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/buffer_head.h>
18 #include <linux/mempool.h>
19 #include <linux/slab.h>
20 #include <linux/idr.h>
21 #include <linux/hdreg.h>
22 #include <linux/delay.h>
24 #include <trace/events/block.h>
26 #define DM_MSG_PREFIX "core"
29 * Cookies are numeric values sent with CHANGE and REMOVE
30 * uevents while resuming, removing or renaming the device.
32 #define DM_COOKIE_ENV_VAR_NAME "DM_COOKIE"
33 #define DM_COOKIE_LENGTH 24
35 static DEFINE_MUTEX(dm_mutex
);
36 static const char *_name
= DM_NAME
;
38 static unsigned int major
= 0;
39 static unsigned int _major
= 0;
41 static DEFINE_SPINLOCK(_minor_lock
);
44 * One of these is allocated per bio.
47 struct mapped_device
*md
;
51 unsigned long start_time
;
52 spinlock_t endio_lock
;
57 * One of these is allocated per target within a bio. Hopefully
58 * this will be simplified out one day.
67 * For request-based dm.
68 * One of these is allocated per request.
70 struct dm_rq_target_io
{
71 struct mapped_device
*md
;
73 struct request
*orig
, clone
;
79 * For request-based dm.
80 * One of these is allocated per bio.
82 struct dm_rq_clone_bio_info
{
84 struct dm_rq_target_io
*tio
;
87 union map_info
*dm_get_mapinfo(struct bio
*bio
)
89 if (bio
&& bio
->bi_private
)
90 return &((struct dm_target_io
*)bio
->bi_private
)->info
;
94 union map_info
*dm_get_rq_mapinfo(struct request
*rq
)
96 if (rq
&& rq
->end_io_data
)
97 return &((struct dm_rq_target_io
*)rq
->end_io_data
)->info
;
100 EXPORT_SYMBOL_GPL(dm_get_rq_mapinfo
);
102 #define MINOR_ALLOCED ((void *)-1)
105 * Bits for the md->flags field.
107 #define DMF_BLOCK_IO_FOR_SUSPEND 0
108 #define DMF_SUSPENDED 1
110 #define DMF_FREEING 3
111 #define DMF_DELETING 4
112 #define DMF_NOFLUSH_SUSPENDING 5
115 * Work processed by per-device workqueue.
117 struct mapped_device
{
118 struct rw_semaphore io_lock
;
119 struct mutex suspend_lock
;
126 struct request_queue
*queue
;
128 /* Protect queue and type against concurrent access. */
129 struct mutex type_lock
;
131 struct gendisk
*disk
;
137 * A list of ios that arrived while we were suspended.
140 wait_queue_head_t wait
;
141 struct work_struct work
;
142 struct bio_list deferred
;
143 spinlock_t deferred_lock
;
146 * Processing queue (flush)
148 struct workqueue_struct
*wq
;
151 * The current mapping.
153 struct dm_table
*map
;
156 * io objects are allocated from here.
167 wait_queue_head_t eventq
;
169 struct list_head uevent_list
;
170 spinlock_t uevent_lock
; /* Protect access to uevent_list */
173 * freeze/thaw support require holding onto a super block
175 struct super_block
*frozen_sb
;
176 struct block_device
*bdev
;
178 /* forced geometry settings */
179 struct hd_geometry geometry
;
181 /* For saving the address of __make_request for request based dm */
182 make_request_fn
*saved_make_request_fn
;
187 /* zero-length flush that will be cloned and submitted to targets */
188 struct bio flush_bio
;
192 * For mempools pre-allocation at the table loading time.
194 struct dm_md_mempools
{
201 static struct kmem_cache
*_io_cache
;
202 static struct kmem_cache
*_tio_cache
;
203 static struct kmem_cache
*_rq_tio_cache
;
204 static struct kmem_cache
*_rq_bio_info_cache
;
206 static int __init
local_init(void)
210 /* allocate a slab for the dm_ios */
211 _io_cache
= KMEM_CACHE(dm_io
, 0);
215 /* allocate a slab for the target ios */
216 _tio_cache
= KMEM_CACHE(dm_target_io
, 0);
218 goto out_free_io_cache
;
220 _rq_tio_cache
= KMEM_CACHE(dm_rq_target_io
, 0);
222 goto out_free_tio_cache
;
224 _rq_bio_info_cache
= KMEM_CACHE(dm_rq_clone_bio_info
, 0);
225 if (!_rq_bio_info_cache
)
226 goto out_free_rq_tio_cache
;
228 r
= dm_uevent_init();
230 goto out_free_rq_bio_info_cache
;
233 r
= register_blkdev(_major
, _name
);
235 goto out_uevent_exit
;
244 out_free_rq_bio_info_cache
:
245 kmem_cache_destroy(_rq_bio_info_cache
);
246 out_free_rq_tio_cache
:
247 kmem_cache_destroy(_rq_tio_cache
);
249 kmem_cache_destroy(_tio_cache
);
251 kmem_cache_destroy(_io_cache
);
256 static void local_exit(void)
258 kmem_cache_destroy(_rq_bio_info_cache
);
259 kmem_cache_destroy(_rq_tio_cache
);
260 kmem_cache_destroy(_tio_cache
);
261 kmem_cache_destroy(_io_cache
);
262 unregister_blkdev(_major
, _name
);
267 DMINFO("cleaned up");
270 static int (*_inits
[])(void) __initdata
= {
280 static void (*_exits
[])(void) = {
290 static int __init
dm_init(void)
292 const int count
= ARRAY_SIZE(_inits
);
296 for (i
= 0; i
< count
; i
++) {
311 static void __exit
dm_exit(void)
313 int i
= ARRAY_SIZE(_exits
);
320 * Block device functions
322 int dm_deleting_md(struct mapped_device
*md
)
324 return test_bit(DMF_DELETING
, &md
->flags
);
327 static int dm_blk_open(struct block_device
*bdev
, fmode_t mode
)
329 struct mapped_device
*md
;
331 mutex_lock(&dm_mutex
);
332 spin_lock(&_minor_lock
);
334 md
= bdev
->bd_disk
->private_data
;
338 if (test_bit(DMF_FREEING
, &md
->flags
) ||
339 dm_deleting_md(md
)) {
345 atomic_inc(&md
->open_count
);
348 spin_unlock(&_minor_lock
);
349 mutex_unlock(&dm_mutex
);
351 return md
? 0 : -ENXIO
;
354 static int dm_blk_close(struct gendisk
*disk
, fmode_t mode
)
356 struct mapped_device
*md
= disk
->private_data
;
358 mutex_lock(&dm_mutex
);
359 atomic_dec(&md
->open_count
);
361 mutex_unlock(&dm_mutex
);
366 int dm_open_count(struct mapped_device
*md
)
368 return atomic_read(&md
->open_count
);
372 * Guarantees nothing is using the device before it's deleted.
374 int dm_lock_for_deletion(struct mapped_device
*md
)
378 spin_lock(&_minor_lock
);
380 if (dm_open_count(md
))
383 set_bit(DMF_DELETING
, &md
->flags
);
385 spin_unlock(&_minor_lock
);
390 static int dm_blk_getgeo(struct block_device
*bdev
, struct hd_geometry
*geo
)
392 struct mapped_device
*md
= bdev
->bd_disk
->private_data
;
394 return dm_get_geometry(md
, geo
);
397 static int dm_blk_ioctl(struct block_device
*bdev
, fmode_t mode
,
398 unsigned int cmd
, unsigned long arg
)
400 struct mapped_device
*md
= bdev
->bd_disk
->private_data
;
401 struct dm_table
*map
= dm_get_live_table(md
);
402 struct dm_target
*tgt
;
405 if (!map
|| !dm_table_get_size(map
))
408 /* We only support devices that have a single target */
409 if (dm_table_get_num_targets(map
) != 1)
412 tgt
= dm_table_get_target(map
, 0);
414 if (dm_suspended_md(md
)) {
419 if (tgt
->type
->ioctl
)
420 r
= tgt
->type
->ioctl(tgt
, cmd
, arg
);
428 static struct dm_io
*alloc_io(struct mapped_device
*md
)
430 return mempool_alloc(md
->io_pool
, GFP_NOIO
);
433 static void free_io(struct mapped_device
*md
, struct dm_io
*io
)
435 mempool_free(io
, md
->io_pool
);
438 static void free_tio(struct mapped_device
*md
, struct dm_target_io
*tio
)
440 mempool_free(tio
, md
->tio_pool
);
443 static struct dm_rq_target_io
*alloc_rq_tio(struct mapped_device
*md
,
446 return mempool_alloc(md
->tio_pool
, gfp_mask
);
449 static void free_rq_tio(struct dm_rq_target_io
*tio
)
451 mempool_free(tio
, tio
->md
->tio_pool
);
454 static struct dm_rq_clone_bio_info
*alloc_bio_info(struct mapped_device
*md
)
456 return mempool_alloc(md
->io_pool
, GFP_ATOMIC
);
459 static void free_bio_info(struct dm_rq_clone_bio_info
*info
)
461 mempool_free(info
, info
->tio
->md
->io_pool
);
464 static int md_in_flight(struct mapped_device
*md
)
466 return atomic_read(&md
->pending
[READ
]) +
467 atomic_read(&md
->pending
[WRITE
]);
470 static void start_io_acct(struct dm_io
*io
)
472 struct mapped_device
*md
= io
->md
;
474 int rw
= bio_data_dir(io
->bio
);
476 io
->start_time
= jiffies
;
478 cpu
= part_stat_lock();
479 part_round_stats(cpu
, &dm_disk(md
)->part0
);
481 dm_disk(md
)->part0
.in_flight
[rw
] = atomic_inc_return(&md
->pending
[rw
]);
484 static void end_io_acct(struct dm_io
*io
)
486 struct mapped_device
*md
= io
->md
;
487 struct bio
*bio
= io
->bio
;
488 unsigned long duration
= jiffies
- io
->start_time
;
490 int rw
= bio_data_dir(bio
);
492 cpu
= part_stat_lock();
493 part_round_stats(cpu
, &dm_disk(md
)->part0
);
494 part_stat_add(cpu
, &dm_disk(md
)->part0
, ticks
[rw
], duration
);
498 * After this is decremented the bio must not be touched if it is
501 dm_disk(md
)->part0
.in_flight
[rw
] = pending
=
502 atomic_dec_return(&md
->pending
[rw
]);
503 pending
+= atomic_read(&md
->pending
[rw
^0x1]);
505 /* nudge anyone waiting on suspend queue */
511 * Add the bio to the list of deferred io.
