Re: [TLS] ticket_lifetime and generic network layer

[Mark Dunn <[e-mail]>]

Thanks for taking the time to read and respond to my queries, my 
understanding of this protocol is incomplete, but I am working hard to 
remedy that.

On 09/02/17 19:05, Ilari Liusvaara wrote:
> On Thu, Feb 09, 2017 at 12:38:50PM -0000, Mark Dunn wrote:
>> I am reading your TLS 3.1 Standard and the mailing list.
>>
>> It looks great.
>>
>> I am particularly interested in using the 0-RTT feature for IoT timestamped
>> data, which would seem immune from replay attacks
>>
>>   
>>
>> I have a couple of questions
>>
>>   
>>
>> 1) The maximum ticket lifetime is set to 7 days. Is this based on hard
>> science or arbitrary?
>>
>> If it is arbitrary then 8 days for weekly intervals or 32 for days for
>> monthly intervals would  make better commercial sense
>>
>>                 (allowing for variability in wake-up times for constrained
>> devices)
> AFAIK, it is arbitrary. However, long validity periods bring security
> issues, with having to store and protect symmetric keys for a long
> time.
If it IS arbitrary, could we have a recommendation rather than a

"Servers MUST NOT use any value more than 604800 seconds (7 days)"
>> 2) Have you considered using TLS for a generic network layer?
> Note that TLS requires in-order reliable delivery (DTLS doesn't, but
> DTLS 1.3 is currently just handwaving), and neither is available below
> transport layer.
>
>
> -Ilari

Yes DTLS, of course you are correct. I will try and rephrase my questions.

I understand "TLS 1.3 requires in-order reliable delivery", but of 
course, if the protocol relies on in-order reliable data delivery then 
it cannot be secure.
The protocol must assume the opposite in fact, that an attacker will 
break boththose requirements in every way possible.
As the TLS 1.3 protocol seems to have been tested(designed) against a 
formal state machine then clearly someone agrees with me.


If the protocol's responsibility is to secure the application data from 
the insecure world of the Internet

Application | TLS :| TCP | IP | 
<-------------------------------------------> | IP | TCP |: TLS | 
Application
\----------------------------------\/---------------------------------/
                                                 Insecure Internet

then the protocol cannot rely on the TCP handshake, neither can it rely 
on TCP's packet ordering. It would seem TCP is largely redundant as IP 
does the fragmentation and re-assembly.
TCP's only role seems to be to hold open the NAT on a firewall!

Which is why when I look at this protocol, I overlook the ordered data 
in-session ability of TLS ( /head/-of-line blocking.. blah, blah)and in 
my blinkered view see the resumption/0-RTT ability as DTLS.
Any protocol which uses IP directly (including TCP) is responsible for 
reliable data delivery whether it is as simple as an ack/nack/retry or 
uses the sophisticated windowing etc. of TCP.
If these protocols sat on (D)TLS then each of these protocols would 
drive their reliable delivery through the TLS state machine. *(this is 
exciting)*

So, for example if TCP relied on TLS instead of the other way around 
then TCP would be protected under TLS's umbrella of security.

Application | TCP | TLS | IP | 
<-------------------------------------------> | IP |: TLS | TCP | 
Application
\----------------------------\/---------------------------/
                                                 Insecure Internet

TCP's sliding window would work, but the syn/syn-ack/ack handshake would 
be sub-optimal again as early data is not available during authentication.
(D)TLS key changes look like they would happen seamlesslyunderneath.

For the Machine to Machine (or Internet of Things) world however, things 
look much rosier (to me) as "Machines Don't Browse!".
Once the server/client authentication handshake is complete:

1) is it possible to use"resumption" as a three way handshake to 
securely deliver a single packet
2) Is it possible to use "0-RTT" as a mechanism to deliver a single 
packet (if slightly less securely)

Thinking this through (slightly) more carefully now, It would not be 
possible to place (D)TLS between the IP layer and Ethernet Layer as 
Ethernet does not have fragmentationand re-assembly
3) Do you agreethat securing virtualisation with (D)TLS should be 
possible (Application / TCP / IP / VXLAN / (D)TLS / IP)
4) Do you think wireless protocols are candidates as many wireless 
protocols have small packets and perform fragmentation/re-assembly.