Re: [OPSAWG] Simplified Alternative to CAPWAP

[Björn Smedman <[e-mail]>]

Hi Rajesh,

On Tue, Feb 25, 2014 at 1:29 AM, Rajesh Pazhyannur (rpazhyan)
<[e-mail]> wrote:
>
> Thanks for the well captured email.  Would like to explore more about what
> you are suggesting.

Thanks, it's much appreciated.

> In the meanwhile,
>
> > 3. The separation of data and control planes that CAPWAP does provide is
> > IMHO incorrect: 802.1X key exchange is considered part of the control plane
> > but the derived encryption keys protect the data plane.
> > This doesn't matter when both control and data plane terminate in the same
> > location (the AC), but becomes a problem when they are separated (encryption
> > keys end up in a different place than the encrypted data!).
>
> Technically speaking, 802.1X protects the 802.11 frames and does not have
> anything to do CAPWAP Data channel. The Data channel is protected using
> DTLS. It is conceivable that 1) the tunnel mode is encapsulating 802.11
> frames (instead of 802.3 frames) and 2) CAPWAP-Data is not encrypted and 3)
> 802.11 encryption is happening at the AC instead of AP. When all three
> conditions hold, what you are saying is true: 802.1X is indeed protecting
> the data plane.

True, in CAPWAP it's a border case that 802.1X key exchange and 802.11
encryption protects the user data plane all the way from the mobile
STA to the AC. But I think there are strong reasons to make this the
default:

* DTLS cannot protect the user data plane inside the WTP. If 802.11
encryption keys and clear-text user data is available inside the WTP
and the WTP is not physically protected then (at least from a
theoretical perspective) the integrity and confidentiality of the data
plane cannot be trusted.

* DTLS encryption is also wasteful: encrypted 802.11 data frames come
in over the air, are unencrypted in the WTP and then encrypted again
with DTLS.

* DTLS also provides authentication, but to me it makes little sense
to authenticate an untrusted party. If the WTP can be under the
physical control of an attacker, then there is no point in
authenticating it against a digital certificate stored in the WTP.

IMHO encrypting the CAPWAP-Data channel with DTLS and authenticating
the WTP is borderline "security by obscurity" in that it relies on an
attacker not being able to gain access to WTP memory. This assumption
may have been relatively true in the corporate WLAN environment that
CAPWAP was designed for, but it's less and less true for carrier Wi-Fi
deployments.

When on the other hand the 802.11 encryption is relied upon to protect
the user data plane the WTP is simply excluded from the security
equation: Even an attacker in complete control of the WTP is no threat
to the integrity or confidentiality of the user data plane (since they
have access to nothing more than the encrypted 802.11 frames which are
also available from the air). That also means there's no real need to
authenticate the WTP, and the security model is greatly simplified.

> > A. The protocol should mandate a single well-defined partitioning of the
> > IEEE 802.11 stack, with 802.1X authentication, key management and encryption
> > all implemented at the tunnel termination point for security reasons.
>
> By this I am assuming that you want encryption at the Tunnel termination
> point rather than the AP. What is motivating this, given that all APs (that
> I know of) implement 802.11 encryption and if security is needed between AP
> and Tunnel termination point then one could use IPSEC or DTLS.

Yes, if piecewise security is sufficient then there are a multitude of
IETF protocols to chose from. But if end-to-end 802.11 encryption all
the way from the mobile STA to a trusted location (e.g. a
service/access router in the network core) is desired then there is
nothing in the IETF playbook except a border case of CAPWAP, which
comes with an entangled control and provisioning/management plane.

Best regards,

Björn

> -----Original Message-----
> From: OPSAWG [mailto:[e-mail]] On Behalf Of Björn Smedman
> Sent: Monday, February 24, 2014 2:57 PM
> To: [e-mail]
> Subject: [OPSAWG] Simplified Alternative to CAPWAP
>
> Dear all,
>
> I've been following the recent CAPWAP activity on this list and thought I
> should share some general thoughts on the protocol, and see if there's
> interest in a broader approach.
>
> The CAPWAP shortcomings discussed are similar to the problems preventing us
> [1] from using it in our Software-Defined Networking
> (SDN) architecture for IEEE 802.11 [2]:
>
> 1. CAPWAP was specified before separation of data, control and management
> planes became the norm in networking architecture.
> Unsurprisingly it provides very little separation of concern, conflating
> everything from firmware updates to configuration of an extra SSID.
>
> 2. CAPWAP attempts to provide provisioning/management functionality that
> overlaps with other IETF standards track protocols, e.g. SNMP and
> NETCONF/YANG, as well as other established management protocols such as
> TR-069.
>
> 3. The separation of data and control planes that CAPWAP does provide is
> IMHO incorrect: 802.1X key exchange is considered part of the control plane
> but the derived encryption keys protect the data plane.
> This doesn't matter when both control and data plane terminate in the same
> location (the AC), but becomes a problem when they are separated (encryption
> keys end up in a different place than the encrypted data!).
>
> 4. CAPWAP leaves many options open to the implementer and is therefore less
> interoperable than desired. To some extent this can be solved with run-time
> negotiation, but such negotiation of security critical aspects (such as
> where encryption keys are derived and kept) makes it impossible to reason
> about the security of the system. This is less than ideal IMHO.
>
> 5. CAPWAP (partly due to 1 and 2 above) depends excessively on details in
> the IEEE 802.11 standard, with each amendment threatening to impact the
> protocol.
>
>
> We would be interested in specifying a simplified tunneling protocol for
> IEEE 802.11 similar to Geneve [3], avoiding the above problems:
>
> A. The protocol should mandate a single well-defined partitioning of the
> IEEE 802.11 stack, with 802.1X authentication, key management and encryption
> all implemented at the tunnel termination point for security reasons.
>
> B. The protocol should mandate a single well-defined frame format relying
> only on a PHY-independent stable subset of IEEE 802.11, ensuring
> compatibility with 11n/ac and (most likely) beyond.
>
> C. Like Geneve the protocol should not provide any provisioning/management
> functionality, instead relying on SNMP, NETCONF/YANG, TR-069 and similar.
>
> D. Like Geneve the protocol should not provide any control plane
> functionality, instead remaining compatible with several Software-Defined
> Networking (SDN) protocols for IEEE 802.11.
>
> E. Each virtual access point in a multi-BSSID (and/or multi-radio) access
> point should be treated as a separate entity associated with a separate
> tunnel.
>
>
> We see strong benefits to such a protocol from an extensibility [4],
> flexibility [5] and security [6] perspective. If there is interest in
> exploring this further please let me know.
>
> Best regards,
>
> Björn Smedman
>
> 1. http://www.anyfinetworks.com
>
> 2. http://anyfi.net and http://www.anyfinetworks.com/products
>
> 3. http://tools.ietf.org/html/draft-gross-geneve-00
>
> 4. http://anyfi.net/documentation#architecture
>
> 5. http://anyfi.net/documentation#a_how_it_all_fits_together
>
> 6. http://anyfi.net/documentation#i_security_model
>
> _______________________________________________
> OPSAWG mailing list
> [e-mail]
> https://www.ietf.org/mailman/listinfo/opsawg
>