Btrfs: change how subvolumes are organized
[linux-2.6-block.git] / fs / btrfs / inode.c
index 9e81f3184f249ebe05dbf66034c4d202b005dc36..6036b36789cc705b6e4fb2a0c892a837cebdbac0 100644 (file)
@@ -2373,6 +2373,69 @@ static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
        return ret;
 }
 
+int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
+                       struct btrfs_root *root,
+                       struct inode *dir, u64 objectid,
+                       const char *name, int name_len)
+{
+       struct btrfs_path *path;
+       struct extent_buffer *leaf;
+       struct btrfs_dir_item *di;
+       struct btrfs_key key;
+       u64 index;
+       int ret;
+
+       path = btrfs_alloc_path();
+       if (!path)
+               return -ENOMEM;
+
+       di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
+                                  name, name_len, -1);
+       BUG_ON(!di || IS_ERR(di));
+
+       leaf = path->nodes[0];
+       btrfs_dir_item_key_to_cpu(leaf, di, &key);
+       WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
+       ret = btrfs_delete_one_dir_name(trans, root, path, di);
+       BUG_ON(ret);
+       btrfs_release_path(root, path);
+
+       ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
+                                objectid, root->root_key.objectid,
+                                dir->i_ino, &index, name, name_len);
+       if (ret < 0) {
+               BUG_ON(ret != -ENOENT);
+               di = btrfs_search_dir_index_item(root, path, dir->i_ino,
+                                                name, name_len);
+               BUG_ON(!di || IS_ERR(di));
+
+               leaf = path->nodes[0];
+               btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
+               btrfs_release_path(root, path);
+               index = key.offset;
+       }
+
+       di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
+                                        index, name, name_len, -1);
+       BUG_ON(!di || IS_ERR(di));
+
+       leaf = path->nodes[0];
+       btrfs_dir_item_key_to_cpu(leaf, di, &key);
+       WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
+       ret = btrfs_delete_one_dir_name(trans, root, path, di);
+       BUG_ON(ret);
+       btrfs_release_path(root, path);
+
+       btrfs_i_size_write(dir, dir->i_size - name_len * 2);
+       dir->i_mtime = dir->i_ctime = CURRENT_TIME;
+       ret = btrfs_update_inode(trans, root, dir);
+       BUG_ON(ret);
+       dir->i_sb->s_dirt = 1;
+
+       btrfs_free_path(path);
+       return 0;
+}
+
 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
 {
        struct inode *inode = dentry->d_inode;
@@ -2382,29 +2445,31 @@ static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
        struct btrfs_trans_handle *trans;
        unsigned long nr = 0;
 
-       /*
-        * the FIRST_FREE_OBJECTID check makes sure we don't try to rmdir
-        * the root of a subvolume or snapshot
-        */
        if (inode->i_size > BTRFS_EMPTY_DIR_SIZE ||
-           inode->i_ino == BTRFS_FIRST_FREE_OBJECTID) {
+           inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
                return -ENOTEMPTY;
-       }
 
        trans = btrfs_start_transaction(root, 1);
        btrfs_set_trans_block_group(trans, dir);
 
+       if (unlikely(inode->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
+               err = btrfs_unlink_subvol(trans, root, dir,
+                                         BTRFS_I(inode)->location.objectid,
+                                         dentry->d_name.name,
+                                         dentry->d_name.len);
+               goto out;
+       }
+
        err = btrfs_orphan_add(trans, inode);
        if (err)
-               goto fail_trans;
+               goto out;
 
        /* now the directory is empty */
        err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
                                 dentry->d_name.name, dentry->d_name.len);
        if (!err)
                btrfs_i_size_write(inode, 0);
-
-fail_trans:
+out:
        nr = trans->blocks_used;
        ret = btrfs_end_transaction_throttle(trans, root);
        btrfs_btree_balance_dirty(root, nr);
@@ -3091,29 +3156,67 @@ out_err:
  * is kind of like crossing a mount point.
  */
 static int fixup_tree_root_location(struct btrfs_root *root,
-                            struct btrfs_key *location,
-                            struct btrfs_root **sub_root,
-                            struct dentry *dentry)
+                                   struct inode *dir,
+                                   struct dentry *dentry,
+                                   struct btrfs_key *location,
+                                   struct btrfs_root **sub_root)
 {
-       struct btrfs_root_item *ri;
+       struct btrfs_path *path;
+       struct btrfs_root *new_root;
+       struct btrfs_root_ref *ref;
+       struct extent_buffer *leaf;
+       int ret;
+       int err = 0;
 
