802.

11i Security Analysis:

Can we build a secure WLAN?

Changhua He
Stanford University
March 24th, 2005
[email protected]
 Outline
 WLAN Security Overview
 Wireless Local Area Networks
 Wireless threats
 Wireless security history
 IEEE 802.11i Standard
 Finite-state Verification of 4-Way Handshake
 More Insecurities and Improvements
 Modular Proof of Security
 Conclusions
802.11 Infrastructure Network

Wired
Network

Can we build a secure WLAN ?

 Wireless Threats
 Passive Eavesdropping/Traffic Analysis
 Easy, most wireless NICs have promiscuous mode
 Message Injection/Active Eavesdropping
 Easy, some techniques to gen. any packet with common NIC,
exploit MAC cooperation to interfere in a timely way
 Message Deletion and Interception
 Possible, interfere packet reception with directional antennas
 Masquerading and Malicious AP
 Easy, MAC address forgeable and s/w available (HostAP)
 Session Hijacking
 Man-in-the-Middle
 Denial-of-Service Attack
 History of Wireless Security
 802.11 (Wired Equivalent Protocol)
 Authentication: Open system (SSID) and Shared Key
 Authorization: some vendor use MAC address filtering
 Confidentiality/Integrity: RC4 + CRC
 Completely insecure [Walker00,Arbaugh01,Wagner01,FMS01 …]
 WPA: Wi-Fi Protected Access
 Authentication: 802.1X
 Confidentiality/Integrity: TKIP
 Reuse the legacy hardware, still problematic [Arbaugh02 …]
 Availability is a big problem
 Frequency jamming: inevitable but expensive and detectable
 Network and upper Layer: depend on protocols
 Link Layer DoS attack: unprotected management frames
[Arbaugh et al 01] [Bellardo et al 03] [Chen04] …
 Outline
 WLAN Security Overview
 IEEE 802.11i Standard
 Data confidentiality and integrity
 Mutual authentication
 Key management protocols
 Finite-state Verification of 4-Way Handshake
 More Insecurities and Improvements
 Modular Proof of Security
 Conclusions
 IEEE 802.11i
 Ratified on June 24, 2004
 Data confidentiality and integrity
 Encryption in Data Link Layer
 WEP: Wired Equivalent Privacy
 TKIP: Temporal Key Integrity Protocol
 CCMP: Counter-mode/CBC-MAC Protocol
• Long term solution, need hardware upgrade
• With a fresh key, 802.11i CCMP is believed secure for data
confidentiality and integrity !
 Mutual authentication
 RSNA: Robust Security Network Association
 EAP-TLS/802.1X/RADIUS
 Key management
 4-Way handshake, Group key handshake
 RSNA Establishment
 RSNA Establishment Procedures
 Network and Security Capability Discovery
 802.11 Open System Authentication and Association
 EAP/802.1X/RADIUS Authentication
 4-Way Handshake
 Group Key Handshake
 Secure Data Communications
 Our security analysis gives:
 can provide satisfactory authentication and key management
 could be problematic in Transient Security Networks (TSN)
 reflection attack could be possible if not implemented correctly
 Availability is still problematic
 A Complete Conversation
Supplicant Authenticator Authentica-
UnAuth/UnAssoc
Auth/Assoc UnAuth/UnAssoc
Auth/Assoc tion Server
802.1X UnBlocked
Blocked 802.1X UnBlocked
Blocked (RADIUS)
No Key
MSK
PMK
New
PTK/GTK
GTK No Key
PMK
New
PTK/GTK
GTK MSK
No Key

