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.
11 #include <linux/init.h>
12 #include <linux/module.h>
13 #include <linux/mutex.h>
14 #include <linux/moduleparam.h>
15 #include <linux/blkpg.h>
16 #include <linux/bio.h>
17 #include <linux/mempool.h>
18 #include <linux/slab.h>
19 #include <linux/idr.h>
20 #include <linux/hdreg.h>
21 #include <linux/delay.h>
22 #include <linux/wait.h>
23 #include <linux/kthread.h>
25 #include <trace/events/block.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
;
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
;
76 * For request-based dm.
77 * One of these is allocated per request.
79 struct dm_rq_target_io
{
80 struct mapped_device
*md
;
82 struct request
*orig
, *clone
;
83 struct kthread_work work
;
89 * For request-based dm - the bio clones we allocate are embedded in these
92 * We allocate these with bio_alloc_bioset, using the front_pad parameter when
93 * the bioset is created - this means the bio has to come at the end of the
96 struct dm_rq_clone_bio_info
{
98 struct dm_rq_target_io
*tio
;
102 union map_info
*dm_get_rq_mapinfo(struct request
*rq
)
104 if (rq
&& rq
->end_io_data
)
105 return &((struct dm_rq_target_io
*)rq
->end_io_data
)->info
;
108 EXPORT_SYMBOL_GPL(dm_get_rq_mapinfo
);
110 #define MINOR_ALLOCED ((void *)-1)
113 * Bits for the md->flags field.
115 #define DMF_BLOCK_IO_FOR_SUSPEND 0
116 #define DMF_SUSPENDED 1
118 #define DMF_FREEING 3
119 #define DMF_DELETING 4
120 #define DMF_NOFLUSH_SUSPENDING 5
121 #define DMF_MERGE_IS_OPTIONAL 6
122 #define DMF_DEFERRED_REMOVE 7
123 #define DMF_SUSPENDED_INTERNALLY 8
126 * A dummy definition to make RCU happy.
127 * struct dm_table should never be dereferenced in this file.
134 * Work processed by per-device workqueue.
136 struct mapped_device
{
137 struct srcu_struct io_barrier
;
138 struct mutex suspend_lock
;
143 * The current mapping.
144 * Use dm_get_live_table{_fast} or take suspend_lock for
147 struct dm_table __rcu
*map
;
149 struct list_head table_devices
;
150 struct mutex table_devices_lock
;
154 struct request_queue
*queue
;
156 /* Protect queue and type against concurrent access. */
157 struct mutex type_lock
;
159 struct target_type
*immutable_target_type
;
161 struct gendisk
*disk
;
167 * A list of ios that arrived while we were suspended.
170 wait_queue_head_t wait
;
171 struct work_struct work
;
172 struct bio_list deferred
;
173 spinlock_t deferred_lock
;
176 * Processing queue (flush)
178 struct workqueue_struct
*wq
;
181 * io objects are allocated from here.
192 wait_queue_head_t eventq
;
194 struct list_head uevent_list
;
195 spinlock_t uevent_lock
; /* Protect access to uevent_list */
198 * freeze/thaw support require holding onto a super block
200 struct super_block
*frozen_sb
;
201 struct block_device
*bdev
;
203 /* forced geometry settings */
204 struct hd_geometry geometry
;
206 /* kobject and completion */
207 struct dm_kobject_holder kobj_holder
;
209 /* zero-length flush that will be cloned and submitted to targets */
210 struct bio flush_bio
;
212 /* the number of internal suspends */
213 unsigned internal_suspend_count
;
215 struct dm_stats stats
;
217 struct kthread_worker kworker
;
218 struct task_struct
*kworker_task
;
222 * For mempools pre-allocation at the table loading time.
224 struct dm_md_mempools
{
230 struct table_device
{
231 struct list_head list
;
233 struct dm_dev dm_dev
;
236 #define RESERVED_BIO_BASED_IOS 16
237 #define RESERVED_REQUEST_BASED_IOS 256
238 #define RESERVED_MAX_IOS 1024
239 static struct kmem_cache
*_io_cache
;
240 static struct kmem_cache
*_rq_tio_cache
;
241 static struct kmem_cache
*_rq_cache
;
244 * Bio-based DM's mempools' reserved IOs set by the user.
246 static unsigned reserved_bio_based_ios
= RESERVED_BIO_BASED_IOS
;
249 * Request-based DM's mempools' reserved IOs set by the user.
251 static unsigned reserved_rq_based_ios
= RESERVED_REQUEST_BASED_IOS
;
253 static unsigned __dm_get_reserved_ios(unsigned *reserved_ios
,
254 unsigned def
, unsigned max
)
256 unsigned ios
= ACCESS_ONCE(*reserved_ios
);
257 unsigned modified_ios
= 0;
265 (void)cmpxchg(reserved_ios
, ios
, modified_ios
);
272 unsigned dm_get_reserved_bio_based_ios(void)
274 return __dm_get_reserved_ios(&reserved_bio_based_ios
,
275 RESERVED_BIO_BASED_IOS
, RESERVED_MAX_IOS
);
277 EXPORT_SYMBOL_GPL(dm_get_reserved_bio_based_ios
);
279 unsigned dm_get_reserved_rq_based_ios(void)
281 return __dm_get_reserved_ios(&reserved_rq_based_ios
,
282 RESERVED_REQUEST_BASED_IOS
, RESERVED_MAX_IOS
);
284 EXPORT_SYMBOL_GPL(dm_get_reserved_rq_based_ios
);
286 static int __init
local_init(void)
290 /* allocate a slab for the dm_ios */
291 _io_cache
= KMEM_CACHE(dm_io
, 0);
295 _rq_tio_cache
= KMEM_CACHE(dm_rq_target_io
, 0);
297 goto out_free_io_cache
;
299 _rq_cache
= kmem_cache_create("dm_clone_request", sizeof(struct request
),
300 __alignof__(struct request
), 0, NULL
);
302 goto out_free_rq_tio_cache
;
304 r
= dm_uevent_init();
306 goto out_free_rq_cache
;
308 deferred_remove_workqueue
= alloc_workqueue("kdmremove", WQ_UNBOUND
, 1);
309 if (!deferred_remove_workqueue
) {
311 goto out_uevent_exit
;
315 r
= register_blkdev(_major
, _name
);
317 goto out_free_workqueue
;
325 destroy_workqueue(deferred_remove_workqueue
);
329 kmem_cache_destroy(_rq_cache
);
330 out_free_rq_tio_cache
:
331 kmem_cache_destroy(_rq_tio_cache
);
333 kmem_cache_destroy(_io_cache
);
338 static void local_exit(void)
340 flush_scheduled_work();
341 destroy_workqueue(deferred_remove_workqueue
);
343 kmem_cache_destroy(_rq_cache
);
344 kmem_cache_destroy(_rq_tio_cache
);
345 kmem_cache_destroy(_io_cache
);
346 unregister_blkdev(_major
, _name
);
351 DMINFO("cleaned up");
354 static int (*_inits
[])(void) __initdata
= {
365 static void (*_exits
[])(void) = {
376 static int __init
dm_init(void)
378 const int count
= ARRAY_SIZE(_inits
);
382 for (i
= 0; i
< count
; i
++) {
397 static void __exit
dm_exit(void)
399 int i
= ARRAY_SIZE(_exits
);
405 * Should be empty by this point.
407 idr_destroy(&_minor_idr
);
411 * Block device functions
413 int dm_deleting_md(struct mapped_device
*md
)
415 return test_bit(DMF_DELETING
, &md
->flags
);
418 static int dm_blk_open(struct block_device
*bdev
, fmode_t mode
)
420 struct mapped_device
*md
;
422 spin_lock(&_minor_lock
);
424 md
= bdev
->bd_disk
->private_data
;
428 if (test_bit(DMF_FREEING
, &md
->flags
) ||
429 dm_deleting_md(md
)) {
435 atomic_inc(&md
->open_count
);
438 spin_unlock(&_minor_lock
);
440 return md
? 0 : -ENXIO
;
443 static void dm_blk_close(struct gendisk
*disk
, fmode_t mode
)
445 struct mapped_device
*md
= disk
->private_data
;
447 spin_lock(&_minor_lock
);
449 if (atomic_dec_and_test(&md
->open_count
) &&
450 (test_bit(DMF_DEFERRED_REMOVE
, &md
->flags
)))
451 queue_work(deferred_remove_workqueue
, &deferred_remove_work
);
455 spin_unlock(&_minor_lock
);
458 int dm_open_count(struct mapped_device
*md
)
460 return atomic_read(&md
->open_count
);
464 * Guarantees nothing is using the device before it's deleted.
466 int dm_lock_for_deletion(struct mapped_device
*md
, bool mark_deferred
, bool only_deferred
)
470 spin_lock(&_minor_lock
);
472 if (dm_open_count(md
)) {
475 set_bit(DMF_DEFERRED_REMOVE
, &md
->flags
);
476 } else if (only_deferred
&& !test_bit(DMF_DEFERRED_REMOVE
, &md
->flags
))
479 set_bit(DMF_DELETING
, &md
->flags
);
481 spin_unlock(&_minor_lock
);
486 int dm_cancel_deferred_remove(struct mapped_device
*md
)
490 spin_lock(&_minor_lock
);
492 if (test_bit(DMF_DELETING
, &md
->flags
))
495 clear_bit(DMF_DEFERRED_REMOVE
, &md
->flags
);
497 spin_unlock(&_minor_lock
);
502 static void do_deferred_remove(struct work_struct
*w
)
504 dm_deferred_remove();
507 sector_t
dm_get_size(struct mapped_device
*md
)
509 return get_capacity(md
->disk
);
512 struct request_queue
*dm_get_md_queue(struct mapped_device
*md
)
517 struct dm_stats
*dm_get_stats(struct mapped_device
*md
)
522 static int dm_blk_getgeo(struct block_device
*bdev
, struct hd_geometry
*geo
)
524 struct mapped_device
*md
= bdev
->bd_disk
->private_data
;
526 return dm_get_geometry(md
, geo
);
529 static int dm_blk_ioctl(struct block_device
*bdev
, fmode_t mode
,
530 unsigned int cmd
, unsigned long arg
)
532 struct mapped_device
*md
= bdev
->bd_disk
->private_data
;
534 struct dm_table
*map
;
535 struct dm_target
*tgt
;
539 map
= dm_get_live_table(md
, &srcu_idx
);
541 if (!map
|| !dm_table_get_size(map
))
544 /* We only support devices that have a single target */
545 if (dm_table_get_num_targets(map
) != 1)
548 tgt
= dm_table_get_target(map
, 0);
549 if (!tgt
->type
->ioctl
)
552 if (dm_suspended_md(md
)) {
557 r
= tgt
->type
->ioctl(tgt
, cmd
, arg
);
560 dm_put_live_table(md
, srcu_idx
);
562 if (r
== -ENOTCONN
) {
570 static struct dm_io
*alloc_io(struct mapped_device
*md
)
572 return mempool_alloc(md
->io_pool
, GFP_NOIO
);
575 static void free_io(struct mapped_device
*md
, struct dm_io
*io
)
577 mempool_free(io
, md
->io_pool
);
580 static void free_tio(struct mapped_device
*md
, struct dm_target_io
*tio
)
582 bio_put(&tio
->clone
);
585 static struct dm_rq_target_io
*alloc_rq_tio(struct mapped_device
*md
,
588 return mempool_alloc(md
->io_pool
, gfp_mask
);
591 static void free_rq_tio(struct dm_rq_target_io
*tio
)
593 mempool_free(tio
, tio
->md
->io_pool
);
596 static struct request
*alloc_clone_request(struct mapped_device
*md
,
599 return mempool_alloc(md
->rq_pool
, gfp_mask
);
602 static void free_clone_request(struct mapped_device
*md
, struct request
*rq
)
604 mempool_free(rq
, md
->rq_pool
);
607 static int md_in_flight(struct mapped_device
*md
)
609 return atomic_read(&md
->pending
[READ
]) +
610 atomic_read(&md
->pending
[WRITE
]);
613 static void start_io_acct(struct dm_io
*io
)
615 struct mapped_device
*md
= io
->md
;
616 struct bio
*bio
= io
->bio
;
618 int rw
= bio_data_dir(bio
);
620 io
->start_time
= jiffies
;
622 cpu
= part_stat_lock();
623 part_round_stats(cpu
, &dm_disk(md
)->part0
);
625 atomic_set(&dm_disk(md
)->part0
.in_flight
[rw
],
626 atomic_inc_return(&md
->pending
[rw
]));
628 if (unlikely(dm_stats_used(&md
->stats
)))
629 dm_stats_account_io(&md
->stats
, bio
->bi_rw
, bio
->bi_iter
.bi_sector
,
630 bio_sectors(bio
), false, 0, &io
->stats_aux
);
633 static void end_io_acct(struct dm_io
*io
)
635 struct mapped_device
*md
= io
->md
;
636 struct bio
*bio
= io
->bio
;
637 unsigned long duration
= jiffies
- io
->start_time
;
639 int rw
= bio_data_dir(bio
);
641 generic_end_io_acct(rw
, &dm_disk(md
)->part0
, io
->start_time
);
643 if (unlikely(dm_stats_used(&md
->stats
)))
644 dm_stats_account_io(&md
->stats
, bio
->bi_rw
, bio
->bi_iter
.bi_sector
,
645 bio_sectors(bio
), true, duration
, &io
->stats_aux
);
648 * After this is decremented the bio must not be touched if it is
651 pending
= atomic_dec_return(&md
->pending
[rw
]);
652 atomic_set(&dm_disk(md
)->part0
.in_flight
[rw
], pending
);
653 pending
+= atomic_read(&md
->pending
[rw
^0x1]);
655 /* nudge anyone waiting on suspend queue */
661 * Add the bio to the list of deferred io.
