WebWorker 封装 JavaScript 沙箱详情

时间：2022-11-13 10:14:24|栏目：JavaScript代码|点击：次

1、场景

在前文 quickjs 封装 JavaScript 沙箱详情已经基于 quickjs 实现了一个沙箱，这里再基于 web worker 实现备用方案。如果你不知道 web worker 是什么或者从未了解过，可以查看 Web Workers API。简而言之，它是一个浏览器实现的多线程，可以运行一段代码在另一个线程，并且提供与之通信的功能。

2、实现 IJavaScriptShadowbox

事实上，web worker 提供了 event emitter 的 api，即 postMessage/onmessage，所以实现非常简单。

实现分为两部分，一部分是在主线程实现 IJavaScriptShadowbox，另一部分则是需要在 web worker 线程实现 IEventEmitter

2.1 主线程的实现

import { IJavaScriptShadowbox } from "./IJavaScriptShadowbox";

export class WebWorkerShadowbox implements IJavaScriptShadowbox {
  destroy(): void {
    this.worker.terminate();
  }

  private worker!: Worker;
  eval(code: string): void {
    const blob = new Blob([code], { type: "application/javascript" });
    this.worker = new Worker(URL.createObjectURL(blob), {
      credentials: "include",
    });
    this.worker.addEventListener("message", (ev) => {
      const msg = ev.data as { channel: string; data: any };
      // console.log('msg.data: ', msg)
      if (!this.listenerMap.has(msg.channel)) {
        return;
      }
      this.listenerMap.get(msg.channel)!.forEach((handle) => {
        handle(msg.data);
      });
    });
  }

  private readonly listenerMap = new Map<string, ((data: any) => void)[]>();
  emit(channel: string, data: any): void {
    this.worker.postMessage({
      channel: channel,
      data,
    });
  }
  on(channel: string, handle: (data: any) => void): void {
    if (!this.listenerMap.has(channel)) {
      this.listenerMap.set(channel, []);
    }
    this.listenerMap.get(channel)!.push(handle);
  }
  offByChannel(channel: string): void {
    this.listenerMap.delete(channel);
  }
}

2.2 web worker 线程的实现

import { IEventEmitter } from "./IEventEmitter";

export class WebWorkerEventEmitter implements IEventEmitter {
  private readonly listenerMap = new Map<string, ((data: any) => void)[]>();

  emit(channel: string, data: any): void {
    postMessage({
      channel: channel,
      data,
    });
  }

  on(channel: string, handle: (data: any) => void): void {
    if (!this.listenerMap.has(channel)) {
      this.listenerMap.set(channel, []);
    }
    this.listenerMap.get(channel)!.push(handle);
  }

  offByChannel(channel: string): void {
    this.listenerMap.delete(channel);
  }

  init() {
    onmessage = (ev) => {
      const msg = ev.data as { channel: string; data: any };
      if (!this.listenerMap.has(msg.channel)) {
        return;
      }
      this.listenerMap.get(msg.channel)!.forEach((handle) => {
        handle(msg.data);
      });
    };
  }

  destroy() {
    this.listenerMap.clear();
    onmessage = null;
  }
}

3、使用 WebWorkerShadowbox/WebWorkerEventEmitter

主线程代码

const shadowbox: IJavaScriptShadowbox = new WebWorkerShadowbox();
shadowbox.on("hello", (name: string) => {
  console.log(`hello ${name}`);
});
// 这里的 code 指的是下面 web worker 线程的代码
shadowbox.eval(code);
shadowbox.emit("open");

web worker 线程代码

const em = new WebWorkerEventEmitter();
em.on("open", () => em.emit("hello", "liuli"));

下面是代码的执行流程示意图；web worker 沙箱实现使用示例代码的执行流程：

4、限制 web worker 全局 api

经大佬 JackWoeker 提醒，web worker 有许多不安全的 api，所以必须限制，包含但不限于以下 api

fetch
indexedDB
performance

事实上，web worker 默认自带了 276 个全局 api，可能比我们想象中多很多。

有篇文章阐述了如何在 web 上通过 performance/SharedArrayBuffer api 做侧信道攻击，即便现在 SharedArrayBuffer api 现在浏览器默认已经禁用了，但天知道还有没有其他方法。所以最安全的方法是设置一个 api 白名单，然后删除掉非白名单的 api。

// whitelistWorkerGlobalScope.ts
/**
 * 设定 web worker 运行时白名单，ban 掉所有不安全的 api
 */
export function whitelistWorkerGlobalScope(list: PropertyKey[]) {
  const whitelist = new Set(list);
  const all = Reflect.ownKeys(globalThis);
  all.forEach((k) => {
    if (whitelist.has(k)) {
      return;
    }
    if (k === "window") {
      console.log("window: ", k);
    }
    Reflect.deleteProperty(globalThis, k);
  });
}

/**
 * 全局值的白名单
 */
const whitelist: (
  | keyof typeof global
  | keyof WindowOrWorkerGlobalScope
  | "console"
)[] = [
  "globalThis",
  "console",
  "setTimeout",
  "clearTimeout",
  "setInterval",
  "clearInterval",
  "postMessage",
  "onmessage",
  "Reflect",
  "Array",
  "Map",
  "Set",
  "Function",
  "Object",
  "Boolean",
  "String",
  "Number",
  "Math",
  "Date",
  "JSON",
];

whitelistWorkerGlobalScope(whitelist);

然后在执行第三方代码前先执行上面的代码

import beforeCode from "./whitelistWorkerGlobalScope.js?raw";

export class WebWorkerShadowbox implements IJavaScriptShadowbox {
  destroy(): void {
    this.worker.terminate();
  }

  private worker!: Worker;
  eval(code: string): void {
    // 这行是关键
    const blob = new Blob([beforeCode + "\n" + code], {
      type: "application/javascript",
    });
    // 其他代码。。。
  }
}

由于我们使用 ts 编写源码，所以还必须将 ts 打包为 js bundle，然后通过 vite 的 ?raw 作为字符串引入，下面吾辈写了一个简单的插件来完成这件事。

import { defineConfig, Plugin } from "vite";
import reactRefresh from "@vitejs/plugin-react-refresh";
import checker from "vite-plugin-checker";
import { build } from "esbuild";
import * as path from "path";

export function buildScript(scriptList: string[]): Plugin {
  const _scriptList = scriptList.map((src) => path.resolve(src));
  async function buildScript(src: string) {
    await build({
      entryPoints: [src],
      outfile: src.slice(0, src.length - 2) + "js",
      format: "iife",
      bundle: true,
      platform: "browser",
      sourcemap: "inline",
      allowOverwrite: true,
    });
    console.log("构建完成: ", path.relative(path.resolve(), src));
  }
  return {
    name: "vite-plugin-build-script",

    async configureServer(server) {
      server.watcher.add(_scriptList);
      const scriptSet = new Set(_scriptList);
      server.watcher.on("change", (filePath) => {
        // console.log('change: ', filePath)
        if (scriptSet.has(filePath)) {
          buildScript(filePath);
        }
      });
    },
    async buildStart() {
      // console.log('buildStart: ', this.meta.watchMode)
      if (this.meta.watchMode) {
        _scriptList.forEach((src) => this.addWatchFile(src));
      }
      await Promise.all(_scriptList.map(buildScript));
    },
  };
}

// https://vitejs.dev/config/
export default defineConfig({
  plugins: [
    reactRefresh(),
    checker({ typescript: true }),
    buildScript([path.resolve("src/utils/app/whitelistWorkerGlobalScope.ts")]),
  ],
});

现在，我们可以看到 web worker 中的全局 api 只有白名单中的那些了。