802.11 Association

EAP/802.1X/RADIUS Authentication
MSK

4-Way Handshake

Group Key Handshake

Data Communication
 Outline
 WLAN Security Overview
 IEEE 802.11i Standard
 Finite-state Verification of 4-Way Handshake
 Murφ Model Checker
 Modeling the 4-way handshake
 Attacks and solutions
 More Insecurities and Improvements
 Modular Proof of Security
 Conclusions
 Finite-State Verification
 Powerful methodology
 Used in protocol and software verifications
 No requirements on implementations
• Comparing to static analysis and dynamic tracing
 Challenges
 Requires complete understanding of what you check
 State space reduction techniques
 Many tools available
 Generic model checking:
• Murφ, Spin, SMV
 Automatic model generation model checking:
• Pathfinder, Bandera, Verisoft
 Murφ Model Checker

 Murφ rules for protocol

participants and the intruder
define a nondeterministic state
transition graph
...  Murφ will exhaustively
... enumerate all graph nodes
 Murφ will verify whether
specified security conditions
hold in every reachable node
Correctness  If not, the path to the violating
condition violated node will describe the attack
 Running Murφ

Informal
Formal Intruder
Protocol
Protocol Model
Description

Murj code Murj code,

RFC, IEEE Std. similar for
all protocols

Analysis
Find error Tool

Specify security
conditions and run Murj
 Diagnosing Errors

 Bad abstraction
 Removed too much detail from the protocol when
constructing the abstract model
 Add the piece that fixes the bug and repeat
 Genuine attack
 Yay! Hooray!
 Attacks found by formal analysis are usually quite strong
• Independent of specific cryptographic schemes
• Independent of implementations, etc.
 Test an implementation of the protocol, if available
 The 4-Way Handshake
Supplicant Authenticator
Auth/Assoc Auth/Assoc
802.1X Blocked 802.1X Blocked
PMK PMK
AA, ANonce, sn, msg1

PTK Derived
SPA, SNonce, SPA RSN IE, sn, msg2, MIC

PTK Derived
Random GTK
AA, ANonce, AA RSN IE, GTK, sn+1, msg3, MIC
SPA, sn+1, msg4, MIC

PTK and GTK PTK and GTK

802.1X Unblocked 802.1X Unblocked
Modeling the 4-Way Handshake
 Authenticators/Supplicants:
 An association is maintained between a pair of authenticator
and supplicant
 Each association has a uniquely shared PMK
 Multiple sequential legitimate handshakes in one association
 Intruder
 Impersonate both supplicant and authenticator
 Eavesdrop, intercept and replay messages
 Compose messages with known nonce and MIC
 Forge fresh Message 1
 Predict and replay nonces for pre-computation of MIC
 Invariants
 Represent the security property, must be true for each state
 Authenticator Msg 1 (Murj)
ruleset i: AuthenticatorId do
ruleset j: AgentId do
rule 20 "Authenticator sends Message 1 to associated Supplicants"
!ismember(j, AuthenticatorId) & -- no message to Authenticators
!ismember(j, IntruderId) & -- no message to Intruders
multisetcount(l:net, true) < NetworkSize &
(aut[i].associations[j].session.state = A_PMKSA |
aut[i].associations[j].session.state = A_DONE)
==>
var
outM: Message; -- outgoing message
begin
outM.source := i;
outM.dest := j;
outM.mtype := M_1;
outM.nonce := freshNonce();
outM.sequence := freshSequence();
outM.address := i;
multisetadd(outM, net);
aut[i].associations[j].peer := j;
aut[i].associations[j].pmk := usePMK(i, j);
aut[i].associations[j].sequence := outM.sequence;
aut[i].associations[j].nonce := outM.nonce;
aut[i].associations[j].session.state:= A_WAITFORNS;
end;
end;
end;
Invariant: Security Properties

invariant "PTKs are consistent and fresh"

forall i: AuthenticatorId do
forall j: SupplicantId do
aut[i].associations[j].session.state = A_DONE

