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 * We do first allocation with GFP_NOIO because dm-mpath and dm-thin must
550 * process messages even if some device is suspended. These messages have a
551 * small fixed number of arguments.
553 * On the other hand, dm-switch needs to process bulk data using messages and
554 * excessive use of GFP_NOIO could cause trouble.
556 static char **realloc_argv(unsigned *array_size
, char **old_argv
)
563 new_size
= *array_size
* 2;
569 argv
= kmalloc(new_size
* sizeof(*argv
), gfp
);
571 memcpy(argv
, old_argv
, *array_size
* sizeof(*argv
));
572 *array_size
= new_size
;
580 * Destructively splits up the argument list to pass to ctr.
582 int dm_split_args(int *argc
, char ***argvp
, char *input
)
584 char *start
, *end
= input
, *out
, **argv
= NULL
;
585 unsigned array_size
= 0;
594 argv
= realloc_argv(&array_size
, argv
);
599 /* Skip whitespace */
600 start
= skip_spaces(end
);
603 break; /* success, we hit the end */
605 /* 'out' is used to remove any back-quotes */
608 /* Everything apart from '\0' can be quoted */
609 if (*end
== '\\' && *(end
+ 1)) {
616 break; /* end of token */
621 /* have we already filled the array ? */
622 if ((*argc
+ 1) > array_size
) {
623 argv
= realloc_argv(&array_size
, argv
);
628 /* we know this is whitespace */
632 /* terminate the string and put it in the array */
643 * Impose necessary and sufficient conditions on a devices's table such
644 * that any incoming bio which respects its logical_block_size can be
645 * processed successfully. If it falls across the boundary between
646 * two or more targets, the size of each piece it gets split into must
647 * be compatible with the logical_block_size of the target processing it.
649 static int validate_hardware_logical_block_alignment(struct dm_table
*table
,
650 struct queue_limits
*limits
)
653 * This function uses arithmetic modulo the logical_block_size
654 * (in units of 512-byte sectors).
656 unsigned short device_logical_block_size_sects
=
657 limits
->logical_block_size
>> SECTOR_SHIFT
;
660 * Offset of the start of the next table entry, mod logical_block_size.
662 unsigned short next_target_start
= 0;
665 * Given an aligned bio that extends beyond the end of a
666 * target, how many sectors must the next target handle?
668 unsigned short remaining
= 0;
670 struct dm_target
*uninitialized_var(ti
);
671 struct queue_limits ti_limits
;
675 * Check each entry in the table in turn.
677 while (i
< dm_table_get_num_targets(table
)) {
678 ti
= dm_table_get_target(table
, i
++);
680 blk_set_stacking_limits(&ti_limits
);
682 /* combine all target devices' limits */
683 if (ti
->type
->iterate_devices
)
684 ti
->type
->iterate_devices(ti
, dm_set_device_limits
,
688 * If the remaining sectors fall entirely within this
689 * table entry are they compatible with its logical_block_size?
691 if (remaining
< ti
->len
&&
692 remaining
& ((ti_limits
.logical_block_size
>>
697 (unsigned short) ((next_target_start
+ ti
->len
) &
698 (device_logical_block_size_sects
- 1));
699 remaining
= next_target_start
?
700 device_logical_block_size_sects
- next_target_start
: 0;
704 DMWARN("%s: table line %u (start sect %llu len %llu) "
705 "not aligned to h/w logical block size %u",
706 dm_device_name(table
->md
), i
,
707 (unsigned long long) ti
->begin
,
708 (unsigned long long) ti
->len
,
709 limits
->logical_block_size
);
716 int dm_table_add_target(struct dm_table
*t
, const char *type
,
717 sector_t start
, sector_t len
, char *params
)
719 int r
= -EINVAL
, argc
;
721 struct dm_target
*tgt
;
724 DMERR("%s: target type %s must appear alone in table",
725 dm_device_name(t
->md
), t
->targets
->type
->name
);
729 if ((r
= check_space(t
)))
732 tgt
= t
->targets
+ t
->num_targets
;
733 memset(tgt
, 0, sizeof(*tgt
));
736 DMERR("%s: zero-length target", dm_device_name(t
->md
));
740 tgt
->type
= dm_get_target_type(type
);
742 DMERR("%s: %s: unknown target type", dm_device_name(t
->md
),
747 if (dm_target_needs_singleton(tgt
->type
)) {
748 if (t
->num_targets
) {
749 DMERR("%s: target type %s must appear alone in table",
750 dm_device_name(t
->md
), type
);
756 if (dm_target_always_writeable(tgt
->type
) && !(t
->mode
& FMODE_WRITE
)) {
757 DMERR("%s: target type %s may not be included in read-only tables",
758 dm_device_name(t
->md
), type
);
762 if (t
->immutable_target_type
) {
763 if (t
->immutable_target_type
!= tgt
->type
) {
764 DMERR("%s: immutable target type %s cannot be mixed with other target types",
765 dm_device_name(t
->md
), t
->immutable_target_type
->name
);
768 } else if (dm_target_is_immutable(tgt
->type
)) {
769 if (t
->num_targets
) {
770 DMERR("%s: immutable target type %s cannot be mixed with other target types",
771 dm_device_name(t
->md
), tgt
->type
->name
);
774 t
->immutable_target_type
= tgt
->type
;
780 tgt
->error
= "Unknown error";
783 * Does this target adjoin the previous one ?
