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/uio.h>
23 #include <linux/hdreg.h>
24 #include <linux/delay.h>
25 #include <linux/wait.h>
27 #include <linux/refcount.h>
29 #define DM_MSG_PREFIX "core"
32 * Cookies are numeric values sent with CHANGE and REMOVE
33 * uevents while resuming, removing or renaming the device.
35 #define DM_COOKIE_ENV_VAR_NAME "DM_COOKIE"
36 #define DM_COOKIE_LENGTH 24
38 static const char *_name
= DM_NAME
;
40 static unsigned int major
= 0;
41 static unsigned int _major
= 0;
43 static DEFINE_IDR(_minor_idr
);
45 static DEFINE_SPINLOCK(_minor_lock
);
47 static void do_deferred_remove(struct work_struct
*w
);
49 static DECLARE_WORK(deferred_remove_work
, do_deferred_remove
);
51 static struct workqueue_struct
*deferred_remove_workqueue
;
53 atomic_t dm_global_event_nr
= ATOMIC_INIT(0);
54 DECLARE_WAIT_QUEUE_HEAD(dm_global_eventq
);
56 void dm_issue_global_event(void)
58 atomic_inc(&dm_global_event_nr
);
59 wake_up(&dm_global_eventq
);
63 * One of these is allocated per bio.
66 struct mapped_device
*md
;
70 unsigned long start_time
;
71 spinlock_t endio_lock
;
72 struct dm_stats_aux stats_aux
;
75 #define MINOR_ALLOCED ((void *)-1)
78 * Bits for the md->flags field.
80 #define DMF_BLOCK_IO_FOR_SUSPEND 0
81 #define DMF_SUSPENDED 1
84 #define DMF_DELETING 4
85 #define DMF_NOFLUSH_SUSPENDING 5
86 #define DMF_DEFERRED_REMOVE 6
87 #define DMF_SUSPENDED_INTERNALLY 7
89 #define DM_NUMA_NODE NUMA_NO_NODE
90 static int dm_numa_node
= DM_NUMA_NODE
;
93 * For mempools pre-allocation at the table loading time.
95 struct dm_md_mempools
{
100 struct table_device
{
101 struct list_head list
;
103 struct dm_dev dm_dev
;
106 static struct kmem_cache
*_io_cache
;
107 static struct kmem_cache
*_rq_tio_cache
;
108 static struct kmem_cache
*_rq_cache
;
111 * Bio-based DM's mempools' reserved IOs set by the user.
113 #define RESERVED_BIO_BASED_IOS 16
114 static unsigned reserved_bio_based_ios
= RESERVED_BIO_BASED_IOS
;
116 static int __dm_get_module_param_int(int *module_param
, int min
, int max
)
118 int param
= READ_ONCE(*module_param
);
119 int modified_param
= 0;
120 bool modified
= true;
123 modified_param
= min
;
124 else if (param
> max
)
125 modified_param
= max
;
130 (void)cmpxchg(module_param
, param
, modified_param
);
131 param
= modified_param
;
137 unsigned __dm_get_module_param(unsigned *module_param
,
138 unsigned def
, unsigned max
)
140 unsigned param
= READ_ONCE(*module_param
);
141 unsigned modified_param
= 0;
144 modified_param
= def
;
145 else if (param
> max
)
146 modified_param
= max
;
148 if (modified_param
) {
149 (void)cmpxchg(module_param
, param
, modified_param
);
150 param
= modified_param
;
156 unsigned dm_get_reserved_bio_based_ios(void)
158 return __dm_get_module_param(&reserved_bio_based_ios
,
159 RESERVED_BIO_BASED_IOS
, DM_RESERVED_MAX_IOS
);
161 EXPORT_SYMBOL_GPL(dm_get_reserved_bio_based_ios
);
163 static unsigned dm_get_numa_node(void)
165 return __dm_get_module_param_int(&dm_numa_node
,
166 DM_NUMA_NODE
, num_online_nodes() - 1);
169 static int __init
local_init(void)
173 /* allocate a slab for the dm_ios */
174 _io_cache
= KMEM_CACHE(dm_io
, 0);
178 _rq_tio_cache
= KMEM_CACHE(dm_rq_target_io
, 0);
180 goto out_free_io_cache
;
182 _rq_cache
= kmem_cache_create("dm_old_clone_request", sizeof(struct request
),
183 __alignof__(struct request
), 0, NULL
);
185 goto out_free_rq_tio_cache
;
187 r
= dm_uevent_init();
189 goto out_free_rq_cache
;
191 deferred_remove_workqueue
= alloc_workqueue("kdmremove", WQ_UNBOUND
, 1);
192 if (!deferred_remove_workqueue
) {
194 goto out_uevent_exit
;
198 r
= register_blkdev(_major
, _name
);
200 goto out_free_workqueue
;
208 destroy_workqueue(deferred_remove_workqueue
);
212 kmem_cache_destroy(_rq_cache
);
213 out_free_rq_tio_cache
:
214 kmem_cache_destroy(_rq_tio_cache
);
216 kmem_cache_destroy(_io_cache
);
221 static void local_exit(void)
223 flush_scheduled_work();
224 destroy_workqueue(deferred_remove_workqueue
);
226 kmem_cache_destroy(_rq_cache
);
227 kmem_cache_destroy(_rq_tio_cache
);
228 kmem_cache_destroy(_io_cache
);
229 unregister_blkdev(_major
, _name
);
234 DMINFO("cleaned up");
237 static int (*_inits
[])(void) __initdata
= {
248 static void (*_exits
[])(void) = {
259 static int __init
dm_init(void)
261 const int count
= ARRAY_SIZE(_inits
);
265 for (i
= 0; i
< count
; i
++) {
280 static void __exit
dm_exit(void)
282 int i
= ARRAY_SIZE(_exits
);
288 * Should be empty by this point.
290 idr_destroy(&_minor_idr
);
294 * Block device functions
296 int dm_deleting_md(struct mapped_device
*md
)
298 return test_bit(DMF_DELETING
, &md
->flags
);
301 static int dm_blk_open(struct block_device
*bdev
, fmode_t mode
)
303 struct mapped_device
*md
;
305 spin_lock(&_minor_lock
);
307 md
= bdev
->bd_disk
->private_data
;
311 if (test_bit(DMF_FREEING
, &md
->flags
) ||
312 dm_deleting_md(md
)) {
318 atomic_inc(&md
->open_count
);
320 spin_unlock(&_minor_lock
);
322 return md
? 0 : -ENXIO
;
325 static void dm_blk_close(struct gendisk
*disk
, fmode_t mode
)
327 struct mapped_device
*md
;
329 spin_lock(&_minor_lock
);
331 md
= disk
->private_data
;
335 if (atomic_dec_and_test(&md
->open_count
) &&
336 (test_bit(DMF_DEFERRED_REMOVE
, &md
->flags
)))
337 queue_work(deferred_remove_workqueue
, &deferred_remove_work
);
341 spin_unlock(&_minor_lock
);
344 int dm_open_count(struct mapped_device
*md
)
346 return atomic_read(&md
->open_count
);
350 * Guarantees nothing is using the device before it's deleted.
352 int dm_lock_for_deletion(struct mapped_device
*md
, bool mark_deferred
, bool only_deferred
)
356 spin_lock(&_minor_lock
);
358 if (dm_open_count(md
)) {
361 set_bit(DMF_DEFERRED_REMOVE
, &md
->flags
);
362 } else if (only_deferred
&& !test_bit(DMF_DEFERRED_REMOVE
, &md
->flags
))
365 set_bit(DMF_DELETING
, &md
->flags
);
367 spin_unlock(&_minor_lock
);
372 int dm_cancel_deferred_remove(struct mapped_device
*md
)
376 spin_lock(&_minor_lock
);
378 if (test_bit(DMF_DELETING
, &md
->flags
))
381 clear_bit(DMF_DEFERRED_REMOVE
, &md
->flags
);
383 spin_unlock(&_minor_lock
);
388 static void do_deferred_remove(struct work_struct
*w
)
390 dm_deferred_remove();
393 sector_t
dm_get_size(struct mapped_device
*md
)
395 return get_capacity(md
->disk
);
398 struct request_queue
*dm_get_md_queue(struct mapped_device
*md
)
403 struct dm_stats
*dm_get_stats(struct mapped_device
*md
)
408 static int dm_blk_getgeo(struct block_device
*bdev
, struct hd_geometry
*geo
)
410 struct mapped_device
*md
= bdev
->bd_disk
->private_data
;
412 return dm_get_geometry(md
, geo
);
415 static int dm_grab_bdev_for_ioctl(struct mapped_device
*md
,
416 struct block_device
**bdev
,
419 struct dm_target
*tgt
;
420 struct dm_table
*map
;
425 map
= dm_get_live_table(md
, &srcu_idx
);
426 if (!map
|| !dm_table_get_size(map
))
429 /* We only support devices that have a single target */
430 if (dm_table_get_num_targets(map
) != 1)
433 tgt
= dm_table_get_target(map
, 0);
434 if (!tgt
->type
->prepare_ioctl
)
437 if (dm_suspended_md(md
)) {
442 r
= tgt
->type
->prepare_ioctl(tgt
, bdev
, mode
);
447 dm_put_live_table(md
, srcu_idx
);
451 dm_put_live_table(md
, srcu_idx
);
452 if (r
== -ENOTCONN
&& !fatal_signal_pending(current
)) {
459 static int dm_blk_ioctl(struct block_device
*bdev
, fmode_t mode
,
460 unsigned int cmd
, unsigned long arg
)
462 struct mapped_device
*md
= bdev
->bd_disk
->private_data
;
465 r
= dm_grab_bdev_for_ioctl(md
, &bdev
, &mode
);
471 * Target determined this ioctl is being issued against a
472 * subset of the parent bdev; require extra privileges.
