2 * Copyright (C) 2011 Red Hat, Inc.
4 * This file is released under the GPL.
7 #include "dm-space-map.h"
8 #include "dm-space-map-common.h"
9 #include "dm-space-map-metadata.h"
11 #include <linux/list.h>
12 #include <linux/slab.h>
13 #include <linux/device-mapper.h>
15 #define DM_MSG_PREFIX "space map metadata"
17 /*----------------------------------------------------------------*/
20 * An edge triggered threshold.
26 dm_block_t current_value
;
27 dm_sm_threshold_fn fn
;
31 static void threshold_init(struct threshold
*t
)
33 t
->threshold_set
= false;
37 static void set_threshold(struct threshold
*t
, dm_block_t value
,
38 dm_sm_threshold_fn fn
, void *context
)
40 t
->threshold_set
= true;
46 static bool below_threshold(struct threshold
*t
, dm_block_t value
)
48 return t
->threshold_set
&& value
<= t
->threshold
;
51 static bool threshold_already_triggered(struct threshold
*t
)
53 return t
->value_set
&& below_threshold(t
, t
->current_value
);
56 static void check_threshold(struct threshold
*t
, dm_block_t value
)
58 if (below_threshold(t
, value
) &&
59 !threshold_already_triggered(t
))
63 t
->current_value
= value
;
66 /*----------------------------------------------------------------*/
69 * Space map interface.
71 * The low level disk format is written using the standard btree and
72 * transaction manager. This means that performing disk operations may
73 * cause us to recurse into the space map in order to allocate new blocks.
74 * For this reason we have a pool of pre-allocated blocks large enough to
75 * service any metadata_ll_disk operation.
79 * FIXME: we should calculate this based on the size of the device.
80 * Only the metadata space map needs this functionality.
82 #define MAX_RECURSIVE_ALLOCATIONS 1024
90 enum block_op_type type
;
94 struct bop_ring_buffer
{
97 struct block_op bops
[MAX_RECURSIVE_ALLOCATIONS
+ 1];
100 static void brb_init(struct bop_ring_buffer
*brb
)
106 static bool brb_empty(struct bop_ring_buffer
*brb
)
108 return brb
->begin
== brb
->end
;
111 static unsigned brb_next(struct bop_ring_buffer
*brb
, unsigned old
)
113 unsigned r
= old
+ 1;
114 return (r
>= (sizeof(brb
->bops
) / sizeof(*brb
->bops
))) ? 0 : r
;
117 static int brb_push(struct bop_ring_buffer
*brb
,
118 enum block_op_type type
, dm_block_t b
)
120 struct block_op
*bop
;
121 unsigned next
= brb_next(brb
, brb
->end
);
124 * We don't allow the last bop to be filled, this way we can
125 * differentiate between full and empty.
127 if (next
== brb
->begin
)
130 bop
= brb
->bops
+ brb
->end
;
139 static int brb_pop(struct bop_ring_buffer
*brb
, struct block_op
*result
)
141 struct block_op
*bop
;
146 bop
= brb
->bops
+ brb
->begin
;
147 result
->type
= bop
->type
;
148 result
->block
= bop
->block
;
150 brb
->begin
= brb_next(brb
, brb
->begin
);
155 /*----------------------------------------------------------------*/
158 struct dm_space_map sm
;
161 struct ll_disk old_ll
;
165 unsigned recursion_count
;
166 unsigned allocated_this_transaction
;
167 struct bop_ring_buffer uncommitted
;
169 struct threshold threshold
;
172 static int add_bop(struct sm_metadata
*smm
, enum block_op_type type
, dm_block_t b
)
174 int r
= brb_push(&smm
->uncommitted
, type
, b
);
177 DMERR("too many recursive allocations");
184 static int commit_bop(struct sm_metadata
*smm
, struct block_op
*op
)
187 enum allocation_event ev
;
191 r
= sm_ll_inc(&smm
->ll
, op
->block
, &ev
);
195 r
= sm_ll_dec(&smm
->ll
, op
->block
, &ev
);
202 static void in(struct sm_metadata
*smm
)
204 smm
->recursion_count
++;
207 static int out(struct sm_metadata
*smm
)
212 * If we're not recursing then very bad things are happening.
