feat(kernel): added unix domain socket subsystem with socketpair, bind/listen/accept/connect, and fcntl by FlareCoding · Pull Request #60 · FlareCoding/StelluxOS

FlareCoding · 2026-03-02T07:28:25Z

Note

High Risk
Adds a new AF_UNIX socket subsystem and multiple syscall handlers, plus changes the core resource I/O ABI to thread per-handle flags (e.g., O_NONBLOCK) through read/write paths. This touches kernel synchronization, handle lifecycle, and VFS interactions, so regressions could affect basic I/O and syscall behavior.

Overview
Adds a new in-kernel UNIX domain stream socket implementation (ring-buffer backed) and wires it into the syscall layer with socket, socketpair, bind, listen, accept, and connect.

Extends the VFS/resource layers to support sockets end-to-end: introduces fs::node_type::socket and fs::socket_node, adds fs::resolve_parent_path and directory create_socket(), and adds a resource_type::SOCKET with new socket resource ops.

Introduces per-handle flags in the handle table and threads them through resource read/write ops; adds a minimal fcntl implementation (F_GETFL/F_SETFL) to control O_NONBLOCK/O_APPEND, plus new error mappings (EAGAIN, EPIPE, EISCONN, etc.).

^{Written by Cursor Bugbot for commit 399e4b6. This will update automatically on new commits. Configure here.}

…d/listen/accept/connect, and fcntl

kernel/resource/handle_table.cpp

userland/apps/init/src/init.c

- alloc_handle: explicitly zero entry.flags on slot reuse to prevent stale O_NONBLOCK from leaking across handle lifetimes - init socketpair demo: guard buf[nr] null-termination with nr >= 0 check to prevent stack buffer underflow when read() returns -1 Co-authored-by: Albert Slepak <FlareCoding@users.noreply.github.com>

kernel/syscall/handlers/sys_socket.cpp

Retrieve per-handle flags via get_handle_object and return EAGAIN immediately when O_NONBLOCK is set and the accept queue is empty, consistent with the read/write paths. Co-authored-by: Albert Slepak <FlareCoding@users.noreply.github.com>

kernel/common/string.cpp

… impl Co-authored-by: Albert Slepak <FlareCoding@users.noreply.github.com>

kernel/syscall/handlers/sys_socket.cpp

…istener - connect: move client_sock mutation before wake_one and insert smp_write barrier so AArch64 readers never see CONNECTED state with a null channel pointer - accept: check ls->closed before returning EAGAIN on nonblock path to avoid infinite retry loops on a permanently closed listener Co-authored-by: Albert Slepak <FlareCoding@users.noreply.github.com>

cursor

Cursor Bugbot has reviewed your changes and found 1 potential issue.

^{Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, have a team admin enable autofix in the Cursor dashboard.}

kernel/syscall/handlers/sys_socket.cpp

EISCONN only for CONNECTED, EINVAL for LISTENING, and allow BOUND sockets to proceed with connect (matching Linux semantics). Co-authored-by: Albert Slepak <FlareCoding@users.noreply.github.com>

feat(kernel): added unix domain socket subsystem with socketpair, bin…

0f82b11

…d/listen/accept/connect, and fcntl