474 if (!capable(CAP_SYS_RAWIO
)) {
476 "%s: sending ioctl %x to DM device without required privilege.",
483 r
= __blkdev_driver_ioctl(bdev
, mode
, cmd
, arg
);
489 static struct dm_io
*alloc_io(struct mapped_device
*md
)
491 return mempool_alloc(md
->io_pool
, GFP_NOIO
);
494 static void free_io(struct mapped_device
*md
, struct dm_io
*io
)
496 mempool_free(io
, md
->io_pool
);
499 static void free_tio(struct dm_target_io
*tio
)
501 bio_put(&tio
->clone
);
504 int md_in_flight(struct mapped_device
*md
)
506 return atomic_read(&md
->pending
[READ
]) +
507 atomic_read(&md
->pending
[WRITE
]);
510 static void start_io_acct(struct dm_io
*io
)
512 struct mapped_device
*md
= io
->md
;
513 struct bio
*bio
= io
->bio
;
515 int rw
= bio_data_dir(bio
);
517 io
->start_time
= jiffies
;
519 cpu
= part_stat_lock();
520 part_round_stats(md
->queue
, cpu
, &dm_disk(md
)->part0
);
522 atomic_set(&dm_disk(md
)->part0
.in_flight
[rw
],
523 atomic_inc_return(&md
->pending
[rw
]));
525 if (unlikely(dm_stats_used(&md
->stats
)))
526 dm_stats_account_io(&md
->stats
, bio_data_dir(bio
),
527 bio
->bi_iter
.bi_sector
, bio_sectors(bio
),
528 false, 0, &io
->stats_aux
);
531 static void end_io_acct(struct dm_io
*io
)
533 struct mapped_device
*md
= io
->md
;
534 struct bio
*bio
= io
->bio
;
535 unsigned long duration
= jiffies
- io
->start_time
;
537 int rw
= bio_data_dir(bio
);
539 generic_end_io_acct(md
->queue
, rw
, &dm_disk(md
)->part0
, io
->start_time
);
541 if (unlikely(dm_stats_used(&md
->stats
)))
542 dm_stats_account_io(&md
->stats
, bio_data_dir(bio
),
543 bio
->bi_iter
.bi_sector
, bio_sectors(bio
),
544 true, duration
, &io
->stats_aux
);
547 * After this is decremented the bio must not be touched if it is
550 pending
= atomic_dec_return(&md
->pending
[rw
]);
551 atomic_set(&dm_disk(md
)->part0
.in_flight
[rw
], pending
);
552 pending
+= atomic_read(&md
->pending
[rw
^0x1]);
554 /* nudge anyone waiting on suspend queue */
560 * Add the bio to the list of deferred io.
562 static void queue_io(struct mapped_device
*md
, struct bio
*bio
)
566 spin_lock_irqsave(&md
->deferred_lock
, flags
);
567 bio_list_add(&md
->deferred
, bio
);
568 spin_unlock_irqrestore(&md
->deferred_lock
, flags
);
569 queue_work(md
->wq
, &md
->work
);
573 * Everyone (including functions in this file), should use this
574 * function to access the md->map field, and make sure they call
575 * dm_put_live_table() when finished.
577 struct dm_table
*dm_get_live_table(struct mapped_device
*md
, int *srcu_idx
) __acquires(md
->io_barrier
)
579 *srcu_idx
= srcu_read_lock(&md
->io_barrier
);
581 return srcu_dereference(md
->map
, &md
->io_barrier
);
584 void dm_put_live_table(struct mapped_device
*md
, int srcu_idx
) __releases(md
->io_barrier
)
586 srcu_read_unlock(&md
->io_barrier
, srcu_idx
);
589 void dm_sync_table(struct mapped_device
*md
)
591 synchronize_srcu(&md
->io_barrier
);
592 synchronize_rcu_expedited();
596 * A fast alternative to dm_get_live_table/dm_put_live_table.
597 * The caller must not block between these two functions.
599 static struct dm_table
*dm_get_live_table_fast(struct mapped_device
*md
) __acquires(RCU
)
602 return rcu_dereference(md
->map
);
605 static void dm_put_live_table_fast(struct mapped_device
*md
) __releases(RCU
)
611 * Open a table device so we can use it as a map destination.
613 static int open_table_device(struct table_device
*td
, dev_t dev
,
614 struct mapped_device
*md
)
616 static char *_claim_ptr
= "I belong to device-mapper";
617 struct block_device
*bdev
;
621 BUG_ON(td
->dm_dev
.bdev
);
623 bdev
= blkdev_get_by_dev(dev
, td
->dm_dev
.mode
| FMODE_EXCL
, _claim_ptr
);
625 return PTR_ERR(bdev
);
627 r
= bd_link_disk_holder(bdev
, dm_disk(md
));
629 blkdev_put(bdev
, td
->dm_dev
.mode
| FMODE_EXCL
);
633 td
->dm_dev
.bdev
= bdev
;
634 td
->dm_dev
.dax_dev
= dax_get_by_host(bdev
->bd_disk
->disk_name
);
639 * Close a table device that we've been using.
641 static void close_table_device(struct table_device
*td
, struct mapped_device
*md
)
643 if (!td
->dm_dev
.bdev
)
646 bd_unlink_disk_holder(td
->dm_dev
.bdev
, dm_disk(md
));
647 blkdev_put(td
->dm_dev
.bdev
, td
->dm_dev
.mode
| FMODE_EXCL
);
648 put_dax(td
->dm_dev
.dax_dev
);
649 td
->dm_dev
.bdev
= NULL
;
650 td
->dm_dev
.dax_dev
= NULL
;
653 static struct table_device
*find_table_device(struct list_head
*l
, dev_t dev
,
655 struct table_device
*td
;
657 list_for_each_entry(td
, l
, list
)
658 if (td
->dm_dev
.bdev
->bd_dev
== dev
&& td
->dm_dev
.mode
== mode
)
664 int dm_get_table_device(struct mapped_device
*md
, dev_t dev
, fmode_t mode
,
665 struct dm_dev
**result
) {
667 struct table_device
*td
;
669 mutex_lock(&md
->table_devices_lock
);
670 td
= find_table_device(&md
->table_devices
, dev
, mode
);
672 td
= kmalloc_node(sizeof(*td
), GFP_KERNEL
, md
->numa_node_id
);
674 mutex_unlock(&md
->table_devices_lock
);
678 td
->dm_dev
.mode
= mode
;
679 td
->dm_dev
.bdev
= NULL
;
681 if ((r
= open_table_device(td
, dev
, md
))) {
682 mutex_unlock(&md
->table_devices_lock
);
687 format_dev_t(td
->dm_dev
.name
, dev
);
689 refcount_set(&td
->count
, 1);
690 list_add(&td
->list
, &md
->table_devices
);
692 refcount_inc(&td
->count
);
694 mutex_unlock(&md
->table_devices_lock
);
696 *result
= &td
->dm_dev
;
699 EXPORT_SYMBOL_GPL(dm_get_table_device
);
701 void dm_put_table_device(struct mapped_device
*md
, struct dm_dev
*d
)
703 struct table_device
*td
= container_of(d
, struct table_device
, dm_dev
);
705 mutex_lock(&md
->table_devices_lock
);
706 if (refcount_dec_and_test(&td
->count
)) {
707 close_table_device(td
, md
);
711 mutex_unlock(&md
->table_devices_lock
);
713 EXPORT_SYMBOL(dm_put_table_device
);
715 static void free_table_devices(struct list_head
*devices
)
717 struct list_head
*tmp
, *next
;
719 list_for_each_safe(tmp
, next
, devices
) {
720 struct table_device
*td
= list_entry(tmp
, struct table_device
, list
);
722 DMWARN("dm_destroy: %s still exists with %d references",
723 td
->dm_dev
.name
, refcount_read(&td
->count
));
729 * Get the geometry associated with a dm device
731 int dm_get_geometry(struct mapped_device
*md
, struct hd_geometry
*geo
)
739 * Set the geometry of a device.
741 int dm_set_geometry(struct mapped_device
*md
, struct hd_geometry
*geo
)
743 sector_t sz
= (sector_t
)geo
->cylinders
* geo
->heads
* geo
->sectors
;
745 if (geo
->start
> sz
) {
746 DMWARN("Start sector is beyond the geometry limits.");
755 /*-----------------------------------------------------------------
757 * A more elegant soln is in the works that uses the queue
758 * merge fn, unfortunately there are a couple of changes to
759 * the block layer that I want to make for this. So in the
760 * interests of getting something for people to use I give
761 * you this clearly demarcated crap.
762 *---------------------------------------------------------------*/
764 static int __noflush_suspending(struct mapped_device
*md
)
766 return test_bit(DMF_NOFLUSH_SUSPENDING
, &md
->flags
);
770 * Decrements the number of outstanding ios that a bio has been
771 * cloned into, completing the original io if necc.
773 static void dec_pending(struct dm_io
*io
, blk_status_t error
)
776 blk_status_t io_error
;
778 struct mapped_device
*md
= io
->md
;
780 /* Push-back supersedes any I/O errors */
781 if (unlikely(error
)) {
782 spin_lock_irqsave(&io
->endio_lock
, flags
);
783 if (!(io
->status
== BLK_STS_DM_REQUEUE
&&
784 __noflush_suspending(md
)))
786 spin_unlock_irqrestore(&io
->endio_lock
, flags
);
789 if (atomic_dec_and_test(&io
->io_count
)) {
790 if (io
->status
== BLK_STS_DM_REQUEUE
) {
792 * Target requested pushing back the I/O.
794 spin_lock_irqsave(&md
->deferred_lock
, flags
);
795 if (__noflush_suspending(md
))
796 bio_list_add_head(&md
->deferred
, io
->bio
);
798 /* noflush suspend was interrupted. */
799 io
->status
= BLK_STS_IOERR
;
800 spin_unlock_irqrestore(&md
->deferred_lock
, flags
);
803 io_error
= io
->status
;
808 if (io_error
== BLK_STS_DM_REQUEUE
)
811 if ((bio
->bi_opf
& REQ_PREFLUSH
) && bio
->bi_iter
.bi_size
) {
813 * Preflush done for flush with data, reissue
814 * without REQ_PREFLUSH.
