#define SEARCH_READ 0
#define SEARCH_WRITE 1
-static int refill_alloc_extent(struct ctree_root *root);
+#define CTREE_EXTENT_PENDING 0
+
int split_node(struct ctree_root *root, struct ctree_path *path, int level);
int split_leaf(struct ctree_root *root, struct ctree_path *path, int data_size);
+struct tree_buffer *alloc_free_block(struct ctree_root *root);
+int free_extent(struct ctree_root *root, u64 blocknr, u64 num_blocks);
static inline void init_path(struct ctree_path *p)
{
unsigned int data_end;
struct ctree_path path;
- refill_alloc_extent(root);
-
/* create a root if there isn't one */
if (!root->node)
BUG();
struct tree_buffer *t;
struct node *node;
int nritems;
+ u64 blocknr;
while(1) {
t = path->nodes[level];
}
node->header.nritems--;
write_tree_block(root, t);
+ blocknr = t->blocknr;
if (node->header.nritems != 0) {
int tslot;
if (slot == 0)
break;
}
level++;
+ free_extent(root, blocknr, 1);
if (!path->nodes[level])
BUG();
}
if (leaf_buf == root->node) {
leaf->header.flags = node_level(0);
write_tree_block(root, leaf_buf);
- } else
+ } else {
del_ptr(root, path, 1);
+ free_extent(root, leaf_buf->blocknr, 1);
+ }
} else {
if (slot == 0)
fixup_low_keys(root, path, &leaf->items[0].key, 1);
return 0;
}
+static int del_pending_extents(struct ctree_root *extent_root)
+{
+ int ret;
+ struct key key;
+ struct tree_buffer *gang[4];
+ int i;
+ struct ctree_path path;
+
+ while(1) {
+ ret = radix_tree_gang_lookup_tag(&extent_root->cache_radix,
+ (void **)gang, 0, ARRAY_SIZE(gang),
+ CTREE_EXTENT_PENDING);
+ if (!ret)
+ break;
+ for (i = 0; i < ret; i++) {
+ key.objectid = gang[i]->blocknr;
+ key.flags = 0;
+ key.offset = 1;
+ init_path(&path);
+ ret = search_slot(extent_root, &key, &path, 0);
+ if (ret) {
+ BUG();
+ // FIXME undo it and return sane
+ return ret;
+ }
+ ret = del_item(extent_root, &path);
+ if (ret) {
+ BUG();
+ return ret;
+ }
+ release_path(extent_root, &path);
+ radix_tree_tag_clear(&extent_root->cache_radix, gang[i]->blocknr,
+ CTREE_EXTENT_PENDING);
+ tree_block_release(extent_root, gang[i]);
+ }
+ }
+ return 0;
+}
+
+int free_extent(struct ctree_root *root, u64 blocknr, u64 num_blocks)
+{
+ struct ctree_path path;
+ struct key key;
+ struct ctree_root *extent_root = root->extent_root;
+ struct tree_buffer *t;
+ int pending_ret;
+ int ret;
+
+ key.objectid = blocknr;
+ key.flags = 0;
+ key.offset = num_blocks;
+ if (root == extent_root) {
+ t = read_tree_block(root, key.objectid);
+ radix_tree_tag_set(&root->cache_radix, key.objectid, CTREE_EXTENT_PENDING);
+ return 0;
+ }
+ init_path(&path);
+ ret = search_slot(extent_root, &key, &path, 0);
+ if (ret)
+ BUG();
+ ret = del_item(extent_root, &path);
+ release_path(extent_root, &path);
+ pending_ret = del_pending_extents(root->extent_root);
+ return ret ? ret : pending_ret;
+}
+
int next_leaf(struct ctree_root *root, struct ctree_path *path)
{
int slot;
return 0;
}
-int alloc_extent(struct ctree_root *orig_root, u64 num_blocks, u64 search_start,
- u64 search_end, u64 owner, struct key *ins)
