Initial commit: Schunix as it stands before the SchOS rename

An x86-64 operating system built from xv6, now largely rewritten: SMP,
a copy-on-write root filesystem (SchFS), PCI/USB/virtio drivers, a TTF
rasteriser, and a Linux-numbered syscall ABI.

No history before this point -- the tree was never under version control.
This commit exists so the rename that follows has something to undo to.

include/socket.h

@@ -0,0 +1,81 @@
+#pragma once
+
+#include "types.h"
+
+// Unix domain sockets.
+//
+// These exist for Schcom. A compositor and its clients have to exchange
+// messages, and a pipe cannot do it: a pipe is one-directional, has no
+// notion of a connection, and gives no way for a server to be *found* by a
+// client that starts later. A socket with a name in the filesystem is
+// exactly the missing piece.
+//
+// AF_INET is deliberately absent. The socket layer is being built around the
+// local case first because that is the one the GUI needs and the one whose
+// semantics are simplest; a network stack plugs in underneath the same
+// syscalls later rather than reshaping them.
+
+#define AF_UNIX         1
+#define AF_LOCAL        1
+#define AF_INET         2       // recognised so it can be refused honestly
+
+#define SOCK_STREAM     1
+#define SOCK_DGRAM      2
+
+#define UNIX_PATH_MAX   108
+
+#define SHUT_RD         0
+#define SHUT_WR         1
+#define SHUT_RDWR       2
+
+struct sockaddr_un {
+  uint16 sun_family;
+  char   sun_path[UNIX_PATH_MAX];
+};
+
+// The generic form syscalls take. Same shape as Linux's, so a musl-linked
+// program's struct lines up byte for byte.
+struct sockaddr {
+  uint16 sa_family;
+  char   sa_data[UNIX_PATH_MAX];
+};
+
+#ifdef _KERNEL_SOCKET_INTERNALS
+
+#define NSOCK       32          // sockets in the system
+#define SOCKBUF     4096        // receive buffer per connected socket
+#define NBACKLOG    8
+
+enum sockstate {
+  SS_FREE = 0,
+  SS_UNBOUND,
+  SS_BOUND,
+  SS_LISTENING,
+  SS_CONNECTED,
+};
+
+// Every field here is protected by the socket table's single lock in
+// kernel/net/unix.c -- there is deliberately no per-socket lock; see the
+// comment at the top of that file for why.
+struct socket {
+  int   ref;
+  int   type;
+  int   state;
+
+  char  name[UNIX_PATH_MAX];    // the bound path, for a listener
+
+  // A connected socket's peer. Writing to a socket appends to the *peer's*
+  // buffer, which is what makes one buffer per socket enough for a
+  // bidirectional connection.
+  struct socket *peer;
+  int   peer_closed;
+
+  char  data[SOCKBUF];
+  uint  nread, nwrite;
+
+  // Connections waiting to be accepted, on a listener.
+  struct socket *queue[NBACKLOG];
+  int   nqueue;
+};
+
+#endif