2 * Copyright (C) 2001 Sistina Software (UK) Limited.
3 * Copyright (C) 2004 Red Hat, Inc. All rights reserved.
5 * This file is released under the GPL.
10 #include <linux/module.h>
11 #include <linux/vmalloc.h>
12 #include <linux/blkdev.h>
13 #include <linux/namei.h>
14 #include <linux/ctype.h>
15 #include <linux/slab.h>
16 #include <linux/interrupt.h>
17 #include <linux/mutex.h>
18 #include <asm/atomic.h>
20 #define DM_MSG_PREFIX "table"
23 #define NODE_SIZE L1_CACHE_BYTES
24 #define KEYS_PER_NODE (NODE_SIZE / sizeof(sector_t))
25 #define CHILDREN_PER_NODE (KEYS_PER_NODE + 1)
28 struct mapped_device
*md
;
33 unsigned int counts
[MAX_DEPTH
]; /* in nodes */
34 sector_t
*index
[MAX_DEPTH
];
36 unsigned int num_targets
;
37 unsigned int num_allocated
;
39 struct dm_target
*targets
;
42 * Indicates the rw permissions for the new logical
43 * device. This should be a combination of FMODE_READ
48 /* a list of devices used by this table */
49 struct list_head devices
;
52 * These are optimistic limits taken from all the
53 * targets, some targets will need smaller limits.
55 struct io_restrictions limits
;
57 /* events get handed up using this callback */
58 void (*event_fn
)(void *);
63 * Similar to ceiling(log_size(n))
65 static unsigned int int_log(unsigned int n
, unsigned int base
)
70 n
= dm_div_up(n
, base
);
78 * Returns the minimum that is _not_ zero, unless both are zero.
80 #define min_not_zero(l, r) (l == 0) ? r : ((r == 0) ? l : min(l, r))
83 * Combine two io_restrictions, always taking the lower value.
85 static void combine_restrictions_low(struct io_restrictions
*lhs
,
86 struct io_restrictions
*rhs
)
89 min_not_zero(lhs
->max_sectors
, rhs
->max_sectors
);
91 lhs
->max_phys_segments
=
92 min_not_zero(lhs
->max_phys_segments
, rhs
->max_phys_segments
);
94 lhs
->max_hw_segments
=
95 min_not_zero(lhs
->max_hw_segments
, rhs
->max_hw_segments
);
97 lhs
->hardsect_size
= max(lhs
->hardsect_size
, rhs
->hardsect_size
);
99 lhs
->max_segment_size
=
100 min_not_zero(lhs
->max_segment_size
, rhs
->max_segment_size
);
102 lhs
->seg_boundary_mask
=
103 min_not_zero(lhs
->seg_boundary_mask
, rhs
->seg_boundary_mask
);
105 lhs
->bounce_pfn
= min_not_zero(lhs
->bounce_pfn
, rhs
->bounce_pfn
);
107 lhs
->no_cluster
|= rhs
->no_cluster
;
111 * Calculate the index of the child node of the n'th node k'th key.
113 static inline unsigned int get_child(unsigned int n
, unsigned int k
)
115 return (n
* CHILDREN_PER_NODE
) + k
;
119 * Return the n'th node of level l from table t.
121 static inline sector_t
*get_node(struct dm_table
*t
,
122 unsigned int l
, unsigned int n
)
124 return t
->index
[l
] + (n
* KEYS_PER_NODE
);
128 * Return the highest key that you could lookup from the n'th
129 * node on level l of the btree.
131 static sector_t
high(struct dm_table
*t
, unsigned int l
, unsigned int n
)
133 for (; l
< t
->depth
- 1; l
++)
134 n
= get_child(n
, CHILDREN_PER_NODE
- 1);
136 if (n
>= t
->counts
[l
])
137 return (sector_t
) - 1;
139 return get_node(t
, l
, n
)[KEYS_PER_NODE
- 1];
143 * Fills in a level of the btree based on the highs of the level
146 static int setup_btree_index(unsigned int l
, struct dm_table
*t
)
151 for (n
= 0U; n
< t
->counts
[l
]; n
++) {
152 node
= get_node(t
, l
, n
);
154 for (k
= 0U; k
< KEYS_PER_NODE
; k
++)
155 node
[k
] = high(t
, l
+ 1, get_child(n
, k
));
161 void *dm_vcalloc(unsigned long nmemb
, unsigned long elem_size
)
167 * Check that we're not going to overflow.