816 bio
->bi_opf
&= ~REQ_PREFLUSH
;
819 /* done with normal IO or empty flush */
820 bio
->bi_status
= io_error
;
826 void disable_write_same(struct mapped_device
*md
)
828 struct queue_limits
*limits
= dm_get_queue_limits(md
);
830 /* device doesn't really support WRITE SAME, disable it */
831 limits
->max_write_same_sectors
= 0;
834 void disable_write_zeroes(struct mapped_device
*md
)
836 struct queue_limits
*limits
= dm_get_queue_limits(md
);
838 /* device doesn't really support WRITE ZEROES, disable it */
839 limits
->max_write_zeroes_sectors
= 0;
842 static void clone_endio(struct bio
*bio
)
844 blk_status_t error
= bio
->bi_status
;
845 struct dm_target_io
*tio
= container_of(bio
, struct dm_target_io
, clone
);
846 struct dm_io
*io
= tio
->io
;
847 struct mapped_device
*md
= tio
->io
->md
;
848 dm_endio_fn endio
= tio
->ti
->type
->end_io
;
850 if (unlikely(error
== BLK_STS_TARGET
)) {
851 if (bio_op(bio
) == REQ_OP_WRITE_SAME
&&
852 !bio
->bi_disk
->queue
->limits
.max_write_same_sectors
)
853 disable_write_same(md
);
854 if (bio_op(bio
) == REQ_OP_WRITE_ZEROES
&&
855 !bio
->bi_disk
->queue
->limits
.max_write_zeroes_sectors
)
856 disable_write_zeroes(md
);
860 int r
= endio(tio
->ti
, bio
, &error
);
862 case DM_ENDIO_REQUEUE
:
863 error
= BLK_STS_DM_REQUEUE
;
867 case DM_ENDIO_INCOMPLETE
:
868 /* The target will handle the io */
871 DMWARN("unimplemented target endio return value: %d", r
);
877 dec_pending(io
, error
);
881 * Return maximum size of I/O possible at the supplied sector up to the current
884 static sector_t
max_io_len_target_boundary(sector_t sector
, struct dm_target
*ti
)
886 sector_t target_offset
= dm_target_offset(ti
, sector
);
888 return ti
->len
- target_offset
;
891 static sector_t
max_io_len(sector_t sector
, struct dm_target
*ti
)
893 sector_t len
= max_io_len_target_boundary(sector
, ti
);
894 sector_t offset
, max_len
;
897 * Does the target need to split even further?
899 if (ti
->max_io_len
) {
900 offset
= dm_target_offset(ti
, sector
);
901 if (unlikely(ti
->max_io_len
& (ti
->max_io_len
- 1)))
902 max_len
= sector_div(offset
, ti
->max_io_len
);
904 max_len
= offset
& (ti
->max_io_len
- 1);
905 max_len
= ti
->max_io_len
- max_len
;
914 int dm_set_target_max_io_len(struct dm_target
*ti
, sector_t len
)
916 if (len
> UINT_MAX
) {
917 DMERR("Specified maximum size of target IO (%llu) exceeds limit (%u)",
918 (unsigned long long)len
, UINT_MAX
);
919 ti
->error
= "Maximum size of target IO is too large";
924 * BIO based queue uses its own splitting. When multipage bvecs
925 * is switched on, size of the incoming bio may be too big to
926 * be handled in some targets, such as crypt.
928 * When these targets are ready for the big bio, we can remove
931 ti
->max_io_len
= min_t(uint32_t, len
, BIO_MAX_PAGES
* PAGE_SIZE
);
935 EXPORT_SYMBOL_GPL(dm_set_target_max_io_len
);
937 static struct dm_target
*dm_dax_get_live_target(struct mapped_device
*md
,
938 sector_t sector
, int *srcu_idx
)
940 struct dm_table
*map
;
941 struct dm_target
*ti
;
943 map
= dm_get_live_table(md
, srcu_idx
);
947 ti
= dm_table_find_target(map
, sector
);
948 if (!dm_target_is_valid(ti
))
954 static long dm_dax_direct_access(struct dax_device
*dax_dev
, pgoff_t pgoff
,
955 long nr_pages
, void **kaddr
, pfn_t
*pfn
)
957 struct mapped_device
*md
= dax_get_private(dax_dev
);
958 sector_t sector
= pgoff
* PAGE_SECTORS
;
959 struct dm_target
*ti
;
960 long len
, ret
= -EIO
;
963 ti
= dm_dax_get_live_target(md
, sector
, &srcu_idx
);
967 if (!ti
->type
->direct_access
)
969 len
= max_io_len(sector
, ti
) / PAGE_SECTORS
;
972 nr_pages
= min(len
, nr_pages
);
973 if (ti
->type
->direct_access
)
974 ret
= ti
->type
->direct_access(ti
, pgoff
, nr_pages
, kaddr
, pfn
);
977 dm_put_live_table(md
, srcu_idx
);
982 static size_t dm_dax_copy_from_iter(struct dax_device
*dax_dev
, pgoff_t pgoff
,
983 void *addr
, size_t bytes
, struct iov_iter
*i
)
985 struct mapped_device
*md
= dax_get_private(dax_dev
);
986 sector_t sector
= pgoff
* PAGE_SECTORS
;
987 struct dm_target
*ti
;
991 ti
= dm_dax_get_live_target(md
, sector
, &srcu_idx
);
995 if (!ti
->type
->dax_copy_from_iter
) {
996 ret
= copy_from_iter(addr
, bytes
, i
);
999 ret
= ti
->type
->dax_copy_from_iter(ti
, pgoff
, addr
, bytes
, i
);
1001 dm_put_live_table(md
, srcu_idx
);
1007 * A target may call dm_accept_partial_bio only from the map routine. It is
1008 * allowed for all bio types except REQ_PREFLUSH.
1010 * dm_accept_partial_bio informs the dm that the target only wants to process
1011 * additional n_sectors sectors of the bio and the rest of the data should be
1012 * sent in a next bio.
1014 * A diagram that explains the arithmetics:
1015 * +--------------------+---------------+-------+
1017 * +--------------------+---------------+-------+
1019 * <-------------- *tio->len_ptr --------------->
1020 * <------- bi_size ------->
1023 * Region 1 was already iterated over with bio_advance or similar function.
1024 * (it may be empty if the target doesn't use bio_advance)
1025 * Region 2 is the remaining bio size that the target wants to process.
1026 * (it may be empty if region 1 is non-empty, although there is no reason
1028 * The target requires that region 3 is to be sent in the next bio.
1030 * If the target wants to receive multiple copies of the bio (via num_*bios, etc),
1031 * the partially processed part (the sum of regions 1+2) must be the same for all
1032 * copies of the bio.
1034 void dm_accept_partial_bio(struct bio
*bio
, unsigned n_sectors
)
1036 struct dm_target_io
*tio
= container_of(bio
, struct dm_target_io
, clone
);
1037 unsigned bi_size
= bio
->bi_iter
.bi_size
>> SECTOR_SHIFT
;
1038 BUG_ON(bio
->bi_opf
& REQ_PREFLUSH
);
1039 BUG_ON(bi_size
> *tio
->len_ptr
);
1040 BUG_ON(n_sectors
> bi_size
);
1041 *tio
->len_ptr
-= bi_size
- n_sectors
;
1042 bio
->bi_iter
.bi_size
= n_sectors
<< SECTOR_SHIFT
;
1044 EXPORT_SYMBOL_GPL(dm_accept_partial_bio
);
1047 * The zone descriptors obtained with a zone report indicate
1048 * zone positions within the target device. The zone descriptors
1049 * must be remapped to match their position within the dm device.
1050 * A target may call dm_remap_zone_report after completion of a
1051 * REQ_OP_ZONE_REPORT bio to remap the zone descriptors obtained
1052 * from the target device mapping to the dm device.
1054 void dm_remap_zone_report(struct dm_target
*ti
, struct bio
*bio
, sector_t start
)
1056 #ifdef CONFIG_BLK_DEV_ZONED
1057 struct dm_target_io
*tio
= container_of(bio
, struct dm_target_io
, clone
);
1058 struct bio
*report_bio
= tio
->io
->bio
;
1059 struct blk_zone_report_hdr
*hdr
= NULL
;
1060 struct blk_zone
*zone
;
1061 unsigned int nr_rep
= 0;
1063 struct bio_vec bvec
;
1064 struct bvec_iter iter
;
1071 * Remap the start sector of the reported zones. For sequential zones,
1072 * also remap the write pointer position.
1074 bio_for_each_segment(bvec
, report_bio
, iter
) {
1075 addr
= kmap_atomic(bvec
.bv_page
);
1077 /* Remember the report header in the first page */
1080 ofst
= sizeof(struct blk_zone_report_hdr
);
1084 /* Set zones start sector */
1085 while (hdr
->nr_zones
&& ofst
< bvec
.bv_len
) {
1087 if (zone
->start
>= start
+ ti
->len
) {
1091 zone
->start
= zone
->start
+ ti
->begin
- start
;
1092 if (zone
->type
!= BLK_ZONE_TYPE_CONVENTIONAL
) {
1093 if (zone
->cond
== BLK_ZONE_COND_FULL
)
1094 zone
->wp
= zone
->start
+ zone
->len
;
1095 else if (zone
->cond
== BLK_ZONE_COND_EMPTY
)
1096 zone
->wp
= zone
->start
;
1098 zone
->wp
= zone
->wp
+ ti
->begin
- start
;
1100 ofst
+= sizeof(struct blk_zone
);
1106 kunmap_atomic(addr
);
1113 hdr
->nr_zones
= nr_rep
;
1117 bio_advance(report_bio
, report_bio
->bi_iter
.bi_size
);
1119 #else /* !CONFIG_BLK_DEV_ZONED */
1120 bio
->bi_status
= BLK_STS_NOTSUPP
;
1123 EXPORT_SYMBOL_GPL(dm_remap_zone_report
);
1126 * Flush current->bio_list when the target map method blocks.
1127 * This fixes deadlocks in snapshot and possibly in other targets.
1130 struct blk_plug plug
;
1131 struct blk_plug_cb cb
;
1134 static void flush_current_bio_list(struct blk_plug_cb
*cb
, bool from_schedule
)
1136 struct dm_offload
*o
= container_of(cb
, struct dm_offload
, cb
);
1137 struct bio_list list
;
1141 INIT_LIST_HEAD(&o
->cb
.list
);
1143 if (unlikely(!current
->bio_list
))
1146 for (i
= 0; i
< 2; i
++) {
1147 list
= current
->bio_list
[i
];
1148 bio_list_init(¤t
->bio_list
[i
]);
1150 while ((bio
= bio_list_pop(&list
))) {
1151 struct bio_set
*bs
= bio
->bi_pool
;
1152 if (unlikely(!bs
) || bs
== fs_bio_set
||
1153 !bs
->rescue_workqueue
) {
1154 bio_list_add(¤t
->bio_list
[i
], bio
);
1158 spin_lock(&bs
->rescue_lock
);
1159 bio_list_add(&bs
->rescue_list
, bio
);
1160 queue_work(bs
->rescue_workqueue
, &bs
->rescue_work
);
1161 spin_unlock(&bs
->rescue_lock
);
1166 static void dm_offload_start(struct dm_offload
*o
)
1168 blk_start_plug(&o
->plug
);
1169 o
->cb
.callback
= flush_current_bio_list
;
1170 list_add(&o
->cb
.list
, ¤t
->plug
->cb_list
);
1173 static void dm_offload_end(struct dm_offload
*o
)
1175 list_del(&o
->cb
.list
);
1176 blk_finish_plug(&o
->plug
);
1179 static void __map_bio(struct dm_target_io
*tio
)
1183 struct dm_offload o
;
1184 struct bio
*clone
= &tio
->clone
;
1185 struct dm_target
*ti
= tio
->ti
;
1187 clone
->bi_end_io
= clone_endio
;
1190 * Map the clone. If r == 0 we don't need to do
1191 * anything, the target has assumed ownership of
1194 atomic_inc(&tio
->io
->io_count
);
1195 sector
= clone
->bi_iter
.bi_sector
;
1197 dm_offload_start(&o
);
1198 r
= ti
->type
->map(ti
, clone
);
1202 case DM_MAPIO_SUBMITTED
:
1204 case DM_MAPIO_REMAPPED
:
1205 /* the bio has been remapped so dispatch it */
1206 trace_block_bio_remap(clone
->bi_disk
->queue
, clone
,
1207 bio_dev(tio
->io
->bio
), sector
);
1208 generic_make_request(clone
);
1211 dec_pending(tio
->io
, BLK_STS_IOERR
);
1214 case DM_MAPIO_REQUEUE
:
1215 dec_pending(tio
->io
, BLK_STS_DM_REQUEUE
);
1219 DMWARN("unimplemented target map return value: %d", r
);
1225 struct mapped_device
*md
;
1226 struct dm_table
*map
;
1230 unsigned sector_count
;
1233 static void bio_setup_sector(struct bio
*bio
, sector_t sector
, unsigned len
)
1235 bio
->bi_iter
.bi_sector
= sector
;
1236 bio
->bi_iter
.bi_size
= to_bytes(len
);
1240 * Creates a bio that consists of range of complete bvecs.
