Btrfs: variable block size support

Signed-off-by: Chris Mason <chris.mason@oracle.com>

fs/btrfs/ctree.h

@@ -5,7 +5,6 @@
 #include "kerncompat.h"
 
 #define BTRFS_MAGIC "_BtRfS_M"
-#define BTRFS_BLOCKSIZE 1024
 
 #define BTRFS_ROOT_TREE_OBJECTID 1
 #define BTRFS_EXTENT_TREE_OBJECTID 2
@@ -52,8 +51,11 @@ struct btrfs_header {
 } __attribute__ ((__packed__));
 
 #define BTRFS_MAX_LEVEL 8
-#define NODEPTRS_PER_BLOCK ((BTRFS_BLOCKSIZE - sizeof(struct btrfs_header)) / \
-			    (sizeof(struct btrfs_disk_key) + sizeof(u64)))
+#define BTRFS_NODEPTRS_PER_BLOCK(r) (((r)->blocksize - \
+			        sizeof(struct btrfs_header)) / \
+			       (sizeof(struct btrfs_disk_key) + sizeof(u64)))
+#define __BTRFS_LEAF_DATA_SIZE(bs) ((bs) - sizeof(struct btrfs_header))
+#define BTRFS_LEAF_DATA_SIZE(r) (__BTRFS_LEAF_DATA_SIZE(r->blocksize))
 
 struct btrfs_buffer;
 
@@ -86,6 +88,7 @@ struct btrfs_root {
 	int ref_cows;
 	struct btrfs_root_item root_item;
 	struct btrfs_key root_key;
+	u32 blocksize;
 };
 
 /*
@@ -97,7 +100,7 @@ struct btrfs_super_block {
 	__le64 blocknr; /* this block number */
 	__le32 csum;
 	__le64 magic;
-	__le16 blocksize;
+	__le32 blocksize;
 	__le64 generation;
 	__le64 root;
 	__le64 total_blocks;
@@ -111,7 +114,7 @@ struct btrfs_super_block {
  */
 struct btrfs_item {
 	struct btrfs_disk_key key;
-	__le16 offset;
+	__le32 offset;
 	__le16 size;
 } __attribute__ ((__packed__));
 
@@ -122,24 +125,23 @@ struct btrfs_item {
  * The data is separate from the items to get the keys closer together
  * during searches.
  */
-#define LEAF_DATA_SIZE (BTRFS_BLOCKSIZE - sizeof(struct btrfs_header))
 struct btrfs_leaf {
 	struct btrfs_header header;
-	union {
-		struct btrfs_item items[LEAF_DATA_SIZE/
-				        sizeof(struct btrfs_item)];
-		u8 data[BTRFS_BLOCKSIZE - sizeof(struct btrfs_header)];
-	};
+	struct btrfs_item items[];
 } __attribute__ ((__packed__));
 
 /*
  * all non-leaf blocks are nodes, they hold only keys and pointers to
  * other blocks
  */
+struct btrfs_key_ptr {
+	struct btrfs_disk_key key;
+	__le64 blockptr;
+} __attribute__ ((__packed__));
+
 struct btrfs_node {
 	struct btrfs_header header;
-	struct btrfs_disk_key keys[NODEPTRS_PER_BLOCK];
-	__le64 blockptrs[NODEPTRS_PER_BLOCK];
+	struct btrfs_key_ptr ptrs[];
 } __attribute__ ((__packed__));
 
 /*
@@ -186,28 +188,28 @@ static inline void btrfs_set_extent_refs(struct btrfs_extent_item *ei, u32 val)
 
 static inline u64 btrfs_node_blockptr(struct btrfs_node *n, int nr)
 {
-	return le64_to_cpu(n->blockptrs[nr]);
+	return le64_to_cpu(n->ptrs[nr].blockptr);
 }
 
 static inline void btrfs_set_node_blockptr(struct btrfs_node *n, int nr,
 					   u64 val)
 {
-	n->blockptrs[nr] = cpu_to_le64(val);
+	n->ptrs[nr].blockptr = cpu_to_le64(val);
 }
 
