Minutes for PERC at interim-2016-perc-4
minutes-interim-2016-perc-4-1

PERC Virtual Interim (4/27/2016)

Raw Notes
------------

Note Taker - Cullen Jennings
-----------------------------
Requirements

- should make sure that we say that if the KMF does like any of the security
profiles that the MDD supports, it just closes the connection

- matt miller was not at meeting but raised concerns on the basic design

* questions about if UDP or TCP would have better odds of getting through
firewalls in the case where the KMF is in a browser and is sending stuff to a
KMF outside the firewall.

If we decide that data channel is out of scope, need to write down that this is
scoped to just SRTP=DTLS for now

Long discussion thinking about different ways to transport.

- EKR has heard no reason why it should be TLS and heard reason it should be UDP

- One possible issue is MTU.
- Paul explained this was not an issue when we exchange key info and all the
things are small. Paul looked at it and does not feel we have an MTU issue.

- Other issue raised is NAT/ FW
- Adam pointed out that is DTLS has to work or this is useless so no worse to
use DTLS

Few questions going forward

- should control of keys and tunnel of message be separated

- should we mux the tunnels onto one 5 tuple

Should use UDP unless someone raises issue beyond FW / MTU

Note Taker - Suhas Nandakumar
----------------------------
 - Requirements should capture MDD vs KMF security profiles mismatch to avoid
 downgrade attacks

 - Matt Miller had few concerns about having HBH association with KMF instead
 of KMF

 - The trust association is with KMF and MDD is un-trusted which implies that
 MDD might not be a
   good place to have its own security association with the end-point (Ekr,
   Cullen)

 - Concerns raised agaisnt using TLS/TCP for the MDD-KMF protocol. UDP is good
 fit and given the
    low bandwidth exchange between KMF & MDD there is no MTU Issue

 - Adam pointed out NAT/F traversal is not an issue especially in the
 Co-Resident KMF use-case.

 - Adam also pointed out Long living TLS connection might get closed by the
 Firewall device, so another reason
    not to go with TLS

 - Minimal changes on the end-point DTLS stack
    + Use raw DTLS instead of special tunnel protocol (Ekr)
    + might need to consider assicaiting

 - consider the design of separating control messages vs dtls-srtp exchange and
 no need to mix them together

 Take aways
   - UDP vs TCP transport. No convincing reasons heard to have TCP as of yet
   - Separation of KMF<->MDD control messages vs end-point<->KMF dtls message
   exchange - For co-resident KMF scenarios , how multiplexing tunnels on the
   same 5-tuple work