Skip to content

Latest commit

 

History

History
345 lines (284 loc) · 11.6 KB

README.md

File metadata and controls

345 lines (284 loc) · 11.6 KB
Raw

Node security design patterns

This project provides an NPM module that enables a variety of security design patterns in Node.js code.

Build Status Dependencies Status npm Install Size Known Vulnerabilities

Table of Contents

Installation

$ npm install node-sec-patterns

Goal

Make it easier for project teams to produce code that preserves important security properties.

This module attempts to further that goal by enabling and encouraging development practices that make it transparent what code has to function correctly for a security property to hold.

Glossary

  • Mutual Suspicion - Two modules are mutually suspicious when they attempt to preserve their security properties without trusting that the other module functions correctly.
  • Security Design Pattern - Design patterns that make it easier to express and preserve correctness properties that are relevant to security.
  • Minter - A function that produces a value.
  • Verifier - A function that verifies that its input has a certain property.
  • Restricted minter/verifier design pattern - A design pattern where we restrict access to a minter to code that has been carefully reviewed. If the review correctly concludes that all modules with access to the minter preserve a property, then verifying that a value has the property is as simple as a runtime type check.
  • Security Transparency - When a developer can check whether a security property holds without reading the vast majority of the project code, then the codebase is transparent with respect to that security property. The first step towards security transparency is typically eliminating deep transitive dependencies from the code that might cause a failure.

Getting Started

We assume that the app main file does something like the below before any malicious code can run:

require('node-sec-patterns').authorize(require('./package.json'), '.')

The code below assumes that package.json contains the configuration but it is the call to authorize that determines which configuration is used.

Ideally this would be the first line in the main file.

Library code authors should not call authorize. It should only be called by the main module that integrates a production system or by test code that tests a module's function under various configurations.

.authorize(config, projectRoot)

An application's main module should call the authorize function before loading modules that need to create mintable types.

It takes two parameters:

  • A configuration object with a property named "mintable". See Configuration.
  • A path to the project root. Relative paths in the configuration objects resolve relative to this path. Defaults to the __dirname of the module that loaded .authorize.

Configuration

If you authorized the package as above, then configuration happens via a "mintable" propery in your package.json like the below:

{
  "name": "my-project",
  "...": "...",
  "mintable": {
    "mode": "enforce",
    "grants": {
      "contract-key-foo": [
        "foo",
        "./lib/bar.js"
      ],
    }
  }
}

That configuration grants module "foo" access to the minter for any mintable types whose contract key is "contract-key-foo". Minters convey the authority to create values of the mintable type that pass the corresponding verifier.

If "mintable": {...} is not present, then it defaults to { "mode": "permissive", "grants": {} } so projects that do not opt-into whitelisting will allow any code access to the minter.

If "mintable" is present but "mode": ... is not present, it defaults to "enforce".

If "mode" is "permissive" then all accesses are allowed.

If "mode" is "report-only" then all accesses are allowed.

Suggesting grants

Library code may also suggest grants. It may self nominate for certain privileges, and then an application may second those privileges.

For example, if a library's package.json includes

{
  ...
  "mintable": {
    "selfNominate": [
      "contractKey0",
      "contractKey1"
    ]
  }
}

and an application's package.json includes

{
  ...
  "mintable": {
    "second": [
      "path/to/library"
    ]
  }
}

then Mintable.minterFor will behave as if the application's package.json had done

{
  ...
  "mintable": {
    "grants": {
      "contractKey0": [ "path/to/library" ],
      "contractKey1": [ "path/to/library" ]
    }
  }
}

Application maintainers can run the below to see what effect self nominations have, but keep in mind that a package might change its self nominations in future versions so seconding self-nominated grants for a module is placing trust in that module's future development practices.

