phc-discussions - Re: [PHC] scripting memory (not so) high

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-next>] [day] [month] [year] [list]

Message-ID: <345989176.572692.1389343069533.open-xchange@email.1and1.com>
Date: Fri, 10 Jan 2014 02:37:49 -0600 (CST)
From: Steve Thomas <steve@...tu.com>
To: discussions@...sword-hashing.net
Subject: Re: [PHC] scripting memory (not so) high

> On January 9, 2014 at 10:32 PM Solar Designer <solar@...nwall.com> wrote:
>
> Disclaimer: there may be bugs in the smhkdf code (e.g., some off-by-ones
> are likely). Implementations in other languages will be needed before
> we can have much confidence in its correctness. Besides, several
> enhancements relative to scrypt's SMix structure may be implemented.
> These possible errors and enhancements may affect smhkdf's performance,
> although I expect its performance to stay within a factor of 2 of the
> numbers given above.
>
> Alexander

Found two:
Line 13: $m_cost = floor(($blocksize + 63) / 64);
Line 22: $mod = ($m_cost * 64) - $blocksize + 1;


BTW I was just working on a scripting language hash. It's about 2.8x faster
with the same defense against using less memory (both aren't that good at
this). For mine $t_cost = 1 runs 5x over the memory. For yours $k = 4 runs 3.4x
over the memory, but accesses 4x-5x blocks. Which is why I went to 5x.

Reference code (ie slow):
function pwhash_ref($pw, $salt, $m_cost, $t_cost)
{
    $m_cost = ($m_cost + 7) & 0xfffffff8;
    $h = $salt . $pw;
    $mem = '';
    for ($i = 0; $i < $m_cost; $i++)
    {
        $h = hash('sha512', $h, true);
        $mem = $h . $mem;
    }

    // Hash mem $t_cost+4 times
    $ctx = hash_init('sha512');
    for ($i = 0; $i < $t_cost + 4; $i++)
    {
        hash_update($ctx, $mem);
    }
    return hash_final($ctx);
}

Optimized code:
function pwhash($pw, $salt, $m_cost, $t_cost)
{
    $h = $salt . $pw;
    $m_cost = ($m_cost + 7) & 0xfffffff8;
    $mem = array();
    $memPos = floor(($m_cost - 1) / 8);
    for ($i = 0; $i <= $memPos; $i++)
    {
        $mem[] = null;
    }
    for ($i = 0; $i < $m_cost; $i += 8)
    {
        $h0 = $h = hash('sha512', $h, true);
        $h1 = $h = hash('sha512', $h, true);
        $h2 = $h = hash('sha512', $h, true);
        $h3 = $h = hash('sha512', $h, true);
        $h4 = $h = hash('sha512', $h, true);
        $h5 = $h = hash('sha512', $h, true);
        $h6 = $h = hash('sha512', $h, true);
        $h7 = $h = hash('sha512', $h, true);
        $mem[$memPos--] = $h7 . $h6 . $h5 . $h4 . $h3 . $h2 . $h1 . $h0;
    }
    $mem = implode('', $mem);

    // Hash mem $t_cost+4 times
    $ctx = hash_init('sha512');
    hash_update($ctx, $mem);
    hash_update($ctx, $mem);
    hash_update($ctx, $mem);
    hash_update($ctx, $mem);
    for ($i = 0; $i < $t_cost; $i++)
    {
        hash_update($ctx, $mem);
    }
    return hash_final($ctx);
}

I originally used md5() then switched to SHA512 because you did :). Also if
hash, hash_init, hash_update, hash_final are not available switching to
md5() and doing the following is a good alternative but does not scale well
with large $t_cost:
    for ($i = -3; $i < $t_cost; $i++)
    {
        $mem = md5($mem, true) . $mem;
    }
    return md5($mem);

It's only a little slower with md5() vs SHA512 with $t_cost = 12. Also I had a
more elaborate optimization for the memory expansion part (5% gain). It was
removed in the SHA512 version because there was no difference in speed.
Content of type "text/html" skipped