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>
24 #include <trace/events/block.h>
26 #define DM_MSG_PREFIX "core"
30 * ratelimit state to be used in DMXXX_LIMIT().
32 DEFINE_RATELIMIT_STATE(dm_ratelimit_state
,
33 DEFAULT_RATELIMIT_INTERVAL
,
34 DEFAULT_RATELIMIT_BURST
);
35 EXPORT_SYMBOL(dm_ratelimit_state
);
39 * Cookies are numeric values sent with CHANGE and REMOVE
40 * uevents while resuming, removing or renaming the device.
42 #define DM_COOKIE_ENV_VAR_NAME "DM_COOKIE"
43 #define DM_COOKIE_LENGTH 24
45 static const char *_name
= DM_NAME
;
47 static unsigned int major
= 0;
48 static unsigned int _major
= 0;
50 static DEFINE_IDR(_minor_idr
);
52 static DEFINE_SPINLOCK(_minor_lock
);
54 static void do_deferred_remove(struct work_struct
*w
);
56 static DECLARE_WORK(deferred_remove_work
, do_deferred_remove
);
58 static struct workqueue_struct
*deferred_remove_workqueue
;
62 * One of these is allocated per bio.
65 struct mapped_device
*md
;
69 unsigned long start_time
;
70 spinlock_t endio_lock
;
71 struct dm_stats_aux stats_aux
;
75 * For request-based dm.
76 * One of these is allocated per request.
78 struct dm_rq_target_io
{
79 struct mapped_device
*md
;
81 struct request
*orig
, clone
;
87 * For request-based dm - the bio clones we allocate are embedded in these
90 * We allocate these with bio_alloc_bioset, using the front_pad parameter when
91 * the bioset is created - this means the bio has to come at the end of the
94 struct dm_rq_clone_bio_info
{
96 struct dm_rq_target_io
*tio
;
100 union map_info
*dm_get_rq_mapinfo(struct request
*rq
)
102 if (rq
&& rq
->end_io_data
)
103 return &((struct dm_rq_target_io
*)rq
->end_io_data
)->info
;
106 EXPORT_SYMBOL_GPL(dm_get_rq_mapinfo
);
108 #define MINOR_ALLOCED ((void *)-1)
111 * Bits for the md->flags field.
113 #define DMF_BLOCK_IO_FOR_SUSPEND 0
114 #define DMF_SUSPENDED 1
116 #define DMF_FREEING 3
117 #define DMF_DELETING 4
118 #define DMF_NOFLUSH_SUSPENDING 5
119 #define DMF_MERGE_IS_OPTIONAL 6
120 #define DMF_DEFERRED_REMOVE 7
121 #define DMF_SUSPENDED_INTERNALLY 8
124 * A dummy definition to make RCU happy.
125 * struct dm_table should never be dereferenced in this file.
132 * Work processed by per-device workqueue.
134 struct mapped_device
{
135 struct srcu_struct io_barrier
;
136 struct mutex suspend_lock
;
141 * The current mapping.
142 * Use dm_get_live_table{_fast} or take suspend_lock for
145 struct dm_table __rcu
*map
;
147 struct list_head table_devices
;
148 struct mutex table_devices_lock
;
152 struct request_queue
*queue
;
154 /* Protect queue and type against concurrent access. */
155 struct mutex type_lock
;
157 struct target_type
*immutable_target_type
;
159 struct gendisk
*disk
;
165 * A list of ios that arrived while we were suspended.
168 wait_queue_head_t wait
;
169 struct work_struct work
;
170 struct bio_list deferred
;
171 spinlock_t deferred_lock
;
174 * Processing queue (flush)
176 struct workqueue_struct
*wq
;
179 * io objects are allocated from here.
189 wait_queue_head_t eventq
;
191 struct list_head uevent_list
;
192 spinlock_t uevent_lock
; /* Protect access to uevent_list */
195 * freeze/thaw support require holding onto a super block
197 struct super_block
*frozen_sb
;
198 struct block_device
*bdev
;
200 /* forced geometry settings */
201 struct hd_geometry geometry
;
203 /* kobject and completion */
204 struct dm_kobject_holder kobj_holder
;
206 /* zero-length flush that will be cloned and submitted to targets */
207 struct bio flush_bio
;
209 /* the number of internal suspends */
210 unsigned internal_suspend_count
;
212 struct dm_stats stats
;
216 * For mempools pre-allocation at the table loading time.
218 struct dm_md_mempools
{
223 struct table_device
{
224 struct list_head list
;
226 struct dm_dev dm_dev
;
229 #define RESERVED_BIO_BASED_IOS 16
230 #define RESERVED_REQUEST_BASED_IOS 256
231 #define RESERVED_MAX_IOS 1024
232 static struct kmem_cache
*_io_cache
;
233 static struct kmem_cache
*_rq_tio_cache
;
236 * Bio-based DM's mempools' reserved IOs set by the user.
238 static unsigned reserved_bio_based_ios
= RESERVED_BIO_BASED_IOS
;
241 * Request-based DM's mempools' reserved IOs set by the user.
243 static unsigned reserved_rq_based_ios
= RESERVED_REQUEST_BASED_IOS
;
245 static unsigned __dm_get_reserved_ios(unsigned *reserved_ios
,
246 unsigned def
, unsigned max
)
248 unsigned ios
= ACCESS_ONCE(*reserved_ios
);
249 unsigned modified_ios
= 0;
257 (void)cmpxchg(reserved_ios
, ios
, modified_ios
);
264 unsigned dm_get_reserved_bio_based_ios(void)
266 return __dm_get_reserved_ios(&reserved_bio_based_ios
,
267 RESERVED_BIO_BASED_IOS
, RESERVED_MAX_IOS
);
269 EXPORT_SYMBOL_GPL(dm_get_reserved_bio_based_ios
);
271 unsigned dm_get_reserved_rq_based_ios(void)
273 return __dm_get_reserved_ios(&reserved_rq_based_ios
,
274 RESERVED_REQUEST_BASED_IOS
, RESERVED_MAX_IOS
);
276 EXPORT_SYMBOL_GPL(dm_get_reserved_rq_based_ios
);
278 static int __init
local_init(void)
282 /* allocate a slab for the dm_ios */
283 _io_cache
= KMEM_CACHE(dm_io
, 0);
287 _rq_tio_cache
= KMEM_CACHE(dm_rq_target_io
, 0);
289 goto out_free_io_cache
;
291 r
= dm_uevent_init();
293 goto out_free_rq_tio_cache
;
295 deferred_remove_workqueue
= alloc_workqueue("kdmremove", WQ_UNBOUND
, 1);
296 if (!deferred_remove_workqueue
) {
298 goto out_uevent_exit
;
302 r
= register_blkdev(_major
, _name
);
304 goto out_free_workqueue
;
312 destroy_workqueue(deferred_remove_workqueue
);
315 out_free_rq_tio_cache
:
316 kmem_cache_destroy(_rq_tio_cache
);
318 kmem_cache_destroy(_io_cache
);
323 static void local_exit(void)
325 flush_scheduled_work();
326 destroy_workqueue(deferred_remove_workqueue
);
328 kmem_cache_destroy(_rq_tio_cache
);
329 kmem_cache_destroy(_io_cache
);
330 unregister_blkdev(_major
, _name
);
335 DMINFO("cleaned up");
338 static int (*_inits
[])(void) __initdata
= {
349 static void (*_exits
[])(void) = {
360 static int __init
dm_init(void)
362 const int count
= ARRAY_SIZE(_inits
);
366 for (i
= 0; i
< count
; i
++) {
381 static void __exit
dm_exit(void)
383 int i
= ARRAY_SIZE(_exits
);
389 * Should be empty by this point.
391 idr_destroy(&_minor_idr
);
395 * Block device functions
397 int dm_deleting_md(struct mapped_device
*md
)
399 return test_bit(DMF_DELETING
, &md
->flags
);
402 static int dm_blk_open(struct block_device
*bdev
, fmode_t mode
)
404 struct mapped_device
*md
;
406 spin_lock(&_minor_lock
);
408 md
= bdev
->bd_disk
->private_data
;
412 if (test_bit(DMF_FREEING
, &md
->flags
) ||
413 dm_deleting_md(md
)) {
419 atomic_inc(&md
->open_count
);
422 spin_unlock(&_minor_lock
);
424 return md
? 0 : -ENXIO
;
427 static void dm_blk_close(struct gendisk
*disk
, fmode_t mode
)
429 struct mapped_device
*md
= disk
->private_data
;
431 spin_lock(&_minor_lock
);
433 if (atomic_dec_and_test(&md
->open_count
) &&
434 (test_bit(DMF_DEFERRED_REMOVE
, &md
->flags
)))
435 queue_work(deferred_remove_workqueue
, &deferred_remove_work
);
439 spin_unlock(&_minor_lock
);
442 int dm_open_count(struct mapped_device
*md
)
444 return atomic_read(&md
->open_count
);
448 * Guarantees nothing is using the device before it's deleted.
