@hoppscotch/cli affected by Sandbox Escape in @hoppscotch/js-sandbox leads to RCE

Observations

The Hoppscotch desktop app takes multiple precautions to be secure against arbitrary JavaScript and system command execution. It does not render user-controlled HTML or Markdown, uses Tauri instead of Electron, and sandboxes pre-request scripts with a simple yet secure implementation using web workers.

Unfortunately, web workers are not available in a pure Node.js application like Hoppscotch CLI. That is why the @hoppscotch/js-sandbox package also provides a Javascript sandbox that uses the Node.js vm module. However, the vm module is not safe for sandboxing untrusted Javascript code, as stated in the documentation. This is because code inside the vm context can break out if it can get a hold of any reference to an object created outside of the vm.

In the case of @hoppscotch/js-sandbox, multiple references to external objects are passed into the vm context to allow pre-request scripts interactions with environment variables and more. But this also allows the pre-request script to escape the sandbox.
packages/hoppscotch-js-sandbox/src/pre-request/node-vm/index.ts

const { pw, updatedEnvs } = getPreRequestScriptMethods(envs)

// Expose pw to the context
context.pw = pw
context.atob = atob
context.btoa = btoa

// Run the pre-request script in the provided context
runInContext(preRequestScript, context)

Exploitation

An attacker can use the exposed pw object reference to escape the sandbox and execute arbitrary system commands using the child_process Node.js module. This PoC pre-request script executes the id > /tmp/pwnd system command as soon as a request is sent.

outside = pw.constructor.constructor('return this')()
outside.process.mainModule.require('child_process').execSync('id > /tmp/pwnd')

An attacker who wants to run arbitrary code on the machine of a victim can create a Hoppscotch collection containing a request with a malicious pre-request script and share it with a victim, using the JSON export feature. The victim then has to run the collection with the Hoppscotch CLI. Then the malicious pre-request script executes.

Impact

This attack gives an attacker arbitrary command execution on the machine of a victim Hoppscotch CLI user. For the attack to succeed, an attacker has to lure the victim into downloading a malicious Hoppscotch collection and running it with the Hoppscotch CLI.

This issue does not impact Hoppscotch Web or Desktop, as they use the safe web worker sandboxing approach.

Recommendations

Hoppscotch CLI and other tools that rely on @hoppscotch/js-sandbox but don’t have access to a browser cannot use the web worker sandbox. For these, you can look into other safe JavaScript sandboxing libraries. We think that isolated-vm looks promising. We discourage the use of vm2, which is deprecated because it has arbitrary bypasses. Alternatively, you can introduce an --enable-scripting flag for the CLI and disable scripting by default. Or you can change the threat model and educate users that they should not run untrusted collections as it can lead to RCE.

Differences from existing CVEs

• nvd.nist.gov/vuln/detail/CVE-2023-37466 : This CVE is regarding an escape of vm2 which we do not even use.