663 static void queue_io(struct mapped_device
*md
, struct bio
*bio
)
667 spin_lock_irqsave(&md
->deferred_lock
, flags
);
668 bio_list_add(&md
->deferred
, bio
);
669 spin_unlock_irqrestore(&md
->deferred_lock
, flags
);
670 queue_work(md
->wq
, &md
->work
);
674 * Everyone (including functions in this file), should use this
675 * function to access the md->map field, and make sure they call
676 * dm_put_live_table() when finished.
678 struct dm_table
*dm_get_live_table(struct mapped_device
*md
, int *srcu_idx
) __acquires(md
->io_barrier
)
680 *srcu_idx
= srcu_read_lock(&md
->io_barrier
);
682 return srcu_dereference(md
->map
, &md
->io_barrier
);
685 void dm_put_live_table(struct mapped_device
*md
, int srcu_idx
) __releases(md
->io_barrier
)
687 srcu_read_unlock(&md
->io_barrier
, srcu_idx
);
690 void dm_sync_table(struct mapped_device
*md
)
692 synchronize_srcu(&md
->io_barrier
);
693 synchronize_rcu_expedited();
697 * A fast alternative to dm_get_live_table/dm_put_live_table.
698 * The caller must not block between these two functions.
700 static struct dm_table
*dm_get_live_table_fast(struct mapped_device
*md
) __acquires(RCU
)
703 return rcu_dereference(md
->map
);
706 static void dm_put_live_table_fast(struct mapped_device
*md
) __releases(RCU
)
712 * Open a table device so we can use it as a map destination.
714 static int open_table_device(struct table_device
*td
, dev_t dev
,
715 struct mapped_device
*md
)
717 static char *_claim_ptr
= "I belong to device-mapper";
718 struct block_device
*bdev
;
722 BUG_ON(td
->dm_dev
.bdev
);
724 bdev
= blkdev_get_by_dev(dev
, td
->dm_dev
.mode
| FMODE_EXCL
, _claim_ptr
);
726 return PTR_ERR(bdev
);
728 r
= bd_link_disk_holder(bdev
, dm_disk(md
));
730 blkdev_put(bdev
, td
->dm_dev
.mode
| FMODE_EXCL
);
734 td
->dm_dev
.bdev
= bdev
;
739 * Close a table device that we've been using.
741 static void close_table_device(struct table_device
*td
, struct mapped_device
*md
)
743 if (!td
->dm_dev
.bdev
)
746 bd_unlink_disk_holder(td
->dm_dev
.bdev
, dm_disk(md
));
747 blkdev_put(td
->dm_dev
.bdev
, td
->dm_dev
.mode
| FMODE_EXCL
);
748 td
->dm_dev
.bdev
= NULL
;
751 static struct table_device
*find_table_device(struct list_head
*l
, dev_t dev
,
753 struct table_device
*td
;
755 list_for_each_entry(td
, l
, list
)
756 if (td
->dm_dev
.bdev
->bd_dev
== dev
&& td
->dm_dev
.mode
== mode
)
762 int dm_get_table_device(struct mapped_device
*md
, dev_t dev
, fmode_t mode
,
763 struct dm_dev
**result
) {
765 struct table_device
*td
;
767 mutex_lock(&md
->table_devices_lock
);
768 td
= find_table_device(&md
->table_devices
, dev
, mode
);
770 td
= kmalloc(sizeof(*td
), GFP_KERNEL
);
772 mutex_unlock(&md
->table_devices_lock
);
776 td
->dm_dev
.mode
= mode
;
777 td
->dm_dev
.bdev
= NULL
;
779 if ((r
= open_table_device(td
, dev
, md
))) {
780 mutex_unlock(&md
->table_devices_lock
);
785 format_dev_t(td
->dm_dev
.name
, dev
);
787 atomic_set(&td
->count
, 0);
788 list_add(&td
->list
, &md
->table_devices
);
790 atomic_inc(&td
->count
);
791 mutex_unlock(&md
->table_devices_lock
);
793 *result
= &td
->dm_dev
;
796 EXPORT_SYMBOL_GPL(dm_get_table_device
);
798 void dm_put_table_device(struct mapped_device
*md
, struct dm_dev
*d
)
800 struct table_device
*td
= container_of(d
, struct table_device
, dm_dev
);
802 mutex_lock(&md
->table_devices_lock
);
803 if (atomic_dec_and_test(&td
->count
)) {
804 close_table_device(td
, md
);
808 mutex_unlock(&md
->table_devices_lock
);
810 EXPORT_SYMBOL(dm_put_table_device
);
812 static void free_table_devices(struct list_head
*devices
)
814 struct list_head
*tmp
, *next
;
816 list_for_each_safe(tmp
, next
, devices
) {
817 struct table_device
*td
= list_entry(tmp
, struct table_device
, list
);
819 DMWARN("dm_destroy: %s still exists with %d references",
820 td
->dm_dev
.name
, atomic_read(&td
->count
));
826 * Get the geometry associated with a dm device
828 int dm_get_geometry(struct mapped_device
*md
, struct hd_geometry
*geo
)
836 * Set the geometry of a device.
838 int dm_set_geometry(struct mapped_device
*md
, struct hd_geometry
*geo
)
840 sector_t sz
= (sector_t
)geo
->cylinders
* geo
->heads
* geo
->sectors
;
842 if (geo
->start
> sz
) {
843 DMWARN("Start sector is beyond the geometry limits.");
852 /*-----------------------------------------------------------------
854 * A more elegant soln is in the works that uses the queue
855 * merge fn, unfortunately there are a couple of changes to
856 * the block layer that I want to make for this. So in the
857 * interests of getting something for people to use I give
858 * you this clearly demarcated crap.
859 *---------------------------------------------------------------*/
861 static int __noflush_suspending(struct mapped_device
*md
)
863 return test_bit(DMF_NOFLUSH_SUSPENDING
, &md
->flags
);
867 * Decrements the number of outstanding ios that a bio has been
868 * cloned into, completing the original io if necc.
870 static void dec_pending(struct dm_io
*io
, int error
)
875 struct mapped_device
*md
= io
->md
;
877 /* Push-back supersedes any I/O errors */
878 if (unlikely(error
)) {
879 spin_lock_irqsave(&io
->endio_lock
, flags
);
880 if (!(io
->error
> 0 && __noflush_suspending(md
)))
882 spin_unlock_irqrestore(&io
->endio_lock
, flags
);
885 if (atomic_dec_and_test(&io
->io_count
)) {
886 if (io
->error
== DM_ENDIO_REQUEUE
) {
888 * Target requested pushing back the I/O.
890 spin_lock_irqsave(&md
->deferred_lock
, flags
);
891 if (__noflush_suspending(md
))
892 bio_list_add_head(&md
->deferred
, io
->bio
);
894 /* noflush suspend was interrupted. */
896 spin_unlock_irqrestore(&md
->deferred_lock
, flags
);
899 io_error
= io
->error
;
904 if (io_error
== DM_ENDIO_REQUEUE
)
907 if ((bio
->bi_rw
& REQ_FLUSH
) && bio
->bi_iter
.bi_size
) {
909 * Preflush done for flush with data, reissue
912 bio
->bi_rw
&= ~REQ_FLUSH
;
915 /* done with normal IO or empty flush */
916 trace_block_bio_complete(md
->queue
, bio
, io_error
);
917 bio_endio(bio
, io_error
);
922 static void disable_write_same(struct mapped_device
*md
)
924 struct queue_limits
*limits
= dm_get_queue_limits(md
);
926 /* device doesn't really support WRITE SAME, disable it */
927 limits
->max_write_same_sectors
= 0;
930 static void clone_endio(struct bio
*bio
, int error
)
933 struct dm_target_io
*tio
= container_of(bio
, struct dm_target_io
, clone
);
934 struct dm_io
*io
= tio
->io
;
935 struct mapped_device
*md
= tio
->io
->md
;
936 dm_endio_fn endio
= tio
->ti
->type
->end_io
;
938 if (!bio_flagged(bio
, BIO_UPTODATE
) && !error
)
942 r
= endio(tio
->ti
, bio
, error
);
943 if (r
< 0 || r
== DM_ENDIO_REQUEUE
)
945 * error and requeue request are handled
949 else if (r
== DM_ENDIO_INCOMPLETE
)
950 /* The target will handle the io */
953 DMWARN("unimplemented target endio return value: %d", r
);
958 if (unlikely(r
== -EREMOTEIO
&& (bio
->bi_rw
& REQ_WRITE_SAME
) &&
959 !bdev_get_queue(bio
->bi_bdev
)->limits
.max_write_same_sectors
))
960 disable_write_same(md
);
963 dec_pending(io
, error
);
967 * Partial completion handling for request-based dm
969 static void end_clone_bio(struct bio
*clone
, int error
)
971 struct dm_rq_clone_bio_info
*info
=
972 container_of(clone
, struct dm_rq_clone_bio_info
, clone
);
973 struct dm_rq_target_io
*tio
= info
->tio
;
974 struct bio
*bio
= info
->orig
;
975 unsigned int nr_bytes
= info
->orig
->bi_iter
.bi_size
;
981 * An error has already been detected on the request.
