phc-discussions - Re: [PHC] pufferfish

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: Mon, 24 Mar 2014 02:47:54 -0700
From: Jeremi Gosney <epixoip@...dshell.nl>
To: discussions@...sword-hashing.net
Subject: Re: [PHC] pufferfish

On 3/24/2014 2:11 AM, beloumi wrote:
>>> 2. Instead of replacing the s-box entries, wouldn't it be easy to expand
>>> them to a memory-hard vector and use them piece by piece in the last
>>> encryption step (that means: using a different s-box for each of the 64
>>> encryptions or - more experimental - for each blowfish round)?
>>> Please excuse that I write a comment although I have not fully reflect
>>> the algorithm...
>> Depending on the m_cost, doing so could break the memory hardness. Since
>> only one s-box would be needed at a time, the attacker could hold each
>> s-box in global memory and swap them into local memory when they are
>> needed. Using each s-box simultaneously prevents this from occuring.
> You are right. And I think using the s-boxes simultaneously would not be
> an easy solution as I expected.

This is what the f-function does, and as you can see from the code it is
not that complicated.

Blowfish's f-function is as follows:

f(x) = ( (S[0][shr(x,24)] + S[1][shr(x,16) & 0xff] )
     ^ S[2][shr(x,8) & 0xff]) + S[3][x & 0xff]

And pufferfish's f-function is:

f(x) = ( (S[0][rotr64(x61)%words] + S[1][rotr64(x,22)%words] )
     ^ S[2][rotr64(x,53)%words]) + S[3][rotr64(x,33)%words]


> But I agree to Bill, that increasing the memory cost of Pufferfish would
> achieve more confidence, even if it is much less than other memory-hard
> schemes.The authors of bcrypt thought (1999) of 14 KB as a „moderate“
> memory amount. Maybe in this time „moderate“ means something like 1 MB.
> I like Bcrypt and I would be glad if we could extend this algorithm for
> the current requirements.

I think you meant 4KB, not 14KB. But yeah, this is what the m_cost
parameter is for. If you want to use 1 MiB, then just set m_cost to
1024. This will generate four s-boxes of 256 KiB each.


> Good luck, Jeremi!

Thank you! :)