513 static void queue_io(struct mapped_device
*md
, struct bio
*bio
)
517 spin_lock_irqsave(&md
->deferred_lock
, flags
);
518 bio_list_add(&md
->deferred
, bio
);
519 spin_unlock_irqrestore(&md
->deferred_lock
, flags
);
520 queue_work(md
->wq
, &md
->work
);
524 * Everyone (including functions in this file), should use this
525 * function to access the md->map field, and make sure they call
526 * dm_table_put() when finished.
528 struct dm_table
*dm_get_live_table(struct mapped_device
*md
)
533 read_lock_irqsave(&md
->map_lock
, flags
);
537 read_unlock_irqrestore(&md
->map_lock
, flags
);
543 * Get the geometry associated with a dm device
545 int dm_get_geometry(struct mapped_device
*md
, struct hd_geometry
*geo
)
553 * Set the geometry of a device.
555 int dm_set_geometry(struct mapped_device
*md
, struct hd_geometry
*geo
)
557 sector_t sz
= (sector_t
)geo
->cylinders
* geo
->heads
* geo
->sectors
;
559 if (geo
->start
> sz
) {
560 DMWARN("Start sector is beyond the geometry limits.");
569 /*-----------------------------------------------------------------
571 * A more elegant soln is in the works that uses the queue
572 * merge fn, unfortunately there are a couple of changes to
573 * the block layer that I want to make for this. So in the
574 * interests of getting something for people to use I give
575 * you this clearly demarcated crap.
576 *---------------------------------------------------------------*/
578 static int __noflush_suspending(struct mapped_device
*md
)
580 return test_bit(DMF_NOFLUSH_SUSPENDING
, &md
->flags
);
584 * Decrements the number of outstanding ios that a bio has been
585 * cloned into, completing the original io if necc.
587 static void dec_pending(struct dm_io
*io
, int error
)
592 struct mapped_device
*md
= io
->md
;
594 /* Push-back supersedes any I/O errors */
595 if (unlikely(error
)) {
596 spin_lock_irqsave(&io
->endio_lock
, flags
);
597 if (!(io
->error
> 0 && __noflush_suspending(md
)))
599 spin_unlock_irqrestore(&io
->endio_lock
, flags
);
602 if (atomic_dec_and_test(&io
->io_count
)) {
603 if (io
->error
== DM_ENDIO_REQUEUE
) {
605 * Target requested pushing back the I/O.
607 spin_lock_irqsave(&md
->deferred_lock
, flags
);
608 if (__noflush_suspending(md
))
609 bio_list_add_head(&md
->deferred
, io
->bio
);
611 /* noflush suspend was interrupted. */
613 spin_unlock_irqrestore(&md
->deferred_lock
, flags
);
616 io_error
= io
->error
;
621 if (io_error
== DM_ENDIO_REQUEUE
)
624 if ((bio
->bi_rw
& REQ_FLUSH
) && bio
->bi_size
) {
626 * Preflush done for flush with data, reissue
629 bio
->bi_rw
&= ~REQ_FLUSH
;
632 /* done with normal IO or empty flush */
633 trace_block_bio_complete(md
->queue
, bio
);
634 bio_endio(bio
, io_error
);
639 static void clone_endio(struct bio
*bio
, int error
)
642 struct dm_target_io
*tio
= bio
->bi_private
;
643 struct dm_io
*io
= tio
->io
;
644 struct mapped_device
*md
= tio
->io
->md
;
645 dm_endio_fn endio
= tio
->ti
->type
->end_io
;
647 if (!bio_flagged(bio
, BIO_UPTODATE
) && !error
)
651 r
= endio(tio
->ti
, bio
, error
, &tio
->info
);
652 if (r
< 0 || r
== DM_ENDIO_REQUEUE
)
654 * error and requeue request are handled
658 else if (r
== DM_ENDIO_INCOMPLETE
)
659 /* The target will handle the io */
662 DMWARN("unimplemented target endio return value: %d", r
);
668 * Store md for cleanup instead of tio which is about to get freed.
670 bio
->bi_private
= md
->bs
;
674 dec_pending(io
, error
);
678 * Partial completion handling for request-based dm
680 static void end_clone_bio(struct bio
*clone
, int error
)
682 struct dm_rq_clone_bio_info
*info
= clone
->bi_private
;
683 struct dm_rq_target_io
*tio
= info
->tio
;
684 struct bio
*bio
= info
->orig
;
685 unsigned int nr_bytes
= info
->orig
->bi_size
;
691 * An error has already been detected on the request.
692 * Once error occurred, just let clone->end_io() handle
698 * Don't notice the error to the upper layer yet.
699 * The error handling decision is made by the target driver,
700 * when the request is completed.
707 * I/O for the bio successfully completed.
708 * Notice the data completion to the upper layer.
712 * bios are processed from the head of the list.
713 * So the completing bio should always be rq->bio.
714 * If it's not, something wrong is happening.
716 if (tio
->orig
->bio
!= bio
)
717 DMERR("bio completion is going in the middle of the request");
720 * Update the original request.
721 * Do not use blk_end_request() here, because it may complete
722 * the original request before the clone, and break the ordering.
724 blk_update_request(tio
->orig
, 0, nr_bytes
);
728 * Don't touch any member of the md after calling this function because
729 * the md may be freed in dm_put() at the end of this function.
730 * Or do dm_get() before calling this function and dm_put() later.
732 static void rq_completed(struct mapped_device
*md
, int rw
, int run_queue
)
734 atomic_dec(&md
->pending
[rw
]);
736 /* nudge anyone waiting on suspend queue */
737 if (!md_in_flight(md
))
741 blk_run_queue(md
->queue
);
744 * dm_put() must be at the end of this function. See the comment above
749 static void free_rq_clone(struct request
*clone
)
751 struct dm_rq_target_io
*tio
= clone
->end_io_data
;
753 blk_rq_unprep_clone(clone
);
758 * Complete the clone and the original request.
759 * Must be called without queue lock.
761 static void dm_end_request(struct request
*clone
, int error
)
763 int rw
= rq_data_dir(clone
);
764 struct dm_rq_target_io
*tio
= clone
->end_io_data
;
765 struct mapped_device
*md
= tio
->md
;
766 struct request
*rq
= tio
->orig
;
768 if (rq
->cmd_type
== REQ_TYPE_BLOCK_PC
) {
769 rq
->errors
= clone
->errors
;
770 rq
->resid_len
= clone
->resid_len
;
774 * We are using the sense buffer of the original
776 * So setting the length of the sense data is enough.
778 rq
->sense_len
= clone
->sense_len
;
781 free_rq_clone(clone
);
782 blk_end_request_all(rq
, error
);
783 rq_completed(md
, rw
, true);
786 static void dm_unprep_request(struct request
*rq
)
788 struct request
*clone
= rq
->special
;
791 rq
->cmd_flags
&= ~REQ_DONTPREP
;
793 free_rq_clone(clone
);
797 * Requeue the original request of a clone.