450 int dm_lock_for_deletion(struct mapped_device
*md
, bool mark_deferred
, bool only_deferred
)
454 spin_lock(&_minor_lock
);
456 if (dm_open_count(md
)) {
459 set_bit(DMF_DEFERRED_REMOVE
, &md
->flags
);
460 } else if (only_deferred
&& !test_bit(DMF_DEFERRED_REMOVE
, &md
->flags
))
463 set_bit(DMF_DELETING
, &md
->flags
);
465 spin_unlock(&_minor_lock
);
470 int dm_cancel_deferred_remove(struct mapped_device
*md
)
474 spin_lock(&_minor_lock
);
476 if (test_bit(DMF_DELETING
, &md
->flags
))
479 clear_bit(DMF_DEFERRED_REMOVE
, &md
->flags
);
481 spin_unlock(&_minor_lock
);
486 static void do_deferred_remove(struct work_struct
*w
)
488 dm_deferred_remove();
491 sector_t
dm_get_size(struct mapped_device
*md
)
493 return get_capacity(md
->disk
);
496 struct request_queue
*dm_get_md_queue(struct mapped_device
*md
)
501 struct dm_stats
*dm_get_stats(struct mapped_device
*md
)
506 static int dm_blk_getgeo(struct block_device
*bdev
, struct hd_geometry
*geo
)
508 struct mapped_device
*md
= bdev
->bd_disk
->private_data
;
510 return dm_get_geometry(md
, geo
);
513 static int dm_blk_ioctl(struct block_device
*bdev
, fmode_t mode
,
514 unsigned int cmd
, unsigned long arg
)
516 struct mapped_device
*md
= bdev
->bd_disk
->private_data
;
518 struct dm_table
*map
;
519 struct dm_target
*tgt
;
523 map
= dm_get_live_table(md
, &srcu_idx
);
525 if (!map
|| !dm_table_get_size(map
))
528 /* We only support devices that have a single target */
529 if (dm_table_get_num_targets(map
) != 1)
532 tgt
= dm_table_get_target(map
, 0);
533 if (!tgt
->type
->ioctl
)
536 if (dm_suspended_md(md
)) {
541 r
= tgt
->type
->ioctl(tgt
, cmd
, arg
);
544 dm_put_live_table(md
, srcu_idx
);
546 if (r
== -ENOTCONN
) {
554 static struct dm_io
*alloc_io(struct mapped_device
*md
)
556 return mempool_alloc(md
->io_pool
, GFP_NOIO
);
559 static void free_io(struct mapped_device
*md
, struct dm_io
*io
)
561 mempool_free(io
, md
->io_pool
);
564 static void free_tio(struct mapped_device
*md
, struct dm_target_io
*tio
)
566 bio_put(&tio
->clone
);
569 static struct dm_rq_target_io
*alloc_rq_tio(struct mapped_device
*md
,
572 return mempool_alloc(md
->io_pool
, gfp_mask
);
575 static void free_rq_tio(struct dm_rq_target_io
*tio
)
577 mempool_free(tio
, tio
->md
->io_pool
);
580 static int md_in_flight(struct mapped_device
*md
)
582 return atomic_read(&md
->pending
[READ
]) +
583 atomic_read(&md
->pending
[WRITE
]);
586 static void start_io_acct(struct dm_io
*io
)
588 struct mapped_device
*md
= io
->md
;
589 struct bio
*bio
= io
->bio
;
591 int rw
= bio_data_dir(bio
);
593 io
->start_time
= jiffies
;
595 cpu
= part_stat_lock();
596 part_round_stats(cpu
, &dm_disk(md
)->part0
);
598 atomic_set(&dm_disk(md
)->part0
.in_flight
[rw
],
599 atomic_inc_return(&md
->pending
[rw
]));
601 if (unlikely(dm_stats_used(&md
->stats
)))
602 dm_stats_account_io(&md
->stats
, bio
->bi_rw
, bio
->bi_iter
.bi_sector
,
603 bio_sectors(bio
), false, 0, &io
->stats_aux
);
606 static void end_io_acct(struct dm_io
*io
)
608 struct mapped_device
*md
= io
->md
;
609 struct bio
*bio
= io
->bio
;
610 unsigned long duration
= jiffies
- io
->start_time
;
612 int rw
= bio_data_dir(bio
);
614 generic_end_io_acct(rw
, &dm_disk(md
)->part0
, io
->start_time
);
616 if (unlikely(dm_stats_used(&md
->stats
)))
617 dm_stats_account_io(&md
->stats
, bio
->bi_rw
, bio
->bi_iter
.bi_sector
,
618 bio_sectors(bio
), true, duration
, &io
->stats_aux
);
621 * After this is decremented the bio must not be touched if it is
624 pending
= atomic_dec_return(&md
->pending
[rw
]);
625 atomic_set(&dm_disk(md
)->part0
.in_flight
[rw
], pending
);
626 pending
+= atomic_read(&md
->pending
[rw
^0x1]);
628 /* nudge anyone waiting on suspend queue */
634 * Add the bio to the list of deferred io.
636 static void queue_io(struct mapped_device
*md
, struct bio
*bio
)
640 spin_lock_irqsave(&md
->deferred_lock
, flags
);
641 bio_list_add(&md
->deferred
, bio
);
642 spin_unlock_irqrestore(&md
->deferred_lock
, flags
);
643 queue_work(md
->wq
, &md
->work
);
647 * Everyone (including functions in this file), should use this
648 * function to access the md->map field, and make sure they call
649 * dm_put_live_table() when finished.
651 struct dm_table
*dm_get_live_table(struct mapped_device
*md
, int *srcu_idx
) __acquires(md
->io_barrier
)
653 *srcu_idx
= srcu_read_lock(&md
->io_barrier
);
655 return srcu_dereference(md
->map
, &md
->io_barrier
);
658 void dm_put_live_table(struct mapped_device
*md
, int srcu_idx
) __releases(md
->io_barrier
)
660 srcu_read_unlock(&md
->io_barrier
, srcu_idx
);
663 void dm_sync_table(struct mapped_device
*md
)
665 synchronize_srcu(&md
->io_barrier
);
666 synchronize_rcu_expedited();
670 * A fast alternative to dm_get_live_table/dm_put_live_table.
671 * The caller must not block between these two functions.
673 static struct dm_table
*dm_get_live_table_fast(struct mapped_device
*md
) __acquires(RCU
)
676 return rcu_dereference(md
->map
);
679 static void dm_put_live_table_fast(struct mapped_device
*md
) __releases(RCU
)
685 * Open a table device so we can use it as a map destination.
687 static int open_table_device(struct table_device
*td
, dev_t dev
,
688 struct mapped_device
*md
)
690 static char *_claim_ptr
= "I belong to device-mapper";
691 struct block_device
*bdev
;
695 BUG_ON(td
->dm_dev
.bdev
);
697 bdev
= blkdev_get_by_dev(dev
, td
->dm_dev
.mode
| FMODE_EXCL
, _claim_ptr
);
699 return PTR_ERR(bdev
);
701 r
= bd_link_disk_holder(bdev
, dm_disk(md
));
703 blkdev_put(bdev
, td
->dm_dev
.mode
| FMODE_EXCL
);
707 td
->dm_dev
.bdev
= bdev
;
712 * Close a table device that we've been using.
714 static void close_table_device(struct table_device
*td
, struct mapped_device
*md
)
716 if (!td
->dm_dev
.bdev
)
719 bd_unlink_disk_holder(td
->dm_dev
.bdev
, dm_disk(md
));
720 blkdev_put(td
->dm_dev
.bdev
, td
->dm_dev
.mode
| FMODE_EXCL
);
721 td
->dm_dev
.bdev
= NULL
;
724 static struct table_device
*find_table_device(struct list_head
*l
, dev_t dev
,
726 struct table_device
*td
;
728 list_for_each_entry(td
, l
, list
)
729 if (td
->dm_dev
.bdev
->bd_dev
== dev
&& td
->dm_dev
.mode
== mode
)
735 int dm_get_table_device(struct mapped_device
*md
, dev_t dev
, fmode_t mode
,
736 struct dm_dev
**result
) {
738 struct table_device
*td
;
740 mutex_lock(&md
->table_devices_lock
);
741 td
= find_table_device(&md
->table_devices
, dev
, mode
);
743 td
= kmalloc(sizeof(*td
), GFP_KERNEL
);
745 mutex_unlock(&md
->table_devices_lock
);
749 td
->dm_dev
.mode
= mode
;
750 td
->dm_dev
.bdev
= NULL
;
752 if ((r
= open_table_device(td
, dev
, md
))) {
753 mutex_unlock(&md
->table_devices_lock
);
758 format_dev_t(td
->dm_dev
.name
, dev
);
760 atomic_set(&td
->count
, 0);
761 list_add(&td
->list
, &md
->table_devices
);
763 atomic_inc(&td
->count
);
764 mutex_unlock(&md
->table_devices_lock
);
766 *result
= &td
->dm_dev
;
769 EXPORT_SYMBOL_GPL(dm_get_table_device
);
771 void dm_put_table_device(struct mapped_device
*md
, struct dm_dev
*d
)
773 struct table_device
*td
= container_of(d
, struct table_device
, dm_dev
);
775 mutex_lock(&md
->table_devices_lock
);
776 if (atomic_dec_and_test(&td
->count
)) {
777 close_table_device(td
, md
);
781 mutex_unlock(&md
->table_devices_lock
);
783 EXPORT_SYMBOL(dm_put_table_device
);
785 static void free_table_devices(struct list_head
*devices
)
787 struct list_head
*tmp
, *next
;
789 list_for_each_safe(tmp
, next
, devices
) {
790 struct table_device
*td
= list_entry(tmp
, struct table_device
, list
);
792 DMWARN("dm_destroy: %s still exists with %d references",
793 td
->dm_dev
.name
, atomic_read(&td
->count
));
799 * Get the geometry associated with a dm device
801 int dm_get_geometry(struct mapped_device
*md
, struct hd_geometry
*geo
)
809 * Set the geometry of a device.
811 int dm_set_geometry(struct mapped_device
*md
, struct hd_geometry
*geo
)
813 sector_t sz
= (sector_t
)geo
->cylinders
* geo
->heads
* geo
->sectors
;
815 if (geo
->start
> sz
) {
816 DMWARN("Start sector is beyond the geometry limits.");
825 /*-----------------------------------------------------------------
827 * A more elegant soln is in the works that uses the queue
828 * merge fn, unfortunately there are a couple of changes to
829 * the block layer that I want to make for this. So in the
830 * interests of getting something for people to use I give
831 * you this clearly demarcated crap.
832 *---------------------------------------------------------------*/
834 static int __noflush_suspending(struct mapped_device
*md
)
836 return test_bit(DMF_NOFLUSH_SUSPENDING
, &md
->flags
);
840 * Decrements the number of outstanding ios that a bio has been
841 * cloned into, completing the original io if necc.
843 static void dec_pending(struct dm_io
*io
, int error
)
848 struct mapped_device
*md
= io
->md
;
850 /* Push-back supersedes any I/O errors */
851 if (unlikely(error
)) {
852 spin_lock_irqsave(&io
->endio_lock
, flags
);
853 if (!(io
->error
> 0 && __noflush_suspending(md
)))
855 spin_unlock_irqrestore(&io
->endio_lock
, flags
);
858 if (atomic_dec_and_test(&io
->io_count
)) {
859 if (io
->error
== DM_ENDIO_REQUEUE
) {
861 * Target requested pushing back the I/O.
863 spin_lock_irqsave(&md
->deferred_lock
, flags
);
864 if (__noflush_suspending(md
))
865 bio_list_add_head(&md
->deferred
, io
->bio
);
867 /* noflush suspend was interrupted. */
869 spin_unlock_irqrestore(&md
->deferred_lock
, flags
);
872 io_error
= io
->error
;
877 if (io_error
== DM_ENDIO_REQUEUE
)
880 if ((bio
->bi_rw
& REQ_FLUSH
) && bio
->bi_iter
.bi_size
) {
882 * Preflush done for flush with data, reissue
885 bio
->bi_rw
&= ~REQ_FLUSH
;
888 /* done with normal IO or empty flush */
889 trace_block_bio_complete(md
->queue
, bio
, io_error
);
890 bio_endio(bio
, io_error
);
895 static void disable_write_same(struct mapped_device
*md
)
897 struct queue_limits
*limits
= dm_get_queue_limits(md
);
899 /* device doesn't really support WRITE SAME, disable it */
900 limits
->max_write_same_sectors
= 0;
903 static void clone_endio(struct bio
*bio
, int error
)
906 struct dm_target_io
*tio
= container_of(bio
, struct dm_target_io
, clone
);
907 struct dm_io
*io
= tio
->io
;
908 struct mapped_device
*md
= tio
->io
->md
;
909 dm_endio_fn endio
= tio
->ti
->type
->end_io
;
911 if (!bio_flagged(bio
, BIO_UPTODATE
) && !error
)
915 r
= endio(tio
->ti
, bio
, error
);
916 if (r
< 0 || r
== DM_ENDIO_REQUEUE
)
918 * error and requeue request are handled
922 else if (r
== DM_ENDIO_INCOMPLETE
)
923 /* The target will handle the io */
926 DMWARN("unimplemented target endio return value: %d", r
);
931 if (unlikely(r
== -EREMOTEIO
&& (bio
->bi_rw
& REQ_WRITE_SAME
) &&
932 !bdev_get_queue(bio
->bi_bdev
)->limits
.max_write_same_sectors
))
933 disable_write_same(md
);
936 dec_pending(io
, error
);
940 * Partial completion handling for request-based dm
942 static void end_clone_bio(struct bio
*clone
, int error
)
944 struct dm_rq_clone_bio_info
*info
=
945 container_of(clone
, struct dm_rq_clone_bio_info
, clone
);
946 struct dm_rq_target_io
*tio
= info
->tio
;
947 struct bio
*bio
= info
->orig
;
948 unsigned int nr_bytes
= info
->orig
->bi_iter
.bi_size
;
954 * An error has already been detected on the request.
