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>
21 #include <linux/blk-mq.h>
22 #include <linux/mount.h>
24 #define DM_MSG_PREFIX "table"
27 #define NODE_SIZE L1_CACHE_BYTES
28 #define KEYS_PER_NODE (NODE_SIZE / sizeof(sector_t))
29 #define CHILDREN_PER_NODE (KEYS_PER_NODE + 1)
32 struct mapped_device
*md
;
37 unsigned int counts
[MAX_DEPTH
]; /* in nodes */
38 sector_t
*index
[MAX_DEPTH
];
40 unsigned int num_targets
;
41 unsigned int num_allocated
;
43 struct dm_target
*targets
;
45 struct target_type
*immutable_target_type
;
46 unsigned integrity_supported
:1;
50 * Indicates the rw permissions for the new logical
51 * device. This should be a combination of FMODE_READ
56 /* a list of devices used by this table */
57 struct list_head devices
;
59 /* events get handed up using this callback */
60 void (*event_fn
)(void *);
63 struct dm_md_mempools
*mempools
;
65 struct list_head target_callbacks
;
69 * Similar to ceiling(log_size(n))
71 static unsigned int int_log(unsigned int n
, unsigned int base
)
76 n
= dm_div_up(n
, base
);
84 * Calculate the index of the child node of the n'th node k'th key.
86 static inline unsigned int get_child(unsigned int n
, unsigned int k
)
88 return (n
* CHILDREN_PER_NODE
) + k
;
92 * Return the n'th node of level l from table t.
94 static inline sector_t
*get_node(struct dm_table
*t
,
95 unsigned int l
, unsigned int n
)
97 return t
->index
[l
] + (n
* KEYS_PER_NODE
);
101 * Return the highest key that you could lookup from the n'th
102 * node on level l of the btree.
104 static sector_t
high(struct dm_table
*t
, unsigned int l
, unsigned int n
)
106 for (; l
< t
->depth
- 1; l
++)
107 n
= get_child(n
, CHILDREN_PER_NODE
- 1);
109 if (n
>= t
->counts
[l
])
110 return (sector_t
) - 1;
112 return get_node(t
, l
, n
)[KEYS_PER_NODE
- 1];
116 * Fills in a level of the btree based on the highs of the level
119 static int setup_btree_index(unsigned int l
, struct dm_table
*t
)
124 for (n
= 0U; n
< t
->counts
[l
]; n
++) {
125 node
= get_node(t
, l
, n
);
127 for (k
= 0U; k
< KEYS_PER_NODE
; k
++)
128 node
[k
] = high(t
, l
+ 1, get_child(n
, k
));
134 void *dm_vcalloc(unsigned long nmemb
, unsigned long elem_size
)
140 * Check that we're not going to overflow.
142 if (nmemb
> (ULONG_MAX
/ elem_size
))
145 size
= nmemb
* elem_size
;
146 addr
= vzalloc(size
);
150 EXPORT_SYMBOL(dm_vcalloc
);
153 * highs, and targets are managed as dynamic arrays during a
156 static int alloc_targets(struct dm_table
*t
, unsigned int num
)
159 struct dm_target
*n_targets
;
162 * Allocate both the target array and offset array at once.
163 * Append an empty entry to catch sectors beyond the end of
166 n_highs
= (sector_t
*) dm_vcalloc(num
+ 1, sizeof(struct dm_target
) +
171 n_targets
= (struct dm_target
*) (n_highs
+ num
);
173 memset(n_highs
, -1, sizeof(*n_highs
) * num
);
176 t
->num_allocated
= num
;
178 t
->targets
= n_targets
;
183 int dm_table_create(struct dm_table
**result
, fmode_t mode
,
184 unsigned num_targets
, struct mapped_device
*md
)
186 struct dm_table
*t
= kzalloc(sizeof(*t
), GFP_KERNEL
);
191 INIT_LIST_HEAD(&t
->devices
);
192 INIT_LIST_HEAD(&t
->target_callbacks
);
195 num_targets
= KEYS_PER_NODE
;
197 num_targets
= dm_round_up(num_targets
, KEYS_PER_NODE
);
204 if (alloc_targets(t
, num_targets
)) {
215 static void free_devices(struct list_head
*devices
, struct mapped_device
*md
)
217 struct list_head
*tmp
, *next
;
219 list_for_each_safe(tmp
, next
, devices
) {
220 struct dm_dev_internal
*dd
=
221 list_entry(tmp
, struct dm_dev_internal
, list
);
222 DMWARN("%s: dm_table_destroy: dm_put_device call missing for %s",
223 dm_device_name(md
), dd
->dm_dev
->name
);
224 dm_put_table_device(md
, dd
->dm_dev
);
229 void dm_table_destroy(struct dm_table
*t
)
236 /* free the indexes */
238 vfree(t
->index
[t
->depth
- 2]);
240 /* free the targets */
241 for (i
= 0; i
< t
->num_targets
; i
++) {
242 struct dm_target
*tgt
= t
->targets
+ i
;
247 dm_put_target_type(tgt
->type
);
252 /* free the device list */
253 free_devices(&t
->devices
, t
->md
);
255 dm_free_md_mempools(t
->mempools
);
261 * See if we've already got a device in the list.
263 static struct dm_dev_internal
*find_device(struct list_head
*l
, dev_t dev
)
265 struct dm_dev_internal
*dd
;
267 list_for_each_entry (dd
, l
, list
)
268 if (dd
->dm_dev
->bdev
->bd_dev
== dev
)
275 * If possible, this checks an area of a destination device is invalid.
