How does the kernel know which file descriptor to write data to after fork() in a concurrent server?

Question

Basic concurrent client/server architecture: There's a main loop listening for requests on a port (for example 3000), after accepting the connection the server spawns a child process that ends up having access to file descriptors where data can be read.

If we have multiple clients connected to the server, the server will have a child process per request. So S1 (child server process) reads data from C1 (a client), S2 reads from C2 and so on. My question is how is it possible that all clients (C1, C2...) are sending information to the same port (3000) and yet the server processes (S1, S2...) are reading only the information sent from the client assigned to them? Where is the multiplexing being done and how?

Answer 1

A TCP/IP connection has both a source port and a destination port, so if the same server connects to another server on port 3000 multiple times, the Linux kernel can sort out the connections because each one has a unique IP + source port + destination port.

This can be seen with the output of netstat when there are active TCP connections, which shows both the local source port and foreign destination port.

As an aside, doing a fork() for each connection is a really bad idea for a server getting any significant load; fork() is a slow, resource-intensive system call. There's a reason nginx is becoming popular; it uses a difficult-to-program fork()-free model for delivering static content.