NextAuthjs Email misdelivery Vulnerability

Summary

NextAuth.js's email sign-in can be forced to deliver authentication emails to an attacker-controlled mailbox due to a bug in nodemailer's address parser used by the project (fixed in nodemailer v7.0.7). A crafted input such as:

"e@attacker.com"@victim.com

is parsed incorrectly and results in the message being delivered to e@attacker.com (attacker) instead of "<e@attacker.com>@victim.com" (the intended recipient at victim.com) in violation of RFC 5321/5322 semantics. This allows an attacker to receive login/verification links or other sensitive emails intended for the victim.

Affected NextAuthjs Version

≤ Version	Afftected
4.24.11	Yes
5.0.0-beta.29	Yes

POC

Example Setup showing misdelivery of email

import NextAuth from "next-auth"
import Nodemailer from "next-auth/providers/nodemailer"
import { PrismaAdapter } from "@auth/prisma-adapter"
import { prisma } from "@/lib/prisma"

export const { handlers, auth, signIn, signOut } = NextAuth({
  adapter: PrismaAdapter(prisma),
  providers: [
    Nodemailer({
      server: {
        host: "127.0.0.1",
        port: 1025,
        ...
      },
      from: "noreply@authjs.dev",
    }),
  ],
  pages: {
    signIn: '/auth/signin',
    verifyRequest: '/auth/verify-request',
  },
})

POST /api/auth/signin/nodemailer HTTP/1.1
Accept-Encoding: gzip, deflate, br, zstd
Cache-Control: no-cache
Connection: keep-alive
Content-Length: 176
DNT: 1
Host: localhost:3000
Origin: http://localhost:3000
Pragma: no-cache
Referer: http://localhost:3000/auth/signin
Sec-Fetch-Dest: empty
Sec-Fetch-Mode: cors
Sec-Fetch-Site: same-origin
User-Agent: Mozilla/5.0 (X11; Linux x86_64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/141.0.0.0 Safari/537.36
accept: */*
accept-language: en-US,en;q=0.9,ta;q=0.8
content-type: application/x-www-form-urlencoded
sec-ch-ua: "Google Chrome";v="141", "Not?A_Brand";v="8", "Chromium";v="141"
sec-ch-ua-mobile: ?0
sec-ch-ua-platform: "Linux"
x-auth-return-redirect: 1

email=%22e%40attacker.coccm%22%40victim.com&csrfToken=90f5e6f48ab577ab011f212011862dcfe546459c23764cf891aab2d176f8d77a&callbackUrl=http%3A%2F%2Flocalhost%3A3000%2Fauth%2Fsignin

Mitigation

Update to nodemailer 7.0.7

Credits

https://zeropath.com/ Helped identify this security issue

Missing proper state, nonce and PKCE checks for OAuth authentication

Impact

next-auth applications using OAuth provider versions before v4.20.1 are affected.

A bad actor who can spy on the victim's network or able to social engineer the victim to click a manipulated login link could intercept and tamper with the authorization URL to log in as the victim, bypassing the CSRF protection.

As an example, an attack can happen in the following scenario.

TL;DR: The attacker steals the victim's authenticated callback by intercepting and tampering with the authorization URL created by next-auth.

1. The victim attempts to log in to the next-auth site. For example https://next-auth-example.vercel.app/

2. next-auth sets the checks cookies according to how the OAuth provider is configured. In this case, state and pkce are set by default for the Google Provider.

3. The attacker intercepts the returned authorization URL, strips away the OAuth check (nonce, state, pkce), and returns the URL without the check to the victim's browser. For example: From https://accounts.google.com/o/oauth2/v2/auth/oauthchooseaccount?client_id=client_id&scope=openid%20email%20profile&response_type=code&redirect_uri=https%3A%2F%2Fnext-auth-example.vercel.app%2Fapi%2Fauth%2Fcallback%2Fgoogle&state=state&code_challenge=code_challenge&code_challenge_method=S256&service=lso&o2v=2&flowName=GeneralOAuthFlow to https://accounts.google.com/o/oauth2/v2/auth/oauthchooseaccount?client_id=client_id&scope=openid%20email%20profile&response_type=code&redirect_uri=https%3A%2F%2Fnext-auth-example.vercel.app%2Fapi%2Fauth%2Fcallback%2Fgoogle&service=lso&o2v=2&flowName=GeneralOAuthFlow. Notice the parameters state, code_challenge and code_verifier are removed from the victim's address bar.

1. The victim attempts to log in using their OAuth account.

2. The Authorization Server logs the victim in and calls back to the next-auth api/auth/callback/:providerIdendpoint. 5.1. The attacker intercepts and logs this callback URL for later use. 5.2. next-auth checks the callback call from OAuth Authorization Server (doesn't have checks) and compares the checks with the cookies set (has checks) at step 2. This check will fail, resulting in the victim isn't logged in. However, at this step, the Authorization Server has already accepted the victim's request to log in and generated/sent a code in the URL.