785 if (!adjoin(t
, tgt
)) {
786 tgt
->error
= "Gap in table";
791 r
= dm_split_args(&argc
, &argv
, params
);
793 tgt
->error
= "couldn't split parameters (insufficient memory)";
797 r
= tgt
->type
->ctr(tgt
, argc
, argv
);
802 t
->highs
[t
->num_targets
++] = tgt
->begin
+ tgt
->len
- 1;
804 if (!tgt
->num_discard_bios
&& tgt
->discards_supported
)
805 DMWARN("%s: %s: ignoring discards_supported because num_discard_bios is zero.",
806 dm_device_name(t
->md
), type
);
811 DMERR("%s: %s: %s", dm_device_name(t
->md
), type
, tgt
->error
);
812 dm_put_target_type(tgt
->type
);
817 * Target argument parsing helpers.
819 static int validate_next_arg(struct dm_arg
*arg
, struct dm_arg_set
*arg_set
,
820 unsigned *value
, char **error
, unsigned grouped
)
822 const char *arg_str
= dm_shift_arg(arg_set
);
826 (sscanf(arg_str
, "%u%c", value
, &dummy
) != 1) ||
827 (*value
< arg
->min
) ||
828 (*value
> arg
->max
) ||
829 (grouped
&& arg_set
->argc
< *value
)) {
837 int dm_read_arg(struct dm_arg
*arg
, struct dm_arg_set
*arg_set
,
838 unsigned *value
, char **error
)
840 return validate_next_arg(arg
, arg_set
, value
, error
, 0);
842 EXPORT_SYMBOL(dm_read_arg
);
844 int dm_read_arg_group(struct dm_arg
*arg
, struct dm_arg_set
*arg_set
,
845 unsigned *value
, char **error
)
847 return validate_next_arg(arg
, arg_set
, value
, error
, 1);
849 EXPORT_SYMBOL(dm_read_arg_group
);
851 const char *dm_shift_arg(struct dm_arg_set
*as
)
864 EXPORT_SYMBOL(dm_shift_arg
);
866 void dm_consume_args(struct dm_arg_set
*as
, unsigned num_args
)
868 BUG_ON(as
->argc
< num_args
);
869 as
->argc
-= num_args
;
870 as
->argv
+= num_args
;
872 EXPORT_SYMBOL(dm_consume_args
);
874 static int dm_table_set_type(struct dm_table
*t
)
877 unsigned bio_based
= 0, request_based
= 0, hybrid
= 0;
878 struct dm_target
*tgt
;
879 struct dm_dev_internal
*dd
;
880 struct list_head
*devices
;
881 unsigned live_md_type
;
883 for (i
= 0; i
< t
->num_targets
; i
++) {
884 tgt
= t
->targets
+ i
;
885 if (dm_target_hybrid(tgt
))
887 else if (dm_target_request_based(tgt
))
892 if (bio_based
&& request_based
) {
893 DMWARN("Inconsistent table: different target types"
894 " can't be mixed up");
899 if (hybrid
&& !bio_based
&& !request_based
) {
901 * The targets can work either way.
902 * Determine the type from the live device.
903 * Default to bio-based if device is new.
