Btrfs: add disk ioctl, mostly working

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

fs/btrfs/disk-io.c

@@ -16,6 +16,28 @@ struct dev_lookup {
 	struct block_device *bdev;
 };
 
+int btrfs_insert_dev_radix(struct btrfs_root *root,
+			   struct block_device *bdev,
+			   u64 block_start,
+			   u64 num_blocks)
+{
+	struct dev_lookup *lookup;
+	char b[BDEVNAME_SIZE];
+	int ret;
+
+	lookup = kmalloc(sizeof(*lookup), GFP_NOFS);
+	if (!lookup)
+		return -ENOMEM;
+	lookup->block_start = block_start;
+	lookup->num_blocks = num_blocks;
+	lookup->bdev = bdev;
+printk("inserting %s into dev radix %Lu %Lu\n", bdevname(bdev, b), block_start, num_blocks);
+
+	ret = radix_tree_insert(&root->fs_info->dev_radix, block_start +
+				num_blocks - 1, lookup);
+	return ret;
+}
+
 u64 bh_blocknr(struct buffer_head *bh)
 {
 	int blkbits = bh->b_page->mapping->host->i_blkbits;
@@ -34,6 +56,8 @@ static int check_tree_block(struct btrfs_root *root, struct buffer_head *buf)
 {
 	struct btrfs_node *node = btrfs_buffer_node(buf);
 	if (bh_blocknr(buf) != btrfs_header_blocknr(&node->header)) {
+		printk(KERN_CRIT "bh_blocknr(buf) is %Lu, header is %Lu\n",
+		       bh_blocknr(buf), btrfs_header_blocknr(&node->header));
 		BUG();
 	}
 	return 0;
@@ -76,11 +100,10 @@ struct buffer_head *btrfs_find_tree_block(struct btrfs_root *root, u64 blocknr)
 	return ret;
 }
 
-static int map_bh_to_logical(struct btrfs_root *root, struct buffer_head *bh,
+int btrfs_map_bh_to_logical(struct btrfs_root *root, struct buffer_head *bh,
 			     u64 logical)
 {
 	struct dev_lookup *lookup[2];
-	char b[BDEVNAME_SIZE];
 
 	int ret;
 
@@ -96,7 +119,6 @@ static int map_bh_to_logical(struct btrfs_root *root, struct buffer_head *bh,
 	}
 	bh->b_bdev = lookup[0]->bdev;
 	bh->b_blocknr = logical - lookup[0]->block_start;
-printk("logical mapping %Lu to %lu bdev  %s\n", logical, bh->b_blocknr, bdevname(bh->b_bdev, b));
 	set_buffer_mapped(bh);
 	ret = 0;
 out:
@@ -126,7 +148,7 @@ struct buffer_head *btrfs_find_create_tree_block(struct btrfs_root *root,
 	bh = head;
 	do {
 		if (!buffer_mapped(bh)) {
-			err = map_bh_to_logical(root, bh, first_block);
+			err = btrfs_map_bh_to_logical(root, bh, first_block);
 			BUG_ON(err);
 		}
 		if (bh_blocknr(bh) == blocknr) {
@@ -150,7 +172,7 @@ static int btree_get_block(struct inode *inode, sector_t iblock,
 {
 	int err;
 	struct btrfs_root *root = BTRFS_I(bh->b_page->mapping->host)->root;
-	err = map_bh_to_logical(root, bh, iblock);
+	err = btrfs_map_bh_to_logical(root, bh, iblock);
 	return err;
 }
 
