* @copyright MMXIII Andreas Fischer * @license http://www.opensource.org/licenses/mit-license.html MIT License */ declare(strict_types=1); namespace phpseclib3\Tests\Unit\Crypt; use phpseclib3\Crypt\RC2; use phpseclib3\Tests\PhpseclibTestCase; class RC2Test extends PhpseclibTestCase { public $engines = [ 'PHP', 'Eval', 'OpenSSL', ]; public function engineVectors(): array { // tests from https://tools.ietf.org/html/rfc2268#page-8 $tests = [ // key, effective key length, plaintext, ciphertext ['0000000000000000', 63, '0000000000000000', 'ebb773f993278eff'], ['ffffffffffffffff', 64, 'ffffffffffffffff', '278b27e42e2f0d49'], ['3000000000000000', 64, '1000000000000001', '30649edf9be7d2c2'], ['88', 64, '0000000000000000', '61a8a244adacccf0'], ['88bca90e90875a', 64, '0000000000000000', '6ccf4308974c267f'], ['88bca90e90875a7f0f79c384627bafb2', 64, '0000000000000000', '1a807d272bbe5db1'], ['88bca90e90875a7f0f79c384627bafb2', 128, '0000000000000000', '2269552ab0f85ca6'], ['88bca90e90875a7f0f79c384627bafb216f80a6f85920584c42fceb0be255daf1e', 129, '0000000000000000', '5b78d3a43dfff1f1'] ]; $result = []; foreach ($this->engines as $engine) { foreach ($tests as $test) { $result[] = [$engine, $test[0], $test[1], $test[2], $test[3]]; } } return $result; } // this test is just confirming RC2's key expansion public function testEncryptPadding(): void { $rc2 = new RC2('ecb'); // unlike Crypt_AES / Crypt_Rijndael, when you tell Crypt_RC2 that the key length is 128-bits the key isn't null padded to that length. // instead, RC2 key expansion is used to extend it out to that length. this isn't done for AES / Rijndael since that doesn't define any // sort of key expansion algorithm. // admittedly, phpseclib is inconsistent in this regard. RC4 and Blowfish support arbitrary key lengths between a certain range, as well, // and they don't have any way to set the key length. but then again, neither do those algorithms have their own key expansion algorithm, // whereas RC2 does. and technically, AES / Rijndael (and even Twofish) don't support arbitrary key lengths - they support variable key // lengths. so in some ways, i suppose this inconsistency somewhat makes sense, although the fact that Crypt_Twofish doesn't have a // setKeyLength() function whereas Crypt_AES / Crypt_Rijndael do not is, itself, an inconsistency. // but that said, Crypt_RC2 is inconsistent in other ways: if you pass a 128-bit (16-byte) key to it via setKey() the key is not treated // as a 128-bit key but rather as a 1024-bit key and is expanded accordingly, not via null padding, but via RC2's key expansion algorithm. // this behavior is in contrast to mcrypt, which extends keys via null padding to 1024 bits. it is also in contrast to OpenSSL, which // extends keys, via null padding, to 128 bits. mcrypt's approach seems preferable as one can simulate 128 bit keys by using RC2's // key expansion algorithm to extend the key to 1024 bits and then changing the first byte of the new key with an inverse pitable mapping. // in contrast, to my knowledge, there is no technique for expanding a key less than 128 bits to 128 bits, via RC2 key expansion. the only // scenario in that regard is null padding. // simple truncation is insufficient, since, quoting RFC2268, "the purpose of the key-expansion algorithm [in RC2] is to modify the key buffer // so that each bit of the expanded key depends in a complicated way on every bit of the supplied input key". // now, to OpenSSL's credit, null padding is internally consistent with OpenSSL. OpenSSL only supports fixed length keys. For rc2, rc4 and // bf (blowfish), all keys are 128 bits (or are null padded / truncated accordingly). to use 40-bit or 64-bit keys with RC4 with OpenSSL you // don't use the rc4 algorithm - you use the rc4-40 or rc4-64 algorithm. and similarly, it's not aes-cbc that you use - it's either aes-128-cbc // or aes-192-cbc or aes-256-cbc. this is in contrast to mcrypt, which (with the exception of RC2) actually supports variable and arbitrary // length keys. // superficially, it seens like Rijndael would be another exception to mcrypt's key length handling, but it in fact is not. the reason being that, // with mcrypt, when you specify MCRYPT_RIJNDAEL_128 or MCRYPT_RIJNDAEL_192 or MCRYPT_RIJNDAEL_256 the numbers at the end aren't referring to the // key length, but rather, the block length. ie. Rijndael, unlike most block ciphers, doesn't just have a variable (but not arbitrary) key length - // it also has a variable block length. AES's block length, however, is not variable, so technically, only MCRYPT_RIJNDAEL_128 is AES. $rc2->setKey(str_repeat('d', 16), 128); $rc2->setPreferredEngine('PHP'); $internal = $rc2->encrypt('d'); $result = pack('H*', 'e3b36057f4821346'); $this->assertEquals($result, $internal, 'Failed asserting that the internal engine produced the correct result'); $rc2->setPreferredEngine('OpenSSL'); if ($rc2->getEngine() == 'OpenSSL') { $openssl = $rc2->encrypt('d'); $this->assertEquals($result, $openssl, 'Failed asserting that the OpenSSL engine produced the correct result'); } else { self::markTestSkipped('Unable to initialize OpenSSL engine'); } } /** * @dataProvider engineVectors */ public function testVectors($engine, $key, $keyLen, $plaintext, $ciphertext): void { $rc2 = new RC2('cbc'); $rc2->disablePadding(); $rc2->setKeyLength($keyLen); $rc2->setKey(pack('H*', $key)); // could also do $rc2->setKey(pack('H*', $key), $keyLen) $rc2->setIV(str_repeat("\0", $rc2->getBlockLength() >> 3)); if (!$rc2->isValidEngine($engine)) { self::markTestSkipped("Unable to initialize $engine engine"); } $rc2->setPreferredEngine($engine); $result = bin2hex($rc2->encrypt(pack('H*', $plaintext))); $this->assertEquals($result, $ciphertext, "Failed asserting that $plaintext yielded expected output in $engine engine"); $result = bin2hex($rc2->decrypt(pack('H*', $ciphertext))); $this->assertEquals($result, $plaintext, "Failed asserting that decrypted result yielded $plaintext as a result in $engine engine"); } }