libcurl.js

libcurl.js

This is an experimental port of libcurl to WebAssembly for use in the browser. It provides an interface compatible with the Fetch API, allowing you to proxy HTTPS requests from the browser with full TLS encryption. Unlike previous implementations, the proxy server cannot read the contents of your requests.

Table of Contents:

• Features
• Building
• Javascript API
  • Importing the Library
  • Making HTTP Requests
  • Creating New HTTP Sessions
  • Creating WebSocket Connections
  • Using TLS Sockets
  • Changing the Network Transport
  • Changing the Websocket Proxy URL
  • Getting Libcurl's Output
  • Getting Error Strings
  • Getting Version Info
  • Getting the CA Certificates Bundle
• Proxy Server
• Project Structure
• Copyright
  • Copyright Notice

Table of contents generated with markdown-toc.

Features:

• Fetch compatible API
• End to end encryption between the browser and the destination server
• Support for up to TLS 1.3
• Support for tunneling HTTP/2 connections
• Support for proxying WebSockets
• Bypass CORS restrictions without compromising on privacy
• Low latency via multiplexing and reusing open connections
• Use raw TLS sockets in the browser
• Custom network transport support
• Works inside web workers without needing special permissions or headers
• Works in all major browsers (Chromium >= 64, Firefox >= 65, Safari >= 14)
• Has the ability to create multiple independent sessions
• Small footprint size (800kb after compression) and low runtime memory usage
• Support for running inside a web worker
• Support for Brotli and gzip compressed responses

Building:

You can build this project by running the following commands:

git clone https://github.com/ading2210/libcurl.js --recursive
cd libcurl.js/client
./build.sh

Make sure you have emscripten, git, and the various C build tools installed. The only OS supported for building libcurl.js is Linux. On Debian-based systems, you can run the following command to install all the dependencies:

sudo apt install make cmake emscripten autoconf automake libtool pkg-config wget xxd jq

The build script will generate client/out/libcurl.js as well as client/out/libcurl.mjs, which is an ES6 module. You can supply the following arguments to the build script to control the build:

• release - Use all optimizations.
• single_file - Include the WASM binary in the outputted JS using base64.
• all - Build twice, once normally, and once as a single file.

Javascript API:

Importing the Library:

To import the library, follow the build instructions in the previous section, and copy client/out/libcurl.js and client/out/libcurl.wasm to a directory of your choice. After the script is loaded, call libcurl.load_wasm, specifying the url of the libcurl.wasm file. You do not need to call libcurl.load_wasm if you use the libcurl_full.js file, as the WASM will be bundled into the JS file.

<script defer src="./out/libcurl.js" onload="libcurl.load_wasm('/out/libcurl.wasm');"></script>

Alternatively, prebuilt versions can be found on NPM and jsDelivr. You can use the following URLs to load libcurl.js from a third party CDN.

https://cdn.jsdelivr.net/npm/libcurl.js@latest/libcurl.js
https://cdn.jsdelivr.net/npm/libcurl.js@latest/libcurl.wasm

To know when libcurl.js has finished loading, you can use the libcurl_load DOM event.

document.addEventListener("libcurl_load", ()=>{
  libcurl.set_websocket(`wss://${location.hostname}/ws/`);
  console.log("libcurl.js ready!");
});

You may also use the libcurl.onload callback, which can be useful for running libcurl.js inside a web worker.

libcurl.onload = () => {
  console.log("libcurl.js ready!");
}

Once loaded, there will be a window.libcurl object which includes all the API functions. The libcurl.ready property can also be used to know if the WASM has loaded.

There are also ES6 modules available if you are using a bundler. The libcurl.mjs and libcurl_full.mjs files provide this functionality, with the former being set as the entry point for the NPM package.

//import the regular version
import { libcurl } from "libcurl.js"; 

//import the version with the wasm included in the js
import { libcurl } from "libcurl.js/bundled"; 

Examples of running libcurl.js on the main thread and in a web worker are available at client/index.html and client/worker.html respectively.

Making HTTP Requests:

To perform HTTP requests, use libcurl.fetch, which takes the same arguments as the browser's regular fetch function. Like the standard Fetch API, libcurl.fetch will also return a Response object.

let r = await libcurl.fetch("https://ading.dev");
console.log(await r.text());

Most of the standard Fetch API's features are supported, with the exception of:

• CORS enforcement
• Caching

Sending cookies is supported, but they will not be automatically sent unless you create a new HTTP session, which is covered in the next section.

