Btrfs: extent fixes

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

fs/btrfs/ctree.c

@@ -354,6 +354,7 @@ int __insert_ptr(struct ctree_root *root,
 		c->header.nritems = 2;
 		c->header.flags = node_level(level);
 		c->header.blocknr = t->blocknr;
+		c->header.parentid = root->node->node.header.parentid;
 		lower = &path->nodes[level-1]->node;
 		if (is_leaf(lower->header.flags))
 			lower_key = &((struct leaf *)lower)->items[0].key;
@@ -363,7 +364,7 @@ int __insert_ptr(struct ctree_root *root,
 		memcpy(c->keys + 1, key, sizeof(struct key));
 		c->blockptrs[0] = path->nodes[level-1]->blocknr;
 		c->blockptrs[1] = blocknr;
-		/* the path has an extra ref to root->node */
+		/* the super has an extra ref to root->node */
 		tree_block_release(root, root->node);
 		root->node = t;
 		t->count++;
@@ -439,6 +440,7 @@ int insert_ptr(struct ctree_root *root,
 		b = &b_buffer->node;
 		b->header.flags = c->header.flags;
 		b->header.blocknr = b_buffer->blocknr;
+		b->header.parentid = root->node->node.header.parentid;
 		mid = (c->header.nritems + 1) / 2;
 		memcpy(b->keys, c->keys + mid,
 			(c->header.nritems - mid) * sizeof(struct key));
@@ -642,6 +644,7 @@ int split_leaf(struct ctree_root *root, struct ctree_path *path, int data_size)
 	right->header.nritems = nritems - mid;
 	right->header.blocknr = right_buffer->blocknr;
 	right->header.flags = node_level(0);
+	right->header.parentid = root->node->node.header.parentid;
 	data_copy_size = l->items[mid].offset + l->items[mid].size -
 			 leaf_data_end(l);
 	memcpy(right->items, l->items + mid,
@@ -689,8 +692,12 @@ int insert_item(struct ctree_root *root, struct key *key,
 	unsigned int data_end;
 	struct ctree_path path;
 
+	refill_alloc_extent(root);
+
 	/* create a root if there isn't one */
 	if (!root->node) {
+		BUG();
+#if 0
 		struct tree_buffer *t;
 		t = alloc_free_block(root);
 		BUG_ON(!t);
@@ -699,6 +706,7 @@ int insert_item(struct ctree_root *root, struct key *key,
 		t->node.header.blocknr = t->blocknr;
 		root->node = t;
 		write_tree_block(root, t);
+#endif
 	}
 	init_path(&path);
 	ret = search_slot(root, key, &path);
@@ -758,7 +766,6 @@ int insert_item(struct ctree_root *root, struct key *key,
 	if (leaf_free_space(leaf) < 0)
 		BUG();
 	release_path(root, &path);
-	refill_alloc_extent(root);
 	return 0;
 }
 
@@ -893,7 +900,7 @@ int next_leaf(struct ctree_root *root, struct ctree_path *path)
 	int level = 1;
 	u64 blocknr;
 	struct tree_buffer *c;
-	struct tree_buffer *next;
+	struct tree_buffer *next = NULL;
 
 	while(level < MAX_LEVEL) {
 		if (!path->nodes[level])
@@ -905,6 +912,8 @@ int next_leaf(struct ctree_root *root, struct ctree_path *path)
 			continue;
 		}
 		blocknr = c->node.blockptrs[slot];
+		if (next)
+			tree_block_release(root, next);
 		next = read_tree_block(root, blocknr);
 		break;
 	}
@@ -922,7 +931,7 @@ int next_leaf(struct ctree_root *root, struct ctree_path *path)
 	return 0;
 }
 
-int alloc_extent(struct ctree_root *root, u64 num_blocks, u64 search_start,
+int alloc_extent(struct ctree_root *orig_root, u64 num_blocks, u64 search_start,
 		 u64 search_end, u64 owner, struct key *ins)
 {
 	struct ctree_path path;
@@ -934,6 +943,7 @@ int alloc_extent(struct ctree_root *root, u64 num_blocks, u64 search_start,
 	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;
@@ -974,13 +984,18 @@ int alloc_extent(struct ctree_root *root, u64 num_blocks, u64 search_start,
 			start_found = 1;
 		last_block = key->objectid + key->offset;
 		path.slots[0]++;
-		printf("last block is not %lu\n", last_block);
 	}
 	// FIXME -ENOSPC
 insert:
+	release_path(root, &path);
 	extent_item.refs = 1;
 	extent_item.owner = owner;
-	ret = insert_item(root, ins, &extent_item, sizeof(extent_item));
+	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 = insert_item(root->extent_root, ins, &extent_item, sizeof(extent_item));
 	return ret;
 }
 