982 * Once error occurred, just let clone->end_io() handle
988 * Don't notice the error to the upper layer yet.
989 * The error handling decision is made by the target driver,
990 * when the request is completed.
997 * I/O for the bio successfully completed.
998 * Notice the data completion to the upper layer.
1002 * bios are processed from the head of the list.
1003 * So the completing bio should always be rq->bio.
1004 * If it's not, something wrong is happening.
1006 if (tio
->orig
->bio
!= bio
)
1007 DMERR("bio completion is going in the middle of the request");
1010 * Update the original request.
1011 * Do not use blk_end_request() here, because it may complete
1012 * the original request before the clone, and break the ordering.
1014 blk_update_request(tio
->orig
, 0, nr_bytes
);
1018 * Don't touch any member of the md after calling this function because
1019 * the md may be freed in dm_put() at the end of this function.
1020 * Or do dm_get() before calling this function and dm_put() later.
1022 static void rq_completed(struct mapped_device
*md
, int rw
, bool run_queue
)
1024 atomic_dec(&md
->pending
[rw
]);
1026 /* nudge anyone waiting on suspend queue */
1027 if (!md_in_flight(md
))
1031 * Run this off this callpath, as drivers could invoke end_io while
1032 * inside their request_fn (and holding the queue lock). Calling
1033 * back into ->request_fn() could deadlock attempting to grab the
1037 blk_run_queue_async(md
->queue
);
1040 * dm_put() must be at the end of this function. See the comment above
1045 static void free_rq_clone(struct request
*clone
)
1047 struct dm_rq_target_io
*tio
= clone
->end_io_data
;
1049 blk_rq_unprep_clone(clone
);
1050 if (clone
->q
&& clone
->q
->mq_ops
)
1051 tio
->ti
->type
->release_clone_rq(clone
);
1053 free_clone_request(tio
->md
, clone
);
1058 * Complete the clone and the original request.
1059 * Must be called without clone's queue lock held,
1060 * see end_clone_request() for more details.
1062 static void dm_end_request(struct request
*clone
, int error
)
1064 int rw
= rq_data_dir(clone
);
1065 struct dm_rq_target_io
*tio
= clone
->end_io_data
;
1066 struct mapped_device
*md
= tio
->md
;
1067 struct request
*rq
= tio
->orig
;
1069 if (rq
->cmd_type
== REQ_TYPE_BLOCK_PC
) {
1070 rq
->errors
= clone
->errors
;
1071 rq
->resid_len
= clone
->resid_len
;
1075 * We are using the sense buffer of the original
1077 * So setting the length of the sense data is enough.
1079 rq
->sense_len
= clone
->sense_len
;
1082 free_rq_clone(clone
);
1083 blk_end_request_all(rq
, error
);
1084 rq_completed(md
, rw
, true);
1087 static void dm_unprep_request(struct request
*rq
)
1089 struct dm_rq_target_io
*tio
= rq
->special
;
1090 struct request
*clone
= tio
->clone
;
1093 rq
->cmd_flags
&= ~REQ_DONTPREP
;
1096 free_rq_clone(clone
);
1100 * Requeue the original request of a clone.
1102 static void dm_requeue_unmapped_original_request(struct mapped_device
*md
,
1105 int rw
= rq_data_dir(rq
);
1106 struct request_queue
*q
= rq
->q
;
1107 unsigned long flags
;
1109 dm_unprep_request(rq
);
1111 spin_lock_irqsave(q
->queue_lock
, flags
);
1112 blk_requeue_request(q
, rq
);
1113 spin_unlock_irqrestore(q
->queue_lock
, flags
);
1115 rq_completed(md
, rw
, false);
1118 static void dm_requeue_unmapped_request(struct request
*clone
)
1120 struct dm_rq_target_io
*tio
= clone
->end_io_data
;
1122 dm_requeue_unmapped_original_request(tio
->md
, tio
->orig
);
1125 static void __stop_queue(struct request_queue
*q
)
1130 static void stop_queue(struct request_queue
*q
)
1132 unsigned long flags
;
1134 spin_lock_irqsave(q
->queue_lock
, flags
);
1136 spin_unlock_irqrestore(q
->queue_lock
, flags
);
1139 static void __start_queue(struct request_queue
*q
)
1141 if (blk_queue_stopped(q
))
1145 static void start_queue(struct request_queue
*q
)
1147 unsigned long flags
;
1149 spin_lock_irqsave(q
->queue_lock
, flags
);
1151 spin_unlock_irqrestore(q
->queue_lock
, flags
);
1154 static void dm_done(struct request
*clone
, int error
, bool mapped
)
1157 struct dm_rq_target_io
*tio
= clone
->end_io_data
;
1158 dm_request_endio_fn rq_end_io
= NULL
;
1161 rq_end_io
= tio
->ti
->type
->rq_end_io
;
1163 if (mapped
&& rq_end_io
)
1164 r
= rq_end_io(tio
->ti
, clone
, error
, &tio
->info
);
1167 if (unlikely(r
== -EREMOTEIO
&& (clone
->cmd_flags
& REQ_WRITE_SAME
) &&
1168 !clone
->q
->limits
.max_write_same_sectors
))
1169 disable_write_same(tio
->md
);
1172 /* The target wants to complete the I/O */
1173 dm_end_request(clone
, r
);
1174 else if (r
== DM_ENDIO_INCOMPLETE
)
1175 /* The target will handle the I/O */
1177 else if (r
== DM_ENDIO_REQUEUE
)
1178 /* The target wants to requeue the I/O */
1179 dm_requeue_unmapped_request(clone
);
1181 DMWARN("unimplemented target endio return value: %d", r
);
1187 * Request completion handler for request-based dm
1189 static void dm_softirq_done(struct request
*rq
)
1192 struct dm_rq_target_io
*tio
= rq
->special
;
1193 struct request
*clone
= tio
->clone
;
1196 blk_end_request_all(rq
, tio
->error
);
1197 rq_completed(tio
->md
, rq_data_dir(rq
), false);
1202 if (rq
->cmd_flags
& REQ_FAILED
)
1205 dm_done(clone
, tio
->error
, mapped
);
1209 * Complete the clone and the original request with the error status
1210 * through softirq context.
1212 static void dm_complete_request(struct request
*rq
, int error
)
1214 struct dm_rq_target_io
*tio
= rq
->special
;
1217 blk_complete_request(rq
);
1221 * Complete the not-mapped clone and the original request with the error status
1222 * through softirq context.
1223 * Target's rq_end_io() function isn't called.
1224 * This may be used when the target's map_rq() or clone_and_map_rq() functions fail.
1226 static void dm_kill_unmapped_request(struct request
*rq
, int error
)
1228 rq
->cmd_flags
|= REQ_FAILED
;
1229 dm_complete_request(rq
, error
);
1233 * Called with the clone's queue lock held
1235 static void end_clone_request(struct request
*clone
, int error
)
1237 struct dm_rq_target_io
*tio
= clone
->end_io_data
;
1239 if (!clone
->q
->mq_ops
) {
1241 * For just cleaning up the information of the queue in which
1242 * the clone was dispatched.
1243 * The clone is *NOT* freed actually here because it is alloced
1244 * from dm own mempool (REQ_ALLOCED isn't set).
1246 __blk_put_request(clone
->q
, clone
);
1250 * Actual request completion is done in a softirq context which doesn't
1251 * hold the clone's queue lock. Otherwise, deadlock could occur because:
1252 * - another request may be submitted by the upper level driver
1253 * of the stacking during the completion
1254 * - the submission which requires queue lock may be done
1255 * against this clone's queue
1257 dm_complete_request(tio
->orig
, error
);
1261 * Return maximum size of I/O possible at the supplied sector up to the current
1264 static sector_t
max_io_len_target_boundary(sector_t sector
, struct dm_target
*ti
)
1266 sector_t target_offset
= dm_target_offset(ti
, sector
);
1268 return ti
->len
- target_offset
;
1271 static sector_t
max_io_len(sector_t sector
, struct dm_target
*ti
)
1273 sector_t len
= max_io_len_target_boundary(sector
, ti
);
1274 sector_t offset
, max_len
;
1277 * Does the target need to split even further?
1279 if (ti
->max_io_len
) {
1280 offset
= dm_target_offset(ti
, sector
);
1281 if (unlikely(ti
->max_io_len
& (ti
->max_io_len
- 1)))
1282 max_len
= sector_div(offset
, ti
->max_io_len
);
1284 max_len
= offset
& (ti
->max_io_len
- 1);
1285 max_len
= ti
->max_io_len
- max_len
;
1294 int dm_set_target_max_io_len(struct dm_target
*ti
, sector_t len
)
1296 if (len
> UINT_MAX
) {
1297 DMERR("Specified maximum size of target IO (%llu) exceeds limit (%u)",
1298 (unsigned long long)len
, UINT_MAX
);
1299 ti
->error
= "Maximum size of target IO is too large";
1303 ti
->max_io_len
= (uint32_t) len
;
1307 EXPORT_SYMBOL_GPL(dm_set_target_max_io_len
);
1310 * A target may call dm_accept_partial_bio only from the map routine. It is
1311 * allowed for all bio types except REQ_FLUSH.
1313 * dm_accept_partial_bio informs the dm that the target only wants to process
1314 * additional n_sectors sectors of the bio and the rest of the data should be
1315 * sent in a next bio.
1317 * A diagram that explains the arithmetics:
1318 * +--------------------+---------------+-------+
1320 * +--------------------+---------------+-------+
1322 * <-------------- *tio->len_ptr --------------->
1323 * <------- bi_size ------->
1326 * Region 1 was already iterated over with bio_advance or similar function.
1327 * (it may be empty if the target doesn't use bio_advance)
1328 * Region 2 is the remaining bio size that the target wants to process.
1329 * (it may be empty if region 1 is non-empty, although there is no reason
1331 * The target requires that region 3 is to be sent in the next bio.
1333 * If the target wants to receive multiple copies of the bio (via num_*bios, etc),
1334 * the partially processed part (the sum of regions 1+2) must be the same for all
1335 * copies of the bio.