+int find_free_extent(struct ctree_root *orig_root, u64 num_blocks, u64 search_start,
+ u64 search_end, struct key *ins)
{
struct ctree_path path;
struct key *key;
u64 last_block;
int start_found = 0;
struct leaf *l;
- struct extent_item extent_item;
struct ctree_root * root = orig_root->extent_root;
init_path(&path);
ins->objectid = search_start;
ins->offset = 0;
ins->flags = 0;
-
- ret = search_slot(root, ins, &path, sizeof(struct extent_item));
+ ret = search_slot(root, ins, &path, 0);
while (1) {
l = &path.nodes[0]->leaf;
slot = path.slots[0];
ins->objectid = search_start;
ins->offset = num_blocks;
hole_size = search_end - search_start;
+ start_found = 1;
goto insert;
}
ins->objectid = last_block;
} else
start_found = 1;
last_block = key->objectid + key->offset;
+insert_failed:
path.slots[0]++;
}
// FIXME -ENOSPC
insert:
+ if (orig_root->extent_root == orig_root) {
+ BUG_ON(num_blocks != 1);
+ if ((root->current_insert.objectid <= ins->objectid &&
+ root->current_insert.objectid + root->current_insert.offset >
+ ins->objectid) ||
+ (root->current_insert.objectid > ins->objectid &&
+ root->current_insert.objectid <= ins->objectid + ins->offset) ||
+ radix_tree_tag_get(&root->cache_radix, ins->objectid,
+ CTREE_EXTENT_PENDING)) {
+ last_block = ins->objectid + 1;
+ search_start = last_block;
+ goto insert_failed;
+ }
+ }
release_path(root, &path);
+ if (ins->offset != 1)
+ BUG();
+ return 0;
+}
+
+static int insert_pending_extents(struct ctree_root *extent_root)
+{
+ int ret;
+ struct key key;
+ struct extent_item item;
+ struct tree_buffer *gang[4];
+ int i;
+
+ // FIXME -ENOSPC
+ item.refs = 1;
+ item.owner = extent_root->node->node.header.parentid;
+ while(1) {
+ ret = radix_tree_gang_lookup_tag(&extent_root->cache_radix,
+ (void **)gang, 0, ARRAY_SIZE(gang),
+ CTREE_EXTENT_PENDING);
+ if (!ret)
+ break;
+ for (i = 0; i < ret; i++) {
+ key.objectid = gang[i]->blocknr;
+ key.flags = 0;
+ key.offset = 1;
+ ret = insert_item(extent_root, &key, &item, sizeof(item));
+ if (ret) {
+ BUG();
+ // FIXME undo it and return sane
+ return ret;
+ }
+ radix_tree_tag_clear(&extent_root->cache_radix, gang[i]->blocknr,
+ CTREE_EXTENT_PENDING);
+ tree_block_release(extent_root, gang[i]);
+ }
+ }
+ return 0;
+}
+
+int alloc_extent(struct ctree_root *root, u64 num_blocks, u64 search_start,
+ u64 search_end, u64 owner, struct key *ins, struct tree_buffer **buf)
+{
+ int ret;
+ int pending_ret;
+ struct extent_item extent_item;
+
extent_item.refs = 1;
extent_item.owner = owner;
- if (root == orig_root && root->reserve_extent->num_blocks == 0) {
- root->reserve_extent->blocknr = ins->objectid;
- root->reserve_extent->num_blocks = ins->offset;
- root->reserve_extent->num_used = 0;
+
+ ret = find_free_extent(root, num_blocks, search_start, search_end, ins);
+ if (ret)
+ return ret;
+
+ if (root != root->extent_root) {
+ memcpy(&root->extent_root->current_insert, ins, sizeof(*ins));
+ ret = insert_item(root->extent_root, ins, &extent_item, sizeof(extent_item));
+ memset(&root->extent_root->current_insert, 0, sizeof(struct key));