169 if (nmemb
> (ULONG_MAX
/ elem_size
))
172 size
= nmemb
* elem_size
;
173 addr
= vmalloc(size
);
175 memset(addr
, 0, size
);
181 * highs, and targets are managed as dynamic arrays during a
184 static int alloc_targets(struct dm_table
*t
, unsigned int num
)
187 struct dm_target
*n_targets
;
188 int n
= t
->num_targets
;
191 * Allocate both the target array and offset array at once.
193 n_highs
= (sector_t
*) dm_vcalloc(num
, sizeof(struct dm_target
) +
198 n_targets
= (struct dm_target
*) (n_highs
+ num
);
201 memcpy(n_highs
, t
->highs
, sizeof(*n_highs
) * n
);
202 memcpy(n_targets
, t
->targets
, sizeof(*n_targets
) * n
);
205 memset(n_highs
+ n
, -1, sizeof(*n_highs
) * (num
- n
));
208 t
->num_allocated
= num
;
210 t
->targets
= n_targets
;
215 int dm_table_create(struct dm_table
**result
, int mode
,
216 unsigned num_targets
, struct mapped_device
*md
)
218 struct dm_table
*t
= kzalloc(sizeof(*t
), GFP_KERNEL
);
223 INIT_LIST_HEAD(&t
->devices
);
224 atomic_set(&t
->holders
, 1);
227 num_targets
= KEYS_PER_NODE
;
229 num_targets
= dm_round_up(num_targets
, KEYS_PER_NODE
);
231 if (alloc_targets(t
, num_targets
)) {
243 int dm_create_error_table(struct dm_table
**result
, struct mapped_device
*md
)
246 sector_t dev_size
= 1;
250 * Find current size of device.
251 * Default to 1 sector if inactive.
253 t
= dm_get_table(md
);
255 dev_size
= dm_table_get_size(t
);
259 r
= dm_table_create(&t
, FMODE_READ
, 1, md
);
263 r
= dm_table_add_target(t
, "error", 0, dev_size
, NULL
);
267 r
= dm_table_complete(t
);
279 EXPORT_SYMBOL_GPL(dm_create_error_table
);
281 static void free_devices(struct list_head
*devices
)
283 struct list_head
*tmp
, *next
;
285 for (tmp
= devices
->next
; tmp
!= devices
; tmp
= next
) {
286 struct dm_dev
*dd
= list_entry(tmp
, struct dm_dev
, list
);
292 static void table_destroy(struct dm_table
*t
)
296 /* free the indexes (see dm_table_complete) */
298 vfree(t
->index
[t
->depth
- 2]);
300 /* free the targets */
301 for (i
= 0; i
< t
->num_targets
; i
++) {
302 struct dm_target
*tgt
= t
->targets
+ i
;
307 dm_put_target_type(tgt
->type
);
312 /* free the device list */
313 if (t
->devices
.next
!= &t
->devices
) {
314 DMWARN("devices still present during destroy: "
315 "dm_table_remove_device calls missing");
317 free_devices(&t
->devices
);
323 void dm_table_get(struct dm_table
*t
)
325 atomic_inc(&t
->holders
);
328 void dm_table_put(struct dm_table
*t
)
333 if (atomic_dec_and_test(&t
->holders
))
338 * Checks to see if we need to extend highs or targets.
340 static inline int check_space(struct dm_table
*t
)
342 if (t
->num_targets
>= t
->num_allocated
)
343 return alloc_targets(t
, t
->num_allocated
* 2);
349 * Convert a device path to a dev_t.
351 static int lookup_device(const char *path
, dev_t
*dev
)
357 if ((r
= path_lookup(path
, LOOKUP_FOLLOW
, &nd
)))
360 inode
= nd
.dentry
->d_inode
;
366 if (!S_ISBLK(inode
->i_mode
)) {
371 *dev
= inode
->i_rdev
;
379 * See if we've already got a device in the list.
