Wiki
Clone wikimeetings / 140616_webex_security
Minutes Webex 16 June 2014, 6TiSCH Security Design Team
Note: timestamps in PDT.
Taking notes (using Etherpad)
- Pascal Thubert
- Michael Richardson
- Thomas Watteyne
Present (alphabetically)
- Giuseppe Piro
- Jonathan Simon
- Maik Seewald
- Michael Richardson
- Nancy Cam-Winget
- Pascal Thubert
- Pat Kinney
- Rene Struik
- Subir Das
- Thomas Watteyne
- Tom Phinney
- Yoshihiro Ohba
Recording
- Webex recording (audio+slides,streaming)
- https://cisco.webex.com/ciscosales/lsr.php?RCID=94dcaeeaca494567b83960d2d7cbad62 [61min]
Slides
- No slides shared through Webex, reusing slides from last call
- https://bitbucket.org/6tisch/meetings/src/master/140609_webex_sec/140609_webex_sec.pdf
Agenda
- Announce ANIMA/UCAN BOF.
http://trac.tools.ietf.org/bof/trac/wiki/WikiStart#UCANApprovedforIETF90
- work on trust/cryptographic details of Joining-Node -> PCE message.
- does it need to be signed?
- 6top ideas for PCE->joining node protocol. Will this work as stock
- CoAP/DTLS, or do we need a proxy to help us?
Minutes
- [07.07] Meeting starts
- [MCR] continue Mozilla Etherpad
reco
- [Michael]
- let's use slides from last week
- new BoF at IETF90: ANIMA/UCAN
http://trac.tools.ietf.org/bof/trac/wiki/WikiStart#UCANApprovedforIETF90
- Core is Michael Behringer's autonomic draft, people should be aware of this effort
- Autonomic Networking Integrated Model and Approach (https://www.ietf.org/mailman/listinfo/anima)
- wanted to talk about messages from slide 3
- [Michael] messages 4 and 5. Need to understand properties of these messages. Have to either encrypt or sign by PSK. If we have private key, why is message 4 encrypted to PCE? Is it enough to sign by joining node? what's the freshness criteria?
- [Rene] what is fresh?
- [Michael] Message 4 could be pre-calculated by joining node long before.
- [Rene] First message by joining node, no guarantee will be fresh.
- [Michael] One of the elements of that message is neighbor info. We could have a nonce based on ASN.
- [Rene] You cannot have any assurance about first packet. The PCE responds.
- [Michael] There could be some freshness in it.
- [Michael] Model trying to see whether possible, is where there is a long period of time between messages 4 or 5.
- [Michael] 5 could start a CoAP/DTLS session, there is no strong relationship between 4 and 5.
- [Michael] PCE could manage order of joining.
- [Michael] If message 4 does not have to be fresh, it can be periodically resent until the node joins the network.
- [Rene] suggesting joining node is sending
- [Thomas] is there any benefits in repeating> that's wasted bandwidth.
- [Michael] if we have an ACK we can avoid it
- [Thomas] The proxy ACKs at L2, it is responsible for the packet after that.
- [Michael] what if lost somewhere up?
- [Thomas] can happen. There can be a very long retry timeout.
- [Michael] agreed. if too many nodes are turned on, the PCE can manage the order and may go out to vendor and get certificate or a security token to complete the join process. Open loops possible, e.g. manual authorization by a human.
- [Michael] We do not want to have to hit a button or something like that.
- [Thomas] Strictly speaking, the joining node is able to keep sync while waiting for reply, keeping in sync with the proxy node and waiting for join response.
- [Michael] Question: is there a reason for message 4 to be encrypted to PCE?
- [Thomas] In a PSK case, encrypting msg 4 shows that the node has the credentials.
- [Michael] Address of PCE could use built into certificate. Is this why encrypt to PCE?
- [Jonathan] came from WirelessHART case, where confidential info is placed there (e.g. HART-specific configuration info). These device details are not publicly broadcast e.g. what kind of device/capabilities.
- [Michael] is there something that would need to be kept private (in the future?). If we do not encrypt to PCE we do not need messages 2 and 3
- [Thomas] this is used for the node to know who it is talking to. Else it could take a long while to figure this is the wrong network. This is really a caching mechanism.
- [Michael] The address of the PCE may well be in the certificate as an attribute.
- [Michael] Certificate response in step 3 cannot contain authentication token that allows node to know which network it joins. This is between messages 5 and 6 when certificate chains or authorization tokens arrive. The certificate response could contain much of the certificate chain but not all. The joining node could authenticate but figure that there is a relationship with vendor.
- [Tom] Example: many networks in a refinery with same owner, since the ExxonMobil owns the whole place. This granularity has to be coming from a tag. Out of the box the device cannot know which part of the network it needs to join.
- [Nancy] we want to endpoint to be provisioned with right certificates. In utility sector service provider (SP) may be like a PG&E, can be provisioning the smart meters with root certificates that indicate which SP network to join. Consumer is very different.
- [Tom] device does not know with the granularity of which block inside the network. Need to provision e.g. with a link back to central system
- [Thomas] to combine the 2, if the device joins the wrong network, PCE can fix that.
- [Tom] can see the joining device probing, trial and error and blacklist networks but that places the burden on the device. The right info is only available in some central system.
- [Jonathan] in HART it is all initiated by joining device
- [Tom] has to, but must come with some ID. Device provisioning may be done by ExxonMobil and then put in the field with a tag name that allows to correlate the device with its role (e.g. street address in smart meters). Devices have differentiable characteristics so confidentiality of the join process is illusional.
- [Michael] concerned about device intercepted on the way to plant. Are we adding some defense by having the device figure the place is correct or not?
- [Tom] would make that an option. Large system owner would tag device for tracking. Need also to look at home owner. No pre-provisioning, single step. Enable security system to add a new device, or manually add a serial number.
- [Michael] not chartered so mush for that in 6TiSCH, expect a network manager. Not to go too heavily towards home.
- [Jonathan] there is an installer that at least can read a do and feed back a central office.
- [Tom] there has to be something that binds the role with the device, e.g. which pipe a corrosion monitor is placed on.
- [Michael] that can be done out-of-band.
- [Tom] yes, there is a mechanism whereby installer says I'm adding that device for that function.
- [Thomas] Big plant case. Out of band provisioning of some data base. Device starts with info about this is the network device must join. See a number of potential proxies. During the real handshake the networks can reject where it is not expected.
- [Michael] could PCEs from different networks communicate to answer to the device which proxy it should join?
- [Tom] device needs positive/negative information to rejoin faster and not retry the wring networks. PANID can be used.
- [Thomas] How does joining node identify network?
- [Tom] device knows which network sent EB.
- [Michael] can be forged.
- [Tom] in large companies, many businesses side-by-side interfering. Lots of rejections in such scenarios.
- [Thomas] PANID is a great filter but not security.
- [Michael] step 2 must indicate what type of certificate chain is expected and proxy will answer I'm ExxonMobil and that root says so.
- [Tom] yes, depending on timing when the device validates the chain.
- [Michael] 7/11 could have purchased the same devices so serial numbers are needed to validate the right device in the right network.
- [Jonathan] proxy can cache some certs.
Updated