Writing an OS Kernel from Scratch – Part 25: Unix Domain Sockets — The Foundation of IPC

Writing an OS Kernel from Scratch – Part 25: Unix Domain Sockets — The Foundation of Inter-Process Communication (IPC)

“The Display Server and Client need efficient communication, but pipes are unidirectional and unstructured. Today, we implement Unix Domain Sockets (AF_UNIX), making IPC as natural as file operations!”

In the previous post, we built the Display Server + Client architecture, but the IPC channel was a simplified pipe with clear limitations:

• Only supports unidirectional communication
• No connection concept → can’t distinguish multiple Clients
• No file descriptor passing → can’t share memory handles

Real IPC needs Unix Domain Sockets:
✅ Bidirectional communication
✅ Supports SOCK_STREAM (stream) and SOCK_DGRAM (datagram)
✅ Can pass file descriptors (like shared memory, device handles)
✅ Addressed via path names (like /tmp/display.sock)

D-Bus, Wayland, and X11 are all based on Unix Domain Sockets!

---

Socket VFS Abstraction Layer

Sockets appear in VFS as special files, but operations go through socket system calls.

Key: The fd returned by socket() is essentially a special file!

---

AF_UNIX Protocol Implementation

1. Socket structure
2. bind: bind path name
3. listen: listen for connections
4. accept: accept connections
5. connect: connect to Server

---

Data Transfer: send/recv

1. Kernel message queue
2. sendmsg (supports FD passing)
3. recvmsg (receive FD)

The essence of FD passing: copying a file object from one process’s fd_table to another!