Virtual TCP/IP stack that can run anywhere (browser, Node.js, Deno, Bun, etc).
- Portable: User-space network stack implemented on top of
lwIP+ WASM - Tun/Tap: L3 and L2 hooks using virtual
TunInterfaceandTapInterface - TCP API: Establish TCP connections over the virtual network stack using clients and servers
- Cross platform: Built on web standard APIs (
ReadableStream,WritableStream, etc) - Lightweight: Less than 100KB
- Fast: Over 500Mbps between stacks
Originally built to communicate with in-browser VMs. Projects like v86 allow you to run a full operating system (like Linux) directly in the browser, which also means you can run Node.js, Postgres, Nginx, or literally any other app in the browser. One of the biggest challenges though, is communicating with this guest OS from the JavaScript host.
With desktop VMs like VMWare, you simply talk to the guest over a network bridge: ethernet frames are forwarded from the VM's virtual NIC to your host's network stack and vice versa. In the browser though, there is no "host network stack" to send frames to - you're stuck with just the ethernet frames. We need a way to communicate at the ethernet (L2) level.
tcpip.js implements the entire network stack in user space and provides APIs to send and receive messages at each layer of the stack (L2, L3, L4). All APIs are built on web standards like ReadableStream, WritableStream, AsyncIterator, etc, so it works on any modern JS runtime (browser, Node.js, Deno, Bun, etc). This library will probably feel similar to Deno's network APIs which strive to be web compliant.
It is implemented on top of lwIP compiled to WASM.
Lightweight IP (lwIP) is a widely adopted network stack written in C. It's primarily used in embedded environments, like in ESP8266 and ESP32 chips, but is also used in unikernels like Unikraft. If you have a smart WiFi device in your home, there's a good chance it's running lwIP!
Because lwIP is both widely adopted and designed for embedded systems, it means that it's battle tested and also lightweight. These are perfect qualities for a WASM lib.
tcpip.js was actually originally written in Go on top of gvisor's tcpip stack, but was later rewritten in C using lwIP to be smaller and faster. It also avoids bundling Go's runtime into the compiled WASM file, which by itself is 1-2MB. lwIP on the other hand is less than 100KB.
NPM
npm i tcpipYarn
yarn add tcpipPNPM
pnpm add tcpipStart by creating a NetworkStack:
import { createStack } from 'tcpip';
const stack = await createStack();Then add a virtual network interface:
const tapInterface = await stack.createTapInterface({
mac: '01:23:45:67:89:ab',
ip: '192.168.1.1/24',
});In this example, we create a tap interface with a MAC address of 01:23:45:67:89:ab and an IP address of 192.168.1.1 on a /24 subnet (192.168.1.0 - 192.168.1.255). For more info on tap and other types of interfaces, see Network interfaces.
Note: this interface is completely virtual within your JS runtime so does not create a real tap interface in your OS.
Next we'll pipe outbound ethernet frames from the tap interface to the VM's virtual NIC (and vice versa):
import { createV86NetworkStream } from '@tcpip/v86';
// ...
const emulator = new V86();
const vmNic = createV86NetworkStream(emulator);
// Forward frames between the tap interface and the VM's NIC
tapInterface.readable.pipeTo(vmNic.writable);
vmNic.readable.pipeTo(tapInterface.writable);This is the virtual equivalent to connecting a patch cable between two physical NICs.
Now that the plumbing is in place, we can start sending TCP packets between our NetworkStack and the VM. Let's assume the VM has an IP address of 192.168.1.2 and is running a TCP server that is listening on port 80.
From our NetworkStack, establish an outbound TCP connection destined to the TCP server running in the VM:
const connection = await stack.connectTcp({
host: '192.168.1.2',
port: 80,
});This method resolves a TcpConnection once the connection is established. It exposes a ReadableStream and WritableStream that you can use to send and receive data over the connection. For more info, see TcpConnection.
Let's send and receive data over the TcpConnection:
const writer = connection.writable.getWriter();
// Send data
await writer.write(new TextEncoder().encode('Hello, world!'));
await writer.close();
// Listen for incoming data
for await (const chunk of connection) {
console.log(new TextDecoder().decode(chunk));
}You can also create a TCP server that listens for incoming connections:
const listener = await stack.listenTcp({
port: 80,
});This method resolves a TcpListener that you can use to accept incoming connections.
