2 * Copyright (C) 2001 Sistina Software (UK) Limited.
3 * Copyright (C) 2004-2008 Red Hat, Inc. All rights reserved.
5 * This file is released under the GPL.
10 #include <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/string.h>
16 #include <linux/slab.h>
17 #include <linux/interrupt.h>
18 #include <linux/mutex.h>
19 #include <linux/delay.h>
20 #include <linux/atomic.h>
22 #define DM_MSG_PREFIX "table"
25 #define NODE_SIZE L1_CACHE_BYTES
26 #define KEYS_PER_NODE (NODE_SIZE / sizeof(sector_t))
27 #define CHILDREN_PER_NODE (KEYS_PER_NODE + 1)
30 * The table has always exactly one reference from either mapped_device->map
31 * or hash_cell->new_map. This reference is not counted in table->holders.
32 * A pair of dm_create_table/dm_destroy_table functions is used for table
33 * creation/destruction.
35 * Temporary references from the other code increase table->holders. A pair
36 * of dm_table_get/dm_table_put functions is used to manipulate it.
38 * When the table is about to be destroyed, we wait for table->holders to
43 struct mapped_device
*md
;
49 unsigned int counts
[MAX_DEPTH
]; /* in nodes */
50 sector_t
*index
[MAX_DEPTH
];
52 unsigned int num_targets
;
53 unsigned int num_allocated
;
55 struct dm_target
*targets
;
57 struct target_type
*immutable_target_type
;
58 unsigned integrity_supported
:1;
62 * Indicates the rw permissions for the new logical
63 * device. This should be a combination of FMODE_READ
68 /* a list of devices used by this table */
69 struct list_head devices
;
71 /* events get handed up using this callback */
72 void (*event_fn
)(void *);
75 struct dm_md_mempools
*mempools
;
77 struct list_head target_callbacks
;
81 * Similar to ceiling(log_size(n))
83 static unsigned int int_log(unsigned int n
, unsigned int base
)
88 n
= dm_div_up(n
, base
);
96 * Calculate the index of the child node of the n'th node k'th key.
98 static inline unsigned int get_child(unsigned int n
, unsigned int k
)
100 return (n
* CHILDREN_PER_NODE
) + k
;
104 * Return the n'th node of level l from table t.
106 static inline sector_t
*get_node(struct dm_table
*t
,
107 unsigned int l
, unsigned int n
)
109 return t
->index
[l
] + (n
* KEYS_PER_NODE
);
113 * Return the highest key that you could lookup from the n'th
114 * node on level l of the btree.
116 static sector_t
high(struct dm_table
*t
, unsigned int l
, unsigned int n
)
118 for (; l
< t
->depth
- 1; l
++)
119 n
= get_child(n
, CHILDREN_PER_NODE
- 1);
121 if (n
>= t
->counts
[l
])
122 return (sector_t
) - 1;
124 return get_node(t
, l
, n
)[KEYS_PER_NODE
- 1];
128 * Fills in a level of the btree based on the highs of the level
131 static int setup_btree_index(unsigned int l
, struct dm_table
*t
)
136 for (n
= 0U; n
< t
->counts
[l
]; n
++) {
137 node
= get_node(t
, l
, n
);
139 for (k
= 0U; k
< KEYS_PER_NODE
; k
++)
140 node
[k
] = high(t
, l
+ 1, get_child(n
, k
));
146 void *dm_vcalloc(unsigned long nmemb
, unsigned long elem_size
)
152 * Check that we're not going to overflow.
154 if (nmemb
> (ULONG_MAX
/ elem_size
))
157 size
= nmemb
* elem_size
;
158 addr
= vzalloc(size
);
162 EXPORT_SYMBOL(dm_vcalloc
);
165 * highs, and targets are managed as dynamic arrays during a
168 static int alloc_targets(struct dm_table
*t
, unsigned int num
)
171 struct dm_target
*n_targets
;
172 int n
= t
->num_targets
;
175 * Allocate both the target array and offset array at once.
176 * Append an empty entry to catch sectors beyond the end of
179 n_highs
= (sector_t
*) dm_vcalloc(num
+ 1, sizeof(struct dm_target
) +
184 n_targets
= (struct dm_target
*) (n_highs
+ num
);
187 memcpy(n_highs
, t
->highs
, sizeof(*n_highs
) * n
);
188 memcpy(n_targets
, t
->targets
, sizeof(*n_targets
) * n
);
191 memset(n_highs
+ n
, -1, sizeof(*n_highs
) * (num
- n
));
194 t
->num_allocated
= num
;
196 t
->targets
= n_targets
;
201 int dm_table_create(struct dm_table
**result
, fmode_t mode
,
202 unsigned num_targets
, struct mapped_device
*md
)
204 struct dm_table
*t
= kzalloc(sizeof(*t
), GFP_KERNEL
);
209 INIT_LIST_HEAD(&t
->devices
);
210 INIT_LIST_HEAD(&t
->target_callbacks
);
211 atomic_set(&t
->holders
, 0);
214 num_targets
= KEYS_PER_NODE
;
216 num_targets
= dm_round_up(num_targets
, KEYS_PER_NODE
);
218 if (alloc_targets(t
, num_targets
)) {
229 static void free_devices(struct list_head
*devices
)
231 struct list_head
*tmp
, *next
;
233 list_for_each_safe(tmp
, next
, devices
) {
234 struct dm_dev_internal
*dd
=
235 list_entry(tmp
, struct dm_dev_internal
, list
);
236 DMWARN("dm_table_destroy: dm_put_device call missing for %s",
242 void dm_table_destroy(struct dm_table
*t
)
249 while (atomic_read(&t
->holders
))
253 /* free the indexes */
255 vfree(t
->index
[t
->depth
- 2]);
257 /* free the targets */
258 for (i
= 0; i
< t
->num_targets
; i
++) {
259 struct dm_target
*tgt
= t
->targets
+ i
;
264 dm_put_target_type(tgt
->type
);
269 /* free the device list */
270 free_devices(&t
->devices
);
272 dm_free_md_mempools(t
->mempools
);
277 void dm_table_get(struct dm_table
*t
)
279 atomic_inc(&t
->holders
);
281 EXPORT_SYMBOL(dm_table_get
);
283 void dm_table_put(struct dm_table
*t
)
288 smp_mb__before_atomic_dec();
289 atomic_dec(&t
->holders
);
291 EXPORT_SYMBOL(dm_table_put
);
294 * Checks to see if we need to extend highs or targets.