@@ -396,6 +418,89 @@ printk("all worked\n");
 	return root;
 }
 
+int btrfs_open_disk(struct btrfs_root *root, u64 block_start, u64 num_blocks,
+		    char *filename, int name_len)
+{
+	char *null_filename;
+	struct block_device *bdev;
+	int ret;
+
+	if (block_start == 0) {
+printk("skipping disk with block_start == 0\n");
+return 0;
+	}
+	null_filename = kmalloc(name_len + 1, GFP_NOFS);
+	if (!null_filename)
+		return -ENOMEM;
+	memcpy(null_filename, filename, name_len);
+	null_filename[name_len] = '\0';
+
+	bdev = open_bdev_excl(null_filename, O_RDWR, root->fs_info->sb);
+	if (IS_ERR(bdev)) {
+		ret = PTR_ERR(bdev);
+		goto out;
+	}
+	set_blocksize(bdev, root->fs_info->sb->s_blocksize);
+	ret = btrfs_insert_dev_radix(root, bdev, block_start, num_blocks);
+	BUG_ON(ret);
+	ret = 0;
+out:
+	kfree(null_filename);
+	return ret;
+}
+
+static int read_device_info(struct btrfs_root *root)
+{
+	struct btrfs_path *path;
+	int ret;
+	struct btrfs_key key;
+	struct btrfs_leaf *leaf;
+	struct btrfs_device_item *dev_item;
+	int nritems;
+	int slot;
+
+	root = root->fs_info->dev_root;
+
+	path = btrfs_alloc_path();
+	if (!path)
+		return -ENOMEM;
+	key.objectid = 0;
+	key.offset = 0;
+	key.flags = 0;
+	btrfs_set_key_type(&key, BTRFS_DEV_ITEM_KEY);
+
+	mutex_lock(&root->fs_info->fs_mutex);
+	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
+	leaf = btrfs_buffer_leaf(path->nodes[0]);
+	nritems = btrfs_header_nritems(&leaf->header);
+	while(1) {
+		slot = path->slots[0];
+		if (slot >= nritems) {
+			ret = btrfs_next_leaf(root, path);
+			if (ret)
+				break;
+			leaf = btrfs_buffer_leaf(path->nodes[0]);
+			nritems = btrfs_header_nritems(&leaf->header);
+			slot = path->slots[0];
+		}
+		btrfs_disk_key_to_cpu(&key, &leaf->items[slot].key);
+		if (btrfs_key_type(&key) != BTRFS_DEV_ITEM_KEY) {
+			path->slots[0]++;
+			continue;
+		}
+		dev_item = btrfs_item_ptr(leaf, slot, struct btrfs_device_item);
+printk("found key %Lu %Lu\n", key.objectid, key.offset);
+		ret = btrfs_open_disk(root, key.objectid, key.offset,
+				      (char *)(dev_item + 1),
+				      btrfs_device_pathlen(dev_item));
+		BUG_ON(ret);
+		path->slots[0]++;
+	}
+	btrfs_free_path(path);
+	mutex_unlock(&root->fs_info->fs_mutex);
+	return 0;
+}
+
 struct btrfs_root *open_ctree(struct super_block *sb)
 {
 	struct btrfs_root *extent_root = kmalloc(sizeof(struct btrfs_root),
@@ -463,17 +568,26 @@ struct btrfs_root *open_ctree(struct super_block *sb)
 	if (!btrfs_super_root(disk_super))
 		return NULL;
 
+	i_size_write(fs_info->btree_inode,
+		     btrfs_super_total_blocks(disk_super) <<
+		     fs_info->btree_inode->i_blkbits);
+
 	radix_tree_delete(&fs_info->dev_radix, (u32)-2);
 	dev_lookup->block_start = btrfs_super_device_block_start(disk_super);
 	dev_lookup->num_blocks = btrfs_super_device_num_blocks(disk_super);
 	ret = radix_tree_insert(&fs_info->dev_radix,
 				dev_lookup->block_start +
-				dev_lookup->num_blocks, dev_lookup);
+				dev_lookup->num_blocks - 1, dev_lookup);
 	BUG_ON(ret);
 
 	fs_info->disk_super = disk_super;
+
 	dev_root->node = read_tree_block(tree_root,
 					  btrfs_super_device_root(disk_super));
+
+	ret = read_device_info(dev_root);
+	BUG_ON(ret);
+
 	tree_root->node = read_tree_block(tree_root,
 					  btrfs_super_root(disk_super));
 	BUG_ON(!tree_root->node);
@@ -562,7 +676,7 @@ static int free_dev_radix(struct btrfs_fs_info *fs_info)
 				close_bdev_excl(lookup[i]->bdev);
 			radix_tree_delete(&fs_info->dev_radix,
 					  lookup[i]->block_start +
-					  lookup[i]->num_blocks);
+					  lookup[i]->num_blocks - 1);
 			kfree(lookup[i]);
 		}
 	}