277 static int device_area_is_invalid(struct dm_target
*ti
, struct dm_dev
*dev
,
278 sector_t start
, sector_t len
, void *data
)
280 struct request_queue
*q
;
281 struct queue_limits
*limits
= data
;
282 struct block_device
*bdev
= dev
->bdev
;
284 i_size_read(bdev
->bd_inode
) >> SECTOR_SHIFT
;
285 unsigned short logical_block_size_sectors
=
286 limits
->logical_block_size
>> SECTOR_SHIFT
;
287 char b
[BDEVNAME_SIZE
];
290 * Some devices exist without request functions,
291 * such as loop devices not yet bound to backing files.
292 * Forbid the use of such devices.
294 q
= bdev_get_queue(bdev
);
295 if (!q
|| !q
->make_request_fn
) {
296 DMWARN("%s: %s is not yet initialised: "
297 "start=%llu, len=%llu, dev_size=%llu",
298 dm_device_name(ti
->table
->md
), bdevname(bdev
, b
),
299 (unsigned long long)start
,
300 (unsigned long long)len
,
301 (unsigned long long)dev_size
);
308 if ((start
>= dev_size
) || (start
+ len
> dev_size
)) {
309 DMWARN("%s: %s too small for target: "
310 "start=%llu, len=%llu, dev_size=%llu",
311 dm_device_name(ti
->table
->md
), bdevname(bdev
, b
),
312 (unsigned long long)start
,
313 (unsigned long long)len
,
314 (unsigned long long)dev_size
);
318 if (logical_block_size_sectors
<= 1)
321 if (start
& (logical_block_size_sectors
- 1)) {
322 DMWARN("%s: start=%llu not aligned to h/w "
323 "logical block size %u of %s",
324 dm_device_name(ti
->table
->md
),
325 (unsigned long long)start
,
326 limits
->logical_block_size
, bdevname(bdev
, b
));
330 if (len
& (logical_block_size_sectors
- 1)) {
331 DMWARN("%s: len=%llu not aligned to h/w "
332 "logical block size %u of %s",
333 dm_device_name(ti
->table
->md
),
334 (unsigned long long)len
,
335 limits
->logical_block_size
, bdevname(bdev
, b
));
343 * This upgrades the mode on an already open dm_dev, being
344 * careful to leave things as they were if we fail to reopen the
345 * device and not to touch the existing bdev field in case
346 * it is accessed concurrently inside dm_table_any_congested().
348 static int upgrade_mode(struct dm_dev_internal
*dd
, fmode_t new_mode
,
349 struct mapped_device
*md
)
352 struct dm_dev
*old_dev
, *new_dev
;
354 old_dev
= dd
->dm_dev
;
356 r
= dm_get_table_device(md
, dd
->dm_dev
->bdev
->bd_dev
,
357 dd
->dm_dev
->mode
| new_mode
, &new_dev
);
361 dd
->dm_dev
= new_dev
;
362 dm_put_table_device(md
, old_dev
);
368 * Add a device to the list, or just increment the usage count if
369 * it's already present.
371 int dm_get_device(struct dm_target
*ti
, const char *path
, fmode_t mode
,
372 struct dm_dev
**result
)
375 dev_t
uninitialized_var(dev
);
376 struct dm_dev_internal
*dd
;
377 struct dm_table
*t
= ti
->table
;
378 struct block_device
*bdev
;
382 /* convert the path to a device */
383 bdev
= lookup_bdev(path
);
385 dev
= name_to_dev_t(path
);
393 dd
= find_device(&t
->devices
, dev
);
395 dd
= kmalloc(sizeof(*dd
), GFP_KERNEL
);
399 if ((r
= dm_get_table_device(t
->md
, dev
, mode
, &dd
->dm_dev
))) {
404 atomic_set(&dd
->count
, 0);
405 list_add(&dd
->list
, &t
->devices
);
407 } else if (dd
->dm_dev
->mode
!= (mode
| dd
->dm_dev
->mode
)) {
408 r
= upgrade_mode(dd
, mode
, t
->md
);
412 atomic_inc(&dd
->count
);
414 *result
= dd
->dm_dev
;
417 EXPORT_SYMBOL(dm_get_device
);
419 static int dm_set_device_limits(struct dm_target
*ti
, struct dm_dev
*dev
,
420 sector_t start
, sector_t len
, void *data
)
422 struct queue_limits
*limits
= data
;
423 struct block_device
*bdev
= dev
->bdev
;
424 struct request_queue
*q
= bdev_get_queue(bdev
);
425 char b
[BDEVNAME_SIZE
];
428 DMWARN("%s: Cannot set limits for nonexistent device %s",
429 dm_device_name(ti
->table
->md
), bdevname(bdev
, b
));
433 if (bdev_stack_limits(limits
, bdev
, start
) < 0)
434 DMWARN("%s: adding target device %s caused an alignment inconsistency: "
435 "physical_block_size=%u, logical_block_size=%u, "
436 "alignment_offset=%u, start=%llu",
437 dm_device_name(ti
->table
->md
), bdevname(bdev
, b
),
438 q
->limits
.physical_block_size
,
439 q
->limits
.logical_block_size
,
440 q
->limits
.alignment_offset
,
441 (unsigned long long) start
<< SECTOR_SHIFT
);
447 * Decrement a device's use count and remove it if necessary.
