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 struct mapped_device
*md
;
35 unsigned int counts
[MAX_DEPTH
]; /* in nodes */
36 sector_t
*index
[MAX_DEPTH
];
38 unsigned int num_targets
;
39 unsigned int num_allocated
;
41 struct dm_target
*targets
;
43 struct target_type
*immutable_target_type
;
44 unsigned integrity_supported
:1;
48 * Indicates the rw permissions for the new logical
49 * device. This should be a combination of FMODE_READ
54 /* a list of devices used by this table */
55 struct list_head devices
;
57 /* events get handed up using this callback */
58 void (*event_fn
)(void *);
61 struct dm_md_mempools
*mempools
;
63 struct list_head target_callbacks
;
67 * Similar to ceiling(log_size(n))
69 static unsigned int int_log(unsigned int n
, unsigned int base
)
74 n
= dm_div_up(n
, base
);
82 * Calculate the index of the child node of the n'th node k'th key.
84 static inline unsigned int get_child(unsigned int n
, unsigned int k
)
86 return (n
* CHILDREN_PER_NODE
) + k
;
90 * Return the n'th node of level l from table t.
92 static inline sector_t
*get_node(struct dm_table
*t
,
93 unsigned int l
, unsigned int n
)
95 return t
->index
[l
] + (n
* KEYS_PER_NODE
);
99 * Return the highest key that you could lookup from the n'th
100 * node on level l of the btree.
102 static sector_t
high(struct dm_table
*t
, unsigned int l
, unsigned int n
)
104 for (; l
< t
->depth
- 1; l
++)
105 n
= get_child(n
, CHILDREN_PER_NODE
- 1);
107 if (n
>= t
->counts
[l
])
108 return (sector_t
) - 1;
110 return get_node(t
, l
, n
)[KEYS_PER_NODE
- 1];
114 * Fills in a level of the btree based on the highs of the level
117 static int setup_btree_index(unsigned int l
, struct dm_table
*t
)
122 for (n
= 0U; n
< t
->counts
[l
]; n
++) {
123 node
= get_node(t
, l
, n
);
125 for (k
= 0U; k
< KEYS_PER_NODE
; k
++)
126 node
[k
] = high(t
, l
+ 1, get_child(n
, k
));
132 void *dm_vcalloc(unsigned long nmemb
, unsigned long elem_size
)
138 * Check that we're not going to overflow.
140 if (nmemb
> (ULONG_MAX
/ elem_size
))
143 size
= nmemb
* elem_size
;
144 addr
= vzalloc(size
);
148 EXPORT_SYMBOL(dm_vcalloc
);
151 * highs, and targets are managed as dynamic arrays during a
154 static int alloc_targets(struct dm_table
*t
, unsigned int num
)
157 struct dm_target
*n_targets
;
158 int n
= t
->num_targets
;
161 * Allocate both the target array and offset array at once.
162 * Append an empty entry to catch sectors beyond the end of
165 n_highs
= (sector_t
*) dm_vcalloc(num
+ 1, sizeof(struct dm_target
) +
170 n_targets
= (struct dm_target
*) (n_highs
+ num
);
173 memcpy(n_highs
, t
->highs
, sizeof(*n_highs
) * n
);
174 memcpy(n_targets
, t
->targets
, sizeof(*n_targets
) * n
);
177 memset(n_highs
+ n
, -1, sizeof(*n_highs
) * (num
- n
));
180 t
->num_allocated
= num
;
182 t
->targets
= n_targets
;
187 int dm_table_create(struct dm_table
**result
, fmode_t mode
,
188 unsigned num_targets
, struct mapped_device
*md
)
190 struct dm_table
*t
= kzalloc(sizeof(*t
), GFP_KERNEL
);
195 INIT_LIST_HEAD(&t
->devices
);
196 INIT_LIST_HEAD(&t
->target_callbacks
);
199 num_targets
= KEYS_PER_NODE
;
201 num_targets
= dm_round_up(num_targets
, KEYS_PER_NODE
);
203 if (alloc_targets(t
, num_targets
)) {
214 static void free_devices(struct list_head
*devices
)
216 struct list_head
*tmp
, *next
;
218 list_for_each_safe(tmp
, next
, devices
) {
219 struct dm_dev_internal
*dd
=
220 list_entry(tmp
, struct dm_dev_internal
, list
);
221 DMWARN("dm_table_destroy: dm_put_device call missing for %s",
227 void dm_table_destroy(struct dm_table
*t
)
234 /* free the indexes */
236 vfree(t
->index
[t
->depth
- 2]);
238 /* free the targets */
239 for (i
= 0; i
< t
->num_targets
; i
++) {
240 struct dm_target
*tgt
= t
->targets
+ i
;
245 dm_put_target_type(tgt
->type
);
250 /* free the device list */
251 free_devices(&t
->devices
);
253 dm_free_md_mempools(t
->mempools
);
259 * Checks to see if we need to extend highs or targets.
261 static inline int check_space(struct dm_table
*t
)
263 if (t
->num_targets
>= t
->num_allocated
)
264 return alloc_targets(t
, t
->num_allocated
* 2);
270 * See if we've already got a device in the list.
272 static struct dm_dev_internal
*find_device(struct list_head
*l
, dev_t dev
)
274 struct dm_dev_internal
*dd
;
276 list_for_each_entry (dd
, l
, list
)
277 if (dd
->dm_dev
.bdev
->bd_dev
== dev
)
284 * Open a device so we can use it as a map destination.