955 * Once error occurred, just let clone->end_io() handle
961 * Don't notice the error to the upper layer yet.
962 * The error handling decision is made by the target driver,
963 * when the request is completed.
970 * I/O for the bio successfully completed.
971 * Notice the data completion to the upper layer.
975 * bios are processed from the head of the list.
976 * So the completing bio should always be rq->bio.
977 * If it's not, something wrong is happening.
979 if (tio
->orig
->bio
!= bio
)
980 DMERR("bio completion is going in the middle of the request");
983 * Update the original request.
984 * Do not use blk_end_request() here, because it may complete
985 * the original request before the clone, and break the ordering.
987 blk_update_request(tio
->orig
, 0, nr_bytes
);
991 * Don't touch any member of the md after calling this function because
992 * the md may be freed in dm_put() at the end of this function.
993 * Or do dm_get() before calling this function and dm_put() later.
995 static void rq_completed(struct mapped_device
*md
, int rw
, int run_queue
)
997 atomic_dec(&md
->pending
[rw
]);
999 /* nudge anyone waiting on suspend queue */
1000 if (!md_in_flight(md
))
1004 * Run this off this callpath, as drivers could invoke end_io while
1005 * inside their request_fn (and holding the queue lock). Calling
1006 * back into ->request_fn() could deadlock attempting to grab the
1010 blk_run_queue_async(md
->queue
);
1013 * dm_put() must be at the end of this function. See the comment above
1018 static void free_rq_clone(struct request
*clone
)
1020 struct dm_rq_target_io
*tio
= clone
->end_io_data
;
1022 blk_rq_unprep_clone(clone
);
1027 * Complete the clone and the original request.
1028 * Must be called without queue lock.
1030 static void dm_end_request(struct request
*clone
, int error
)
1032 int rw
= rq_data_dir(clone
);
1033 struct dm_rq_target_io
*tio
= clone
->end_io_data
;
1034 struct mapped_device
*md
= tio
->md
;
1035 struct request
*rq
= tio
->orig
;
1037 if (rq
->cmd_type
== REQ_TYPE_BLOCK_PC
) {
1038 rq
->errors
= clone
->errors
;
1039 rq
->resid_len
= clone
->resid_len
;
1043 * We are using the sense buffer of the original
1045 * So setting the length of the sense data is enough.
1047 rq
->sense_len
= clone
->sense_len
;
1050 free_rq_clone(clone
);
1051 blk_end_request_all(rq
, error
);
1052 rq_completed(md
, rw
, true);
1055 static void dm_unprep_request(struct request
*rq
)
1057 struct request
*clone
= rq
->special
;
1060 rq
->cmd_flags
&= ~REQ_DONTPREP
;
1062 free_rq_clone(clone
);
1066 * Requeue the original request of a clone.
1068 void dm_requeue_unmapped_request(struct request
*clone
)
1070 int rw
= rq_data_dir(clone
);
1071 struct dm_rq_target_io
*tio
= clone
->end_io_data
;
1072 struct mapped_device
*md
= tio
->md
;
1073 struct request
*rq
= tio
->orig
;
1074 struct request_queue
*q
= rq
->q
;
1075 unsigned long flags
;
1077 dm_unprep_request(rq
);
1079 spin_lock_irqsave(q
->queue_lock
, flags
);
1080 blk_requeue_request(q
, rq
);
1081 spin_unlock_irqrestore(q
->queue_lock
, flags
);
1083 rq_completed(md
, rw
, 0);
1085 EXPORT_SYMBOL_GPL(dm_requeue_unmapped_request
);
1087 static void __stop_queue(struct request_queue
*q
)
1092 static void stop_queue(struct request_queue
*q
)
1094 unsigned long flags
;
1096 spin_lock_irqsave(q
->queue_lock
, flags
);
1098 spin_unlock_irqrestore(q
->queue_lock
, flags
);
1101 static void __start_queue(struct request_queue
*q
)
1103 if (blk_queue_stopped(q
))
1107 static void start_queue(struct request_queue
*q
)
1109 unsigned long flags
;
1111 spin_lock_irqsave(q
->queue_lock
, flags
);
1113 spin_unlock_irqrestore(q
->queue_lock
, flags
);
1116 static void dm_done(struct request
*clone
, int error
, bool mapped
)
1119 struct dm_rq_target_io
*tio
= clone
->end_io_data
;
1120 dm_request_endio_fn rq_end_io
= NULL
;
1123 rq_end_io
= tio
->ti
->type
->rq_end_io
;
1125 if (mapped
&& rq_end_io
)
1126 r
= rq_end_io(tio
->ti
, clone
, error
, &tio
->info
);
1129 if (unlikely(r
== -EREMOTEIO
&& (clone
->cmd_flags
& REQ_WRITE_SAME
) &&
1130 !clone
->q
->limits
.max_write_same_sectors
))
1131 disable_write_same(tio
->md
);
1134 /* The target wants to complete the I/O */
1135 dm_end_request(clone
, r
);
1136 else if (r
== DM_ENDIO_INCOMPLETE
)
1137 /* The target will handle the I/O */
1139 else if (r
== DM_ENDIO_REQUEUE
)
1140 /* The target wants to requeue the I/O */
1141 dm_requeue_unmapped_request(clone
);
1143 DMWARN("unimplemented target endio return value: %d", r
);
1149 * Request completion handler for request-based dm
1151 static void dm_softirq_done(struct request
*rq
)
1154 struct request
*clone
= rq
->completion_data
;
1155 struct dm_rq_target_io
*tio
= clone
->end_io_data
;
1157 if (rq
->cmd_flags
& REQ_FAILED
)
1160 dm_done(clone
, tio
->error
, mapped
);
1164 * Complete the clone and the original request with the error status
1165 * through softirq context.
1167 static void dm_complete_request(struct request
*clone
, int error
)
1169 struct dm_rq_target_io
*tio
= clone
->end_io_data
;
1170 struct request
*rq
= tio
->orig
;
1173 rq
->completion_data
= clone
;
1174 blk_complete_request(rq
);
1178 * Complete the not-mapped clone and the original request with the error status
1179 * through softirq context.
1180 * Target's rq_end_io() function isn't called.
1181 * This may be used when the target's map_rq() function fails.
1183 void dm_kill_unmapped_request(struct request
*clone
, int error
)
1185 struct dm_rq_target_io
*tio
= clone
->end_io_data
;
1186 struct request
*rq
= tio
->orig
;
1188 rq
->cmd_flags
|= REQ_FAILED
;
1189 dm_complete_request(clone
, error
);
1191 EXPORT_SYMBOL_GPL(dm_kill_unmapped_request
);
1194 * Called with the queue lock held
1196 static void end_clone_request(struct request
*clone
, int error
)
1199 * For just cleaning up the information of the queue in which
1200 * the clone was dispatched.
1201 * The clone is *NOT* freed actually here because it is alloced from
1202 * dm own mempool and REQ_ALLOCED isn't set in clone->cmd_flags.
1204 __blk_put_request(clone
->q
, clone
);
1207 * Actual request completion is done in a softirq context which doesn't
1208 * hold the queue lock. Otherwise, deadlock could occur because:
1209 * - another request may be submitted by the upper level driver
1210 * of the stacking during the completion
1211 * - the submission which requires queue lock may be done
1212 * against this queue
1214 dm_complete_request(clone
, error
);
1218 * Return maximum size of I/O possible at the supplied sector up to the current
1221 static sector_t
max_io_len_target_boundary(sector_t sector
, struct dm_target
*ti
)
1223 sector_t target_offset
= dm_target_offset(ti
, sector
);
1225 return ti
->len
- target_offset
;
1228 static sector_t
max_io_len(sector_t sector
, struct dm_target
*ti
)
1230 sector_t len
= max_io_len_target_boundary(sector
, ti
);
1231 sector_t offset
, max_len
;
1234 * Does the target need to split even further?
1236 if (ti
->max_io_len
) {
1237 offset
= dm_target_offset(ti
, sector
);
1238 if (unlikely(ti
->max_io_len
& (ti
->max_io_len
- 1)))
1239 max_len
= sector_div(offset
, ti
->max_io_len
);
1241 max_len
= offset
& (ti
->max_io_len
- 1);
1242 max_len
= ti
->max_io_len
- max_len
;
1251 int dm_set_target_max_io_len(struct dm_target
*ti
, sector_t len
)
1253 if (len
> UINT_MAX
) {
1254 DMERR("Specified maximum size of target IO (%llu) exceeds limit (%u)",
1255 (unsigned long long)len
, UINT_MAX
);
1256 ti
->error
= "Maximum size of target IO is too large";
1260 ti
->max_io_len
= (uint32_t) len
;
1264 EXPORT_SYMBOL_GPL(dm_set_target_max_io_len
);
1267 * A target may call dm_accept_partial_bio only from the map routine. It is
1268 * allowed for all bio types except REQ_FLUSH.
1270 * dm_accept_partial_bio informs the dm that the target only wants to process
1271 * additional n_sectors sectors of the bio and the rest of the data should be
1272 * sent in a next bio.
1274 * A diagram that explains the arithmetics:
1275 * +--------------------+---------------+-------+
1277 * +--------------------+---------------+-------+
1279 * <-------------- *tio->len_ptr --------------->
1280 * <------- bi_size ------->
1283 * Region 1 was already iterated over with bio_advance or similar function.
1284 * (it may be empty if the target doesn't use bio_advance)
1285 * Region 2 is the remaining bio size that the target wants to process.
1286 * (it may be empty if region 1 is non-empty, although there is no reason
1288 * The target requires that region 3 is to be sent in the next bio.
1290 * If the target wants to receive multiple copies of the bio (via num_*bios, etc),
1291 * the partially processed part (the sum of regions 1+2) must be the same for all
1292 * copies of the bio.
