Follow @Openwall on Twitter for new release announcements and other news
[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Date: Sun, 21 Dec 2014 23:27:19 -0800
From: Walter Parker <walterp@...il.com>
To: oss-security@...ts.openwall.com
Subject: Re: can we talk about secure time?

On Sun, Dec 21, 2014 at 10:39 PM, Hanno Böck <hanno@...eck.de> wrote:

> On Sun, 21 Dec 2014 23:30:10 -0700
> Kurt Seifried <kseifried@...hat.com> wrote:
>
> > Having to reconcile multiple logs/events across widely distributed
> > systems, especially in high volume situations, 1-2 seconds is a deal
> > breaker. Or people running SCADA systems for industrial plants. Or
> > people that run financial systems.
>
> I don't think this contradicts my statement that average consumer hw
> doesn't need the high accuracy of ntp :-)
>
> This is something that I don't want to see in the Linux/xBSD world. A
split made between the software that the "average consumer" uses and the
software that prosumers and professionals use for basic system services
like timekeeping. We should all be be using the good stuff...

>From what has been said so far, there appears to be two sides:

One that points out that security in ntp is week therefore MITM attacks can
be done. So the idea is to remove ntp and replace it something else, which
would be using the time field in TLS transactions. Most arguments seem to
come from the assuming the TLS servers are secure and have the correct
time. Nothing has be said about how well the TLS servers would work as time
servers

The other points out that ntp is a time protocol that specializes in
getting and keeping time correct. It is designed to allow time to be set
with high level of precision and accuracy (milliseconds on regular systems,
microseconds when it really matters).

Has there been any studies on the quality of TLS for a timeserver? From
comments, to me, it looks like the TLS side is rationalizing the 1 second
limit on TLS because they want want to use TLS as is, without further
researching/refining the topic.

If we are going to build a secure time protocol, the timekeeping piece is
at least as important as the security piece.



> > So it's not an either/or situation (care about security, or have
> > accurate time, sometimes we need both).
>
> Yeah, I totally agree that this would be the desired thing to have.
> However the facts are that at the moment we don't. And imho for
> consumer HW the slight inaccuracy of tlsdate doesn't matter, while the
> insecurity of ntp does (as the very practical hsts attack has shown).
>
> I read these days that the Linux foundation is sponsoring some work on
> NTP. Anyone involved in this and can comment whether secure
> authentication for NTP is something that's being looked at or if it is
> only about creating a better implementation of the ntp software?
>
>
Given there are 100's of millions of NTP clients in the field right now,
where would we get the trusted TLS servers to use as time servers. We can't
trust the web server that we are trying to connect to, as if it is a MITM
attack, it could also have bad time.


Walter


-- 
The greatest dangers to liberty lurk in insidious encroachment by men of
zeal, well-meaning but without understanding.   -- Justice Louis D. Brandeis

Powered by blists - more mailing lists

Please check out the Open Source Software Security Wiki, which is counterpart to this mailing list.

Confused about mailing lists and their use? Read about mailing lists on Wikipedia and check out these guidelines on proper formatting of your messages.