1242 static int clone_bio(struct dm_target_io
*tio
, struct bio
*bio
,
1243 sector_t sector
, unsigned len
)
1245 struct bio
*clone
= &tio
->clone
;
1247 __bio_clone_fast(clone
, bio
);
1249 if (unlikely(bio_integrity(bio
) != NULL
)) {
1252 if (unlikely(!dm_target_has_integrity(tio
->ti
->type
) &&
1253 !dm_target_passes_integrity(tio
->ti
->type
))) {
1254 DMWARN("%s: the target %s doesn't support integrity data.",
1255 dm_device_name(tio
->io
->md
),
1256 tio
->ti
->type
->name
);
1260 r
= bio_integrity_clone(clone
, bio
, GFP_NOIO
);
1265 if (bio_op(bio
) != REQ_OP_ZONE_REPORT
)
1266 bio_advance(clone
, to_bytes(sector
- clone
->bi_iter
.bi_sector
));
1267 clone
->bi_iter
.bi_size
= to_bytes(len
);
1269 if (unlikely(bio_integrity(bio
) != NULL
))
1270 bio_integrity_trim(clone
);
1275 static struct dm_target_io
*alloc_tio(struct clone_info
*ci
,
1276 struct dm_target
*ti
,
1277 unsigned target_bio_nr
)
1279 struct dm_target_io
*tio
;
1282 clone
= bio_alloc_bioset(GFP_NOIO
, 0, ci
->md
->bs
);
1283 tio
= container_of(clone
, struct dm_target_io
, clone
);
1287 tio
->target_bio_nr
= target_bio_nr
;
1292 static void __clone_and_map_simple_bio(struct clone_info
*ci
,
1293 struct dm_target
*ti
,
1294 unsigned target_bio_nr
, unsigned *len
)
1296 struct dm_target_io
*tio
= alloc_tio(ci
, ti
, target_bio_nr
);
1297 struct bio
*clone
= &tio
->clone
;
1301 __bio_clone_fast(clone
, ci
->bio
);
1303 bio_setup_sector(clone
, ci
->sector
, *len
);
1308 static void __send_duplicate_bios(struct clone_info
*ci
, struct dm_target
*ti
,
1309 unsigned num_bios
, unsigned *len
)
1311 unsigned target_bio_nr
;
1313 for (target_bio_nr
= 0; target_bio_nr
< num_bios
; target_bio_nr
++)
1314 __clone_and_map_simple_bio(ci
, ti
, target_bio_nr
, len
);
1317 static int __send_empty_flush(struct clone_info
*ci
)
1319 unsigned target_nr
= 0;
1320 struct dm_target
*ti
;
1322 BUG_ON(bio_has_data(ci
->bio
));
1323 while ((ti
= dm_table_get_target(ci
->map
, target_nr
++)))
1324 __send_duplicate_bios(ci
, ti
, ti
->num_flush_bios
, NULL
);
1329 static int __clone_and_map_data_bio(struct clone_info
*ci
, struct dm_target
*ti
,
1330 sector_t sector
, unsigned *len
)
1332 struct bio
*bio
= ci
->bio
;
1333 struct dm_target_io
*tio
;
1334 unsigned target_bio_nr
;
1335 unsigned num_target_bios
= 1;
1339 * Does the target want to receive duplicate copies of the bio?
1341 if (bio_data_dir(bio
) == WRITE
&& ti
->num_write_bios
)
1342 num_target_bios
= ti
->num_write_bios(ti
, bio
);
1344 for (target_bio_nr
= 0; target_bio_nr
< num_target_bios
; target_bio_nr
++) {
1345 tio
= alloc_tio(ci
, ti
, target_bio_nr
);
1347 r
= clone_bio(tio
, bio
, sector
, *len
);
1358 typedef unsigned (*get_num_bios_fn
)(struct dm_target
*ti
);
1360 static unsigned get_num_discard_bios(struct dm_target
*ti
)
1362 return ti
->num_discard_bios
;
1365 static unsigned get_num_write_same_bios(struct dm_target
*ti
)
1367 return ti
->num_write_same_bios
;
1370 static unsigned get_num_write_zeroes_bios(struct dm_target
*ti
)
1372 return ti
->num_write_zeroes_bios
;
1375 typedef bool (*is_split_required_fn
)(struct dm_target
*ti
);
1377 static bool is_split_required_for_discard(struct dm_target
*ti
)
1379 return ti
->split_discard_bios
;
1382 static int __send_changing_extent_only(struct clone_info
*ci
,
1383 get_num_bios_fn get_num_bios
,
1384 is_split_required_fn is_split_required
)
1386 struct dm_target
*ti
;
1391 ti
= dm_table_find_target(ci
->map
, ci
->sector
);
1392 if (!dm_target_is_valid(ti
))
1396 * Even though the device advertised support for this type of
1397 * request, that does not mean every target supports it, and
1398 * reconfiguration might also have changed that since the
1399 * check was performed.
1401 num_bios
= get_num_bios
? get_num_bios(ti
) : 0;
1405 if (is_split_required
&& !is_split_required(ti
))
1406 len
= min((sector_t
)ci
->sector_count
, max_io_len_target_boundary(ci
->sector
, ti
));
1408 len
= min((sector_t
)ci
->sector_count
, max_io_len(ci
->sector
, ti
));
1410 __send_duplicate_bios(ci
, ti
, num_bios
, &len
);
1413 } while (ci
->sector_count
-= len
);
1418 static int __send_discard(struct clone_info
*ci
)
1420 return __send_changing_extent_only(ci
, get_num_discard_bios
,
1421 is_split_required_for_discard
);
1424 static int __send_write_same(struct clone_info
*ci
)
1426 return __send_changing_extent_only(ci
, get_num_write_same_bios
, NULL
);
1429 static int __send_write_zeroes(struct clone_info
*ci
)
1431 return __send_changing_extent_only(ci
, get_num_write_zeroes_bios
, NULL
);
1435 * Select the correct strategy for processing a non-flush bio.
1437 static int __split_and_process_non_flush(struct clone_info
*ci
)
1439 struct bio
*bio
= ci
->bio
;
1440 struct dm_target
*ti
;
1444 if (unlikely(bio_op(bio
) == REQ_OP_DISCARD
))
1445 return __send_discard(ci
);
1446 else if (unlikely(bio_op(bio
) == REQ_OP_WRITE_SAME
))
1447 return __send_write_same(ci
);
1448 else if (unlikely(bio_op(bio
) == REQ_OP_WRITE_ZEROES
))
1449 return __send_write_zeroes(ci
);
1451 ti
= dm_table_find_target(ci
->map
, ci
->sector
);
1452 if (!dm_target_is_valid(ti
))
1455 if (bio_op(bio
) == REQ_OP_ZONE_REPORT
)
1456 len
= ci
->sector_count
;
1458 len
= min_t(sector_t
, max_io_len(ci
->sector
, ti
),
1461 r
= __clone_and_map_data_bio(ci
, ti
, ci
->sector
, &len
);
1466 ci
->sector_count
-= len
;
1472 * Entry point to split a bio into clones and submit them to the targets.
1474 static void __split_and_process_bio(struct mapped_device
*md
,
1475 struct dm_table
*map
, struct bio
*bio
)
1477 struct clone_info ci
;
1480 if (unlikely(!map
)) {
1487 ci
.io
= alloc_io(md
);
1489 atomic_set(&ci
.io
->io_count
, 1);
1492 spin_lock_init(&ci
.io
->endio_lock
);
1493 ci
.sector
= bio
->bi_iter
.bi_sector
;
1495 start_io_acct(ci
.io
);
1497 if (bio
->bi_opf
& REQ_PREFLUSH
) {
1498 ci
.bio
= &ci
.md
->flush_bio
;
1499 ci
.sector_count
= 0;
1500 error
= __send_empty_flush(&ci
);
1501 /* dec_pending submits any data associated with flush */
1502 } else if (bio_op(bio
) == REQ_OP_ZONE_RESET
) {
1504 ci
.sector_count
= 0;
1505 error
= __split_and_process_non_flush(&ci
);
1508 ci
.sector_count
= bio_sectors(bio
);
1509 while (ci
.sector_count
&& !error
)
1510 error
= __split_and_process_non_flush(&ci
);
1513 /* drop the extra reference count */
1514 dec_pending(ci
.io
, errno_to_blk_status(error
));
1516 /*-----------------------------------------------------------------
1518 *---------------------------------------------------------------*/
1521 * The request function that just remaps the bio built up by
1524 static blk_qc_t
dm_make_request(struct request_queue
*q
, struct bio
*bio
)
1526 int rw
= bio_data_dir(bio
);
1527 struct mapped_device
*md
= q
->queuedata
;
1529 struct dm_table
*map
;
1531 map
= dm_get_live_table(md
, &srcu_idx
);
1533 generic_start_io_acct(q
, rw
, bio_sectors(bio
), &dm_disk(md
)->part0
);
1535 /* if we're suspended, we have to queue this io for later */
1536 if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND
, &md
->flags
))) {
1537 dm_put_live_table(md
, srcu_idx
);
1539 if (!(bio
->bi_opf
& REQ_RAHEAD
))
1543 return BLK_QC_T_NONE
;
1546 __split_and_process_bio(md
, map
, bio
);
1547 dm_put_live_table(md
, srcu_idx
);
1548 return BLK_QC_T_NONE
;
1551 static int dm_any_congested(void *congested_data
, int bdi_bits
)
1554 struct mapped_device
*md
= congested_data
;
1555 struct dm_table
*map
;
1557 if (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND
, &md
->flags
)) {
1558 if (dm_request_based(md
)) {
1560 * With request-based DM we only need to check the
1561 * top-level queue for congestion.