449 void dm_put_device(struct dm_target
*ti
, struct dm_dev
*d
)
452 struct list_head
*devices
= &ti
->table
->devices
;
453 struct dm_dev_internal
*dd
;
455 list_for_each_entry(dd
, devices
, list
) {
456 if (dd
->dm_dev
== d
) {
462 DMWARN("%s: device %s not in table devices list",
463 dm_device_name(ti
->table
->md
), d
->name
);
466 if (atomic_dec_and_test(&dd
->count
)) {
467 dm_put_table_device(ti
->table
->md
, d
);
472 EXPORT_SYMBOL(dm_put_device
);
475 * Checks to see if the target joins onto the end of the table.
477 static int adjoin(struct dm_table
*table
, struct dm_target
*ti
)
479 struct dm_target
*prev
;
481 if (!table
->num_targets
)
484 prev
= &table
->targets
[table
->num_targets
- 1];
485 return (ti
->begin
== (prev
->begin
+ prev
->len
));
489 * Used to dynamically allocate the arg array.
491 * We do first allocation with GFP_NOIO because dm-mpath and dm-thin must
492 * process messages even if some device is suspended. These messages have a
493 * small fixed number of arguments.
495 * On the other hand, dm-switch needs to process bulk data using messages and
496 * excessive use of GFP_NOIO could cause trouble.
498 static char **realloc_argv(unsigned *array_size
, char **old_argv
)
505 new_size
= *array_size
* 2;
511 argv
= kmalloc(new_size
* sizeof(*argv
), gfp
);
513 memcpy(argv
, old_argv
, *array_size
* sizeof(*argv
));
514 *array_size
= new_size
;
522 * Destructively splits up the argument list to pass to ctr.
524 int dm_split_args(int *argc
, char ***argvp
, char *input
)
526 char *start
, *end
= input
, *out
, **argv
= NULL
;
527 unsigned array_size
= 0;
536 argv
= realloc_argv(&array_size
, argv
);
541 /* Skip whitespace */
542 start
= skip_spaces(end
);
545 break; /* success, we hit the end */
547 /* 'out' is used to remove any back-quotes */
550 /* Everything apart from '\0' can be quoted */
551 if (*end
== '\\' && *(end
+ 1)) {
558 break; /* end of token */
563 /* have we already filled the array ? */
564 if ((*argc
+ 1) > array_size
) {
565 argv
= realloc_argv(&array_size
, argv
);
570 /* we know this is whitespace */
574 /* terminate the string and put it in the array */
585 * Impose necessary and sufficient conditions on a devices's table such
586 * that any incoming bio which respects its logical_block_size can be
587 * processed successfully. If it falls across the boundary between
588 * two or more targets, the size of each piece it gets split into must
589 * be compatible with the logical_block_size of the target processing it.
591 static int validate_hardware_logical_block_alignment(struct dm_table
*table
,
592 struct queue_limits
*limits
)
595 * This function uses arithmetic modulo the logical_block_size
596 * (in units of 512-byte sectors).
598 unsigned short device_logical_block_size_sects
=
599 limits
->logical_block_size
>> SECTOR_SHIFT
;
602 * Offset of the start of the next table entry, mod logical_block_size.
604 unsigned short next_target_start
= 0;
607 * Given an aligned bio that extends beyond the end of a
608 * target, how many sectors must the next target handle?
610 unsigned short remaining
= 0;
612 struct dm_target
*uninitialized_var(ti
);
613 struct queue_limits ti_limits
;
617 * Check each entry in the table in turn.
619 while (i
< dm_table_get_num_targets(table
)) {
620 ti
= dm_table_get_target(table
, i
++);
622 blk_set_stacking_limits(&ti_limits
);
624 /* combine all target devices' limits */
625 if (ti
->type
->iterate_devices
)
626 ti
->type
->iterate_devices(ti
, dm_set_device_limits
,
630 * If the remaining sectors fall entirely within this
631 * table entry are they compatible with its logical_block_size?
633 if (remaining
< ti
->len
&&
634 remaining
& ((ti_limits
.logical_block_size
>>
639 (unsigned short) ((next_target_start
+ ti
->len
) &
640 (device_logical_block_size_sects
- 1));
641 remaining
= next_target_start
?
642 device_logical_block_size_sects
- next_target_start
: 0;
646 DMWARN("%s: table line %u (start sect %llu len %llu) "
647 "not aligned to h/w logical block size %u",
648 dm_device_name(table
->md
), i
,
649 (unsigned long long) ti
->begin
,
650 (unsigned long long) ti
->len
,
651 limits
->logical_block_size
);
658 int dm_table_add_target(struct dm_table
*t
, const char *type
,
659 sector_t start
, sector_t len
, char *params
)
661 int r
= -EINVAL
, argc
;
663 struct dm_target
*tgt
;
666 DMERR("%s: target type %s must appear alone in table",
667 dm_device_name(t
->md
), t
->targets
->type
->name
);
671 BUG_ON(t
->num_targets
>= t
->num_allocated
);
673 tgt
= t
->targets
+ t
->num_targets
;
674 memset(tgt
, 0, sizeof(*tgt
));
677 DMERR("%s: zero-length target", dm_device_name(t
->md
));
681 tgt
->type
= dm_get_target_type(type
);
683 DMERR("%s: %s: unknown target type", dm_device_name(t
->md
),
688 if (dm_target_needs_singleton(tgt
->type
)) {
689 if (t
->num_targets
) {
690 DMERR("%s: target type %s must appear alone in table",
691 dm_device_name(t
->md
), type
);
697 if (dm_target_always_writeable(tgt
->type
) && !(t
->mode
& FMODE_WRITE
)) {
698 DMERR("%s: target type %s may not be included in read-only tables",
699 dm_device_name(t
->md
), type
);
703 if (t
->immutable_target_type
) {
704 if (t
->immutable_target_type
!= tgt
->type
) {
705 DMERR("%s: immutable target type %s cannot be mixed with other target types",
706 dm_device_name(t
->md
), t
->immutable_target_type
->name
);
709 } else if (dm_target_is_immutable(tgt
->type
)) {
710 if (t
->num_targets
) {
711 DMERR("%s: immutable target type %s cannot be mixed with other target types",
712 dm_device_name(t
->md
), tgt
->type
->name
);
715 t
->immutable_target_type
= tgt
->type
;
721 tgt
->error
= "Unknown error";
724 * Does this target adjoin the previous one ?