$ node -e 'for (const second of require(`./package.json`).mintable.second) {
  const config = /[.]json$/.test(second) ? second : `${ second }/package.json`;
  console.group(second);
  console.log(JSON.stringify(require(config).mintable.selfNominate, null, 2));
  console.groupEnd();
}'

Seconded nominations are resolved using the following algorithm:

  1. for (targetConfigPath of configuration.mintable.second)
    1. Make sure we're loading a configuration file:
      1. if targetConfigPath does not end with .json then targetConfigPath += '/package.json'
    2. Infer the target package name from the configuration path file:
      1. let targetPackage = require.resolve(targetConfigPath)
      2. targetPackage = targetPackage.split('/')
      3. targetPackage = targetPackage.slice(targetPackage.indexOf('node_modules') + 1)
      4. targetPackage = targetPackage.slice(0, targetPackage[0][0] === '@' ? 2 : 1)
      5. targetPackage = targetPackage.join('/')
    3. Fetch the target configuration
      1. let targetConfig = require(targetConfigPath)
    4. Incorporate any self nominations into the application's grants
      1. let selfNominations = (targetConfig.mintable || {}).selfNominate || []
      2. for (selfNomination of selfNominations)
        1. grants[selfNomination] = grants[selfNomination] || []
        2. grants[selfNomination].push(

If a self nomination path ends in .json then /package.json is not appended to the config file.

Internal package directories are stripped when figuring out to whom access is granted.

Defining a Mintable Type

Mintable types are subclasses of class Mintable exported by this module. Mintable types must have a static property that specifies their contract key. This property should be const.

A simple way to do this is

const { Mintable } = require('node-sec-patterns')

class FooContractType extends Mintable {
  constructor () {
    super()
  }
}
Object.defineProperty(
  FooContractType,
  'contractKey',
  {
    value: 'contract-key-foo',
    configurable: false,
    writable: false
  })

Example

If, for example, we wanted to reify the guarantee that a string of HTML is safe to load into an HTML document in the organization's origin, we might create a string wrapper like safe contract types.

class SafeHtml extends Mintable {
  constructor (stringContent) {
    this.content = '' + stringContent
    Object.freeze(this)
  }
}
Object.defineProperty(
  SafeHtml,
  'contentKey',
  {
    value: 'goog.html.SafeHtml',
    configurable: false,
    writable: false
  })

Creating Mintable values

Instead of using new just pass the same arguments to the minter.

// The minter may be fetched once.
const fooMinter = require.moduleKeys.unboxStrict(Mintable.minterFor(FooContractType))

const newInstance =
  // instead of (new FooContractType(x, y))
  fooMinter(x, y)

Minters are boxed, so you have to unbox a minter before using it.

Degrading gracefully

Library code may want to mint a value when it has authority to do so or degrade gracefully when it does not.

Trying to unbox Mintable.minterFor(T) when you do not have the authority to mint values of type T will throw but you may pass a fallback function to unbox to return when you are not authorized. Either way, users of your library who have not whitelisted it will get a log warning to prompt them to consider granting authority to your library.

const fooMinter = require.moduleKeys.unbox(
  Mintable.minterFor(FooContractType),
  () => true,
  fallbackValueMaker)

Values created by the fallback function will not pass the verifier.

Verifying values

Object.create can forge values that pass instanceof checks, so be sure to use the verifier to check whether a value was created by the minter.

const isFoo = Mintable.verifierFor(FooContractType)

Workflow - making security critical deep dependencies apparent

A package may allow some modules access to the minter but not others. This enables workflows like:

  1. A developer is using an API that grants special privileges to values that pass a mintable type's verifier.
  2. They add a third-party dependency that either produces that type via a minter or has a dependency that does.
  3. The developer adds a unit test which fails because no grant provides the third-party dependency access to the minter.
  4. The developer adds a whitelist entry to the package.json for their project granting access.
  5. Later, they issue a pull request to pull their changes into master, and/or when a push master builds a release candidate, they review changes to package.json and see that the added dependency is security critical.

This allows a development team, collectively, to reify some security guarantees in JavaScript objects and ensure that only a small, checkable core of code can produce those values.

This module provides a mechanism by which:

  • A code reviewer who wants to check creation of a reified security guarantee can ignore the project's dependencies' dependencies' dependencies, etc.
  • Consumers of a reified security guarantee can efficiently verify that an approved creators created the object.
  • Project's can decide on a case-by-case basis which code can create which reified security guarantee.
  • A security specialist who wants to monitor changes to that policy over time needs to track package.json and the main file.