1563 r
= md
->queue
->backing_dev_info
->wb
.state
& bdi_bits
;
1565 map
= dm_get_live_table_fast(md
);
1567 r
= dm_table_any_congested(map
, bdi_bits
);
1568 dm_put_live_table_fast(md
);
1575 /*-----------------------------------------------------------------
1576 * An IDR is used to keep track of allocated minor numbers.
1577 *---------------------------------------------------------------*/
1578 static void free_minor(int minor
)
1580 spin_lock(&_minor_lock
);
1581 idr_remove(&_minor_idr
, minor
);
1582 spin_unlock(&_minor_lock
);
1586 * See if the device with a specific minor # is free.
1588 static int specific_minor(int minor
)
1592 if (minor
>= (1 << MINORBITS
))
1595 idr_preload(GFP_KERNEL
);
1596 spin_lock(&_minor_lock
);
1598 r
= idr_alloc(&_minor_idr
, MINOR_ALLOCED
, minor
, minor
+ 1, GFP_NOWAIT
);
1600 spin_unlock(&_minor_lock
);
1603 return r
== -ENOSPC
? -EBUSY
: r
;
1607 static int next_free_minor(int *minor
)
1611 idr_preload(GFP_KERNEL
);
1612 spin_lock(&_minor_lock
);
1614 r
= idr_alloc(&_minor_idr
, MINOR_ALLOCED
, 0, 1 << MINORBITS
, GFP_NOWAIT
);
1616 spin_unlock(&_minor_lock
);
1624 static const struct block_device_operations dm_blk_dops
;
1625 static const struct dax_operations dm_dax_ops
;
1627 static void dm_wq_work(struct work_struct
*work
);
1629 void dm_init_md_queue(struct mapped_device
*md
)
1632 * Initialize data that will only be used by a non-blk-mq DM queue
1633 * - must do so here (in alloc_dev callchain) before queue is used
1635 md
->queue
->queuedata
= md
;
1636 md
->queue
->backing_dev_info
->congested_data
= md
;
1639 void dm_init_normal_md_queue(struct mapped_device
*md
)
1641 md
->use_blk_mq
= false;
1642 dm_init_md_queue(md
);
1645 * Initialize aspects of queue that aren't relevant for blk-mq
1647 md
->queue
->backing_dev_info
->congested_fn
= dm_any_congested
;
1650 static void cleanup_mapped_device(struct mapped_device
*md
)
1653 destroy_workqueue(md
->wq
);
1654 if (md
->kworker_task
)
1655 kthread_stop(md
->kworker_task
);
1656 mempool_destroy(md
->io_pool
);
1658 bioset_free(md
->bs
);
1661 kill_dax(md
->dax_dev
);
1662 put_dax(md
->dax_dev
);
1667 spin_lock(&_minor_lock
);
1668 md
->disk
->private_data
= NULL
;
1669 spin_unlock(&_minor_lock
);
1670 del_gendisk(md
->disk
);
1675 blk_cleanup_queue(md
->queue
);
1677 cleanup_srcu_struct(&md
->io_barrier
);
1684 dm_mq_cleanup_mapped_device(md
);
1688 * Allocate and initialise a blank device with a given minor.
1690 static struct mapped_device
*alloc_dev(int minor
)
1692 int r
, numa_node_id
= dm_get_numa_node();
1693 struct dax_device
*dax_dev
;
1694 struct mapped_device
*md
;
1697 md
= kvzalloc_node(sizeof(*md
), GFP_KERNEL
, numa_node_id
);
1699 DMWARN("unable to allocate device, out of memory.");
1703 if (!try_module_get(THIS_MODULE
))
1704 goto bad_module_get
;
1706 /* get a minor number for the dev */
1707 if (minor
== DM_ANY_MINOR
)
1708 r
= next_free_minor(&minor
);
1710 r
= specific_minor(minor
);
1714 r
= init_srcu_struct(&md
->io_barrier
);
1716 goto bad_io_barrier
;
1718 md
->numa_node_id
= numa_node_id
;
1719 md
->use_blk_mq
= dm_use_blk_mq_default();
1720 md
->init_tio_pdu
= false;
1721 md
->type
= DM_TYPE_NONE
;
1722 mutex_init(&md
->suspend_lock
);
1723 mutex_init(&md
->type_lock
);
1724 mutex_init(&md
->table_devices_lock
);
1725 spin_lock_init(&md
->deferred_lock
);
1726 atomic_set(&md
->holders
, 1);
1727 atomic_set(&md
->open_count
, 0);
1728 atomic_set(&md
->event_nr
, 0);
1729 atomic_set(&md
->uevent_seq
, 0);
1730 INIT_LIST_HEAD(&md
->uevent_list
);
1731 INIT_LIST_HEAD(&md
->table_devices
);
1732 spin_lock_init(&md
->uevent_lock
);
1734 md
->queue
= blk_alloc_queue_node(GFP_KERNEL
, numa_node_id
);
1738 dm_init_md_queue(md
);
1740 md
->disk
= alloc_disk_node(1, numa_node_id
);
1744 atomic_set(&md
->pending
[0], 0);
1745 atomic_set(&md
->pending
[1], 0);
1746 init_waitqueue_head(&md
->wait
);
1747 INIT_WORK(&md
->work
, dm_wq_work
);
1748 init_waitqueue_head(&md
->eventq
);
1749 init_completion(&md
->kobj_holder
.completion
);
1750 md
->kworker_task
= NULL
;
1752 md
->disk
->major
= _major
;
1753 md
->disk
->first_minor
= minor
;
1754 md
->disk
->fops
= &dm_blk_dops
;
1755 md
->disk
->queue
= md
->queue
;
1756 md
->disk
->private_data
= md
;
1757 sprintf(md
->disk
->disk_name
, "dm-%d", minor
);
1759 dax_dev
= alloc_dax(md
, md
->disk
->disk_name
, &dm_dax_ops
);
1762 md
->dax_dev
= dax_dev
;
1764 add_disk_no_queue_reg(md
->disk
);
1765 format_dev_t(md
->name
, MKDEV(_major
, minor
));
1767 md
->wq
= alloc_workqueue("kdmflush", WQ_MEM_RECLAIM
, 0);
1771 md
->bdev
= bdget_disk(md
->disk
, 0);
1775 bio_init(&md
->flush_bio
, NULL
, 0);
1776 bio_set_dev(&md
->flush_bio
, md
->bdev
);
1777 md
->flush_bio
.bi_opf
= REQ_OP_WRITE
| REQ_PREFLUSH
| REQ_SYNC
;
1779 dm_stats_init(&md
->stats
);
1781 /* Populate the mapping, nobody knows we exist yet */
1782 spin_lock(&_minor_lock
);
1783 old_md
= idr_replace(&_minor_idr
, md
, minor
);
1784 spin_unlock(&_minor_lock
);
1786 BUG_ON(old_md
!= MINOR_ALLOCED
);
1791 cleanup_mapped_device(md
);
1795 module_put(THIS_MODULE
);
1801 static void unlock_fs(struct mapped_device
*md
);
1803 static void free_dev(struct mapped_device
*md
)
1805 int minor
= MINOR(disk_devt(md
->disk
));
1809 cleanup_mapped_device(md
);
1811 free_table_devices(&md
->table_devices
);
1812 dm_stats_cleanup(&md
->stats
);
1815 module_put(THIS_MODULE
);
1819 static void __bind_mempools(struct mapped_device
*md
, struct dm_table
*t
)
1821 struct dm_md_mempools
*p
= dm_table_get_md_mempools(t
);
1824 /* The md already has necessary mempools. */
1825 if (dm_table_bio_based(t
)) {
1827 * Reload bioset because front_pad may have changed
1828 * because a different table was loaded.
1830 bioset_free(md
->bs
);
1835 * There's no need to reload with request-based dm
1836 * because the size of front_pad doesn't change.
1837 * Note for future: If you are to reload bioset,
1838 * prep-ed requests in the queue may refer
1839 * to bio from the old bioset, so you must walk
1840 * through the queue to unprep.
1845 BUG_ON(!p
|| md
->io_pool
|| md
->bs
);
1847 md
->io_pool
= p
->io_pool
;
1853 /* mempool bind completed, no longer need any mempools in the table */
1854 dm_table_free_md_mempools(t
);
1858 * Bind a table to the device.
1860 static void event_callback(void *context
)
1862 unsigned long flags
;
1864 struct mapped_device
*md
= (struct mapped_device
*) context
;
1866 spin_lock_irqsave(&md
->uevent_lock
, flags
);
1867 list_splice_init(&md
->uevent_list
, &uevents
);
1868 spin_unlock_irqrestore(&md
->uevent_lock
, flags
);
1870 dm_send_uevents(&uevents
, &disk_to_dev(md
->disk
)->kobj
);
1872 atomic_inc(&md
->event_nr
);
1873 wake_up(&md
->eventq
);
1874 dm_issue_global_event();
1878 * Protected by md->suspend_lock obtained by dm_swap_table().
1880 static void __set_size(struct mapped_device
*md
, sector_t size
)
1882 lockdep_assert_held(&md
->suspend_lock
);
1884 set_capacity(md
->disk
, size
);
1886 i_size_write(md
->bdev
->bd_inode
, (loff_t
)size
<< SECTOR_SHIFT
);
1890 * Returns old map, which caller must destroy.
1892 static struct dm_table
*__bind(struct mapped_device
*md
, struct dm_table
*t
,
1893 struct queue_limits
*limits
)
1895 struct dm_table
*old_map
;
1896 struct request_queue
*q
= md
->queue
;
1899 lockdep_assert_held(&md
->suspend_lock
);
1901 size
= dm_table_get_size(t
);
1904 * Wipe any geometry if the size of the table changed.