214 if (!smm
->recursion_count
) {
215 DMERR("lost track of recursion depth");
219 if (smm
->recursion_count
== 1) {
220 while (!brb_empty(&smm
->uncommitted
)) {
223 r
= brb_pop(&smm
->uncommitted
, &bop
);
225 DMERR("bug in bop ring buffer");
229 r
= commit_bop(smm
, &bop
);
235 smm
->recursion_count
--;
241 * When using the out() function above, we often want to combine an error
242 * code for the operation run in the recursive context with that from
245 static int combine_errors(int r1
, int r2
)
250 static int recursing(struct sm_metadata
*smm
)
252 return smm
->recursion_count
;
255 static void sm_metadata_destroy(struct dm_space_map
*sm
)
257 struct sm_metadata
*smm
= container_of(sm
, struct sm_metadata
, sm
);
262 static int sm_metadata_get_nr_blocks(struct dm_space_map
*sm
, dm_block_t
*count
)
264 struct sm_metadata
*smm
= container_of(sm
, struct sm_metadata
, sm
);
266 *count
= smm
->ll
.nr_blocks
;
271 static int sm_metadata_get_nr_free(struct dm_space_map
*sm
, dm_block_t
*count
)
273 struct sm_metadata
*smm
= container_of(sm
, struct sm_metadata
, sm
);
275 *count
= smm
->old_ll
.nr_blocks
- smm
->old_ll
.nr_allocated
-
276 smm
->allocated_this_transaction
;
281 static int sm_metadata_get_count(struct dm_space_map
*sm
, dm_block_t b
,
286 struct sm_metadata
*smm
= container_of(sm
, struct sm_metadata
, sm
);
287 unsigned adjustment
= 0;
290 * We may have some uncommitted adjustments to add. This list
291 * should always be really short.
293 for (i
= smm
->uncommitted
.begin
;
294 i
!= smm
->uncommitted
.end
;
295 i
= brb_next(&smm
->uncommitted
, i
)) {
296 struct block_op
*op
= smm
->uncommitted
.bops
+ i
;
312 r
= sm_ll_lookup(&smm
->ll
, b
, result
);
316 *result
+= adjustment
;
321 static int sm_metadata_count_is_more_than_one(struct dm_space_map
*sm
,
322 dm_block_t b
, int *result
)
324 int r
, adjustment
= 0;
326 struct sm_metadata
*smm
= container_of(sm
, struct sm_metadata
, sm
);
330 * We may have some uncommitted adjustments to add. This list
331 * should always be really short.
333 for (i
= smm
->uncommitted
.begin
;
334 i
!= smm
->uncommitted
.end
;
335 i
= brb_next(&smm
->uncommitted
, i
)) {
337 struct block_op
*op
= smm
->uncommitted
.bops
+ i
;
353 if (adjustment
> 1) {
358 r
= sm_ll_lookup_bitmap(&smm
->ll
, b
, &rc
);
364 * We err on the side of caution, and always return true.
