graphql-scalars PhoneNumber scalar is vulnerable to regular expression denial of service

Summary

The regular expression behind GraphQLPhoneNumber since 1.24.1 backtracks exponentially. A string made of a run of digits followed by one character that makes the match fail (for example "1".repeat(30) + "x") blocks the JavaScript thread for seconds, and every additional digit doubles the time.

graphql-js calls parseLiteral while validating a document and parseValue while coercing variables, so the regex runs for every request that merely contains such a value in a PhoneNumber position, before execution and before any authentication or authorization step of the server. One small unauthenticated request is enough to stall a server process.

Affected versions

• graphql-scalars >= 1.24.1, <= 2.0.0 (2.0.0 is the current latest; the default branch still has the same pattern).
• Not affected: 1.24.0 and the earlier versions I checked (1.22.4, 1.23.0), which use the linear /^\+[1-9]\d{6,14}$/.
• Introduced by #2719 (c7f58da16, "PhoneNumber scalar is not fully compatible with E.164 format so updated regex to support phone number with extension"), released as 1.24.1 on 2025-01-29.

The payload was ran against 1.22.4, 1.23.0, 1.24.0, 1.24.1, 1.24.2, 1.25.0, 1.26.0 and 2.0.0: the first three reject it instantly, the other five show the exponential growth.

Root cause

src/scalars/PhoneNumber.ts:

const PHONE_NUMBER_REGEX = /^\+?\d{1,3}(\-|\x20)?\(?\d+\)?((\-|\x20)?\d+)+$/;

In \d+\)?((\-|\x20)?\d+)+$ the separator is optional, so a run of n digits can be split across the inner \d+ and the outer + iterations in a number of ways that is exponential in n. When the overall match fails, for instance on a trailing non-digit, the engine tries all of them.

Proof of concept

1. The scalar alone (npm i graphql graphql-scalars@2.0.0):

import { GraphQLPhoneNumber } from 'graphql-scalars';

for (const ok of ['+41791234567', '079 123 45 67', '+1 (123) 456-7890', 'nope']) {
  try { GraphQLPhoneNumber.parseValue(ok); } catch {}
}
for (const digits of [24, 26, 28, 30]) {
  const start = performance.now();
  try { GraphQLPhoneNumber.parseValue('1'.repeat(digits) + 'x'); } catch {}
  console.log(`${digits} digits: ${((performance.now() - start) / 1000).toFixed(2)} s`);
}

Measured warm, i.e. after ordinary inputs have been validated like in a running server:

The time doubles with every digit. The first execution of the regex in a fresh process is about 5x slower because V8 interprets a regexp on its first run (Node 22.19.0, 28 digits: 7.8 s).

2. Through graphql-js, no server and no authentication: a schema with a single echo(phone: PhoneNumber) field, 26 digits + x, Node 22.19.0, warm:

3. Also reproduced end to end against a GraphQL Yoga server: an anonymous HTTP request with a 27-digit value plus one invalid character stalled the whole process (a concurrent trivial { __typename } request waited as long). A mutation that requires authentication was affected equally when called anonymously, because document validation runs before any authorization in resolvers or context.

Impact

Denial of service. Any GraphQL server that exposes PhoneNumber on an argument, an input object field or a variable can be stalled by a request of about 30 characters; roughly 40 characters would keep a process busy for hours (extrapolated from the doubling, not measured). serialize uses the same regex, so unvalidated stored values returned through a PhoneNumber field are affected too. The actual severity depends on the scalar being applied to untrusted input, which is the typical case.

Suggested fix

An equivalent pattern without ambiguous splits:

const PHONE_NUMBER_REGEX = /^\+?\d{1,3}[- ]?\(?\d(?:\d*\))?(?:[- ]?\d)+$/;

It accepts exactly the same strings as the current one, so there is no behaviour change: with N = (S?\d+)+ = (S?\d)+ a digit run is just repeated single-digit iterations, hence \d+\)?N = \d(\d*\))?N. Verification:

• exhaustive comparison of old and new pattern on every string up to length 9 over 1 + - space ( ) x: 47,079,208 strings (1,832 accepted), 0 differences;
• every sequence of up to 6 tokens from 1 11 111 1111 11111 + - space ( ) x: 1,948,717 strings (100,380 accepted), 0 differences;
• adversarial inputs (digit runs plus a failing character, digit/space and digit/dash alternations, parenthesis variants) scale linearly: 1,000,000 characters take about 10 ms, where the current pattern needs 0.3 s for 26.

Optionally also reject values above a sane length before matching; the grammar currently has no upper bound and E.164 allows at most 15 digits. Above roughly 10 million characters V8 aborts the match with a RangeError (stack) after about 25 ms instead of returning false.

A regression test that fails with the current implementation: GraphQLPhoneNumber.parseValue('1'.repeat(40) + 'x') must throw its validation error immediately; with the current pattern it does not return in any reasonable time.

Workarounds for users until a fix is released

• Replace the scalar in the schema with one that uses the pattern above.
• Run Node with --enable-experimental-regexp-engine-on-excessive-backtracks, or call v8.setFlagsFromString('--enable-experimental-regexp-engine-on-excessive-backtracks') before the regex first runs. V8 then falls back to its linear-time engine after excessive backtracking; verified on Node 22.19, 24.15 and 26.10. The flag is not accepted in NODE_OPTIONS.
• Pinning graphql-scalars to <= 1.24.0 restores a linear pattern but changes which values are accepted (only + prefixed E.164).

Coordination

We have not disclosed this publicly. We maintain Unchained Engine (@unchainedshop/api), a downstream project that exposes this scalar on public input fields and will need to ship a mitigation of its own, so we would like to coordinate the timing of any public disclosure with you. Please let us know your preferred timeline, and whether you would like us to prepare a pull request (we can work in a temporary private fork if you enable one).