1906 if (size
!= dm_get_size(md
))
1907 memset(&md
->geometry
, 0, sizeof(md
->geometry
));
1909 __set_size(md
, size
);
1911 dm_table_event_callback(t
, event_callback
, md
);
1914 * The queue hasn't been stopped yet, if the old table type wasn't
1915 * for request-based during suspension. So stop it to prevent
1916 * I/O mapping before resume.
1917 * This must be done before setting the queue restrictions,
1918 * because request-based dm may be run just after the setting.
1920 if (dm_table_request_based(t
)) {
1923 * Leverage the fact that request-based DM targets are
1924 * immutable singletons and establish md->immutable_target
1925 * - used to optimize both dm_request_fn and dm_mq_queue_rq
1927 md
->immutable_target
= dm_table_get_immutable_target(t
);
1930 __bind_mempools(md
, t
);
1932 old_map
= rcu_dereference_protected(md
->map
, lockdep_is_held(&md
->suspend_lock
));
1933 rcu_assign_pointer(md
->map
, (void *)t
);
1934 md
->immutable_target_type
= dm_table_get_immutable_target_type(t
);
1936 dm_table_set_restrictions(t
, q
, limits
);
1944 * Returns unbound table for the caller to free.
1946 static struct dm_table
*__unbind(struct mapped_device
*md
)
1948 struct dm_table
*map
= rcu_dereference_protected(md
->map
, 1);
1953 dm_table_event_callback(map
, NULL
, NULL
);
1954 RCU_INIT_POINTER(md
->map
, NULL
);
1961 * Constructor for a new device.
1963 int dm_create(int minor
, struct mapped_device
**result
)
1965 struct mapped_device
*md
;
1967 md
= alloc_dev(minor
);
1978 * Functions to manage md->type.
1979 * All are required to hold md->type_lock.
1981 void dm_lock_md_type(struct mapped_device
*md
)
1983 mutex_lock(&md
->type_lock
);
1986 void dm_unlock_md_type(struct mapped_device
*md
)
1988 mutex_unlock(&md
->type_lock
);
1991 void dm_set_md_type(struct mapped_device
*md
, enum dm_queue_mode type
)
1993 BUG_ON(!mutex_is_locked(&md
->type_lock
));
1997 enum dm_queue_mode
dm_get_md_type(struct mapped_device
*md
)
2002 struct target_type
*dm_get_immutable_target_type(struct mapped_device
*md
)
2004 return md
->immutable_target_type
;
2008 * The queue_limits are only valid as long as you have a reference
2011 struct queue_limits
*dm_get_queue_limits(struct mapped_device
*md
)
2013 BUG_ON(!atomic_read(&md
->holders
));
2014 return &md
->queue
->limits
;
2016 EXPORT_SYMBOL_GPL(dm_get_queue_limits
);
2019 * Setup the DM device's queue based on md's type
2021 int dm_setup_md_queue(struct mapped_device
*md
, struct dm_table
*t
)
2024 struct queue_limits limits
;
2025 enum dm_queue_mode type
= dm_get_md_type(md
);
2028 case DM_TYPE_REQUEST_BASED
:
2029 r
= dm_old_init_request_queue(md
, t
);
2031 DMERR("Cannot initialize queue for request-based mapped device");
2035 case DM_TYPE_MQ_REQUEST_BASED
:
2036 r
= dm_mq_init_request_queue(md
, t
);
2038 DMERR("Cannot initialize queue for request-based dm-mq mapped device");
2042 case DM_TYPE_BIO_BASED
:
2043 case DM_TYPE_DAX_BIO_BASED
:
2044 dm_init_normal_md_queue(md
);
2045 blk_queue_make_request(md
->queue
, dm_make_request
);
2047 * DM handles splitting bios as needed. Free the bio_split bioset
2048 * since it won't be used (saves 1 process per bio-based DM device).
2050 bioset_free(md
->queue
->bio_split
);
2051 md
->queue
->bio_split
= NULL
;
2053 if (type
== DM_TYPE_DAX_BIO_BASED
)
2054 queue_flag_set_unlocked(QUEUE_FLAG_DAX
, md
->queue
);
2061 r
= dm_calculate_queue_limits(t
, &limits
);
2063 DMERR("Cannot calculate initial queue limits");
2066 dm_table_set_restrictions(t
, md
->queue
, &limits
);
2067 blk_register_queue(md
->disk
);
2072 struct mapped_device
*dm_get_md(dev_t dev
)
2074 struct mapped_device
*md
;
2075 unsigned minor
= MINOR(dev
);
2077 if (MAJOR(dev
) != _major
|| minor
>= (1 << MINORBITS
))
2080 spin_lock(&_minor_lock
);
2082 md
= idr_find(&_minor_idr
, minor
);
2083 if (!md
|| md
== MINOR_ALLOCED
|| (MINOR(disk_devt(dm_disk(md
))) != minor
) ||
2084 test_bit(DMF_FREEING
, &md
->flags
) || dm_deleting_md(md
)) {
2090 spin_unlock(&_minor_lock
);
2094 EXPORT_SYMBOL_GPL(dm_get_md
);
2096 void *dm_get_mdptr(struct mapped_device
*md
)
2098 return md
->interface_ptr
;
2101 void dm_set_mdptr(struct mapped_device
*md
, void *ptr
)
2103 md
->interface_ptr
= ptr
;
2106 void dm_get(struct mapped_device
*md
)
2108 atomic_inc(&md
->holders
);
2109 BUG_ON(test_bit(DMF_FREEING
, &md
->flags
));
2112 int dm_hold(struct mapped_device
*md
)
2114 spin_lock(&_minor_lock
);
2115 if (test_bit(DMF_FREEING
, &md
->flags
)) {
2116 spin_unlock(&_minor_lock
);
2120 spin_unlock(&_minor_lock
);
2123 EXPORT_SYMBOL_GPL(dm_hold
);
2125 const char *dm_device_name(struct mapped_device
*md
)
2129 EXPORT_SYMBOL_GPL(dm_device_name
);
2131 static void __dm_destroy(struct mapped_device
*md
, bool wait
)
2133 struct request_queue
*q
= dm_get_md_queue(md
);
2134 struct dm_table
*map
;
2139 spin_lock(&_minor_lock
);
2140 idr_replace(&_minor_idr
, MINOR_ALLOCED
, MINOR(disk_devt(dm_disk(md
))));
2141 set_bit(DMF_FREEING
, &md
->flags
);
2142 spin_unlock(&_minor_lock
);
2144 blk_set_queue_dying(q
);
2146 if (dm_request_based(md
) && md
->kworker_task
)
2147 kthread_flush_worker(&md
->kworker
);
2150 * Take suspend_lock so that presuspend and postsuspend methods
2151 * do not race with internal suspend.
2153 mutex_lock(&md
->suspend_lock
);
2154 map
= dm_get_live_table(md
, &srcu_idx
);
2155 if (!dm_suspended_md(md
)) {
2156 dm_table_presuspend_targets(map
);
2157 dm_table_postsuspend_targets(map
);
2159 /* dm_put_live_table must be before msleep, otherwise deadlock is possible */
2160 dm_put_live_table(md
, srcu_idx
);
2161 mutex_unlock(&md
->suspend_lock
);
2164 * Rare, but there may be I/O requests still going to complete,
2165 * for example. Wait for all references to disappear.
2166 * No one should increment the reference count of the mapped_device,
2167 * after the mapped_device state becomes DMF_FREEING.
2170 while (atomic_read(&md
->holders
))
2172 else if (atomic_read(&md
->holders
))
2173 DMWARN("%s: Forcibly removing mapped_device still in use! (%d users)",
2174 dm_device_name(md
), atomic_read(&md
->holders
));
2177 dm_table_destroy(__unbind(md
));
2181 void dm_destroy(struct mapped_device
*md
)
2183 __dm_destroy(md
, true);
2186 void dm_destroy_immediate(struct mapped_device
*md
)
2188 __dm_destroy(md
, false);
2191 void dm_put(struct mapped_device
*md
)
2193 atomic_dec(&md
->holders
);
2195 EXPORT_SYMBOL_GPL(dm_put
);
2197 static int dm_wait_for_completion(struct mapped_device
*md
, long task_state
)
2203 prepare_to_wait(&md
->wait
, &wait
, task_state
);
2205 if (!md_in_flight(md
))
2208 if (signal_pending_state(task_state
, current
)) {
2215 finish_wait(&md
->wait
, &wait
);
2221 * Process the deferred bios
2223 static void dm_wq_work(struct work_struct
*work
)
2225 struct mapped_device
*md
= container_of(work
, struct mapped_device
,
2229 struct dm_table
*map
;
2231 map
= dm_get_live_table(md
, &srcu_idx
);
2233 while (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND
, &md
->flags
)) {
2234 spin_lock_irq(&md
->deferred_lock
);
2235 c
= bio_list_pop(&md
->deferred
);
2236 spin_unlock_irq(&md
->deferred_lock
);
2241 if (dm_request_based(md
))
2242 generic_make_request(c
);
2244 __split_and_process_bio(md
, map
, c
);
2247 dm_put_live_table(md
, srcu_idx
);
2250 static void dm_queue_flush(struct mapped_device
*md
)
2252 clear_bit(DMF_BLOCK_IO_FOR_SUSPEND
, &md
->flags
);
2253 smp_mb__after_atomic();
2254 queue_work(md
->wq
, &md
->work
);
2258 * Swap in a new table, returning the old one for the caller to destroy.
2260 struct dm_table
*dm_swap_table(struct mapped_device
*md
, struct dm_table
*table
)
2262 struct dm_table
*live_map
= NULL
, *map
= ERR_PTR(-EINVAL
);
2263 struct queue_limits limits
;
2266 mutex_lock(&md
->suspend_lock
);
2268 /* device must be suspended */
2269 if (!dm_suspended_md(md
))
2273 * If the new table has no data devices, retain the existing limits.