286 static int open_dev(struct dm_dev_internal
*d
, dev_t dev
,
287 struct mapped_device
*md
)
289 static char *_claim_ptr
= "I belong to device-mapper";
290 struct block_device
*bdev
;
294 BUG_ON(d
->dm_dev
.bdev
);
296 bdev
= blkdev_get_by_dev(dev
, d
->dm_dev
.mode
| FMODE_EXCL
, _claim_ptr
);
298 return PTR_ERR(bdev
);
300 r
= bd_link_disk_holder(bdev
, dm_disk(md
));
302 blkdev_put(bdev
, d
->dm_dev
.mode
| FMODE_EXCL
);
306 d
->dm_dev
.bdev
= bdev
;
311 * Close a device that we've been using.
313 static void close_dev(struct dm_dev_internal
*d
, struct mapped_device
*md
)
318 bd_unlink_disk_holder(d
->dm_dev
.bdev
, dm_disk(md
));
319 blkdev_put(d
->dm_dev
.bdev
, d
->dm_dev
.mode
| FMODE_EXCL
);
320 d
->dm_dev
.bdev
= NULL
;
324 * If possible, this checks an area of a destination device is invalid.
326 static int device_area_is_invalid(struct dm_target
*ti
, struct dm_dev
*dev
,
327 sector_t start
, sector_t len
, void *data
)
329 struct request_queue
*q
;
330 struct queue_limits
*limits
= data
;
331 struct block_device
*bdev
= dev
->bdev
;
333 i_size_read(bdev
->bd_inode
) >> SECTOR_SHIFT
;
334 unsigned short logical_block_size_sectors
=
335 limits
->logical_block_size
>> SECTOR_SHIFT
;
336 char b
[BDEVNAME_SIZE
];
339 * Some devices exist without request functions,
340 * such as loop devices not yet bound to backing files.
341 * Forbid the use of such devices.
343 q
= bdev_get_queue(bdev
);
344 if (!q
|| !q
->make_request_fn
) {
345 DMWARN("%s: %s is not yet initialised: "
346 "start=%llu, len=%llu, dev_size=%llu",
347 dm_device_name(ti
->table
->md
), bdevname(bdev
, b
),
348 (unsigned long long)start
,
349 (unsigned long long)len
,
350 (unsigned long long)dev_size
);
357 if ((start
>= dev_size
) || (start
+ len
> dev_size
)) {
358 DMWARN("%s: %s too small for target: "
359 "start=%llu, len=%llu, dev_size=%llu",
360 dm_device_name(ti
->table
->md
), bdevname(bdev
, b
),
361 (unsigned long long)start
,
362 (unsigned long long)len
,
363 (unsigned long long)dev_size
);
367 if (logical_block_size_sectors
<= 1)
370 if (start
& (logical_block_size_sectors
- 1)) {
371 DMWARN("%s: start=%llu not aligned to h/w "
372 "logical block size %u of %s",
373 dm_device_name(ti
->table
->md
),
374 (unsigned long long)start
,
375 limits
->logical_block_size
, bdevname(bdev
, b
));
379 if (len
& (logical_block_size_sectors
- 1)) {
380 DMWARN("%s: len=%llu not aligned to h/w "
381 "logical block size %u of %s",
382 dm_device_name(ti
->table
->md
),
383 (unsigned long long)len
,
384 limits
->logical_block_size
, bdevname(bdev
, b
));
392 * This upgrades the mode on an already open dm_dev, being
393 * careful to leave things as they were if we fail to reopen the
394 * device and not to touch the existing bdev field in case
395 * it is accessed concurrently inside dm_table_any_congested().
397 static int upgrade_mode(struct dm_dev_internal
*dd
, fmode_t new_mode
,
398 struct mapped_device
*md
)
401 struct dm_dev_internal dd_new
, dd_old
;
403 dd_new
= dd_old
= *dd
;
405 dd_new
.dm_dev
.mode
|= new_mode
;
406 dd_new
.dm_dev
.bdev
= NULL
;
408 r
= open_dev(&dd_new
, dd
->dm_dev
.bdev
->bd_dev
, md
);
412 dd
->dm_dev
.mode
|= new_mode
;
413 close_dev(&dd_old
, md
);
419 * Add a device to the list, or just increment the usage count if
420 * it's already present.
