Nodemailer: Quadratic (O(n^2)) SMTP client reply parsing lets a malicious server or MITM cause a pre-auth remote denial of service

Summary

The SMTP client response parser reassembles multiline server replies by appending each continuation line onto a single response-queue entry and re-testing whether that entry is still a partial (continuation) reply on every line. The partial-response test runs str.slice(str.lastIndexOf('\n') + 1) on the growing entry, which forces a full flatten of the entire accumulated string on each line. A malicious or compromised SMTP server (or an on-path attacker on a plaintext or pre-STARTTLS hop) that streams continuation lines and never sends a completion line drives O(n^2) CPU on the single-threaded Node.js event loop before any authentication, freezing the whole process for an attacker-controlled duration.

Root Cause

The invariant that a server cannot cause superlinear client-side work while a reply is being reassembled is violated. In src/smtp-connection/index.ts, _onData splits inbound bytes into lines and, while the last queued entry is a continuation reply, concatenates each new line onto that same entry with this._responseQueue[len - 1] += '\n' + lines[i]. On every iteration it calls isPartialResponse(lastline), which evaluates /^\d+-/.test(str.slice(str.lastIndexOf('\n') + 1)). Because the queued entry grows by one line per continuation line and is stored as a V8 rope (ConsString), each lastIndexOf/slice forces a full string flatten of length proportional to the accumulated reply. Executed once per line over n continuation lines, this is O(n^2) CPU. While the reply stays partial the function early-returns and never resets the entry, so the growth and the per-line reflatten continue unbounded. There is no response-size or line-count cap in the connection code; the only limits are time-based (greetingTimeout, and an inactivity socketTimeout), and neither bounds accumulated reply size.

Affected file and functions (verified at the published v10.0.9 artifact, in src/ and in the built dist/cjs and dist/esm output):
- src/smtp-connection/index.ts: isPartialResponse (the /^\d+-/ test over str.slice(str.lastIndexOf('\n')+1))
- src/smtp-connection/index.ts: SMTPConnection._onData (the continuation-line append loop with no size cap)

Impact

An attacker who is, or can impersonate, the SMTP peer the client connects to can freeze the connecting Node.js process. Because Node.js is single-threaded, event-loop starvation halts all other work in that process (every other connection, request, and timer), not just the offending SMTP session. The reply is parsed during the greeting and EHLO phases, before the client authenticates or issues any command, so no credentials on the client are required. The blocking duration scales with the volume of continuation data the attacker streams and can be sustained indefinitely by continuing to stream, and memory for the accumulated entry also grows without bound. Independent reproduction on Node.js 24 measured roughly 11.6 seconds of event-loop blocking for about 640 KB of continuation data and roughly 47 seconds for about 1.3 MB, scaling quadratically. This is a repeatable remote denial of service against the sending process.

Proof of Concept

Node.js 24; a raw TCP listener acting as a hostile SMTP server; the client uses the shipped SMTPConnection (require('nodemailer/lib/smtp-connection')).

Steps: (1) start a TCP server that accepts a connection and streams continuation lines of the form 220-flood...\r\n in large batches, never emitting a completion line such as 220 ready; (2) point an SMTPConnection at it and observe the client event loop stall while the single response-queue entry grows and is reflattened on every line. A minimal microbenchmark of the exact parsing logic demonstrates the complexity:

// Reproduces isPartialResponse + the _onData append loop verbatim (Node 24).
function isPartialResponse(str) { return /^\d+-/.test(str.slice(str.lastIndexOf('\n') + 1)); }
function run(n) {
    const queue = [];
    let remainder = '';
    function onData(chunk) {
        const lines = (remainder + chunk).split(/\r?\n/);
        remainder = lines.pop();
        for (let i = 0; i < lines.length; i++) {
            if (queue.length && isPartialResponse(queue[queue.length - 1])) {
                queue[queue.length - 1] += '\n' + lines[i];
                continue;
            }
            queue.push(lines[i]);
        }
    }
    let buf = '';
    for (let i = 0; i < n; i++) buf += '220-flood line ' + i + '\r\n';
    const t = Date.now();
    onData(buf);
    return Date.now() - t;
}
for (const n of [8000, 16000, 32000, 64000]) console.log(n, run(n) + 'ms');

8000   26.5ms
16000  357.6ms
32000  2712.0ms
64000  11645.1ms
(2x input yields roughly 4x-7x time: quadratic. ~640 KB of continuation data blocks the event loop ~11.6s.)