1294 void dm_accept_partial_bio(struct bio
*bio
, unsigned n_sectors
)
1296 struct dm_target_io
*tio
= container_of(bio
, struct dm_target_io
, clone
);
1297 unsigned bi_size
= bio
->bi_iter
.bi_size
>> SECTOR_SHIFT
;
1298 BUG_ON(bio
->bi_rw
& REQ_FLUSH
);
1299 BUG_ON(bi_size
> *tio
->len_ptr
);
1300 BUG_ON(n_sectors
> bi_size
);
1301 *tio
->len_ptr
-= bi_size
- n_sectors
;
1302 bio
->bi_iter
.bi_size
= n_sectors
<< SECTOR_SHIFT
;
1304 EXPORT_SYMBOL_GPL(dm_accept_partial_bio
);
1306 static void __map_bio(struct dm_target_io
*tio
)
1310 struct mapped_device
*md
;
1311 struct bio
*clone
= &tio
->clone
;
1312 struct dm_target
*ti
= tio
->ti
;
1314 clone
->bi_end_io
= clone_endio
;
1317 * Map the clone. If r == 0 we don't need to do
1318 * anything, the target has assumed ownership of
1321 atomic_inc(&tio
->io
->io_count
);
1322 sector
= clone
->bi_iter
.bi_sector
;
1323 r
= ti
->type
->map(ti
, clone
);
1324 if (r
== DM_MAPIO_REMAPPED
) {
1325 /* the bio has been remapped so dispatch it */
1327 trace_block_bio_remap(bdev_get_queue(clone
->bi_bdev
), clone
,
1328 tio
->io
->bio
->bi_bdev
->bd_dev
, sector
);
1330 generic_make_request(clone
);
1331 } else if (r
< 0 || r
== DM_MAPIO_REQUEUE
) {
1332 /* error the io and bail out, or requeue it if needed */
1334 dec_pending(tio
->io
, r
);
1337 DMWARN("unimplemented target map return value: %d", r
);
1343 struct mapped_device
*md
;
1344 struct dm_table
*map
;
1348 unsigned sector_count
;
1351 static void bio_setup_sector(struct bio
*bio
, sector_t sector
, unsigned len
)
1353 bio
->bi_iter
.bi_sector
= sector
;
1354 bio
->bi_iter
.bi_size
= to_bytes(len
);
1358 * Creates a bio that consists of range of complete bvecs.
1360 static void clone_bio(struct dm_target_io
*tio
, struct bio
*bio
,
1361 sector_t sector
, unsigned len
)
1363 struct bio
*clone
= &tio
->clone
;
1365 __bio_clone_fast(clone
, bio
);
1367 if (bio_integrity(bio
))
1368 bio_integrity_clone(clone
, bio
, GFP_NOIO
);
1370 bio_advance(clone
, to_bytes(sector
- clone
->bi_iter
.bi_sector
));
1371 clone
->bi_iter
.bi_size
= to_bytes(len
);
1373 if (bio_integrity(bio
))
1374 bio_integrity_trim(clone
, 0, len
);
1377 static struct dm_target_io
*alloc_tio(struct clone_info
*ci
,
1378 struct dm_target
*ti
,
1379 unsigned target_bio_nr
)
1381 struct dm_target_io
*tio
;
1384 clone
= bio_alloc_bioset(GFP_NOIO
, 0, ci
->md
->bs
);
1385 tio
= container_of(clone
, struct dm_target_io
, clone
);
1389 tio
->target_bio_nr
= target_bio_nr
;
1394 static void __clone_and_map_simple_bio(struct clone_info
*ci
,
1395 struct dm_target
*ti
,
1396 unsigned target_bio_nr
, unsigned *len
)
1398 struct dm_target_io
*tio
= alloc_tio(ci
, ti
, target_bio_nr
);
1399 struct bio
*clone
= &tio
->clone
;
1403 __bio_clone_fast(clone
, ci
->bio
);
1405 bio_setup_sector(clone
, ci
->sector
, *len
);
1410 static void __send_duplicate_bios(struct clone_info
*ci
, struct dm_target
*ti
,
1411 unsigned num_bios
, unsigned *len
)
1413 unsigned target_bio_nr
;
1415 for (target_bio_nr
= 0; target_bio_nr
< num_bios
; target_bio_nr
++)
1416 __clone_and_map_simple_bio(ci
, ti
, target_bio_nr
, len
);
1419 static int __send_empty_flush(struct clone_info
*ci
)
1421 unsigned target_nr
= 0;
1422 struct dm_target
*ti
;
1424 BUG_ON(bio_has_data(ci
->bio
));
1425 while ((ti
= dm_table_get_target(ci
->map
, target_nr
++)))
1426 __send_duplicate_bios(ci
, ti
, ti
->num_flush_bios
, NULL
);
1431 static void __clone_and_map_data_bio(struct clone_info
*ci
, struct dm_target
*ti
,
1432 sector_t sector
, unsigned *len
)
1434 struct bio
*bio
= ci
->bio
;
1435 struct dm_target_io
*tio
;
1436 unsigned target_bio_nr
;
1437 unsigned num_target_bios
= 1;
1440 * Does the target want to receive duplicate copies of the bio?
1442 if (bio_data_dir(bio
) == WRITE
&& ti
->num_write_bios
)
1443 num_target_bios
= ti
->num_write_bios(ti
, bio
);
1445 for (target_bio_nr
= 0; target_bio_nr
< num_target_bios
; target_bio_nr
++) {
1446 tio
= alloc_tio(ci
, ti
, target_bio_nr
);
1448 clone_bio(tio
, bio
, sector
, *len
);
1453 typedef unsigned (*get_num_bios_fn
)(struct dm_target
*ti
);
1455 static unsigned get_num_discard_bios(struct dm_target
*ti
)
1457 return ti
->num_discard_bios
;
1460 static unsigned get_num_write_same_bios(struct dm_target
*ti
)
1462 return ti
->num_write_same_bios
;
1465 typedef bool (*is_split_required_fn
)(struct dm_target
*ti
);
1467 static bool is_split_required_for_discard(struct dm_target
*ti
)
1469 return ti
->split_discard_bios
;
1472 static int __send_changing_extent_only(struct clone_info
*ci
,
1473 get_num_bios_fn get_num_bios
,
1474 is_split_required_fn is_split_required
)
1476 struct dm_target
*ti
;
1481 ti
= dm_table_find_target(ci
->map
, ci
->sector
);
1482 if (!dm_target_is_valid(ti
))
1486 * Even though the device advertised support for this type of
1487 * request, that does not mean every target supports it, and
1488 * reconfiguration might also have changed that since the
1489 * check was performed.
1491 num_bios
= get_num_bios
? get_num_bios(ti
) : 0;
1495 if (is_split_required
&& !is_split_required(ti
))
1496 len
= min((sector_t
)ci
->sector_count
, max_io_len_target_boundary(ci
->sector
, ti
));
1498 len
= min((sector_t
)ci
->sector_count
, max_io_len(ci
->sector
, ti
));
1500 __send_duplicate_bios(ci
, ti
, num_bios
, &len
);
1503 } while (ci
->sector_count
-= len
);
1508 static int __send_discard(struct clone_info
*ci
)
1510 return __send_changing_extent_only(ci
, get_num_discard_bios
,
1511 is_split_required_for_discard
);
1514 static int __send_write_same(struct clone_info
*ci
)
1516 return __send_changing_extent_only(ci
, get_num_write_same_bios
, NULL
);
1520 * Select the correct strategy for processing a non-flush bio.
1522 static int __split_and_process_non_flush(struct clone_info
*ci
)
1524 struct bio
*bio
= ci
->bio
;
1525 struct dm_target
*ti
;
1528 if (unlikely(bio
->bi_rw
& REQ_DISCARD
))
1529 return __send_discard(ci
);
1530 else if (unlikely(bio
->bi_rw
& REQ_WRITE_SAME
))
1531 return __send_write_same(ci
);
1533 ti
= dm_table_find_target(ci
->map
, ci
->sector
);
1534 if (!dm_target_is_valid(ti
))
1537 len
= min_t(sector_t
, max_io_len(ci
->sector
, ti
), ci
->sector_count
);
1539 __clone_and_map_data_bio(ci
, ti
, ci
->sector
, &len
);
1542 ci
->sector_count
-= len
;
1548 * Entry point to split a bio into clones and submit them to the targets.
1550 static void __split_and_process_bio(struct mapped_device
*md
,
1551 struct dm_table
*map
, struct bio
*bio
)
1553 struct clone_info ci
;
1556 if (unlikely(!map
)) {
1563 ci
.io
= alloc_io(md
);
1565 atomic_set(&ci
.io
->io_count
, 1);
1568 spin_lock_init(&ci
.io
->endio_lock
);
1569 ci
.sector
= bio
->bi_iter
.bi_sector
;
1571 start_io_acct(ci
.io
);
1573 if (bio
->bi_rw
& REQ_FLUSH
) {
1574 ci
.bio
= &ci
.md
->flush_bio
;
1575 ci
.sector_count
= 0;
1576 error
= __send_empty_flush(&ci
);
1577 /* dec_pending submits any data associated with flush */
1580 ci
.sector_count
= bio_sectors(bio
);
1581 while (ci
.sector_count
&& !error
)
1582 error
= __split_and_process_non_flush(&ci
);
1585 /* drop the extra reference count */
1586 dec_pending(ci
.io
, error
);
1588 /*-----------------------------------------------------------------
1590 *---------------------------------------------------------------*/
1592 static int dm_merge_bvec(struct request_queue
*q
,
1593 struct bvec_merge_data
*bvm
,
1594 struct bio_vec
*biovec
)
1596 struct mapped_device
*md
= q
->queuedata
;
1597 struct dm_table
*map
= dm_get_live_table_fast(md
);
1598 struct dm_target
*ti
;
1599 sector_t max_sectors
;
1605 ti
= dm_table_find_target(map
, bvm
->bi_sector
);
1606 if (!dm_target_is_valid(ti
))
1610 * Find maximum amount of I/O that won't need splitting
1612 max_sectors
= min(max_io_len(bvm
->bi_sector
, ti
),
1613 (sector_t
) queue_max_sectors(q
));
1614 max_size
= (max_sectors
<< SECTOR_SHIFT
) - bvm
->bi_size
;
1615 if (unlikely(max_size
< 0)) /* this shouldn't _ever_ happen */
1619 * merge_bvec_fn() returns number of bytes
1620 * it can accept at this offset
1621 * max is precomputed maximal io size
1623 if (max_size
&& ti
->type
->merge
)
1624 max_size
= ti
->type
->merge(ti
, bvm
, biovec
, max_size
);
1626 * If the target doesn't support merge method and some of the devices
1627 * provided their merge_bvec method (we know this by looking for the
1628 * max_hw_sectors that dm_set_device_limits may set), then we can't
1629 * allow bios with multiple vector entries. So always set max_size
1630 * to 0, and the code below allows just one page.