381 static struct dm_dev
*find_device(struct list_head
*l
, dev_t dev
)
385 list_for_each_entry (dd
, l
, list
)
386 if (dd
->bdev
->bd_dev
== dev
)
393 * Open a device so we can use it as a map destination.
395 static int open_dev(struct dm_dev
*d
, dev_t dev
, struct mapped_device
*md
)
397 static char *_claim_ptr
= "I belong to device-mapper";
398 struct block_device
*bdev
;
404 bdev
= open_by_devnum(dev
, d
->mode
);
406 return PTR_ERR(bdev
);
407 r
= bd_claim_by_disk(bdev
, _claim_ptr
, dm_disk(md
));
416 * Close a device that we've been using.
418 static void close_dev(struct dm_dev
*d
, struct mapped_device
*md
)
423 bd_release_from_disk(d
->bdev
, dm_disk(md
));
429 * If possible, this checks an area of a destination device is valid.
431 static int check_device_area(struct dm_dev
*dd
, sector_t start
, sector_t len
)
433 sector_t dev_size
= dd
->bdev
->bd_inode
->i_size
>> SECTOR_SHIFT
;
438 return ((start
< dev_size
) && (len
<= (dev_size
- start
)));
442 * This upgrades the mode on an already open dm_dev. Being
443 * careful to leave things as they were if we fail to reopen the
446 static int upgrade_mode(struct dm_dev
*dd
, int new_mode
, struct mapped_device
*md
)
449 struct dm_dev dd_copy
;
450 dev_t dev
= dd
->bdev
->bd_dev
;
454 dd
->mode
|= new_mode
;
456 r
= open_dev(dd
, dev
, md
);
458 close_dev(&dd_copy
, md
);
466 * Add a device to the list, or just increment the usage count if
467 * it's already present.
469 static int __table_get_device(struct dm_table
*t
, struct dm_target
*ti
,
470 const char *path
, sector_t start
, sector_t len
,
471 int mode
, struct dm_dev
**result
)
476 unsigned int major
, minor
;
480 if (sscanf(path
, "%u:%u", &major
, &minor
) == 2) {
481 /* Extract the major/minor numbers */
482 dev
= MKDEV(major
, minor
);
483 if (MAJOR(dev
) != major
|| MINOR(dev
) != minor
)
486 /* convert the path to a device */
487 if ((r
= lookup_device(path
, &dev
)))
491 dd
= find_device(&t
->devices
, dev
);
493 dd
= kmalloc(sizeof(*dd
), GFP_KERNEL
);
500 if ((r
= open_dev(dd
, dev
, t
->md
))) {
505 format_dev_t(dd
->name
, dev
);
507 atomic_set(&dd
->count
, 0);
508 list_add(&dd
->list
, &t
->devices
);
510 } else if (dd
->mode
!= (mode
| dd
->mode
)) {
511 r
= upgrade_mode(dd
, mode
, t
->md
);
515 atomic_inc(&dd
->count
);
517 if (!check_device_area(dd
, start
, len
)) {
518 DMWARN("device %s too small for target", path
);
519 dm_put_device(ti
, dd
);
528 void dm_set_device_limits(struct dm_target
*ti
, struct block_device
*bdev
)
530 struct request_queue
*q
= bdev_get_queue(bdev
);
531 struct io_restrictions
*rs
= &ti
->limits
;
534 * Combine the device limits low.
536 * FIXME: if we move an io_restriction struct
537 * into q this would just be a call to
538 * combine_restrictions_low()
541 min_not_zero(rs
->max_sectors
, q
->max_sectors
);
543 /* FIXME: Device-Mapper on top of RAID-0 breaks because DM
544 * currently doesn't honor MD's merge_bvec_fn routine.
545 * In this case, we'll force DM to use PAGE_SIZE or
546 * smaller I/O, just to be safe. A better fix is in the
547 * works, but add this for the time being so it will at
548 * least operate correctly.
