doc.: IEEE /1062r0 Submission September 2004 F. Bersani, France Telecom R&DSlide 1 Dominos, bonds and watches: discussion of some security requirements for TGr Florent Bersani, France Telecom R&D
doc.: IEEE /1062r0 Submission September 2004 F. Bersani, France Telecom R&DSlide 2 Goal of this presentation Loren Adams once said: "What is understood need not be discussed" This presentation is about discussing some tentative security requirements listed in document IEEE /10048r0: –Sharing of the PMK –Binding of the PMK –Timeliness of messages... and make sure 09/13 discussion is captured
doc.: IEEE /1062r0 Submission September 2004 F. Bersani, France Telecom R&DSlide 3 Sharing of the PMK Sharing of the PMK is explicitly prohibited by IEEE i: –Page 38, clause item 5 "An IEEE 802.1X AS never exposes the common symmetric key to any party except the AP with which the STA is currently communicating. (...)"
doc.: IEEE /1062r0 Submission September 2004 F. Bersani, France Telecom R&DSlide 4 Sharing of the PMK Sharing of the PMK is explicitly prohibited by IEEE i: –Page 38, clause item 5 ctd. "It implies that the AS itself is never compromised. It also implies that the IEEE 802.1X AS is embedded in the AP, or the AP is physically secure and the AS and the AP lie entirely within the same administrative domain."
doc.: IEEE /1062r0 Submission September 2004 F. Bersani, France Telecom R&DSlide 5 Sharing of the PMK Yet practice does not seem to comply with this requirement (e.g., RADIUS proxying), should TG r? Possible rationale for this requirement: –The domino effect –Sound cryptographic practice
doc.: IEEE /1062r0 Submission September 2004 F. Bersani, France Telecom R&DSlide 6 Sharing of the PMK This is reflected in document IEEE /10048r0, requirements: –#2: "In particular, the pairs and must have cryptographically independent PMKs, or all the i security claims are voided. This rules out sharing PMKs among different APs." –#7: "A PMK shall never be shared between APs"
doc.: IEEE /1062r0 Submission September 2004 F. Bersani, France Telecom R&DSlide 7 Sharing of the PMK The domino effect –Housley, R., "Key Management in AAA", Presentation to the AAA WG at IETF 56, –"Compromise of a single authenticator cannot compromise any other part of the system, including session keys and long-term secrets."
doc.: IEEE /1062r0 Submission September 2004 F. Bersani, France Telecom R&DSlide 8 Sharing of the PMK The domino effect –Compromise of long term secrets would be really painful... but PMKs are not long term (except in PSK mode) –Domino protection is useful when the network has the possibility to inform STA of the compromised AP Could also very well do it for PMKID... if it is not "too broadly shared"
doc.: IEEE /1062r0 Submission September 2004 F. Bersani, France Telecom R&DSlide 9 Sharing of the PMK Sound cryptographic practice –Reusing a nonce can void security properties (e.g., Counter mode) However, if nonce is random then collision probability can be bounded –A key has limited lifetime Sharing it, speeds up burning out. For 128 bit block length, assuming 54 Mbit/s, 0.4*10**14 s with 1 AP –PMK is generally for a short period of time and used in a somewhat conservative scheme
doc.: IEEE /1062r0 Submission September 2004 F. Bersani, France Telecom R&DSlide 10 Sharing of the PMK Anyway what is an AP? Not particularly advocating sharing the PMKs... just debating!
doc.: IEEE /1062r0 Submission September 2004 F. Bersani, France Telecom R&DSlide 11 Binding the PMK This "requirement" appeared during discussion of IEEE /10048r0 Perhaps alluded to in requirement #9 "The only visible identifier seems to be BSSID." The idea would be that: –A PMK should only be used by a pair –This restriction should be "incorporated" in the PMK
doc.: IEEE /1062r0 Submission September 2004 F. Bersani, France Telecom R&DSlide 12 Binding the PMK Is the PTK bound to something? –It incorporates nonces and MAC addresses... –Yet this does not per se preclude sharing of the PTK! How can we prevent two parties that have agreed to do so to abuse a key usage??? What does "binding the keys" mean? –EAP channel binding?
doc.: IEEE /1062r0 Submission September 2004 F. Bersani, France Telecom R&DSlide 13 Timeliness of messages This is reflected in document IEEE /10048r0, requirements –#11 "An AP shall test the liveness of the mobile STA at reassociation when the mobile STA does a secure fast transition to it. This is required to synchronize replay counters." –#12. "A mobile STA shall test the liveness of the AP at reassociation when it does a secure fast transition to a new AP. This is required to synchronize replay counters"
doc.: IEEE /1062r0 Submission September 2004 F. Bersani, France Telecom R&DSlide 14 Timeliness of messages The attack scenario: CCMP protected frame, PN=23: "Sell stocks, now!" Attacker delays the frame (and possibly subsequent traffic) This is not a replay: client only sees the delayed frame once CCMP protected frame, PN=23: "Sell stocks, now!"
doc.: IEEE /1062r0 Submission September 2004 F. Bersani, France Telecom R&DSlide 15 Timeliness of messages i does not include such a feature, why: –Hard to do? (timestamps,...) –Not necessary? (client won't stay associated with big delays) Should TGr? –What if pre-auth is allowed to go up to deriving a PTK: there can be some time between derivation and usage –Avoid DoS and black-holes
doc.: IEEE /1062r0 Submission September 2004 F. Bersani, France Telecom R&DSlide 16 Timeliness of messages Solution alluded to to provided timeliness: periodically make a protected resynch of replay counters –Kind of a key confirmation –Tradeoff between the frequency of this resynch and the "timeliness" protection
doc.: IEEE /1062r0 Submission September 2004 F. Bersani, France Telecom R&DSlide 17 Conclusion TG r PAR: "Security must not be decreased as a result of the enhancement." To what extent has TG r to comply with TG i? –Is there an objective measure of security diminution? –TG r can break some of TG i assumptions, yet prove that it remains secure... H2 make Apples to apples comparison between securtiy of the proposals?