905 live_md_type
= dm_get_md_type(t
->md
);
906 if (live_md_type
== DM_TYPE_REQUEST_BASED
)
913 /* We must use this table as bio-based */
914 t
->type
= DM_TYPE_BIO_BASED
;
918 BUG_ON(!request_based
); /* No targets in this table */
920 /* Non-request-stackable devices can't be used for request-based dm */
921 devices
= dm_table_get_devices(t
);
922 list_for_each_entry(dd
, devices
, list
) {
923 if (!blk_queue_stackable(bdev_get_queue(dd
->dm_dev
.bdev
))) {
924 DMWARN("table load rejected: including"
925 " non-request-stackable devices");
931 * Request-based dm supports only tables that have a single target now.
932 * To support multiple targets, request splitting support is needed,
933 * and that needs lots of changes in the block-layer.
934 * (e.g. request completion process for partial completion.)
936 if (t
->num_targets
> 1) {
937 DMWARN("Request-based dm doesn't support multiple targets yet");
941 t
->type
= DM_TYPE_REQUEST_BASED
;
946 unsigned dm_table_get_type(struct dm_table
*t
)
951 struct target_type
*dm_table_get_immutable_target_type(struct dm_table
*t
)
953 return t
->immutable_target_type
;
956 bool dm_table_request_based(struct dm_table
*t
)
958 return dm_table_get_type(t
) == DM_TYPE_REQUEST_BASED
;
961 int dm_table_alloc_md_mempools(struct dm_table
*t
)
963 unsigned type
= dm_table_get_type(t
);
964 unsigned per_bio_data_size
= 0;
965 struct dm_target
*tgt
;
968 if (unlikely(type
== DM_TYPE_NONE
)) {
969 DMWARN("no table type is set, can't allocate mempools");
973 if (type
== DM_TYPE_BIO_BASED
)
974 for (i
= 0; i
< t
->num_targets
; i
++) {
975 tgt
= t
->targets
+ i
;
976 per_bio_data_size
= max(per_bio_data_size
, tgt
->per_bio_data_size
);
979 t
->mempools
= dm_alloc_md_mempools(type
, t
->integrity_supported
, per_bio_data_size
);
986 void dm_table_free_md_mempools(struct dm_table
*t
)
988 dm_free_md_mempools(t
->mempools
);
992 struct dm_md_mempools
*dm_table_get_md_mempools(struct dm_table
*t
)
997 static int setup_indexes(struct dm_table
*t
)
1000 unsigned int total
= 0;
1003 /* allocate the space for *all* the indexes */
1004 for (i
= t
->depth
- 2; i
>= 0; i
--) {
1005 t
->counts
[i
] = dm_div_up(t
->counts
[i
+ 1], CHILDREN_PER_NODE
);
1006 total
+= t
->counts
[i
];
1009 indexes
= (sector_t
*) dm_vcalloc(total
, (unsigned long) NODE_SIZE
);
1013 /* set up internal nodes, bottom-up */
1014 for (i
= t
->depth
- 2; i
>= 0; i
--) {
1015 t
->index
[i
] = indexes
;
1016 indexes
+= (KEYS_PER_NODE
* t
->counts
[i
]);
1017 setup_btree_index(i
, t
);
1024 * Builds the btree to index the map.
1026 static int dm_table_build_index(struct dm_table
*t
)
1029 unsigned int leaf_nodes
;
1031 /* how many indexes will the btree have ? */
1032 leaf_nodes
= dm_div_up(t
->num_targets
, KEYS_PER_NODE
);
1033 t
->depth
= 1 + int_log(leaf_nodes
, CHILDREN_PER_NODE
);
1035 /* leaf layer has already been set up */
1036 t
->counts
[t
->depth
- 1] = leaf_nodes
;
1037 t
->index
[t
->depth
- 1] = t
->highs
;
1040 r
= setup_indexes(t
);
1046 * Get a disk whose integrity profile reflects the table's profile.
1047 * If %match_all is true, all devices' profiles must match.
1048 * If %match_all is false, all devices must at least have an
1049 * allocated integrity profile; but uninitialized is ok.
1050 * Returns NULL if integrity support was inconsistent or unavailable.