The response may contain multiple HTTP headers with the same name, which the Headers object isn't able to properly represent. If this matters to you, use response.raw_headers, which is an array of key value pairs, instead of response.headers. There is support for streaming the response body using a ReadableStream, as well as canceling requests using an AbortSignal. All requests made using this method share the same connection pool, which has a limit of 50 active TCP connections.

Creating New HTTP Sessions:

To create new sessions for HTTP requests, use the libcurl.HTTPSession class. The constructor for this class takes the following arguments:

• options - An optional object with various settings.

The valid HTTP session settings are:

• enable_cookies - A boolean which indicate whether or not cookies should be persisted within the session.
• cookie_jar - A string containing the data in the cookie jar file. This should have been exported from a previous session. For more information on the format for this file, see the curl documentation.

Each HTTP session has the following methods available:

• fetch - Identical to the libcurl.fetch function but only creates connections in this session.
• set_connections - Set the connection limits. This takes two arguments, the first being the limit for the connection cache, and the second being the max number of active connections.
• export_cookies - Export any cookies which were recorded in the session. This will return an empty string if cookies are disabled or no cookies have been set yet.
• close - Close all connections and clean up the session. You must call this after you are done using the session, otherwise it will leak memory.

Creating WebSocket Connections:

To use WebSockets, create a libcurl.CurlWebSocket object, which takes the following arguments:

• url - The Websocket URL.
• protocols - A optional list of websocket subprotocols, as an array of strings.
• options - An optional object with extra settings to pass to curl.

The valid WebSocket options are:

• headers - HTTP request headers for the websocket handshake.
• verbose - A boolean flag that toggles the verbose libcurl output. This verbose output will be passed to the function defined in libcurl.stderr, which is console.warn by default.

The following callbacks are available:

• CurlWebSocket.onopen - Called when the websocket is successfully connected.
• CurlWebSocket.onmessage - Called when a websocket message is received from the server. The data is passed to the first argument of the function, and it will be either a Uint8Array or a string, depending on the type of message.
• CurlWebSocket.onclose - Called when the websocket is cleanly closed with no error.
• CurlWebSocket.onerror - Called when the websocket encounters an unexpected error. The error code is passed to the first argument of the function.

The CurlWebSocket.send function can be used to send data to the websocket. The only argument is the data that is to be sent, which must be either a string or a Uint8Array.

You can call CurlWebSocket.close to close and clean up the websocket.

let ws = new libcurl.CurlWebSocket("wss://echo.websocket.org", [], {verbose: 1});
ws.onopen = () => {
  console.log("ws connected!");
  ws.send("hello".repeat(100));
};
ws.onmessage = (data) => {
  console.log(data);
};

You can also use the libcurl.WebSocket object, which works identically to the regular WebSocket object. It uses the same arguments as the simpler CurlWebSocket API.

let ws = new libcurl.WebSocket("wss://echo.websocket.org");
ws.addEventListener("open", () => {
  console.log("ws connected!");
  ws.send("hello".repeat(128));
});
ws.addEventListener("message", (event) => {
  console.log(event.data);
});

Using TLS Sockets:

Raw TLS sockets can be created with the libcurl.TLSSocket class, which takes the following arguments:

• host - The hostname to connect to.
• port - The TCP port to connect to.
• options - An optional object with extra settings to pass to curl.

The valid TLS socket options are:

• verbose - A boolean flag that toggles the verbose libcurl output.

The callbacks work similarly to the libcurl.CurlWebSocket object, with the main difference being that the onmessage callback always returns a Uint8Array.

The TLSSocket.send function can be used to send data to the socket. The only argument is the data that is to be sent, which must be a Uint8Array.