550 if (q
->merge_bvec_fn
)
552 min_not_zero(rs
->max_sectors
,
553 (unsigned int) (PAGE_SIZE
>> 9));
555 rs
->max_phys_segments
=
556 min_not_zero(rs
->max_phys_segments
,
557 q
->max_phys_segments
);
559 rs
->max_hw_segments
=
560 min_not_zero(rs
->max_hw_segments
, q
->max_hw_segments
);
562 rs
->hardsect_size
= max(rs
->hardsect_size
, q
->hardsect_size
);
564 rs
->max_segment_size
=
565 min_not_zero(rs
->max_segment_size
, q
->max_segment_size
);
567 rs
->seg_boundary_mask
=
568 min_not_zero(rs
->seg_boundary_mask
,
569 q
->seg_boundary_mask
);
571 rs
->bounce_pfn
= min_not_zero(rs
->bounce_pfn
, q
->bounce_pfn
);
573 rs
->no_cluster
|= !test_bit(QUEUE_FLAG_CLUSTER
, &q
->queue_flags
);
575 EXPORT_SYMBOL_GPL(dm_set_device_limits
);
577 int dm_get_device(struct dm_target
*ti
, const char *path
, sector_t start
,
578 sector_t len
, int mode
, struct dm_dev
**result
)
580 int r
= __table_get_device(ti
->table
, ti
, path
,
581 start
, len
, mode
, result
);
584 dm_set_device_limits(ti
, (*result
)->bdev
);
590 * Decrement a devices use count and remove it if necessary.
592 void dm_put_device(struct dm_target
*ti
, struct dm_dev
*dd
)
594 if (atomic_dec_and_test(&dd
->count
)) {
595 close_dev(dd
, ti
->table
->md
);
602 * Checks to see if the target joins onto the end of the table.
604 static int adjoin(struct dm_table
*table
, struct dm_target
*ti
)
606 struct dm_target
*prev
;
608 if (!table
->num_targets
)
611 prev
= &table
->targets
[table
->num_targets
- 1];
612 return (ti
->begin
== (prev
->begin
+ prev
->len
));
616 * Used to dynamically allocate the arg array.
618 static char **realloc_argv(unsigned *array_size
, char **old_argv
)
623 new_size
= *array_size
? *array_size
* 2 : 64;
624 argv
= kmalloc(new_size
* sizeof(*argv
), GFP_KERNEL
);
626 memcpy(argv
, old_argv
, *array_size
* sizeof(*argv
));
627 *array_size
= new_size
;
635 * Destructively splits up the argument list to pass to ctr.
637 int dm_split_args(int *argc
, char ***argvp
, char *input
)
639 char *start
, *end
= input
, *out
, **argv
= NULL
;
640 unsigned array_size
= 0;
649 argv
= realloc_argv(&array_size
, argv
);
656 /* Skip whitespace */
657 while (*start
&& isspace(*start
))
661 break; /* success, we hit the end */
663 /* 'out' is used to remove any back-quotes */
666 /* Everything apart from '\0' can be quoted */
667 if (*end
== '\\' && *(end
+ 1)) {
674 break; /* end of token */
679 /* have we already filled the array ? */
680 if ((*argc
+ 1) > array_size
) {
681 argv
= realloc_argv(&array_size
, argv
);
686 /* we know this is whitespace */
690 /* terminate the string and put it in the array */
700 static void check_for_valid_limits(struct io_restrictions
*rs
)
702 if (!rs
->max_sectors
)
703 rs
->max_sectors
= SAFE_MAX_SECTORS
;
704 if (!rs
->max_phys_segments
)
705 rs
->max_phys_segments
= MAX_PHYS_SEGMENTS
;
706 if (!rs
->max_hw_segments
)
707 rs
->max_hw_segments
= MAX_HW_SEGMENTS
;
708 if (!rs
->hardsect_size
)
709 rs
->hardsect_size
= 1 << SECTOR_SHIFT
;
710 if (!rs
->max_segment_size
)
711 rs
->max_segment_size
= MAX_SEGMENT_SIZE
;
712 if (!rs
->seg_boundary_mask
)
713 rs
->seg_boundary_mask
= -1;
718 int dm_table_add_target(struct dm_table
*t
, const char *type
,
719 sector_t start
, sector_t len
, char *params
)
721 int r
= -EINVAL
, argc
;
723 struct dm_target
*tgt
;
725 if ((r
= check_space(t
)))
728 tgt
= t
->targets
+ t
->num_targets
;
729 memset(tgt
, 0, sizeof(*tgt
));
732 DMERR("%s: zero-length target", dm_device_name(t
->md
));
736 tgt
->type
= dm_get_target_type(type
);
738 DMERR("%s: %s: unknown target type", dm_device_name(t
->md
),
746 tgt
->error
= "Unknown error";
749 * Does this target adjoin the previous one ?
