phc-discussions - Re: [PHC] Specification of a modular crypt format (2)

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, 30 Sep 2015 10:24:40 +0100
From: Hugo Landau <hlandau@...ever.net>
To: discussions@...sword-hashing.net
Subject: Re: [PHC] Specification of a modular crypt format (2)

On 28/09/15 12:12, Thomas Pornin wrote:
> Hello all,
>
> here are my revised specification for the modular crypt format, and
> example code. Here are the changes since the last version:
This seems good. A few small issues:

- By allowing variable hash lengths but not specifying the hash length
in the parameter section, a salt string is ambiguous and cannot be used
to unambiguously generate a hash. This introduces a dependency on an
implicit parameter, the hash length, which needs to be communicated
separately from the salt string. I should be able to unambiguously
generate a password using only a salt string. I shouldn't have to append
a dummy all-zeroes hash as a byzantine way of communicating the hash
length, or add a separate parameter just for the hash length.

Either the hash length should be explicitly declared, or (better) hash
length variation should be removed on the grounds that there's no clear
use case for it. If a use case comes into being, support for different
hash lengths can be introduced at that time.

- I think the acceptance of non-canonical strings should be changed from
'MAY' to 'SHOULD NOT'. Thus this is allowed but discouraged.

Of course this leads to the question of whether this will encourage
implementations to reject non-canonical strings and thereby introduce
fragility (see Postel's law). I don't think this applies since this is a
new application without any burden of compatibility with existing
implementations.

Generally speaking, if you accept non-canonical strings, there is going
to end up being some implementation which takes the liberty of
generating them. Whereas if you reject them from the beginning, all
implementations will make sure they generate only canonical strings.

In other words, I don't think Postel's law properly applies for new
protocols. If there is any case in which the principle should be
adopted, it is after many implementations are available and the effort
of standards to encourage conformance has failed and has ceased to have
any influence, in which case implementations get updated to reflect reality.