As best as I can understand things, it's not possible to get started with 
authentication using current technology without already knowing the time.  
You don't have to know it very accurately, but it can't be wildly off or you 
are opening a crack for MITM attacks.

For example, page 18, section 9.3 of draft-ietf-ntp-network-time-security-13 
says:
  "To solve this chicken-egg problem, the client has to rely on external 
means."

If we are serious about security, the problem with getting started needs to 
be acknowledged prominently and frequently rather than buried in a paragraph 
that most implementers will skim over.  That holds for both the NTS document 
and all documents that reference it.

Is getting started complicated enough that we should put the discussion in 
another document?  Has anybody thought about how close you need to know the 
time when getting started?


A variation on this problem is that DNSSEC requries time, so you can't use 
DNS to locate time servers until you know the time.  Again, you don't have to 
know it precisely, just close enough.

Page 16, section 8.2 of draft-ietf-ntp-using-nts-for-ntp-04 says:
"Therefore, NTS is unable to provide secure usage of NTP pools."  I think 
that's misleading.  It doesn't include the non-pool usage of DNS and you can 
use a pool after you know the time if the pool (and your API) supports DNSSEC.

Is there an API to require the use of DNSSEC when doing DNS lookups?

A well managed /etc/hosts (or equivalent) might be good enough, but the "well 
managed" part seems to be asking for trouble.


-- 
These are my opinions.  I hate spam.



_______________________________________________
TICTOC mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/tictoc

Reply via email to