1337 void dm_accept_partial_bio(struct bio
*bio
, unsigned n_sectors
)
1339 struct dm_target_io
*tio
= container_of(bio
, struct dm_target_io
, clone
);
1340 unsigned bi_size
= bio
->bi_iter
.bi_size
>> SECTOR_SHIFT
;
1341 BUG_ON(bio
->bi_rw
& REQ_FLUSH
);
1342 BUG_ON(bi_size
> *tio
->len_ptr
);
1343 BUG_ON(n_sectors
> bi_size
);
1344 *tio
->len_ptr
-= bi_size
- n_sectors
;
1345 bio
->bi_iter
.bi_size
= n_sectors
<< SECTOR_SHIFT
;
1347 EXPORT_SYMBOL_GPL(dm_accept_partial_bio
);
1349 static void __map_bio(struct dm_target_io
*tio
)
1353 struct mapped_device
*md
;
1354 struct bio
*clone
= &tio
->clone
;
1355 struct dm_target
*ti
= tio
->ti
;
1357 clone
->bi_end_io
= clone_endio
;
1360 * Map the clone. If r == 0 we don't need to do
1361 * anything, the target has assumed ownership of
1364 atomic_inc(&tio
->io
->io_count
);
1365 sector
= clone
->bi_iter
.bi_sector
;
1366 r
= ti
->type
->map(ti
, clone
);
1367 if (r
== DM_MAPIO_REMAPPED
) {
1368 /* the bio has been remapped so dispatch it */
1370 trace_block_bio_remap(bdev_get_queue(clone
->bi_bdev
), clone
,
1371 tio
->io
->bio
->bi_bdev
->bd_dev
, sector
);
1373 generic_make_request(clone
);
1374 } else if (r
< 0 || r
== DM_MAPIO_REQUEUE
) {
1375 /* error the io and bail out, or requeue it if needed */
1377 dec_pending(tio
->io
, r
);
1380 DMWARN("unimplemented target map return value: %d", r
);
1386 struct mapped_device
*md
;
1387 struct dm_table
*map
;
1391 unsigned sector_count
;
1394 static void bio_setup_sector(struct bio
*bio
, sector_t sector
, unsigned len
)
1396 bio
->bi_iter
.bi_sector
= sector
;
1397 bio
->bi_iter
.bi_size
= to_bytes(len
);
1401 * Creates a bio that consists of range of complete bvecs.
1403 static void clone_bio(struct dm_target_io
*tio
, struct bio
*bio
,
1404 sector_t sector
, unsigned len
)
1406 struct bio
*clone
= &tio
->clone
;
1408 __bio_clone_fast(clone
, bio
);
1410 if (bio_integrity(bio
))
1411 bio_integrity_clone(clone
, bio
, GFP_NOIO
);
1413 bio_advance(clone
, to_bytes(sector
- clone
->bi_iter
.bi_sector
));
1414 clone
->bi_iter
.bi_size
= to_bytes(len
);
1416 if (bio_integrity(bio
))
1417 bio_integrity_trim(clone
, 0, len
);
1420 static struct dm_target_io
*alloc_tio(struct clone_info
*ci
,
1421 struct dm_target
*ti
,
1422 unsigned target_bio_nr
)
1424 struct dm_target_io
*tio
;
1427 clone
= bio_alloc_bioset(GFP_NOIO
, 0, ci
->md
->bs
);
1428 tio
= container_of(clone
, struct dm_target_io
, clone
);
1432 tio
->target_bio_nr
= target_bio_nr
;
1437 static void __clone_and_map_simple_bio(struct clone_info
*ci
,
1438 struct dm_target
*ti
,
1439 unsigned target_bio_nr
, unsigned *len
)
1441 struct dm_target_io
*tio
= alloc_tio(ci
, ti
, target_bio_nr
);
1442 struct bio
*clone
= &tio
->clone
;
1446 __bio_clone_fast(clone
, ci
->bio
);
1448 bio_setup_sector(clone
, ci
->sector
, *len
);
1453 static void __send_duplicate_bios(struct clone_info
*ci
, struct dm_target
*ti
,
1454 unsigned num_bios
, unsigned *len
)
1456 unsigned target_bio_nr
;
1458 for (target_bio_nr
= 0; target_bio_nr
< num_bios
; target_bio_nr
++)
1459 __clone_and_map_simple_bio(ci
, ti
, target_bio_nr
, len
);
1462 static int __send_empty_flush(struct clone_info
*ci
)
1464 unsigned target_nr
= 0;
1465 struct dm_target
*ti
;
1467 BUG_ON(bio_has_data(ci
->bio
));
1468 while ((ti
= dm_table_get_target(ci
->map
, target_nr
++)))
1469 __send_duplicate_bios(ci
, ti
, ti
->num_flush_bios
, NULL
);
1474 static void __clone_and_map_data_bio(struct clone_info
*ci
, struct dm_target
*ti
,
1475 sector_t sector
, unsigned *len
)
1477 struct bio
*bio
= ci
->bio
;
1478 struct dm_target_io
*tio
;
1479 unsigned target_bio_nr
;
1480 unsigned num_target_bios
= 1;
1483 * Does the target want to receive duplicate copies of the bio?
1485 if (bio_data_dir(bio
) == WRITE
&& ti
->num_write_bios
)
1486 num_target_bios
= ti
->num_write_bios(ti
, bio
);
1488 for (target_bio_nr
= 0; target_bio_nr
< num_target_bios
; target_bio_nr
++) {
1489 tio
= alloc_tio(ci
, ti
, target_bio_nr
);
1491 clone_bio(tio
, bio
, sector
, *len
);
1496 typedef unsigned (*get_num_bios_fn
)(struct dm_target
*ti
);
1498 static unsigned get_num_discard_bios(struct dm_target
*ti
)
1500 return ti
->num_discard_bios
;
1503 static unsigned get_num_write_same_bios(struct dm_target
*ti
)
1505 return ti
->num_write_same_bios
;
1508 typedef bool (*is_split_required_fn
)(struct dm_target
*ti
);
1510 static bool is_split_required_for_discard(struct dm_target
*ti
)
1512 return ti
->split_discard_bios
;
1515 static int __send_changing_extent_only(struct clone_info
*ci
,
1516 get_num_bios_fn get_num_bios
,
1517 is_split_required_fn is_split_required
)
1519 struct dm_target
*ti
;
1524 ti
= dm_table_find_target(ci
->map
, ci
->sector
);
1525 if (!dm_target_is_valid(ti
))
1529 * Even though the device advertised support for this type of
1530 * request, that does not mean every target supports it, and
1531 * reconfiguration might also have changed that since the
1532 * check was performed.
1534 num_bios
= get_num_bios
? get_num_bios(ti
) : 0;
1538 if (is_split_required
&& !is_split_required(ti
))
1539 len
= min((sector_t
)ci
->sector_count
, max_io_len_target_boundary(ci
->sector
, ti
));
1541 len
= min((sector_t
)ci
->sector_count
, max_io_len(ci
->sector
, ti
));
1543 __send_duplicate_bios(ci
, ti
, num_bios
, &len
);
1546 } while (ci
->sector_count
-= len
);
1551 static int __send_discard(struct clone_info
*ci
)
1553 return __send_changing_extent_only(ci
, get_num_discard_bios
,
1554 is_split_required_for_discard
);
1557 static int __send_write_same(struct clone_info
*ci
)
1559 return __send_changing_extent_only(ci
, get_num_write_same_bios
, NULL
);
1563 * Select the correct strategy for processing a non-flush bio.
1565 static int __split_and_process_non_flush(struct clone_info
*ci
)
1567 struct bio
*bio
= ci
->bio
;
1568 struct dm_target
*ti
;
1571 if (unlikely(bio
->bi_rw
& REQ_DISCARD
))
1572 return __send_discard(ci
);
1573 else if (unlikely(bio
->bi_rw
& REQ_WRITE_SAME
))
1574 return __send_write_same(ci
);
1576 ti
= dm_table_find_target(ci
->map
, ci
->sector
);
1577 if (!dm_target_is_valid(ti
))
1580 len
= min_t(sector_t
, max_io_len(ci
->sector
, ti
), ci
->sector_count
);
1582 __clone_and_map_data_bio(ci
, ti
, ci
->sector
, &len
);
1585 ci
->sector_count
-= len
;
1591 * Entry point to split a bio into clones and submit them to the targets.
1593 static void __split_and_process_bio(struct mapped_device
*md
,
1594 struct dm_table
*map
, struct bio
*bio
)
1596 struct clone_info ci
;
1599 if (unlikely(!map
)) {
1606 ci
.io
= alloc_io(md
);
1608 atomic_set(&ci
.io
->io_count
, 1);
1611 spin_lock_init(&ci
.io
->endio_lock
);
1612 ci
.sector
= bio
->bi_iter
.bi_sector
;
1614 start_io_acct(ci
.io
);
1616 if (bio
->bi_rw
& REQ_FLUSH
) {
1617 ci
.bio
= &ci
.md
->flush_bio
;
1618 ci
.sector_count
= 0;
1619 error
= __send_empty_flush(&ci
);
1620 /* dec_pending submits any data associated with flush */
1623 ci
.sector_count
= bio_sectors(bio
);
1624 while (ci
.sector_count
&& !error
)
1625 error
= __split_and_process_non_flush(&ci
);
1628 /* drop the extra reference count */
1629 dec_pending(ci
.io
, error
);
1631 /*-----------------------------------------------------------------
1633 *---------------------------------------------------------------*/
1635 static int dm_merge_bvec(struct request_queue
*q
,
1636 struct bvec_merge_data
*bvm
,
1637 struct bio_vec
*biovec
)
1639 struct mapped_device
*md
= q
->queuedata
;
1640 struct dm_table
*map
= dm_get_live_table_fast(md
);
1641 struct dm_target
*ti
;
1642 sector_t max_sectors
;
1648 ti
= dm_table_find_target(map
, bvm
->bi_sector
);
1649 if (!dm_target_is_valid(ti
))
1653 * Find maximum amount of I/O that won't need splitting
1655 max_sectors
= min(max_io_len(bvm
->bi_sector
, ti
),
1656 (sector_t
) queue_max_sectors(q
));
1657 max_size
= (max_sectors
<< SECTOR_SHIFT
) - bvm
->bi_size
;
1658 if (unlikely(max_size
< 0)) /* this shouldn't _ever_ happen */
1662 * merge_bvec_fn() returns number of bytes
1663 * it can accept at this offset
1664 * max is precomputed maximal io size
1666 if (max_size
&& ti
->type
->merge
)
1667 max_size
= ti
->type
->merge(ti
, bvm
, biovec
, max_size
);
1669 * If the target doesn't support merge method and some of the devices
1670 * provided their merge_bvec method (we know this by looking for the
1671 * max_hw_sectors that dm_set_device_limits may set), then we can't
1672 * allow bios with multiple vector entries. So always set max_size
1673 * to 0, and the code below allows just one page.