296 static inline int check_space(struct dm_table
*t
)
298 if (t
->num_targets
>= t
->num_allocated
)
299 return alloc_targets(t
, t
->num_allocated
* 2);
305 * See if we've already got a device in the list.
307 static struct dm_dev_internal
*find_device(struct list_head
*l
, dev_t dev
)
309 struct dm_dev_internal
*dd
;
311 list_for_each_entry (dd
, l
, list
)
312 if (dd
->dm_dev
.bdev
->bd_dev
== dev
)
319 * Open a device so we can use it as a map destination.
321 static int open_dev(struct dm_dev_internal
*d
, dev_t dev
,
322 struct mapped_device
*md
)
324 static char *_claim_ptr
= "I belong to device-mapper";
325 struct block_device
*bdev
;
329 BUG_ON(d
->dm_dev
.bdev
);
331 bdev
= blkdev_get_by_dev(dev
, d
->dm_dev
.mode
| FMODE_EXCL
, _claim_ptr
);
333 return PTR_ERR(bdev
);
335 r
= bd_link_disk_holder(bdev
, dm_disk(md
));
337 blkdev_put(bdev
, d
->dm_dev
.mode
| FMODE_EXCL
);
341 d
->dm_dev
.bdev
= bdev
;
346 * Close a device that we've been using.
348 static void close_dev(struct dm_dev_internal
*d
, struct mapped_device
*md
)
353 bd_unlink_disk_holder(d
->dm_dev
.bdev
, dm_disk(md
));
354 blkdev_put(d
->dm_dev
.bdev
, d
->dm_dev
.mode
| FMODE_EXCL
);
355 d
->dm_dev
.bdev
= NULL
;
359 * If possible, this checks an area of a destination device is invalid.
361 static int device_area_is_invalid(struct dm_target
*ti
, struct dm_dev
*dev
,
362 sector_t start
, sector_t len
, void *data
)
364 struct request_queue
*q
;
365 struct queue_limits
*limits
= data
;
366 struct block_device
*bdev
= dev
->bdev
;
368 i_size_read(bdev
->bd_inode
) >> SECTOR_SHIFT
;
369 unsigned short logical_block_size_sectors
=
370 limits
->logical_block_size
>> SECTOR_SHIFT
;
371 char b
[BDEVNAME_SIZE
];
374 * Some devices exist without request functions,
375 * such as loop devices not yet bound to backing files.
376 * Forbid the use of such devices.
378 q
= bdev_get_queue(bdev
);
379 if (!q
|| !q
->make_request_fn
) {
380 DMWARN("%s: %s is not yet initialised: "
381 "start=%llu, len=%llu, dev_size=%llu",
382 dm_device_name(ti
->table
->md
), bdevname(bdev
, b
),
383 (unsigned long long)start
,
384 (unsigned long long)len
,
385 (unsigned long long)dev_size
);
392 if ((start
>= dev_size
) || (start
+ len
> dev_size
)) {
393 DMWARN("%s: %s too small for target: "
394 "start=%llu, len=%llu, dev_size=%llu",
395 dm_device_name(ti
->table
->md
), bdevname(bdev
, b
),
396 (unsigned long long)start
,
397 (unsigned long long)len
,
398 (unsigned long long)dev_size
);
402 if (logical_block_size_sectors
<= 1)
405 if (start
& (logical_block_size_sectors
- 1)) {
406 DMWARN("%s: start=%llu not aligned to h/w "
407 "logical block size %u of %s",
408 dm_device_name(ti
->table
->md
),
409 (unsigned long long)start
,
410 limits
->logical_block_size
, bdevname(bdev
, b
));
414 if (len
& (logical_block_size_sectors
- 1)) {
415 DMWARN("%s: len=%llu not aligned to h/w "
416 "logical block size %u of %s",
417 dm_device_name(ti
->table
->md
),
418 (unsigned long long)len
,
419 limits
->logical_block_size
, bdevname(bdev
, b
));
427 * This upgrades the mode on an already open dm_dev, being
428 * careful to leave things as they were if we fail to reopen the
429 * device and not to touch the existing bdev field in case
430 * it is accessed concurrently inside dm_table_any_congested().
432 static int upgrade_mode(struct dm_dev_internal
*dd
, fmode_t new_mode
,
433 struct mapped_device
*md
)
436 struct dm_dev_internal dd_new
, dd_old
;
438 dd_new
= dd_old
= *dd
;
440 dd_new
.dm_dev
.mode
|= new_mode
;
441 dd_new
.dm_dev
.bdev
= NULL
;
443 r
= open_dev(&dd_new
, dd
->dm_dev
.bdev
->bd_dev
, md
);
447 dd
->dm_dev
.mode
|= new_mode
;
448 close_dev(&dd_old
, md
);
454 * Add a device to the list, or just increment the usage count if
455 * it's already present.