422 int dm_get_device(struct dm_target
*ti
, const char *path
, fmode_t mode
,
423 struct dm_dev
**result
)
426 dev_t
uninitialized_var(dev
);
427 struct dm_dev_internal
*dd
;
428 unsigned int major
, minor
;
429 struct dm_table
*t
= ti
->table
;
434 if (sscanf(path
, "%u:%u%c", &major
, &minor
, &dummy
) == 2) {
435 /* Extract the major/minor numbers */
436 dev
= MKDEV(major
, minor
);
437 if (MAJOR(dev
) != major
|| MINOR(dev
) != minor
)
440 /* convert the path to a device */
441 struct block_device
*bdev
= lookup_bdev(path
);
444 return PTR_ERR(bdev
);
449 dd
= find_device(&t
->devices
, dev
);
451 dd
= kmalloc(sizeof(*dd
), GFP_KERNEL
);
455 dd
->dm_dev
.mode
= mode
;
456 dd
->dm_dev
.bdev
= NULL
;
458 if ((r
= open_dev(dd
, dev
, t
->md
))) {
463 format_dev_t(dd
->dm_dev
.name
, dev
);
465 atomic_set(&dd
->count
, 0);
466 list_add(&dd
->list
, &t
->devices
);
468 } else if (dd
->dm_dev
.mode
!= (mode
| dd
->dm_dev
.mode
)) {
469 r
= upgrade_mode(dd
, mode
, t
->md
);
473 atomic_inc(&dd
->count
);
475 *result
= &dd
->dm_dev
;
478 EXPORT_SYMBOL(dm_get_device
);
480 int dm_set_device_limits(struct dm_target
*ti
, struct dm_dev
*dev
,
481 sector_t start
, sector_t len
, void *data
)
483 struct queue_limits
*limits
= data
;
484 struct block_device
*bdev
= dev
->bdev
;
485 struct request_queue
*q
= bdev_get_queue(bdev
);
486 char b
[BDEVNAME_SIZE
];
489 DMWARN("%s: Cannot set limits for nonexistent device %s",
490 dm_device_name(ti
->table
->md
), bdevname(bdev
, b
));
494 if (bdev_stack_limits(limits
, bdev
, start
) < 0)
495 DMWARN("%s: adding target device %s caused an alignment inconsistency: "
496 "physical_block_size=%u, logical_block_size=%u, "
497 "alignment_offset=%u, start=%llu",
498 dm_device_name(ti
->table
->md
), bdevname(bdev
, b
),
499 q
->limits
.physical_block_size
,
500 q
->limits
.logical_block_size
,
501 q
->limits
.alignment_offset
,
502 (unsigned long long) start
<< SECTOR_SHIFT
);
505 * Check if merge fn is supported.
506 * If not we'll force DM to use PAGE_SIZE or
507 * smaller I/O, just to be safe.
509 if (dm_queue_merge_is_compulsory(q
) && !ti
->type
->merge
)
510 blk_limits_max_hw_sectors(limits
,
511 (unsigned int) (PAGE_SIZE
>> 9));
514 EXPORT_SYMBOL_GPL(dm_set_device_limits
);
517 * Decrement a device's use count and remove it if necessary.
519 void dm_put_device(struct dm_target
*ti
, struct dm_dev
*d
)
521 struct dm_dev_internal
*dd
= container_of(d
, struct dm_dev_internal
,
524 if (atomic_dec_and_test(&dd
->count
)) {
525 close_dev(dd
, ti
->table
->md
);
530 EXPORT_SYMBOL(dm_put_device
);
533 * Checks to see if the target joins onto the end of the table.
535 static int adjoin(struct dm_table
*table
, struct dm_target
*ti
)
537 struct dm_target
*prev
;
539 if (!table
->num_targets
)
542 prev
= &table
->targets
[table
->num_targets
- 1];
543 return (ti
->begin
== (prev
->begin
+ prev
->len
));
547 * Used to dynamically allocate the arg array.
549 static char **realloc_argv(unsigned *array_size
, char **old_argv
)
554 new_size
= *array_size
? *array_size
* 2 : 64;
555 argv
= kmalloc(new_size
* sizeof(*argv
), GFP_KERNEL
);
557 memcpy(argv
, old_argv
, *array_size
* sizeof(*argv
));
558 *array_size
= new_size
;
566 * Destructively splits up the argument list to pass to ctr.
568 int dm_split_args(int *argc
, char ***argvp
, char *input
)
570 char *start
, *end
= input
, *out
, **argv
= NULL
;
571 unsigned array_size
= 0;
580 argv
= realloc_argv(&array_size
, argv
);
585 /* Skip whitespace */
586 start
= skip_spaces(end
);
589 break; /* success, we hit the end */
591 /* 'out' is used to remove any back-quotes */
594 /* Everything apart from '\0' can be quoted */
595 if (*end
== '\\' && *(end
+ 1)) {
602 break; /* end of token */
607 /* have we already filled the array ? */
608 if ((*argc
+ 1) > array_size
) {
609 argv
= realloc_argv(&array_size
, argv
);
614 /* we know this is whitespace */
618 /* terminate the string and put it in the array */
629 * Impose necessary and sufficient conditions on a devices's table such
630 * that any incoming bio which respects its logical_block_size can be
631 * processed successfully. If it falls across the boundary between
632 * two or more targets, the size of each piece it gets split into must
633 * be compatible with the logical_block_size of the target processing it.
635 static int validate_hardware_logical_block_alignment(struct dm_table
*table
,
636 struct queue_limits
*limits
)
639 * This function uses arithmetic modulo the logical_block_size
640 * (in units of 512-byte sectors).
642 unsigned short device_logical_block_size_sects
=
643 limits
->logical_block_size
>> SECTOR_SHIFT
;
646 * Offset of the start of the next table entry, mod logical_block_size.
648 unsigned short next_target_start
= 0;
651 * Given an aligned bio that extends beyond the end of a
652 * target, how many sectors must the next target handle?
654 unsigned short remaining
= 0;
656 struct dm_target
*uninitialized_var(ti
);
657 struct queue_limits ti_limits
;
661 * Check each entry in the table in turn.
663 while (i
< dm_table_get_num_targets(table
)) {
664 ti
= dm_table_get_target(table
, i
++);
666 blk_set_stacking_limits(&ti_limits
);
668 /* combine all target devices' limits */
669 if (ti
->type
->iterate_devices
)
670 ti
->type
->iterate_devices(ti
, dm_set_device_limits
,
674 * If the remaining sectors fall entirely within this
675 * table entry are they compatible with its logical_block_size?