2274 * This helps multipath with queue_if_no_path if all paths disappear,
2275 * then new I/O is queued based on these limits, and then some paths
2278 if (dm_table_has_no_data_devices(table
)) {
2279 live_map
= dm_get_live_table_fast(md
);
2281 limits
= md
->queue
->limits
;
2282 dm_put_live_table_fast(md
);
2286 r
= dm_calculate_queue_limits(table
, &limits
);
2293 map
= __bind(md
, table
, &limits
);
2294 dm_issue_global_event();
2297 mutex_unlock(&md
->suspend_lock
);
2302 * Functions to lock and unlock any filesystem running on the
2305 static int lock_fs(struct mapped_device
*md
)
2309 WARN_ON(md
->frozen_sb
);
2311 md
->frozen_sb
= freeze_bdev(md
->bdev
);
2312 if (IS_ERR(md
->frozen_sb
)) {
2313 r
= PTR_ERR(md
->frozen_sb
);
2314 md
->frozen_sb
= NULL
;
2318 set_bit(DMF_FROZEN
, &md
->flags
);
2323 static void unlock_fs(struct mapped_device
*md
)
2325 if (!test_bit(DMF_FROZEN
, &md
->flags
))
2328 thaw_bdev(md
->bdev
, md
->frozen_sb
);
2329 md
->frozen_sb
= NULL
;
2330 clear_bit(DMF_FROZEN
, &md
->flags
);
2334 * @suspend_flags: DM_SUSPEND_LOCKFS_FLAG and/or DM_SUSPEND_NOFLUSH_FLAG
2335 * @task_state: e.g. TASK_INTERRUPTIBLE or TASK_UNINTERRUPTIBLE
2336 * @dmf_suspended_flag: DMF_SUSPENDED or DMF_SUSPENDED_INTERNALLY
2338 * If __dm_suspend returns 0, the device is completely quiescent
2339 * now. There is no request-processing activity. All new requests
2340 * are being added to md->deferred list.
2342 static int __dm_suspend(struct mapped_device
*md
, struct dm_table
*map
,
2343 unsigned suspend_flags
, long task_state
,
2344 int dmf_suspended_flag
)
2346 bool do_lockfs
= suspend_flags
& DM_SUSPEND_LOCKFS_FLAG
;
2347 bool noflush
= suspend_flags
& DM_SUSPEND_NOFLUSH_FLAG
;
2350 lockdep_assert_held(&md
->suspend_lock
);
2353 * DMF_NOFLUSH_SUSPENDING must be set before presuspend.
2354 * This flag is cleared before dm_suspend returns.
2357 set_bit(DMF_NOFLUSH_SUSPENDING
, &md
->flags
);
2359 pr_debug("%s: suspending with flush\n", dm_device_name(md
));
2362 * This gets reverted if there's an error later and the targets
2363 * provide the .presuspend_undo hook.
2365 dm_table_presuspend_targets(map
);
2368 * Flush I/O to the device.
2369 * Any I/O submitted after lock_fs() may not be flushed.
2370 * noflush takes precedence over do_lockfs.
2371 * (lock_fs() flushes I/Os and waits for them to complete.)
2373 if (!noflush
&& do_lockfs
) {
2376 dm_table_presuspend_undo_targets(map
);
2382 * Here we must make sure that no processes are submitting requests
2383 * to target drivers i.e. no one may be executing
2384 * __split_and_process_bio. This is called from dm_request and
2387 * To get all processes out of __split_and_process_bio in dm_request,
2388 * we take the write lock. To prevent any process from reentering
2389 * __split_and_process_bio from dm_request and quiesce the thread
2390 * (dm_wq_work), we set BMF_BLOCK_IO_FOR_SUSPEND and call
2391 * flush_workqueue(md->wq).
2393 set_bit(DMF_BLOCK_IO_FOR_SUSPEND
, &md
->flags
);
2395 synchronize_srcu(&md
->io_barrier
);
2398 * Stop md->queue before flushing md->wq in case request-based
2399 * dm defers requests to md->wq from md->queue.
2401 if (dm_request_based(md
)) {
2402 dm_stop_queue(md
->queue
);
2403 if (md
->kworker_task
)
2404 kthread_flush_worker(&md
->kworker
);
2407 flush_workqueue(md
->wq
);
2410 * At this point no more requests are entering target request routines.
2411 * We call dm_wait_for_completion to wait for all existing requests
2414 r
= dm_wait_for_completion(md
, task_state
);
2416 set_bit(dmf_suspended_flag
, &md
->flags
);
2419 clear_bit(DMF_NOFLUSH_SUSPENDING
, &md
->flags
);
2421 synchronize_srcu(&md
->io_barrier
);
2423 /* were we interrupted ? */
2427 if (dm_request_based(md
))
2428 dm_start_queue(md
->queue
);
2431 dm_table_presuspend_undo_targets(map
);
2432 /* pushback list is already flushed, so skip flush */
2439 * We need to be able to change a mapping table under a mounted
2440 * filesystem. For example we might want to move some data in
2441 * the background. Before the table can be swapped with
2442 * dm_bind_table, dm_suspend must be called to flush any in
2443 * flight bios and ensure that any further io gets deferred.
2446 * Suspend mechanism in request-based dm.
2448 * 1. Flush all I/Os by lock_fs() if needed.
2449 * 2. Stop dispatching any I/O by stopping the request_queue.
2450 * 3. Wait for all in-flight I/Os to be completed or requeued.
2452 * To abort suspend, start the request_queue.
2454 int dm_suspend(struct mapped_device
*md
, unsigned suspend_flags
)
2456 struct dm_table
*map
= NULL
;
2460 mutex_lock_nested(&md
->suspend_lock
, SINGLE_DEPTH_NESTING
);
2462 if (dm_suspended_md(md
)) {
2467 if (dm_suspended_internally_md(md
)) {
2468 /* already internally suspended, wait for internal resume */
2469 mutex_unlock(&md
->suspend_lock
);
2470 r
= wait_on_bit(&md
->flags
, DMF_SUSPENDED_INTERNALLY
, TASK_INTERRUPTIBLE
);
2476 map
= rcu_dereference_protected(md
->map
, lockdep_is_held(&md
->suspend_lock
));
2478 r
= __dm_suspend(md
, map
, suspend_flags
, TASK_INTERRUPTIBLE
, DMF_SUSPENDED
);
2482 dm_table_postsuspend_targets(map
);
2485 mutex_unlock(&md
->suspend_lock
);
2489 static int __dm_resume(struct mapped_device
*md
, struct dm_table
*map
)
2492 int r
= dm_table_resume_targets(map
);
2500 * Flushing deferred I/Os must be done after targets are resumed
2501 * so that mapping of targets can work correctly.
2502 * Request-based dm is queueing the deferred I/Os in its request_queue.
2504 if (dm_request_based(md
))
2505 dm_start_queue(md
->queue
);
2512 int dm_resume(struct mapped_device
*md
)
2515 struct dm_table
*map
= NULL
;
2519 mutex_lock_nested(&md
->suspend_lock
, SINGLE_DEPTH_NESTING
);
2521 if (!dm_suspended_md(md
))
2524 if (dm_suspended_internally_md(md
)) {
2525 /* already internally suspended, wait for internal resume */
2526 mutex_unlock(&md
->suspend_lock
);
2527 r
= wait_on_bit(&md
->flags
, DMF_SUSPENDED_INTERNALLY
, TASK_INTERRUPTIBLE
);
2533 map
= rcu_dereference_protected(md
->map
, lockdep_is_held(&md
->suspend_lock
));
2534 if (!map
|| !dm_table_get_size(map
))
2537 r
= __dm_resume(md
, map
);
2541 clear_bit(DMF_SUSPENDED
, &md
->flags
);
2543 mutex_unlock(&md
->suspend_lock
);
2549 * Internal suspend/resume works like userspace-driven suspend. It waits
2550 * until all bios finish and prevents issuing new bios to the target drivers.
2551 * It may be used only from the kernel.
2554 static void __dm_internal_suspend(struct mapped_device
*md
, unsigned suspend_flags
)
2556 struct dm_table
*map
= NULL
;
2558 lockdep_assert_held(&md
->suspend_lock
);
2560 if (md
->internal_suspend_count
++)
2561 return; /* nested internal suspend */
2563 if (dm_suspended_md(md
)) {
2564 set_bit(DMF_SUSPENDED_INTERNALLY
, &md
->flags
);
2565 return; /* nest suspend */
2568 map
= rcu_dereference_protected(md
->map
, lockdep_is_held(&md
->suspend_lock
));
2571 * Using TASK_UNINTERRUPTIBLE because only NOFLUSH internal suspend is
2572 * supported. Properly supporting a TASK_INTERRUPTIBLE internal suspend
2573 * would require changing .presuspend to return an error -- avoid this
2574 * until there is a need for more elaborate variants of internal suspend.
2576 (void) __dm_suspend(md
, map
, suspend_flags
, TASK_UNINTERRUPTIBLE
,
2577 DMF_SUSPENDED_INTERNALLY
);
2579 dm_table_postsuspend_targets(map
);
2582 static void __dm_internal_resume(struct mapped_device
*md
)
2584 BUG_ON(!md
->internal_suspend_count
);
2586 if (--md
->internal_suspend_count
)
2587 return; /* resume from nested internal suspend */
2589 if (dm_suspended_md(md
))
2590 goto done
; /* resume from nested suspend */
2593 * NOTE: existing callers don't need to call dm_table_resume_targets
2594 * (which may fail -- so best to avoid it for now by passing NULL map)
2596 (void) __dm_resume(md
, NULL
);
2599 clear_bit(DMF_SUSPENDED_INTERNALLY
, &md
->flags
);
2600 smp_mb__after_atomic();
2601 wake_up_bit(&md
->flags
, DMF_SUSPENDED_INTERNALLY
);
2604 void dm_internal_suspend_noflush(struct mapped_device
*md
)
2606 mutex_lock(&md
->suspend_lock
);
2607 __dm_internal_suspend(md
, DM_SUSPEND_NOFLUSH_FLAG
);
2608 mutex_unlock(&md
->suspend_lock
);
2610 EXPORT_SYMBOL_GPL(dm_internal_suspend_noflush
);
2612 void dm_internal_resume(struct mapped_device
*md
)
2614 mutex_lock(&md
->suspend_lock
);
2615 __dm_internal_resume(md
);
2616 mutex_unlock(&md
->suspend_lock
);
2618 EXPORT_SYMBOL_GPL(dm_internal_resume
);
2621 * Fast variants of internal suspend/resume hold md->suspend_lock,
2622 * which prevents interaction with userspace-driven suspend.
2625 void dm_internal_suspend_fast(struct mapped_device
*md
)
2627 mutex_lock(&md
->suspend_lock
);
2628 if (dm_suspended_md(md
) || dm_suspended_internally_md(md
))
2631 set_bit(DMF_BLOCK_IO_FOR_SUSPEND
, &md
->flags
);
2632 synchronize_srcu(&md
->io_barrier
);
2633 flush_workqueue(md
->wq
);
2634 dm_wait_for_completion(md
, TASK_UNINTERRUPTIBLE
);
2636 EXPORT_SYMBOL_GPL(dm_internal_suspend_fast
);
2638 void dm_internal_resume_fast(struct mapped_device
*md
)
2640 if (dm_suspended_md(md
) || dm_suspended_internally_md(md
))
2646 mutex_unlock(&md
->suspend_lock
);
2648 EXPORT_SYMBOL_GPL(dm_internal_resume_fast
);
2650 /*-----------------------------------------------------------------
2651 * Event notification.
