Re: [I2nsf] [IPsec] draft-abad-i2nsf-sdn-ipsec-flow-protection

["Valery Smyslov" <[e-mail]>]

Hi Gabriel,

 

I think that at this point the discussion is not very productive.

I admit that I’m not very familiar with SDNs, so I have to 

blindly trust you when you state that the SDN Controller

knows everything and is able to control everything,

so it is like God. Probably this is true.

 

I just want to reiterate, that while security architecture

with central key distribution is definitely feasible and

it is feasible to make it secure, my strong opinion is that

it is still a large step backward from End-to-End security

model and it is much more fragile. And I agree with Tero that

“EE simplicity” argument in most cases doesn’t look 

reasonable to buy. Probably you can add justifications

to this argument, e.g. by providing estimates how much

resources you are going to save on EE if you get rid of IKE

(but leave IPsec, TLS and so on). 

 

Regards,

Valery.

 

From: Gabriel Lopez [mailto:[e-mail]] 
Sent: Wednesday, July 19, 2017 5:21 PM
To: Valery Smyslov
Cc: Alejandro Pérez Méndez; [e-mail]; IPsecME WG; Rafa Marin Lopez
Subject: Re: [I2nsf] [IPsec] draft-abad-i2nsf-sdn-ipsec-flow-protection

 

Hi Valery,

 

El 19 jul 2017, a las 13:54, Valery Smyslov <[e-mail]> escribió:

 

Hi Alejandro,




It is more fragile too. You must perform periodical rekey (update keys)
and this must be done synchronously.

You have to do it by pairs, does not seem that difficult. And, as IKE
does, you create the new ones and, once created, delete the old ones. I
don't see the synchrony problem that important.

In ideal world - yes. In real world - I'm not so sure.
Imagine you have an SA expired and the SDN installs a new SA
on the peers, but one of SDN-peer TLS connection failed because off
the temporary network problem, so you have a new SA partially installed.
What is the peer that didn't receive a new SA supposed to do?
Continue to use an old one despite the fact that it is expired?
Block all traffic? Something else?

In fact, I think the SDN-based approach performs even better than IKE in
this regards.
Imagine what happens if the corresponding IKE rekey process fails for
the very same temporary network problem. In the best case, CHILD SAs are
deleted after a hard expiration, and they will need to be re-created
when triggered by the SPD again. This is roughly identical to the SDN
approach. But, typically, when an IKE rekey fails, the initiator will
likely close the entire IKE_SA thinking the other peer is down, which
would result into having to recreate the IKE_SA (including the DH
exchange), and all the associated CHILD_SAs afterwards.


Exactly, that's what IKE will do. But this is reasonable, because if IKE
messages cannot go between peers, it is most probably that 
IPsec won't go either (especially in case of UDP encapsulation).
With SDN the situation is a bit different. The network problem 
takes place on SDN-EE path, while EE-EE path works well,
but the peers cannot communicate, because SDN fails to provide
the keys in time. Note, that rekey may take place quite often, depending 
on the algorithms involved. 

 

[Gabi] This kind of strong requirements on the controller availability and workload is assumed in the SDN paradigm. Let’s think in a L2 OpenFlow controller for example, where the L2-switch has to forward a copy of the incoming frame before to forward it.







How NAT traversal is to be done in IKE-less case? I understand, that
NATs are also controlled by SDN, but does SDN pre-install NAT mappings?


That's a good question. I would say so, yes.


So, SDN needs to synchronously configure one more entity (NAT)
for IPsec to start working? 

 

[Gabi] If NAT is required the controller should know that, the IPsec configuration required to cross the NAT should be applied by the Security Controller . The configuration of the NAT entity may be configured independently (manually or not, note that there are Yang models for NAT (https://tools.ietf.org/html/draft-sivakumar-yang-nat-07)).









But, would that generate problems if the NAT box is not included in your
SDN (e.g. it belongs to the mall centre or similar)?


In this case you first need to use UDP encapsulation and second need to send
NAT keepalive messages periodically. Usually it is IKE who  sends
these messages (I admit that you can also sends them from the kernel).
IKE also determines if there is a NAT in between and which peer is 
behind the NAT. In case the NAT is out of SDN control, who will do this job?

 

[Gabi] the Controller should know that there is NAT in the scenario.




What is supposed to be done if packet with invalid AH/ESP SPI is received?

 

 





Clearly, the packet itself is dropped, but later IKEv2 extensions (namely
RFC6290, Quick Crash Detection) allows to send IKE notification
to the peer with a security token to help peers quickly recover from the crash.
What is supposed to be done in IKE-less case? Or do you think  that
with SDN such things like non-synchronized IPsec states never happen?

 

[Gabi] If it is an audiatable event by the kernel, the peer can notify the controller about that (some examples of notifications are already included in the yang model)

 

Thanks for the comments, this discussion is really interesting ….

 

Regards, Gabi.


Regards,
Valery. 




These are exactly the sort of situations that need to be figured out.


I believe there are many of them.

Regards,
Valery.

_______________________________________________
I2nsf mailing list
[e-mail]
https://www.ietf.org/mailman/listinfo/i2nsf

 

 

-----------------------------------------------------------
Gabriel López Millán
Departamento de Ingeniería de la Información y las Comunicaciones
University of Murcia
Spain
Tel: [phone]
Fax: [phone]
email: [e-mail]