799 void dm_requeue_unmapped_request(struct request
*clone
)
801 int rw
= rq_data_dir(clone
);
802 struct dm_rq_target_io
*tio
= clone
->end_io_data
;
803 struct mapped_device
*md
= tio
->md
;
804 struct request
*rq
= tio
->orig
;
805 struct request_queue
*q
= rq
->q
;
808 dm_unprep_request(rq
);
810 spin_lock_irqsave(q
->queue_lock
, flags
);
811 if (elv_queue_empty(q
))
813 blk_requeue_request(q
, rq
);
814 spin_unlock_irqrestore(q
->queue_lock
, flags
);
816 rq_completed(md
, rw
, 0);
818 EXPORT_SYMBOL_GPL(dm_requeue_unmapped_request
);
820 static void __stop_queue(struct request_queue
*q
)
825 static void stop_queue(struct request_queue
*q
)
829 spin_lock_irqsave(q
->queue_lock
, flags
);
831 spin_unlock_irqrestore(q
->queue_lock
, flags
);
834 static void __start_queue(struct request_queue
*q
)
836 if (blk_queue_stopped(q
))
840 static void start_queue(struct request_queue
*q
)
844 spin_lock_irqsave(q
->queue_lock
, flags
);
846 spin_unlock_irqrestore(q
->queue_lock
, flags
);
849 static void dm_done(struct request
*clone
, int error
, bool mapped
)
852 struct dm_rq_target_io
*tio
= clone
->end_io_data
;
853 dm_request_endio_fn rq_end_io
= tio
->ti
->type
->rq_end_io
;
855 if (mapped
&& rq_end_io
)
856 r
= rq_end_io(tio
->ti
, clone
, error
, &tio
->info
);
859 /* The target wants to complete the I/O */
860 dm_end_request(clone
, r
);
861 else if (r
== DM_ENDIO_INCOMPLETE
)
862 /* The target will handle the I/O */
864 else if (r
== DM_ENDIO_REQUEUE
)
865 /* The target wants to requeue the I/O */
866 dm_requeue_unmapped_request(clone
);
868 DMWARN("unimplemented target endio return value: %d", r
);
874 * Request completion handler for request-based dm
876 static void dm_softirq_done(struct request
*rq
)
879 struct request
*clone
= rq
->completion_data
;
880 struct dm_rq_target_io
*tio
= clone
->end_io_data
;
882 if (rq
->cmd_flags
& REQ_FAILED
)
885 dm_done(clone
, tio
->error
, mapped
);
889 * Complete the clone and the original request with the error status
890 * through softirq context.
892 static void dm_complete_request(struct request
*clone
, int error
)
894 struct dm_rq_target_io
*tio
= clone
->end_io_data
;
895 struct request
*rq
= tio
->orig
;
898 rq
->completion_data
= clone
;
899 blk_complete_request(rq
);
903 * Complete the not-mapped clone and the original request with the error status
904 * through softirq context.
905 * Target's rq_end_io() function isn't called.
906 * This may be used when the target's map_rq() function fails.
908 void dm_kill_unmapped_request(struct request
*clone
, int error
)
910 struct dm_rq_target_io
*tio
= clone
->end_io_data
;
911 struct request
*rq
= tio
->orig
;
913 rq
->cmd_flags
|= REQ_FAILED
;
914 dm_complete_request(clone
, error
);
916 EXPORT_SYMBOL_GPL(dm_kill_unmapped_request
);
919 * Called with the queue lock held
921 static void end_clone_request(struct request
*clone
, int error
)
924 * For just cleaning up the information of the queue in which
925 * the clone was dispatched.
926 * The clone is *NOT* freed actually here because it is alloced from
927 * dm own mempool and REQ_ALLOCED isn't set in clone->cmd_flags.
929 __blk_put_request(clone
->q
, clone
);
932 * Actual request completion is done in a softirq context which doesn't
933 * hold the queue lock. Otherwise, deadlock could occur because:
934 * - another request may be submitted by the upper level driver
935 * of the stacking during the completion
936 * - the submission which requires queue lock may be done
939 dm_complete_request(clone
, error
);
943 * Return maximum size of I/O possible at the supplied sector up to the current
946 static sector_t
max_io_len_target_boundary(sector_t sector
, struct dm_target
*ti
)
948 sector_t target_offset
= dm_target_offset(ti
, sector
);
950 return ti
->len
- target_offset
;
953 static sector_t
max_io_len(sector_t sector
, struct dm_target
*ti
)
955 sector_t len
= max_io_len_target_boundary(sector
, ti
);
958 * Does the target need to split even further ?
962 sector_t offset
= dm_target_offset(ti
, sector
);
963 boundary
= ((offset
+ ti
->split_io
) & ~(ti
->split_io
- 1))
972 static void __map_bio(struct dm_target
*ti
, struct bio
*clone
,
973 struct dm_target_io
*tio
)
977 struct mapped_device
*md
;
979 clone
->bi_end_io
= clone_endio
;
980 clone
->bi_private
= tio
;
983 * Map the clone. If r == 0 we don't need to do
984 * anything, the target has assumed ownership of
987 atomic_inc(&tio
->io
->io_count
);
988 sector
= clone
->bi_sector
;
989 r
= ti
->type
->map(ti
, clone
, &tio
->info
);
990 if (r
== DM_MAPIO_REMAPPED
) {
991 /* the bio has been remapped so dispatch it */
993 trace_block_remap(bdev_get_queue(clone
->bi_bdev
), clone
,
994 tio
->io
->bio
->bi_bdev
->bd_dev
, sector
);
996 generic_make_request(clone
);
997 } else if (r
< 0 || r
== DM_MAPIO_REQUEUE
) {
998 /* error the io and bail out, or requeue it if needed */
1000 dec_pending(tio
->io
, r
);
1002 * Store bio_set for cleanup.
1004 clone
->bi_private
= md
->bs
;
1008 DMWARN("unimplemented target map return value: %d", r
);
1014 struct mapped_device
*md
;
1015 struct dm_table
*map
;
1019 sector_t sector_count
;
1023 static void dm_bio_destructor(struct bio
*bio
)
1025 struct bio_set
*bs
= bio
->bi_private
;
1031 * Creates a little bio that just does part of a bvec.
1033 static struct bio
*split_bvec(struct bio
*bio
, sector_t sector
,
1034 unsigned short idx
, unsigned int offset
,
1035 unsigned int len
, struct bio_set
*bs
)
1038 struct bio_vec
*bv
= bio
->bi_io_vec
+ idx
;
1040 clone
= bio_alloc_bioset(GFP_NOIO
, 1, bs
);
1041 clone
->bi_destructor
= dm_bio_destructor
;
1042 *clone
->bi_io_vec
= *bv
;
1044 clone
->bi_sector
= sector
;
1045 clone
->bi_bdev
= bio
->bi_bdev
;
1046 clone
->bi_rw
= bio
->bi_rw
;
1048 clone
->bi_size
= to_bytes(len
);
1049 clone
->bi_io_vec
->bv_offset
= offset
;
1050 clone
->bi_io_vec
->bv_len
= clone
->bi_size
;
1051 clone
->bi_flags
|= 1 << BIO_CLONED
;
1053 if (bio_integrity(bio
)) {
1054 bio_integrity_clone(clone
, bio
, GFP_NOIO
, bs
);
1055 bio_integrity_trim(clone
,
1056 bio_sector_offset(bio
, idx
, offset
), len
);
1063 * Creates a bio that consists of range of complete bvecs.
1065 static struct bio
*clone_bio(struct bio
*bio
, sector_t sector
,
1066 unsigned short idx
, unsigned short bv_count
,
1067 unsigned int len
, struct bio_set
*bs
)
1071 clone
= bio_alloc_bioset(GFP_NOIO
, bio
->bi_max_vecs
, bs
);
1072 __bio_clone(clone
, bio
);
1073 clone
->bi_destructor
= dm_bio_destructor
;
1074 clone
->bi_sector
= sector
;
1075 clone
->bi_idx
= idx
;
1076 clone
->bi_vcnt
= idx
+ bv_count
;
1077 clone
->bi_size
= to_bytes(len
);
1078 clone
->bi_flags
&= ~(1 << BIO_SEG_VALID
);
1080 if (bio_integrity(bio
)) {
1081 bio_integrity_clone(clone
, bio
, GFP_NOIO
, bs
);
1083 if (idx
!= bio
->bi_idx
|| clone
->bi_size
< bio
->bi_size
)
1084 bio_integrity_trim(clone
,
1085 bio_sector_offset(bio
, idx
, 0), len
);
1091 static struct dm_target_io
*alloc_tio(struct clone_info
*ci
,
1092 struct dm_target
*ti
)
1094 struct dm_target_io
*tio
= mempool_alloc(ci
->md
->tio_pool
, GFP_NOIO
);
1098 memset(&tio
->info
, 0, sizeof(tio
->info
));
1103 static void __issue_target_request(struct clone_info
*ci
, struct dm_target
*ti
,
1104 unsigned request_nr
, sector_t len
)
1106 struct dm_target_io
*tio
= alloc_tio(ci
, ti
);
1109 tio
->info
.target_request_nr
= request_nr
;
1112 * Discard requests require the bio's inline iovecs be initialized.