457 int dm_get_device(struct dm_target
*ti
, const char *path
, fmode_t mode
,
458 struct dm_dev
**result
)
461 dev_t
uninitialized_var(dev
);
462 struct dm_dev_internal
*dd
;
463 unsigned int major
, minor
;
464 struct dm_table
*t
= ti
->table
;
469 if (sscanf(path
, "%u:%u%c", &major
, &minor
, &dummy
) == 2) {
470 /* Extract the major/minor numbers */
471 dev
= MKDEV(major
, minor
);
472 if (MAJOR(dev
) != major
|| MINOR(dev
) != minor
)
475 /* convert the path to a device */
476 struct block_device
*bdev
= lookup_bdev(path
);
479 return PTR_ERR(bdev
);
484 dd
= find_device(&t
->devices
, dev
);
486 dd
= kmalloc(sizeof(*dd
), GFP_KERNEL
);
490 dd
->dm_dev
.mode
= mode
;
491 dd
->dm_dev
.bdev
= NULL
;
493 if ((r
= open_dev(dd
, dev
, t
->md
))) {
498 format_dev_t(dd
->dm_dev
.name
, dev
);
500 atomic_set(&dd
->count
, 0);
501 list_add(&dd
->list
, &t
->devices
);
503 } else if (dd
->dm_dev
.mode
!= (mode
| dd
->dm_dev
.mode
)) {
504 r
= upgrade_mode(dd
, mode
, t
->md
);
508 atomic_inc(&dd
->count
);
510 *result
= &dd
->dm_dev
;
513 EXPORT_SYMBOL(dm_get_device
);
515 int dm_set_device_limits(struct dm_target
*ti
, struct dm_dev
*dev
,
516 sector_t start
, sector_t len
, void *data
)
518 struct queue_limits
*limits
= data
;
519 struct block_device
*bdev
= dev
->bdev
;
520 struct request_queue
*q
= bdev_get_queue(bdev
);
521 char b
[BDEVNAME_SIZE
];
524 DMWARN("%s: Cannot set limits for nonexistent device %s",
525 dm_device_name(ti
->table
->md
), bdevname(bdev
, b
));
529 if (bdev_stack_limits(limits
, bdev
, start
) < 0)
530 DMWARN("%s: adding target device %s caused an alignment inconsistency: "
531 "physical_block_size=%u, logical_block_size=%u, "
532 "alignment_offset=%u, start=%llu",
533 dm_device_name(ti
->table
->md
), bdevname(bdev
, b
),
534 q
->limits
.physical_block_size
,
535 q
->limits
.logical_block_size
,
536 q
->limits
.alignment_offset
,
537 (unsigned long long) start
<< SECTOR_SHIFT
);
540 * Check if merge fn is supported.
541 * If not we'll force DM to use PAGE_SIZE or
542 * smaller I/O, just to be safe.
544 if (dm_queue_merge_is_compulsory(q
) && !ti
->type
->merge
)
545 blk_limits_max_hw_sectors(limits
,
546 (unsigned int) (PAGE_SIZE
>> 9));
549 EXPORT_SYMBOL_GPL(dm_set_device_limits
);
552 * Decrement a device's use count and remove it if necessary.
554 void dm_put_device(struct dm_target
*ti
, struct dm_dev
*d
)
556 struct dm_dev_internal
*dd
= container_of(d
, struct dm_dev_internal
,
559 if (atomic_dec_and_test(&dd
->count
)) {
560 close_dev(dd
, ti
->table
->md
);
565 EXPORT_SYMBOL(dm_put_device
);
568 * Checks to see if the target joins onto the end of the table.
570 static int adjoin(struct dm_table
*table
, struct dm_target
*ti
)
572 struct dm_target
*prev
;
574 if (!table
->num_targets
)
577 prev
= &table
->targets
[table
->num_targets
- 1];
578 return (ti
->begin
== (prev
->begin
+ prev
->len
));
582 * Used to dynamically allocate the arg array.
584 static char **realloc_argv(unsigned *array_size
, char **old_argv
)
589 new_size
= *array_size
? *array_size
* 2 : 64;
590 argv
= kmalloc(new_size
* sizeof(*argv
), GFP_KERNEL
);
592 memcpy(argv
, old_argv
, *array_size
* sizeof(*argv
));
593 *array_size
= new_size
;
601 * Destructively splits up the argument list to pass to ctr.
603 int dm_split_args(int *argc
, char ***argvp
, char *input
)
605 char *start
, *end
= input
, *out
, **argv
= NULL
;
606 unsigned array_size
= 0;
615 argv
= realloc_argv(&array_size
, argv
);
620 /* Skip whitespace */
621 start
= skip_spaces(end
);
624 break; /* success, we hit the end */
626 /* 'out' is used to remove any back-quotes */
629 /* Everything apart from '\0' can be quoted */
630 if (*end
== '\\' && *(end
+ 1)) {
637 break; /* end of token */
642 /* have we already filled the array ? */
643 if ((*argc
+ 1) > array_size
) {
644 argv
= realloc_argv(&array_size
, argv
);
649 /* we know this is whitespace */
653 /* terminate the string and put it in the array */
664 * Impose necessary and sufficient conditions on a devices's table such
665 * that any incoming bio which respects its logical_block_size can be
666 * processed successfully. If it falls across the boundary between
667 * two or more targets, the size of each piece it gets split into must
668 * be compatible with the logical_block_size of the target processing it.
670 static int validate_hardware_logical_block_alignment(struct dm_table
*table
,
671 struct queue_limits
*limits
)
674 * This function uses arithmetic modulo the logical_block_size
675 * (in units of 512-byte sectors).