1632 else if (queue_max_hw_sectors(q
) <= PAGE_SIZE
>> 9)
1636 dm_put_live_table_fast(md
);
1638 * Always allow an entire first page
1640 if (max_size
<= biovec
->bv_len
&& !(bvm
->bi_size
>> SECTOR_SHIFT
))
1641 max_size
= biovec
->bv_len
;
1647 * The request function that just remaps the bio built up by
1650 static void _dm_request(struct request_queue
*q
, struct bio
*bio
)
1652 int rw
= bio_data_dir(bio
);
1653 struct mapped_device
*md
= q
->queuedata
;
1655 struct dm_table
*map
;
1657 map
= dm_get_live_table(md
, &srcu_idx
);
1659 generic_start_io_acct(rw
, bio_sectors(bio
), &dm_disk(md
)->part0
);
1661 /* if we're suspended, we have to queue this io for later */
1662 if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND
, &md
->flags
))) {
1663 dm_put_live_table(md
, srcu_idx
);
1665 if (bio_rw(bio
) != READA
)
1672 __split_and_process_bio(md
, map
, bio
);
1673 dm_put_live_table(md
, srcu_idx
);
1677 int dm_request_based(struct mapped_device
*md
)
1679 return blk_queue_stackable(md
->queue
);
1682 static void dm_request(struct request_queue
*q
, struct bio
*bio
)
1684 struct mapped_device
*md
= q
->queuedata
;
1686 if (dm_request_based(md
))
1687 blk_queue_bio(q
, bio
);
1689 _dm_request(q
, bio
);
1692 void dm_dispatch_request(struct request
*rq
)
1696 if (blk_queue_io_stat(rq
->q
))
1697 rq
->cmd_flags
|= REQ_IO_STAT
;
1699 rq
->start_time
= jiffies
;
1700 r
= blk_insert_cloned_request(rq
->q
, rq
);
1702 dm_complete_request(rq
, r
);
1704 EXPORT_SYMBOL_GPL(dm_dispatch_request
);
1706 static int dm_rq_bio_constructor(struct bio
*bio
, struct bio
*bio_orig
,
1709 struct dm_rq_target_io
*tio
= data
;
1710 struct dm_rq_clone_bio_info
*info
=
1711 container_of(bio
, struct dm_rq_clone_bio_info
, clone
);
1713 info
->orig
= bio_orig
;
1715 bio
->bi_end_io
= end_clone_bio
;
1720 static int setup_clone(struct request
*clone
, struct request
*rq
,
1721 struct dm_rq_target_io
*tio
)
1725 blk_rq_init(NULL
, clone
);
1726 r
= blk_rq_prep_clone(clone
, rq
, tio
->md
->bs
, GFP_ATOMIC
,
1727 dm_rq_bio_constructor
, tio
);
1731 clone
->cmd
= rq
->cmd
;
1732 clone
->cmd_len
= rq
->cmd_len
;
1733 clone
->sense
= rq
->sense
;
1734 clone
->end_io
= end_clone_request
;
1735 clone
->end_io_data
= tio
;
1740 static struct request
*clone_rq(struct request
*rq
, struct mapped_device
*md
,
1743 struct request
*clone
;
1744 struct dm_rq_target_io
*tio
;
1746 tio
= alloc_rq_tio(md
, gfp_mask
);
1754 memset(&tio
->info
, 0, sizeof(tio
->info
));
1756 clone
= &tio
->clone
;
1757 if (setup_clone(clone
, rq
, tio
)) {
1767 * Called with the queue lock held.
1769 static int dm_prep_fn(struct request_queue
*q
, struct request
*rq
)
1771 struct mapped_device
*md
= q
->queuedata
;
1772 struct request
*clone
;
1774 if (unlikely(rq
->special
)) {
1775 DMWARN("Already has something in rq->special.");
1776 return BLKPREP_KILL
;
1779 clone
= clone_rq(rq
, md
, GFP_ATOMIC
);
1781 return BLKPREP_DEFER
;
1783 rq
->special
= clone
;
1784 rq
->cmd_flags
|= REQ_DONTPREP
;
1791 * 0 : the request has been processed (not requeued)
1792 * !0 : the request has been requeued
1794 static int map_request(struct dm_target
*ti
, struct request
*clone
,
1795 struct mapped_device
*md
)
1797 int r
, requeued
= 0;
1798 struct dm_rq_target_io
*tio
= clone
->end_io_data
;
1801 r
= ti
->type
->map_rq(ti
, clone
, &tio
->info
);
1803 case DM_MAPIO_SUBMITTED
:
1804 /* The target has taken the I/O to submit by itself later */
1806 case DM_MAPIO_REMAPPED
:
1807 /* The target has remapped the I/O so dispatch it */
1808 trace_block_rq_remap(clone
->q
, clone
, disk_devt(dm_disk(md
)),
1809 blk_rq_pos(tio
->orig
));
1810 dm_dispatch_request(clone
);
1812 case DM_MAPIO_REQUEUE
:
1813 /* The target wants to requeue the I/O */
1814 dm_requeue_unmapped_request(clone
);
1819 DMWARN("unimplemented target map return value: %d", r
);
1823 /* The target wants to complete the I/O */
1824 dm_kill_unmapped_request(clone
, r
);
1831 static struct request
*dm_start_request(struct mapped_device
*md
, struct request
*orig
)
1833 struct request
*clone
;
1835 blk_start_request(orig
);
1836 clone
= orig
->special
;
1837 atomic_inc(&md
->pending
[rq_data_dir(clone
)]);
1840 * Hold the md reference here for the in-flight I/O.
1841 * We can't rely on the reference count by device opener,
1842 * because the device may be closed during the request completion
1843 * when all bios are completed.
1844 * See the comment in rq_completed() too.
1852 * q->request_fn for request-based dm.
1853 * Called with the queue lock held.
1855 static void dm_request_fn(struct request_queue
*q
)
1857 struct mapped_device
*md
= q
->queuedata
;
1859 struct dm_table
*map
= dm_get_live_table(md
, &srcu_idx
);
1860 struct dm_target
*ti
;
1861 struct request
*rq
, *clone
;
1865 * For suspend, check blk_queue_stopped() and increment
1866 * ->pending within a single queue_lock not to increment the
1867 * number of in-flight I/Os after the queue is stopped in
1870 while (!blk_queue_stopped(q
)) {
1871 rq
= blk_peek_request(q
);
1875 /* always use block 0 to find the target for flushes for now */
1877 if (!(rq
->cmd_flags
& REQ_FLUSH
))
1878 pos
= blk_rq_pos(rq
);
1880 ti
= dm_table_find_target(map
, pos
);
1881 if (!dm_target_is_valid(ti
)) {
1883 * Must perform setup, that dm_done() requires,
1884 * before calling dm_kill_unmapped_request
1886 DMERR_LIMIT("request attempted access beyond the end of device");
1887 clone
= dm_start_request(md
, rq
);
1888 dm_kill_unmapped_request(clone
, -EIO
);
1892 if (ti
->type
->busy
&& ti
->type
->busy(ti
))
1895 clone
= dm_start_request(md
, rq
);
1897 spin_unlock(q
->queue_lock
);
1898 if (map_request(ti
, clone
, md
))
1901 BUG_ON(!irqs_disabled());
1902 spin_lock(q
->queue_lock
);
1908 BUG_ON(!irqs_disabled());
1909 spin_lock(q
->queue_lock
);
1912 blk_delay_queue(q
, HZ
/ 10);
1914 dm_put_live_table(md
, srcu_idx
);
1917 int dm_underlying_device_busy(struct request_queue
*q
)
1919 return blk_lld_busy(q
);
1921 EXPORT_SYMBOL_GPL(dm_underlying_device_busy
);
1923 static int dm_lld_busy(struct request_queue
*q
)
1926 struct mapped_device
*md
= q
->queuedata
;
1927 struct dm_table
*map
= dm_get_live_table_fast(md
);
1929 if (!map
|| test_bit(DMF_BLOCK_IO_FOR_SUSPEND
, &md
->flags
))
1932 r
= dm_table_any_busy_target(map
);
1934 dm_put_live_table_fast(md
);
1939 static int dm_any_congested(void *congested_data
, int bdi_bits
)
1942 struct mapped_device
*md
= congested_data
;
1943 struct dm_table
*map
;
1945 if (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND
, &md
->flags
)) {
1946 map
= dm_get_live_table_fast(md
);
1949 * Request-based dm cares about only own queue for
1950 * the query about congestion status of request_queue
1952 if (dm_request_based(md
))
1953 r
= md
->queue
->backing_dev_info
.state
&
1956 r
= dm_table_any_congested(map
, bdi_bits
);
1958 dm_put_live_table_fast(md
);
1964 /*-----------------------------------------------------------------
1965 * An IDR is used to keep track of allocated minor numbers.
1966 *---------------------------------------------------------------*/
1967 static void free_minor(int minor
)
1969 spin_lock(&_minor_lock
);
1970 idr_remove(&_minor_idr
, minor
);
1971 spin_unlock(&_minor_lock
);
1975 * See if the device with a specific minor # is free.
1977 static int specific_minor(int minor
)
1981 if (minor
>= (1 << MINORBITS
))
1984 idr_preload(GFP_KERNEL
);
1985 spin_lock(&_minor_lock
);
1987 r
= idr_alloc(&_minor_idr
, MINOR_ALLOCED
, minor
, minor
+ 1, GFP_NOWAIT
);
1989 spin_unlock(&_minor_lock
);
1992 return r
== -ENOSPC
? -EBUSY
: r
;
1996 static int next_free_minor(int *minor
)
2000 idr_preload(GFP_KERNEL
);
2001 spin_lock(&_minor_lock
);
2003 r
= idr_alloc(&_minor_idr
, MINOR_ALLOCED
, 0, 1 << MINORBITS
, GFP_NOWAIT
);
2005 spin_unlock(&_minor_lock
);
2013 static const struct block_device_operations dm_blk_dops
;
2015 static void dm_wq_work(struct work_struct
*work
);
2017 static void dm_init_md_queue(struct mapped_device
*md
)
2020 * Request-based dm devices cannot be stacked on top of bio-based dm
2021 * devices. The type of this dm device has not been decided yet.
2022 * The type is decided at the first table loading time.
2023 * To prevent problematic device stacking, clear the queue flag
2024 * for request stacking support until then.
2026 * This queue is new, so no concurrency on the queue_flags.
2028 queue_flag_clear_unlocked(QUEUE_FLAG_STACKABLE
, md
->queue
);
2030 md
->queue
->queuedata
= md
;
2031 md
->queue
->backing_dev_info
.congested_fn
= dm_any_congested
;
2032 md
->queue
->backing_dev_info
.congested_data
= md
;
2033 blk_queue_make_request(md
->queue
, dm_request
);
2034 blk_queue_bounce_limit(md
->queue
, BLK_BOUNCE_ANY
);
2035 blk_queue_merge_bvec(md
->queue
, dm_merge_bvec
);
2039 * Allocate and initialise a blank device with a given minor.