1113 * ci->bio->bi_max_vecs is BIO_INLINE_VECS anyway, for both flush
1114 * and discard, so no need for concern about wasted bvec allocations.
1116 clone
= bio_alloc_bioset(GFP_NOIO
, ci
->bio
->bi_max_vecs
, ci
->md
->bs
);
1117 __bio_clone(clone
, ci
->bio
);
1118 clone
->bi_destructor
= dm_bio_destructor
;
1120 clone
->bi_sector
= ci
->sector
;
1121 clone
->bi_size
= to_bytes(len
);
1124 __map_bio(ti
, clone
, tio
);
1127 static void __issue_target_requests(struct clone_info
*ci
, struct dm_target
*ti
,
1128 unsigned num_requests
, sector_t len
)
1130 unsigned request_nr
;
1132 for (request_nr
= 0; request_nr
< num_requests
; request_nr
++)
1133 __issue_target_request(ci
, ti
, request_nr
, len
);
1136 static int __clone_and_map_empty_flush(struct clone_info
*ci
)
1138 unsigned target_nr
= 0;
1139 struct dm_target
*ti
;
1141 BUG_ON(bio_has_data(ci
->bio
));
1142 while ((ti
= dm_table_get_target(ci
->map
, target_nr
++)))
1143 __issue_target_requests(ci
, ti
, ti
->num_flush_requests
, 0);
1149 * Perform all io with a single clone.
1151 static void __clone_and_map_simple(struct clone_info
*ci
, struct dm_target
*ti
)
1153 struct bio
*clone
, *bio
= ci
->bio
;
1154 struct dm_target_io
*tio
;
1156 tio
= alloc_tio(ci
, ti
);
1157 clone
= clone_bio(bio
, ci
->sector
, ci
->idx
,
1158 bio
->bi_vcnt
- ci
->idx
, ci
->sector_count
,
1160 __map_bio(ti
, clone
, tio
);
1161 ci
->sector_count
= 0;
1164 static int __clone_and_map_discard(struct clone_info
*ci
)
1166 struct dm_target
*ti
;
1170 ti
= dm_table_find_target(ci
->map
, ci
->sector
);
1171 if (!dm_target_is_valid(ti
))
1175 * Even though the device advertised discard support,
1176 * reconfiguration might have changed that since the
1177 * check was performed.
1179 if (!ti
->num_discard_requests
)
1182 len
= min(ci
->sector_count
, max_io_len_target_boundary(ci
->sector
, ti
));
1184 __issue_target_requests(ci
, ti
, ti
->num_discard_requests
, len
);
1187 } while (ci
->sector_count
-= len
);
1192 static int __clone_and_map(struct clone_info
*ci
)
1194 struct bio
*clone
, *bio
= ci
->bio
;
1195 struct dm_target
*ti
;
1196 sector_t len
= 0, max
;
1197 struct dm_target_io
*tio
;
1199 if (unlikely(bio
->bi_rw
& REQ_DISCARD
))
1200 return __clone_and_map_discard(ci
);
1202 ti
= dm_table_find_target(ci
->map
, ci
->sector
);
1203 if (!dm_target_is_valid(ti
))
1206 max
= max_io_len(ci
->sector
, ti
);
1208 if (ci
->sector_count
<= max
) {
1210 * Optimise for the simple case where we can do all of
1211 * the remaining io with a single clone.
1213 __clone_and_map_simple(ci
, ti
);
1215 } else if (to_sector(bio
->bi_io_vec
[ci
->idx
].bv_len
) <= max
) {
1217 * There are some bvecs that don't span targets.
1218 * Do as many of these as possible.
1221 sector_t remaining
= max
;
1224 for (i
= ci
->idx
; remaining
&& (i
< bio
->bi_vcnt
); i
++) {
1225 bv_len
= to_sector(bio
->bi_io_vec
[i
].bv_len
);
1227 if (bv_len
> remaining
)
1230 remaining
-= bv_len
;
1234 tio
= alloc_tio(ci
, ti
);
1235 clone
= clone_bio(bio
, ci
->sector
, ci
->idx
, i
- ci
->idx
, len
,
1237 __map_bio(ti
, clone
, tio
);
1240 ci
->sector_count
-= len
;
1245 * Handle a bvec that must be split between two or more targets.
1247 struct bio_vec
*bv
= bio
->bi_io_vec
+ ci
->idx
;
1248 sector_t remaining
= to_sector(bv
->bv_len
);
1249 unsigned int offset
= 0;
1253 ti
= dm_table_find_target(ci
->map
, ci
->sector
);
1254 if (!dm_target_is_valid(ti
))
1257 max
= max_io_len(ci
->sector
, ti
);
1260 len
= min(remaining
, max
);
1262 tio
= alloc_tio(ci
, ti
);
1263 clone
= split_bvec(bio
, ci
->sector
, ci
->idx
,
1264 bv
->bv_offset
+ offset
, len
,
1267 __map_bio(ti
, clone
, tio
);
1270 ci
->sector_count
-= len
;
1271 offset
+= to_bytes(len
);
1272 } while (remaining
-= len
);
1281 * Split the bio into several clones and submit it to targets.
1283 static void __split_and_process_bio(struct mapped_device
*md
, struct bio
*bio
)
1285 struct clone_info ci
;
1288 ci
.map
= dm_get_live_table(md
);
1289 if (unlikely(!ci
.map
)) {
1295 ci
.io
= alloc_io(md
);
1297 atomic_set(&ci
.io
->io_count
, 1);
1300 spin_lock_init(&ci
.io
->endio_lock
);
1301 ci
.sector
= bio
->bi_sector
;
1302 ci
.idx
= bio
->bi_idx
;
1304 start_io_acct(ci
.io
);
1305 if (bio
->bi_rw
& REQ_FLUSH
) {
1306 ci
.bio
= &ci
.md
->flush_bio
;
1307 ci
.sector_count
= 0;
1308 error
= __clone_and_map_empty_flush(&ci
);
1309 /* dec_pending submits any data associated with flush */
1312 ci
.sector_count
= bio_sectors(bio
);
1313 while (ci
.sector_count
&& !error
)
1314 error
= __clone_and_map(&ci
);
1317 /* drop the extra reference count */
1318 dec_pending(ci
.io
, error
);
1319 dm_table_put(ci
.map
);
1321 /*-----------------------------------------------------------------
1323 *---------------------------------------------------------------*/
1325 static int dm_merge_bvec(struct request_queue
*q
,
1326 struct bvec_merge_data
*bvm
,
1327 struct bio_vec
*biovec
)
1329 struct mapped_device
*md
= q
->queuedata
;
1330 struct dm_table
*map
= dm_get_live_table(md
);
1331 struct dm_target
*ti
;
1332 sector_t max_sectors
;
1338 ti
= dm_table_find_target(map
, bvm
->bi_sector
);
1339 if (!dm_target_is_valid(ti
))
1343 * Find maximum amount of I/O that won't need splitting
1345 max_sectors
= min(max_io_len(bvm
->bi_sector
, ti
),
1346 (sector_t
) BIO_MAX_SECTORS
);
1347 max_size
= (max_sectors
<< SECTOR_SHIFT
) - bvm
->bi_size
;
1352 * merge_bvec_fn() returns number of bytes
1353 * it can accept at this offset
1354 * max is precomputed maximal io size
1356 if (max_size
&& ti
->type
->merge
)
1357 max_size
= ti
->type
->merge(ti
, bvm
, biovec
, max_size
);
1359 * If the target doesn't support merge method and some of the devices
1360 * provided their merge_bvec method (we know this by looking at
1361 * queue_max_hw_sectors), then we can't allow bios with multiple vector
1362 * entries. So always set max_size to 0, and the code below allows
1365 else if (queue_max_hw_sectors(q
) <= PAGE_SIZE
>> 9)
1374 * Always allow an entire first page
1376 if (max_size
<= biovec
->bv_len
&& !(bvm
->bi_size
>> SECTOR_SHIFT
))
1377 max_size
= biovec
->bv_len
;
1383 * The request function that just remaps the bio built up by
1386 static int _dm_request(struct request_queue
*q
, struct bio
*bio
)
1388 int rw
= bio_data_dir(bio
);
1389 struct mapped_device
*md
= q
->queuedata
;
1392 down_read(&md
->io_lock
);
1394 cpu
= part_stat_lock();
1395 part_stat_inc(cpu
, &dm_disk(md
)->part0
, ios
[rw
]);
1396 part_stat_add(cpu
, &dm_disk(md
)->part0
, sectors
[rw
], bio_sectors(bio
));
1399 /* if we're suspended, we have to queue this io for later */
1400 if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND
, &md
->flags
))) {
1401 up_read(&md
->io_lock
);
1403 if (bio_rw(bio