751 if (!adjoin(t
, tgt
)) {
752 tgt
->error
= "Gap in table";
757 r
= dm_split_args(&argc
, &argv
, params
);
759 tgt
->error
= "couldn't split parameters (insufficient memory)";
763 r
= tgt
->type
->ctr(tgt
, argc
, argv
);
768 t
->highs
[t
->num_targets
++] = tgt
->begin
+ tgt
->len
- 1;
770 /* FIXME: the plan is to combine high here and then have
771 * the merge fn apply the target level restrictions. */
772 combine_restrictions_low(&t
->limits
, &tgt
->limits
);
776 DMERR("%s: %s: %s", dm_device_name(t
->md
), type
, tgt
->error
);
777 dm_put_target_type(tgt
->type
);
781 static int setup_indexes(struct dm_table
*t
)
784 unsigned int total
= 0;
787 /* allocate the space for *all* the indexes */
788 for (i
= t
->depth
- 2; i
>= 0; i
--) {
789 t
->counts
[i
] = dm_div_up(t
->counts
[i
+ 1], CHILDREN_PER_NODE
);
790 total
+= t
->counts
[i
];
793 indexes
= (sector_t
*) dm_vcalloc(total
, (unsigned long) NODE_SIZE
);
797 /* set up internal nodes, bottom-up */
798 for (i
= t
->depth
- 2, total
= 0; i
>= 0; i
--) {
799 t
->index
[i
] = indexes
;
800 indexes
+= (KEYS_PER_NODE
* t
->counts
[i
]);
801 setup_btree_index(i
, t
);
808 * Builds the btree to index the map.
810 int dm_table_complete(struct dm_table
*t
)
813 unsigned int leaf_nodes
;
815 check_for_valid_limits(&t
->limits
);
817 /* how many indexes will the btree have ? */
818 leaf_nodes
= dm_div_up(t
->num_targets
, KEYS_PER_NODE
);
819 t
->depth
= 1 + int_log(leaf_nodes
, CHILDREN_PER_NODE
);
821 /* leaf layer has already been set up */
822 t
->counts
[t
->depth
- 1] = leaf_nodes
;
823 t
->index
[t
->depth
- 1] = t
->highs
;
826 r
= setup_indexes(t
);
831 static DEFINE_MUTEX(_event_lock
);
832 void dm_table_event_callback(struct dm_table
*t
,
833 void (*fn
)(void *), void *context
)
835 mutex_lock(&_event_lock
);
837 t
->event_context
= context
;
838 mutex_unlock(&_event_lock
);
841 void dm_table_event(struct dm_table
*t
)
844 * You can no longer call dm_table_event() from interrupt
845 * context, use a bottom half instead.
847 BUG_ON(in_interrupt());
849 mutex_lock(&_event_lock
);
851 t
->event_fn(t
->event_context
);
852 mutex_unlock(&_event_lock
);
855 sector_t
dm_table_get_size(struct dm_table
*t
)
857 return t
->num_targets
? (t
->highs
[t
->num_targets
- 1] + 1) : 0;
860 struct dm_target
*dm_table_get_target(struct dm_table
*t
, unsigned int index
)
862 if (index
>= t
->num_targets
)
865 return t
->targets
+ index
;
869 * Search the btree for the correct target.