677 unsigned short device_logical_block_size_sects
=
678 limits
->logical_block_size
>> SECTOR_SHIFT
;
681 * Offset of the start of the next table entry, mod logical_block_size.
683 unsigned short next_target_start
= 0;
686 * Given an aligned bio that extends beyond the end of a
687 * target, how many sectors must the next target handle?
689 unsigned short remaining
= 0;
691 struct dm_target
*uninitialized_var(ti
);
692 struct queue_limits ti_limits
;
696 * Check each entry in the table in turn.
698 while (i
< dm_table_get_num_targets(table
)) {
699 ti
= dm_table_get_target(table
, i
++);
701 blk_set_stacking_limits(&ti_limits
);
703 /* combine all target devices' limits */
704 if (ti
->type
->iterate_devices
)
705 ti
->type
->iterate_devices(ti
, dm_set_device_limits
,
709 * If the remaining sectors fall entirely within this
710 * table entry are they compatible with its logical_block_size?
712 if (remaining
< ti
->len
&&
713 remaining
& ((ti_limits
.logical_block_size
>>
718 (unsigned short) ((next_target_start
+ ti
->len
) &
719 (device_logical_block_size_sects
- 1));
720 remaining
= next_target_start
?
721 device_logical_block_size_sects
- next_target_start
: 0;
725 DMWARN("%s: table line %u (start sect %llu len %llu) "
726 "not aligned to h/w logical block size %u",
727 dm_device_name(table
->md
), i
,
728 (unsigned long long) ti
->begin
,
729 (unsigned long long) ti
->len
,
730 limits
->logical_block_size
);
737 int dm_table_add_target(struct dm_table
*t
, const char *type
,
738 sector_t start
, sector_t len
, char *params
)
740 int r
= -EINVAL
, argc
;
742 struct dm_target
*tgt
;
745 DMERR("%s: target type %s must appear alone in table",
746 dm_device_name(t
->md
), t
->targets
->type
->name
);
750 if ((r
= check_space(t
)))
753 tgt
= t
->targets
+ t
->num_targets
;
754 memset(tgt
, 0, sizeof(*tgt
));
757 DMERR("%s: zero-length target", dm_device_name(t
->md
));
761 tgt
->type
= dm_get_target_type(type
);
763 DMERR("%s: %s: unknown target type", dm_device_name(t
->md
),
768 if (dm_target_needs_singleton(tgt
->type
)) {
769 if (t
->num_targets
) {
770 DMERR("%s: target type %s must appear alone in table",
771 dm_device_name(t
->md
), type
);
777 if (dm_target_always_writeable(tgt
->type
) && !(t
->mode
& FMODE_WRITE
)) {
778 DMERR("%s: target type %s may not be included in read-only tables",
779 dm_device_name(t
->md
), type
);
783 if (t
->immutable_target_type
) {
784 if (t
->immutable_target_type
!= tgt
->type
) {
785 DMERR("%s: immutable target type %s cannot be mixed with other target types",
786 dm_device_name(t
->md
), t
->immutable_target_type
->name
);
789 } else if (dm_target_is_immutable(tgt
->type
)) {
790 if (t
->num_targets
) {
791 DMERR("%s: immutable target type %s cannot be mixed with other target types",
792 dm_device_name(t
->md
), tgt
->type
->name
);
795 t
->immutable_target_type
= tgt
->type
;
801 tgt
->error
= "Unknown error";
804 * Does this target adjoin the previous one ?
806 if (!adjoin(t
, tgt
)) {
807 tgt
->error
= "Gap in table";
812 r
= dm_split_args(&argc
, &argv
, params
);
814 tgt
->error
= "couldn't split parameters (insufficient memory)";
818 r
= tgt
->type
->ctr(tgt
, argc
, argv
);
823 t
->highs
[t
->num_targets
++] = tgt
->begin
+ tgt
->len
- 1;
825 if (!tgt
->num_discard_bios
&& tgt
->discards_supported
)
826 DMWARN("%s: %s: ignoring discards_supported because num_discard_bios is zero.",
827 dm_device_name(t
->md
), type
);
832 DMERR("%s: %s: %s", dm_device_name(t
->md
), type
, tgt
->error
);
833 dm_put_target_type(tgt
->type
);
838 * Target argument parsing helpers.