1675 else if (queue_max_hw_sectors(q
) <= PAGE_SIZE
>> 9)
1679 dm_put_live_table_fast(md
);
1681 * Always allow an entire first page
1683 if (max_size
<= biovec
->bv_len
&& !(bvm
->bi_size
>> SECTOR_SHIFT
))
1684 max_size
= biovec
->bv_len
;
1690 * The request function that just remaps the bio built up by
1693 static void _dm_request(struct request_queue
*q
, struct bio
*bio
)
1695 int rw
= bio_data_dir(bio
);
1696 struct mapped_device
*md
= q
->queuedata
;
1698 struct dm_table
*map
;
1700 map
= dm_get_live_table(md
, &srcu_idx
);
1702 generic_start_io_acct(rw
, bio_sectors(bio
), &dm_disk(md
)->part0
);
1704 /* if we're suspended, we have to queue this io for later */
1705 if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND
, &md
->flags
))) {
1706 dm_put_live_table(md
, srcu_idx
);
1708 if (bio_rw(bio
) != READA
)
1715 __split_and_process_bio(md
, map
, bio
);
1716 dm_put_live_table(md
, srcu_idx
);
1720 int dm_request_based(struct mapped_device
*md
)
1722 return blk_queue_stackable(md
->queue
);
1725 static void dm_request(struct request_queue
*q
, struct bio
*bio
)
1727 struct mapped_device
*md
= q
->queuedata
;
1729 if (dm_request_based(md
))
1730 blk_queue_bio(q
, bio
);
1732 _dm_request(q
, bio
);
1735 static void dm_dispatch_clone_request(struct request
*clone
, struct request
*rq
)
1739 if (blk_queue_io_stat(clone
->q
))
1740 clone
->cmd_flags
|= REQ_IO_STAT
;
1742 clone
->start_time
= jiffies
;
1743 r
= blk_insert_cloned_request(clone
->q
, clone
);
1745 /* must complete clone in terms of original request */
1746 dm_complete_request(rq
, r
);
1749 static int dm_rq_bio_constructor(struct bio
*bio
, struct bio
*bio_orig
,
1752 struct dm_rq_target_io
*tio
= data
;
1753 struct dm_rq_clone_bio_info
*info
=
1754 container_of(bio
, struct dm_rq_clone_bio_info
, clone
);
1756 info
->orig
= bio_orig
;
1758 bio
->bi_end_io
= end_clone_bio
;
1763 static int setup_clone(struct request
*clone
, struct request
*rq
,
1764 struct dm_rq_target_io
*tio
, gfp_t gfp_mask
)
1768 r
= blk_rq_prep_clone(clone
, rq
, tio
->md
->bs
, gfp_mask
,
1769 dm_rq_bio_constructor
, tio
);
1773 clone
->cmd
= rq
->cmd
;
1774 clone
->cmd_len
= rq
->cmd_len
;
1775 clone
->sense
= rq
->sense
;
1776 clone
->end_io
= end_clone_request
;
1777 clone
->end_io_data
= tio
;
1784 static struct request
*clone_rq(struct request
*rq
, struct mapped_device
*md
,
1785 struct dm_rq_target_io
*tio
, gfp_t gfp_mask
)
1787 struct request
*clone
= alloc_clone_request(md
, gfp_mask
);
1792 blk_rq_init(NULL
, clone
);
1793 if (setup_clone(clone
, rq
, tio
, gfp_mask
)) {
1795 free_clone_request(md
, clone
);
1802 static void map_tio_request(struct kthread_work
*work
);
1804 static struct dm_rq_target_io
*prep_tio(struct request
*rq
,
1805 struct mapped_device
*md
, gfp_t gfp_mask
)
1807 struct dm_rq_target_io
*tio
;
1809 struct dm_table
*table
;
1811 tio
= alloc_rq_tio(md
, gfp_mask
);
1820 memset(&tio
->info
, 0, sizeof(tio
->info
));
1821 init_kthread_work(&tio
->work
, map_tio_request
);
1823 table
= dm_get_live_table(md
, &srcu_idx
);
1824 if (!dm_table_mq_request_based(table
)) {
1825 if (!clone_rq(rq
, md
, tio
, gfp_mask
)) {
1826 dm_put_live_table(md
, srcu_idx
);
1831 dm_put_live_table(md
, srcu_idx
);
1837 * Called with the queue lock held.
1839 static int dm_prep_fn(struct request_queue
*q
, struct request
*rq
)
1841 struct mapped_device
*md
= q
->queuedata
;
1842 struct dm_rq_target_io
*tio
;
1844 if (unlikely(rq
->special
)) {
1845 DMWARN("Already has something in rq->special.");
1846 return BLKPREP_KILL
;
1849 tio
= prep_tio(rq
, md
, GFP_ATOMIC
);
1851 return BLKPREP_DEFER
;
1854 rq
->cmd_flags
|= REQ_DONTPREP
;
1861 * 0 : the request has been processed
1862 * DM_MAPIO_REQUEUE : the original request needs to be requeued
1863 * < 0 : the request was completed due to failure
1865 static int map_request(struct dm_target
*ti
, struct request
*rq
,
1866 struct mapped_device
*md
)
1869 struct dm_rq_target_io
*tio
= rq
->special
;
1870 struct request
*clone
= NULL
;
1874 r
= ti
->type
->map_rq(ti
, clone
, &tio
->info
);
1876 r
= ti
->type
->clone_and_map_rq(ti
, rq
, &tio
->info
, &clone
);
1878 /* The target wants to complete the I/O */
1879 dm_kill_unmapped_request(rq
, r
);
1883 return DM_MAPIO_REQUEUE
;
1884 if (setup_clone(clone
, rq
, tio
, GFP_KERNEL
)) {
1886 ti
->type
->release_clone_rq(clone
);
1887 return DM_MAPIO_REQUEUE
;
1892 case DM_MAPIO_SUBMITTED
:
1893 /* The target has taken the I/O to submit by itself later */
1895 case DM_MAPIO_REMAPPED
:
1896 /* The target has remapped the I/O so dispatch it */
1897 trace_block_rq_remap(clone
->q
, clone
, disk_devt(dm_disk(md
)),
1899 dm_dispatch_clone_request(clone
, rq
);
1901 case DM_MAPIO_REQUEUE
:
1902 /* The target wants to requeue the I/O */
1903 dm_requeue_unmapped_request(clone
);
1907 DMWARN("unimplemented target map return value: %d", r
);
1911 /* The target wants to complete the I/O */
1912 dm_kill_unmapped_request(rq
, r
);
1919 static void map_tio_request(struct kthread_work
*work
)
1921 struct dm_rq_target_io
*tio
= container_of(work
, struct dm_rq_target_io
, work
);
1922 struct request
*rq
= tio
->orig
;
1923 struct mapped_device
*md
= tio
->md
;
1925 if (map_request(tio
->ti
, rq
, md
) == DM_MAPIO_REQUEUE
)
1926 dm_requeue_unmapped_original_request(md
, rq
);
1929 static void dm_start_request(struct mapped_device
*md
, struct request
*orig
)
1931 blk_start_request(orig
);
1932 atomic_inc(&md
->pending
[rq_data_dir(orig
)]);
1935 * Hold the md reference here for the in-flight I/O.
1936 * We can't rely on the reference count by device opener,
1937 * because the device may be closed during the request completion
1938 * when all bios are completed.
1939 * See the comment in rq_completed() too.
1945 * q->request_fn for request-based dm.
1946 * Called with the queue lock held.
1948 static void dm_request_fn(struct request_queue
*q
)
1950 struct mapped_device
*md
= q
->queuedata
;
1952 struct dm_table
*map
= dm_get_live_table(md
, &srcu_idx
);
1953 struct dm_target
*ti
;
1955 struct dm_rq_target_io
*tio
;
1959 * For suspend, check blk_queue_stopped() and increment
1960 * ->pending within a single queue_lock not to increment the
1961 * number of in-flight I/Os after the queue is stopped in
1964 while (!blk_queue_stopped(q
)) {
1965 rq
= blk_peek_request(q
);
1969 /* always use block 0 to find the target for flushes for now */
1971 if (!(rq
->cmd_flags
& REQ_FLUSH
))
1972 pos
= blk_rq_pos(rq
);
1974 ti
= dm_table_find_target(map
, pos
);
1975 if (!dm_target_is_valid(ti
)) {
1977 * Must perform setup, that rq_completed() requires,
1978 * before calling dm_kill_unmapped_request
1980 DMERR_LIMIT("request attempted access beyond the end of device");
1981 dm_start_request(md
, rq
);
1982 dm_kill_unmapped_request(rq
, -EIO
);
1986 if (ti
->type
->busy
&& ti
->type
->busy(ti
))
1989 dm_start_request(md
, rq
);
1992 /* Establish tio->ti before queuing work (map_tio_request) */
1994 queue_kthread_work(&md
->kworker
, &tio
->work
);
1995 BUG_ON(!irqs_disabled());
2001 blk_delay_queue(q
, HZ
/ 10);
2003 dm_put_live_table(md
, srcu_idx
);
2006 int dm_underlying_device_busy(struct request_queue
*q
)
2008 return blk_lld_busy(q
);
2010 EXPORT_SYMBOL_GPL(dm_underlying_device_busy
);
2012 static int dm_lld_busy(struct request_queue
*q
)
2015 struct mapped_device
*md
= q
->queuedata
;
2016 struct dm_table
*map
= dm_get_live_table_fast(md
);
2018 if (!map
|| test_bit(DMF_BLOCK_IO_FOR_SUSPEND
, &md
->flags
))
2021 r
= dm_table_any_busy_target(map
);
2023 dm_put_live_table_fast(md
);
2028 static int dm_any_congested(void *congested_data
, int bdi_bits
)
2031 struct mapped_device
*md
= congested_data
;
2032 struct dm_table
*map
;
2034 if (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND
, &md
->flags
)) {
2035 map
= dm_get_live_table_fast(md
);
2038 * Request-based dm cares about only own queue for
2039 * the query about congestion status of request_queue
2041 if (dm_request_based(md
))
2042 r
= md
->queue
->backing_dev_info
.state
&
2045 r
= dm_table_any_congested(map
, bdi_bits
);
2047 dm_put_live_table_fast(md
);
2053 /*-----------------------------------------------------------------
2054 * An IDR is used to keep track of allocated minor numbers.
2055 *---------------------------------------------------------------*/
2056 static void free_minor(int minor
)
2058 spin_lock(&_minor_lock
);
2059 idr_remove(&_minor_idr
, minor
);
2060 spin_unlock(&_minor_lock
);
2064 * See if the device with a specific minor # is free.
2066 static int specific_minor(int minor
)
2070 if (minor
>= (1 << MINORBITS
))
2073 idr_preload(GFP_KERNEL
);
2074 spin_lock(&_minor_lock
);
2076 r
= idr_alloc(&_minor_idr
, MINOR_ALLOCED
, minor
, minor
+ 1, GFP_NOWAIT
);
2078 spin_unlock(&_minor_lock
);
2081 return r
== -ENOSPC
? -EBUSY
: r
;
2085 static int next_free_minor(int *minor
)
2089 idr_preload(GFP_KERNEL
);
2090 spin_lock(&_minor_lock
);
2092 r
= idr_alloc(&_minor_idr
, MINOR_ALLOCED
, 0, 1 << MINORBITS
, GFP_NOWAIT
);
2094 spin_unlock(&_minor_lock
);
2102 static const struct block_device_operations dm_blk_dops
;
2104 static void dm_wq_work(struct work_struct
*work
);
2106 static void dm_init_md_queue(struct mapped_device
*md
)
2109 * Request-based dm devices cannot be stacked on top of bio-based dm
2110 * devices. The type of this dm device has not been decided yet.
2111 * The type is decided at the first table loading time.
2112 * To prevent problematic device stacking, clear the queue flag
2113 * for request stacking support until then.
2115 * This queue is new, so no concurrency on the queue_flags.
2117 queue_flag_clear_unlocked(QUEUE_FLAG_STACKABLE
, md
->queue
);
2119 md
->queue
->queuedata
= md
;
2120 md
->queue
->backing_dev_info
.congested_fn
= dm_any_congested
;
2121 md
->queue
->backing_dev_info
.congested_data
= md
;
2122 blk_queue_make_request(md
->queue
, dm_request
);
2123 blk_queue_bounce_limit(md
->queue
, BLK_BOUNCE_ANY
);
2124 blk_queue_merge_bvec(md
->queue
, dm_merge_bvec
);
2128 * Allocate and initialise a blank device with a given minor.