+ pending_ret = insert_pending_extents(root->extent_root);
+ if (ret)
+ return ret;
+ if (pending_ret)
+ return pending_ret;
+ *buf = find_tree_block(root, ins->objectid);
+ return 0;
}
- ret = insert_item(root->extent_root, ins, &extent_item, sizeof(extent_item));
- return ret;
+ /* we're allocating an extent for the extent tree, don't recurse */
+ BUG_ON(ins->offset != 1);
+ *buf = find_tree_block(root, ins->objectid);
+ BUG_ON(!*buf);
+ radix_tree_tag_set(&root->cache_radix, ins->objectid, CTREE_EXTENT_PENDING);
+ (*buf)->count++;
+ return 0;
+
}
-static int refill_alloc_extent(struct ctree_root *root)
+struct tree_buffer *alloc_free_block(struct ctree_root *root)
{
- struct alloc_extent *ae = root->alloc_extent;
- struct key key;
+ struct key ins;
int ret;
- int min_blocks = MAX_LEVEL * 2;
+ struct tree_buffer *buf = NULL;
- if (ae->num_blocks > ae->num_used && ae->num_blocks - ae->num_used >
- min_blocks)
- return 0;
- ae = root->reserve_extent;
- if (ae->num_blocks > ae->num_used) {
- if (root->alloc_extent->num_blocks == 0) {
- /* we should swap reserve/alloc_extent when alloc
- * fills up
- */
- BUG();
- }
- if (ae->num_blocks - ae->num_used < min_blocks)
- BUG();
- return 0;
+ ret = alloc_extent(root, 1, 0, (unsigned long)-1, root->node->node.header.parentid,
+ &ins, &buf);
+
+ if (ret) {
+ BUG();
+ return NULL;
}
- ret = alloc_extent(root,
- min_blocks * 2, 0, (unsigned long)-1,
- root->node->node.header.parentid, &key);
- ae->blocknr = key.objectid;
- ae->num_blocks = key.offset;
- ae->num_used = 0;
- return ret;
+ if (root != root->extent_root)
+ BUG_ON(radix_tree_tag_get(&root->extent_root->cache_radix, buf->blocknr,
+ CTREE_EXTENT_PENDING));
+ return buf;
}
void print_leaf(struct leaf *l)
print_tree(root, root->node);
printf("map tree\n");
print_tree(root->extent_root, root->extent_root->node);
+ fflush(stdout);
srand(55);
for (i = 0; i < run_size; i++) {
if (!ret)
tree_size++;
}
- printf("root used: %lu\n", root->alloc_extent->num_used);
- printf("root tree\n");
- // print_tree(root, root->node);
- printf("map tree\n");
- printf("map used: %lu\n", root->extent_root->alloc_extent->num_used);
- // print_tree(root->extent_root, root->extent_root->node);
write_ctree_super(root, &super);
close_ctree(root);
ret = insert_item(root, &ins, buf, strlen(buf));
if (!ret)
tree_size++;
+ if (i >= 5) {
+ struct key ugh;
+ ugh.objectid = 5;
+ ugh.flags = 0;
+ ugh.offset = 0;
+ init_path(&path);
+ ret = search_slot(root, &ugh, &path, 0);
+ if (ret) {
+ print_tree(root, root->node);
+ printf("unable to find 5 %d\n", num);
+ exit(1);
+ }
+ release_path(root, &path);
+
+ }
}
write_ctree_super(root, &super);
close_ctree(root);
root = open_ctree("dbfile", &super);
- printf("starting search2\n");
srand(128);
+ printf("starting search2\n");
for (i = 0; i < run_size; i++) {
num = next_key(i, max_key);
ins.objectid = num;
write_ctree_super(root, &super);
close_ctree(root);
printf("tree size is now %d\n", tree_size);
+ printf("map tree\n");
+ print_tree(root->extent_root, root->extent_root->node);
return 0;
}