@@ -991,7 +1006,6 @@ static int refill_alloc_extent(struct ctree_root *root)
 	int ret;
 	int min_blocks = MAX_LEVEL * 2;
 
-	printf("refill alloc root %p, numused %lu total %lu\n", root, ae->num_used, ae->num_blocks);
 	if (ae->num_blocks > ae->num_used && ae->num_blocks - ae->num_used >
 	    min_blocks)
 		return 0;
@@ -1007,9 +1021,9 @@ static int refill_alloc_extent(struct ctree_root *root)
 			BUG();
 		return 0;
 	}
-	// FIXME, this recurses
-	ret = alloc_extent(root->extent_root,
-			   min_blocks * 2, 0, (unsigned long)-1, 0, &key);
+	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;
@@ -1021,6 +1035,7 @@ void print_leaf(struct leaf *l)
 	int i;
 	int nr = l->header.nritems;
 	struct item *item;
+	struct extent_item *ei;
 	printf("leaf %lu total ptrs %d free space %d\n", l->header.blocknr, nr,
 	       leaf_free_space(l));
 	fflush(stdout);
@@ -1032,6 +1047,8 @@ void print_leaf(struct leaf *l)
 			item->offset, item->size);
 		fflush(stdout);
 		printf("\t\titem data %.*s\n", item->size, l->data+item->offset);
+		ei = (struct extent_item *)(l->data + item->offset);
+		printf("\t\textent data %u %lu\n", ei->refs, ei->owner);
 		fflush(stdout);
 	}
 }
@@ -1080,8 +1097,8 @@ void print_tree(struct ctree_root *root, struct tree_buffer *t)
 
 /* for testing only */
 int next_key(int i, int max_key) {
-	// return rand() % max_key;
-	return i;
+	return rand() % max_key;
+	// return i;
 }
 
 int main() {
@@ -1092,15 +1109,20 @@ int main() {
 	int i;
 	int num;
 	int ret;
-	int run_size = 256;
+	int run_size = 10000;
 	int max_key = 100000000;
 	int tree_size = 0;
 	struct ctree_path path;
+	struct ctree_super_block super;
 
 	radix_tree_init();
 
 
-	root = open_ctree("dbfile");
+	root = open_ctree("dbfile", &super);
+	printf("root tree\n");
+	print_tree(root, root->node);
+	printf("map tree\n");
+	print_tree(root->extent_root, root->extent_root->node);
 
 	srand(55);
 	for (i = 0; i < run_size; i++) {
@@ -1112,22 +1134,20 @@ int main() {
 		ins.objectid = num;
 		ins.offset = 0;
 		ins.flags = 0;
-		printf("insert %d\n", i);
 		ret = insert_item(root, &ins, buf, strlen(buf));
 		if (!ret)
 			tree_size++;
-		printf("done insert %d\n", i);
 	}
 	printf("root used: %lu\n", root->alloc_extent->num_used);
 	printf("root tree\n");
-	print_tree(root, root->node);
+	// 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);
-	exit(1);
-
+	// print_tree(root->extent_root, root->extent_root->node);
+	write_ctree_super(root, &super);
 	close_ctree(root);
-	root = open_ctree("dbfile");
+
+	root = open_ctree("dbfile", &super);
 	printf("starting search\n");
 	srand(55);
 	for (i = 0; i < run_size; i++) {
@@ -1142,8 +1162,9 @@ int main() {
 		}
 		release_path(root, &path);
 	}
+	write_ctree_super(root, &super);
 	close_ctree(root);
-	root = open_ctree("dbfile");
+	root = open_ctree("dbfile", &super);
 	printf("node %p level %d total ptrs %d free spc %lu\n", root->node,
 	        node_level(root->node->node.header.flags),
 		root->node->node.header.nritems,
@@ -1174,8 +1195,9 @@ int main() {
 		if (!ret)
 			tree_size++;
 	}
+	write_ctree_super(root, &super);
 	close_ctree(root);
-	root = open_ctree("dbfile");
+	root = open_ctree("dbfile", &super);
 	printf("starting search2\n");
 	srand(128);
 	for (i = 0; i < run_size; i++) {
@@ -1221,6 +1243,7 @@ int main() {
 		}
 		release_path(root, &path);
 	}
+	write_ctree_super(root, &super);
 	close_ctree(root);
 	printf("tree size is now %d\n", tree_size);
 	return 0;