677 if (remaining
< ti
->len
&&
678 remaining
& ((ti_limits
.logical_block_size
>>
683 (unsigned short) ((next_target_start
+ ti
->len
) &
684 (device_logical_block_size_sects
- 1));
685 remaining
= next_target_start
?
686 device_logical_block_size_sects
- next_target_start
: 0;
690 DMWARN("%s: table line %u (start sect %llu len %llu) "
691 "not aligned to h/w logical block size %u",
692 dm_device_name(table
->md
), i
,
693 (unsigned long long) ti
->begin
,
694 (unsigned long long) ti
->len
,
695 limits
->logical_block_size
);
702 int dm_table_add_target(struct dm_table
*t
, const char *type
,
703 sector_t start
, sector_t len
, char *params
)
705 int r
= -EINVAL
, argc
;
707 struct dm_target
*tgt
;
710 DMERR("%s: target type %s must appear alone in table",
711 dm_device_name(t
->md
), t
->targets
->type
->name
);
715 if ((r
= check_space(t
)))
718 tgt
= t
->targets
+ t
->num_targets
;
719 memset(tgt
, 0, sizeof(*tgt
));
722 DMERR("%s: zero-length target", dm_device_name(t
->md
));
726 tgt
->type
= dm_get_target_type(type
);
728 DMERR("%s: %s: unknown target type", dm_device_name(t
->md
),
733 if (dm_target_needs_singleton(tgt
->type
)) {
734 if (t
->num_targets
) {
735 DMERR("%s: target type %s must appear alone in table",
736 dm_device_name(t
->md
), type
);
742 if (dm_target_always_writeable(tgt
->type
) && !(t
->mode
& FMODE_WRITE
)) {
743 DMERR("%s: target type %s may not be included in read-only tables",
744 dm_device_name(t
->md
), type
);
748 if (t
->immutable_target_type
) {
749 if (t
->immutable_target_type
!= tgt
->type
) {
750 DMERR("%s: immutable target type %s cannot be mixed with other target types",
751 dm_device_name(t
->md
), t
->immutable_target_type
->name
);
754 } else if (dm_target_is_immutable(tgt
->type
)) {
755 if (t
->num_targets
) {
756 DMERR("%s: immutable target type %s cannot be mixed with other target types",
757 dm_device_name(t
->md
), tgt
->type
->name
);
760 t
->immutable_target_type
= tgt
->type
;
766 tgt
->error
= "Unknown error";
769 * Does this target adjoin the previous one ?
771 if (!adjoin(t
, tgt
)) {
772 tgt
->error
= "Gap in table";
777 r
= dm_split_args(&argc
, &argv
, params
);
779 tgt
->error
= "couldn't split parameters (insufficient memory)";
783 r
= tgt
->type
->ctr(tgt
, argc
, argv
);
788 t
->highs
[t
->num_targets
++] = tgt
->begin
+ tgt
->len
- 1;
790 if (!tgt
->num_discard_bios
&& tgt
->discards_supported
)
791 DMWARN("%s: %s: ignoring discards_supported because num_discard_bios is zero.",
792 dm_device_name(t
->md
), type
);
797 DMERR("%s: %s: %s", dm_device_name(t
->md
), type
, tgt
->error
);
798 dm_put_target_type(tgt
->type
);
803 * Target argument parsing helpers.
805 static int validate_next_arg(struct dm_arg
*arg
, struct dm_arg_set
*arg_set
,
806 unsigned *value
, char **error
, unsigned grouped
)
808 const char *arg_str
= dm_shift_arg(arg_set
);
812 (sscanf(arg_str
, "%u%c", value
, &dummy
) != 1) ||
813 (*value
< arg
->min
) ||
814 (*value
> arg
->max
) ||
815 (grouped
&& arg_set
->argc
< *value
)) {
823 int dm_read_arg(struct dm_arg
*arg
, struct dm_arg_set
*arg_set
,
824 unsigned *value
, char **error
)
826 return validate_next_arg(arg
, arg_set
, value
, error
, 0);
828 EXPORT_SYMBOL(dm_read_arg
);
830 int dm_read_arg_group(struct dm_arg
*arg
, struct dm_arg_set
*arg_set
,
831 unsigned *value
, char **error
)
833 return validate_next_arg(arg
, arg_set
, value
, error
, 1);
835 EXPORT_SYMBOL(dm_read_arg_group
);
837 const char *dm_shift_arg(struct dm_arg_set
*as
)
850 EXPORT_SYMBOL(dm_shift_arg
);
852 void dm_consume_args(struct dm_arg_set
*as
, unsigned num_args
)
854 BUG_ON(as
->argc
< num_args
);
855 as
->argc
-= num_args
;
856 as
->argv
+= num_args
;
858 EXPORT_SYMBOL(dm_consume_args
);
860 static int dm_table_set_type(struct dm_table
*t
)
863 unsigned bio_based
= 0, request_based
= 0;
864 struct dm_target
*tgt
;
865 struct dm_dev_internal
*dd
;
866 struct list_head
*devices
;
868 for (i
= 0; i
< t
->num_targets
; i
++) {
869 tgt
= t
->targets
+ i
;
870 if (dm_target_request_based(tgt
))
875 if (bio_based
&& request_based
) {
876 DMWARN("Inconsistent table: different target types"
877 " can't be mixed up");
883 /* We must use this table as bio-based */
884 t
->type
= DM_TYPE_BIO_BASED
;
888 BUG_ON(!request_based
); /* No targets in this table */
890 /* Non-request-stackable devices can't be used for request-based dm */
891 devices
= dm_table_get_devices(t
);
892 list_for_each_entry(dd
, devices
, list
) {
893 if (!blk_queue_stackable(bdev_get_queue(dd
->dm_dev
.bdev
))) {
894 DMWARN("table load rejected: including"
895 " non-request-stackable devices");
901 * Request-based dm supports only tables that have a single target now.