) != READA
)
1410 __split_and_process_bio(md
, bio
);
1411 up_read(&md
->io_lock
);
1415 static int dm_make_request(struct request_queue
*q
, struct bio
*bio
)
1417 struct mapped_device
*md
= q
->queuedata
;
1419 return md
->saved_make_request_fn(q
, bio
); /* call __make_request() */
1422 static int dm_request_based(struct mapped_device
*md
)
1424 return blk_queue_stackable(md
->queue
);
1427 static int dm_request(struct request_queue
*q
, struct bio
*bio
)
1429 struct mapped_device
*md
= q
->queuedata
;
1431 if (dm_request_based(md
))
1432 return dm_make_request(q
, bio
);
1434 return _dm_request(q
, bio
);
1437 void dm_dispatch_request(struct request
*rq
)
1441 if (blk_queue_io_stat(rq
->q
))
1442 rq
->cmd_flags
|= REQ_IO_STAT
;
1444 rq
->start_time
= jiffies
;
1445 r
= blk_insert_cloned_request(rq
->q
, rq
);
1447 dm_complete_request(rq
, r
);
1449 EXPORT_SYMBOL_GPL(dm_dispatch_request
);
1451 static void dm_rq_bio_destructor(struct bio
*bio
)
1453 struct dm_rq_clone_bio_info
*info
= bio
->bi_private
;
1454 struct mapped_device
*md
= info
->tio
->md
;
1456 free_bio_info(info
);
1457 bio_free(bio
, md
->bs
);
1460 static int dm_rq_bio_constructor(struct bio
*bio
, struct bio
*bio_orig
,
1463 struct dm_rq_target_io
*tio
= data
;
1464 struct mapped_device
*md
= tio
->md
;
1465 struct dm_rq_clone_bio_info
*info
= alloc_bio_info(md
);
1470 info
->orig
= bio_orig
;
1472 bio
->bi_end_io
= end_clone_bio
;
1473 bio
->bi_private
= info
;
1474 bio
->bi_destructor
= dm_rq_bio_destructor
;
1479 static int setup_clone(struct request
*clone
, struct request
*rq
,
1480 struct dm_rq_target_io
*tio
)
1484 r
= blk_rq_prep_clone(clone
, rq
, tio
->md
->bs
, GFP_ATOMIC
,
1485 dm_rq_bio_constructor
, tio
);
1489 clone
->cmd
= rq
->cmd
;
1490 clone
->cmd_len
= rq
->cmd_len
;
1491 clone
->sense
= rq
->sense
;
1492 clone
->buffer
= rq
->buffer
;
1493 clone
->end_io
= end_clone_request
;
1494 clone
->end_io_data
= tio
;
1499 static struct request
*clone_rq(struct request
*rq
, struct mapped_device
*md
,
1502 struct request
*clone
;
1503 struct dm_rq_target_io
*tio
;
1505 tio
= alloc_rq_tio(md
, gfp_mask
);
1513 memset(&tio
->info
, 0, sizeof(tio
->info
));
1515 clone
= &tio
->clone
;
1516 if (setup_clone(clone
, rq
, tio
)) {
1526 * Called with the queue lock held.
1528 static int dm_prep_fn(struct request_queue
*q
, struct request
*rq
)
1530 struct mapped_device
*md
= q
->queuedata
;
1531 struct request
*clone
;
1533 if (unlikely(rq
->special
)) {
1534 DMWARN("Already has something in rq->special.");
1535 return BLKPREP_KILL
;
1538 clone
= clone_rq(rq
, md
, GFP_ATOMIC
);
1540 return BLKPREP_DEFER
;
1542 rq
->special
= clone
;
1543 rq
->cmd_flags
|= REQ_DONTPREP
;
1550 * 0 : the request has been processed (not requeued)
1551 * !0 : the request has been requeued
1553 static int map_request(struct dm_target
*ti
, struct request
*clone
,
1554 struct mapped_device
*md
)
1556 int r
, requeued
= 0;
1557 struct dm_rq_target_io
*tio
= clone
->end_io_data
;
1560 * Hold the md reference here for the in-flight I/O.
1561 * We can't rely on the reference count by device opener,
1562 * because the device may be closed during the request completion
1563 * when all bios are completed.
1564 * See the comment in rq_completed() too.
1569 r
= ti
->type
->map_rq(ti
, clone
, &tio
->info
);
1571 case DM_MAPIO_SUBMITTED
:
1572 /* The target has taken the I/O to submit by itself later */
1574 case DM_MAPIO_REMAPPED
:
1575 /* The target has remapped the I/O so dispatch it */
1576 trace_block_rq_remap(clone
->q
, clone
, disk_devt(dm_disk(md
)),
1577 blk_rq_pos(tio
->orig
));
1578 dm_dispatch_request(clone
);
1580 case DM_MAPIO_REQUEUE
:
1581 /* The target wants to requeue the I/O */
1582 dm_requeue_unmapped_request(clone
);
1587 DMWARN("unimplemented target map return value: %d", r
);
1591 /* The target wants to complete the I/O */
1592 dm_kill_unmapped_request(clone
, r
);
1600 * q->request_fn for request-based dm.
1601 * Called with the queue lock held.
1603 static void dm_request_fn(struct request_queue
*q
)
1605 struct mapped_device
*md
= q
->queuedata
;
1606 struct dm_table
*map
= dm_get_live_table(md
);
1607 struct dm_target
*ti
;
1608 struct request
*rq
, *clone
;
1612 * For suspend, check blk_queue_stopped() and increment
1613 * ->pending within a single queue_lock not to increment the
1614 * number of in-flight I/Os after the queue is stopped in
1617 while (!blk_queue_plugged(q
) && !blk_queue_stopped(q
)) {
1618 rq
= blk_peek_request(q
);
1622 /* always use block 0 to find the target for flushes for now */
1624 if (!(rq
->cmd_flags
& REQ_FLUSH
))
1625 pos
= blk_rq_pos(rq
);
1627 ti
= dm_table_find_target(map
, pos
);
1628 BUG_ON(!dm_target_is_valid(ti
));
1630 if (ti
->type
->busy
&& ti
->type
->busy(ti
))
1633 blk_start_request(rq
);
1634 clone
= rq
->special
;
1635 atomic_inc(&md
->pending
[rq_data_dir(clone
)]);
1637 spin_unlock(q
->queue_lock
);
1638 if (map_request(ti
, clone
, md
))
1641 spin_lock_irq(q
->queue_lock
);
1647 spin_lock_irq(q
->queue_lock
);
1650 if (!elv_queue_empty(q
))
1651 /* Some requests still remain, retry later */
1660 int dm_underlying_device_busy(struct request_queue
*q
)
1662 return blk_lld_busy(q
);
1664 EXPORT_SYMBOL_GPL(dm_underlying_device_busy
);
1666 static int dm_lld_busy(struct request_queue
*q
)
1669 struct mapped_device
*md
= q
->queuedata
;
1670 struct dm_table
*map
= dm_get_live_table(md
);
1672 if (!map
|| test_bit(DMF_BLOCK_IO_FOR_SUSPEND
, &md
->flags
))
1675 r
= dm_table_any_busy_target(map
);
1682 static void dm_unplug_all(struct request_queue
*q
)
1684 struct mapped_device
*md
= q
->queuedata
;
1685 struct dm_table
*map
= dm_get_live_table(md
);
1688 if (dm_request_based(md
))
1689 generic_unplug_device(q
);
1691 dm_table_unplug_all(map
);
1696 static int dm_any_congested(void *congested_data
, int bdi_bits
)
1699 struct mapped_device
*md
= congested_data
;
1700 struct dm_table
*map
;
1702 if (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND
, &md
->flags
)) {
1703 map
= dm_get_live_table(md
);
1706 * Request-based dm cares about only own queue for
1707 * the query about congestion status of request_queue
1709 if (dm_request_based(md
))
1710 r
= md
->queue
->backing_dev_info
.state
&
1713 r
= dm_table_any_congested(map
, bdi_bits
);
1722 /*-----------------------------------------------------------------
1723 * An IDR is used to keep track of allocated minor numbers.
1724 *---------------------------------------------------------------*/
1725 static DEFINE_IDR(_minor_idr
);
1727 static void free_minor(int minor
)
1729 spin_lock(&_minor_lock
);
1730 idr_remove(&_minor_idr
, minor
);
1731 spin_unlock(&_minor_lock
);
1735 * See if the device with a specific minor # is free.