1052 static struct gendisk
* dm_table_get_integrity_disk(struct dm_table
*t
,
1055 struct list_head
*devices
= dm_table_get_devices(t
);
1056 struct dm_dev_internal
*dd
= NULL
;
1057 struct gendisk
*prev_disk
= NULL
, *template_disk
= NULL
;
1059 list_for_each_entry(dd
, devices
, list
) {
1060 template_disk
= dd
->dm_dev
.bdev
->bd_disk
;
1061 if (!blk_get_integrity(template_disk
))
1063 if (!match_all
&& !blk_integrity_is_initialized(template_disk
))
1064 continue; /* skip uninitialized profiles */
1065 else if (prev_disk
&&
1066 blk_integrity_compare(prev_disk
, template_disk
) < 0)
1068 prev_disk
= template_disk
;
1071 return template_disk
;
1075 DMWARN("%s: integrity not set: %s and %s profile mismatch",
1076 dm_device_name(t
->md
),
1077 prev_disk
->disk_name
,
1078 template_disk
->disk_name
);
1083 * Register the mapped device for blk_integrity support if
1084 * the underlying devices have an integrity profile. But all devices
1085 * may not have matching profiles (checking all devices isn't reliable
1086 * during table load because this table may use other DM device(s) which
1087 * must be resumed before they will have an initialized integity profile).
1088 * Stacked DM devices force a 2 stage integrity profile validation:
1089 * 1 - during load, validate all initialized integrity profiles match
1090 * 2 - during resume, validate all integrity profiles match
1092 static int dm_table_prealloc_integrity(struct dm_table
*t
, struct mapped_device
*md
)
1094 struct gendisk
*template_disk
= NULL
;
1096 template_disk
= dm_table_get_integrity_disk(t
, false);
1100 if (!blk_integrity_is_initialized(dm_disk(md
))) {
1101 t
->integrity_supported
= 1;
1102 return blk_integrity_register(dm_disk(md
), NULL
);
1106 * If DM device already has an initalized integrity
1107 * profile the new profile should not conflict.
1109 if (blk_integrity_is_initialized(template_disk
) &&
1110 blk_integrity_compare(dm_disk(md
), template_disk
) < 0) {
1111 DMWARN("%s: conflict with existing integrity profile: "
1112 "%s profile mismatch",
1113 dm_device_name(t
->md
),
1114 template_disk
->disk_name
);
1118 /* Preserve existing initialized integrity profile */
1119 t
->integrity_supported
= 1;
1124 * Prepares the table for use by building the indices,
1125 * setting the type, and allocating mempools.
1127 int dm_table_complete(struct dm_table
*t
)
1131 r
= dm_table_set_type(t
);
1133 DMERR("unable to set table type");
1137 r
= dm_table_build_index(t
);
1139 DMERR("unable to build btrees");
1143 r
= dm_table_prealloc_integrity(t
, t
->md
);
1145 DMERR("could not register integrity profile.");
1149 r
= dm_table_alloc_md_mempools(t
);
1151 DMERR("unable to allocate mempools");
1156 static DEFINE_MUTEX(_event_lock
);
1157 void dm_table_event_callback(struct dm_table
*t
,
1158 void (*fn
)(void *), void *context
)
1160 mutex_lock(&_event_lock
);
1162 t
->event_context
= context
;
1163 mutex_unlock(&_event_lock
);
1166 void dm_table_event(struct dm_table
*t
)
1169 * You can no longer call dm_table_event() from interrupt
1170 * context, use a bottom half instead.
1172 BUG_ON(in_interrupt());
1174 mutex_lock(&_event_lock
);
1176 t
->event_fn(t
->event_context
);
1177 mutex_unlock(&_event_lock
);
1179 EXPORT_SYMBOL(dm_table_event
);
1181 sector_t
dm_table_get_size(struct dm_table
*t
)
1183 return t
->num_targets
? (t
->highs
[t
->num_targets
- 1] + 1) : 0;
1185 EXPORT_SYMBOL(dm_table_get_size
);
1187 struct dm_target
*dm_table_get_target(struct dm_table
*t
, unsigned int index
)
1189 if (index
>= t
->num_targets
)
1192 return t
->targets
+ index
;
1196 * Search the btree for the correct target.
1198 * Caller should check returned pointer with dm_target_is_valid()
1199 * to trap I/O beyond end of device.
1201 struct dm_target
*dm_table_find_target(struct dm_table
*t
, sector_t sector
)
1203 unsigned int l
, n
= 0, k
= 0;
1206 for (l
= 0; l
< t
->depth
; l
++) {
1207 n
= get_child(n
, k
);
1208 node
= get_node(t
, l
, n
);
1210 for (k
= 0; k
< KEYS_PER_NODE
; k
++)
1211 if (node
[k
] >= sector
)
1215 return &t
->targets
[(KEYS_PER_NODE
* n
) + k
];
1218 static int count_device(struct dm_target
*ti
, struct dm_dev
*dev
,
1219 sector_t start
, sector_t len
, void *data
)
1221 unsigned *num_devices
= data
;
1229 * Check whether a table has no data devices attached using each
1230 * target's iterate_devices method.
