Newsgroups: php.internals
Path: news.php.net
Xref: news.php.net php.internals:72410
Mailing-List: contact internals-help@lists.php.net; run by ezmlm
Received-SPF: error (pb1.pair.com: domain lerdorf.com from 209.85.160.171 cause and error)
Content-Type: text/plain;
	charset=us-ascii
Mime-Version: 1.0 (1.0)
In-Reply-To: <52F63DDE.6090600@lsces.co.uk>
Date: Sat, 8 Feb 2014 12:09:39 -0500
Cc: PHP internals <internals@lists.php.net>
Content-Transfer-Encoding: quoted-printable
Message-ID: <1D9112B0-8383-4C19-90AC-6B2F249C049F@lerdorf.com>
References: <52F61A78.1020401@lsces.co.uk> <CAF+90c9ZmeysC88ghHk50rpv_8T4jYhGCWEDwZxOR5EmPuDDLg@mail.gmail.com> <52F62B08.6050201@ajf.me> <52F63DDE.6090600@lsces.co.uk>
To: Lester Caine <lester@lsces.co.uk>
Subject: Re: [PHP-DEV] Security Diligence
From: rasmus@lerdorf.com (Rasmus Lerdorf)

On Feb 8, 2014, at 9:23, Lester Caine <lester@lsces.co.uk> wrote:
>=20
> Specifically looking at the 'timing attack', as I understand it, if a comp=
arison process scans all elements and simply sets a flag when failure is det=
ected which is not used until all characters have been processed. Which is t=
he reason for establishing 'safely' the number of characters involved. Using=
 64bit functions rather than 32bit will also change the way that process wor=
ks?

I think the one simple detail you are missing is that many/most string compa=
risons will short-circuit as soon as it hits a character that doesn't match.=
 If you think about that a little bit I am sure you can think of a nice algo=
rithm for narrowing in on the password character by character. That is, if '=
aaaa' consistently takes longer to compare against the password compared to '=
bbbb' then there is a good chance that the first character of the password i=
s 'a'. Looking at the firebird implementation that seems to be the case ther=
e as well. Of course, network jitter and other factors make this a complicat=
ed attack, but that doesn't change the fact that there is a faint signal the=
re and with enough iterations it has been proven to be possible to separate t=
hat signal from the noise.=20

As a fun little experiment for yourself, try implementing an iterative timin=
g attack against your local Firebird database password and see if you can fi=
nd the signal in the noise.

-Rasmus=