1737 static int specific_minor(int minor
)
1741 if (minor
>= (1 << MINORBITS
))
1744 r
= idr_pre_get(&_minor_idr
, GFP_KERNEL
);
1748 spin_lock(&_minor_lock
);
1750 if (idr_find(&_minor_idr
, minor
)) {
1755 r
= idr_get_new_above(&_minor_idr
, MINOR_ALLOCED
, minor
, &m
);
1760 idr_remove(&_minor_idr
, m
);
1766 spin_unlock(&_minor_lock
);
1770 static int next_free_minor(int *minor
)
1774 r
= idr_pre_get(&_minor_idr
, GFP_KERNEL
);
1778 spin_lock(&_minor_lock
);
1780 r
= idr_get_new(&_minor_idr
, MINOR_ALLOCED
, &m
);
1784 if (m
>= (1 << MINORBITS
)) {
1785 idr_remove(&_minor_idr
, m
);
1793 spin_unlock(&_minor_lock
);
1797 static const struct block_device_operations dm_blk_dops
;
1799 static void dm_wq_work(struct work_struct
*work
);
1801 static void dm_init_md_queue(struct mapped_device
*md
)
1804 * Request-based dm devices cannot be stacked on top of bio-based dm
1805 * devices. The type of this dm device has not been decided yet.
1806 * The type is decided at the first table loading time.
1807 * To prevent problematic device stacking, clear the queue flag
1808 * for request stacking support until then.
1810 * This queue is new, so no concurrency on the queue_flags.
1812 queue_flag_clear_unlocked(QUEUE_FLAG_STACKABLE
, md
->queue
);
1814 md
->queue
->queuedata
= md
;
1815 md
->queue
->backing_dev_info
.congested_fn
= dm_any_congested
;
1816 md
->queue
->backing_dev_info
.congested_data
= md
;
1817 blk_queue_make_request(md
->queue
, dm_request
);
1818 blk_queue_bounce_limit(md
->queue
, BLK_BOUNCE_ANY
);
1819 md
->queue
->unplug_fn
= dm_unplug_all
;
1820 blk_queue_merge_bvec(md
->queue
, dm_merge_bvec
);
1821 blk_queue_flush(md
->queue
, REQ_FLUSH
| REQ_FUA
);
1825 * Allocate and initialise a blank device with a given minor.
1827 static struct mapped_device
*alloc_dev(int minor
)
1830 struct mapped_device
*md
= kzalloc(sizeof(*md
), GFP_KERNEL
);
1834 DMWARN("unable to allocate device, out of memory.");
1838 if (!try_module_get(THIS_MODULE
))
1839 goto bad_module_get
;
1841 /* get a minor number for the dev */
1842 if (minor
== DM_ANY_MINOR
)
1843 r
= next_free_minor(&minor
);
1845 r
= specific_minor(minor
);
1849 md
->type
= DM_TYPE_NONE
;
1850 init_rwsem(&md
->io_lock
);
1851 mutex_init(&md
->suspend_lock
);
1852 mutex_init(&md
->type_lock
);
1853 spin_lock_init(&md
->deferred_lock
);
1854 rwlock_init(&md
->map_lock
);
1855 atomic_set(&md
->holders
, 1);
1856 atomic_set(&md
->open_count
, 0);
1857 atomic_set(&md
->event_nr
, 0);
1858 atomic_set(&md
->uevent_seq
, 0);
1859 INIT_LIST_HEAD(&md
->uevent_list
);
1860 spin_lock_init(&md
->uevent_lock
);
1862 md
->queue
= blk_alloc_queue(GFP_KERNEL
);
1866 dm_init_md_queue(md
);
1868 md
->disk
= alloc_disk(1);
1872 atomic_set(&md
->pending
[0], 0);
1873 atomic_set(&md
->pending
[1], 0);
1874 init_waitqueue_head(&md
->wait
);
1875 INIT_WORK(&md
->work
, dm_wq_work
);
1876 init_waitqueue_head(&md
->eventq
);
1878 md
->disk
->major
= _major
;
1879 md
->disk
->first_minor
= minor
;
1880 md
->disk
->fops
= &dm_blk_dops
;
1881 md
->disk
->queue
= md
->queue
;
1882 md
->disk
->private_data
= md
;
1883 sprintf(md
->disk
->disk_name
, "dm-%d", minor
);
1885 format_dev_t(md
->name
, MKDEV(_major
, minor
));
1887 md
->wq
= create_singlethread_workqueue("kdmflush");
1891 md
->bdev
= bdget_disk(md
->disk
, 0);
1895 bio_init(&md
->flush_bio
);
1896 md
->flush_bio
.bi_bdev
= md
->bdev
;
1897 md
->flush_bio
.bi_rw
= WRITE_FLUSH
;
1899 /* Populate the mapping, nobody knows we exist yet */
1900 spin_lock(&_minor_lock
);
1901 old_md
= idr_replace(&_minor_idr
, md
, minor
);
1902 spin_unlock(&_minor_lock
);
1904 BUG_ON(old_md
!= MINOR_ALLOCED
);
1909 destroy_workqueue(md
->wq
);
1911 del_gendisk(md
->disk
);
1914 blk_cleanup_queue(md
->queue
);
1918 module_put(THIS_MODULE
);
1924 static void unlock_fs(struct mapped_device
*md
);
1926 static void free_dev(struct mapped_device
*md
)
1928 int minor
= MINOR(disk_devt(md
->disk
));
1932 destroy_workqueue(md
->wq
);
1934 mempool_destroy(md
->tio_pool
);
1936 mempool_destroy(md
->io_pool
);
1938 bioset_free(md
->bs
);
1939 blk_integrity_unregister(md
->disk
);
1940 del_gendisk(md
->disk
);
1943 spin_lock(&_minor_lock
);
1944 md
->disk
->private_data
= NULL
;
1945 spin_unlock(&_minor_lock
);
1948 blk_cleanup_queue(md
->queue
);
1949 module_put(THIS_MODULE
);
1953 static void __bind_mempools(struct mapped_device
*md
, struct dm_table
*t
)
1955 struct dm_md_mempools
*p
;
1957 if (md
->io_pool
&& md
->tio_pool
&& md
->bs
)
1958 /* the md already has necessary mempools */
1961 p
= dm_table_get_md_mempools(t
);
1962 BUG_ON(!p
|| md
->io_pool
|| md
->tio_pool
|| md
->bs
);
1964 md
->io_pool
= p
->io_pool
;
1966 md
->tio_pool
= p
->tio_pool
;
1972 /* mempool bind completed, now no need any mempools in the table */
1973 dm_table_free_md_mempools(t
);
1977 * Bind a table to the device.
1979 static void event_callback(void *context
)
1981 unsigned long flags
;
1983 struct mapped_device
*md
= (struct mapped_device
*) context
;
1985 spin_lock_irqsave(&md
->uevent_lock
, flags
);
1986 list_splice_init(&md
->uevent_list
, &uevents
);
1987 spin_unlock_irqrestore(&md
->uevent_lock
, flags
);
1989 dm_send_uevents(&uevents
, &disk_to_dev(md
->disk
)->kobj
);
1991 atomic_inc(&md
->event_nr
);
1992 wake_up(&md
->eventq
);
1995 static void __set_size(struct mapped_device
*md
, sector_t size
)
1997 set_capacity(md
->disk
, size
);
1999 mutex_lock(&md
->bdev
->bd_inode
->i_mutex
);
2000 i_size_write(md
->bdev
->bd_inode
, (loff_t
)size
<< SECTOR_SHIFT
);
2001 mutex_unlock(&md
->bdev
->bd_inode
->i_mutex
);
2005 * Returns old map, which caller must destroy.
2007 static struct dm_table
*__bind(struct mapped_device
*md
, struct dm_table
*t
,
2008 struct queue_limits
*limits
)
2010 struct dm_table
*old_map
;
2011 struct request_queue
*q
= md
->queue
;
2013 unsigned long flags
;
2015 size
= dm_table_get_size(t
);
2018 * Wipe any geometry if the size of the table changed.
2020 if (size
!= get_capacity(md
->disk
))
2021 memset(&md
->geometry
, 0, sizeof(md
->geometry
));
2023 __set_size(md
, size
);
2025 dm_table_event_callback(t
, event_callback
, md
);
2028 * The queue hasn't been stopped yet, if the old table type wasn't
2029 * for request-based during suspension. So stop it to prevent
2030 * I/O mapping before resume.
2031 * This must be done before setting the queue restrictions,
2032 * because request-based dm may be run just after the setting.
2034 if (dm_table_request_based(t
) && !blk_queue_stopped(q
))
2037 __bind_mempools(md
, t
);
2039 write_lock_irqsave(&md
->map_lock
, flags
);
2042 dm_table_set_restrictions(t
, q
, limits
);
2043 write_unlock_irqrestore(&md
->map_lock
, flags
);
2049 * Returns unbound table for the caller to free.