2652 *---------------------------------------------------------------*/
2653 int dm_kobject_uevent(struct mapped_device
*md
, enum kobject_action action
,
2656 char udev_cookie
[DM_COOKIE_LENGTH
];
2657 char *envp
[] = { udev_cookie
, NULL
};
2660 return kobject_uevent(&disk_to_dev(md
->disk
)->kobj
, action
);
2662 snprintf(udev_cookie
, DM_COOKIE_LENGTH
, "%s=%u",
2663 DM_COOKIE_ENV_VAR_NAME
, cookie
);
2664 return kobject_uevent_env(&disk_to_dev(md
->disk
)->kobj
,
2669 uint32_t dm_next_uevent_seq(struct mapped_device
*md
)
2671 return atomic_add_return(1, &md
->uevent_seq
);
2674 uint32_t dm_get_event_nr(struct mapped_device
*md
)
2676 return atomic_read(&md
->event_nr
);
2679 int dm_wait_event(struct mapped_device
*md
, int event_nr
)
2681 return wait_event_interruptible(md
->eventq
,
2682 (event_nr
!= atomic_read(&md
->event_nr
)));
2685 void dm_uevent_add(struct mapped_device
*md
, struct list_head
*elist
)
2687 unsigned long flags
;
2689 spin_lock_irqsave(&md
->uevent_lock
, flags
);
2690 list_add(elist
, &md
->uevent_list
);
2691 spin_unlock_irqrestore(&md
->uevent_lock
, flags
);
2695 * The gendisk is only valid as long as you have a reference
2698 struct gendisk
*dm_disk(struct mapped_device
*md
)
2702 EXPORT_SYMBOL_GPL(dm_disk
);
2704 struct kobject
*dm_kobject(struct mapped_device
*md
)
2706 return &md
->kobj_holder
.kobj
;
2709 struct mapped_device
*dm_get_from_kobject(struct kobject
*kobj
)
2711 struct mapped_device
*md
;
2713 md
= container_of(kobj
, struct mapped_device
, kobj_holder
.kobj
);
2715 spin_lock(&_minor_lock
);
2716 if (test_bit(DMF_FREEING
, &md
->flags
) || dm_deleting_md(md
)) {
2722 spin_unlock(&_minor_lock
);
2727 int dm_suspended_md(struct mapped_device
*md
)
2729 return test_bit(DMF_SUSPENDED
, &md
->flags
);
2732 int dm_suspended_internally_md(struct mapped_device
*md
)
2734 return test_bit(DMF_SUSPENDED_INTERNALLY
, &md
->flags
);
2737 int dm_test_deferred_remove_flag(struct mapped_device
*md
)
2739 return test_bit(DMF_DEFERRED_REMOVE
, &md
->flags
);
2742 int dm_suspended(struct dm_target
*ti
)
2744 return dm_suspended_md(dm_table_get_md(ti
->table
));
2746 EXPORT_SYMBOL_GPL(dm_suspended
);
2748 int dm_noflush_suspending(struct dm_target
*ti
)
2750 return __noflush_suspending(dm_table_get_md(ti
->table
));
2752 EXPORT_SYMBOL_GPL(dm_noflush_suspending
);
2754 struct dm_md_mempools
*dm_alloc_md_mempools(struct mapped_device
*md
, enum dm_queue_mode type
,
2755 unsigned integrity
, unsigned per_io_data_size
)
2757 struct dm_md_mempools
*pools
= kzalloc_node(sizeof(*pools
), GFP_KERNEL
, md
->numa_node_id
);
2758 unsigned int pool_size
= 0;
2759 unsigned int front_pad
;
2765 case DM_TYPE_BIO_BASED
:
2766 case DM_TYPE_DAX_BIO_BASED
:
2767 pool_size
= dm_get_reserved_bio_based_ios();
2768 front_pad
= roundup(per_io_data_size
, __alignof__(struct dm_target_io
)) + offsetof(struct dm_target_io
, clone
);
2770 pools
->io_pool
= mempool_create_slab_pool(pool_size
, _io_cache
);
2771 if (!pools
->io_pool
)
2774 case DM_TYPE_REQUEST_BASED
:
2775 case DM_TYPE_MQ_REQUEST_BASED
:
2776 pool_size
= dm_get_reserved_rq_based_ios();
2777 front_pad
= offsetof(struct dm_rq_clone_bio_info
, clone
);
2778 /* per_io_data_size is used for blk-mq pdu at queue allocation */
2784 pools
->bs
= bioset_create(pool_size
, front_pad
, BIOSET_NEED_RESCUER
);
2788 if (integrity
&& bioset_integrity_create(pools
->bs
, pool_size
))
2794 dm_free_md_mempools(pools
);
2799 void dm_free_md_mempools(struct dm_md_mempools
*pools
)
2804 mempool_destroy(pools
->io_pool
);
2807 bioset_free(pools
->bs
);
2819 static int dm_call_pr(struct block_device
*bdev
, iterate_devices_callout_fn fn
,
2822 struct mapped_device
*md
= bdev
->bd_disk
->private_data
;
2823 struct dm_table
*table
;
2824 struct dm_target
*ti
;
2825 int ret
= -ENOTTY
, srcu_idx
;
2827 table
= dm_get_live_table(md
, &srcu_idx
);
2828 if (!table
|| !dm_table_get_size(table
))
2831 /* We only support devices that have a single target */
2832 if (dm_table_get_num_targets(table
) != 1)
2834 ti
= dm_table_get_target(table
, 0);
2837 if (!ti
->type
->iterate_devices
)
2840 ret
= ti
->type
->iterate_devices(ti
, fn
, data
);
2842 dm_put_live_table(md
, srcu_idx
);
2847 * For register / unregister we need to manually call out to every path.
2849 static int __dm_pr_register(struct dm_target
*ti
, struct dm_dev
*dev
,
2850 sector_t start
, sector_t len
, void *data
)
2852 struct dm_pr
*pr
= data
;
2853 const struct pr_ops
*ops
= dev
->bdev
->bd_disk
->fops
->pr_ops
;
2855 if (!ops
|| !ops
->pr_register
)
2857 return ops
->pr_register(dev
->bdev
, pr
->old_key
, pr
->new_key
, pr
->flags
);
2860 static int dm_pr_register(struct block_device
*bdev
, u64 old_key
, u64 new_key
,
2871 ret
= dm_call_pr(bdev
, __dm_pr_register
, &pr
);
2872 if (ret
&& new_key
) {
2873 /* unregister all paths if we failed to register any path */
2874 pr
.old_key
= new_key
;
2877 pr
.fail_early
= false;
2878 dm_call_pr(bdev
, __dm_pr_register
, &pr
);
2884 static int dm_pr_reserve(struct block_device
*bdev
, u64 key
, enum pr_type type
,
2887 struct mapped_device
*md
= bdev
->bd_disk
->private_data
;
2888 const struct pr_ops
*ops
;
2892 r
= dm_grab_bdev_for_ioctl(md
, &bdev
, &mode
);
2896 ops
= bdev
->bd_disk
->fops
->pr_ops
;
2897 if (ops
&& ops
->pr_reserve
)
2898 r
= ops
->pr_reserve(bdev
, key
, type
, flags
);
2906 static int dm_pr_release(struct block_device
*bdev
, u64 key
, enum pr_type type
)
2908 struct mapped_device
*md
= bdev
->bd_disk
->private_data
;
2909 const struct pr_ops
*ops
;
2913 r
= dm_grab_bdev_for_ioctl(md
, &bdev
, &mode
);
2917 ops
= bdev
->bd_disk
->fops
->pr_ops
;
2918 if (ops
&& ops
->pr_release
)
2919 r
= ops
->pr_release(bdev
, key
, type
);
2927 static int dm_pr_preempt(struct block_device
*bdev
, u64 old_key
, u64 new_key
,
2928 enum pr_type type
, bool abort
)
2930 struct mapped_device
*md
= bdev
->bd_disk
->private_data
;
2931 const struct pr_ops
*ops
;
2935 r
= dm_grab_bdev_for_ioctl(md
, &bdev
, &mode
);
2939 ops
= bdev
->bd_disk
->fops
->pr_ops
;
2940 if (ops
&& ops
->pr_preempt
)
2941 r
= ops
->pr_preempt(bdev
, old_key
, new_key
, type
, abort
);
2949 static int dm_pr_clear(struct block_device
*bdev
, u64 key
)
2951 struct mapped_device
*md
= bdev
->bd_disk
->private_data
;
2952 const struct pr_ops
*ops
;
2956 r
= dm_grab_bdev_for_ioctl(md
, &bdev
, &mode
);
2960 ops
= bdev
->bd_disk
->fops
->pr_ops
;
2961 if (ops
&& ops
->pr_clear
)
2962 r
= ops
->pr_clear(bdev
, key
);
2970 static const struct pr_ops dm_pr_ops
= {
2971 .pr_register
= dm_pr_register
,
2972 .pr_reserve
= dm_pr_reserve
,
2973 .pr_release
= dm_pr_release
,
2974 .pr_preempt
= dm_pr_preempt
,
2975 .pr_clear
= dm_pr_clear
,
2978 static const struct block_device_operations dm_blk_dops
= {
2979 .open
= dm_blk_open
,
2980 .release
= dm_blk_close
,
2981 .ioctl
= dm_blk_ioctl
,
2982 .getgeo
= dm_blk_getgeo
,
2983 .pr_ops
= &dm_pr_ops
,
2984 .owner
= THIS_MODULE
2987 static const struct dax_operations dm_dax_ops
= {
2988 .direct_access
= dm_dax_direct_access
,
2989 .copy_from_iter
= dm_dax_copy_from_iter
,
2995 module_init(dm_init
);
2996 module_exit(dm_exit
);
2998 module_param(major
, uint
, 0);
2999 MODULE_PARM_DESC(major
, "The major number of the device mapper");
3001 module_param(reserved_bio_based_ios
, uint
, S_IRUGO
| S_IWUSR
);
3002 MODULE_PARM_DESC(reserved_bio_based_ios
, "Reserved IOs in bio-based mempools");
3004 module_param(dm_numa_node
, int, S_IRUGO
| S_IWUSR
);
3005 MODULE_PARM_DESC(dm_numa_node
, "NUMA node for DM device memory allocations");
3007 MODULE_DESCRIPTION(DM_NAME
" driver");
3008 MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
3009 MODULE_LICENSE("GPL");