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_peek(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
;
153 static int brb_pop(struct bop_ring_buffer
*brb
)
155 struct block_op
*bop
;
160 bop
= brb
->bops
+ brb
->begin
;
161 brb
->begin
= brb_next(brb
, brb
->begin
);
166 /*----------------------------------------------------------------*/
169 struct dm_space_map sm
;
172 struct ll_disk old_ll
;
176 unsigned recursion_count
;
177 unsigned allocated_this_transaction
;
178 struct bop_ring_buffer uncommitted
;
180 struct threshold threshold
;
183 static int add_bop(struct sm_metadata
*smm
, enum block_op_type type
, dm_block_t b
)
185 int r
= brb_push(&smm
->uncommitted
, type
, b
);
188 DMERR("too many recursive allocations");
195 static int commit_bop(struct sm_metadata
*smm
, struct block_op
*op
)
198 enum allocation_event ev
;
202 r
= sm_ll_inc(&smm
->ll
, op
->block
, &ev
);
206 r
= sm_ll_dec(&smm
->ll
, op
->block
, &ev
);
213 static void in(struct sm_metadata
*smm
)
215 smm
->recursion_count
++;
218 static int apply_bops(struct sm_metadata
*smm
)
222 while (!brb_empty(&smm
->uncommitted
)) {
225 r
= brb_peek(&smm
->uncommitted
, &bop
);
227 DMERR("bug in bop ring buffer");
231 r
= commit_bop(smm
, &bop
);
235 brb_pop(&smm
->uncommitted
);
241 static int out(struct sm_metadata
*smm
)
246 * If we're not recursing then very bad things are happening.
248 if (!smm
->recursion_count
) {
249 DMERR("lost track of recursion depth");
253 if (smm
->recursion_count
== 1)
256 smm
->recursion_count
--;
262 * When using the out() function above, we often want to combine an error
263 * code for the operation run in the recursive context with that from
266 static int combine_errors(int r1
, int r2
)
271 static int recursing(struct sm_metadata
*smm
)
273 return smm
->recursion_count
;
276 static void sm_metadata_destroy(struct dm_space_map
*sm
)
278 struct sm_metadata
*smm
= container_of(sm
, struct sm_metadata
, sm
);
283 static int sm_metadata_get_nr_blocks(struct dm_space_map
*sm
, dm_block_t
*count
)
285 struct sm_metadata
*smm
= container_of(sm
, struct sm_metadata
, sm
);
287 *count
= smm
->ll
.nr_blocks
;
292 static int sm_metadata_get_nr_free(struct dm_space_map
*sm
, dm_block_t
*count
)
294 struct sm_metadata
*smm
= container_of(sm
, struct sm_metadata
, sm
);
296 *count
= smm
->old_ll
.nr_blocks
- smm
->old_ll
.nr_allocated
-
297 smm
->allocated_this_transaction
;
302 static int sm_metadata_get_count(struct dm_space_map
*sm
, dm_block_t b
,
307 struct sm_metadata
*smm
= container_of(sm
, struct sm_metadata
, sm
);
308 unsigned adjustment
= 0;
311 * We may have some uncommitted adjustments to add. This list
312 * should always be really short.
314 for (i
= smm
->uncommitted
.begin
;
315 i
!= smm
->uncommitted
.end
;
316 i
= brb_next(&smm
->uncommitted
, i
)) {
317 struct block_op
*op
= smm
->uncommitted
.bops
+ i
;
333 r
= sm_ll_lookup(&smm
->ll
, b
, result
);
337 *result
+= adjustment
;
342 static int sm_metadata_count_is_more_than_one(struct dm_space_map
*sm
,
343 dm_block_t b
, int *result
)
345 int r
, adjustment
= 0;
347 struct sm_metadata
*smm
= container_of(sm
, struct sm_metadata
, sm
);
351 * We may have some uncommitted adjustments to add. This list
352 * should always be really short.
354 for (i
= smm
->uncommitted
.begin
;
355 i
!= smm
->uncommitted
.end
;
356 i
= brb_next(&smm
->uncommitted
, i
)) {
358 struct block_op
*op
= smm
->uncommitted
.bops
+ i
;
374 if (adjustment
> 1) {
379 r
= sm_ll_lookup_bitmap(&smm
->ll
, b
, &rc
);
385 * We err on the side of caution, and always return true.