fs/btrfs/ctree.h

@@ -57,6 +57,11 @@ struct ctree_root_info {
 	u64 snapuuid[2]; /* root specific uuid */
 } __attribute__ ((__packed__));
 
+struct ctree_super_block {
+	struct ctree_root_info root_info;
+	struct ctree_root_info extent_info;
+} __attribute__ ((__packed__));
+
 struct item {
 	struct key key;
 	u16 offset;

fs/btrfs/disk-io.c

@@ -15,7 +15,7 @@ static int allocated_blocks = 0;
 static int get_free_block(struct ctree_root *root, u64 *block)
 {
 	struct stat st;
-	int ret;
+	int ret = 0;
 
 	if (root->alloc_extent->num_used >= root->alloc_extent->num_blocks)
 		return -1;
@@ -30,9 +30,14 @@ static int get_free_block(struct ctree_root *root, u64 *block)
 	}
 	st.st_size = 0;
 	ret = fstat(root->fp, &st);
-	if (st.st_size < (*block + 1) * CTREE_BLOCKSIZE)
+	if (st.st_size < (*block + 1) * CTREE_BLOCKSIZE) {
 		ret = ftruncate(root->fp,
 				(*block + 1) * CTREE_BLOCKSIZE);
+		if (ret) {
+			perror("ftruncate");
+			exit(1);
+		}
+	}
 	return ret;
 }
 
@@ -81,11 +86,7 @@ struct tree_buffer *read_tree_block(struct ctree_root *root, u64 blocknr)
 	buf = radix_tree_lookup(&root->cache_radix, blocknr);
 	if (buf) {
 		buf->count++;
-		if (buf->blocknr != blocknr)
-			BUG();
-		if (buf->blocknr != buf->node.header.blocknr)
-			BUG();
-		return buf;
+		goto test;
 	}
 	buf = alloc_tree_block(root, blocknr);
 	if (!buf)
@@ -95,8 +96,11 @@ struct tree_buffer *read_tree_block(struct ctree_root *root, u64 blocknr)
 		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)
+		BUG();
 	return buf;
 }
 
@@ -111,36 +115,30 @@ int write_tree_block(struct ctree_root *root, struct tree_buffer *buf)
 	ret = pwrite(root->fp, &buf->node, CTREE_BLOCKSIZE, offset);
 	if (ret != CTREE_BLOCKSIZE)
 		return ret;
-	if (buf == root->node)
-		return update_root_block(root);
 	return 0;
 }
 
-struct ctree_super_block {
-	struct ctree_root_info root_info;
-	struct ctree_root_info extent_info;
-} __attribute__ ((__packed__));
-
 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;
-	INIT_RADIX_TREE(&root->cache_radix, GFP_KERNEL);
-	printf("setup done reading root %p, used %lu\n", root, root->alloc_extent->num_used);
+	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;
 }
 
-struct ctree_root *open_ctree(char *filename)
+struct ctree_root *open_ctree(char *filename, struct ctree_super_block *super)
 {
 	struct ctree_root *root = malloc(sizeof(struct ctree_root));
 	struct ctree_root *extent_root = malloc(sizeof(struct ctree_root));
-	struct ctree_super_block super;
 	int fp;
 	int ret;
 
@@ -149,48 +147,61 @@ struct ctree_root *open_ctree(char *filename)
 		free(root);
 		return NULL;
 	}
-	ret = pread(fp, &super, sizeof(struct ctree_super_block),
+	ret = pread(fp, super, sizeof(struct ctree_super_block),
 		     CTREE_SUPER_INFO_OFFSET(CTREE_BLOCKSIZE));
 	if (ret == 0) {
 		ret = mkfs(fp);
 		if (ret)
 			return NULL;
-		ret = pread(fp, &super, sizeof(struct ctree_super_block),
+		ret = pread(fp, super, sizeof(struct ctree_super_block),
 			     CTREE_SUPER_INFO_OFFSET(CTREE_BLOCKSIZE));
 		if (ret != sizeof(struct ctree_super_block))
 			return NULL;
 	}
 	BUG_ON(ret < 0);
-	__setup_root(root, extent_root, &super.root_info, fp);
-	__setup_root(extent_root, extent_root, &super.extent_info, fp);
+	__setup_root(root, extent_root, &super->root_info, fp);
+	__setup_root(extent_root, extent_root, &super->extent_info, fp);
 	return root;
 }
 
