Pure-Elixir codec for the Linux FUSE kernel protocol, targeting FUSE_KERNEL_VERSION 7.31 (the version exposed by libfuse 3.10+ and the Linux 5.4+ kernel).

The codec operates on binaries only — no I/O, no dependency on the /dev/fuse NIF. A typical read path looks like:

{:ok, request_bytes} = Wick.Native.read_frame(fd)

case Wick.Protocol.decode_request(request_bytes) do
  {:ok, :lookup, header, %Wick.Protocol.LookupRequest{name: name}} ->
    reply = %Wick.Protocol.EntryReply{nodeid: ..., attr: ...}
    bytes = Wick.Protocol.encode_response(header.unique, reply, 0)
    :ok = Wick.Native.write_frame(fd, bytes)
  # ...
end

Wire rules

• All multi-byte integers are little-endian. (The kernel UAPI is documented as native-endian, but every Linux platform Wick targets is little-endian.)
• Every struct is padded to an 8-byte boundary. All padding / unused* / dummy fields are wire-significant — encode them as zero; ignore them on decode.
• fuse_in_header.len / fuse_out_header.len is the total message length, header inclusive.
• Error replies carry fuse_out_header only (no body). error is 0 for success or a negative POSIX errno.
• Filenames in request bodies are NUL-terminated.

Supported opcodes

See opcode_to_atom/1 and atom_to_opcode/1 for the supported opcode set (the ~20 opcodes required for the read-path + write-path + metadata operations — INIT, LOOKUP, GETATTR, SETATTR, READDIR, READ, WRITE, OPEN, RELEASE, CREATE, MKDIR, UNLINK, RMDIR, RENAME, STATFS, FLUSH, FSYNC, FORGET, BATCH_FORGET, DESTROY).

Extended opcodes for xattrs (SETXATTR / GETXATTR / LISTXATTR / REMOVEXATTR) are supported as of #671. The lock opcodes (GETLK / SETLK / SETLKW) are decoded as of #672 — handlers route them by lk_flags & FUSE_LK_FLOCK to either FLOCK whole-file logic (#672, #677) or byte-range fcntl (#674, #681). INTERRUPT (cancellation of a queued SETLKW) is supported as of #675. The remaining extended opcodes (IOCTL, POLL) are out of scope.