389 *result
= rc
+ adjustment
> 1;
394 static int sm_metadata_set_count(struct dm_space_map
*sm
, dm_block_t b
,
398 enum allocation_event ev
;
399 struct sm_metadata
*smm
= container_of(sm
, struct sm_metadata
, sm
);
401 if (smm
->recursion_count
) {
402 DMERR("cannot recurse set_count()");
407 r
= sm_ll_insert(&smm
->ll
, b
, count
, &ev
);
410 return combine_errors(r
, r2
);
413 static int sm_metadata_inc_block(struct dm_space_map
*sm
, dm_block_t b
)
416 enum allocation_event ev
;
417 struct sm_metadata
*smm
= container_of(sm
, struct sm_metadata
, sm
);
420 r
= add_bop(smm
, BOP_INC
, b
);
423 r
= sm_ll_inc(&smm
->ll
, b
, &ev
);
427 return combine_errors(r
, r2
);
430 static int sm_metadata_dec_block(struct dm_space_map
*sm
, dm_block_t b
)
433 enum allocation_event ev
;
434 struct sm_metadata
*smm
= container_of(sm
, struct sm_metadata
, sm
);
437 r
= add_bop(smm
, BOP_DEC
, b
);
440 r
= sm_ll_dec(&smm
->ll
, b
, &ev
);
444 return combine_errors(r
, r2
);
447 static int sm_metadata_new_block_(struct dm_space_map
*sm
, dm_block_t
*b
)
450 enum allocation_event ev
;
451 struct sm_metadata
*smm
= container_of(sm
, struct sm_metadata
, sm
);
453 r
= sm_ll_find_free_block(&smm
->old_ll
, smm
->begin
, smm
->old_ll
.nr_blocks
, b
);
460 r
= add_bop(smm
, BOP_INC
, *b
);
463 r
= sm_ll_inc(&smm
->ll
, *b
, &ev
);
468 smm
->allocated_this_transaction
++;
470 return combine_errors(r
, r2
);
473 static int sm_metadata_new_block(struct dm_space_map
*sm
, dm_block_t
*b
)
476 struct sm_metadata
*smm
= container_of(sm
, struct sm_metadata
, sm
);
478 int r
= sm_metadata_new_block_(sm
, b
);
480 DMERR_LIMIT("unable to allocate new metadata block");
484 r
= sm_metadata_get_nr_free(sm
, &count
);
486 DMERR_LIMIT("couldn't get free block count");
490 check_threshold(&smm
->threshold
, count
);
495 static int sm_metadata_commit(struct dm_space_map
*sm
)
498 struct sm_metadata
*smm
= container_of(sm
, struct sm_metadata
, sm
);
500 r
= sm_ll_commit(&smm
->ll
);
504 memcpy(&smm
->old_ll
, &smm
->ll
, sizeof(smm
->old_ll
));
506 smm
->allocated_this_transaction
= 0;
511 static int sm_metadata_register_threshold_callback(struct dm_space_map
*sm
,
512 dm_block_t threshold
,
513 dm_sm_threshold_fn fn
,
516 struct sm_metadata
*smm
= container_of(sm
, struct sm_metadata
, sm
);
518 set_threshold(&smm
->threshold
, threshold
, fn
, context
);
523 static int sm_metadata_root_size(struct dm_space_map
*sm
, size_t *result
)
525 *result
= sizeof(struct disk_sm_root
);
530 static int sm_metadata_copy_root(struct dm_space_map
*sm
, void *where_le
, size_t max
)
532 struct sm_metadata
*smm
= container_of(sm
, struct sm_metadata
, sm
);
533 struct disk_sm_root root_le
;
535 root_le
.nr_blocks
= cpu_to_le64(smm
->ll
.nr_blocks
);
536 root_le
.nr_allocated
= cpu_to_le64(smm
->ll
.nr_allocated
);
537 root_le
.bitmap_root
= cpu_to_le64(smm
->ll
.bitmap_root
);
538 root_le
.ref_count_root
= cpu_to_le64(smm
->ll
.ref_count_root
);
540 if (max
< sizeof(root_le
))
543 memcpy(where_le
, &root_le
, sizeof(root_le
));
548 static int sm_metadata_extend(struct dm_space_map
*sm
, dm_block_t extra_blocks
);
550 static struct dm_space_map ops
= {
551 .destroy
= sm_metadata_destroy
,
552 .extend
= sm_metadata_extend
,
553 .get_nr_blocks
= sm_metadata_get_nr_blocks
,
554 .get_nr_free
= sm_metadata_get_nr_free
,
555 .get_count
= sm_metadata_get_count
,
556 .count_is_more_than_one
= sm_metadata_count_is_more_than_one
,
557 .set_count
= sm_metadata_set_count
,
558 .inc_block
= sm_metadata_inc_block
,
559 .dec_block
= sm_metadata_dec_block
,
560 .new_block
= sm_metadata_new_block
,
561 .commit
= sm_metadata_commit
,
562 .root_size
= sm_metadata_root_size
,
563 .copy_root
= sm_metadata_copy_root
,
564 .register_threshold_callback
= sm_metadata_register_threshold_callback
567 /*----------------------------------------------------------------*/
570 * When a new space map is created that manages its own space. We use
571 * this tiny bootstrap allocator.