1231 * Returns false if the result is unknown because a target doesn't
1232 * support iterate_devices.
1234 bool dm_table_has_no_data_devices(struct dm_table
*table
)
1236 struct dm_target
*uninitialized_var(ti
);
1237 unsigned i
= 0, num_devices
= 0;
1239 while (i
< dm_table_get_num_targets(table
)) {
1240 ti
= dm_table_get_target(table
, i
++);
1242 if (!ti
->type
->iterate_devices
)
1245 ti
->type
->iterate_devices(ti
, count_device
, &num_devices
);
1254 * Establish the new table's queue_limits and validate them.
1256 int dm_calculate_queue_limits(struct dm_table
*table
,
1257 struct queue_limits
*limits
)
1259 struct dm_target
*uninitialized_var(ti
);
1260 struct queue_limits ti_limits
;
1263 blk_set_stacking_limits(limits
);
1265 while (i
< dm_table_get_num_targets(table
)) {
1266 blk_set_stacking_limits(&ti_limits
);
1268 ti
= dm_table_get_target(table
, i
++);
1270 if (!ti
->type
->iterate_devices
)
1271 goto combine_limits
;
1274 * Combine queue limits of all the devices this target uses.
1276 ti
->type
->iterate_devices(ti
, dm_set_device_limits
,
1279 /* Set I/O hints portion of queue limits */
1280 if (ti
->type
->io_hints
)
1281 ti
->type
->io_hints(ti
, &ti_limits
);
1284 * Check each device area is consistent with the target's
1285 * overall queue limits.
1287 if (ti
->type
->iterate_devices(ti
, device_area_is_invalid
,
1293 * Merge this target's queue limits into the overall limits
1296 if (blk_stack_limits(limits
, &ti_limits
, 0) < 0)
1297 DMWARN("%s: adding target device "
1298 "(start sect %llu len %llu) "
1299 "caused an alignment inconsistency",
1300 dm_device_name(table
->md
),
1301 (unsigned long long) ti
->begin
,
1302 (unsigned long long) ti
->len
);
1305 return validate_hardware_logical_block_alignment(table
, limits
);
1309 * Set the integrity profile for this device if all devices used have
1310 * matching profiles. We're quite deep in the resume path but still
1311 * don't know if all devices (particularly DM devices this device
1312 * may be stacked on) have matching profiles. Even if the profiles
1313 * don't match we have no way to fail (to resume) at this point.
1315 static void dm_table_set_integrity(struct dm_table
*t
)
1317 struct gendisk
*template_disk
= NULL
;
1319 if (!blk_get_integrity(dm_disk(t
->md
)))
1322 template_disk
= dm_table_get_integrity_disk(t
, true);
1324 blk_integrity_register(dm_disk(t
->md
),
1325 blk_get_integrity(template_disk
));
1326 else if (blk_integrity_is_initialized(dm_disk(t
->md
)))
1327 DMWARN("%s: device no longer has a valid integrity profile",
1328 dm_device_name(t
->md
));
1330 DMWARN("%s: unable to establish an integrity profile",
1331 dm_device_name(t
->md
));
1334 static int device_flush_capable(struct dm_target
*ti
, struct dm_dev
*dev
,
1335 sector_t start
, sector_t len
, void *data
)
1337 unsigned flush
= (*(unsigned *)data
);
1338 struct request_queue
*q
= bdev_get_queue(dev
->bdev
);
1340 return q
&& (q
->flush_flags
& flush
);
1343 static bool dm_table_supports_flush(struct dm_table
*t
, unsigned flush
)
1345 struct dm_target
*ti
;
1349 * Require at least one underlying device to support flushes.
1350 * t->devices includes internal dm devices such as mirror logs
1351 * so we need to use iterate_devices here, which targets
1352 * supporting flushes must provide.