-       if (btrfs_key_type(location) != BTRFS_ROOT_ITEM_KEY)
-               return 0;
-       if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
-               return 0;
+       path = btrfs_alloc_path();
+       if (!path) {
+               err = -ENOMEM;
+               goto out;
+       }
 
-       *sub_root = btrfs_read_fs_root(root->fs_info, location,
-                                       dentry->d_name.name,
-                                       dentry->d_name.len);
-       if (IS_ERR(*sub_root))
-               return PTR_ERR(*sub_root);
+       err = -ENOENT;
+       ret = btrfs_find_root_ref(root->fs_info->tree_root, path,
+                                 BTRFS_I(dir)->root->root_key.objectid,
+                                 location->objectid);
+       if (ret) {
+               if (ret < 0)
+                       err = ret;
+               goto out;
+       }
 
-       ri = &(*sub_root)->root_item;
-       location->objectid = btrfs_root_dirid(ri);
-       btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
-       location->offset = 0;
+       leaf = path->nodes[0];
+       ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
+       if (btrfs_root_ref_dirid(leaf, ref) != dir->i_ino ||
+           btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
+               goto out;
 
-       return 0;
+       ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
+                                  (unsigned long)(ref + 1),
+                                  dentry->d_name.len);
+       if (ret)
+               goto out;
+
+       btrfs_release_path(root->fs_info->tree_root, path);
+
+       new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
+       if (IS_ERR(new_root)) {
+               err = PTR_ERR(new_root);
+               goto out;
+       }
+
+       if (btrfs_root_refs(&new_root->root_item) == 0) {
+               err = -ENOENT;
+               goto out;
+       }
+
+       *sub_root = new_root;
+       location->objectid = btrfs_root_dirid(&new_root->root_item);
+       location->type = BTRFS_INODE_ITEM_KEY;
+       location->offset = 0;
+       err = 0;
+out:
+       btrfs_free_path(path);
+       return err;
 }
 
 static void inode_tree_add(struct inode *inode)
@@ -3246,11 +3349,34 @@ struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
        return inode;
 }
 
+static struct inode *new_simple_dir(struct super_block *s,
+                                   struct btrfs_key *key,
+                                   struct btrfs_root *root)
+{
+       struct inode *inode = new_inode(s);
+
+       if (!inode)
+               return ERR_PTR(-ENOMEM);
+
+       init_btrfs_i(inode);
+
+       BTRFS_I(inode)->root = root;
+       memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
+       BTRFS_I(inode)->dummy_inode = 1;
+
+       inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
+       inode->i_op = &simple_dir_inode_operations;
+       inode->i_fop = &simple_dir_operations;
+       inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
+       inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
+
+       return inode;
+}
+
 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
 {
        struct inode *inode;
-       struct btrfs_inode *bi = BTRFS_I(dir);
-       struct btrfs_root *root = bi->root;
+       struct btrfs_root *root = BTRFS_I(dir)->root;
        struct btrfs_root *sub_root = root;
        struct btrfs_key location;
        int ret;
@@ -3263,17 +3389,25 @@ struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
        if (ret < 0)
                return ERR_PTR(ret);
 
-       inode = NULL;
-       if (location.objectid) {
-               ret = fixup_tree_root_location(root, &location, &sub_root,
-                                               dentry);
-               if (ret < 0)
-                       return ERR_PTR(ret);
-               if (ret > 0)
-                       return ERR_PTR(-ENOENT);
+       if (location.objectid == 0)
+               return NULL;
+
+       if (location.type == BTRFS_INODE_ITEM_KEY) {
+               inode = btrfs_iget(dir->i_sb, &location, root);
+               return inode;
+       }
+
+       BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
+
+       ret = fixup_tree_root_location(root, dir, dentry,
+                                      &location, &sub_root);
+       if (ret < 0) {
+               if (ret != -ENOENT)
+                       inode = ERR_PTR(ret);
+               else
+                       inode = new_simple_dir(dir->i_sb, &location, sub_root);
+       } else {
                inode = btrfs_iget(dir->i_sb, &location, sub_root);
-               if (IS_ERR(inode))
-                       return ERR_CAST(inode);
        }
        return inode;
 }
@@ -3283,9 +3417,6 @@ static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
 {
        struct inode *inode;
 