902 * To support multiple targets, request splitting support is needed,
903 * and that needs lots of changes in the block-layer.
904 * (e.g. request completion process for partial completion.)
906 if (t
->num_targets
> 1) {
907 DMWARN("Request-based dm doesn't support multiple targets yet");
911 t
->type
= DM_TYPE_REQUEST_BASED
;
916 unsigned dm_table_get_type(struct dm_table
*t
)
921 struct target_type
*dm_table_get_immutable_target_type(struct dm_table
*t
)
923 return t
->immutable_target_type
;
926 bool dm_table_request_based(struct dm_table
*t
)
928 return dm_table_get_type(t
) == DM_TYPE_REQUEST_BASED
;
931 int dm_table_alloc_md_mempools(struct dm_table
*t
)
933 unsigned type
= dm_table_get_type(t
);
934 unsigned per_bio_data_size
= 0;
935 struct dm_target
*tgt
;
938 if (unlikely(type
== DM_TYPE_NONE
)) {
939 DMWARN("no table type is set, can't allocate mempools");
943 if (type
== DM_TYPE_BIO_BASED
)
944 for (i
= 0; i
< t
->num_targets
; i
++) {
945 tgt
= t
->targets
+ i
;
946 per_bio_data_size
= max(per_bio_data_size
, tgt
->per_bio_data_size
);
949 t
->mempools
= dm_alloc_md_mempools(type
, t
->integrity_supported
, per_bio_data_size
);
956 void dm_table_free_md_mempools(struct dm_table
*t
)
958 dm_free_md_mempools(t
->mempools
);
962 struct dm_md_mempools
*dm_table_get_md_mempools(struct dm_table
*t
)
967 static int setup_indexes(struct dm_table
*t
)
970 unsigned int total
= 0;
973 /* allocate the space for *all* the indexes */
974 for (i
= t
->depth
- 2; i
>= 0; i
--) {
975 t
->counts
[i
] = dm_div_up(t
->counts
[i
+ 1], CHILDREN_PER_NODE
);
976 total
+= t
->counts
[i
];
979 indexes
= (sector_t
*) dm_vcalloc(total
, (unsigned long) NODE_SIZE
);
983 /* set up internal nodes, bottom-up */
984 for (i
= t
->depth
- 2; i
>= 0; i
--) {
985 t
->index
[i
] = indexes
;
986 indexes
+= (KEYS_PER_NODE
* t
->counts
[i
]);
987 setup_btree_index(i
, t
);
994 * Builds the btree to index the map.
996 static int dm_table_build_index(struct dm_table
*t
)
999 unsigned int leaf_nodes
;
1001 /* how many indexes will the btree have ? */
1002 leaf_nodes
= dm_div_up(t
->num_targets
, KEYS_PER_NODE
);
1003 t
->depth
= 1 + int_log(leaf_nodes
, CHILDREN_PER_NODE
);
1005 /* leaf layer has already been set up */
1006 t
->counts
[t
->depth
- 1] = leaf_nodes
;
1007 t
->index
[t
->depth
- 1] = t
->highs
;
1010 r
= setup_indexes(t
);
1016 * Get a disk whose integrity profile reflects the table's profile.
1017 * If %match_all is true, all devices' profiles must match.
1018 * If %match_all is false, all devices must at least have an
1019 * allocated integrity profile; but uninitialized is ok.
1020 * Returns NULL if integrity support was inconsistent or unavailable.
1022 static struct gendisk
* dm_table_get_integrity_disk(struct dm_table
*t
,
1025 struct list_head
*devices
= dm_table_get_devices(t
);
1026 struct dm_dev_internal
*dd
= NULL
;
1027 struct gendisk
*prev_disk
= NULL
, *template_disk
= NULL
;
1029 list_for_each_entry(dd
, devices
, list
) {
1030 template_disk
= dd
->dm_dev
.bdev
->bd_disk
;
1031 if (!blk_get_integrity(template_disk
))
1033 if (!match_all
&& !blk_integrity_is_initialized(template_disk
))
1034 continue; /* skip uninitialized profiles */
1035 else if (prev_disk
&&
1036 blk_integrity_compare(prev_disk
, template_disk
) < 0)
1038 prev_disk
= template_disk
;
1041 return template_disk
;
1045 DMWARN("%s: integrity not set: %s and %s profile mismatch",
1046 dm_device_name(t
->md
),
1047 prev_disk
->disk_name
,
1048 template_disk
->disk_name
);
1053 * Register the mapped device for blk_integrity support if
1054 * the underlying devices have an integrity profile. But all devices
1055 * may not have matching profiles (checking all devices isn't reliable
1056 * during table load because this table may use other DM device(s) which
1057 * must be resumed before they will have an initialized integity profile).