840 static int validate_next_arg(struct dm_arg
*arg
, struct dm_arg_set
*arg_set
,
841 unsigned *value
, char **error
, unsigned grouped
)
843 const char *arg_str
= dm_shift_arg(arg_set
);
847 (sscanf(arg_str
, "%u%c", value
, &dummy
) != 1) ||
848 (*value
< arg
->min
) ||
849 (*value
> arg
->max
) ||
850 (grouped
&& arg_set
->argc
< *value
)) {
858 int dm_read_arg(struct dm_arg
*arg
, struct dm_arg_set
*arg_set
,
859 unsigned *value
, char **error
)
861 return validate_next_arg(arg
, arg_set
, value
, error
, 0);
863 EXPORT_SYMBOL(dm_read_arg
);
865 int dm_read_arg_group(struct dm_arg
*arg
, struct dm_arg_set
*arg_set
,
866 unsigned *value
, char **error
)
868 return validate_next_arg(arg
, arg_set
, value
, error
, 1);
870 EXPORT_SYMBOL(dm_read_arg_group
);
872 const char *dm_shift_arg(struct dm_arg_set
*as
)
885 EXPORT_SYMBOL(dm_shift_arg
);
887 void dm_consume_args(struct dm_arg_set
*as
, unsigned num_args
)
889 BUG_ON(as
->argc
< num_args
);
890 as
->argc
-= num_args
;
891 as
->argv
+= num_args
;
893 EXPORT_SYMBOL(dm_consume_args
);
895 static int dm_table_set_type(struct dm_table
*t
)
898 unsigned bio_based
= 0, request_based
= 0;
899 struct dm_target
*tgt
;
900 struct dm_dev_internal
*dd
;
901 struct list_head
*devices
;
903 for (i
= 0; i
< t
->num_targets
; i
++) {
904 tgt
= t
->targets
+ i
;
905 if (dm_target_request_based(tgt
))
910 if (bio_based
&& request_based
) {
911 DMWARN("Inconsistent table: different target types"
912 " can't be mixed up");
918 /* We must use this table as bio-based */
919 t
->type
= DM_TYPE_BIO_BASED
;
923 BUG_ON(!request_based
); /* No targets in this table */
925 /* Non-request-stackable devices can't be used for request-based dm */
926 devices
= dm_table_get_devices(t
);
927 list_for_each_entry(dd
, devices
, list
) {
928 if (!blk_queue_stackable(bdev_get_queue(dd
->dm_dev
.bdev
))) {
929 DMWARN("table load rejected: including"
930 " non-request-stackable devices");
936 * Request-based dm supports only tables that have a single target now.
937 * To support multiple targets, request splitting support is needed,
938 * and that needs lots of changes in the block-layer.
939 * (e.g. request completion process for partial completion.)
941 if (t
->num_targets
> 1) {
942 DMWARN("Request-based dm doesn't support multiple targets yet");
946 t
->type
= DM_TYPE_REQUEST_BASED
;
951 unsigned dm_table_get_type(struct dm_table
*t
)
956 struct target_type
*dm_table_get_immutable_target_type(struct dm_table
*t
)
958 return t
->immutable_target_type
;
961 bool dm_table_request_based(struct dm_table
*t
)
963 return dm_table_get_type(t
) == DM_TYPE_REQUEST_BASED
;
966 int dm_table_alloc_md_mempools(struct dm_table
*t
)
968 unsigned type
= dm_table_get_type(t
);
969 unsigned per_bio_data_size
= 0;
970 struct dm_target
*tgt
;
973 if (unlikely(type
== DM_TYPE_NONE
)) {
974 DMWARN("no table type is set, can't allocate mempools");
978 if (type
== DM_TYPE_BIO_BASED
)
979 for (i
= 0; i
< t
->num_targets
; i
++) {
980 tgt
= t
->targets
+ i
;
981 per_bio_data_size
= max(per_bio_data_size
, tgt
->per_bio_data_size
);
984 t
->mempools
= dm_alloc_md_mempools(type
, t
->integrity_supported
, per_bio_data_size
);
991 void dm_table_free_md_mempools(struct dm_table
*t
)
993 dm_free_md_mempools(t
->mempools
);
997 struct dm_md_mempools
*dm_table_get_md_mempools(struct dm_table
*t
)
1002 static int setup_indexes(struct dm_table
*t
)
1005 unsigned int total
= 0;
1008 /* allocate the space for *all* the indexes */
1009 for (i
= t
->depth
- 2; i
>= 0; i
--) {
1010 t
->counts
[i
] = dm_div_up(t
->counts
[i
+ 1], CHILDREN_PER_NODE
);
1011 total
+= t
->counts
[i
];
1014 indexes
= (sector_t
*) dm_vcalloc(total
, (unsigned long) NODE_SIZE
);
1018 /* set up internal nodes, bottom-up */
1019 for (i
= t
->depth
- 2; i
>= 0; i
--) {
1020 t
->index
[i
] = indexes
;
1021 indexes
+= (KEYS_PER_NODE
* t
->counts
[i
]);
1022 setup_btree_index(i
, t
);
1029 * Builds the btree to index the map.
1031 static int dm_table_build_index(struct dm_table
*t
)
1034 unsigned int leaf_nodes
;
1036 /* how many indexes will the btree have ? */
1037 leaf_nodes
= dm_div_up(t
->num_targets
, KEYS_PER_NODE
);
1038 t
->depth
= 1 + int_log(leaf_nodes
, CHILDREN_PER_NODE
);
1040 /* leaf layer has already been set up */
1041 t
->counts
[t
->depth
- 1] = leaf_nodes
;
1042 t
->index
[t
->depth
- 1] = t
->highs
;
1045 r
= setup_indexes(t
);
1051 * Get a disk whose integrity profile reflects the table's profile.
1052 * If %match_all is true, all devices' profiles must match.
1053 * If %match_all is false, all devices must at least have an
1054 * allocated integrity profile; but uninitialized is ok.
1055 * Returns NULL if integrity support was inconsistent or unavailable.
1057 static struct gendisk
* dm_table_get_integrity_disk(struct dm_table
*t
,
1060 struct list_head
*devices
= dm_table_get_devices(t
);
1061 struct dm_dev_internal
*dd
= NULL
;
1062 struct gendisk
*prev_disk
= NULL
, *template_disk
= NULL
;
1064 list_for_each_entry(dd
, devices
, list
) {
1065 template_disk
= dd
->dm_dev
.bdev
->bd_disk
;
1066 if (!blk_get_integrity(template_disk
))
1068 if (!match_all
&& !blk_integrity_is_initialized(template_disk
))
1069 continue; /* skip uninitialized profiles */
1070 else if (prev_disk
&&
1071 blk_integrity_compare(prev_disk
, template_disk
) < 0)
1073 prev_disk
= template_disk
;
1076 return template_disk
;
1080 DMWARN("%s: integrity not set: %s and %s profile mismatch",
1081 dm_device_name(t
->md
),
1082 prev_disk
->disk_name
,
1083 template_disk
->disk_name
);
1088 * Register the mapped device for blk_integrity support if
1089 * the underlying devices have an integrity profile. But all devices
1090 * may not have matching profiles (checking all devices isn't reliable
1091 * during table load because this table may use other DM device(s) which
1092 * must be resumed before they will have an initialized integity profile).