2130 static struct mapped_device
*alloc_dev(int minor
)
2133 struct mapped_device
*md
= kzalloc(sizeof(*md
), GFP_KERNEL
);
2137 DMWARN("unable to allocate device, out of memory.");
2141 if (!try_module_get(THIS_MODULE
))
2142 goto bad_module_get
;
2144 /* get a minor number for the dev */
2145 if (minor
== DM_ANY_MINOR
)
2146 r
= next_free_minor(&minor
);
2148 r
= specific_minor(minor
);
2152 r
= init_srcu_struct(&md
->io_barrier
);
2154 goto bad_io_barrier
;
2156 md
->type
= DM_TYPE_NONE
;
2157 mutex_init(&md
->suspend_lock
);
2158 mutex_init(&md
->type_lock
);
2159 mutex_init(&md
->table_devices_lock
);
2160 spin_lock_init(&md
->deferred_lock
);
2161 atomic_set(&md
->holders
, 1);
2162 atomic_set(&md
->open_count
, 0);
2163 atomic_set(&md
->event_nr
, 0);
2164 atomic_set(&md
->uevent_seq
, 0);
2165 INIT_LIST_HEAD(&md
->uevent_list
);
2166 INIT_LIST_HEAD(&md
->table_devices
);
2167 spin_lock_init(&md
->uevent_lock
);
2169 md
->queue
= blk_alloc_queue(GFP_KERNEL
);
2173 dm_init_md_queue(md
);
2175 md
->disk
= alloc_disk(1);
2179 atomic_set(&md
->pending
[0], 0);
2180 atomic_set(&md
->pending
[1], 0);
2181 init_waitqueue_head(&md
->wait
);
2182 INIT_WORK(&md
->work
, dm_wq_work
);
2183 init_waitqueue_head(&md
->eventq
);
2184 init_completion(&md
->kobj_holder
.completion
);
2185 md
->kworker_task
= NULL
;
2187 md
->disk
->major
= _major
;
2188 md
->disk
->first_minor
= minor
;
2189 md
->disk
->fops
= &dm_blk_dops
;
2190 md
->disk
->queue
= md
->queue
;
2191 md
->disk
->private_data
= md
;
2192 sprintf(md
->disk
->disk_name
, "dm-%d", minor
);
2194 format_dev_t(md
->name
, MKDEV(_major
, minor
));
2196 md
->wq
= alloc_workqueue("kdmflush", WQ_MEM_RECLAIM
, 0);
2200 md
->bdev
= bdget_disk(md
->disk
, 0);
2204 bio_init(&md
->flush_bio
);
2205 md
->flush_bio
.bi_bdev
= md
->bdev
;
2206 md
->flush_bio
.bi_rw
= WRITE_FLUSH
;
2208 dm_stats_init(&md
->stats
);
2210 /* Populate the mapping, nobody knows we exist yet */
2211 spin_lock(&_minor_lock
);
2212 old_md
= idr_replace(&_minor_idr
, md
, minor
);
2213 spin_unlock(&_minor_lock
);
2215 BUG_ON(old_md
!= MINOR_ALLOCED
);
2220 destroy_workqueue(md
->wq
);
2222 del_gendisk(md
->disk
);
2225 blk_cleanup_queue(md
->queue
);
2227 cleanup_srcu_struct(&md
->io_barrier
);
2231 module_put(THIS_MODULE
);
2237 static void unlock_fs(struct mapped_device
*md
);
2239 static void free_dev(struct mapped_device
*md
)
2241 int minor
= MINOR(disk_devt(md
->disk
));
2245 destroy_workqueue(md
->wq
);
2247 if (md
->kworker_task
)
2248 kthread_stop(md
->kworker_task
);
2250 mempool_destroy(md
->io_pool
);
2252 mempool_destroy(md
->rq_pool
);
2254 bioset_free(md
->bs
);
2255 blk_integrity_unregister(md
->disk
);
2256 del_gendisk(md
->disk
);
2257 cleanup_srcu_struct(&md
->io_barrier
);
2258 free_table_devices(&md
->table_devices
);
2261 spin_lock(&_minor_lock
);
2262 md
->disk
->private_data
= NULL
;
2263 spin_unlock(&_minor_lock
);
2266 blk_cleanup_queue(md
->queue
);
2267 dm_stats_cleanup(&md
->stats
);
2268 module_put(THIS_MODULE
);
2272 static void __bind_mempools(struct mapped_device
*md
, struct dm_table
*t
)
2274 struct dm_md_mempools
*p
= dm_table_get_md_mempools(t
);
2276 if (md
->io_pool
&& md
->bs
) {
2277 /* The md already has necessary mempools. */
2278 if (dm_table_get_type(t
) == DM_TYPE_BIO_BASED
) {
2280 * Reload bioset because front_pad may have changed
2281 * because a different table was loaded.
2283 bioset_free(md
->bs
);
2288 * There's no need to reload with request-based dm
2289 * because the size of front_pad doesn't change.
2290 * Note for future: If you are to reload bioset,
2291 * prep-ed requests in the queue may refer
2292 * to bio from the old bioset, so you must walk
2293 * through the queue to unprep.
2298 BUG_ON(!p
|| md
->io_pool
|| md
->rq_pool
|| md
->bs
);
2300 md
->io_pool
= p
->io_pool
;
2302 md
->rq_pool
= p
->rq_pool
;
2308 /* mempool bind completed, now no need any mempools in the table */
2309 dm_table_free_md_mempools(t
);
2313 * Bind a table to the device.
2315 static void event_callback(void *context
)
2317 unsigned long flags
;
2319 struct mapped_device
*md
= (struct mapped_device
*) context
;
2321 spin_lock_irqsave(&md
->uevent_lock
, flags
);
2322 list_splice_init(&md
->uevent_list
, &uevents
);
2323 spin_unlock_irqrestore(&md
->uevent_lock
, flags
);
2325 dm_send_uevents(&uevents
, &disk_to_dev(md
->disk
)->kobj
);
2327 atomic_inc(&md
->event_nr
);
2328 wake_up(&md
->eventq
);
2332 * Protected by md->suspend_lock obtained by dm_swap_table().
2334 static void __set_size(struct mapped_device
*md
, sector_t size
)
2336 set_capacity(md
->disk
, size
);
2338 i_size_write(md
->bdev
->bd_inode
, (loff_t
)size
<< SECTOR_SHIFT
);
2342 * Return 1 if the queue has a compulsory merge_bvec_fn function.
2344 * If this function returns 0, then the device is either a non-dm
2345 * device without a merge_bvec_fn, or it is a dm device that is
2346 * able to split any bios it receives that are too big.
2348 int dm_queue_merge_is_compulsory(struct request_queue
*q
)
2350 struct mapped_device
*dev_md
;
2352 if (!q
->merge_bvec_fn
)
2355 if (q
->make_request_fn
== dm_request
) {
2356 dev_md
= q
->queuedata
;
2357 if (test_bit(DMF_MERGE_IS_OPTIONAL
, &dev_md
->flags
))
2364 static int dm_device_merge_is_compulsory(struct dm_target
*ti
,
2365 struct dm_dev
*dev
, sector_t start
,
2366 sector_t len
, void *data
)
2368 struct block_device
*bdev
= dev
->bdev
;
2369 struct request_queue
*q
= bdev_get_queue(bdev
);
2371 return dm_queue_merge_is_compulsory(q
);
2375 * Return 1 if it is acceptable to ignore merge_bvec_fn based
2376 * on the properties of the underlying devices.
2378 static int dm_table_merge_is_optional(struct dm_table
*table
)
2381 struct dm_target
*ti
;
2383 while (i
< dm_table_get_num_targets(table
)) {
2384 ti
= dm_table_get_target(table
, i
++);
2386 if (ti
->type
->iterate_devices
&&
2387 ti
->type
->iterate_devices(ti
, dm_device_merge_is_compulsory
, NULL
))
2395 * Returns old map, which caller must destroy.
2397 static struct dm_table
*__bind(struct mapped_device
*md
, struct dm_table
*t
,
2398 struct queue_limits
*limits
)
2400 struct dm_table
*old_map
;
2401 struct request_queue
*q
= md
->queue
;
2403 int merge_is_optional
;
2405 size
= dm_table_get_size(t
);
2408 * Wipe any geometry if the size of the table changed.
2410 if (size
!= dm_get_size(md
))
2411 memset(&md
->geometry
, 0, sizeof(md
->geometry
));
2413 __set_size(md
, size
);
2415 dm_table_event_callback(t
, event_callback
, md
);
2418 * The queue hasn't been stopped yet, if the old table type wasn't
2419 * for request-based during suspension. So stop it to prevent
2420 * I/O mapping before resume.
2421 * This must be done before setting the queue restrictions,
2422 * because request-based dm may be run just after the setting.
2424 if (dm_table_request_based(t
) && !blk_queue_stopped(q
))
2427 __bind_mempools(md
, t
);
2429 merge_is_optional
= dm_table_merge_is_optional(t
);
2431 old_map
= rcu_dereference_protected(md
->map
, lockdep_is_held(&md
->suspend_lock
));
2432 rcu_assign_pointer(md
->map
, t
);
2433 md
->immutable_target_type
= dm_table_get_immutable_target_type(t
);
2435 dm_table_set_restrictions(t
, q
, limits
);
2436 if (merge_is_optional
)
2437 set_bit(DMF_MERGE_IS_OPTIONAL
, &md
->flags
);
2439 clear_bit(DMF_MERGE_IS_OPTIONAL
, &md
->flags
);
2447 * Returns unbound table for the caller to free.
2449 static struct dm_table
*__unbind(struct mapped_device
*md
)
2451 struct dm_table
*map
= rcu_dereference_protected(md
->map
, 1);
2456 dm_table_event_callback(map
, NULL
, NULL
);
2457 RCU_INIT_POINTER(md
->map
, NULL
);
2464 * Constructor for a new device.
2466 int dm_create(int minor
, struct mapped_device
**result
)
2468 struct mapped_device
*md
;
2470 md
= alloc_dev(minor
);
2481 * Functions to manage md->type.
2482 * All are required to hold md->type_lock.
2484 void dm_lock_md_type(struct mapped_device
*md
)
2486 mutex_lock(&md
->type_lock
);
2489 void dm_unlock_md_type(struct mapped_device
*md
)
2491 mutex_unlock(&md
->type_lock
);
2494 void dm_set_md_type(struct mapped_device
*md
, unsigned type
)
2496 BUG_ON(!mutex_is_locked(&md
->type_lock
));
2500 unsigned dm_get_md_type(struct mapped_device
*md
)
2502 BUG_ON(!mutex_is_locked(&md
->type_lock
));
2506 static bool dm_md_type_request_based(struct mapped_device
*md
)
2508 unsigned table_type
= dm_get_md_type(md
);
2510 return (table_type
== DM_TYPE_REQUEST_BASED
||
2511 table_type
== DM_TYPE_MQ_REQUEST_BASED
);
2514 struct target_type
*dm_get_immutable_target_type(struct mapped_device
*md
)
2516 return md
->immutable_target_type
;
2520 * The queue_limits are only valid as long as you have a reference
2523 struct queue_limits
*dm_get_queue_limits(struct mapped_device
*md
)
2525 BUG_ON(!atomic_read(&md
->holders
));
2526 return &md
->queue
->limits
;
2528 EXPORT_SYMBOL_GPL(dm_get_queue_limits
);
2531 * Fully initialize a request-based queue (->elevator, ->request_fn, etc).
