phc-discussions - Re: [PHC] Minor bug in yescrypt

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] [day] [month] [year] [list]

Message-ID: <20150324032027.GA2156@openwall.com>
Date: Tue, 24 Mar 2015 06:20:27 +0300
From: Solar Designer <solar@...nwall.com>
To: discussions@...sword-hashing.net
Subject: Re: [PHC] Minor bug in yescrypt

Hi Steve,

Thank you for reviewing my code!

On Mon, Mar 23, 2015 at 09:34:56PM -0500, Steve Thomas wrote:
> file yescrypt-ref.c, line 728, in yescrypt_kdf():
> 
> - size_t dklen = g ? sizeof(dk) : buflen;
> + size_t dklen = buflen;
> 
> This needs to be this way since upgrades break if they are not the same length.

No, that's deliberate.  Both server relief and hash upgrades are only
defined for exactly 256-bit hashes.  I need to document this (just wrote
down a note to this extent, thanks!)

Technically, longer than 256-bit hashes may also be upgraded, with the
upgrades using their initial 256 bits, but maybe it's better to declare
this unsupported.

Shorter than 256-bit hashes can't be upgraded, because they need a
256-bit previous hash for the upgrades, but they can nevertheless be
computed right away with g > 0 if the caller so desires for whatever
reason.  And there's a valid reason why they might: g > 0 increases
resistance to garbage collector attacks.

Part of the rationale for the above is that I opted to address the
confusion reported here:

http://thread.gmane.org/gmane.comp.security.phc/2237/focus=2239

by having yescrypt's output for a shorter length always be a prefix of
its output for a longer length.  scrypt has the same property, and I
felt it'd be cleaner to preserve it for all of yescrypt's modes as well.
(I'd appreciate more testing for this property, to ensure there are no
exceptions to it in current yescrypt.)

Unfortunately, this has some complexity cost, such as on producing
exactly 256-bit intermediate derived keys and having a 256-bit buffer
to put them into even if the requested final output size is shorter (so
can't just use the output buffer unconditionally).  It's not just the
line you quoted above, but several other lines as well.

Thanks again,

Alexander