-int close_ctree(struct ctree_root *root)
+static int __update_root(struct ctree_root *root, struct ctree_root_info *info)
 {
-	close(root->fp);
-	if (root->node)
-		tree_block_release(root, root->node);
-	free(root);
-	printf("on close %d blocks are allocated\n", allocated_blocks);
+	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;
 }
 
-int update_root_block(struct ctree_root *root)
+int write_ctree_super(struct ctree_root *root, struct ctree_super_block *s)
 {
 	int ret;
-	u64 root_block = root->node->blocknr;
-
-	ret = pwrite(root->fp, &root_block, sizeof(u64), 0);
-	if (ret != sizeof(u64))
+	__update_root(root, &s->root_info);
+	__update_root(root->extent_root, &s->extent_info);
+	ret = pwrite(root->fp, s, sizeof(*s), CTREE_SUPER_INFO_OFFSET(CTREE_BLOCKSIZE));
+	if (ret != sizeof(*s)) {
+		fprintf(stderr, "failed to write new super block err %d\n", ret);
 		return ret;
+	}
+	return 0;
+}
+
+int close_ctree(struct ctree_root *root)
+{
+	close(root->fp);
+	if (root->node)
+		tree_block_release(root, root->node);
+	if (root->extent_root->node)
+		tree_block_release(root->extent_root, root->extent_root->node);
+	free(root);
+	printf("on close %d blocks are allocated\n", allocated_blocks);
 	return 0;
 }
 
 void tree_block_release(struct ctree_root *root, struct tree_buffer *buf)
 {
-	return;
 	buf->count--;
+	if (buf->count < 0)
+		BUG();
 	if (buf->count == 0) {
 		if (!radix_tree_lookup(&root->cache_radix, buf->blocknr))
 			BUG();

fs/btrfs/disk-io.h

@@ -12,11 +12,11 @@ struct tree_buffer {
 
 struct tree_buffer *read_tree_block(struct ctree_root *root, u64 blocknr);
 int write_tree_block(struct ctree_root *root, struct tree_buffer *buf);
-struct ctree_root *open_ctree(char *filename);
+struct ctree_root *open_ctree(char *filename, struct ctree_super_block *s);
 int close_ctree(struct ctree_root *root);
 void tree_block_release(struct ctree_root *root, struct tree_buffer *buf);
 struct tree_buffer *alloc_free_block(struct ctree_root *root);
-int update_root_block(struct ctree_root *root);
+int write_ctree_super(struct ctree_root *root, struct ctree_super_block *s);
 int mkfs(int fd);
 
 #define CTREE_SUPER_INFO_OFFSET(bs) (16 * (bs))

fs/btrfs/mkfs.c

@@ -18,12 +18,13 @@ int mkfs(int fd)
 	struct extent_item extent_item;
 	int ret;
 
+	/* setup the super block area */
 	memset(info, 0, sizeof(info));
 	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 = 20;
+	info[0].alloc_extent.num_blocks = 64;
 	/* 0-17 are used (inclusive) */
 	info[0].alloc_extent.num_used = 18;
 
@@ -31,12 +32,14 @@ int mkfs(int fd)
 	info[1].objectid = 2;
 	info[1].tree_root = 64;
 	info[1].alloc_extent.blocknr = 64;
-	info[1].alloc_extent.num_blocks = 8;
+	info[1].alloc_extent.num_blocks = 64;
 	info[1].alloc_extent.num_used = 1;
 	ret = pwrite(fd, info, sizeof(info),
 		     CTREE_SUPER_INFO_OFFSET(CTREE_BLOCKSIZE));
 	if (ret != sizeof(info))
 		return -1;
+
+	/* create leaves for the tree root and extent root */
 	memset(&empty_leaf, 0, sizeof(empty_leaf));
 	empty_leaf.header.parentid = 1;
 	empty_leaf.header.blocknr = 17;