1093 * Stacked DM devices force a 2 stage integrity profile validation:
1094 * 1 - during load, validate all initialized integrity profiles match
1095 * 2 - during resume, validate all integrity profiles match
1097 static int dm_table_prealloc_integrity(struct dm_table
*t
, struct mapped_device
*md
)
1099 struct gendisk
*template_disk
= NULL
;
1101 template_disk
= dm_table_get_integrity_disk(t
, false);
1105 if (!blk_integrity_is_initialized(dm_disk(md
))) {
1106 t
->integrity_supported
= 1;
1107 return blk_integrity_register(dm_disk(md
), NULL
);
1111 * If DM device already has an initalized integrity
1112 * profile the new profile should not conflict.
1114 if (blk_integrity_is_initialized(template_disk
) &&
1115 blk_integrity_compare(dm_disk(md
), template_disk
) < 0) {
1116 DMWARN("%s: conflict with existing integrity profile: "
1117 "%s profile mismatch",
1118 dm_device_name(t
->md
),
1119 template_disk
->disk_name
);
1123 /* Preserve existing initialized integrity profile */
1124 t
->integrity_supported
= 1;
1129 * Prepares the table for use by building the indices,
1130 * setting the type, and allocating mempools.
1132 int dm_table_complete(struct dm_table
*t
)
1136 r
= dm_table_set_type(t
);
1138 DMERR("unable to set table type");
1142 r
= dm_table_build_index(t
);
1144 DMERR("unable to build btrees");
1148 r
= dm_table_prealloc_integrity(t
, t
->md
);
1150 DMERR("could not register integrity profile.");
1154 r
= dm_table_alloc_md_mempools(t
);
1156 DMERR("unable to allocate mempools");
1161 static DEFINE_MUTEX(_event_lock
);
1162 void dm_table_event_callback(struct dm_table
*t
,
1163 void (*fn
)(void *), void *context
)
1165 mutex_lock(&_event_lock
);
1167 t
->event_context
= context
;
1168 mutex_unlock(&_event_lock
);
1171 void dm_table_event(struct dm_table
*t
)
1174 * You can no longer call dm_table_event() from interrupt
1175 * context, use a bottom half instead.
1177 BUG_ON(in_interrupt());
1179 mutex_lock(&_event_lock
);
1181 t
->event_fn(t
->event_context
);
1182 mutex_unlock(&_event_lock
);
1184 EXPORT_SYMBOL(dm_table_event
);
1186 sector_t
dm_table_get_size(struct dm_table
*t
)
1188 return t
->num_targets
? (t
->highs
[t
->num_targets
- 1] + 1) : 0;
1190 EXPORT_SYMBOL(dm_table_get_size
);
1192 struct dm_target
*dm_table_get_target(struct dm_table
*t
, unsigned int index
)
1194 if (index
>= t
->num_targets
)
1197 return t
->targets
+ index
;
1201 * Search the btree for the correct target.
1203 * Caller should check returned pointer with dm_target_is_valid()
1204 * to trap I/O beyond end of device.
1206 struct dm_target
*dm_table_find_target(struct dm_table
*t
, sector_t sector
)
1208 unsigned int l
, n
= 0, k
= 0;
1211 for (l
= 0; l
< t
->depth
; l
++) {
1212 n
= get_child(n
, k
);
1213 node
= get_node(t
, l
, n
);
1215 for (k
= 0; k
< KEYS_PER_NODE
; k
++)
1216 if (node
[k
] >= sector
)
1220 return &t
->targets
[(KEYS_PER_NODE
* n
) + k
];
1223 static int count_device(struct dm_target
*ti
, struct dm_dev
*dev
,
1224 sector_t start
, sector_t len
, void *data
)
1226 unsigned *num_devices
= data
;
1234 * Check whether a table has no data devices attached using each
1235 * target's iterate_devices method.
1236 * Returns false if the result is unknown because a target doesn't
1237 * support iterate_devices.
1239 bool dm_table_has_no_data_devices(struct dm_table
*table
)
1241 struct dm_target
*uninitialized_var(ti
);
1242 unsigned i
= 0, num_devices
= 0;
1244 while (i
< dm_table_get_num_targets(table
)) {
1245 ti
= dm_table_get_target(table
, i
++);
1247 if (!ti
->type
->iterate_devices
)
1250 ti
->type
->iterate_devices(ti
, count_device
, &num_devices
);
1259 * Establish the new table's queue_limits and validate them.
1261 int dm_calculate_queue_limits(struct dm_table
*table
,
1262 struct queue_limits
*limits
)
1264 struct dm_target
*uninitialized_var(ti
);
1265 struct queue_limits ti_limits
;
1268 blk_set_stacking_limits(limits
);
1270 while (i
< dm_table_get_num_targets(table
)) {
1271 blk_set_stacking_limits(&ti_limits
);
1273 ti
= dm_table_get_target(table
, i
++);
1275 if (!ti
->type
->iterate_devices
)
1276 goto combine_limits
;
1279 * Combine queue limits of all the devices this target uses.
1281 ti
->type
->iterate_devices(ti
, dm_set_device_limits
,
1284 /* Set I/O hints portion of queue limits */
1285 if (ti
->type
->io_hints
)
1286 ti
->type
->io_hints(ti
, &ti_limits
);
1289 * Check each device area is consistent with the target's
1290 * overall queue limits.