fs/btrfs/disk-io.h

@@ -42,4 +42,10 @@ int btrfs_csum_data(struct btrfs_root * root, char *data, size_t len,
 struct btrfs_root *btrfs_read_fs_root(struct btrfs_fs_info *fs_info,
 				      struct btrfs_key *location);
 u64 bh_blocknr(struct buffer_head *bh);
+int btrfs_insert_dev_radix(struct btrfs_root *root,
+			   struct block_device *bdev,
+			   u64 block_start,
+			   u64 num_blocks);
+int btrfs_map_bh_to_logical(struct btrfs_root *root, struct buffer_head *bh,
+			     u64 logical);
 #endif

fs/btrfs/ioctl.h

@@ -10,4 +10,6 @@ struct btrfs_ioctl_vol_args {
 
 #define BTRFS_IOC_SNAP_CREATE _IOW(BTRFS_IOCTL_MAGIC, 1, \
 				   struct btrfs_ioctl_vol_args)
+#define BTRFS_IOC_ADD_DISK _IOW(BTRFS_IOCTL_MAGIC, 2, \
+				   struct btrfs_ioctl_vol_args)
 #endif

fs/btrfs/super.c

@@ -1073,6 +1073,7 @@ static int btrfs_get_block_lock(struct inode *inode, sector_t iblock,
 
 		}
 		map_bh(result, inode->i_sb, blocknr);
+		btrfs_map_bh_to_logical(root, result, blocknr);
 		goto out;
 	}
 
@@ -1092,7 +1093,8 @@ static int btrfs_get_block_lock(struct inode *inode, sector_t iblock,
 	extent_end = extent_start + btrfs_file_extent_num_blocks(item);
 	if (iblock >= extent_start && iblock < extent_end) {
 		err = 0;
-		map_bh(result, inode->i_sb, blocknr + iblock - extent_start);
+		btrfs_map_bh_to_logical(root, result, blocknr + iblock -
+					extent_start);
 		goto out;
 	}
 allocate:
@@ -1112,6 +1114,7 @@ static int btrfs_get_block_lock(struct inode *inode, sector_t iblock,
 	set_buffer_new(result);
 	map_bh(result, inode->i_sb, blocknr);
 
+	btrfs_map_bh_to_logical(root, result, blocknr);
 out:
 	btrfs_release_path(root, path);
 	btrfs_free_path(path);
@@ -1153,12 +1156,6 @@ static int btrfs_readpage(struct file *file, struct page *page)
 	return mpage_readpage(page, btrfs_get_block);
 }
 
-static int btrfs_readpages(struct file *file, struct address_space *mapping,
-			   struct list_head *pages, unsigned nr_pages)
-{
-	return mpage_readpages(mapping, pages, nr_pages, btrfs_get_block);
-}
-
 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
 {
 	return nobh_writepage(page, btrfs_get_block, wbc);
@@ -1831,12 +1828,81 @@ printk("adding snapshot name %.*s root %Lu %Lu %u\n", namelen, name, key.objecti
 	return 0;
 }
 