2533 static int dm_init_request_based_queue(struct mapped_device
*md
)
2535 struct request_queue
*q
= NULL
;
2537 if (md
->queue
->elevator
)
2540 /* Fully initialize the queue */
2541 q
= blk_init_allocated_queue(md
->queue
, dm_request_fn
, NULL
);
2546 dm_init_md_queue(md
);
2547 blk_queue_softirq_done(md
->queue
, dm_softirq_done
);
2548 blk_queue_prep_rq(md
->queue
, dm_prep_fn
);
2549 blk_queue_lld_busy(md
->queue
, dm_lld_busy
);
2551 /* Also initialize the request-based DM worker thread */
2552 init_kthread_worker(&md
->kworker
);
2553 md
->kworker_task
= kthread_run(kthread_worker_fn
, &md
->kworker
,
2554 "kdmwork-%s", dm_device_name(md
));
2556 elv_register_queue(md
->queue
);
2562 * Setup the DM device's queue based on md's type
2564 int dm_setup_md_queue(struct mapped_device
*md
)
2566 if (dm_md_type_request_based(md
) && !dm_init_request_based_queue(md
)) {
2567 DMWARN("Cannot initialize queue for request-based mapped device");
2574 struct mapped_device
*dm_get_md(dev_t dev
)
2576 struct mapped_device
*md
;
2577 unsigned minor
= MINOR(dev
);
2579 if (MAJOR(dev
) != _major
|| minor
>= (1 << MINORBITS
))
2582 spin_lock(&_minor_lock
);
2584 md
= idr_find(&_minor_idr
, minor
);
2586 if ((md
== MINOR_ALLOCED
||
2587 (MINOR(disk_devt(dm_disk(md
))) != minor
) ||
2588 dm_deleting_md(md
) ||
2589 test_bit(DMF_FREEING
, &md
->flags
))) {
2597 spin_unlock(&_minor_lock
);
2601 EXPORT_SYMBOL_GPL(dm_get_md
);
2603 void *dm_get_mdptr(struct mapped_device
*md
)
2605 return md
->interface_ptr
;
2608 void dm_set_mdptr(struct mapped_device
*md
, void *ptr
)
2610 md
->interface_ptr
= ptr
;
2613 void dm_get(struct mapped_device
*md
)
2615 atomic_inc(&md
->holders
);
2616 BUG_ON(test_bit(DMF_FREEING
, &md
->flags
));
2619 const char *dm_device_name(struct mapped_device
*md
)
2623 EXPORT_SYMBOL_GPL(dm_device_name
);
2625 static void __dm_destroy(struct mapped_device
*md
, bool wait
)
2627 struct dm_table
*map
;
2632 spin_lock(&_minor_lock
);
2633 map
= dm_get_live_table(md
, &srcu_idx
);
2634 idr_replace(&_minor_idr
, MINOR_ALLOCED
, MINOR(disk_devt(dm_disk(md
))));
2635 set_bit(DMF_FREEING
, &md
->flags
);
2636 spin_unlock(&_minor_lock
);
2638 if (dm_request_based(md
))
2639 flush_kthread_worker(&md
->kworker
);
2641 if (!dm_suspended_md(md
)) {
2642 dm_table_presuspend_targets(map
);
2643 dm_table_postsuspend_targets(map
);
2646 /* dm_put_live_table must be before msleep, otherwise deadlock is possible */
2647 dm_put_live_table(md
, srcu_idx
);
2650 * Rare, but there may be I/O requests still going to complete,
2651 * for example. Wait for all references to disappear.
2652 * No one should increment the reference count of the mapped_device,
2653 * after the mapped_device state becomes DMF_FREEING.
2656 while (atomic_read(&md
->holders
))
2658 else if (atomic_read(&md
->holders
))
2659 DMWARN("%s: Forcibly removing mapped_device still in use! (%d users)",
2660 dm_device_name(md
), atomic_read(&md
->holders
));
2663 dm_table_destroy(__unbind(md
));
2667 void dm_destroy(struct mapped_device
*md
)
2669 __dm_destroy(md
, true);
2672 void dm_destroy_immediate(struct mapped_device
*md
)
2674 __dm_destroy(md
, false);
2677 void dm_put(struct mapped_device
*md
)
2679 atomic_dec(&md
->holders
);
2681 EXPORT_SYMBOL_GPL(dm_put
);
2683 static int dm_wait_for_completion(struct mapped_device
*md
, int interruptible
)
2686 DECLARE_WAITQUEUE(wait
, current
);
2688 add_wait_queue(&md
->wait
, &wait
);
2691 set_current_state(interruptible
);
2693 if (!md_in_flight(md
))
2696 if (interruptible
== TASK_INTERRUPTIBLE
&&
2697 signal_pending(current
)) {
2704 set_current_state(TASK_RUNNING
);
2706 remove_wait_queue(&md
->wait
, &wait
);
2712 * Process the deferred bios
2714 static void dm_wq_work(struct work_struct
*work
)
2716 struct mapped_device
*md
= container_of(work
, struct mapped_device
,
2720 struct dm_table
*map
;
2722 map
= dm_get_live_table(md
, &srcu_idx
);
2724 while (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND
, &md
->flags
)) {
2725 spin_lock_irq(&md
->deferred_lock
);
2726 c
= bio_list_pop(&md
->deferred
);
2727 spin_unlock_irq(&md
->deferred_lock
);
2732 if (dm_request_based(md
))
2733 generic_make_request(c
);
2735 __split_and_process_bio(md
, map
, c
);
2738 dm_put_live_table(md
, srcu_idx
);
2741 static void dm_queue_flush(struct mapped_device
*md
)
2743 clear_bit(DMF_BLOCK_IO_FOR_SUSPEND
, &md
->flags
);
2744 smp_mb__after_atomic();
2745 queue_work(md
->wq
, &md
->work
);
2749 * Swap in a new table, returning the old one for the caller to destroy.
2751 struct dm_table
*dm_swap_table(struct mapped_device
*md
, struct dm_table
*table
)
2753 struct dm_table
*live_map
= NULL
, *map
= ERR_PTR(-EINVAL
);
2754 struct queue_limits limits
;
2757 mutex_lock(&md
->suspend_lock
);
2759 /* device must be suspended */
2760 if (!dm_suspended_md(md
))
2764 * If the new table has no data devices, retain the existing limits.
2765 * This helps multipath with queue_if_no_path if all paths disappear,
2766 * then new I/O is queued based on these limits, and then some paths
2769 if (dm_table_has_no_data_devices(table
)) {
2770 live_map
= dm_get_live_table_fast(md
);
2772 limits
= md
->queue
->limits
;
2773 dm_put_live_table_fast(md
);
2777 r
= dm_calculate_queue_limits(table
, &limits
);
2784 map
= __bind(md
, table
, &limits
);
2787 mutex_unlock(&md
->suspend_lock
);
2792 * Functions to lock and unlock any filesystem running on the
2795 static int lock_fs(struct mapped_device
*md
)
2799 WARN_ON(md
->frozen_sb
);
2801 md
->frozen_sb
= freeze_bdev(md
->bdev
);
2802 if (IS_ERR(md
->frozen_sb
)) {
2803 r
= PTR_ERR(md
->frozen_sb
);
2804 md
->frozen_sb
= NULL
;
2808 set_bit(DMF_FROZEN
, &md
->flags
);
2813 static void unlock_fs(struct mapped_device
*md
)
2815 if (!test_bit(DMF_FROZEN
, &md
->flags
))
2818 thaw_bdev(md
->bdev
, md
->frozen_sb
);
2819 md
->frozen_sb
= NULL
;
2820 clear_bit(DMF_FROZEN
, &md
->flags
);
2824 * If __dm_suspend returns 0, the device is completely quiescent
2825 * now. There is no request-processing activity. All new requests
2826 * are being added to md->deferred list.
2828 * Caller must hold md->suspend_lock
2830 static int __dm_suspend(struct mapped_device
*md
, struct dm_table
*map
,
2831 unsigned suspend_flags
, int interruptible
)
2833 bool do_lockfs
= suspend_flags
& DM_SUSPEND_LOCKFS_FLAG
;
2834 bool noflush
= suspend_flags
& DM_SUSPEND_NOFLUSH_FLAG
;
2838 * DMF_NOFLUSH_SUSPENDING must be set before presuspend.
2839 * This flag is cleared before dm_suspend returns.
2842 set_bit(DMF_NOFLUSH_SUSPENDING
, &md
->flags
);
2845 * This gets reverted if there's an error later and the targets
2846 * provide the .presuspend_undo hook.
2848 dm_table_presuspend_targets(map
);
2851 * Flush I/O to the device.
2852 * Any I/O submitted after lock_fs() may not be flushed.
2853 * noflush takes precedence over do_lockfs.
2854 * (lock_fs() flushes I/Os and waits for them to complete.)
2856 if (!noflush
&& do_lockfs
) {
2859 dm_table_presuspend_undo_targets(map
);
2865 * Here we must make sure that no processes are submitting requests
2866 * to target drivers i.e. no one may be executing
2867 * __split_and_process_bio. This is called from dm_request and
2870 * To get all processes out of __split_and_process_bio in dm_request,
2871 * we take the write lock. To prevent any process from reentering
2872 * __split_and_process_bio from dm_request and quiesce the thread
2873 * (dm_wq_work), we set BMF_BLOCK_IO_FOR_SUSPEND and call
2874 * flush_workqueue(md->wq).
2876 set_bit(DMF_BLOCK_IO_FOR_SUSPEND
, &md
->flags
);
2878 synchronize_srcu(&md
->io_barrier
);
2881 * Stop md->queue before flushing md->wq in case request-based
2882 * dm defers requests to md->wq from md->queue.
2884 if (dm_request_based(md
)) {
2885 stop_queue(md
->queue
);
2886 flush_kthread_worker(&md
->kworker
);
2889 flush_workqueue(md
->wq
);
2892 * At this point no more requests are entering target request routines.
2893 * We call dm_wait_for_completion to wait for all existing requests
2896 r
= dm_wait_for_completion(md
, interruptible
);
2899 clear_bit(DMF_NOFLUSH_SUSPENDING
, &md
->flags
);
2901 synchronize_srcu(&md
->io_barrier
);
2903 /* were we interrupted ? */
2907 if (dm_request_based(md
))
2908 start_queue(md
->queue
);
2911 dm_table_presuspend_undo_targets(map
);
2912 /* pushback list is already flushed, so skip flush */
2919 * We need to be able to change a mapping table under a mounted
2920 * filesystem. For example we might want to move some data in
2921 * the background. Before the table can be swapped with
2922 * dm_bind_table, dm_suspend must be called to flush any in
2923 * flight bios and ensure that any further io gets deferred.
2926 * Suspend mechanism in request-based dm.
2928 * 1. Flush all I/Os by lock_fs() if needed.
2929 * 2. Stop dispatching any I/O by stopping the request_queue.
2930 * 3. Wait for all in-flight I/Os to be completed or requeued.
2932 * To abort suspend, start the request_queue.