2051 static struct dm_table
*__unbind(struct mapped_device
*md
)
2053 struct dm_table
*map
= md
->map
;
2054 unsigned long flags
;
2059 dm_table_event_callback(map
, NULL
, NULL
);
2060 write_lock_irqsave(&md
->map_lock
, flags
);
2062 write_unlock_irqrestore(&md
->map_lock
, flags
);
2068 * Constructor for a new device.
2070 int dm_create(int minor
, struct mapped_device
**result
)
2072 struct mapped_device
*md
;
2074 md
= alloc_dev(minor
);
2085 * Functions to manage md->type.
2086 * All are required to hold md->type_lock.
2088 void dm_lock_md_type(struct mapped_device
*md
)
2090 mutex_lock(&md
->type_lock
);
2093 void dm_unlock_md_type(struct mapped_device
*md
)
2095 mutex_unlock(&md
->type_lock
);
2098 void dm_set_md_type(struct mapped_device
*md
, unsigned type
)
2103 unsigned dm_get_md_type(struct mapped_device
*md
)
2109 * Fully initialize a request-based queue (->elevator, ->request_fn, etc).
2111 static int dm_init_request_based_queue(struct mapped_device
*md
)
2113 struct request_queue
*q
= NULL
;
2115 if (md
->queue
->elevator
)
2118 /* Fully initialize the queue */
2119 q
= blk_init_allocated_queue(md
->queue
, dm_request_fn
, NULL
);
2124 md
->saved_make_request_fn
= md
->queue
->make_request_fn
;
2125 dm_init_md_queue(md
);
2126 blk_queue_softirq_done(md
->queue
, dm_softirq_done
);
2127 blk_queue_prep_rq(md
->queue
, dm_prep_fn
);
2128 blk_queue_lld_busy(md
->queue
, dm_lld_busy
);
2130 elv_register_queue(md
->queue
);
2136 * Setup the DM device's queue based on md's type
2138 int dm_setup_md_queue(struct mapped_device
*md
)
2140 if ((dm_get_md_type(md
) == DM_TYPE_REQUEST_BASED
) &&
2141 !dm_init_request_based_queue(md
)) {
2142 DMWARN("Cannot initialize queue for request-based mapped device");
2149 static struct mapped_device
*dm_find_md(dev_t dev
)
2151 struct mapped_device
*md
;
2152 unsigned minor
= MINOR(dev
);
2154 if (MAJOR(dev
) != _major
|| minor
>= (1 << MINORBITS
))
2157 spin_lock(&_minor_lock
);
2159 md
= idr_find(&_minor_idr
, minor
);
2160 if (md
&& (md
== MINOR_ALLOCED
||
2161 (MINOR(disk_devt(dm_disk(md
))) != minor
) ||
2162 dm_deleting_md(md
) ||
2163 test_bit(DMF_FREEING
, &md
->flags
))) {
2169 spin_unlock(&_minor_lock
);
2174 struct mapped_device
*dm_get_md(dev_t dev
)
2176 struct mapped_device
*md
= dm_find_md(dev
);
2184 void *dm_get_mdptr(struct mapped_device
*md
)
2186 return md
->interface_ptr
;
2189 void dm_set_mdptr(struct mapped_device
*md
, void *ptr
)
2191 md
->interface_ptr
= ptr
;
2194 void dm_get(struct mapped_device
*md
)
2196 atomic_inc(&md
->holders
);
2197 BUG_ON(test_bit(DMF_FREEING
, &md
->flags
));
2200 const char *dm_device_name(struct mapped_device
*md
)
2204 EXPORT_SYMBOL_GPL(dm_device_name
);
2206 static void __dm_destroy(struct mapped_device
*md
, bool wait
)
2208 struct dm_table
*map
;
2212 spin_lock(&_minor_lock
);
2213 map
= dm_get_live_table(md
);
2214 idr_replace(&_minor_idr
, MINOR_ALLOCED
, MINOR(disk_devt(dm_disk(md
))));
2215 set_bit(DMF_FREEING
, &md
->flags
);
2216 spin_unlock(&_minor_lock
);
2218 if (!dm_suspended_md(md
)) {
2219 dm_table_presuspend_targets(map
);
2220 dm_table_postsuspend_targets(map
);
2224 * Rare, but there may be I/O requests still going to complete,
2225 * for example. Wait for all references to disappear.
2226 * No one should increment the reference count of the mapped_device,
2227 * after the mapped_device state becomes DMF_FREEING.
2230 while (atomic_read(&md
->holders
))
2232 else if (atomic_read(&md
->holders
))
2233 DMWARN("%s: Forcibly removing mapped_device still in use! (%d users)",
2234 dm_device_name(md
), atomic_read(&md
->holders
));
2238 dm_table_destroy(__unbind(md
));
2242 void dm_destroy(struct mapped_device
*md
)
2244 __dm_destroy(md
, true);
2247 void dm_destroy_immediate(struct mapped_device
*md
)
2249 __dm_destroy(md
, false);
2252 void dm_put(struct mapped_device
*md
)
2254 atomic_dec(&md
->holders
);
2256 EXPORT_SYMBOL_GPL(dm_put
);
2258 static int dm_wait_for_completion(struct mapped_device
*md
, int interruptible
)
2261 DECLARE_WAITQUEUE(wait
, current
);
2263 dm_unplug_all(md
->queue
);
2265 add_wait_queue(&md
->wait
, &wait
);
2268 set_current_state(interruptible
);
2271 if (!md_in_flight(md
))
2274 if (interruptible
== TASK_INTERRUPTIBLE
&&
2275 signal_pending(current
)) {
2282 set_current_state(TASK_RUNNING
);
2284 remove_wait_queue(&md
->wait
, &wait
);
2290 * Process the deferred bios
2292 static void dm_wq_work(struct work_struct
*work
)
2294 struct mapped_device
*md
= container_of(work
, struct mapped_device
,
2298 down_read(&md
->io_lock
);
2300 while (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND
, &md
->flags
)) {
2301 spin_lock_irq(&md
->deferred_lock
);
2302 c
= bio_list_pop(&md
->deferred
);
2303 spin_unlock_irq(&md
->deferred_lock
);
2308 up_read(&md
->io_lock
);
2310 if (dm_request_based(md
))
2311 generic_make_request(c
);
2313 __split_and_process_bio(md
, c
);
2315 down_read(&md
->io_lock
);
2318 up_read(&md
->io_lock
);
2321 static void dm_queue_flush(struct mapped_device
*md
)
2323 clear_bit(DMF_BLOCK_IO_FOR_SUSPEND
, &md
->flags
);
2324 smp_mb__after_clear_bit();
2325 queue_work(md
->wq
, &md
->work
);
2329 * Swap in a new table, returning the old one for the caller to destroy.
2331 struct dm_table
*dm_swap_table(struct mapped_device
*md
, struct dm_table
*table
)
2333 struct dm_table
*map
= ERR_PTR(-EINVAL
);
2334 struct queue_limits limits
;
2337 mutex_lock(&md
->suspend_lock
);
2339 /* device must be suspended */
2340 if (!dm_suspended_md(md
))
2343 r
= dm_calculate_queue_limits(table
, &limits
);
2349 map
= __bind(md
, table
, &limits
);
2352 mutex_unlock(&md
->suspend_lock
);
2357 * Functions to lock and unlock any filesystem running on the
2360 static int lock_fs(struct mapped_device
*md
)
2364 WARN_ON(md
->frozen_sb
);
2366 md
->frozen_sb
= freeze_bdev(md
->bdev
);
2367 if (IS_ERR(md
->frozen_sb
)) {
2368 r
= PTR_ERR(md
->frozen_sb
);
2369 md
->frozen_sb
= NULL
;
2373 set_bit(DMF_FROZEN
, &md
->flags
);
2378 static void unlock_fs(struct mapped_device
*md
)
2380 if (!test_bit(DMF_FROZEN
, &md
->flags
))
2383 thaw_bdev(md
->bdev
, md
->frozen_sb
);
2384 md
->frozen_sb
= NULL
;
2385 clear_bit(DMF_FROZEN
, &md
->flags
);
2389 * We need to be able to change a mapping table under a mounted
2390 * filesystem. For example we might want to move some data in
2391 * the background. Before the table can be swapped with
2392 * dm_bind_table, dm_suspend must be called to flush any in
2393 * flight bios and ensure that any further io gets deferred.
2396 * Suspend mechanism in request-based dm.
2398 * 1. Flush all I/Os by lock_fs() if needed.
2399 * 2. Stop dispatching any I/O by stopping the request_queue.
2400 * 3. Wait for all in-flight I/Os to be completed or requeued.
2402 * To abort suspend, start the request_queue.