+static int add_disk(struct btrfs_root *root, char *name, int namelen)
+{
+	struct block_device *bdev;
+	struct btrfs_path *path;
+	struct super_block *sb = root->fs_info->sb;
+	struct btrfs_root *dev_root = root->fs_info->dev_root;
+	struct btrfs_trans_handle *trans;
+	struct btrfs_device_item *dev_item;
+	struct btrfs_key key;
+	u16 item_size;
+	u64 num_blocks;
+	u64 new_blocks;
+	int ret;
+printk("adding disk %s\n", name);
+	path = btrfs_alloc_path();
+	if (!path)
+		return -ENOMEM;
+	num_blocks = btrfs_super_total_blocks(root->fs_info->disk_super);
+	bdev = open_bdev_excl(name, O_RDWR, sb);
+	if (IS_ERR(bdev)) {
+		ret = PTR_ERR(bdev);
+printk("open bdev excl failed ret %d\n", ret);
+		goto out_nolock;
+	}
+	set_blocksize(bdev, sb->s_blocksize);
+	new_blocks = bdev->bd_inode->i_size >> sb->s_blocksize_bits;
+	key.objectid = num_blocks;
+	key.offset = new_blocks;
+	key.flags = 0;
+	btrfs_set_key_type(&key, BTRFS_DEV_ITEM_KEY);
+
+	mutex_lock(&dev_root->fs_info->fs_mutex);
+	trans = btrfs_start_transaction(dev_root, 1);
+	item_size = sizeof(*dev_item) + namelen;
+printk("insert empty on %Lu %Lu %u size %d\n", num_blocks, new_blocks, key.flags, item_size);
+	ret = btrfs_insert_empty_item(trans, dev_root, path, &key, item_size);
+	if (ret) {
+printk("insert failed %d\n", ret);
+		close_bdev_excl(bdev);
+		if (ret > 0)
+			ret = -EEXIST;
+		goto out;
+	}
+	dev_item = btrfs_item_ptr(btrfs_buffer_leaf(path->nodes[0]),
+				  path->slots[0], struct btrfs_device_item);
+	btrfs_set_device_pathlen(dev_item, namelen);
+	memcpy(dev_item + 1, name, namelen);
+	mark_buffer_dirty(path->nodes[0]);
+
+	ret = btrfs_insert_dev_radix(root, bdev, num_blocks, new_blocks);
+
+	if (!ret) {
+		btrfs_set_super_total_blocks(root->fs_info->disk_super,
+					     num_blocks + new_blocks);
+		i_size_write(root->fs_info->btree_inode,
+			     (num_blocks + new_blocks) <<
+			     root->fs_info->btree_inode->i_blkbits);
+	}
+
+out:
+	ret = btrfs_commit_transaction(trans, dev_root);
+	BUG_ON(ret);
+	mutex_unlock(&root->fs_info->fs_mutex);
+out_nolock:
+	btrfs_free_path(path);
+
+	return ret;
+}
+
 static int btrfs_ioctl(struct inode *inode, struct file *filp, unsigned int
 		       cmd, unsigned long arg)
 {
 	struct btrfs_root *root = BTRFS_I(inode)->root;
 	struct btrfs_ioctl_vol_args vol_args;
-	int ret;
+	int ret = 0;
 	int namelen;
 	struct btrfs_path *path;
 	u64 root_dirid;
@@ -1869,10 +1935,21 @@ static int btrfs_ioctl(struct inode *inode, struct file *filp, unsigned int
 			ret = create_snapshot(root, vol_args.name, namelen);
 		WARN_ON(ret);
 		break;
+	case BTRFS_IOC_ADD_DISK:
+		if (copy_from_user(&vol_args,
+				   (struct btrfs_ioctl_vol_args __user *)arg,
+				   sizeof(vol_args)))
+			return -EFAULT;
+		namelen = strlen(vol_args.name);
+		if (namelen > BTRFS_VOL_NAME_MAX)
+			return -EINVAL;
+		vol_args.name[namelen] = '\0';
+		ret = add_disk(root, vol_args.name, namelen);
+		break;
 	default:
 		return -ENOTTY;
 	}
-	return 0;
+	return ret;
 }
 
 static struct kmem_cache *btrfs_inode_cachep;
@@ -2004,7 +2081,6 @@ static struct file_operations btrfs_dir_file_operations = {
 
 static struct address_space_operations btrfs_aops = {
 	.readpage	= btrfs_readpage,
-	.readpages	= btrfs_readpages,
 	.writepage	= btrfs_writepage,
 	.sync_page	= block_sync_page,
 	.prepare_write	= btrfs_prepare_write,

fs/btrfs/transaction.c

@@ -116,7 +116,13 @@ int btrfs_commit_tree_roots(struct btrfs_trans_handle *trans,
 	struct btrfs_fs_info *fs_info = root->fs_info;
 	struct btrfs_root *tree_root = fs_info->tree_root;
 	struct btrfs_root *extent_root = fs_info->extent_root;
+	struct btrfs_root *dev_root = fs_info->dev_root;
 
+	if (btrfs_super_device_root(fs_info->disk_super) !=
+	    bh_blocknr(dev_root->node)) {
+		btrfs_set_super_device_root(fs_info->disk_super,
+					    bh_blocknr(dev_root->node));
+	}
 	while(1) {
 		old_extent_block = btrfs_root_blocknr(&extent_root->root_item);
 		if (old_extent_block == bh_blocknr(extent_root->node))