726 if (!adjoin(t
, tgt
)) {
727 tgt
->error
= "Gap in table";
732 r
= dm_split_args(&argc
, &argv
, params
);
734 tgt
->error
= "couldn't split parameters (insufficient memory)";
738 r
= tgt
->type
->ctr(tgt
, argc
, argv
);
743 t
->highs
[t
->num_targets
++] = tgt
->begin
+ tgt
->len
- 1;
745 if (!tgt
->num_discard_bios
&& tgt
->discards_supported
)
746 DMWARN("%s: %s: ignoring discards_supported because num_discard_bios is zero.",
747 dm_device_name(t
->md
), type
);
752 DMERR("%s: %s: %s", dm_device_name(t
->md
), type
, tgt
->error
);
753 dm_put_target_type(tgt
->type
);
758 * Target argument parsing helpers.
760 static int validate_next_arg(struct dm_arg
*arg
, struct dm_arg_set
*arg_set
,
761 unsigned *value
, char **error
, unsigned grouped
)
763 const char *arg_str
= dm_shift_arg(arg_set
);
767 (sscanf(arg_str
, "%u%c", value
, &dummy
) != 1) ||
768 (*value
< arg
->min
) ||
769 (*value
> arg
->max
) ||
770 (grouped
&& arg_set
->argc
< *value
)) {
778 int dm_read_arg(struct dm_arg
*arg
, struct dm_arg_set
*arg_set
,
779 unsigned *value
, char **error
)
781 return validate_next_arg(arg
, arg_set
, value
, error
, 0);
783 EXPORT_SYMBOL(dm_read_arg
);
785 int dm_read_arg_group(struct dm_arg
*arg
, struct dm_arg_set
*arg_set
,
786 unsigned *value
, char **error
)
788 return validate_next_arg(arg
, arg_set
, value
, error
, 1);
790 EXPORT_SYMBOL(dm_read_arg_group
);
792 const char *dm_shift_arg(struct dm_arg_set
*as
)
805 EXPORT_SYMBOL(dm_shift_arg
);
807 void dm_consume_args(struct dm_arg_set
*as
, unsigned num_args
)
809 BUG_ON(as
->argc
< num_args
);
810 as
->argc
-= num_args
;
811 as
->argv
+= num_args
;
813 EXPORT_SYMBOL(dm_consume_args
);
815 static bool __table_type_request_based(unsigned table_type
)
817 return (table_type
== DM_TYPE_REQUEST_BASED
||
818 table_type
== DM_TYPE_MQ_REQUEST_BASED
);
821 static int dm_table_set_type(struct dm_table
*t
)
824 unsigned bio_based
= 0, request_based
= 0, hybrid
= 0;
825 bool use_blk_mq
= false;
826 struct dm_target
*tgt
;
827 struct dm_dev_internal
*dd
;
828 struct list_head
*devices
;
829 unsigned live_md_type
= dm_get_md_type(t
->md
);
831 for (i
= 0; i
< t
->num_targets
; i
++) {
832 tgt
= t
->targets
+ i
;
833 if (dm_target_hybrid(tgt
))
835 else if (dm_target_request_based(tgt
))
840 if (bio_based
&& request_based
) {
841 DMWARN("Inconsistent table: different target types"
842 " can't be mixed up");
847 if (hybrid
&& !bio_based
&& !request_based
) {
849 * The targets can work either way.
850 * Determine the type from the live device.
851 * Default to bio-based if device is new.
853 if (__table_type_request_based(live_md_type
))
860 /* We must use this table as bio-based */
861 t
->type
= DM_TYPE_BIO_BASED
;
865 BUG_ON(!request_based
); /* No targets in this table */
868 * Request-based dm supports only tables that have a single target now.
869 * To support multiple targets, request splitting support is needed,
870 * and that needs lots of changes in the block-layer.
871 * (e.g. request completion process for partial completion.)