// TcpListener is an async iterable that yields TcpConnections
for await (const connection of listener) {
const writer = connection.writable.getWriter();
// Send data
await writer.write(new TextEncoder().encode('Hello, world!'));
await writer.close();
// Listen for incoming data
for await (const chunk of connection) {
console.log(new TextDecoder().decode(chunk));
}
}For more info, see TcpListener.
3 types of interfaces are available:
- Loopback: Loop packets back onto itself (ie.
localhost) - Tun: Hook into IP packets (L3)
- Tap: Hook into ethernet frames (L2)
These interfaces are designed to resemble their counterparts in a traditional host network stack.
A loopback interface simply forwards packets back on to itself. It's akin to 127.0.0.1 (localhost) on a traditional network stack.
const loopbackInterface = await stack.createLoopbackInterface({
ip: '127.0.0.1/8',
});Note that NetworkStack will automatically create a single loopback interface with the above configuration by default. If you prefer to manage all loopback interfaces manually, you can disable the default loopback interface:
const stack = await createStack({
initializeLoopback: false,
});Loopback interfaces are useful when you want to both listen for and establish TCP connections on the same virtual stack without needing to forward packets to a real network interface.
const listener = await stack.listenTcp({
port: 80,
});
const connection = await stack.connectTcp({
host: '127.0.0.1',
port: 80,
});You can create as many loopback interfaces as you wish.
A tun interface hooks into inbound and outbound IP packets (L3).
const tunInterface = await stack.createTunInterface({
ip: '192.168.1.1/24',
});It exposes a ReadableStream and WritableStream as the underlying APIs to send and receive IP packets. It also implements the async iterable protocol for convenience.
interface TunInterface {
readable: ReadableStream<Uint8Array>;
writable: WritableStream<Uint8Array>;
listen(): AsyncIterableIterator<Uint8Array>;
[Symbol.asyncIterator](): AsyncIterableIterator<Uint8Array>;
}Use a TunInterface to forward IP packets to another device that also communicates over IP (L3). In practice tun interfaces are most often used to implement VPNs. If instead you're looking to forward ethernet (L2) frames to a virtual NIC (like v86), use a TapInterface.
In the case of a VPN, would typically pipe the packet streams over another transport and vice versa:
// Connect the tun interface with some transport
tunInterface.readable.pipeTo(someTransport.writable);
someTransport.readable.pipeTo(tunInterface.writable);Important: Tun interfaces will only listen for IP packets after you explicitly start listening (ie. by locking the readable stream). The following methods will lock the readable stream and begin buffering packets:
tunInterface.listen()for await (const packet of tunInterface) { ... }tunInterface.readable.getReader()tunInterface.readable.pipeThrough()tunInterface.readable.pipeTo()tunInterface.readable.tee()
The reason for this is that, unlike TCP, raw IP packets have no form of flow control (back pressure) and buffering packets without a reader will result in memory exhaustion. If you plan to hook into IP packets, be sure to lock the stream before sending data on the stack, otherwise packets will be dropped. Then once listening begins, be sure to regularly read packets to avoid memory exhaustion.
const tunInterface = await stack.createTunInterface({
ip: '192.168.1.1/24',
});
// First call `pipeTo()` to begin listening
tunInterface.readable.pipeTo(vmNic.writable);
vmNic.readable.pipeTo(tunInterface.writable);
// Then send data through the stack (like TCP)
const connection = await stack.connectTcp({
host: '192.168.1.2',
port: 80,
});
...You can create as many tun interfaces as you wish.
A tap interface hooks into inbound and outbound ethernet frames (L2).
const tapInterface = await stack.createTapInterface({
mac: '01:23:45:67:89:ab',
ip: '196.168.1.1/24',
});It exposes a ReadableStream and WritableStream as the underlying APIs to send and receive ethernet frames. It also implements the async iterable protocol for convenience.
interface TapInterface {
readable: ReadableStream<Uint8Array>;
writable: WritableStream<Uint8Array>;
listen(): AsyncIterableIterator<Uint8Array>;
[Symbol.asyncIterator](): AsyncIterableIterator<Uint8Array>;
}Use a TapInterface to forward ethernet frames to another device that also communicates over ethernet (L2). You would typically use this to forward ethernet frames to a virtual NIC (like v86):
import { createV86NetworkStream } from '@tcpip/v86';
// ...
const emulator = new V86();
const vmNic = createV86NetworkStream(emulator);
// Connect the tap interface with the VM's virtual NIC
tapInterface.readable.pipeTo(vmNic.writable);
vmNic.readable.pipeTo(tapInterface.writable);TapInterface has full ARP support, so it will both respond to ARP requests and send ARP requests for unknown MAC addresses.
