Security #

Executing arbitrary expressions like enabled by the expression parser of mathjs involves a risk in general. When you’re using mathjs to let users execute arbitrary expressions, it’s good to take a moment to think about possible security and stability implications, especially when running the code server side.

Security risks #

A user could try to inject malicious JavaScript code via the expression parser. The expression parser of mathjs offers a sandboxed environment to execute expressions which should make this impossible. It’s possible though that there are unknown security vulnerabilities, so it’s important to be careful, especially when allowing server side execution of arbitrary expressions.

The expression parser of mathjs parses the input in a controlled way into an expression tree or abstract syntax tree (AST). In a “compile” step, it does as much as possible preprocessing on the static parts of the expression, and creates a fast performing function which can be used to evaluate the expression repeatedly using a dynamically passed scope.

The parser actively prevents access to JavaScripts internal eval and new Function which are the main cause of security attacks. Mathjs versions 4 and newer does not use JavaScript’s eval under the hood. Version 3 and older did use eval for the compile step. This is not directly a security issue but results in a larger possible attack surface.

When running a node.js server, it’s good to be aware of the different types of security risks. The risk when running inside a browser may be limited, though it’s good to be aware of Cross side scripting (XSS) vulnerabilities. A nice overview of the security risks of node.js servers is listed in the article Node.js security checklist by Gergely Nemeth.

Less vulnerable expression parser #

There is a small number of functions which yield the biggest security risk in the expression parser:

To make the expression parser less vulnerable whilst still supporting most functionality, these functions can be disabled:

import { create, all } from 'mathjs'

const math = create(all)
const limitedEvaluate = math.evaluate

math.import({
  // most important (hardly any functional impact)
  'import':     function () { throw new Error('Function import is disabled') },
  'createUnit': function () { throw new Error('Function createUnit is disabled') },
  'reviver':    function () { throw new Error('Function reviver is disabled') },

  // extra (has functional impact)
  'evaluate':   function () { throw new Error('Function evaluate is disabled') },
  'parse':      function () { throw new Error('Function parse is disabled') },
  'simplify':   function () { throw new Error('Function simplify is disabled') },
  'derivative': function () { throw new Error('Function derivative is disabled') },
  'resolve':    function () { throw new Error('Function resolve is disabled') },
}, { override: true })

console.log(limitedEvaluate('sqrt(16)'))     // Ok, 4
console.log(limitedEvaluate('parse("2+3")')) // Error: Function parse is disabled

Found a security vulnerability? Please report in private! #

You found a security vulnerability? Awesome! We hope you don’t have bad intentions and want to help fix the issue. Please report the vulnerability in a private way by contacting one of the maintainers via mail or another private channel. That way we can work together on a fix before sharing the issue with everybody including the bad guys.

Stability risks #

A user could accidentally or on purpose execute a heavy expression like creating a huge matrix. That can let the JavaScript engine run out of memory or freeze it when the CPU goes to 100% for a long time.

To protect against this sort of issue, one can run the expression parser in a separate Web Worker or child_process, so it can’t affect the main process. The workers can be killed when it runs for too long or consumes too much memory. A useful library in this regard is workerpool, which makes it easy to manage a pool of workers in both browser and node.js.

Fork me on GitHub