873 if (t
->num_targets
> 1) {
874 DMWARN("Request-based dm doesn't support multiple targets yet");
878 /* Non-request-stackable devices can't be used for request-based dm */
879 devices
= dm_table_get_devices(t
);
880 list_for_each_entry(dd
, devices
, list
) {
881 struct request_queue
*q
= bdev_get_queue(dd
->dm_dev
->bdev
);
883 if (!blk_queue_stackable(q
)) {
884 DMERR("table load rejected: including"
885 " non-request-stackable devices");
894 /* verify _all_ devices in the table are blk-mq devices */
895 list_for_each_entry(dd
, devices
, list
)
896 if (!bdev_get_queue(dd
->dm_dev
->bdev
)->mq_ops
) {
897 DMERR("table load rejected: not all devices"
898 " are blk-mq request-stackable");
901 t
->type
= DM_TYPE_MQ_REQUEST_BASED
;
903 } else if (list_empty(devices
) && __table_type_request_based(live_md_type
)) {
904 /* inherit live MD type */
905 t
->type
= live_md_type
;
908 t
->type
= DM_TYPE_REQUEST_BASED
;
913 unsigned dm_table_get_type(struct dm_table
*t
)
918 struct target_type
*dm_table_get_immutable_target_type(struct dm_table
*t
)
920 return t
->immutable_target_type
;
923 bool dm_table_request_based(struct dm_table
*t
)
925 return __table_type_request_based(dm_table_get_type(t
));
928 bool dm_table_mq_request_based(struct dm_table
*t
)
930 return dm_table_get_type(t
) == DM_TYPE_MQ_REQUEST_BASED
;
933 static int dm_table_alloc_md_mempools(struct dm_table
*t
, struct mapped_device
*md
)
935 unsigned type
= dm_table_get_type(t
);
936 unsigned per_bio_data_size
= 0;
937 struct dm_target
*tgt
;
940 if (unlikely(type
== DM_TYPE_NONE
)) {
941 DMWARN("no table type is set, can't allocate mempools");
945 if (type
== DM_TYPE_BIO_BASED
)
946 for (i
= 0; i
< t
->num_targets
; i
++) {
947 tgt
= t
->targets
+ i
;
948 per_bio_data_size
= max(per_bio_data_size
, tgt
->per_bio_data_size
);
951 t
->mempools
= dm_alloc_md_mempools(md
, type
, t
->integrity_supported
, per_bio_data_size
);
958 void dm_table_free_md_mempools(struct dm_table
*t
)
960 dm_free_md_mempools(t
->mempools
);
964 struct dm_md_mempools
*dm_table_get_md_mempools(struct dm_table
*t
)
969 static int setup_indexes(struct dm_table
*t
)
972 unsigned int total
= 0;
975 /* allocate the space for *all* the indexes */
976 for (i
= t
->depth
- 2; i
>= 0; i
--) {
977 t
->counts
[i
] = dm_div_up(t
->counts
[i
+ 1], CHILDREN_PER_NODE
);
978 total
+= t
->counts
[i
];
981 indexes
= (sector_t
*) dm_vcalloc(total
, (unsigned long) NODE_SIZE
);
985 /* set up internal nodes, bottom-up */
986 for (i
= t
->depth
- 2; i
>= 0; i
--) {
987 t
->index
[i
] = indexes
;
988 indexes
+= (KEYS_PER_NODE
* t
->counts
[i
]);
989 setup_btree_index(i
, t
);
996 * Builds the btree to index the map.
998 static int dm_table_build_index(struct dm_table
*t
)
1001 unsigned int leaf_nodes
;
1003 /* how many indexes will the btree have ? */
1004 leaf_nodes
= dm_div_up(t
->num_targets
, KEYS_PER_NODE
);
1005 t
->depth
= 1 + int_log(leaf_nodes
, CHILDREN_PER_NODE
);
1007 /* leaf layer has already been set up */
1008 t
->counts
[t
->depth
- 1] = leaf_nodes
;
1009 t
->index
[t
->depth
- 1] = t
->highs
;
1012 r
= setup_indexes(t
);
1018 * Get a disk whose integrity profile reflects the table's profile.
1019 * If %match_all is true, all devices' profiles must match.
1020 * If %match_all is false, all devices must at least have an
1021 * allocated integrity profile; but uninitialized is ok.
1022 * Returns NULL if integrity support was inconsistent or unavailable.
1024 static struct gendisk
* dm_table_get_integrity_disk(struct dm_table
*t
,
1027 struct list_head
*devices
= dm_table_get_devices(t
);
1028 struct dm_dev_internal
*dd
= NULL
;
1029 struct gendisk
*prev_disk
= NULL
, *template_disk
= NULL
;
1031 list_for_each_entry(dd
, devices
, list
) {
1032 template_disk
= dd
->dm_dev
->bdev
->bd_disk
;
1033 if (!blk_get_integrity(template_disk
))
1035 if (!match_all
&& !blk_integrity_is_initialized(template_disk
))
1036 continue; /* skip uninitialized profiles */
1037 else if (prev_disk
&&
1038 blk_integrity_compare(prev_disk
, template_disk
) < 0)
1040 prev_disk
= template_disk
;
1043 return template_disk
;
1047 DMWARN("%s: integrity not set: %s and %s profile mismatch",
1048 dm_device_name(t
->md
),
1049 prev_disk
->disk_name
,
1050 template_disk
->disk_name
);
1055 * Register the mapped device for blk_integrity support if
1056 * the underlying devices have an integrity profile. But all devices
1057 * may not have matching profiles (checking all devices isn't reliable
1058 * during table load because this table may use other DM device(s) which
1059 * must be resumed before they will have an initialized integity profile).
1060 * Stacked DM devices force a 2 stage integrity profile validation:
1061 * 1 - during load, validate all initialized integrity profiles match
1062 * 2 - during resume, validate all integrity profiles match
1064 static int dm_table_prealloc_integrity(struct dm_table
*t
, struct mapped_device
*md
)
1066 struct gendisk
*template_disk
= NULL
;
1068 template_disk
= dm_table_get_integrity_disk(t
, false);
1072 if (!blk_integrity_is_initialized(dm_disk(md
))) {
1073 t
->integrity_supported
= 1;
1074 return blk_integrity_register(dm_disk(md
), NULL
);
1078 * If DM device already has an initalized integrity
1079 * profile the new profile should not conflict.
1081 if (blk_integrity_is_initialized(template_disk
) &&
1082 blk_integrity_compare(dm_disk(md
), template_disk
) < 0) {
1083 DMWARN("%s: conflict with existing integrity profile: "
1084 "%s profile mismatch",
1085 dm_device_name(t
->md
),
1086 template_disk
->disk_name
);
1090 /* Preserve existing initialized integrity profile */
1091 t
->integrity_supported
= 1;
1096 * Prepares the table for use by building the indices,
1097 * setting the type, and allocating mempools.