-static inline u16 btrfs_item_offset(struct btrfs_item *item)
+static inline u32 btrfs_item_offset(struct btrfs_item *item)
 {
-	return le16_to_cpu(item->offset);
+	return le32_to_cpu(item->offset);
 }
 
-static inline void btrfs_set_item_offset(struct btrfs_item *item, u16 val)
+static inline void btrfs_set_item_offset(struct btrfs_item *item, u32 val)
 {
-	item->offset = cpu_to_le16(val);
+	item->offset = cpu_to_le32(val);
 }
 
-static inline u16 btrfs_item_end(struct btrfs_item *item)
+static inline u32 btrfs_item_end(struct btrfs_item *item)
 {
-	return le16_to_cpu(item->offset) + le16_to_cpu(item->size);
+	return le32_to_cpu(item->offset) + le16_to_cpu(item->size);
 }
 
 static inline u16 btrfs_item_size(struct btrfs_item *item)
@@ -390,20 +392,26 @@ static inline void btrfs_set_super_blocks_used(struct btrfs_super_block *s,
 	s->blocks_used = cpu_to_le64(val);
 }
 
-static inline u16 btrfs_super_blocksize(struct btrfs_super_block *s)
+static inline u32 btrfs_super_blocksize(struct btrfs_super_block *s)
 {
-	return le16_to_cpu(s->blocksize);
+	return le32_to_cpu(s->blocksize);
 }
 
 static inline void btrfs_set_super_blocksize(struct btrfs_super_block *s,
-						u16 val)
+						u32 val)
+{
+	s->blocksize = cpu_to_le32(val);
+}
+
+static inline u8 *btrfs_leaf_data(struct btrfs_leaf *l)
 {
-	s->blocksize = cpu_to_le16(val);
+	return (u8 *)l->items;
 }
 
 /* helper function to cast into the data area of the leaf. */
 #define btrfs_item_ptr(leaf, slot, type) \
-	((type *)((leaf)->data + btrfs_item_offset((leaf)->items + (slot))))
+	((type *)(btrfs_leaf_data(leaf) + \
+	btrfs_item_offset((leaf)->items + (slot))))
 
 struct btrfs_buffer *btrfs_alloc_free_block(struct btrfs_root *root);
 int btrfs_inc_ref(struct btrfs_root *root, struct btrfs_buffer *buf);
@@ -416,7 +424,7 @@ int btrfs_del_item(struct btrfs_root *root, struct btrfs_path *path);
 int btrfs_insert_item(struct btrfs_root *root, struct btrfs_key *key,
 		void *data, int data_size);
 int btrfs_next_leaf(struct btrfs_root *root, struct btrfs_path *path);
-int btrfs_leaf_free_space(struct btrfs_leaf *leaf);
+int btrfs_leaf_free_space(struct btrfs_root *root, struct btrfs_leaf *leaf);
 int btrfs_drop_snapshot(struct btrfs_root *root, struct btrfs_buffer *snap);
 int btrfs_finish_extent_commit(struct btrfs_root *root);
 int btrfs_del_root(struct btrfs_root *root, struct btrfs_key *key);

fs/btrfs/disk-io.c

@@ -46,7 +46,8 @@ struct btrfs_buffer *alloc_tree_block(struct btrfs_root *root, u64 blocknr)
 {
 	struct btrfs_buffer *buf;
 	int ret;
-	buf = malloc(sizeof(struct btrfs_buffer));
+
+	buf = malloc(sizeof(struct btrfs_buffer) + root->blocksize);
 	if (!buf)
 		return buf;
 	allocated_blocks++;
@@ -84,7 +85,7 @@ struct btrfs_buffer *find_tree_block(struct btrfs_root *root, u64 blocknr)
 
 struct btrfs_buffer *read_tree_block(struct btrfs_root *root, u64 blocknr)
 {
-	loff_t offset = blocknr * BTRFS_BLOCKSIZE;
+	loff_t offset = blocknr * root->blocksize;
 	struct btrfs_buffer *buf;
 	int ret;
 
