Why is using salt more secure?

Question

Storing the hash of users' passwords, e.g. in a database, is insecure since human passwords are vulnerable to dictionary attacks. Everyone suggests that this is mitigated via the use of salts, but the salt is considered non-sensitive and does not need to be protected.

In the event that the attacker has the salt how has his dictionary attack become more difficult than before? Does not having access to the salt, effectively, remove its usefulness?

Answer 1

Salt doesn't protect you against a lone attacker who is only after one password. An attacker who just wants to break one password will calculate hash(salt + guess) instead of hash(guess) (if the password scheme is hash(salt+password)).

Salt helps if the attacker wants to break many passwords. This is usually the case. Sometimes the attacker is attacking a site and wants to break into an account on that site, any account. Without a salt, most of the attacker's work can be used for all accounts, so she can test each of her attempts against all accounts at once. With a correctly-chosen salt (i.e. if no two accounts have the same salt), the attacker has to start over for each hashed password.

Furthermore, in a sense, all password cracking attempts are attempting to crack all accounts passwords at once. That's because hashes can be precomputed; all it takes is for someone to generate a table of hashes — or more efficiently, a rainbow table — and that initial work can be distributed to multiple attackers who can use it on any account database that uses the same password hashing algorithm. Again, salt makes these precomputations useless.

A brute-force password attack can be summarized like this:

1. Make any precomputation that the attacker deems useful, such as building a rainbow table (which is an efficient way to represent a table mapping hashes to common passwords).
2. For every one of the n accounts that the attacker is interested in breaking in, and for every one of the p password guesses that the attacker includes in her dictionary, test whether hash(guess[i]) = hashed_password[j].

In a naive approach, the second step requires n × p hash computations to try all guesses against all accounts. However, if the first step already calculated all the possible hashes, then the second step requires no hash computation at all, just testing whether each hashed_password is in the precomputed database, so the attack requires just n table lookups (this can even be sped up, but we've already gone from n x p slow computations¹ down to n table lookups).

If each password has a different salt, then in order to be helpful, the precomputation would have to include an entry for every possible salt value. If the salt is large enough, the precomputation is infeasible. If the precomputation doesn't take the salt into account, it won't be useful to speed up the second step, because any cryptographic hash function “mixes” its input: knowing the hash of UIOQWHHXpassword doesn't help compute the hash of NUIASZNApassword. Even to attack a single account, the attacker needs to perform p hash computations to try all guesses, already an improvement on the single table lookup that would be sufficient if the attacker has a precomputed dictionary.

¹ _{A password should not be stored as a hash such as SHA-1, but using a slower hash function such as bcrypt or scrypt or PBKDF2.}

Answer 2

I'm going to explain each level of security for database stored password, maybe this will help clarify the importance of salt.

Level 1: Password stored in clear in the database.

This is just pure stupidity. But sometimes, big company get their database compromised and, oh surprise, all password are stored in clear. What a shame!

Level 2: Password stored using an hashing algorithm.

This is a step toward more security. If an attacker get the hashed password, he can't still login using this credential to the service. He will first have to "unhash" the password using a Rainbow Table or by brute forcing it. That means it require for the attacker a bit more work, but it can still be possible to retrieve the original password.

Level 3: Password stored using an hashing algorithm with a salt.

This starts to be interesting. As we saw on Level 2, an attacker that has access to the hashed password can either brute force it or use a rainbow table. Adding a salt to the original password make the rainbow table totally useless because they don't take into consideration the salt. It still possible to get the original string using a rainbow table, but it would mean that in the rainbow table, the password+salt exists. Since a salt is generally very long, the odd having a hashed value of a string(40+) is almost impossible. The only solution left is brute force.

Level 4: Password stored using an hashing algorithm with a salt and a peper.

This is very interesting (it's the reason why I'm posting my answer since the actual ones are already interesting).

I recommend you to use an hashing algorithm like BCrypt, SCrypt or PBKDF2 since they aren't beek broken so far, and are very slow to hack, increasing the time to break one, and then a complete database of passwords.

The difference between salt and pepper is their location. The salt is generally stored in the database but the pepper is located in the code. This can seems odd but the original idea is that a database can be compromised, but not the code, leaving the attacker a salt and a list of useless hashed password. Moreover, it add even more complexity to the hashed password, making rainbow tables completely useless (if we suppose they were still a bit useful with a salt!).

In the case that only the database is compromised, even the brute force is then useless since the attacker is not looking for one value (the original password) but for two values (the original password AND the pepper).

Answer 3

If your hashes are unsalted I can generate a whole dictionaries worth of hashes and store them in a database (special kind called a rainbow table, it's more efficient for large numbers of hashes) , then all I need to do is look them up. Basically one giant memory vs. CPU time optimization. Also, if I see two users with the same hash I know they are using the same password.

A salt makes every users password hash unique, and if you do it right it'll likely be unique even if they use that same password on some other site (not uncommon), and so I would actually have to take word, apply salt and hash it, repeat for next entry in dictionary.

Answer 4

Unsalted hashes are vulnerable to lookup attacks. There are freely available databases containing millions of password hashes (mainly MD5 and SHA1) which can be used to look up the plaintext of a hash. These may be based on standard relational databases, or on specialized database formats like rainbow tables.

So, here's an example:
7c6a61c68ef8b9b6b061b28c348bc1ed7921cb53 (unsalted)
2d67062d67b4d0eaca31a768e901d04982de7352 (salted with "RxB6aE" prefix)

A quick lookup on the first hash will reveal the plaintext to be "passw0rd". However, the latter isn't very likely to be found. In the case where the salt isn't known by the attacker, it makes dictionary and bruteforce attacks difficult.

If you're looking to secure passwords in a database, you should use something like bcrypt. It uses a large number of iterations of a hash algorithm to make each computation slow. As such, bruteforce attacks, dictionary attacks and the building of rainbow tables are highly infeasible.