1058 * Stacked DM devices force a 2 stage integrity profile validation:
1059 * 1 - during load, validate all initialized integrity profiles match
1060 * 2 - during resume, validate all integrity profiles match
1062 static int dm_table_prealloc_integrity(struct dm_table
*t
, struct mapped_device
*md
)
1064 struct gendisk
*template_disk
= NULL
;
1066 template_disk
= dm_table_get_integrity_disk(t
, false);
1070 if (!blk_integrity_is_initialized(dm_disk(md
))) {
1071 t
->integrity_supported
= 1;
1072 return blk_integrity_register(dm_disk(md
), NULL
);
1076 * If DM device already has an initalized integrity
1077 * profile the new profile should not conflict.
1079 if (blk_integrity_is_initialized(template_disk
) &&
1080 blk_integrity_compare(dm_disk(md
), template_disk
) < 0) {
1081 DMWARN("%s: conflict with existing integrity profile: "
1082 "%s profile mismatch",
1083 dm_device_name(t
->md
),
1084 template_disk
->disk_name
);
1088 /* Preserve existing initialized integrity profile */
1089 t
->integrity_supported
= 1;
1094 * Prepares the table for use by building the indices,
1095 * setting the type, and allocating mempools.
1097 int dm_table_complete(struct dm_table
*t
)
1101 r
= dm_table_set_type(t
);
1103 DMERR("unable to set table type");
1107 r
= dm_table_build_index(t
);
1109 DMERR("unable to build btrees");
1113 r
= dm_table_prealloc_integrity(t
, t
->md
);
1115 DMERR("could not register integrity profile.");
1119 r
= dm_table_alloc_md_mempools(t
);
1121 DMERR("unable to allocate mempools");
1126 static DEFINE_MUTEX(_event_lock
);
1127 void dm_table_event_callback(struct dm_table
*t
,
1128 void (*fn
)(void *), void *context
)
1130 mutex_lock(&_event_lock
);
1132 t
->event_context
= context
;
1133 mutex_unlock(&_event_lock
);
1136 void dm_table_event(struct dm_table
*t
)
1139 * You can no longer call dm_table_event() from interrupt
1140 * context, use a bottom half instead.
1142 BUG_ON(in_interrupt());
1144 mutex_lock(&_event_lock
);
1146 t
->event_fn(t
->event_context
);
1147 mutex_unlock(&_event_lock
);
1149 EXPORT_SYMBOL(dm_table_event
);
1151 sector_t
dm_table_get_size(struct dm_table
*t
)
1153 return t
->num_targets
? (t
->highs
[t
->num_targets
- 1] + 1) : 0;
1155 EXPORT_SYMBOL(dm_table_get_size
);
1157 struct dm_target
*dm_table_get_target(struct dm_table
*t
, unsigned int index
)
1159 if (index
>= t
->num_targets
)
1162 return t
->targets
+ index
;
1166 * Search the btree for the correct target.
1168 * Caller should check returned pointer with dm_target_is_valid()
1169 * to trap I/O beyond end of device.
1171 struct dm_target
*dm_table_find_target(struct dm_table
*t
, sector_t sector
)
1173 unsigned int l
, n
= 0, k
= 0;
1176 for (l
= 0; l
< t
->depth
; l
++) {
1177 n
= get_child(n
, k
);
1178 node
= get_node(t
, l
, n
);
1180 for (k
= 0; k
< KEYS_PER_NODE
; k
++)
1181 if (node
[k
] >= sector
)
1185 return &t
->targets
[(KEYS_PER_NODE
* n
) + k
];
1188 static int count_device(struct dm_target
*ti
, struct dm_dev
*dev
,
1189 sector_t start
, sector_t len
, void *data
)
1191 unsigned *num_devices
= data
;
1199 * Check whether a table has no data devices attached using each
1200 * target's iterate_devices method.
1201 * Returns false if the result is unknown because a target doesn't
1202 * support iterate_devices.
1204 bool dm_table_has_no_data_devices(struct dm_table
*table
)
1206 struct dm_target
*uninitialized_var(ti
);
1207 unsigned i
= 0, num_devices
= 0;
1209 while (i
< dm_table_get_num_targets(table
)) {
1210 ti
= dm_table_get_target(table
, i
++);
1212 if (!ti
->type
->iterate_devices
)
1215 ti
->type
->iterate_devices(ti
, count_device
, &num_devices
);
1224 * Establish the new table's queue_limits and validate them.
1226 int dm_calculate_queue_limits(struct dm_table
*table
,
1227 struct queue_limits
*limits
)
1229 struct dm_target
*uninitialized_var(ti
);
1230 struct queue_limits ti_limits
;
1233 blk_set_stacking_limits(limits
);
1235 while (i
< dm_table_get_num_targets(table
)) {
1236 blk_set_stacking_limits(&ti_limits
);
1238 ti
= dm_table_get_target(table
, i
++);
1240 if (!ti
->type
->iterate_devices
)
1241 goto combine_limits
;
1244 * Combine queue limits of all the devices this target uses.
1246 ti
->type
->iterate_devices(ti
, dm_set_device_limits
,
1249 /* Set I/O hints portion of queue limits */
1250 if (ti
->type
->io_hints
)
1251 ti
->type
->io_hints(ti
, &ti_limits
);
1254 * Check each device area is consistent with the target's
1255 * overall queue limits.