573 static void sm_bootstrap_destroy(struct dm_space_map
*sm
)
577 static int sm_bootstrap_extend(struct dm_space_map
*sm
, dm_block_t extra_blocks
)
579 DMERR("bootstrap doesn't support extend");
584 static int sm_bootstrap_get_nr_blocks(struct dm_space_map
*sm
, dm_block_t
*count
)
586 struct sm_metadata
*smm
= container_of(sm
, struct sm_metadata
, sm
);
588 *count
= smm
->ll
.nr_blocks
;
593 static int sm_bootstrap_get_nr_free(struct dm_space_map
*sm
, dm_block_t
*count
)
595 struct sm_metadata
*smm
= container_of(sm
, struct sm_metadata
, sm
);
597 *count
= smm
->ll
.nr_blocks
- smm
->begin
;
602 static int sm_bootstrap_get_count(struct dm_space_map
*sm
, dm_block_t b
,
605 struct sm_metadata
*smm
= container_of(sm
, struct sm_metadata
, sm
);
607 *result
= (b
< smm
->begin
) ? 1 : 0;
612 static int sm_bootstrap_count_is_more_than_one(struct dm_space_map
*sm
,
613 dm_block_t b
, int *result
)
620 static int sm_bootstrap_set_count(struct dm_space_map
*sm
, dm_block_t b
,
623 DMERR("bootstrap doesn't support set_count");
628 static int sm_bootstrap_new_block(struct dm_space_map
*sm
, dm_block_t
*b
)
630 struct sm_metadata
*smm
= container_of(sm
, struct sm_metadata
, sm
);
633 * We know the entire device is unused.
635 if (smm
->begin
== smm
->ll
.nr_blocks
)
643 static int sm_bootstrap_inc_block(struct dm_space_map
*sm
, dm_block_t b
)
645 struct sm_metadata
*smm
= container_of(sm
, struct sm_metadata
, sm
);
647 return add_bop(smm
, BOP_INC
, b
);
650 static int sm_bootstrap_dec_block(struct dm_space_map
*sm
, dm_block_t b
)
652 struct sm_metadata
*smm
= container_of(sm
, struct sm_metadata
, sm
);
654 return add_bop(smm
, BOP_DEC
, b
);
657 static int sm_bootstrap_commit(struct dm_space_map
*sm
)
662 static int sm_bootstrap_root_size(struct dm_space_map
*sm
, size_t *result
)
664 DMERR("bootstrap doesn't support root_size");
669 static int sm_bootstrap_copy_root(struct dm_space_map
*sm
, void *where
,
672 DMERR("bootstrap doesn't support copy_root");
677 static struct dm_space_map bootstrap_ops
= {
678 .destroy
= sm_bootstrap_destroy
,
679 .extend
= sm_bootstrap_extend
,
680 .get_nr_blocks
= sm_bootstrap_get_nr_blocks
,
681 .get_nr_free
= sm_bootstrap_get_nr_free
,
682 .get_count
= sm_bootstrap_get_count
,
683 .count_is_more_than_one
= sm_bootstrap_count_is_more_than_one
,
684 .set_count
= sm_bootstrap_set_count
,
685 .inc_block
= sm_bootstrap_inc_block
,
686 .dec_block
= sm_bootstrap_dec_block
,
687 .new_block
= sm_bootstrap_new_block
,
688 .commit
= sm_bootstrap_commit
,
689 .root_size
= sm_bootstrap_root_size
,
690 .copy_root
= sm_bootstrap_copy_root
,
691 .register_threshold_callback
= NULL
694 /*----------------------------------------------------------------*/
696 static int sm_metadata_extend(struct dm_space_map
*sm
, dm_block_t extra_blocks
)
699 struct sm_metadata
*smm
= container_of(sm
, struct sm_metadata
, sm
);
700 dm_block_t old_len
= smm
->ll
.nr_blocks
;
703 * Flick into a mode where all blocks get allocated in the new area.