@@ -95,8 +96,8 @@ struct btrfs_buffer *read_tree_block(struct btrfs_root *root, u64 blocknr)
 		buf = alloc_tree_block(root, blocknr);
 		if (!buf)
 			return NULL;
-		ret = pread(root->fp, &buf->node, BTRFS_BLOCKSIZE, offset);
-		if (ret != BTRFS_BLOCKSIZE) {
+		ret = pread(root->fp, &buf->node, root->blocksize, offset);
+		if (ret != root->blocksize) {
 			free(buf);
 			return NULL;
 		}
@@ -127,13 +128,13 @@ int clean_tree_block(struct btrfs_root *root, struct btrfs_buffer *buf)
 int write_tree_block(struct btrfs_root *root, struct btrfs_buffer *buf)
 {
 	u64 blocknr = buf->blocknr;
-	loff_t offset = blocknr * BTRFS_BLOCKSIZE;
+	loff_t offset = blocknr * root->blocksize;
 	int ret;
 
 	if (buf->blocknr != btrfs_header_blocknr(&buf->node.header))
 		BUG();
-	ret = pwrite(root->fp, &buf->node, BTRFS_BLOCKSIZE, offset);
-	if (ret != BTRFS_BLOCKSIZE)
+	ret = pwrite(root->fp, &buf->node, root->blocksize, offset);
+	if (ret != root->blocksize)
 		return ret;
 	return 0;
 }
@@ -215,34 +216,37 @@ int btrfs_commit_transaction(struct btrfs_root *root,
 	return ret;
 }
 
-static int __setup_root(struct btrfs_root *root, u64 objectid, int fp)
+static int __setup_root(struct btrfs_super_block *super,
+			struct btrfs_root *root, u64 objectid, int fp)
 {
 	INIT_LIST_HEAD(&root->trans);
 	INIT_LIST_HEAD(&root->cache);
 	root->cache_size = 0;
 	root->fp = fp;
 	root->node = NULL;
 	root->commit_root = NULL;
+	root->blocksize = btrfs_super_blocksize(super);
+	root->ref_cows = 0;
 	memset(&root->current_insert, 0, sizeof(root->current_insert));
 	memset(&root->last_insert, 0, sizeof(root->last_insert));
 	memset(&root->root_key, 0, sizeof(root->root_key));
 	memset(&root->root_item, 0, sizeof(root->root_item));
 	return 0;
 }
 
-static int find_and_setup_root(struct btrfs_root *tree_root, u64 objectid,
-			struct btrfs_root *root, int fp)
+static int find_and_setup_root(struct btrfs_super_block *super,
+			       struct btrfs_root *tree_root, u64 objectid,
+			       struct btrfs_root *root, int fp)
 {
 	int ret;
 
-	__setup_root(root, objectid, fp);
+	__setup_root(super, root, objectid, fp);
 	ret = btrfs_find_last_root(tree_root, objectid,
 				   &root->root_item, &root->root_key);
 	BUG_ON(ret);
 
 	root->node = read_tree_block(root,
 				     btrfs_root_blocknr(&root->root_item));
-	root->ref_cows = 0;
 	BUG_ON(!root->node);
 	return 0;
 }
@@ -277,28 +281,28 @@ struct btrfs_root *open_ctree(char *filename, struct btrfs_super_block *super)
 	INIT_RADIX_TREE(&tree_root->cache_radix, GFP_KERNEL);
 
 	ret = pread(fp, super, sizeof(struct btrfs_super_block),
-		     BTRFS_SUPER_INFO_OFFSET(BTRFS_BLOCKSIZE));
+		     BTRFS_SUPER_INFO_OFFSET);
 	if (ret == 0 || btrfs_super_root(super) == 0) {
 		printf("making new FS!\n");
-		ret = mkfs(fp, 0, BTRFS_BLOCKSIZE);
+		ret = mkfs(fp, 0, 1024);
 		if (ret)
 			return NULL;
 		ret = pread(fp, super, sizeof(struct btrfs_super_block),
-			     BTRFS_SUPER_INFO_OFFSET(BTRFS_BLOCKSIZE));
+			     BTRFS_SUPER_INFO_OFFSET);
 		if (ret != sizeof(struct btrfs_super_block))
 			return NULL;
 	}
 	BUG_ON(ret < 0);
 