Attack Chain

1. Exposure: The client opens an SMTP connection whose peer is attacker-controlled or attacker-influenced: a compromised or malicious configured server, a tenant-supplied outbound server in a multi-tenant "bring your own SMTP" deployment, a deliverability or mailbox-probing tool that connects to arbitrary servers, or an on-path attacker on a plaintext or pre-STARTTLS hop.
2. Control: The peer fully controls the bytes of its greeting and EHLO replies, including whether a reply ever completes.
3. Path: Inbound bytes reach the socket data handler _onData, the sole data handler; every reply phase traverses the same continuation-append loop.
4. Guard: No response-size or line-count cap exists. The greeting timer is time-based and cleared after the greeting, and its callback is itself starved while _onData runs synchronously; the inactivity timer resets on every inbound chunk, so an actively streaming attacker never trips it.
5. Primitive: Each continuation line appended to the single growing queue entry triggers a full string flatten in isPartialResponse, yielding O(n^2) synchronous CPU on the event loop.
6. Result: Whole-process event-loop starvation for an attacker-controlled, sustainable duration, plus unbounded memory growth: a remote denial of service reachable before authentication.

Bypass Evidence

The relevant guard would be a cap on accumulated reply size or line count. A search of the connection module (maxSize, maxLength, responseQueue.length >, byte or line ceilings) finds none; the only bounds are greetingTimeout and the inactivity-based socketTimeout, both time-based and neither bounding accumulated size. Disproof attempts: the greeting timeout does not help because it is time-based, fires only during the greeting phase, is cleared once the greeting completes, and its callback is delayed while _onData runs synchronously; a well-behaved server does not trigger the issue because a completion line resets the entry and keeps cost linear, but the modeled attacker is a hostile peer; segmented delivery does not help the defender because the response queue and line remainder persist across _onData calls so the accumulation and quadratic cost survive chunking; there is no safer alternate caller because _onData is the only data handler and the append loop is on every reply phase. Note on scope: core nodemailer connects to a configured host and does not perform MX resolution, so this is not reachable through a direct-to-MX transport; the malicious-peer, multi-tenant tenant-supplied-host, deliverability-probe, and MITM paths establish reachability.

Affected Versions

• Ecosystem: npm
• Package: nodemailer
• Confirmed affected range: <= 10.0.9
• Latest release checked: nodemailer 10.0.9 (npm registry; vulnerable code verified at git tag v10.0.9 and in the published dist output)
• Fix status: 10.0.10

The quadratic reassembly shape is long-standing and predates the addressparser complexity fix family, so the exact introducing release is not positively established and the lower bound is left open; the upper bound is verified against the current published artifact, which is vulnerable with no fix.

Suggested Fix

Enforce that reassembling a multiline reply is linear in the reply size and that the reply size is bounded. Track the running length of the accumulated entry and the continuation state without re-scanning the whole entry on every line (for example, test the newly arrived line rather than slicing the tail of the growing accumulator, or collect lines in an array and join once at completion). Independently, impose a configurable maximum on the accumulated reply size or continuation-line count and fail the connection when it is exceeded, so a peer that never completes a reply cannot force unbounded work or memory. The remediation must cover every reply phase (greeting, EHLO, and command responses), because all of them flow through the same handler. As an interim mitigation, only connect to trusted SMTP servers over authenticated TLS (avoiding plaintext or STARTTLS-strippable hops), and in multi-tenant deployments isolate outbound SMTP sessions to tenant-supplied hosts so one tenant cannot starve a shared process.

Reported by zx (GitHub: @manus-pi).