1292 if (ti
->type
->iterate_devices(ti
, device_area_is_invalid
,
1298 * Merge this target's queue limits into the overall limits
1301 if (blk_stack_limits(limits
, &ti_limits
, 0) < 0)
1302 DMWARN("%s: adding target device "
1303 "(start sect %llu len %llu) "
1304 "caused an alignment inconsistency",
1305 dm_device_name(table
->md
),
1306 (unsigned long long) ti
->begin
,
1307 (unsigned long long) ti
->len
);
1310 return validate_hardware_logical_block_alignment(table
, limits
);
1314 * Set the integrity profile for this device if all devices used have
1315 * matching profiles. We're quite deep in the resume path but still
1316 * don't know if all devices (particularly DM devices this device
1317 * may be stacked on) have matching profiles. Even if the profiles
1318 * don't match we have no way to fail (to resume) at this point.
1320 static void dm_table_set_integrity(struct dm_table
*t
)
1322 struct gendisk
*template_disk
= NULL
;
1324 if (!blk_get_integrity(dm_disk(t
->md
)))
1327 template_disk
= dm_table_get_integrity_disk(t
, true);
1329 blk_integrity_register(dm_disk(t
->md
),
1330 blk_get_integrity(template_disk
));
1331 else if (blk_integrity_is_initialized(dm_disk(t
->md
)))
1332 DMWARN("%s: device no longer has a valid integrity profile",
1333 dm_device_name(t
->md
));
1335 DMWARN("%s: unable to establish an integrity profile",
1336 dm_device_name(t
->md
));
1339 static int device_flush_capable(struct dm_target
*ti
, struct dm_dev
*dev
,
1340 sector_t start
, sector_t len
, void *data
)
1342 unsigned flush
= (*(unsigned *)data
);
1343 struct request_queue
*q
= bdev_get_queue(dev
->bdev
);
1345 return q
&& (q
->flush_flags
& flush
);
1348 static bool dm_table_supports_flush(struct dm_table
*t
, unsigned flush
)
1350 struct dm_target
*ti
;
1354 * Require at least one underlying device to support flushes.
1355 * t->devices includes internal dm devices such as mirror logs
1356 * so we need to use iterate_devices here, which targets
1357 * supporting flushes must provide.
1359 while (i
< dm_table_get_num_targets(t
)) {
1360 ti
= dm_table_get_target(t
, i
++);
1362 if (!ti
->num_flush_bios
)
1365 if (ti
->flush_supported
)
1368 if (ti
->type
->iterate_devices
&&
1369 ti
->type
->iterate_devices(ti
, device_flush_capable
, &flush
))
1376 static bool dm_table_discard_zeroes_data(struct dm_table
*t
)
1378 struct dm_target
*ti
;
1381 /* Ensure that all targets supports discard_zeroes_data. */
1382 while (i
< dm_table_get_num_targets(t
)) {
1383 ti
= dm_table_get_target(t
, i
++);
1385 if (ti
->discard_zeroes_data_unsupported
)
1392 static int device_is_nonrot(struct dm_target
*ti
, struct dm_dev
*dev
,
1393 sector_t start
, sector_t len
, void *data
)
1395 struct request_queue
*q
= bdev_get_queue(dev
->bdev
);
1397 return q
&& blk_queue_nonrot(q
);
1400 static int device_is_not_random(struct dm_target
*ti
, struct dm_dev
*dev
,
1401 sector_t start
, sector_t len
, void *data
)
1403 struct request_queue
*q
= bdev_get_queue(dev
->bdev
);
1405 return q
&& !blk_queue_add_random(q
);
1408 static bool dm_table_all_devices_attribute(struct dm_table
*t
,
1409 iterate_devices_callout_fn func
)
1411 struct dm_target
*ti
;
1414 while (i
< dm_table_get_num_targets(t
)) {
1415 ti
= dm_table_get_target(t
, i
++);
1417 if (!ti
->type
->iterate_devices
||
1418 !ti
->type
->iterate_devices(ti
, func
, NULL
))
1425 static int device_not_write_same_capable(struct dm_target
*ti
, struct dm_dev
*dev
,
1426 sector_t start
, sector_t len
, void *data
)
1428 struct request_queue
*q
= bdev_get_queue(dev
->bdev
);
1430 return q
&& !q
->limits
.max_write_same_sectors
;
1433 static bool dm_table_supports_write_same(struct dm_table
*t
)
1435 struct dm_target
*ti
;
1438 while (i
< dm_table_get_num_targets(t
)) {
1439 ti
= dm_table_get_target(t
, i
++);
1441 if (!ti
->num_write_same_bios
)
1444 if (!ti
->type
->iterate_devices
||
1445 !ti
->type
->iterate_devices(ti
, device_not_write_same_capable
, NULL
))
1452 void dm_table_set_restrictions(struct dm_table
*t
, struct request_queue
*q
,
1453 struct queue_limits
*limits
)
1458 * Copy table's limits to the DM device's request_queue
1460 q
->limits
= *limits
;
1462 if (!dm_table_supports_discards(t
))
1463 queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD
, q
);
1465 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD
, q
);
1467 if (dm_table_supports_flush(t
, REQ_FLUSH
)) {
1469 if (dm_table_supports_flush(t
, REQ_FUA
))
1472 blk_queue_flush(q
, flush
);
1474 if (!dm_table_discard_zeroes_data(t
))
1475 q
->limits
.discard_zeroes_data
= 0;
1477 /* Ensure that all underlying devices are non-rotational. */
1478 if (dm_table_all_devices_attribute(t
, device_is_nonrot
))
1479 queue_flag_set_unlocked(QUEUE_FLAG_NONROT
, q
);
1481 queue_flag_clear_unlocked(QUEUE_FLAG_NONROT
, q
);
1483 if (!dm_table_supports_write_same(t
))
1484 q
->limits
.max_write_same_sectors
= 0;
1486 dm_table_set_integrity(t
);
1489 * Determine whether or not this queue's I/O timings contribute
1490 * to the entropy pool, Only request-based targets use this.
