Migrate Oracle authentication to table

Question

I have a web application based on Java/Spring that uses Oracle 11g. Currently, the users authenticate via username/password directly against the system table SYS.USER$ on login.

This has to change, so we created a (regular) new table to store all the user data there. We inserted all existing passwords to the newly created table. However, the passwords seem to be encrypted/hashed in a way that's described by this site

One example: Once the user enters XXXXX, the database stores 07E4898C06DEF253.

I want to perform the authentication with the old passwords stored in the new (regular) table. My problem is that I don't know how to verify the existing passwords since I don't know exactly how they have been hashed/encrypted.

I played around with ora_hash and dbms_obfuscation_toolkit.DESDecrypt, but none of these gave me a correct result. I know the correct password for my user and I can see Oracle's generated value for this one, but I can't reproduce the way Oracle generally "handles" the password data.

Is there any way to solve this problem without resetting all passwords?

Answer 1

Adapting the Java implementation you linked to in a comment, which is close but isn't quite using the salt properly:

import java.security.MessageDigest;
import java.util.Formatter;

class Main{

    public static String calculateHash(String password) throws Exception{
        MessageDigest crypt = MessageDigest.getInstance("SHA-1");

        String encodedPassword = "S:71752CE0530476A8B2E0DD218AE59CB71B211D7E1DB70EE23BFB23BDFD48";

        // Convert password to bytes
        byte[] bPassword = password.getBytes("UTF-8");

        // Get salt from encoded password
        String salt = encodedPassword.substring(42, 62);
        System.out.println("Salt is " + salt);

        // Convert salt from hex back to bytes
        // based on http://stackoverflow.com/a/140861/266304
        int len = salt.length();
        byte[] bSalt = new byte[len / 2];
        for (int i = 0; i < len; i += 2) {
            bSalt[i / 2] = (byte) ((Character.digit(salt.charAt(i), 16) << 4)
                + Character.digit(salt.charAt(i+1), 16));
        }

        // Add converted salt to password bytes
        // based on http://stackoverflow.com/a/80503/266304
        byte[] bData = new byte[bPassword.length + bSalt.length];
        System.arraycopy(bPassword, 0, bData, 0, bPassword.length);
        System.arraycopy(bSalt, 0, bData, bPassword.length, bSalt.length);

        // Hash the final byte array
        crypt.update(bData);
        byte bHash[] = crypt.digest();

        Formatter formatter = new Formatter();
        for (byte b : bHash)
        {
            formatter.format("%02x", b);
        }

        System.out.println("Expected      " + encodedPassword.substring(2,42));

        return formatter.toString().toUpperCase();
    }

    public static void main(String[] args) throws Exception {
        System.out.println("The result is " + calculateHash("ZK3002"));
    }
}

Which gives output:

Salt is 1DB70EE23BFB23BDFD48
Expected      71752CE0530476A8B2E0DD218AE59CB71B211D7E
The result is 71752CE0530476A8B2E0DD218AE59CB71B211D7E

The PL/SQL version involves some conversion; dbms_crypto.hash() takes a RAW argument, so you have to convert the plain-text password to RAW, then concatenate the extracted salt - which is already hex. (In the PL/SQL version in Pete Finnigan's blog you may notice that he has an explicit hextoraw call, so I'm simplifying a bit). So the argument passed to dbms_crypto.hash for your example would be the hex (OK, raw) equivalent of ZK3002, which is 5A4B33303032, with the hex salt concatenated to that; so 5A4B333030321DB70EE23BFB23BDFD48.

For the Java version you pass a byte array, but that means you need to convert the salt extracted from the stored password back from hex before tacking it on to the password; and since it's unlikely to have a useful string representation you might as well put it straight into a byte array. So, convert the password to a byte array, convert the salt into a byte array, and stick the two arrays together. That then becomes the value you pass to MessageDigest.

You can compare the hash this produces with the Oracle-hashed version, skipping the initial S: and the embedded salt.