705 smm
->begin
= old_len
;
706 memcpy(sm
, &bootstrap_ops
, sizeof(*sm
));
711 r
= sm_ll_extend(&smm
->ll
, extra_blocks
);
716 * We repeatedly increment then commit until the commit doesn't
717 * allocate any new blocks.
720 for (i
= old_len
; !r
&& i
< smm
->begin
; i
++)
721 r
= add_bop(smm
, BOP_INC
, i
);
726 old_len
= smm
->begin
;
730 DMERR("%s: apply_bops failed", __func__
);
734 r
= sm_ll_commit(&smm
->ll
);
738 } while (old_len
!= smm
->begin
);
742 * Switch back to normal behaviour.
744 memcpy(sm
, &ops
, sizeof(*sm
));
748 /*----------------------------------------------------------------*/
750 struct dm_space_map
*dm_sm_metadata_init(void)
752 struct sm_metadata
*smm
;
754 smm
= kmalloc(sizeof(*smm
), GFP_KERNEL
);
756 return ERR_PTR(-ENOMEM
);
758 memcpy(&smm
->sm
, &ops
, sizeof(smm
->sm
));
763 int dm_sm_metadata_create(struct dm_space_map
*sm
,
764 struct dm_transaction_manager
*tm
,
765 dm_block_t nr_blocks
,
766 dm_block_t superblock
)
770 struct sm_metadata
*smm
= container_of(sm
, struct sm_metadata
, sm
);
772 smm
->begin
= superblock
+ 1;
773 smm
->recursion_count
= 0;
774 smm
->allocated_this_transaction
= 0;
775 brb_init(&smm
->uncommitted
);
776 threshold_init(&smm
->threshold
);
778 memcpy(&smm
->sm
, &bootstrap_ops
, sizeof(smm
->sm
));
780 r
= sm_ll_new_metadata(&smm
->ll
, tm
);
784 if (nr_blocks
> DM_SM_METADATA_MAX_BLOCKS
)
785 nr_blocks
= DM_SM_METADATA_MAX_BLOCKS
;
786 r
= sm_ll_extend(&smm
->ll
, nr_blocks
);
790 memcpy(&smm
->sm
, &ops
, sizeof(smm
->sm
));
793 * Now we need to update the newly created data structures with the
794 * allocated blocks that they were built from.
796 for (i
= superblock
; !r
&& i
< smm
->begin
; i
++)
797 r
= add_bop(smm
, BOP_INC
, i
);
804 DMERR("%s: apply_bops failed", __func__
);
808 return sm_metadata_commit(sm
);
811 int dm_sm_metadata_open(struct dm_space_map
*sm
,
812 struct dm_transaction_manager
*tm
,
813 void *root_le
, size_t len
)
816 struct sm_metadata
*smm
= container_of(sm
, struct sm_metadata
, sm
);
818 r
= sm_ll_open_metadata(&smm
->ll
, tm
, root_le
, len
);
823 smm
->recursion_count
= 0;
824 smm
->allocated_this_transaction
= 0;
825 brb_init(&smm
->uncommitted
);
826 threshold_init(&smm
->threshold
);
828 memcpy(&smm
->old_ll
, &smm
->ll
, sizeof(smm
->old_ll
));