1354 while (i
< dm_table_get_num_targets(t
)) {
1355 ti
= dm_table_get_target(t
, i
++);
1357 if (!ti
->num_flush_bios
)
1360 if (ti
->flush_supported
)
1363 if (ti
->type
->iterate_devices
&&
1364 ti
->type
->iterate_devices(ti
, device_flush_capable
, &flush
))
1371 static bool dm_table_discard_zeroes_data(struct dm_table
*t
)
1373 struct dm_target
*ti
;
1376 /* Ensure that all targets supports discard_zeroes_data. */
1377 while (i
< dm_table_get_num_targets(t
)) {
1378 ti
= dm_table_get_target(t
, i
++);
1380 if (ti
->discard_zeroes_data_unsupported
)
1387 static int device_is_nonrot(struct dm_target
*ti
, struct dm_dev
*dev
,
1388 sector_t start
, sector_t len
, void *data
)
1390 struct request_queue
*q
= bdev_get_queue(dev
->bdev
);
1392 return q
&& blk_queue_nonrot(q
);
1395 static int device_is_not_random(struct dm_target
*ti
, struct dm_dev
*dev
,
1396 sector_t start
, sector_t len
, void *data
)
1398 struct request_queue
*q
= bdev_get_queue(dev
->bdev
);
1400 return q
&& !blk_queue_add_random(q
);
1403 static bool dm_table_all_devices_attribute(struct dm_table
*t
,
1404 iterate_devices_callout_fn func
)
1406 struct dm_target
*ti
;
1409 while (i
< dm_table_get_num_targets(t
)) {
1410 ti
= dm_table_get_target(t
, i
++);
1412 if (!ti
->type
->iterate_devices
||
1413 !ti
->type
->iterate_devices(ti
, func
, NULL
))
1420 static int device_not_write_same_capable(struct dm_target
*ti
, struct dm_dev
*dev
,
1421 sector_t start
, sector_t len
, void *data
)
1423 struct request_queue
*q
= bdev_get_queue(dev
->bdev
);
1425 return q
&& !q
->limits
.max_write_same_sectors
;
1428 static bool dm_table_supports_write_same(struct dm_table
*t
)
1430 struct dm_target
*ti
;
1433 while (i
< dm_table_get_num_targets(t
)) {
1434 ti
= dm_table_get_target(t
, i
++);
1436 if (!ti
->num_write_same_bios
)
1439 if (!ti
->type
->iterate_devices
||
1440 ti
->type
->iterate_devices(ti
, device_not_write_same_capable
, NULL
))
1447 void dm_table_set_restrictions(struct dm_table
*t
, struct request_queue
*q
,
1448 struct queue_limits
*limits
)
1453 * Copy table's limits to the DM device's request_queue
1455 q
->limits
= *limits
;
1457 if (!dm_table_supports_discards(t
))
1458 queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD
, q
);
1460 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD
, q
);
1462 if (dm_table_supports_flush(t
, REQ_FLUSH
)) {
1464 if (dm_table_supports_flush(t
, REQ_FUA
))
1467 blk_queue_flush(q
, flush
);
1469 if (!dm_table_discard_zeroes_data(t
))
1470 q
->limits
.discard_zeroes_data
= 0;
1472 /* Ensure that all underlying devices are non-rotational. */
1473 if (dm_table_all_devices_attribute(t
, device_is_nonrot
))
1474 queue_flag_set_unlocked(QUEUE_FLAG_NONROT
, q
);
1476 queue_flag_clear_unlocked(QUEUE_FLAG_NONROT
, q
);
1478 if (!dm_table_supports_write_same(t
))
1479 q
->limits
.max_write_same_sectors
= 0;
1481 dm_table_set_integrity(t
);
1484 * Determine whether or not this queue's I/O timings contribute
1485 * to the entropy pool, Only request-based targets use this.
1486 * Clear QUEUE_FLAG_ADD_RANDOM if any underlying device does not
1489 if (blk_queue_add_random(q
) && dm_table_all_devices_attribute(t
, device_is_not_random
))
1490 queue_flag_clear_unlocked(QUEUE_FLAG_ADD_RANDOM
, q
);
1493 * QUEUE_FLAG_STACKABLE must be set after all queue settings are
1494 * visible to other CPUs because, once the flag is set, incoming bios
1495 * are processed by request-based dm, which refers to the queue
1497 * Until the flag set, bios are passed to bio-based dm and queued to
1498 * md->deferred where queue settings are not needed yet.
1499 * Those bios are passed to request-based dm at the resume time.
