phc-discussions - Re: [PHC] Upgrade HKDF to HKDF2?

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite for Android: free password hash cracker in your pocket

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Date: Sun, 9 Mar 2014 10:56:20 -0400
From: Bill Cox <waywardgeek@...il.com>
To: discussions@...sword-hashing.net
Subject: Re: [PHC] Upgrade HKDF to HKDF2?

On Sun, Mar 9, 2014 at 10:28 AM, Jeremi Gosney <epixoip@...dshell.nl> wrote:
> On 3/8/2014 10:48 AM, Bill Cox wrote:
>> HMAC: Input collisions galore!  Really?!?  Why are we using this for
>> hashing passwords???
>> PBKDF2: Inherits HMAC collisions
>
> HMAC itself isn't the problem. PBKDF2 doesn't inherit an issue from
> HMAC, it creates an issue with HMAC.
>
> Collisions only occur when using the password as the key and the salt as
> the message. Using the salt as the key and the password as the message
> resolves this issue. Generate salts that are the exactly $blocksize in
> length, and you don't have to worry about null padding issues, either.
>
> PBKDF2PEBKAC.

Got it.  HKDF gets the order right, and avoids the input collisions.
It also relies on the underlying hash function to do any padding.  In
coarse detail, an attack only might gain knowledge of how many hash
blocks long a password is, and assuming it's <= 32, SHA-256 would be
find (I might have the password offset wrong).

I still see code like memcpy(buf, password) in the hashing functions,
though.  This runs in time proportional to passwordLength.  Why don't
we instead do some larger constant runtime loop, like:

for(i = 0; i < passwordLength; i += 256) {
    for(j = 0; j < 256; j++) {
        buf[i] = password[(i + j) % passwordLength];
    }
    Hash_Update(hashCtx, buf, 256);
}

That would run in constant time assuming passwords are <= 256 long....
wouldn't i?  Maybe I'd need to write my own mod function, in case some
platforms have time dependent IDIV instructions.  Still, wouldn't it
be better to have a longer pad and to not use memcpy on the password?

Thanks,
Bill