368 *result
= rc
+ adjustment
> 1;
373 static int sm_metadata_set_count(struct dm_space_map
*sm
, dm_block_t b
,
377 enum allocation_event ev
;
378 struct sm_metadata
*smm
= container_of(sm
, struct sm_metadata
, sm
);
380 if (smm
->recursion_count
) {
381 DMERR("cannot recurse set_count()");
386 r
= sm_ll_insert(&smm
->ll
, b
, count
, &ev
);
389 return combine_errors(r
, r2
);
392 static int sm_metadata_inc_block(struct dm_space_map
*sm
, dm_block_t b
)
395 enum allocation_event ev
;
396 struct sm_metadata
*smm
= container_of(sm
, struct sm_metadata
, sm
);
399 r
= add_bop(smm
, BOP_INC
, b
);
402 r
= sm_ll_inc(&smm
->ll
, b
, &ev
);
406 return combine_errors(r
, r2
);
409 static int sm_metadata_dec_block(struct dm_space_map
*sm
, dm_block_t b
)
412 enum allocation_event ev
;
413 struct sm_metadata
*smm
= container_of(sm
, struct sm_metadata
, sm
);
416 r
= add_bop(smm
, BOP_DEC
, b
);
419 r
= sm_ll_dec(&smm
->ll
, b
, &ev
);
423 return combine_errors(r
, r2
);
426 static int sm_metadata_new_block_(struct dm_space_map
*sm
, dm_block_t
*b
)
429 enum allocation_event ev
;
430 struct sm_metadata
*smm
= container_of(sm
, struct sm_metadata
, sm
);
432 r
= sm_ll_find_free_block(&smm
->old_ll
, smm
->begin
, smm
->old_ll
.nr_blocks
, b
);
439 r
= add_bop(smm
, BOP_INC
, *b
);
442 r
= sm_ll_inc(&smm
->ll
, *b
, &ev
);
447 smm
->allocated_this_transaction
++;
449 return combine_errors(r
, r2
);
452 static int sm_metadata_new_block(struct dm_space_map
*sm
, dm_block_t
*b
)
455 struct sm_metadata
*smm
= container_of(sm
, struct sm_metadata
, sm
);
457 int r
= sm_metadata_new_block_(sm
, b
);
459 DMERR("unable to allocate new metadata block");
463 r
= sm_metadata_get_nr_free(sm
, &count
);
465 DMERR("couldn't get free block count");
469 check_threshold(&smm
->threshold
, count
);
474 static int sm_metadata_commit(struct dm_space_map
*sm
)
477 struct sm_metadata
*smm
= container_of(sm
, struct sm_metadata
, sm
);
479 r
= sm_ll_commit(&smm
->ll
);
483 memcpy(&smm
->old_ll
, &smm
->ll
, sizeof(smm
->old_ll
));
485 smm
->allocated_this_transaction
= 0;
490 static int sm_metadata_register_threshold_callback(struct dm_space_map
*sm
,
491 dm_block_t threshold
,
492 dm_sm_threshold_fn fn
,
495 struct sm_metadata
*smm
= container_of(sm
, struct sm_metadata
, sm
);
497 set_threshold(&smm
->threshold
, threshold
, fn
, context
);
502 static int sm_metadata_root_size(struct dm_space_map
*sm
, size_t *result
)
504 *result
= sizeof(struct disk_sm_root
);
509 static int sm_metadata_copy_root(struct dm_space_map
*sm
, void *where_le
, size_t max
)
511 struct sm_metadata
*smm
= container_of(sm
, struct sm_metadata
, sm
);
512 struct disk_sm_root root_le
;
514 root_le
.nr_blocks
= cpu_to_le64(smm
->ll
.nr_blocks
);
515 root_le
.nr_allocated
= cpu_to_le64(smm
->ll
.nr_allocated
);
516 root_le
.bitmap_root
= cpu_to_le64(smm
->ll
.bitmap_root
);
517 root_le
.ref_count_root
= cpu_to_le64(smm
->ll
.ref_count_root
);
519 if (max
< sizeof(root_le
))
522 memcpy(where_le
, &root_le
, sizeof(root_le
));
527 static int sm_metadata_extend(struct dm_space_map
*sm
, dm_block_t extra_blocks
);
529 static struct dm_space_map ops
= {
530 .destroy
= sm_metadata_destroy
,
531 .extend
= sm_metadata_extend
,
532 .get_nr_blocks
= sm_metadata_get_nr_blocks
,
533 .get_nr_free
= sm_metadata_get_nr_free
,
534 .get_count
= sm_metadata_get_count
,
535 .count_is_more_than_one
= sm_metadata_count_is_more_than_one
,
536 .set_count
= sm_metadata_set_count
,
537 .inc_block
= sm_metadata_inc_block
,
538 .dec_block
= sm_metadata_dec_block
,
539 .new_block
= sm_metadata_new_block
,
540 .commit
= sm_metadata_commit
,
541 .root_size
= sm_metadata_root_size
,
542 .copy_root
= sm_metadata_copy_root
,
543 .register_threshold_callback
= sm_metadata_register_threshold_callback
546 /*----------------------------------------------------------------*/
549 * When a new space map is created that manages its own space. We use
550 * this tiny bootstrap allocator.