static int allocated_blocks = 0;
-static int get_free_block(struct ctree_root *root, u64 *block)
+static int check_tree_block(struct ctree_root *root, struct tree_buffer *buf)
{
- struct stat st;
- int ret = 0;
-
- if (root->alloc_extent->num_used >= root->alloc_extent->num_blocks)
- return -1;
-
- *block = root->alloc_extent->blocknr + root->alloc_extent->num_used;
- root->alloc_extent->num_used += 1;
- if (root->alloc_extent->num_used >= root->alloc_extent->num_blocks) {
- struct alloc_extent *ae = root->alloc_extent;
- root->alloc_extent = root->reserve_extent;
- root->reserve_extent = ae;
- ae->num_blocks = 0;
- }
- st.st_size = 0;
- ret = fstat(root->fp, &st);
- if (st.st_size < (*block + 1) * CTREE_BLOCKSIZE) {
- ret = ftruncate(root->fp,
- (*block + 1) * CTREE_BLOCKSIZE);
- if (ret) {
- perror("ftruncate");
- exit(1);
- }
- }
- return ret;
+ if (buf->blocknr != buf->node.header.blocknr)
+ BUG();
+ if (root->node && buf->node.header.parentid != root->node->node.header.parentid)
+ BUG();
+ return 0;
}
struct tree_buffer *alloc_tree_block(struct ctree_root *root, u64 blocknr)
return buf;
}
-struct tree_buffer *alloc_free_block(struct ctree_root *root)
+struct tree_buffer *find_tree_block(struct ctree_root *root, u64 blocknr)
{
- u64 free_block;
- int ret;
- struct tree_buffer * buf;
- ret = get_free_block(root, &free_block);
- if (ret) {
- BUG();
- return NULL;
+ struct tree_buffer *buf;
+ buf = radix_tree_lookup(&root->cache_radix, blocknr);
+ if (buf) {
+ buf->count++;
+ } else {
+ buf = alloc_tree_block(root, blocknr);
+ if (!buf) {
+ BUG();
+ return NULL;
+ }
}
- buf = alloc_tree_block(root, free_block);
- if (!buf)
- BUG();
return buf;
}
+
struct tree_buffer *read_tree_block(struct ctree_root *root, u64 blocknr)
{
loff_t offset = blocknr * CTREE_BLOCKSIZE;
buf = radix_tree_lookup(&root->cache_radix, blocknr);
if (buf) {
buf->count++;
- goto test;
- }
- buf = alloc_tree_block(root, blocknr);
- if (!buf)
- return NULL;
- ret = pread(root->fp, &buf->node, CTREE_BLOCKSIZE, offset);
- if (ret != CTREE_BLOCKSIZE) {
- free(buf);
- return NULL;
+ } else {
+ buf = alloc_tree_block(root, blocknr);
+ if (!buf)
+ return NULL;
+ ret = pread(root->fp, &buf->node, CTREE_BLOCKSIZE, offset);
+ if (ret != CTREE_BLOCKSIZE) {
+ free(buf);
+ return NULL;
+ }
}
-test:
- if (buf->blocknr != buf->node.header.blocknr)
- BUG();
- if (root->node && buf->node.header.parentid != root->node->node.header.parentid)
+ if (check_tree_block(root, buf))
BUG();
return buf;
}
static int __setup_root(struct ctree_root *root, struct ctree_root *extent_root,
struct ctree_root_info *info, int fp)
{
- INIT_RADIX_TREE(&root->cache_radix, GFP_KERNEL);
root->fp = fp;
root->node = NULL;
root->node = read_tree_block(root, info->tree_root);
root->extent_root = extent_root;
- memcpy(&root->ai1, &info->alloc_extent, sizeof(info->alloc_extent));
- memcpy(&root->ai2, &info->reserve_extent, sizeof(info->reserve_extent));
- root->alloc_extent = &root->ai1;
- root->reserve_extent = &root->ai2;