3. The attacker now has an authorization URL with the code that the AS will exchange for valid access_token/id_token and can log in as the victim automatically. They can open a new browser window and paste in the URL logged at step 5.1 and log in as the victim.

Patches

We patched the vulnerability in next-auth v4.20.1 To upgrade, run one of the following: npm i next-auth@latest yarn add next-auth@latest pnpm add next-auth@latest

Workarounds

Upgrading to latest is the recommended way to fix this issue. However, using Advanced Initialization, developers can manually check the callback request for state, pkce, and nonce against the provider configuration, and abort the sign-in process if there is a mismatch. Check out the source code for help.

References

• CSRF
• PKCE vs nonce
• OAuth provider options
• checks provider config

Possible user mocking that bypasses basic authentication

Impact

next-auth applications prior to version 4.24.5 that rely on the default Middleware authorization are affected.

A bad actor could create an empty/mock user, by getting hold of a NextAuth.js-issued JWT from an interrupted OAuth sign-in flow (state, PKCE or nonce).

Manually overriding the next-auth.session-token cookie value with this non-related JWT would let the user simulate a logged in user, albeit having no user information associated with it. (The only property on this user is an opaque randomly generated string).

This vulnerability does not give access to other users' data, neither to resources that require proper authorization via scopes or other means. The created mock user has no information associated with it (ie. no name, email, access_token, etc.)

This vulnerability can be exploited by bad actors to peek at logged in user states (e.g. dashboard layout).

Note: Regardless of the vulnerability, the existence of a NextAuth.js session state can provide simple authentication, but not authorization in your applications. For role-based access control, you can check out our guide.

Patches

We patched the vulnerability in next-auth v4.24.5. To upgrade, run one of the following:

npm i next-auth@latest

yarn add next-auth@latest

pnpm add next-auth@latest

Workarounds

Upgrading to latest is the recommended way to fix this issue. However, using a custom authorization callback for Middleware, developers can manually do a basic authentication:

// middleware.ts
import { withAuth } from "next-auth/middleware"

export default withAuth(/*your middleware function*/, {
  // checking the existence of any property - besides `value` which might be a random string - on the `token` object is sufficient to prevent this vulnerability
  callbacks: { authorized: ({ token }) => !!token?.email }
})

References

• NextAuth.js Middleware
• Role-based access contorl (RBAC) guide

NextAuth.js before 4.10.3 and 3.29.10 sending verification requests (magic link) to unwanted emails

Impact

next-auth users who are using the EmailProvider either in versions before 4.10.3 or 3.29.10 are affected.

If an attacker could forge a request that sent a comma-separated list of emails (eg.: attacker@attacker.com,victim@victim.com) to the sign-in endpoint, NextAuth.js would send emails to both the attacker and the victim's e-mail addresses. The attacker could then login as a newly created user with the email being attacker@attacker.com,victim@victim.com. This means that basic authorization like email.endsWith("@victim.com") in the signIn callback would fail to communicate a threat to the developer and would let the attacker bypass authorization, even with an @attacker.com address.

Patches

We patched this vulnerability in v4.10.3 and v3.29.10 by normalizing the email value that is sent to the sign-in endpoint before accessing it anywhere else. We also added a normalizeIdentifier callback on the EmailProvider configuration, where you can further tweak your requirements for what your system considers a valid e-mail address. (E.g.: strict RFC2821 compliance)

To upgrade, run one of the following: sh npm i next-auth@latest sh yarn add next-auth@latest sh pnpm add next-auth@latest

(This will update to the latest v4 version, but you can change latest to 3 if you want to stay on v3. This is not recommended. v3 is unmaintained.)

Workarounds

If for some reason you cannot upgrade, you can normalize the incoming request like the following, using Advanced Initialization: ```ts // pages/api/auth/[...nextauth].ts

function normalize(identifier) { // Get the first two elements only, // separated by @ from user input. let [local, domain] = identifier.toLowerCase().trim().split(@) // The part before @ can contain a , // but we remove it on the domain part domain = domain.split(,)[0] return ${local}@${domain} }

export default async function handler(req, res) { if (req.body.email) req.body.email = normalize(req.body.email) return await NextAuth(req, res, {/* your options */ }) } ```

References

• EmailProvider: https://next-auth.js.org/providers/email
• Normalize the email address: https://next-auth.js.org/providers/email#normalizing-the-email-address
• Email syntax: https://en.wikipedia.org/wiki/Email_address#Local-part
• signIn callback: https://next-auth.js.org/configuration/callbacks#sign-in-callback
• Advanced Initialization: https://next-auth.js.org/configuration/initialization#advanced-initialization
• nodemailer address: https://nodemailer.com/message/addresses

For more information

If you have any concerns, we request responsible disclosure, outlined here: https://next-auth.js.org/security#reporting-a-vulnerability

Timeline

The issue was reported 26th of July, a response was sent out in less than 1 hour and after identifying the issue a patch was published within 5 working days.

Acknowledgments

We would like to thank Socket for disclosing this vulnerability in a responsible manner and following up until it got published.