1257 if (ti
->type
->iterate_devices(ti
, device_area_is_invalid
,
1263 * Merge this target's queue limits into the overall limits
1266 if (blk_stack_limits(limits
, &ti_limits
, 0) < 0)
1267 DMWARN("%s: adding target device "
1268 "(start sect %llu len %llu) "
1269 "caused an alignment inconsistency",
1270 dm_device_name(table
->md
),
1271 (unsigned long long) ti
->begin
,
1272 (unsigned long long) ti
->len
);
1275 return validate_hardware_logical_block_alignment(table
, limits
);
1279 * Set the integrity profile for this device if all devices used have
1280 * matching profiles. We're quite deep in the resume path but still
1281 * don't know if all devices (particularly DM devices this device
1282 * may be stacked on) have matching profiles. Even if the profiles
1283 * don't match we have no way to fail (to resume) at this point.
1285 static void dm_table_set_integrity(struct dm_table
*t
)
1287 struct gendisk
*template_disk
= NULL
;
1289 if (!blk_get_integrity(dm_disk(t
->md
)))
1292 template_disk
= dm_table_get_integrity_disk(t
, true);
1294 blk_integrity_register(dm_disk(t
->md
),
1295 blk_get_integrity(template_disk
));
1296 else if (blk_integrity_is_initialized(dm_disk(t
->md
)))
1297 DMWARN("%s: device no longer has a valid integrity profile",
1298 dm_device_name(t
->md
));
1300 DMWARN("%s: unable to establish an integrity profile",
1301 dm_device_name(t
->md
));
1304 static int device_flush_capable(struct dm_target
*ti
, struct dm_dev
*dev
,
1305 sector_t start
, sector_t len
, void *data
)
1307 unsigned flush
= (*(unsigned *)data
);
1308 struct request_queue
*q
= bdev_get_queue(dev
->bdev
);
1310 return q
&& (q
->flush_flags
& flush
);
1313 static bool dm_table_supports_flush(struct dm_table
*t
, unsigned flush
)
1315 struct dm_target
*ti
;
1319 * Require at least one underlying device to support flushes.
1320 * t->devices includes internal dm devices such as mirror logs
1321 * so we need to use iterate_devices here, which targets
1322 * supporting flushes must provide.
1324 while (i
< dm_table_get_num_targets(t
)) {
1325 ti
= dm_table_get_target(t
, i
++);
1327 if (!ti
->num_flush_bios
)
1330 if (ti
->flush_supported
)
1333 if (ti
->type
->iterate_devices
&&
1334 ti
->type
->iterate_devices(ti
, device_flush_capable
, &flush
))
1341 static bool dm_table_discard_zeroes_data(struct dm_table
*t
)
1343 struct dm_target
*ti
;
1346 /* Ensure that all targets supports discard_zeroes_data. */
1347 while (i
< dm_table_get_num_targets(t
)) {
1348 ti
= dm_table_get_target(t
, i
++);
1350 if (ti
->discard_zeroes_data_unsupported
)
1357 static int device_is_nonrot(struct dm_target
*ti
, struct dm_dev
*dev
,
1358 sector_t start
, sector_t len
, void *data
)
1360 struct request_queue
*q
= bdev_get_queue(dev
->bdev
);
1362 return q
&& blk_queue_nonrot(q
);
1365 static int device_is_not_random(struct dm_target
*ti
, struct dm_dev
*dev
,
1366 sector_t start
, sector_t len
, void *data
)
1368 struct request_queue
*q
= bdev_get_queue(dev
->bdev
);
1370 return q
&& !blk_queue_add_random(q
);
1373 static bool dm_table_all_devices_attribute(struct dm_table
*t
,
1374 iterate_devices_callout_fn func
)
1376 struct dm_target
*ti
;
1379 while (i
< dm_table_get_num_targets(t
)) {
1380 ti
= dm_table_get_target(t
, i
++);
1382 if (!ti
->type
->iterate_devices
||
1383 !ti
->type
->iterate_devices(ti
, func
, NULL
))
1390 static int device_not_write_same_capable(struct dm_target
*ti
, struct dm_dev
*dev
,
1391 sector_t start
, sector_t len
, void *data
)
1393 struct request_queue
*q
= bdev_get_queue(dev
->bdev
);
1395 return q
&& !q
->limits
.max_write_same_sectors
;
1398 static bool dm_table_supports_write_same(struct dm_table
*t
)
1400 struct dm_target
*ti
;
1403 while (i
< dm_table_get_num_targets(t
)) {
1404 ti
= dm_table_get_target(t
, i
++);
1406 if (!ti
->num_write_same_bios
)
1409 if (!ti
->type
->iterate_devices
||
1410 ti
->type
->iterate_devices(ti
, device_not_write_same_capable
, NULL
))
1417 void dm_table_set_restrictions(struct dm_table
*t
, struct request_queue
*q
,
1418 struct queue_limits
*limits
)
1423 * Copy table's limits to the DM device's request_queue
1425 q
->limits
= *limits
;
1427 if (!dm_table_supports_discards(t
))
1428 queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD
, q
);
1430 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD
, q
);
1432 if (dm_table_supports_flush(t
, REQ_FLUSH
)) {
1434 if (dm_table_supports_flush(t
, REQ_FUA
))
1437 blk_queue_flush(q
, flush
);
1439 if (!dm_table_discard_zeroes_data(t
))
1440 q
->limits
.discard_zeroes_data
= 0;
1442 /* Ensure that all underlying devices are non-rotational. */
1443 if (dm_table_all_devices_attribute(t
, device_is_nonrot
))
1444 queue_flag_set_unlocked(QUEUE_FLAG_NONROT
, q
);
1446 queue_flag_clear_unlocked(QUEUE_FLAG_NONROT
, q
);
1448 if (!dm_table_supports_write_same(t
))
1449 q
->limits
.max_write_same_sectors
= 0;
1451 dm_table_set_integrity(t
);
1454 * Determine whether or not this queue's I/O timings contribute
1455 * to the entropy pool, Only request-based targets use this.