- printf("setup done reading root %p, used %lu available %lu\n", root, root->alloc_extent->num_used, root->alloc_extent->num_blocks);
- printf("setup done reading root %p, reserve used %lu available %lu\n", root, root->reserve_extent->num_used, root->reserve_extent->num_blocks);
return 0;
}
free(root);
return NULL;
}
+ INIT_RADIX_TREE(&root->cache_radix, GFP_KERNEL);
+ INIT_RADIX_TREE(&extent_root->cache_radix, GFP_KERNEL);
ret = pread(fp, super, sizeof(struct ctree_super_block),
CTREE_SUPER_INFO_OFFSET(CTREE_BLOCKSIZE));
if (ret == 0 || super->root_info.tree_root == 0) {
static int __update_root(struct ctree_root *root, struct ctree_root_info *info)
{
info->tree_root = root->node->blocknr;
- memcpy(&info->alloc_extent, root->alloc_extent, sizeof(struct alloc_extent));
- memcpy(&info->reserve_extent, root->reserve_extent, sizeof(struct alloc_extent));
return 0;
}
void tree_block_release(struct ctree_root *root, struct tree_buffer *buf)
{
buf->count--;
+ write_tree_block(root, buf);
if (buf->count < 0)
BUG();
if (buf->count == 0) {
info[0].blocknr = 16;
info[0].objectid = 1;
info[0].tree_root = 17;
- info[0].alloc_extent.blocknr = 0;
- info[0].alloc_extent.num_blocks = 64;
- /* 0-17 are used (inclusive) */
- info[0].alloc_extent.num_used = 18;
info[1].blocknr = 16;
info[1].objectid = 2;
- info[1].tree_root = 64;
- info[1].alloc_extent.blocknr = 64;
- info[1].alloc_extent.num_blocks = 64;
- info[1].alloc_extent.num_used = 1;
+ info[1].tree_root = 18;
ret = pwrite(fd, info, sizeof(info),
CTREE_SUPER_INFO_OFFSET(CTREE_BLOCKSIZE));
if (ret != sizeof(info))
return -1;
empty_leaf.header.parentid = 2;
- empty_leaf.header.blocknr = 64;
- empty_leaf.header.nritems = 2;
+ empty_leaf.header.blocknr = 18;
+ empty_leaf.header.nritems = 3;
+
+ /* item1, reserve blocks 0-16 */
item.key.objectid = 0;
- item.key.offset = 64;
+ item.key.offset = 17;
item.key.flags = 0;
item.offset = LEAF_DATA_SIZE - sizeof(struct extent_item);
item.size = sizeof(struct extent_item);
extent_item.refs = 1;
- extent_item.owner = 1;
+ extent_item.owner = 0;
memcpy(empty_leaf.items, &item, sizeof(item));
memcpy(empty_leaf.data + item.offset, &extent_item, item.size);
- item.key.objectid = 64;
- item.key.offset = 64;
+
+ /* item2, give block 17 to the root */
+ item.key.objectid = 17;
+ item.key.offset = 1;
item.offset = LEAF_DATA_SIZE - sizeof(struct extent_item) * 2;
- extent_item.owner = 2;
+ extent_item.owner = 1;
memcpy(empty_leaf.items + 1, &item, sizeof(item));
memcpy(empty_leaf.data + item.offset, &extent_item, item.size);
- ret = pwrite(fd, &empty_leaf, sizeof(empty_leaf), 64 * CTREE_BLOCKSIZE);
+
+ /* item3, give block 18 for the extent root */
+ item.key.objectid = 18;
+ item.key.offset = 1;
+ item.offset = LEAF_DATA_SIZE - sizeof(struct extent_item) * 3;
+ extent_item.owner = 2;
+ memcpy(empty_leaf.items + 2, &item, sizeof(item));
+ memcpy(empty_leaf.data + item.offset, &extent_item, item.size);
+ ret = pwrite(fd, &empty_leaf, sizeof(empty_leaf), 18 * CTREE_BLOCKSIZE);
if (ret != sizeof(empty_leaf))
return -1;
return 0;
git://ftp.safe.ca / safe / jmp / linux-2.6 / commitdiff