->io_pool);
2692 bioset_free(pools->bs);
2704 static int dm_call_pr(struct block_device *bdev, iterate_devices_callout_fn fn,
2707 struct mapped_device *md = bdev->bd_disk->private_data;
2708 struct dm_table *table;
2709 struct dm_target *ti;
2710 int ret = -ENOTTY, srcu_idx;
2712 table = dm_get_live_table(md, &srcu_idx);
2713 if (!table || !dm_table_get_size(table))
2716 /* We only support devices that have a single target */
2717 if (dm_table_get_num_targets(table) != 1)
2719 ti = dm_table_get_target(table, 0);
2722 if (!ti->type->iterate_devices)
2725 ret = ti->type->iterate_devices(ti, fn, data);
2727 dm_put_live_table(md, srcu_idx);
2732 * For register / unregister we need to manually call out to every path.
2734 static int __dm_pr_register(struct dm_target *ti, struct dm_dev *dev,
2735 sector_t start, sector_t len, void *data)
2737 struct dm_pr *pr = data;
2738 const struct pr_ops *ops = dev->bdev->bd_disk->fops->pr_ops;
2740 if (!ops || !ops->pr_register)
2742 return ops->pr_register(dev->bdev, pr->old_key, pr->new_key, pr->flags);
2745 static int dm_pr_register(struct block_device *bdev, u64 old_key, u64 new_key,
2756 ret = dm_call_pr(bdev, __dm_pr_register, &pr);
2757 if (ret && new_key) {
2758 /* unregister all paths if we failed to register any path */
2759 pr.old_key = new_key;
2762 pr.fail_early = false;
2763 dm_call_pr(bdev, __dm_pr_register, &pr);
2769 static int dm_pr_reserve(struct block_device *bdev, u64 key, enum pr_type type,
2772 struct mapped_device *md = bdev->bd_disk->private_data;
2773 const struct pr_ops *ops;
2777 r = dm_grab_bdev_for_ioctl(md, &bdev, &mode);
2781 ops = bdev->bd_disk->fops->pr_ops;
2782 if (ops && ops->pr_reserve)
2783 r = ops->pr_reserve(bdev, key, type, flags);
2791 static int dm_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
2793 struct mapped_device *md = bdev->bd_disk->private_data;
2794 const struct pr_ops *ops;
2798 r = dm_grab_bdev_for_ioctl(md, &bdev, &mode);
2802 ops = bdev->bd_disk->fops->pr_ops;
2803 if (ops && ops->pr_release)
2804 r = ops->pr_release(bdev, key, type);
2812 static int dm_pr_preempt(struct block_device *bdev, u64 old_key, u64 new_key,
2813 enum pr_type type, bool abort)
2815 struct mapped_device *md = bdev->bd_disk->private_data;
2816 const struct pr_ops *ops;
2820 r = dm_grab_bdev_for_ioctl(md, &bdev, &mode);
2824 ops = bdev->bd_disk->fops->pr_ops;
2825 if (ops && ops->pr_preempt)
2826 r = ops->pr_preempt(bdev, old_key, new_key, type, abort);
2834 static int dm_pr_clear(struct block_device *bdev, u64 key)
2836 struct mapped_device *md = bdev->bd_disk->private_data;
2837 const struct pr_ops *ops;
2841 r = dm_grab_bdev_for_ioctl(md, &bdev, &mode);
2845 ops = bdev->bd_disk->fops->pr_ops;
2846 if (ops && ops->pr_clear)
2847 r = ops->pr_clear(bdev, key);
2855 static const struct pr_ops dm_pr_ops = {
2856 .pr_register = dm_pr_register,
2857 .pr_reserve = dm_pr_reserve,
2858 .pr_release = dm_pr_release,
2859 .pr_preempt = dm_pr_preempt,
2860 .pr_clear = dm_pr_clear,
2863 static const struct block_device_operations dm_blk_dops = {
2864 .open = dm_blk_open,
2865 .release = dm_blk_close,
2866 .ioctl = dm_blk_ioctl,
2867 .getgeo = dm_blk_getgeo,
2868 .pr_ops = &dm_pr_ops,
2869 .owner = THIS_MODULE
2872 static const struct dax_operations dm_dax_ops = {
2873 .direct_access = dm_dax_direct_access,
2879 module_init(dm_init);
2880 module_exit(dm_exit);
2882 module_param(major, uint, 0);
2883 MODULE_PARM_DESC(major, "The major number of the device mapper");
2885 module_param(reserved_bio_based_ios, uint, S_IRUGO | S_IWUSR);
2886 MODULE_PARM_DESC(reserved_bio_based_ios, "Reserved IOs in bio-based mempools");
2888 module_param(dm_numa_node, int, S_IRUGO | S_IWUSR);
2889 MODULE_PARM_DESC(dm_numa_node, "NUMA node for DM device memory allocations");
2891 MODULE_DESCRIPTION(DM_NAME " driver");
2892 MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
2893 MODULE_LICENSE("GPL");