phc-discussions - Re: Argon2 improvement thread

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: Wed, 22 Jul 2015 10:34:31 +0000 (UTC)
From: EddyHawk <quarsicon@...oo.com>
To: discussions@...sword-hashing.net
Subject: Re: Argon2 improvement thread

Congratulations to PHC!

As my input, I don't know whether this is possible/better/already done
to Argon2, but I prefer single unified PHC winner, like Argon2-hi-sd-hi:
 ---
 Phase1: expand/RAM-fill (write) from password hash / output of previous
          iteration.
         1 pass of 1/2 Argon2i (2 rounds of Blake2b).
         sequential addressing.
        
 Phase2: salt-dependent-only random word swaps (read and write).
         1 pass of Argon2d (4 rounds of Blake2b, or even 2 too).
         optional blamka or maxform can be applied here (once).

 Phase3: RAM state 'hashing' (read).
         1 pass of 1/2 Argon2i (2 rounds of Blake2b).
         sequential addressing.
         prevents salt-derived swaps being trivially inverted.
         outputs tag/hash for next iteration / as the final output.
 ---
to get the same performance as Argon2 hybrid. In other words,
the single independent phase in split into two lighter phases
before and after the non-independent phase.

Additionaly, 3-pass hybrid above forms a single iteration (as time_cost).
Thus, for very low mem_cost (say 8-64 KiB), we simply applies (& can afford)
much higher iterations (say thousands) for this new Argon2, therefore
making it more or less behave like Bcrypt (obtaining free Bcrypt-like GPU
resistance without having to incorporate such thing explicitly).
At very high mem_cost, we can only afford a few iterations, but Bcrypt-like
GPU resistance is no longer matters here (and thus naturally being
turned off by the choice of much less iterations).

Summary:
-Hybrid like Lyra2, but
-Fully immune to cache-timing-attacks like Catena
 (all cache-timing measurements will only leak information
  about the salt, which is usually already known/public),
-Faster than Argon2i (2 passes worthy instead of 3 passes),
-Nice placement of optional (slow) features, automatic L1 cache defense,
-Neat time_cost = iteration, like PbKDF2.