871 struct dm_target
*dm_table_find_target(struct dm_table
*t
, sector_t sector
)
873 unsigned int l
, n
= 0, k
= 0;
876 for (l
= 0; l
< t
->depth
; l
++) {
878 node
= get_node(t
, l
, n
);
880 for (k
= 0; k
< KEYS_PER_NODE
; k
++)
881 if (node
[k
] >= sector
)
885 return &t
->targets
[(KEYS_PER_NODE
* n
) + k
];
888 void dm_table_set_restrictions(struct dm_table
*t
, struct request_queue
*q
)
891 * Make sure we obey the optimistic sub devices
894 blk_queue_max_sectors(q
, t
->limits
.max_sectors
);
895 q
->max_phys_segments
= t
->limits
.max_phys_segments
;
896 q
->max_hw_segments
= t
->limits
.max_hw_segments
;
897 q
->hardsect_size
= t
->limits
.hardsect_size
;
898 q
->max_segment_size
= t
->limits
.max_segment_size
;
899 q
->seg_boundary_mask
= t
->limits
.seg_boundary_mask
;
900 q
->bounce_pfn
= t
->limits
.bounce_pfn
;
901 if (t
->limits
.no_cluster
)
902 q
->queue_flags
&= ~(1 << QUEUE_FLAG_CLUSTER
);
904 q
->queue_flags
|= (1 << QUEUE_FLAG_CLUSTER
);
908 unsigned int dm_table_get_num_targets(struct dm_table
*t
)
910 return t
->num_targets
;
913 struct list_head
*dm_table_get_devices(struct dm_table
*t
)
918 int dm_table_get_mode(struct dm_table
*t
)
923 static void suspend_targets(struct dm_table
*t
, unsigned postsuspend
)
925 int i
= t
->num_targets
;
926 struct dm_target
*ti
= t
->targets
;
930 if (ti
->type
->postsuspend
)
931 ti
->type
->postsuspend(ti
);
932 } else if (ti
->type
->presuspend
)
933 ti
->type
->presuspend(ti
);
939 void dm_table_presuspend_targets(struct dm_table
*t
)
944 return suspend_targets(t
, 0);
947 void dm_table_postsuspend_targets(struct dm_table
*t
)
952 return suspend_targets(t
, 1);
955 int dm_table_resume_targets(struct dm_table
*t
)
959 for (i
= 0; i
< t
->num_targets
; i
++) {
960 struct dm_target
*ti
= t
->targets
+ i
;
962 if (!ti
->type
->preresume
)
965 r
= ti
->type
->preresume(ti
);
970 for (i
= 0; i
< t
->num_targets
; i
++) {
971 struct dm_target
*ti
= t
->targets
+ i
;
973 if (ti
->type
->resume
)
974 ti
->type
->resume(ti
);
980 int dm_table_any_congested(struct dm_table
*t
, int bdi_bits
)
982 struct list_head
*d
, *devices
;
985 devices
= dm_table_get_devices(t
);
986 for (d
= devices
->next
; d
!= devices
; d
= d
->next
) {
987 struct dm_dev
*dd
= list_entry(d
, struct dm_dev
, list
);
988 struct request_queue
*q
= bdev_get_queue(dd
->bdev
);
989 r
|= bdi_congested(&q
->backing_dev_info
, bdi_bits
);
995 void dm_table_unplug_all(struct dm_table
*t
)
997 struct list_head
*d
, *devices
= dm_table_get_devices(t
);
999 for (d
= devices
->next
; d
!= devices
; d
= d
->next
) {
1000 struct dm_dev
*dd
= list_entry(d
, struct dm_dev
, list
);
1001 struct request_queue
*q
= bdev_get_queue(dd
->bdev
);
1008 struct mapped_device
*dm_table_get_md(struct dm_table
*t
)
1015 EXPORT_SYMBOL(dm_vcalloc
);
1016 EXPORT_SYMBOL(dm_get_device
);
1017 EXPORT_SYMBOL(dm_put_device
);
1018 EXPORT_SYMBOL(dm_table_event
);
1019 EXPORT_SYMBOL(dm_table_get_size
);
1020 EXPORT_SYMBOL(dm_table_get_mode
);
1021 EXPORT_SYMBOL(dm_table_get_md
);
1022 EXPORT_SYMBOL(dm_table_put
);
1023 EXPORT_SYMBOL(dm_table_get
);
1024 EXPORT_SYMBOL(dm_table_unplug_all
);