Are there known issues with the C function crypt()? - c

Question

I use crypt() to encrypt passwords for a project of mine. When the password is chosen by the user, it's encrypted like this:
password = crypt(<password chosen>, <user's account name>)
The problem is when the user logs in using their password. If what they type doesn't match their password, it should enter this if check:
if (strcmp(crypt(<what user types in as password>, <user's account name>), <user's encrypted password>)) {
//...
}
It doesn't in one certain scenario. Let's say their password is 'asdf'. If they enter 'asdf' with any random trailing characters, such as 'asdffffff' or 'asdf339sfd', it still accepts the password. It seems to ignore everything after 'asdf'.
Is this a known issue with crypt? Is there another way to encrypt passwords?

Answer 1

The second argument to crypt is not supposed to be the user's account name. It's supposed to be a setting string. Setting strings look like this:
$2b$07$fQuDK3TaQP4sw6IX6iVcTw
The $2b$07$ part tells crypt which password-hashing algorithm to use, and the string of random characters that follows is the salt. The salt must be different for each user, but it is not supposed to have any correlation with the user's account name. It doesn't technically have to be random, but it's critical that it be different for each user, and it needs to change every time the user changes their password, so best practice is to use a long string drawn from a cryptographic PRNG.
When you authenticate a user who has logged in before, you use the stored hashed password as the setting string:
char *new_hash = crypt("password typed in", "stored hash");
if (new_hash && !strcmp(new_hash, "stored hash")) {
// user has successfully logged in
}
This works because the stored hashed password always begins with the setting string that was used to create it in the first place, and crypt is coded to look only at the setting-string part.
(Also note the null check; some implementations of crypt can fail and report failure by returning a null pointer.)
When you create a new account or change a password, you have to generate a new setting string. If you have the function crypt_gensalt, use that:
char *new_setting = crypt_gensalt("$2b$", 0, 0, 0);
if (new_setting) {
char *new_hash = crypt("user's new password", new_setting);
// ...
} else {
// halt and catch fire
}
If you don't have crypt_gensalt you have to implement it yourself, unfortunately. (To make matters worse, some Unixes have the crypt_gensalt whose documentation I linked to above, and others have a different version, with the same name, doing the same job, but taking different arguments. Time to dust off your Autoconf skillz!)
Now you know all of that, I can explain why
password = crypt("password chosen", "user's account name");
appeared to work but truncated the password. Your user account names probably begin with at least two alphanumeric characters, right? Like, "Ma[ya]", or "zw[ol]"? Unfortunately, any two alphanumeric characters make a valid setting string ... that selects one of the oldest and least secure password-hashing algorithms known to science, descrypt. (It was pretty good when it was invented ... in the mid-1970s. Nowadays, it can be cracked by brute force no matter what the password is.) One of the many problems with this algorithm is that it truncates all passwords to eight characters. asdf and asdfhjkl hash to different things, but asdfhjkl and asdfhjkl1234 hash to the same thing.
The cure for this is to use crypt_gensalt or equivalent to select a modern algorithm. All the modern algorithms accept arbitrarily long passphrases.