552 static void sm_bootstrap_destroy(struct dm_space_map
*sm
)
556 static int sm_bootstrap_extend(struct dm_space_map
*sm
, dm_block_t extra_blocks
)
558 DMERR("bootstrap doesn't support extend");
563 static int sm_bootstrap_get_nr_blocks(struct dm_space_map
*sm
, dm_block_t
*count
)
565 struct sm_metadata
*smm
= container_of(sm
, struct sm_metadata
, sm
);
567 *count
= smm
->ll
.nr_blocks
;
572 static int sm_bootstrap_get_nr_free(struct dm_space_map
*sm
, dm_block_t
*count
)
574 struct sm_metadata
*smm
= container_of(sm
, struct sm_metadata
, sm
);
576 *count
= smm
->ll
.nr_blocks
- smm
->begin
;
581 static int sm_bootstrap_get_count(struct dm_space_map
*sm
, dm_block_t b
,
584 struct sm_metadata
*smm
= container_of(sm
, struct sm_metadata
, sm
);
586 return b
< smm
->begin
? 1 : 0;
589 static int sm_bootstrap_count_is_more_than_one(struct dm_space_map
*sm
,
590 dm_block_t b
, int *result
)
597 static int sm_bootstrap_set_count(struct dm_space_map
*sm
, dm_block_t b
,
600 DMERR("bootstrap doesn't support set_count");
605 static int sm_bootstrap_new_block(struct dm_space_map
*sm
, dm_block_t
*b
)
607 struct sm_metadata
*smm
= container_of(sm
, struct sm_metadata
, sm
);
610 * We know the entire device is unused.
612 if (smm
->begin
== smm
->ll
.nr_blocks
)
620 static int sm_bootstrap_inc_block(struct dm_space_map
*sm
, dm_block_t b
)
622 struct sm_metadata
*smm
= container_of(sm
, struct sm_metadata
, sm
);
624 return add_bop(smm
, BOP_INC
, b
);
627 static int sm_bootstrap_dec_block(struct dm_space_map
*sm
, dm_block_t b
)
629 struct sm_metadata
*smm
= container_of(sm
, struct sm_metadata
, sm
);
631 return add_bop(smm
, BOP_DEC
, b
);
634 static int sm_bootstrap_commit(struct dm_space_map
*sm
)
639 static int sm_bootstrap_root_size(struct dm_space_map
*sm
, size_t *result
)
641 DMERR("bootstrap doesn't support root_size");
646 static int sm_bootstrap_copy_root(struct dm_space_map
*sm
, void *where
,
649 DMERR("bootstrap doesn't support copy_root");
654 static struct dm_space_map bootstrap_ops
= {
655 .destroy
= sm_bootstrap_destroy
,
656 .extend
= sm_bootstrap_extend
,
657 .get_nr_blocks
= sm_bootstrap_get_nr_blocks
,
658 .get_nr_free
= sm_bootstrap_get_nr_free
,
659 .get_count
= sm_bootstrap_get_count
,
660 .count_is_more_than_one
= sm_bootstrap_count_is_more_than_one
,
661 .set_count
= sm_bootstrap_set_count
,
662 .inc_block
= sm_bootstrap_inc_block
,
663 .dec_block
= sm_bootstrap_dec_block
,
664 .new_block
= sm_bootstrap_new_block
,
665 .commit
= sm_bootstrap_commit
,
666 .root_size
= sm_bootstrap_root_size
,
667 .copy_root
= sm_bootstrap_copy_root
,
668 .register_threshold_callback
= NULL
671 /*----------------------------------------------------------------*/
673 static int sm_metadata_extend(struct dm_space_map
*sm
, dm_block_t extra_blocks
)
676 enum allocation_event ev
;
677 struct sm_metadata
*smm
= container_of(sm
, struct sm_metadata
, sm
);
678 dm_block_t old_len
= smm
->ll
.nr_blocks
;
681 * Flick into a mode where all blocks get allocated in the new area.