2041 static struct mapped_device
*alloc_dev(int minor
)
2044 struct mapped_device
*md
= kzalloc(sizeof(*md
), GFP_KERNEL
);
2048 DMWARN("unable to allocate device, out of memory.");
2052 if (!try_module_get(THIS_MODULE
))
2053 goto bad_module_get
;
2055 /* get a minor number for the dev */
2056 if (minor
== DM_ANY_MINOR
)
2057 r
= next_free_minor(&minor
);
2059 r
= specific_minor(minor
);
2063 r
= init_srcu_struct(&md
->io_barrier
);
2065 goto bad_io_barrier
;
2067 md
->type
= DM_TYPE_NONE
;
2068 mutex_init(&md
->suspend_lock
);
2069 mutex_init(&md
->type_lock
);
2070 mutex_init(&md
->table_devices_lock
);
2071 spin_lock_init(&md
->deferred_lock
);
2072 atomic_set(&md
->holders
, 1);
2073 atomic_set(&md
->open_count
, 0);
2074 atomic_set(&md
->event_nr
, 0);
2075 atomic_set(&md
->uevent_seq
, 0);
2076 INIT_LIST_HEAD(&md
->uevent_list
);
2077 INIT_LIST_HEAD(&md
->table_devices
);
2078 spin_lock_init(&md
->uevent_lock
);
2080 md
->queue
= blk_alloc_queue(GFP_KERNEL
);
2084 dm_init_md_queue(md
);
2086 md
->disk
= alloc_disk(1);
2090 atomic_set(&md
->pending
[0], 0);
2091 atomic_set(&md
->pending
[1], 0);
2092 init_waitqueue_head(&md
->wait
);
2093 INIT_WORK(&md
->work
, dm_wq_work
);
2094 init_waitqueue_head(&md
->eventq
);
2095 init_completion(&md
->kobj_holder
.completion
);
2097 md
->disk
->major
= _major
;
2098 md
->disk
->first_minor
= minor
;
2099 md
->disk
->fops
= &dm_blk_dops
;
2100 md
->disk
->queue
= md
->queue
;
2101 md
->disk
->private_data
= md
;
2102 sprintf(md
->disk
->disk_name
, "dm-%d", minor
);
2104 format_dev_t(md
->name
, MKDEV(_major
, minor
));
2106 md
->wq
= alloc_workqueue("kdmflush", WQ_MEM_RECLAIM
, 0);
2110 md
->bdev
= bdget_disk(md
->disk
, 0);
2114 bio_init(&md
->flush_bio
);
2115 md
->flush_bio
.bi_bdev
= md
->bdev
;
2116 md
->flush_bio
.bi_rw
= WRITE_FLUSH
;
2118 dm_stats_init(&md
->stats
);
2120 /* Populate the mapping, nobody knows we exist yet */
2121 spin_lock(&_minor_lock
);
2122 old_md
= idr_replace(&_minor_idr
, md
, minor
);
2123 spin_unlock(&_minor_lock
);
2125 BUG_ON(old_md
!= MINOR_ALLOCED
);
2130 destroy_workqueue(md
->wq
);
2132 del_gendisk(md
->disk
);
2135 blk_cleanup_queue(md
->queue
);
2137 cleanup_srcu_struct(&md
->io_barrier
);
2141 module_put(THIS_MODULE
);
2147 static void unlock_fs(struct mapped_device
*md
);
2149 static void free_dev(struct mapped_device
*md
)
2151 int minor
= MINOR(disk_devt(md
->disk
));
2155 destroy_workqueue(md
->wq
);
2157 mempool_destroy(md
->io_pool
);
2159 bioset_free(md
->bs
);
2160 blk_integrity_unregister(md
->disk
);
2161 del_gendisk(md
->disk
);
2162 cleanup_srcu_struct(&md
->io_barrier
);
2163 free_table_devices(&md
->table_devices
);
2166 spin_lock(&_minor_lock
);
2167 md
->disk
->private_data
= NULL
;
2168 spin_unlock(&_minor_lock
);
2171 blk_cleanup_queue(md
->queue
);
2172 dm_stats_cleanup(&md
->stats
);
2173 module_put(THIS_MODULE
);
2177 static void __bind_mempools(struct mapped_device
*md
, struct dm_table
*t
)
2179 struct dm_md_mempools
*p
= dm_table_get_md_mempools(t
);
2181 if (md
->io_pool
&& md
->bs
) {
2182 /* The md already has necessary mempools. */
2183 if (dm_table_get_type(t
) == DM_TYPE_BIO_BASED
) {
2185 * Reload bioset because front_pad may have changed
2186 * because a different table was loaded.
2188 bioset_free(md
->bs
);
2191 } else if (dm_table_get_type(t
) == DM_TYPE_REQUEST_BASED
) {
2193 * There's no need to reload with request-based dm
2194 * because the size of front_pad doesn't change.
2195 * Note for future: If you are to reload bioset,
2196 * prep-ed requests in the queue may refer
2197 * to bio from the old bioset, so you must walk
2198 * through the queue to unprep.
2204 BUG_ON(!p
|| md
->io_pool
|| md
->bs
);
2206 md
->io_pool
= p
->io_pool
;
2212 /* mempool bind completed, now no need any mempools in the table */
2213 dm_table_free_md_mempools(t
);
2217 * Bind a table to the device.
2219 static void event_callback(void *context
)
2221 unsigned long flags
;
2223 struct mapped_device
*md
= (struct mapped_device
*) context
;
2225 spin_lock_irqsave(&md
->uevent_lock
, flags
);
2226 list_splice_init(&md
->uevent_list
, &uevents
);
2227 spin_unlock_irqrestore(&md
->uevent_lock
, flags
);
2229 dm_send_uevents(&uevents
, &disk_to_dev(md
->disk
)->kobj
);
2231 atomic_inc(&md
->event_nr
);
2232 wake_up(&md
->eventq
);
2236 * Protected by md->suspend_lock obtained by dm_swap_table().
2238 static void __set_size(struct mapped_device
*md
, sector_t size
)
2240 set_capacity(md
->disk
, size
);
2242 i_size_write(md
->bdev
->bd_inode
, (loff_t
)size
<< SECTOR_SHIFT
);
2246 * Return 1 if the queue has a compulsory merge_bvec_fn function.
2248 * If this function returns 0, then the device is either a non-dm
2249 * device without a merge_bvec_fn, or it is a dm device that is
2250 * able to split any bios it receives that are too big.
2252 int dm_queue_merge_is_compulsory(struct request_queue
*q
)
2254 struct mapped_device
*dev_md
;
2256 if (!q
->merge_bvec_fn
)
2259 if (q
->make_request_fn
== dm_request
) {
2260 dev_md
= q
->queuedata
;
2261 if (test_bit(DMF_MERGE_IS_OPTIONAL
, &dev_md
->flags
))
2268 static int dm_device_merge_is_compulsory(struct dm_target
*ti
,
2269 struct dm_dev
*dev
, sector_t start
,
2270 sector_t len
, void *data
)
2272 struct block_device
*bdev
= dev
->bdev
;
2273 struct request_queue
*q
= bdev_get_queue(bdev
);
2275 return dm_queue_merge_is_compulsory(q
);
2279 * Return 1 if it is acceptable to ignore merge_bvec_fn based
2280 * on the properties of the underlying devices.
2282 static int dm_table_merge_is_optional(struct dm_table
*table
)
2285 struct dm_target
*ti
;
2287 while (i
< dm_table_get_num_targets(table
)) {
2288 ti
= dm_table_get_target(table
, i
++);
2290 if (ti
->type
->iterate_devices
&&
2291 ti
->type
->iterate_devices(ti
, dm_device_merge_is_compulsory
, NULL
))
2299 * Returns old map, which caller must destroy.
2301 static struct dm_table
*__bind(struct mapped_device
*md
, struct dm_table
*t
,
2302 struct queue_limits
*limits
)
2304 struct dm_table
*old_map
;
2305 struct request_queue
*q
= md
->queue
;
2307 int merge_is_optional
;
2309 size
= dm_table_get_size(t
);
2312 * Wipe any geometry if the size of the table changed.
2314 if (size
!= dm_get_size(md
))
2315 memset(&md
->geometry
, 0, sizeof(md
->geometry
));
2317 __set_size(md
, size
);
2319 dm_table_event_callback(t
, event_callback
, md
);
2322 * The queue hasn't been stopped yet, if the old table type wasn't
2323 * for request-based during suspension. So stop it to prevent
2324 * I/O mapping before resume.
2325 * This must be done before setting the queue restrictions,
2326 * because request-based dm may be run just after the setting.
2328 if (dm_table_request_based(t
) && !blk_queue_stopped(q
))
2331 __bind_mempools(md
, t
);
2333 merge_is_optional
= dm_table_merge_is_optional(t
);
2335 old_map
= rcu_dereference_protected(md
->map
, lockdep_is_held(&md
->suspend_lock
));
2336 rcu_assign_pointer(md
->map
, t
);
2337 md
->immutable_target_type
= dm_table_get_immutable_target_type(t
);
2339 dm_table_set_restrictions(t
, q
, limits
);
2340 if (merge_is_optional
)
2341 set_bit(DMF_MERGE_IS_OPTIONAL
, &md
->flags
);
2343 clear_bit(DMF_MERGE_IS_OPTIONAL
, &md
->flags
);
2351 * Returns unbound table for the caller to free.
2353 static struct dm_table
*__unbind(struct mapped_device
*md
)
2355 struct dm_table
*map
= rcu_dereference_protected(md
->map
, 1);
2360 dm_table_event_callback(map
, NULL
, NULL
);
2361 RCU_INIT_POINTER(md
->map
, NULL
);
2368 * Constructor for a new device.
2370 int dm_create(int minor
, struct mapped_device
**result
)
2372 struct mapped_device
*md
;
2374 md
= alloc_dev(minor
);
2385 * Functions to manage md->type.
2386 * All are required to hold md->type_lock.
2388 void dm_lock_md_type(struct mapped_device
*md
)
2390 mutex_lock(&md
->type_lock
);
2393 void dm_unlock_md_type(struct mapped_device
*md
)
2395 mutex_unlock(&md
->type_lock
);
2398 void dm_set_md_type(struct mapped_device
*md
, unsigned type
)
2400 BUG_ON(!mutex_is_locked(&md
->type_lock
));
2404 unsigned dm_get_md_type(struct mapped_device
*md
)
2406 BUG_ON(!mutex_is_locked(&md
->type_lock
));
2410 struct target_type
*dm_get_immutable_target_type(struct mapped_device
*md
)
2412 return md
->immutable_target_type
;
2416 * The queue_limits are only valid as long as you have a reference
2419 struct queue_limits
*dm_get_queue_limits(struct mapped_device
*md
)
2421 BUG_ON(!atomic_read(&md
->holders
));
2422 return &md
->queue
->limits
;
2424 EXPORT_SYMBOL_GPL(dm_get_queue_limits
);
2427 * Fully initialize a request-based queue (->elevator, ->request_fn, etc).