-       if (dentry->d_name.len > BTRFS_NAME_LEN)
-               return ERR_PTR(-ENAMETOOLONG);
-
        inode = btrfs_lookup_dentry(dir, dentry);
        if (IS_ERR(inode))
                return ERR_CAST(inode);
@@ -3691,26 +3822,35 @@ int btrfs_add_link(struct btrfs_trans_handle *trans,
                   struct inode *parent_inode, struct inode *inode,
                   const char *name, int name_len, int add_backref, u64 index)
 {
-       int ret;
+       int ret = 0;
        struct btrfs_key key;
        struct btrfs_root *root = BTRFS_I(parent_inode)->root;
 
-       key.objectid = inode->i_ino;
-       btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
-       key.offset = 0;
+       if (unlikely(inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
+               memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
+       } else {
+               key.objectid = inode->i_ino;
+               btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
+               key.offset = 0;
+       }
+
+       if (unlikely(inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
+               ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
+                                        key.objectid, root->root_key.objectid,
+                                        parent_inode->i_ino,
+                                        index, name, name_len);
+       } else if (add_backref) {
+               ret = btrfs_insert_inode_ref(trans, root,
+                                            name, name_len, inode->i_ino,
+                                            parent_inode->i_ino, index);
+       }
 
-       ret = btrfs_insert_dir_item(trans, root, name, name_len,
-                                   parent_inode->i_ino,
-                                   &key, btrfs_inode_type(inode),
-                                   index);
        if (ret == 0) {
-               if (add_backref) {
-                       ret = btrfs_insert_inode_ref(trans, root,
-                                                    name, name_len,
-                                                    inode->i_ino,
-                                                    parent_inode->i_ino,
-                                                    index);
-               }
+               ret = btrfs_insert_dir_item(trans, root, name, name_len,
+                                           parent_inode->i_ino, &key,
+                                           btrfs_inode_type(inode), index);
+               BUG_ON(ret);
+
                btrfs_i_size_write(parent_inode, parent_inode->i_size +
                                   name_len * 2);
                parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
@@ -4800,31 +4940,29 @@ static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
 {
        struct btrfs_trans_handle *trans;
        struct btrfs_root *root = BTRFS_I(old_dir)->root;
+       struct btrfs_root *dest = BTRFS_I(new_dir)->root;
        struct inode *new_inode = new_dentry->d_inode;
        struct inode *old_inode = old_dentry->d_inode;
        struct timespec ctime = CURRENT_TIME;
        u64 index = 0;
+       u64 root_objectid;
        int ret;
 
-       /* we're not allowed to rename between subvolumes */
-       if (BTRFS_I(old_inode)->root->root_key.objectid !=
-           BTRFS_I(new_dir)->root->root_key.objectid)
+       /* we only allow rename subvolume link between subvolumes */
+       if (old_inode->i_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
                return -EXDEV;
 
-       if (S_ISDIR(old_inode->i_mode) && new_inode &&
-           new_inode->i_size > BTRFS_EMPTY_DIR_SIZE) {
+       if (old_inode->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
+           (new_inode && new_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID))
                return -ENOTEMPTY;
-       }
 
-       /* to rename a snapshot or subvolume, we need to juggle the
-        * backrefs.  This isn't coded yet
-        */
-       if (old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
-               return -EXDEV;
+       if (S_ISDIR(old_inode->i_mode) && new_inode &&
+           new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
+               return -ENOTEMPTY;
 
        ret = btrfs_check_metadata_free_space(root);
        if (ret)
-               goto out_unlock;
+               return ret;
 
        /*
         * we're using rename to replace one file with another.
@@ -4837,6 +4975,9 @@ static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
 
        trans = btrfs_start_transaction(root, 1);
 
+       if (dest != root)
+               btrfs_record_root_in_trans(trans, dest);
+
        /*
         * make sure the inode gets flushed if it is replacing
         * something.
@@ -4846,18 +4987,22 @@ static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
                btrfs_add_ordered_operation(trans, root, old_inode);
        }
 