-	__setup_root(tree_root, BTRFS_ROOT_TREE_OBJECTID, fp);
+	__setup_root(super, tree_root, BTRFS_ROOT_TREE_OBJECTID, fp);
 	tree_root->node = read_tree_block(tree_root, btrfs_super_root(super));
 	BUG_ON(!tree_root->node);
 
-	ret = find_and_setup_root(tree_root, BTRFS_EXTENT_TREE_OBJECTID,
+	ret = find_and_setup_root(super, tree_root, BTRFS_EXTENT_TREE_OBJECTID,
 				  extent_root, fp);
 	BUG_ON(ret);
 
-	ret = find_and_setup_root(tree_root, BTRFS_FS_TREE_OBJECTID,
+	ret = find_and_setup_root(super, tree_root, BTRFS_FS_TREE_OBJECTID,
 				  root, fp);
 	BUG_ON(ret);
 
@@ -313,7 +317,7 @@ int write_ctree_super(struct btrfs_root *root, struct btrfs_super_block *s)
 	int ret;
 	btrfs_set_super_root(s, root->tree_root->node->blocknr);
 	ret = pwrite(root->fp, s, sizeof(*s),
-		     BTRFS_SUPER_INFO_OFFSET(BTRFS_BLOCKSIZE));
+		     BTRFS_SUPER_INFO_OFFSET);
 	if (ret != sizeof(*s)) {
 		fprintf(stderr, "failed to write new super block err %d\n", ret);
 		return ret;

fs/btrfs/disk-io.h

@@ -5,12 +5,12 @@
 struct btrfs_buffer {
 	u64 blocknr;
 	int count;
+	struct list_head dirty;
+	struct list_head cache;
 	union {
 		struct btrfs_node node;
 		struct btrfs_leaf leaf;
 	};
-	struct list_head dirty;
-	struct list_head cache;
 };
 
 struct btrfs_buffer *read_tree_block(struct btrfs_root *root, u64 blocknr);
@@ -24,9 +24,8 @@ struct btrfs_root *open_ctree(char *filename, struct btrfs_super_block *s);
 int close_ctree(struct btrfs_root *root, struct btrfs_super_block *s);
 void btrfs_block_release(struct btrfs_root *root, struct btrfs_buffer *buf);
 int write_ctree_super(struct btrfs_root *root, struct btrfs_super_block *s);
-int mkfs(int fd, u64 num_blocks, u16 blocksize);
-
+int mkfs(int fd, u64 num_blocks, u32 blocksize);
 
-#define BTRFS_SUPER_INFO_OFFSET(bs) (16 * (bs))
+#define BTRFS_SUPER_INFO_OFFSET (16 * 1024)
 
 #endif

fs/btrfs/extent-tree.c

@@ -143,7 +143,6 @@ static int __free_extent(struct btrfs_root *root, u64 blocknr, u64 num_blocks)
 	struct btrfs_key key;
 	struct btrfs_root *extent_root = root->extent_root;
 	int ret;
-	struct btrfs_item *item;
 	struct btrfs_extent_item *ei;
 	struct btrfs_key ins;
 	u32 refs;
@@ -161,9 +160,8 @@ static int __free_extent(struct btrfs_root *root, u64 blocknr, u64 num_blocks)
 		printf("failed to find %Lu\n", key.objectid);
 		BUG();
 	}
-	item = path.nodes[0]->leaf.items + path.slots[0];
-	ei = (struct btrfs_extent_item *)(path.nodes[0]->leaf.data +
-				    btrfs_item_offset(item));
+	ei = btrfs_item_ptr(&path.nodes[0]->leaf, path.slots[0],
+			    struct btrfs_extent_item);
 	BUG_ON(ei->refs == 0);
 	refs = btrfs_extent_refs(ei) - 1;
 	btrfs_set_extent_refs(ei, refs);