2429 static int dm_init_request_based_queue(struct mapped_device
*md
)
2431 struct request_queue
*q
= NULL
;
2433 if (md
->queue
->elevator
)
2436 /* Fully initialize the queue */
2437 q
= blk_init_allocated_queue(md
->queue
, dm_request_fn
, NULL
);
2442 dm_init_md_queue(md
);
2443 blk_queue_softirq_done(md
->queue
, dm_softirq_done
);
2444 blk_queue_prep_rq(md
->queue
, dm_prep_fn
);
2445 blk_queue_lld_busy(md
->queue
, dm_lld_busy
);
2447 elv_register_queue(md
->queue
);
2453 * Setup the DM device's queue based on md's type
2455 int dm_setup_md_queue(struct mapped_device
*md
)
2457 if ((dm_get_md_type(md
) == DM_TYPE_REQUEST_BASED
) &&
2458 !dm_init_request_based_queue(md
)) {
2459 DMWARN("Cannot initialize queue for request-based mapped device");
2466 static struct mapped_device
*dm_find_md(dev_t dev
)
2468 struct mapped_device
*md
;
2469 unsigned minor
= MINOR(dev
);
2471 if (MAJOR(dev
) != _major
|| minor
>= (1 << MINORBITS
))
2474 spin_lock(&_minor_lock
);
2476 md
= idr_find(&_minor_idr
, minor
);
2477 if (md
&& (md
== MINOR_ALLOCED
||
2478 (MINOR(disk_devt(dm_disk(md
))) != minor
) ||
2479 dm_deleting_md(md
) ||
2480 test_bit(DMF_FREEING
, &md
->flags
))) {
2486 spin_unlock(&_minor_lock
);
2491 struct mapped_device
*dm_get_md(dev_t dev
)
2493 struct mapped_device
*md
= dm_find_md(dev
);
2500 EXPORT_SYMBOL_GPL(dm_get_md
);
2502 void *dm_get_mdptr(struct mapped_device
*md
)
2504 return md
->interface_ptr
;
2507 void dm_set_mdptr(struct mapped_device
*md
, void *ptr
)
2509 md
->interface_ptr
= ptr
;
2512 void dm_get(struct mapped_device
*md
)
2514 atomic_inc(&md
->holders
);
2515 BUG_ON(test_bit(DMF_FREEING
, &md
->flags
));
2518 const char *dm_device_name(struct mapped_device
*md
)
2522 EXPORT_SYMBOL_GPL(dm_device_name
);
2524 static void __dm_destroy(struct mapped_device
*md
, bool wait
)
2526 struct dm_table
*map
;
2531 spin_lock(&_minor_lock
);
2532 map
= dm_get_live_table(md
, &srcu_idx
);
2533 idr_replace(&_minor_idr
, MINOR_ALLOCED
, MINOR(disk_devt(dm_disk(md
))));
2534 set_bit(DMF_FREEING
, &md
->flags
);
2535 spin_unlock(&_minor_lock
);
2537 if (!dm_suspended_md(md
)) {
2538 dm_table_presuspend_targets(map
);
2539 dm_table_postsuspend_targets(map
);
2542 /* dm_put_live_table must be before msleep, otherwise deadlock is possible */
2543 dm_put_live_table(md
, srcu_idx
);
2546 * Rare, but there may be I/O requests still going to complete,
2547 * for example. Wait for all references to disappear.
2548 * No one should increment the reference count of the mapped_device,
2549 * after the mapped_device state becomes DMF_FREEING.
2552 while (atomic_read(&md
->holders
))
2554 else if (atomic_read(&md
->holders
))
2555 DMWARN("%s: Forcibly removing mapped_device still in use! (%d users)",
2556 dm_device_name(md
), atomic_read(&md
->holders
));
2559 dm_table_destroy(__unbind(md
));
2563 void dm_destroy(struct mapped_device
*md
)
2565 __dm_destroy(md
, true);
2568 void dm_destroy_immediate(struct mapped_device
*md
)
2570 __dm_destroy(md
, false);
2573 void dm_put(struct mapped_device
*md
)
2575 atomic_dec(&md
->holders
);
2577 EXPORT_SYMBOL_GPL(dm_put
);
2579 static int dm_wait_for_completion(struct mapped_device
*md
, int interruptible
)
2582 DECLARE_WAITQUEUE(wait
, current
);
2584 add_wait_queue(&md
->wait
, &wait
);
2587 set_current_state(interruptible
);
2589 if (!md_in_flight(md
))
2592 if (interruptible
== TASK_INTERRUPTIBLE
&&
2593 signal_pending(current
)) {
2600 set_current_state(TASK_RUNNING
);
2602 remove_wait_queue(&md
->wait
, &wait
);
2608 * Process the deferred bios
2610 static void dm_wq_work(struct work_struct
*work
)
2612 struct mapped_device
*md
= container_of(work
, struct mapped_device
,
2616 struct dm_table
*map
;
2618 map
= dm_get_live_table(md
, &srcu_idx
);
2620 while (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND
, &md
->flags
)) {
2621 spin_lock_irq(&md
->deferred_lock
);
2622 c
= bio_list_pop(&md
->deferred
);
2623 spin_unlock_irq(&md
->deferred_lock
);
2628 if (dm_request_based(md
))
2629 generic_make_request(c
);
2631 __split_and_process_bio(md
, map
, c
);
2634 dm_put_live_table(md
, srcu_idx
);
2637 static void dm_queue_flush(struct mapped_device
*md
)
2639 clear_bit(DMF_BLOCK_IO_FOR_SUSPEND
, &md
->flags
);
2640 smp_mb__after_atomic();
2641 queue_work(md
->wq
, &md
->work
);
2645 * Swap in a new table, returning the old one for the caller to destroy.
2647 struct dm_table
*dm_swap_table(struct mapped_device
*md
, struct dm_table
*table
)
2649 struct dm_table
*live_map
= NULL
, *map
= ERR_PTR(-EINVAL
);
2650 struct queue_limits limits
;
2653 mutex_lock(&md
->suspend_lock
);
2655 /* device must be suspended */
2656 if (!dm_suspended_md(md
))
2660 * If the new table has no data devices, retain the existing limits.
2661 * This helps multipath with queue_if_no_path if all paths disappear,
2662 * then new I/O is queued based on these limits, and then some paths
2665 if (dm_table_has_no_data_devices(table
)) {
2666 live_map
= dm_get_live_table_fast(md
);
2668 limits
= md
->queue
->limits
;
2669 dm_put_live_table_fast(md
);
2673 r
= dm_calculate_queue_limits(table
, &limits
);
2680 map
= __bind(md
, table
, &limits
);
2683 mutex_unlock(&md
->suspend_lock
);
2688 * Functions to lock and unlock any filesystem running on the
2691 static int lock_fs(struct mapped_device
*md
)
2695 WARN_ON(md
->frozen_sb
);
2697 md
->frozen_sb
= freeze_bdev(md
->bdev
);
2698 if (IS_ERR(md
->frozen_sb
)) {
2699 r
= PTR_ERR(md
->frozen_sb
);
2700 md
->frozen_sb
= NULL
;
2704 set_bit(DMF_FROZEN
, &md
->flags
);
2709 static void unlock_fs(struct mapped_device
*md
)
2711 if (!test_bit(DMF_FROZEN
, &md
->flags
))
2714 thaw_bdev(md
->bdev
, md
->frozen_sb
);
2715 md
->frozen_sb
= NULL
;
2716 clear_bit(DMF_FROZEN
, &md
->flags
);
2720 * If __dm_suspend returns 0, the device is completely quiescent
2721 * now. There is no request-processing activity. All new requests
2722 * are being added to md->deferred list.
2724 * Caller must hold md->suspend_lock
2726 static int __dm_suspend(struct mapped_device
*md
, struct dm_table
*map
,
2727 unsigned suspend_flags
, int interruptible
)
2729 bool do_lockfs
= suspend_flags
& DM_SUSPEND_LOCKFS_FLAG
;
2730 bool noflush
= suspend_flags
& DM_SUSPEND_NOFLUSH_FLAG
;
2734 * DMF_NOFLUSH_SUSPENDING must be set before presuspend.
2735 * This flag is cleared before dm_suspend returns.
2738 set_bit(DMF_NOFLUSH_SUSPENDING
, &md
->flags
);
2741 * This gets reverted if there's an error later and the targets
2742 * provide the .presuspend_undo hook.
2744 dm_table_presuspend_targets(map
);
2747 * Flush I/O to the device.
2748 * Any I/O submitted after lock_fs() may not be flushed.
2749 * noflush takes precedence over do_lockfs.
2750 * (lock_fs() flushes I/Os and waits for them to complete.)
2752 if (!noflush
&& do_lockfs
) {
2755 dm_table_presuspend_undo_targets(map
);
2761 * Here we must make sure that no processes are submitting requests
2762 * to target drivers i.e. no one may be executing
2763 * __split_and_process_bio. This is called from dm_request and
2766 * To get all processes out of __split_and_process_bio in dm_request,
2767 * we take the write lock. To prevent any process from reentering
2768 * __split_and_process_bio from dm_request and quiesce the thread
2769 * (dm_wq_work), we set BMF_BLOCK_IO_FOR_SUSPEND and call
2770 * flush_workqueue(md->wq).
2772 set_bit(DMF_BLOCK_IO_FOR_SUSPEND
, &md
->flags
);
2774 synchronize_srcu(&md
->io_barrier
);
2777 * Stop md->queue before flushing md->wq in case request-based
2778 * dm defers requests to md->wq from md->queue.
2780 if (dm_request_based(md
))
2781 stop_queue(md
->queue
);
2783 flush_workqueue(md
->wq
);
2786 * At this point no more requests are entering target request routines.
2787 * We call dm_wait_for_completion to wait for all existing requests
2790 r
= dm_wait_for_completion(md
, interruptible
);
2793 clear_bit(DMF_NOFLUSH_SUSPENDING
, &md
->flags
);
2795 synchronize_srcu(&md
->io_barrier
);
2797 /* were we interrupted ? */
2801 if (dm_request_based(md
))
2802 start_queue(md
->queue
);
2805 dm_table_presuspend_undo_targets(map
);
2806 /* pushback list is already flushed, so skip flush */
2813 * We need to be able to change a mapping table under a mounted
2814 * filesystem. For example we might want to move some data in
2815 * the background. Before the table can be swapped with
2816 * dm_bind_table, dm_suspend must be called to flush any in
2817 * flight bios and ensure that any further io gets deferred.
2820 * Suspend mechanism in request-based dm.
2822 * 1. Flush all I/Os by lock_fs() if needed.
2823 * 2. Stop dispatching any I/O by stopping the request_queue.
2824 * 3. Wait for all in-flight I/Os to be completed or requeued.
2826 * To abort suspend, start the request_queue.