You can use the TLSSocket.close function to close the socket.

let socket = new libcurl.TLSSocket("ading.dev", 443, {verbose: 1});
socket.onopen = () => {
  console.log("socket connected!");
  let str = "GET / HTTP/1.1\r\nHost: ading.dev\r\nConnection: close\r\n\r\n";
  socket.send(new TextEncoder().encode(str));
};
socket.onmessage = (data) => {
  console.log(new TextDecoder().decode(data));
};

Changing the Network Transport:

You can change the underlying network transport by setting libcurl.transport. The following values are accepted:

• "wisp" - Use the Wisp protocol. This is the default and the fastest option, since it multiplexes several TCP connections on the same websocket.
• "wsproxy" - Use the wsproxy protocol, where a new websocket is created for each TCP connection. For example, connecting to wss://example.com/ading.dev:443 would open a new TCP connection to ading.dev:443.
• Any custom class - Use a custom network protocol. If you pass in custom code here, it must be roughly conformant with the standard WebSocket API. The URL that is passed into this fake websocket always looks like "wss://example.com/ws/ading.dev:443", where wss://example.com/ws/ is the proxy server URL, and ading.dev:443 is the destination server.

Changing the Websocket Proxy URL:

You can change the URL of the websocket proxy by using libcurl.set_websocket.

libcurl.set_websocket("ws://localhost:6001/");

If the websocket proxy URL is not set and one of the other API functions is called, an error will be thrown. Note that this URL must end with a trailing slash.

Getting Libcurl's Output:

If you want more information about a connection, you can pass the _libcurl_verbose argument to the libcurl.fetch function. These are the same messages that you would see if you ran curl -v on the command line.

await libcurl.fetch("https://example.com", {_libcurl_verbose: 1});

By default this will print the output to the browser console, but you can set libcurl.stdout and libcurl.stderr to intercept these messages. This callback will be executed on every line of text that libcurl outputs.

libcurl.stderr = (text) => {document.body.innerHTML += text};

Libcurl.js will also output some error messages to the browser console. You can intercept these messages by setting the libcurl.logger callback, which takes two arguments:

• type - The type of message. This will be one of the following: "log", "warn", "error"
• text - The text that is to be logged.

This may be useful if you are running libcurl.js inside a web worker and do not have access to the regular console API.

Getting Error Strings:

Libcurl.js reports errors based on the error codes defined by the libcurl C library. The libcurl.get_error_string function can be used to get an error string from an error code.

console.log(libcurl.get_error_string(56));
//"Failure when receiving data from the peer"

Getting Version Info:

You can get version information from the libcurl.version object. This object will also contain the versions of all the C libraries that libcurl.js uses. libcurl.version.lib returns the version of libcurl.js itself.

Getting the CA Certificates Bundle:

You can get the CA cert bundle that libcurl uses by calling libcurl.get_cacert. The function will return a string with the certificates in PEM format. The cert bundle comes from the official curl website, which is extracted from the Mozilla Firefox source code.

Proxy Server:

The proxy server consists of a standard Wisp server, allowing multiple TCP connections to share the same websocket.

To host the proxy server, run the following commands:

git clone https://github.com/ading2210/libcurl.js --recursive
cd libcurl.js
server/run.sh --static=./client

For a full list of server arguments, see the wisp-server-python documentation.

Project Structure:

• client - Contains all the client-side code.
  • fragments - Various patches for the JS that emscripten produces. The script which does the patching can be found at client/tools/patch_js.py.
  • javascript - All the code for the Javascript API, and for interfacing with the compiled C code.
  • libcurl - The C code that interfaces with the libcurl library and gets compiled by emscripten.
  • tests - Unit tests and the scripts for running them.
  • tools - Helper shell scripts for the build process, and for compiling the various C libraries.
  • wisp_client - A submodule for the Wisp client library.
• server - Contains all the server-side code for running the websocket proxy server.
  • wisp_server - A submodule for the Python Wisp server.

Copyright:

This project is licensed under the GNU AGPL v3.

Copyright Notice:

ading2210/libcurl.js - A port of libcurl to WASM for use in the browser.
Copyright (C) 2023 ading2210

This program is free software: you can redistribute it and/or modify
it under the terms of the GNU Affero General Public License as published by
the Free Software Foundation, either version 3 of the License, or
(at your option) any later version.

This program is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
GNU Affero General Public License for more details.

You should have received a copy of the GNU Affero General Public License
along with this program.  If not, see <https://www.gnu.org/licenses/>.

Changelog

v0.6.0 (3/20/24):

• Refactor JS and C code
• Allow for multiple sessions with separate connection pools
• Switch to wolfSSL instead of OpenSSL for significantly smaller binaries
• Add support for sending cookies automatically
• Add NPM export for bundled version