GHSA-7w52-7jvm-m9vw

Summary

There is a Proof of Concept which is able to enumerate the usernames of administrator users. This was possible by performing a timing attack.

Details

The faulty code exists in src/Core/Framework/Api/OAuth/UserRepository.php:

public function getUserEntityByUserCredentials(
        string $username,
        #[\SensitiveParameter]
        string $password,
        string $grantType,
        ClientEntityInterface $clientEntity
    ): ?UserEntityInterface {
        if ($this->loginConfigService->getConfig()?->useDefault === false) {
            // never allow login via password if the default login is disabled (e.g. using SSO only)
            return null;
        }

        $builder = $this->connection->createQueryBuilder();
        $user = $builder->select('user.id', 'user.password')
            ->from('user')
            ->where('username = :username')
            ->setParameter('username', $username)
            ->fetchAssociative();

        // PATH 1: EARLY RETURN WHEN USERNAME IS NOT FOUND
        if (!$user) {
            return null;
        }

        // PATH 2: VERIFY PASSWORD IF USER IS FOUND
        if (!password_verify($password, (string) $user['password'])) {
            return null;
        }

        return new User(Uuid::fromBytesToHex($user['id']));
    }

Subroutine getUserEntityByUserCredentials() is called when an auth request is send to api/oauth/token. If the given username is not found an early return is done (PATH 1). Only if the user is found we verify the password using password_verify.

PHP method password_verify by default uses hashing algorithm Argon2id which by design is intentionally 'slow' by introducing a timing cost to an attempt to bruteforce hashes more costly.

Since password_verify has a notable executable time, PATH 2 where an user is found and verified will be slower on average then PATH 1 where we do an early return for non-existing users.

Proposed fix

Before doing the early return, password_verify a dummy hash.

Impact

1. More targeted dictionary/bruteforce attacks.
2. Spear phishing / eases social engineering.
3. Credential stuffing from other data leaks.

Authors

Niel Duysters (@NielDuysters) and Thomas Brankaer (@tbrankaer)