2828 int dm_suspend(struct mapped_device
*md
, unsigned suspend_flags
)
2830 struct dm_table
*map
= NULL
;
2834 mutex_lock_nested(&md
->suspend_lock
, SINGLE_DEPTH_NESTING
);
2836 if (dm_suspended_md(md
)) {
2841 if (dm_suspended_internally_md(md
)) {
2842 /* already internally suspended, wait for internal resume */
2843 mutex_unlock(&md
->suspend_lock
);
2844 r
= wait_on_bit(&md
->flags
, DMF_SUSPENDED_INTERNALLY
, TASK_INTERRUPTIBLE
);
2850 map
= rcu_dereference_protected(md
->map
, lockdep_is_held(&md
->suspend_lock
));
2852 r
= __dm_suspend(md
, map
, suspend_flags
, TASK_INTERRUPTIBLE
);
2856 set_bit(DMF_SUSPENDED
, &md
->flags
);
2858 dm_table_postsuspend_targets(map
);
2861 mutex_unlock(&md
->suspend_lock
);
2865 static int __dm_resume(struct mapped_device
*md
, struct dm_table
*map
)
2868 int r
= dm_table_resume_targets(map
);
2876 * Flushing deferred I/Os must be done after targets are resumed
2877 * so that mapping of targets can work correctly.
2878 * Request-based dm is queueing the deferred I/Os in its request_queue.
2880 if (dm_request_based(md
))
2881 start_queue(md
->queue
);
2888 int dm_resume(struct mapped_device
*md
)
2891 struct dm_table
*map
= NULL
;
2894 mutex_lock_nested(&md
->suspend_lock
, SINGLE_DEPTH_NESTING
);
2896 if (!dm_suspended_md(md
))
2899 if (dm_suspended_internally_md(md
)) {
2900 /* already internally suspended, wait for internal resume */
2901 mutex_unlock(&md
->suspend_lock
);
2902 r
= wait_on_bit(&md
->flags
, DMF_SUSPENDED_INTERNALLY
, TASK_INTERRUPTIBLE
);
2908 map
= rcu_dereference_protected(md
->map
, lockdep_is_held(&md
->suspend_lock
));
2909 if (!map
|| !dm_table_get_size(map
))
2912 r
= __dm_resume(md
, map
);
2916 clear_bit(DMF_SUSPENDED
, &md
->flags
);
2920 mutex_unlock(&md
->suspend_lock
);
2926 * Internal suspend/resume works like userspace-driven suspend. It waits
2927 * until all bios finish and prevents issuing new bios to the target drivers.
2928 * It may be used only from the kernel.
2931 static void __dm_internal_suspend(struct mapped_device
*md
, unsigned suspend_flags
)
2933 struct dm_table
*map
= NULL
;
2935 if (md
->internal_suspend_count
++)
2936 return; /* nested internal suspend */
2938 if (dm_suspended_md(md
)) {
2939 set_bit(DMF_SUSPENDED_INTERNALLY
, &md
->flags
);
2940 return; /* nest suspend */
2943 map
= rcu_dereference_protected(md
->map
, lockdep_is_held(&md
->suspend_lock
));
2946 * Using TASK_UNINTERRUPTIBLE because only NOFLUSH internal suspend is
2947 * supported. Properly supporting a TASK_INTERRUPTIBLE internal suspend
2948 * would require changing .presuspend to return an error -- avoid this
2949 * until there is a need for more elaborate variants of internal suspend.
2951 (void) __dm_suspend(md
, map
, suspend_flags
, TASK_UNINTERRUPTIBLE
);
2953 set_bit(DMF_SUSPENDED_INTERNALLY
, &md
->flags
);
2955 dm_table_postsuspend_targets(map
);
2958 static void __dm_internal_resume(struct mapped_device
*md
)
2960 BUG_ON(!md
->internal_suspend_count
);
2962 if (--md
->internal_suspend_count
)
2963 return; /* resume from nested internal suspend */
2965 if (dm_suspended_md(md
))
2966 goto done
; /* resume from nested suspend */
2969 * NOTE: existing callers don't need to call dm_table_resume_targets
2970 * (which may fail -- so best to avoid it for now by passing NULL map)
2972 (void) __dm_resume(md
, NULL
);
2975 clear_bit(DMF_SUSPENDED_INTERNALLY
, &md
->flags
);
2976 smp_mb__after_atomic();
2977 wake_up_bit(&md
->flags
, DMF_SUSPENDED_INTERNALLY
);
2980 void dm_internal_suspend_noflush(struct mapped_device
*md
)
2982 mutex_lock(&md
->suspend_lock
);
2983 __dm_internal_suspend(md
, DM_SUSPEND_NOFLUSH_FLAG
);
2984 mutex_unlock(&md
->suspend_lock
);
2986 EXPORT_SYMBOL_GPL(dm_internal_suspend_noflush
);
2988 void dm_internal_resume(struct mapped_device
*md
)
2990 mutex_lock(&md
->suspend_lock
);
2991 __dm_internal_resume(md
);
2992 mutex_unlock(&md
->suspend_lock
);
2994 EXPORT_SYMBOL_GPL(dm_internal_resume
);
2997 * Fast variants of internal suspend/resume hold md->suspend_lock,
2998 * which prevents interaction with userspace-driven suspend.
3001 void dm_internal_suspend_fast(struct mapped_device
*md
)
3003 mutex_lock(&md
->suspend_lock
);
3004 if (dm_suspended_md(md
) || dm_suspended_internally_md(md
))
3007 set_bit(DMF_BLOCK_IO_FOR_SUSPEND
, &md
->flags
);
3008 synchronize_srcu(&md
->io_barrier
);
3009 flush_workqueue(md
->wq
);
3010 dm_wait_for_completion(md
, TASK_UNINTERRUPTIBLE
);
3013 void dm_internal_resume_fast(struct mapped_device
*md
)
3015 if (dm_suspended_md(md
) || dm_suspended_internally_md(md
))
3021 mutex_unlock(&md
->suspend_lock
);
3024 /*-----------------------------------------------------------------
3025 * Event notification.
3026 *---------------------------------------------------------------*/
3027 int dm_kobject_uevent(struct mapped_device
*md
, enum kobject_action action
,
3030 char udev_cookie
[DM_COOKIE_LENGTH
];
3031 char *envp
[] = { udev_cookie
, NULL
};
3034 return kobject_uevent(&disk_to_dev(md
->disk
)->kobj
, action
);
3036 snprintf(udev_cookie
, DM_COOKIE_LENGTH
, "%s=%u",
3037 DM_COOKIE_ENV_VAR_NAME
, cookie
);
3038 return kobject_uevent_env(&disk_to_dev(md
->disk
)->kobj
,
3043 uint32_t dm_next_uevent_seq(struct mapped_device
*md
)
3045 return atomic_add_return(1, &md
->uevent_seq
);
3048 uint32_t dm_get_event_nr(struct mapped_device
*md
)
3050 return atomic_read(&md
->event_nr
);
3053 int dm_wait_event(struct mapped_device
*md
, int event_nr
)
3055 return wait_event_interruptible(md
->eventq
,
3056 (event_nr
!= atomic_read(&md
->event_nr
)));
3059 void dm_uevent_add(struct mapped_device
*md
, struct list_head
*elist
)
3061 unsigned long flags
;
3063 spin_lock_irqsave(&md
->uevent_lock
, flags
);
3064 list_add(elist
, &md
->uevent_list
);
3065 spin_unlock_irqrestore(&md
->uevent_lock
, flags
);
3069 * The gendisk is only valid as long as you have a reference
3072 struct gendisk
*dm_disk(struct mapped_device
*md
)
3077 struct kobject
*dm_kobject(struct mapped_device
*md
)
3079 return &md
->kobj_holder
.kobj
;
3082 struct mapped_device
*dm_get_from_kobject(struct kobject
*kobj
)
3084 struct mapped_device
*md
;
3086 md
= container_of(kobj
, struct mapped_device
, kobj_holder
.kobj
);
3088 if (test_bit(DMF_FREEING
, &md
->flags
) ||
3096 int dm_suspended_md(struct mapped_device
*md
)
3098 return test_bit(DMF_SUSPENDED
, &md
->flags
);
3101 int dm_suspended_internally_md(struct mapped_device
*md
)
3103 return test_bit(DMF_SUSPENDED_INTERNALLY
, &md
->flags
);
3106 int dm_test_deferred_remove_flag(struct mapped_device
*md
)
3108 return test_bit(DMF_DEFERRED_REMOVE
, &md
->flags
);
3111 int dm_suspended(struct dm_target
*ti
)
3113 return dm_suspended_md(dm_table_get_md(ti
->table
));
3115 EXPORT_SYMBOL_GPL(dm_suspended
);
3117 int dm_noflush_suspending(struct dm_target
*ti
)
3119 return __noflush_suspending(dm_table_get_md(ti
->table
));
3121 EXPORT_SYMBOL_GPL(dm_noflush_suspending
);
3123 struct dm_md_mempools
*dm_alloc_md_mempools(unsigned type
, unsigned integrity
, unsigned per_bio_data_size
)
3125 struct dm_md_mempools
*pools
= kzalloc(sizeof(*pools
), GFP_KERNEL
);
3126 struct kmem_cache
*cachep
;
3127 unsigned int pool_size
;
3128 unsigned int front_pad
;
3133 if (type
== DM_TYPE_BIO_BASED
) {
3135 pool_size
= dm_get_reserved_bio_based_ios();
3136 front_pad
= roundup(per_bio_data_size
, __alignof__(struct dm_target_io
)) + offsetof(struct dm_target_io
, clone
);
3137 } else if (type
== DM_TYPE_REQUEST_BASED
) {
3138 cachep
= _rq_tio_cache
;
3139 pool_size
= dm_get_reserved_rq_based_ios();
3140 front_pad
= offsetof(struct dm_rq_clone_bio_info
, clone
);
3141 /* per_bio_data_size is not used. See __bind_mempools(). */
3142 WARN_ON(per_bio_data_size
!= 0);
3146 pools
->io_pool
= mempool_create_slab_pool(pool_size
, cachep
);
3147 if (!pools
->io_pool
)
3150 pools
->bs
= bioset_create_nobvec(pool_size
, front_pad
);
3154 if (integrity
&& bioset_integrity_create(pools
->bs
, pool_size
))
3160 dm_free_md_mempools(pools
);
3165 void dm_free_md_mempools(struct dm_md_mempools
*pools
)
3171 mempool_destroy(pools
->io_pool
);
3174 bioset_free(pools
->bs
);
3179 static const struct block_device_operations dm_blk_dops
= {
3180 .open
= dm_blk_open
,
3181 .release
= dm_blk_close
,
3182 .ioctl
= dm_blk_ioctl
,
3183 .getgeo
= dm_blk_getgeo
,
3184 .owner
= THIS_MODULE
3190 module_init(dm_init
);
3191 module_exit(dm_exit
);
3193 module_param(major
, uint
, 0);
3194 MODULE_PARM_DESC(major
, "The major number of the device mapper");
3196 module_param(reserved_bio_based_ios
, uint
, S_IRUGO
| S_IWUSR
);
3197 MODULE_PARM_DESC(reserved_bio_based_ios
, "Reserved IOs in bio-based mempools");
3199 module_param(reserved_rq_based_ios
, uint
, S_IRUGO
| S_IWUSR
);
3200 MODULE_PARM_DESC(reserved_rq_based_ios
, "Reserved IOs in request-based mempools");
3202 MODULE_DESCRIPTION(DM_NAME
" driver");
3203 MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
3204 MODULE_LICENSE("GPL");