1502 if (dm_table_request_based(t
))
1503 queue_flag_set_unlocked(QUEUE_FLAG_STACKABLE
, q
);
1506 unsigned int dm_table_get_num_targets(struct dm_table
*t
)
1508 return t
->num_targets
;
1511 struct list_head
*dm_table_get_devices(struct dm_table
*t
)
1516 fmode_t
dm_table_get_mode(struct dm_table
*t
)
1520 EXPORT_SYMBOL(dm_table_get_mode
);
1522 static void suspend_targets(struct dm_table
*t
, unsigned postsuspend
)
1524 int i
= t
->num_targets
;
1525 struct dm_target
*ti
= t
->targets
;
1529 if (ti
->type
->postsuspend
)
1530 ti
->type
->postsuspend(ti
);
1531 } else if (ti
->type
->presuspend
)
1532 ti
->type
->presuspend(ti
);
1538 void dm_table_presuspend_targets(struct dm_table
*t
)
1543 suspend_targets(t
, 0);
1546 void dm_table_postsuspend_targets(struct dm_table
*t
)
1551 suspend_targets(t
, 1);
1554 int dm_table_resume_targets(struct dm_table
*t
)
1558 for (i
= 0; i
< t
->num_targets
; i
++) {
1559 struct dm_target
*ti
= t
->targets
+ i
;
1561 if (!ti
->type
->preresume
)
1564 r
= ti
->type
->preresume(ti
);
1566 DMERR("%s: %s: preresume failed, error = %d",
1567 dm_device_name(t
->md
), ti
->type
->name
, r
);
1572 for (i
= 0; i
< t
->num_targets
; i
++) {
1573 struct dm_target
*ti
= t
->targets
+ i
;
1575 if (ti
->type
->resume
)
1576 ti
->type
->resume(ti
);
1582 void dm_table_add_target_callbacks(struct dm_table
*t
, struct dm_target_callbacks
*cb
)
1584 list_add(&cb
->list
, &t
->target_callbacks
);
1586 EXPORT_SYMBOL_GPL(dm_table_add_target_callbacks
);
1588 int dm_table_any_congested(struct dm_table
*t
, int bdi_bits
)
1590 struct dm_dev_internal
*dd
;
1591 struct list_head
*devices
= dm_table_get_devices(t
);
1592 struct dm_target_callbacks
*cb
;
1595 list_for_each_entry(dd
, devices
, list
) {
1596 struct request_queue
*q
= bdev_get_queue(dd
->dm_dev
.bdev
);
1597 char b
[BDEVNAME_SIZE
];
1600 r
|= bdi_congested(&q
->backing_dev_info
, bdi_bits
);
1602 DMWARN_LIMIT("%s: any_congested: nonexistent device %s",
1603 dm_device_name(t
->md
),
1604 bdevname(dd
->dm_dev
.bdev
, b
));
1607 list_for_each_entry(cb
, &t
->target_callbacks
, list
)
1608 if (cb
->congested_fn
)
1609 r
|= cb
->congested_fn(cb
, bdi_bits
);
1614 int dm_table_any_busy_target(struct dm_table
*t
)
1617 struct dm_target
*ti
;
1619 for (i
= 0; i
< t
->num_targets
; i
++) {
1620 ti
= t
->targets
+ i
;
1621 if (ti
->type
->busy
&& ti
->type
->busy(ti
))
1628 struct mapped_device
*dm_table_get_md(struct dm_table
*t
)
1632 EXPORT_SYMBOL(dm_table_get_md
);
1634 static int device_discard_capable(struct dm_target
*ti
, struct dm_dev
*dev
,
1635 sector_t start
, sector_t len
, void *data
)
1637 struct request_queue
*q
= bdev_get_queue(dev
->bdev
);
1639 return q
&& blk_queue_discard(q
);
1642 bool dm_table_supports_discards(struct dm_table
*t
)
1644 struct dm_target
*ti
;
1648 * Unless any target used by the table set discards_supported,
1649 * require at least one underlying device to support discards.
1650 * t->devices includes internal dm devices such as mirror logs
1651 * so we need to use iterate_devices here, which targets
1652 * supporting discard selectively must provide.
1654 while (i
< dm_table_get_num_targets(t
)) {
1655 ti
= dm_table_get_target(t
, i
++);
1657 if (!ti
->num_discard_bios
)
1660 if (ti
->discards_supported
)
1663 if (ti
->type
->iterate_devices
&&
1664 ti
->type
->iterate_devices(ti
, device_discard_capable
, NULL
))