The filesystem is a very primitive variation of unix v6 filesystem. It's neither COW (as I've initially planned) nor uses write-ahead log, so it's unreliable on both data and metadata level. Some repair mechanisms could be implemented though (like deallocating inodes which have no links from other nodes), but I didn't bother to implement them. The on-disk representation of the filesystem is following: Pages: |--- 1 --|-- 2 .. n -----|------ n+1 --------|-- n+2 ... maxBlocks --| Data: | Header | Inodes blocks | Free block bitmap | Data | Runtime File data structure is a thin wrapper for Inode. When the File is open, inodes along the whole path to this file are being locked as task requires, but this restriction unnecessary and could be relaxed since we separate links (references to the node from other nodes) and refs (references from runtime objects i.e. ffs.File), and we could easily hold a ref to unlinked file and continue reading it even if it was removed from the file tree. I also didn't implement walkTree iterator for recursive tree walkin since I didn't have much time. Hope the rest works more or less.