1456 * Clear QUEUE_FLAG_ADD_RANDOM if any underlying device does not
1459 if (blk_queue_add_random(q
) && dm_table_all_devices_attribute(t
, device_is_not_random
))
1460 queue_flag_clear_unlocked(QUEUE_FLAG_ADD_RANDOM
, q
);
1463 * QUEUE_FLAG_STACKABLE must be set after all queue settings are
1464 * visible to other CPUs because, once the flag is set, incoming bios
1465 * are processed by request-based dm, which refers to the queue
1467 * Until the flag set, bios are passed to bio-based dm and queued to
1468 * md->deferred where queue settings are not needed yet.
1469 * Those bios are passed to request-based dm at the resume time.
1472 if (dm_table_request_based(t
))
1473 queue_flag_set_unlocked(QUEUE_FLAG_STACKABLE
, q
);
1476 unsigned int dm_table_get_num_targets(struct dm_table
*t
)
1478 return t
->num_targets
;
1481 struct list_head
*dm_table_get_devices(struct dm_table
*t
)
1486 fmode_t
dm_table_get_mode(struct dm_table
*t
)
1490 EXPORT_SYMBOL(dm_table_get_mode
);
1492 static void suspend_targets(struct dm_table
*t
, unsigned postsuspend
)
1494 int i
= t
->num_targets
;
1495 struct dm_target
*ti
= t
->targets
;
1499 if (ti
->type
->postsuspend
)
1500 ti
->type
->postsuspend(ti
);
1501 } else if (ti
->type
->presuspend
)
1502 ti
->type
->presuspend(ti
);
1508 void dm_table_presuspend_targets(struct dm_table
*t
)
1513 suspend_targets(t
, 0);
1516 void dm_table_postsuspend_targets(struct dm_table
*t
)
1521 suspend_targets(t
, 1);
1524 int dm_table_resume_targets(struct dm_table
*t
)
1528 for (i
= 0; i
< t
->num_targets
; i
++) {
1529 struct dm_target
*ti
= t
->targets
+ i
;
1531 if (!ti
->type
->preresume
)
1534 r
= ti
->type
->preresume(ti
);
1539 for (i
= 0; i
< t
->num_targets
; i
++) {
1540 struct dm_target
*ti
= t
->targets
+ i
;
1542 if (ti
->type
->resume
)
1543 ti
->type
->resume(ti
);
1549 void dm_table_add_target_callbacks(struct dm_table
*t
, struct dm_target_callbacks
*cb
)
1551 list_add(&cb
->list
, &t
->target_callbacks
);
1553 EXPORT_SYMBOL_GPL(dm_table_add_target_callbacks
);
1555 int dm_table_any_congested(struct dm_table
*t
, int bdi_bits
)
1557 struct dm_dev_internal
*dd
;
1558 struct list_head
*devices
= dm_table_get_devices(t
);
1559 struct dm_target_callbacks
*cb
;
1562 list_for_each_entry(dd
, devices
, list
) {
1563 struct request_queue
*q
= bdev_get_queue(dd
->dm_dev
.bdev
);
1564 char b
[BDEVNAME_SIZE
];
1567 r
|= bdi_congested(&q
->backing_dev_info
, bdi_bits
);
1569 DMWARN_LIMIT("%s: any_congested: nonexistent device %s",
1570 dm_device_name(t
->md
),
1571 bdevname(dd
->dm_dev
.bdev
, b
));
1574 list_for_each_entry(cb
, &t
->target_callbacks
, list
)
1575 if (cb
->congested_fn
)
1576 r
|= cb
->congested_fn(cb
, bdi_bits
);
1581 int dm_table_any_busy_target(struct dm_table
*t
)
1584 struct dm_target
*ti
;
1586 for (i
= 0; i
< t
->num_targets
; i
++) {
1587 ti
= t
->targets
+ i
;
1588 if (ti
->type
->busy
&& ti
->type
->busy(ti
))
1595 struct mapped_device
*dm_table_get_md(struct dm_table
*t
)
1599 EXPORT_SYMBOL(dm_table_get_md
);
1601 static int device_discard_capable(struct dm_target
*ti
, struct dm_dev
*dev
,
1602 sector_t start
, sector_t len
, void *data
)
1604 struct request_queue
*q
= bdev_get_queue(dev
->bdev
);
1606 return q
&& blk_queue_discard(q
);
1609 bool dm_table_supports_discards(struct dm_table
*t
)
1611 struct dm_target
*ti
;
1615 * Unless any target used by the table set discards_supported,
1616 * require at least one underlying device to support discards.
1617 * t->devices includes internal dm devices such as mirror logs
1618 * so we need to use iterate_devices here, which targets
1619 * supporting discard selectively must provide.
1621 while (i
< dm_table_get_num_targets(t
)) {
1622 ti
= dm_table_get_target(t
, i
++);
1624 if (!ti
->num_discard_bios
)
1627 if (ti
->discards_supported
)
1630 if (ti
->type
->iterate_devices
&&
1631 ti
->type
->iterate_devices(ti
, device_discard_capable
, NULL
))