1491 * Clear QUEUE_FLAG_ADD_RANDOM if any underlying device does not
1494 if (blk_queue_add_random(q
) && dm_table_all_devices_attribute(t
, device_is_not_random
))
1495 queue_flag_clear_unlocked(QUEUE_FLAG_ADD_RANDOM
, q
);
1498 * QUEUE_FLAG_STACKABLE must be set after all queue settings are
1499 * visible to other CPUs because, once the flag is set, incoming bios
1500 * are processed by request-based dm, which refers to the queue
1502 * Until the flag set, bios are passed to bio-based dm and queued to
1503 * md->deferred where queue settings are not needed yet.
1504 * Those bios are passed to request-based dm at the resume time.
1507 if (dm_table_request_based(t
))
1508 queue_flag_set_unlocked(QUEUE_FLAG_STACKABLE
, q
);
1511 unsigned int dm_table_get_num_targets(struct dm_table
*t
)
1513 return t
->num_targets
;
1516 struct list_head
*dm_table_get_devices(struct dm_table
*t
)
1521 fmode_t
dm_table_get_mode(struct dm_table
*t
)
1525 EXPORT_SYMBOL(dm_table_get_mode
);
1527 static void suspend_targets(struct dm_table
*t
, unsigned postsuspend
)
1529 int i
= t
->num_targets
;
1530 struct dm_target
*ti
= t
->targets
;
1534 if (ti
->type
->postsuspend
)
1535 ti
->type
->postsuspend(ti
);
1536 } else if (ti
->type
->presuspend
)
1537 ti
->type
->presuspend(ti
);
1543 void dm_table_presuspend_targets(struct dm_table
*t
)
1548 suspend_targets(t
, 0);
1551 void dm_table_postsuspend_targets(struct dm_table
*t
)
1556 suspend_targets(t
, 1);
1559 int dm_table_resume_targets(struct dm_table
*t
)
1563 for (i
= 0; i
< t
->num_targets
; i
++) {
1564 struct dm_target
*ti
= t
->targets
+ i
;
1566 if (!ti
->type
->preresume
)
1569 r
= ti
->type
->preresume(ti
);
1574 for (i
= 0; i
< t
->num_targets
; i
++) {
1575 struct dm_target
*ti
= t
->targets
+ i
;
1577 if (ti
->type
->resume
)
1578 ti
->type
->resume(ti
);
1584 void dm_table_add_target_callbacks(struct dm_table
*t
, struct dm_target_callbacks
*cb
)
1586 list_add(&cb
->list
, &t
->target_callbacks
);
1588 EXPORT_SYMBOL_GPL(dm_table_add_target_callbacks
);
1590 int dm_table_any_congested(struct dm_table
*t
, int bdi_bits
)
1592 struct dm_dev_internal
*dd
;
1593 struct list_head
*devices
= dm_table_get_devices(t
);
1594 struct dm_target_callbacks
*cb
;
1597 list_for_each_entry(dd
, devices
, list
) {
1598 struct request_queue
*q
= bdev_get_queue(dd
->dm_dev
.bdev
);
1599 char b
[BDEVNAME_SIZE
];
1602 r
|= bdi_congested(&q
->backing_dev_info
, bdi_bits
);
1604 DMWARN_LIMIT("%s: any_congested: nonexistent device %s",
1605 dm_device_name(t
->md
),
1606 bdevname(dd
->dm_dev
.bdev
, b
));
1609 list_for_each_entry(cb
, &t
->target_callbacks
, list
)
1610 if (cb
->congested_fn
)
1611 r
|= cb
->congested_fn(cb
, bdi_bits
);
1616 int dm_table_any_busy_target(struct dm_table
*t
)
1619 struct dm_target
*ti
;
1621 for (i
= 0; i
< t
->num_targets
; i
++) {
1622 ti
= t
->targets
+ i
;
1623 if (ti
->type
->busy
&& ti
->type
->busy(ti
))
1630 struct mapped_device
*dm_table_get_md(struct dm_table
*t
)
1634 EXPORT_SYMBOL(dm_table_get_md
);
1636 static int device_discard_capable(struct dm_target
*ti
, struct dm_dev
*dev
,
1637 sector_t start
, sector_t len
, void *data
)
1639 struct request_queue
*q
= bdev_get_queue(dev
->bdev
);
1641 return q
&& blk_queue_discard(q
);
1644 bool dm_table_supports_discards(struct dm_table
*t
)
1646 struct dm_target
*ti
;
1650 * Unless any target used by the table set discards_supported,
1651 * require at least one underlying device to support discards.
1652 * t->devices includes internal dm devices such as mirror logs
1653 * so we need to use iterate_devices here, which targets
1654 * supporting discard selectively must provide.
1656 while (i
< dm_table_get_num_targets(t
)) {
1657 ti
= dm_table_get_target(t
, i
++);
1659 if (!ti
->num_discard_bios
)
1662 if (ti
->discards_supported
)
1665 if (ti
->type
->iterate_devices
&&
1666 ti
->type
->iterate_devices(ti
, device_discard_capable
, NULL
))