TODO.md

mysqlfs TODO

• Verify whether timestamp updates (ctime, atime, mtime) are always consistent.

• Handle deletion of files that are still in use more safely. Deleting an open file can currently purge data blocks immediately, which does not match normal Unix expectations for deleted-but-still-open files.

• Improve connection-pool robustness. Avoid returning -EMFILE when no pooled connection is immediately available. Also remove the dependency on the deprecated MySQL MYSQL_OPT_RECONNECT option. A better approach would be to treat a dead connection as an operation failure, drop it from the pool, and reopen a fresh connection explicitly on the next checkout instead of relying on transparent reconnect inside the client library.

• Improve security behavior. mysqlfs stores permission metadata, but it still relies mainly on default_permissions for enforcement.

• Add path-to-inode and inode-to-stat caches.

  • getattr() would benefit significantly.
  • query_inode_full() should eventually support walking a path in chunks from an intermediate inode instead of always building one large SQL query from the root.
  • query_getattr() could eventually expose the inode it already resolves so mysqlfs_getattr() can avoid a second path lookup.

• Add file buffering. Running a database query after every write() is still too expensive.

• Add support for xattrs and ACLs. FUSE exposes methods for xattrs, but mysqlfs does not implement them yet.

• Add explicit fallocate() support. Sparse file semantics already work through normal read, write, and truncate operations, but mysqlfs does not currently implement the dedicated FUSE fallocate path.

• Plan a migration to the libfuse 3.x interfaces. This should be treated as an API-compatibility pass rather than just adding new callbacks: existing operations such as getattr(), rename(), readdir(), truncate(), and timestamp handling need to be checked against the newer signatures and semantics.