2934 int dm_suspend(struct mapped_device
*md
, unsigned suspend_flags
)
2936 struct dm_table
*map
= NULL
;
2940 mutex_lock_nested(&md
->suspend_lock
, SINGLE_DEPTH_NESTING
);
2942 if (dm_suspended_md(md
)) {
2947 if (dm_suspended_internally_md(md
)) {
2948 /* already internally suspended, wait for internal resume */
2949 mutex_unlock(&md
->suspend_lock
);
2950 r
= wait_on_bit(&md
->flags
, DMF_SUSPENDED_INTERNALLY
, TASK_INTERRUPTIBLE
);
2956 map
= rcu_dereference_protected(md
->map
, lockdep_is_held(&md
->suspend_lock
));
2958 r
= __dm_suspend(md
, map
, suspend_flags
, TASK_INTERRUPTIBLE
);
2962 set_bit(DMF_SUSPENDED
, &md
->flags
);
2964 dm_table_postsuspend_targets(map
);
2967 mutex_unlock(&md
->suspend_lock
);
2971 static int __dm_resume(struct mapped_device
*md
, struct dm_table
*map
)
2974 int r
= dm_table_resume_targets(map
);
2982 * Flushing deferred I/Os must be done after targets are resumed
2983 * so that mapping of targets can work correctly.
2984 * Request-based dm is queueing the deferred I/Os in its request_queue.
2986 if (dm_request_based(md
))
2987 start_queue(md
->queue
);
2994 int dm_resume(struct mapped_device
*md
)
2997 struct dm_table
*map
= NULL
;
3000 mutex_lock_nested(&md
->suspend_lock
, SINGLE_DEPTH_NESTING
);
3002 if (!dm_suspended_md(md
))
3005 if (dm_suspended_internally_md(md
)) {
3006 /* already internally suspended, wait for internal resume */
3007 mutex_unlock(&md
->suspend_lock
);
3008 r
= wait_on_bit(&md
->flags
, DMF_SUSPENDED_INTERNALLY
, TASK_INTERRUPTIBLE
);
3014 map
= rcu_dereference_protected(md
->map
, lockdep_is_held(&md
->suspend_lock
));
3015 if (!map
|| !dm_table_get_size(map
))
3018 r
= __dm_resume(md
, map
);
3022 clear_bit(DMF_SUSPENDED
, &md
->flags
);
3026 mutex_unlock(&md
->suspend_lock
);
3032 * Internal suspend/resume works like userspace-driven suspend. It waits
3033 * until all bios finish and prevents issuing new bios to the target drivers.
3034 * It may be used only from the kernel.
3037 static void __dm_internal_suspend(struct mapped_device
*md
, unsigned suspend_flags
)
3039 struct dm_table
*map
= NULL
;
3041 if (md
->internal_suspend_count
++)
3042 return; /* nested internal suspend */
3044 if (dm_suspended_md(md
)) {
3045 set_bit(DMF_SUSPENDED_INTERNALLY
, &md
->flags
);
3046 return; /* nest suspend */
3049 map
= rcu_dereference_protected(md
->map
, lockdep_is_held(&md
->suspend_lock
));
3052 * Using TASK_UNINTERRUPTIBLE because only NOFLUSH internal suspend is
3053 * supported. Properly supporting a TASK_INTERRUPTIBLE internal suspend
3054 * would require changing .presuspend to return an error -- avoid this
3055 * until there is a need for more elaborate variants of internal suspend.
3057 (void) __dm_suspend(md
, map
, suspend_flags
, TASK_UNINTERRUPTIBLE
);
3059 set_bit(DMF_SUSPENDED_INTERNALLY
, &md
->flags
);
3061 dm_table_postsuspend_targets(map
);
3064 static void __dm_internal_resume(struct mapped_device
*md
)
3066 BUG_ON(!md
->internal_suspend_count
);
3068 if (--md
->internal_suspend_count
)
3069 return; /* resume from nested internal suspend */
3071 if (dm_suspended_md(md
))
3072 goto done
; /* resume from nested suspend */
3075 * NOTE: existing callers don't need to call dm_table_resume_targets
3076 * (which may fail -- so best to avoid it for now by passing NULL map)
3078 (void) __dm_resume(md
, NULL
);
3081 clear_bit(DMF_SUSPENDED_INTERNALLY
, &md
->flags
);
3082 smp_mb__after_atomic();
3083 wake_up_bit(&md
->flags
, DMF_SUSPENDED_INTERNALLY
);
3086 void dm_internal_suspend_noflush(struct mapped_device
*md
)
3088 mutex_lock(&md
->suspend_lock
);
3089 __dm_internal_suspend(md
, DM_SUSPEND_NOFLUSH_FLAG
);
3090 mutex_unlock(&md
->suspend_lock
);
3092 EXPORT_SYMBOL_GPL(dm_internal_suspend_noflush
);
3094 void dm_internal_resume(struct mapped_device
*md
)
3096 mutex_lock(&md
->suspend_lock
);
3097 __dm_internal_resume(md
);
3098 mutex_unlock(&md
->suspend_lock
);
3100 EXPORT_SYMBOL_GPL(dm_internal_resume
);
3103 * Fast variants of internal suspend/resume hold md->suspend_lock,
3104 * which prevents interaction with userspace-driven suspend.
3107 void dm_internal_suspend_fast(struct mapped_device
*md
)
3109 mutex_lock(&md
->suspend_lock
);
3110 if (dm_suspended_md(md
) || dm_suspended_internally_md(md
))
3113 set_bit(DMF_BLOCK_IO_FOR_SUSPEND
, &md
->flags
);
3114 synchronize_srcu(&md
->io_barrier
);
3115 flush_workqueue(md
->wq
);
3116 dm_wait_for_completion(md
, TASK_UNINTERRUPTIBLE
);
3119 void dm_internal_resume_fast(struct mapped_device
*md
)
3121 if (dm_suspended_md(md
) || dm_suspended_internally_md(md
))
3127 mutex_unlock(&md
->suspend_lock
);
3130 /*-----------------------------------------------------------------
3131 * Event notification.
3132 *---------------------------------------------------------------*/
3133 int dm_kobject_uevent(struct mapped_device
*md
, enum kobject_action action
,
3136 char udev_cookie
[DM_COOKIE_LENGTH
];
3137 char *envp
[] = { udev_cookie
, NULL
};
3140 return kobject_uevent(&disk_to_dev(md
->disk
)->kobj
, action
);
3142 snprintf(udev_cookie
, DM_COOKIE_LENGTH
, "%s=%u",
3143 DM_COOKIE_ENV_VAR_NAME
, cookie
);
3144 return kobject_uevent_env(&disk_to_dev(md
->disk
)->kobj
,
3149 uint32_t dm_next_uevent_seq(struct mapped_device
*md
)
3151 return atomic_add_return(1, &md
->uevent_seq
);
3154 uint32_t dm_get_event_nr(struct mapped_device
*md
)
3156 return atomic_read(&md
->event_nr
);
3159 int dm_wait_event(struct mapped_device
*md
, int event_nr
)
3161 return wait_event_interruptible(md
->eventq
,
3162 (event_nr
!= atomic_read(&md
->event_nr
)));
3165 void dm_uevent_add(struct mapped_device
*md
, struct list_head
*elist
)
3167 unsigned long flags
;
3169 spin_lock_irqsave(&md
->uevent_lock
, flags
);
3170 list_add(elist
, &md
->uevent_list
);
3171 spin_unlock_irqrestore(&md
->uevent_lock
, flags
);
3175 * The gendisk is only valid as long as you have a reference
3178 struct gendisk
*dm_disk(struct mapped_device
*md
)
3183 struct kobject
*dm_kobject(struct mapped_device
*md
)
3185 return &md
->kobj_holder
.kobj
;
3188 struct mapped_device
*dm_get_from_kobject(struct kobject
*kobj
)
3190 struct mapped_device
*md
;
3192 md
= container_of(kobj
, struct mapped_device
, kobj_holder
.kobj
);
3194 if (test_bit(DMF_FREEING
, &md
->flags
) ||
3202 int dm_suspended_md(struct mapped_device
*md
)
3204 return test_bit(DMF_SUSPENDED
, &md
->flags
);
3207 int dm_suspended_internally_md(struct mapped_device
*md
)
3209 return test_bit(DMF_SUSPENDED_INTERNALLY
, &md
->flags
);
3212 int dm_test_deferred_remove_flag(struct mapped_device
*md
)
3214 return test_bit(DMF_DEFERRED_REMOVE
, &md
->flags
);
3217 int dm_suspended(struct dm_target
*ti
)
3219 return dm_suspended_md(dm_table_get_md(ti
->table
));
3221 EXPORT_SYMBOL_GPL(dm_suspended
);
3223 int dm_noflush_suspending(struct dm_target
*ti
)
3225 return __noflush_suspending(dm_table_get_md(ti
->table
));
3227 EXPORT_SYMBOL_GPL(dm_noflush_suspending
);
3229 struct dm_md_mempools
*dm_alloc_md_mempools(unsigned type
, unsigned integrity
, unsigned per_bio_data_size
)
3231 struct dm_md_mempools
*pools
= kzalloc(sizeof(*pools
), GFP_KERNEL
);
3232 struct kmem_cache
*cachep
;
3233 unsigned int pool_size
= 0;
3234 unsigned int front_pad
;
3240 case DM_TYPE_BIO_BASED
:
3242 pool_size
= dm_get_reserved_bio_based_ios();
3243 front_pad
= roundup(per_bio_data_size
, __alignof__(struct dm_target_io
)) + offsetof(struct dm_target_io
, clone
);
3245 case DM_TYPE_REQUEST_BASED
:
3246 pool_size
= dm_get_reserved_rq_based_ios();
3247 pools
->rq_pool
= mempool_create_slab_pool(pool_size
, _rq_cache
);
3248 if (!pools
->rq_pool
)
3250 /* fall through to setup remaining rq-based pools */
3251 case DM_TYPE_MQ_REQUEST_BASED
:
3252 cachep
= _rq_tio_cache
;
3254 pool_size
= dm_get_reserved_rq_based_ios();
3255 front_pad
= offsetof(struct dm_rq_clone_bio_info
, clone
);
3256 /* per_bio_data_size is not used. See __bind_mempools(). */
3257 WARN_ON(per_bio_data_size
!= 0);
3263 pools
->io_pool
= mempool_create_slab_pool(pool_size
, cachep
);
3264 if (!pools
->io_pool
)
3267 pools
->bs
= bioset_create_nobvec(pool_size
, front_pad
);
3271 if (integrity
&& bioset_integrity_create(pools
->bs
, pool_size
))
3277 dm_free_md_mempools(pools
);
3282 void dm_free_md_mempools(struct dm_md_mempools
*pools
)
3288 mempool_destroy(pools
->io_pool
);
3291 mempool_destroy(pools
->rq_pool
);
3294 bioset_free(pools
->bs
);
3299 static const struct block_device_operations dm_blk_dops
= {
3300 .open
= dm_blk_open
,
3301 .release
= dm_blk_close
,
3302 .ioctl
= dm_blk_ioctl
,
3303 .getgeo
= dm_blk_getgeo
,
3304 .owner
= THIS_MODULE
3310 module_init(dm_init
);
3311 module_exit(dm_exit
);
3313 module_param(major
, uint
, 0);
3314 MODULE_PARM_DESC(major
, "The major number of the device mapper");
3316 module_param(reserved_bio_based_ios
, uint
, S_IRUGO
| S_IWUSR
);
3317 MODULE_PARM_DESC(reserved_bio_based_ios
, "Reserved IOs in bio-based mempools");
3319 module_param(reserved_rq_based_ios
, uint
, S_IRUGO
| S_IWUSR
);
3320 MODULE_PARM_DESC(reserved_rq_based_ios
, "Reserved IOs in request-based mempools");
3322 MODULE_DESCRIPTION(DM_NAME
" driver");
3323 MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
3324 MODULE_LICENSE("GPL");