1099 int dm_table_complete(struct dm_table
*t
)
1103 r
= dm_table_set_type(t
);
1105 DMERR("unable to set table type");
1109 r
= dm_table_build_index(t
);
1111 DMERR("unable to build btrees");
1115 r
= dm_table_prealloc_integrity(t
, t
->md
);
1117 DMERR("could not register integrity profile.");
1121 r
= dm_table_alloc_md_mempools(t
, t
->md
);
1123 DMERR("unable to allocate mempools");
1128 static DEFINE_MUTEX(_event_lock
);
1129 void dm_table_event_callback(struct dm_table
*t
,
1130 void (*fn
)(void *), void *context
)
1132 mutex_lock(&_event_lock
);
1134 t
->event_context
= context
;
1135 mutex_unlock(&_event_lock
);
1138 void dm_table_event(struct dm_table
*t
)
1141 * You can no longer call dm_table_event() from interrupt
1142 * context, use a bottom half instead.
1144 BUG_ON(in_interrupt());
1146 mutex_lock(&_event_lock
);
1148 t
->event_fn(t
->event_context
);
1149 mutex_unlock(&_event_lock
);
1151 EXPORT_SYMBOL(dm_table_event
);
1153 sector_t
dm_table_get_size(struct dm_table
*t
)
1155 return t
->num_targets
? (t
->highs
[t
->num_targets
- 1] + 1) : 0;
1157 EXPORT_SYMBOL(dm_table_get_size
);
1159 struct dm_target
*dm_table_get_target(struct dm_table
*t
, unsigned int index
)
1161 if (index
>= t
->num_targets
)
1164 return t
->targets
+ index
;
1168 * Search the btree for the correct target.
1170 * Caller should check returned pointer with dm_target_is_valid()
1171 * to trap I/O beyond end of device.
1173 struct dm_target
*dm_table_find_target(struct dm_table
*t
, sector_t sector
)
1175 unsigned int l
, n
= 0, k
= 0;
1178 for (l
= 0; l
< t
->depth
; l
++) {
1179 n
= get_child(n
, k
);
1180 node
= get_node(t
, l
, n
);
1182 for (k
= 0; k
< KEYS_PER_NODE
; k
++)
1183 if (node
[k
] >= sector
)
1187 return &t
->targets
[(KEYS_PER_NODE
* n
) + k
];
1190 static int count_device(struct dm_target
*ti
, struct dm_dev
*dev
,
1191 sector_t start
, sector_t len
, void *data
)
1193 unsigned *num_devices
= data
;
1201 * Check whether a table has no data devices attached using each
1202 * target's iterate_devices method.
1203 * Returns false if the result is unknown because a target doesn't
1204 * support iterate_devices.
1206 bool dm_table_has_no_data_devices(struct dm_table
*table
)
1208 struct dm_target
*uninitialized_var(ti
);
1209 unsigned i
= 0, num_devices
= 0;
1211 while (i
< dm_table_get_num_targets(table
)) {
1212 ti
= dm_table_get_target(table
, i
++);
1214 if (!ti
->type
->iterate_devices
)
1217 ti
->type
->iterate_devices(ti
, count_device
, &num_devices
);
1226 * Establish the new table's queue_limits and validate them.
1228 int dm_calculate_queue_limits(struct dm_table
*table
,
1229 struct queue_limits
*limits
)
1231 struct dm_target
*uninitialized_var(ti
);
1232 struct queue_limits ti_limits
;
1235 blk_set_stacking_limits(limits
);
1237 while (i
< dm_table_get_num_targets(table
)) {
1238 blk_set_stacking_limits(&ti_limits
);
1240 ti
= dm_table_get_target(table
, i
++);
1242 if (!ti
->type
->iterate_devices
)
1243 goto combine_limits
;
1246 * Combine queue limits of all the devices this target uses.
1248 ti
->type
->iterate_devices(ti
, dm_set_device_limits
,
1251 /* Set I/O hints portion of queue limits */
1252 if (ti
->type
->io_hints
)
1253 ti
->type
->io_hints(ti
, &ti_limits
);
1256 * Check each device area is consistent with the target's
1257 * overall queue limits.
1259 if (ti
->type
->iterate_devices(ti
, device_area_is_invalid
,
1265 * Merge this target's queue limits into the overall limits
1268 if (blk_stack_limits(limits
, &ti_limits
, 0) < 0)
1269 DMWARN("%s: adding target device "
1270 "(start sect %llu len %llu) "
1271 "caused an alignment inconsistency",
1272 dm_device_name(table
->md
),
1273 (unsigned long long) ti
->begin
,
1274 (unsigned long long) ti
->len
);
1277 return validate_hardware_logical_block_alignment(table
, limits
);
1281 * Set the integrity profile for this device if all devices used have
1282 * matching profiles. We're quite deep in the resume path but still
1283 * don't know if all devices (particularly DM devices this device
1284 * may be stacked on) have matching profiles. Even if the profiles
1285 * don't match we have no way to fail (to resume) at this point.
