Walk through the socket calls for a TCP client and a TCP server.

A client creates a socket with socket(AF_INET, SOCK_STREAM, 0) for TCP, fills in a sockaddr_in with the server's IP and port (port converted with htons), then calls connect(), which performs the three-way handshake; once connected it exchanges data with send()/recv() and calls close() when done. A server creates a socket, calls bind() to claim a local IP and port, listen() to mark it as accepting connections and set a backlog, then accept() which blocks until a client connects and returns a new socket dedicated to that client (the original listening socket keeps accepting more); it then uses send/recv on the per-client socket and closes it when the conversation ends. The key subtlety people miss is that accept() returns a separate connected socket per client, so the server can keep listening on the original while servicing accepted ones (often one per thread/task, or multiplexed with select). UDP skips connect/listen/accept entirely: create a SOCK_DGRAM socket, optionally bind() a local port, and use sendto/recvfrom with the peer address on each call.