-       /*
-        * this is an ugly little race, but the rename is required to make
-        * sure that if we crash, the inode is either at the old name
-        * or the new one.  pinning the log transaction lets us make sure
-        * we don't allow a log commit to come in after we unlink the
-        * name but before we add the new name back in.
-        */
-       btrfs_pin_log_trans(root);
+       if (old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID) {
+               /* force full log commit if subvolume involved. */
+               root->fs_info->last_trans_log_full_commit = trans->transid;
+       } else {
+               /*
+                * this is an ugly little race, but the rename is required
+                * to make sure that if we crash, the inode is either at the
+                * old name or the new one.  pinning the log transaction lets
+                * us make sure we don't allow a log commit to come in after
+                * we unlink the name but before we add the new name back in.
+                */
+               btrfs_pin_log_trans(root);
+       }
 
        btrfs_set_trans_block_group(trans, new_dir);
 
-       btrfs_inc_nlink(old_dentry->d_inode);
        old_dir->i_ctime = old_dir->i_mtime = ctime;
        new_dir->i_ctime = new_dir->i_mtime = ctime;
        old_inode->i_ctime = ctime;
@@ -4865,47 +5010,58 @@ static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
        if (old_dentry->d_parent != new_dentry->d_parent)
                btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
 
-       ret = btrfs_unlink_inode(trans, root, old_dir, old_dentry->d_inode,
-                                old_dentry->d_name.name,
-                                old_dentry->d_name.len);
-       if (ret)
-               goto out_fail;
+       if (unlikely(old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
+               root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
+               ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
+                                       old_dentry->d_name.name,
+                                       old_dentry->d_name.len);
+       } else {
+               btrfs_inc_nlink(old_dentry->d_inode);
+               ret = btrfs_unlink_inode(trans, root, old_dir,
+                                        old_dentry->d_inode,
+                                        old_dentry->d_name.name,
+                                        old_dentry->d_name.len);
+       }
+       BUG_ON(ret);
 
        if (new_inode) {
                new_inode->i_ctime = CURRENT_TIME;
-               ret = btrfs_unlink_inode(trans, root, new_dir,
-                                        new_dentry->d_inode,
-                                        new_dentry->d_name.name,
-                                        new_dentry->d_name.len);
-               if (ret)
-                       goto out_fail;
+               if (unlikely(new_inode->i_ino ==
+                            BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
+                       root_objectid = BTRFS_I(new_inode)->location.objectid;
+                       ret = btrfs_unlink_subvol(trans, dest, new_dir,
+                                               root_objectid,
+                                               new_dentry->d_name.name,
+                                               new_dentry->d_name.len);
+                       BUG_ON(new_inode->i_nlink == 0);
+               } else {
+                       ret = btrfs_unlink_inode(trans, dest, new_dir,
+                                                new_dentry->d_inode,
+                                                new_dentry->d_name.name,
+                                                new_dentry->d_name.len);
+               }
+               BUG_ON(ret);
                if (new_inode->i_nlink == 0) {
                        ret = btrfs_orphan_add(trans, new_dentry->d_inode);
-                       if (ret)
-                               goto out_fail;
+                       BUG_ON(ret);
                }
-
        }
        ret = btrfs_set_inode_index(new_dir, &index);
-       if (ret)
-               goto out_fail;
+       BUG_ON(ret);
 
-       ret = btrfs_add_link(trans, new_dentry->d_parent->d_inode,
-                            old_inode, new_dentry->d_name.name,
+       ret = btrfs_add_link(trans, new_dir, old_inode,
+                            new_dentry->d_name.name,
                             new_dentry->d_name.len, 1, index);
-       if (ret)
-               goto out_fail;
+       BUG_ON(ret);
 
-       btrfs_log_new_name(trans, old_inode, old_dir,
-                                      new_dentry->d_parent);
-out_fail:
+       if (old_inode->i_ino != BTRFS_FIRST_FREE_OBJECTID) {
+               btrfs_log_new_name(trans, old_inode, old_dir,
+                                  new_dentry->d_parent);
+               btrfs_end_log_trans(root);
+       }
 
-       /* this btrfs_end_log_trans just allows the current
-        * log-sub transaction to complete
-        */
-       btrfs_end_log_trans(root);
        btrfs_end_transaction_throttle(trans, root);
-out_unlock:
+
        return ret;
 }