1287 static void dm_table_set_integrity(struct dm_table
*t
)
1289 struct gendisk
*template_disk
= NULL
;
1291 if (!blk_get_integrity(dm_disk(t
->md
)))
1294 template_disk
= dm_table_get_integrity_disk(t
, true);
1296 blk_integrity_register(dm_disk(t
->md
),
1297 blk_get_integrity(template_disk
));
1298 else if (blk_integrity_is_initialized(dm_disk(t
->md
)))
1299 DMWARN("%s: device no longer has a valid integrity profile",
1300 dm_device_name(t
->md
));
1302 DMWARN("%s: unable to establish an integrity profile",
1303 dm_device_name(t
->md
));
1306 static int device_flush_capable(struct dm_target
*ti
, struct dm_dev
*dev
,
1307 sector_t start
, sector_t len
, void *data
)
1309 unsigned flush
= (*(unsigned *)data
);
1310 struct request_queue
*q
= bdev_get_queue(dev
->bdev
);
1312 return q
&& (q
->flush_flags
& flush
);
1315 static bool dm_table_supports_flush(struct dm_table
*t
, unsigned flush
)
1317 struct dm_target
*ti
;
1321 * Require at least one underlying device to support flushes.
1322 * t->devices includes internal dm devices such as mirror logs
1323 * so we need to use iterate_devices here, which targets
1324 * supporting flushes must provide.
1326 while (i
< dm_table_get_num_targets(t
)) {
1327 ti
= dm_table_get_target(t
, i
++);
1329 if (!ti
->num_flush_bios
)
1332 if (ti
->flush_supported
)
1335 if (ti
->type
->iterate_devices
&&
1336 ti
->type
->iterate_devices(ti
, device_flush_capable
, &flush
))
1343 static bool dm_table_discard_zeroes_data(struct dm_table
*t
)
1345 struct dm_target
*ti
;
1348 /* Ensure that all targets supports discard_zeroes_data. */
1349 while (i
< dm_table_get_num_targets(t
)) {
1350 ti
= dm_table_get_target(t
, i
++);
1352 if (ti
->discard_zeroes_data_unsupported
)
1359 static int device_is_nonrot(struct dm_target
*ti
, struct dm_dev
*dev
,
1360 sector_t start
, sector_t len
, void *data
)
1362 struct request_queue
*q
= bdev_get_queue(dev
->bdev
);
1364 return q
&& blk_queue_nonrot(q
);
1367 static int device_is_not_random(struct dm_target
*ti
, struct dm_dev
*dev
,
1368 sector_t start
, sector_t len
, void *data
)
1370 struct request_queue
*q
= bdev_get_queue(dev
->bdev
);
1372 return q
&& !blk_queue_add_random(q
);
1375 static int queue_supports_sg_merge(struct dm_target
*ti
, struct dm_dev
*dev
,
1376 sector_t start
, sector_t len
, void *data
)
1378 struct request_queue
*q
= bdev_get_queue(dev
->bdev
);
1380 return q
&& !test_bit(QUEUE_FLAG_NO_SG_MERGE
, &q
->queue_flags
);
1383 static bool dm_table_all_devices_attribute(struct dm_table
*t
,
1384 iterate_devices_callout_fn func
)
1386 struct dm_target
*ti
;
1389 while (i
< dm_table_get_num_targets(t
)) {
1390 ti
= dm_table_get_target(t
, i
++);
1392 if (!ti
->type
->iterate_devices
||
1393 !ti
->type
->iterate_devices(ti
, func
, NULL
))
1400 static int device_not_write_same_capable(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
&& !q
->limits
.max_write_same_sectors
;
1408 static bool dm_table_supports_write_same(struct dm_table
*t
)
1410 struct dm_target
*ti
;
1413 while (i
< dm_table_get_num_targets(t
)) {
1414 ti
= dm_table_get_target(t
, i
++);
1416 if (!ti
->num_write_same_bios
)
1419 if (!ti
->type
->iterate_devices
||
1420 ti
->type
->iterate_devices(ti
, device_not_write_same_capable
, NULL
))
1427 static int device_discard_capable(struct dm_target
*ti
, struct dm_dev
*dev
,
1428 sector_t start
, sector_t len
, void *data
)
1430 struct request_queue
*q
= bdev_get_queue(dev
->bdev
);
1432 return q
&& blk_queue_discard(q
);
1435 static bool dm_table_supports_discards(struct dm_table
*t
)
1437 struct dm_target
*ti
;
1441 * Unless any target used by the table set discards_supported,
1442 * require at least one underlying device to support discards.
1443 * t->devices includes internal dm devices such as mirror logs
1444 * so we need to use iterate_devices here, which targets
1445 * supporting discard selectively must provide.
1447 while (i
< dm_table_get_num_targets(t
)) {
1448 ti
= dm_table_get_target(t
, i
++);
1450 if (!ti
->num_discard_bios
)
1453 if (ti
->discards_supported
)
1456 if (ti
->type
->iterate_devices
&&
1457 ti
->type
->iterate_devices(ti
, device_discard_capable
, NULL
))
1464 void dm_table_set_restrictions(struct dm_table
*t
, struct request_queue
*q
,
1465 struct queue_limits
*limits
)
1470 * Copy table's limits to the DM device's request_queue
1472 q
->limits
= *limits
;
1474 if (!dm_table_supports_discards(t
))
1475 queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD
, q
);
1477 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD
, q
);
1479 if (dm_table_supports_flush(t
, REQ_FLUSH
)) {
1481 if (dm_table_supports_flush(t
, REQ_FUA
))
1484 blk_queue_flush(q
, flush
);
1486 if (!dm_table_discard_zeroes_data(t
))
1487 q
->limits
.discard_zeroes_data
= 0;
1489 /* Ensure that all underlying devices are non-rotational. */
1490 if (dm_table_all_devices_attribute(t
, device_is_nonrot
))
1491 queue_flag_set_unlocked(QUEUE_FLAG_NONROT
, q
);
1493 queue_flag_clear_unlocked(QUEUE_FLAG_NONROT
, q
);
1495 if (!dm_table_supports_write_same(t
))
1496 q
->limits
.max_write_same_sectors
= 0;
1498 if (dm_table_all_devices_attribute(t
, queue_supports_sg_merge
))
1499 queue_flag_clear_unlocked(QUEUE_FLAG_NO_SG_MERGE
, q
);
1501 queue_flag_set_unlocked(QUEUE_FLAG_NO_SG_MERGE
, q
);
1503 dm_table_set_integrity(t
);
1506 * Determine whether or not this queue's I/O timings contribute
1507 * to the entropy pool, Only request-based targets use this.
