[tor-talk] secure and simple network time (hack)

[Jacob Appelbaum]

On 02/20/2012 03:31 PM, Maxim Kammerer wrote:
> On Mon, Feb 20, 2012 at 22:30, Jacob Appelbaum <jacob at appelbaum.net> wrote:
>> Some people have taken to setting clocks with HTTP headers but I think that's a nightmare - not only
>> because people will parse the header with questionable code but also because of latency, amongst other things.
> 
> What questionable code? HTTP Date: header is standard (RFC 1123).

Whenever you parse a thing, there is code to do the parsing. That's the
code that is questionable. For tlsdate, I cast a few bytes as an
integer. It's well, quite straight forward to audit.

> HTPDate (C version) [1] does a rather good job of maintaining time
> from such headers, and with an obvious header parsing vulnerability
> fix and some improvements / feature additions [2] it is used in
> Liberté Linux without issues. The only downside is lack of https
> support.
> 

Totally unrelated to HTPDate - without https support, I think it suffers
from a pretty serious problem: it isn't securely transfering the data
from the server or authenticating it.

I haven't audited HTPDate but I generally think that using the TLS data
provides for all of the properties I'd want and almost zero issues that
I don't.

> [1] http://www.clevervest.com/htp/
> [2] https://github.com/mkdesu/liberte/blob/master/src/usr/local/portage/net-misc/htpdate/files/htpdate-1.0.4-robustness.patch
> 
>> Currently tlsdate only has one way to verify certificates to ensure that
>> the connection is secure - namely, it's the usual CA racket.
> 
> Does it mean that verification will fail if the clock is several years
> behind, for instance?
> 

It might. It depends. I have some things in the works to solve this
issue. One idea that I've been toying with is to include a few hard
coded times - compile time as an epoch, rather than 1970, recent time
fetches rather than 1970 and so on.

This will work securely with a clock drift of years but probably not
forty years.

>> I'd love some code review but also just some feedback.
> 
> Does become_nobody() drop group privileges as well?

It should eventually. It did at one point. I think right now that it
does not. I think that's a vote for doing so again. :)

> Is operation over
> Tor supported (I don't see any proxy handling)?
> 

If you `usewithtor` it should work without issues.

All the best,
Jacob