2404 int dm_suspend(struct mapped_device
*md
, unsigned suspend_flags
)
2406 struct dm_table
*map
= NULL
;
2408 int do_lockfs
= suspend_flags
& DM_SUSPEND_LOCKFS_FLAG
? 1 : 0;
2409 int noflush
= suspend_flags
& DM_SUSPEND_NOFLUSH_FLAG
? 1 : 0;
2411 mutex_lock(&md
->suspend_lock
);
2413 if (dm_suspended_md(md
)) {
2418 map
= dm_get_live_table(md
);
2421 * DMF_NOFLUSH_SUSPENDING must be set before presuspend.
2422 * This flag is cleared before dm_suspend returns.
2425 set_bit(DMF_NOFLUSH_SUSPENDING
, &md
->flags
);
2427 /* This does not get reverted if there's an error later. */
2428 dm_table_presuspend_targets(map
);
2431 * Flush I/O to the device.
2432 * Any I/O submitted after lock_fs() may not be flushed.
2433 * noflush takes precedence over do_lockfs.
2434 * (lock_fs() flushes I/Os and waits for them to complete.)
2436 if (!noflush
&& do_lockfs
) {
2443 * Here we must make sure that no processes are submitting requests
2444 * to target drivers i.e. no one may be executing
2445 * __split_and_process_bio. This is called from dm_request and
2448 * To get all processes out of __split_and_process_bio in dm_request,
2449 * we take the write lock. To prevent any process from reentering
2450 * __split_and_process_bio from dm_request and quiesce the thread
2451 * (dm_wq_work), we set BMF_BLOCK_IO_FOR_SUSPEND and call
2452 * flush_workqueue(md->wq).
2454 down_write(&md
->io_lock
);
2455 set_bit(DMF_BLOCK_IO_FOR_SUSPEND
, &md
->flags
);
2456 up_write(&md
->io_lock
);
2459 * Stop md->queue before flushing md->wq in case request-based
2460 * dm defers requests to md->wq from md->queue.
2462 if (dm_request_based(md
))
2463 stop_queue(md
->queue
);
2465 flush_workqueue(md
->wq
);
2468 * At this point no more requests are entering target request routines.
2469 * We call dm_wait_for_completion to wait for all existing requests
2472 r
= dm_wait_for_completion(md
, TASK_INTERRUPTIBLE
);
2474 down_write(&md
->io_lock
);
2476 clear_bit(DMF_NOFLUSH_SUSPENDING
, &md
->flags
);
2477 up_write(&md
->io_lock
);
2479 /* were we interrupted ? */
2483 if (dm_request_based(md
))
2484 start_queue(md
->queue
);
2487 goto out
; /* pushback list is already flushed, so skip flush */
2491 * If dm_wait_for_completion returned 0, the device is completely
2492 * quiescent now. There is no request-processing activity. All new
2493 * requests are being added to md->deferred list.
2496 set_bit(DMF_SUSPENDED
, &md
->flags
);
2498 dm_table_postsuspend_targets(map
);
2504 mutex_unlock(&md
->suspend_lock
);
2508 int dm_resume(struct mapped_device
*md
)
2511 struct dm_table
*map
= NULL
;
2513 mutex_lock(&md
->suspend_lock
);
2514 if (!dm_suspended_md(md
))
2517 map
= dm_get_live_table(md
);
2518 if (!map
|| !dm_table_get_size(map
))
2521 r
= dm_table_resume_targets(map
);
2528 * Flushing deferred I/Os must be done after targets are resumed
2529 * so that mapping of targets can work correctly.
2530 * Request-based dm is queueing the deferred I/Os in its request_queue.
2532 if (dm_request_based(md
))
2533 start_queue(md
->queue
);
2537 clear_bit(DMF_SUSPENDED
, &md
->flags
);
2539 dm_table_unplug_all(map
);
2543 mutex_unlock(&md
->suspend_lock
);
2548 /*-----------------------------------------------------------------
2549 * Event notification.
2550 *---------------------------------------------------------------*/
2551 int dm_kobject_uevent(struct mapped_device
*md
, enum kobject_action action
,
2554 char udev_cookie
[DM_COOKIE_LENGTH
];
2555 char *envp
[] = { udev_cookie
, NULL
};
2558 return kobject_uevent(&disk_to_dev(md
->disk
)->kobj
, action
);
2560 snprintf(udev_cookie
, DM_COOKIE_LENGTH
, "%s=%u",
2561 DM_COOKIE_ENV_VAR_NAME
, cookie
);
2562 return kobject_uevent_env(&disk_to_dev(md
->disk
)->kobj
,
2567 uint32_t dm_next_uevent_seq(struct mapped_device
*md
)
2569 return atomic_add_return(1, &md
->uevent_seq
);
2572 uint32_t dm_get_event_nr(struct mapped_device
*md
)
2574 return atomic_read(&md
->event_nr
);
2577 int dm_wait_event(struct mapped_device
*md
, int event_nr
)
2579 return wait_event_interruptible(md
->eventq
,
2580 (event_nr
!= atomic_read(&md
->event_nr
)));
2583 void dm_uevent_add(struct mapped_device
*md
, struct list_head
*elist
)
2585 unsigned long flags
;
2587 spin_lock_irqsave(&md
->uevent_lock
, flags
);
2588 list_add(elist
, &md
->uevent_list
);
2589 spin_unlock_irqrestore(&md
->uevent_lock
, flags
);
2593 * The gendisk is only valid as long as you have a reference
2596 struct gendisk
*dm_disk(struct mapped_device
*md
)
2601 struct kobject
*dm_kobject(struct mapped_device
*md
)
2607 * struct mapped_device should not be exported outside of dm.c
2608 * so use this check to verify that kobj is part of md structure
2610 struct mapped_device
*dm_get_from_kobject(struct kobject
*kobj
)
2612 struct mapped_device
*md
;
2614 md
= container_of(kobj
, struct mapped_device
, kobj
);
2615 if (&md
->kobj
!= kobj
)
2618 if (test_bit(DMF_FREEING
, &md
->flags
) ||
2626 int dm_suspended_md(struct mapped_device
*md
)
2628 return test_bit(DMF_SUSPENDED
, &md
->flags
);
2631 int dm_suspended(struct dm_target
*ti
)
2633 return dm_suspended_md(dm_table_get_md(ti
->table
));
2635 EXPORT_SYMBOL_GPL(dm_suspended
);
2637 int dm_noflush_suspending(struct dm_target
*ti
)
2639 return __noflush_suspending(dm_table_get_md(ti
->table
));
2641 EXPORT_SYMBOL_GPL(dm_noflush_suspending
);
2643 struct dm_md_mempools
*dm_alloc_md_mempools(unsigned type
)
2645 struct dm_md_mempools
*pools
= kmalloc(sizeof(*pools
), GFP_KERNEL
);
2650 pools
->io_pool
= (type
== DM_TYPE_BIO_BASED
) ?
2651 mempool_create_slab_pool(MIN_IOS
, _io_cache
) :
2652 mempool_create_slab_pool(MIN_IOS
, _rq_bio_info_cache
);
2653 if (!pools
->io_pool
)
2654 goto free_pools_and_out
;
2656 pools
->tio_pool
= (type
== DM_TYPE_BIO_BASED
) ?
2657 mempool_create_slab_pool(MIN_IOS
, _tio_cache
) :
2658 mempool_create_slab_pool(MIN_IOS
, _rq_tio_cache
);
2659 if (!pools
->tio_pool
)
2660 goto free_io_pool_and_out
;
2662 pools
->bs
= (type
== DM_TYPE_BIO_BASED
) ?
2663 bioset_create(16, 0) : bioset_create(MIN_IOS
, 0);
2665 goto free_tio_pool_and_out
;
2669 free_tio_pool_and_out
:
2670 mempool_destroy(pools
->tio_pool
);
2672 free_io_pool_and_out
:
2673 mempool_destroy(pools
->io_pool
);
2681 void dm_free_md_mempools(struct dm_md_mempools
*pools
)
2687 mempool_destroy(pools
->io_pool
);
2689 if (pools
->tio_pool
)
2690 mempool_destroy(pools
->tio_pool
);
2693 bioset_free(pools
->bs
);
2698 static const struct block_device_operations dm_blk_dops
= {
2699 .open
= dm_blk_open
,
2700 .release
= dm_blk_close
,
2701 .ioctl
= dm_blk_ioctl
,
2702 .getgeo
= dm_blk_getgeo
,
2703 .owner
= THIS_MODULE
2706 EXPORT_SYMBOL(dm_get_mapinfo
);
2711 module_init(dm_init
);
2712 module_exit(dm_exit
);
2714 module_param(major
, uint
, 0);
2715 MODULE_PARM_DESC(major
, "The major number of the device mapper");
2716 MODULE_DESCRIPTION(DM_NAME
" driver");
2717 MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
2718 MODULE_LICENSE("GPL");