1508 * Clear QUEUE_FLAG_ADD_RANDOM if any underlying device does not
1511 if (blk_queue_add_random(q
) && dm_table_all_devices_attribute(t
, device_is_not_random
))
1512 queue_flag_clear_unlocked(QUEUE_FLAG_ADD_RANDOM
, q
);
1515 * QUEUE_FLAG_STACKABLE must be set after all queue settings are
1516 * visible to other CPUs because, once the flag is set, incoming bios
1517 * are processed by request-based dm, which refers to the queue
1519 * Until the flag set, bios are passed to bio-based dm and queued to
1520 * md->deferred where queue settings are not needed yet.
1521 * Those bios are passed to request-based dm at the resume time.
1524 if (dm_table_request_based(t
))
1525 queue_flag_set_unlocked(QUEUE_FLAG_STACKABLE
, q
);
1528 unsigned int dm_table_get_num_targets(struct dm_table
*t
)
1530 return t
->num_targets
;
1533 struct list_head
*dm_table_get_devices(struct dm_table
*t
)
1538 fmode_t
dm_table_get_mode(struct dm_table
*t
)
1542 EXPORT_SYMBOL(dm_table_get_mode
);
1550 static void suspend_targets(struct dm_table
*t
, enum suspend_mode mode
)
1552 int i
= t
->num_targets
;
1553 struct dm_target
*ti
= t
->targets
;
1558 if (ti
->type
->presuspend
)
1559 ti
->type
->presuspend(ti
);
1561 case PRESUSPEND_UNDO
:
1562 if (ti
->type
->presuspend_undo
)
1563 ti
->type
->presuspend_undo(ti
);
1566 if (ti
->type
->postsuspend
)
1567 ti
->type
->postsuspend(ti
);
1574 void dm_table_presuspend_targets(struct dm_table
*t
)
1579 suspend_targets(t
, PRESUSPEND
);
1582 void dm_table_presuspend_undo_targets(struct dm_table
*t
)
1587 suspend_targets(t
, PRESUSPEND_UNDO
);
1590 void dm_table_postsuspend_targets(struct dm_table
*t
)
1595 suspend_targets(t
, POSTSUSPEND
);
1598 int dm_table_resume_targets(struct dm_table
*t
)
1602 for (i
= 0; i
< t
->num_targets
; i
++) {
1603 struct dm_target
*ti
= t
->targets
+ i
;
1605 if (!ti
->type
->preresume
)
1608 r
= ti
->type
->preresume(ti
);
1610 DMERR("%s: %s: preresume failed, error = %d",
1611 dm_device_name(t
->md
), ti
->type
->name
, r
);
1616 for (i
= 0; i
< t
->num_targets
; i
++) {
1617 struct dm_target
*ti
= t
->targets
+ i
;
1619 if (ti
->type
->resume
)
1620 ti
->type
->resume(ti
);
1626 void dm_table_add_target_callbacks(struct dm_table
*t
, struct dm_target_callbacks
*cb
)
1628 list_add(&cb
->list
, &t
->target_callbacks
);
1630 EXPORT_SYMBOL_GPL(dm_table_add_target_callbacks
);
1632 int dm_table_any_congested(struct dm_table
*t
, int bdi_bits
)
1634 struct dm_dev_internal
*dd
;
1635 struct list_head
*devices
= dm_table_get_devices(t
);
1636 struct dm_target_callbacks
*cb
;
1639 list_for_each_entry(dd
, devices
, list
) {
1640 struct request_queue
*q
= bdev_get_queue(dd
->dm_dev
->bdev
);
1641 char b
[BDEVNAME_SIZE
];
1644 r
|= bdi_congested(&q
->backing_dev_info
, bdi_bits
);
1646 DMWARN_LIMIT("%s: any_congested: nonexistent device %s",
1647 dm_device_name(t
->md
),
1648 bdevname(dd
->dm_dev
->bdev
, b
));
1651 list_for_each_entry(cb
, &t
->target_callbacks
, list
)
1652 if (cb
->congested_fn
)
1653 r
|= cb
->congested_fn(cb
, bdi_bits
);
1658 struct mapped_device
*dm_table_get_md(struct dm_table
*t
)
1662 EXPORT_SYMBOL(dm_table_get_md
);
1664 void dm_table_run_md_queue_async(struct dm_table
*t
)
1666 struct mapped_device
*md
;
1667 struct request_queue
*queue
;
1668 unsigned long flags
;
1670 if (!dm_table_request_based(t
))
1673 md
= dm_table_get_md(t
);
1674 queue
= dm_get_md_queue(md
);
1677 blk_mq_run_hw_queues(queue
, true);
1679 spin_lock_irqsave(queue
->queue_lock
, flags
);
1680 blk_run_queue_async(queue
);
1681 spin_unlock_irqrestore(queue
->queue_lock
, flags
);
1685 EXPORT_SYMBOL(dm_table_run_md_queue_async
);