Important: Tap interfaces will only listen for ethernet frames after you explicitly start listening (ie. by locking the readable stream). The following methods will lock the readable stream and begin buffering frames:
tapInterface.listen()for await (const frame of tapInterface) { ... }tapInterface.readable.getReader()tapInterface.readable.pipeThrough()tapInterface.readable.pipeTo()tapInterface.readable.tee()
The reason for this is that, unlike TCP, raw ethernet frames have no form of flow control (back pressure) and buffering frames without a reader will result in memory exhaustion. If you plan to hook into ethernet frames, be sure to lock the stream before sending data on the stack, otherwise frames will be dropped. Then once listening begins, be sure to regularly read frames to avoid memory exhaustion.
const tapInterface = await stack.createTapInterface({
mac: '01:23:45:67:89:ab',
ip: '196.168.1.1/24',
});
// First call `pipeTo()` to begin listening
tapInterface.readable.pipeTo(vmNic.writable);
vmNic.readable.pipeTo(tapInterface.writable);
// Then send data through the stack (like TCP)
const connection = await stack.connectTcp({
host: '192.168.1.2',
port: 80,
});
...You can create as many tap interfaces as you wish.
Looking for another type of interface? See Future plans.
You can remove any network interface from the stack by calling removeInterface():
await stack.removeInterface(tapInterface);You can retrieve all interfaces on the stack via the interfaces property:
const allInterfaces = stack.interfaces;The TCP API allows you to establish TCP connections over the virtual network stack using clients and servers.
To establish an outbound TCP connection, call connectTcp():
const connection = await stack.connectTcp({
host: '192.168.1.2',
port: 80,
});connectTcp() returns a Promise<TcpConnection> that resolves once the connection is established. See TcpConnection.
Note that DNS resolution is not yet supported, so you must provide the IP address of the host you wish to connect to. See Future plans.
To create a TCP server that listens for incoming connections, call listenTcp():
const listener = await stack.listenTcp({
port: 80,
});listenTcp() returns a Promise<TcpListener> that resolves once the server is listening. See TcpListener.
A TcpListener is an async iterable that yields TcpConnections:
interface TcpListener {
[Symbol.asyncIterator](): AsyncIterableIterator<TcpConnection>;
}You can accept incoming connections by iterating over the TcpListener using the for await syntax:
for await (const connection of listener) {
// Process incoming connection
}This should feel similar to Deno's TCP listener.
A TcpConnection represents an established TCP connection. It exposes a ReadableStream and WritableStream as the underlying APIs to send and receive data. It also implements the async iterable protocol for convenience.
interface TcpConnection {
readable: ReadableStream<Uint8Array>;
writable: WritableStream<Uint8Array>;
close(): Promise<void>;
[Symbol.asyncIterator](): AsyncIterableIterator<Uint8Array>;
}You would typically read incoming data by iterating over the TcpConnection using the for await syntax:
for await (const chunk of connection) {
console.log(new TextDecoder().decode(chunk));
}But you can also read data from the readable stream directly by acquiring a reader:
const reader = connection.readable.getReader();
// Read data
const { value, done } = await reader.read();To send data, you would typically acquire a writer from the writable stream:
const writer = connection.writable.getWriter();
// Send data
await writer.write(new TextEncoder().encode('Hello, world!'));As with any web stream, you can pipe data through a transform stream:
const decompressedStream = connection.readable.pipeThrough(
new DecompressionStream('gzip')
);Or pipe it to a writable stream:
connection.readable.pipeTo(someWritableStream);You could, for example, build an echo server by piping the connection's readable stream back onto its own writable stream:
connection.readable.pipeTo(connection.writable);In practice you might use piping to connect TCP streams to a higher-level protocol library or to proxy connections over another transport.
To close the connection, call close():
await connection.close();- HTTP API
- UDP API
- ICMP (ping) API
- DHCP API
- DNS API
- mDNS API
- Hosts file
- Bridge interface
- Experimental Wireguard interface
- Node.js net polyfill
- Deno net polyfill
MIT