683 smm
->begin
= old_len
;
684 memcpy(sm
, &bootstrap_ops
, sizeof(*sm
));
689 r
= sm_ll_extend(&smm
->ll
, extra_blocks
);
694 * We repeatedly increment then commit until the commit doesn't
695 * allocate any new blocks.
698 for (i
= old_len
; !r
&& i
< smm
->begin
; i
++) {
699 r
= sm_ll_inc(&smm
->ll
, i
, &ev
);
703 old_len
= smm
->begin
;
705 r
= sm_ll_commit(&smm
->ll
);
709 } while (old_len
!= smm
->begin
);
713 * Switch back to normal behaviour.
715 memcpy(sm
, &ops
, sizeof(*sm
));
719 /*----------------------------------------------------------------*/
721 struct dm_space_map
*dm_sm_metadata_init(void)
723 struct sm_metadata
*smm
;
725 smm
= kmalloc(sizeof(*smm
), GFP_KERNEL
);
727 return ERR_PTR(-ENOMEM
);
729 memcpy(&smm
->sm
, &ops
, sizeof(smm
->sm
));
734 int dm_sm_metadata_create(struct dm_space_map
*sm
,
735 struct dm_transaction_manager
*tm
,
736 dm_block_t nr_blocks
,
737 dm_block_t superblock
)
741 enum allocation_event ev
;
742 struct sm_metadata
*smm
= container_of(sm
, struct sm_metadata
, sm
);
744 smm
->begin
= superblock
+ 1;
745 smm
->recursion_count
= 0;
746 smm
->allocated_this_transaction
= 0;
747 brb_init(&smm
->uncommitted
);
748 threshold_init(&smm
->threshold
);
750 memcpy(&smm
->sm
, &bootstrap_ops
, sizeof(smm
->sm
));
752 r
= sm_ll_new_metadata(&smm
->ll
, tm
);
756 r
= sm_ll_extend(&smm
->ll
, nr_blocks
);
760 memcpy(&smm
->sm
, &ops
, sizeof(smm
->sm
));
763 * Now we need to update the newly created data structures with the
764 * allocated blocks that they were built from.
766 for (i
= superblock
; !r
&& i
< smm
->begin
; i
++)
767 r
= sm_ll_inc(&smm
->ll
, i
, &ev
);
772 return sm_metadata_commit(sm
);
775 int dm_sm_metadata_open(struct dm_space_map
*sm
,
776 struct dm_transaction_manager
*tm
,
777 void *root_le
, size_t len
)
780 struct sm_metadata
*smm
= container_of(sm
, struct sm_metadata
, sm
);
782 r
= sm_ll_open_metadata(&smm
->ll
, tm
, root_le
, len
);
787 smm
->recursion_count
= 0;
788 smm
->allocated_this_transaction
= 0;
789 brb_init(&smm
->uncommitted
);
790 threshold_init(&smm
->threshold
);
792 memcpy(&smm
->old_ll
, &smm
->ll
, sizeof(smm
->old_ll
));