->

(sup[j].associations[i].session.state = S_DONE &

ptkEqual(aut[i].associations[j].session.ptk,
sup[j].associations[i].session.ptk) &
aut[i].associations[j].sid = sup[j].associations[i].sid) |
(sup[j].associations[i].session.state = S_PTKSA &
aut[i].associations[j].sid <= sup[j].associations[i].sid)
end;
end;
 Forged Message 1 Attack
Supplicant Authenticator
Auth/Assoc Auth/Assoc
802.1X Blocked 802.1X Blocked
PMK PMK
AA, ANonce, sn, msg1

PTK Derived
SPA, SNonce, SPA RSN IE, sn, msg2, MIC

AA, ANonce, sn, msg1 PTK Derived

Random GTK
AA, ANonce, AA RSN IE, GTK, sn+1, msg3, MIC
SPA, sn+1, msg4, MIC
AA, ANonce, sn, msg1

PTK and GTK PTK and GTK

802.1X Unblocked 802.1X Unblocked
 4-Way Handshake Blocking
Supplicant Authenticator
Auth/Assoc Auth/Assoc
802.1X Blocked 802.1X Blocked
PMK PMK
AA, ANonce, sn, msg1

PTK Derived
SPA, SNonce, SPA RSN IE, sn, msg2, MIC
AA, ANonce[1], sn, msg1 PTK Derived
Random GTK
AA, ANonce[n], sn, msg1

AA, ANonce, AA RSN IE, GTK, sn+1, msg3, MIC

SPA, sn+1, msg4, MIC

PTK and GTK PTK and GTK

802.1X Unblocked 802.1X Unblocked
4-Way Blocking: Solution 1

Random-Drop
Queue:

Randomly drop a
stored entry to
adopt the state
for the incoming
Message 1 if the
queue is filled.

Not so effective
 4-Way Blocking: Solution 2

 Authenticate Message 1
 To reuse the algorithms/hardware, set nonces to special
values, e.g., 0, and derive PTK.
 Calculate MIC for Msg 1 using the derived PTK
 Good solution if PMK is fresh
 If PSK and cached PMK, replay attacks !
 Add a monotonically increasing global sequence counter
 Use local time in authenticator side
 Sufficient space in Message 1 ( 8-octet sequence field )
 No worry about time synchronization

Modifications on packet format

 4-Way Blocking: Solution 3
 Re-Use Nonce
 Supplicant re-use SNonce until one 4-way handshake
completes successfully
 Derive correct PTK from Message 3
 Authenticator may (or may not) re-use ANonce
 Solve the problem, but
 More computations in the supplicant
 Attacker might gather more infomation about PMK by
playing with Message 1, ok if PMK is strong
PTK=PRF{PMK, AA||SPA||ANonce||SNonce}

Performance Degradation
 4-Way Blocking: Solution 4
 Combined solution
 Supplicant re-use SNonce
 Store one entry of ANonce and PTK for the first Message 1
 If nonce in Message 3 matches the entry, use PTK directly;
otherwise derive PTK again and use it.
 Advantages
 Eliminate the memory DoS attack
 Ensure performance in “friendly” scenarios
 Only minor modifications on the algorithm in the Supplicant
• No modifications on the packet format
 Adopted by TGi
 Simple solution, but not immediate
 Practical considerations, not designing a new protocol
 Reflection Attack
Adversary Legitimate
Impersonates Devices
Communicating Authenticator and
Peers Supplicant
{A1, Nonce1, sn, msg1}
{A2, Nonce1, sn, msg1}
{A1, Nonce2, RSN IE, sn, msg2, MIC}
{A2, Nonce2, RSN IE, sn, msg2, MIC}
{A1, Nonce1, RSN IE, GTK, sn+1, msg3, MIC}
{A2, Nonce1, RSN IE, GTK, sn+1, msg3, MIC}
{A1, sn+1, msg4, MIC}
{SPA, sn+1, msg4, MIC}

Bogus Authentication Peers Authenticated

 Reflection Attack: Solutions
 Possible in ad hoc networks
 Each participant plays the role of both authenticator and
supplicant
 Less damage if strong confidentiality adopted
 Adversary fool the peers to send packets
 Cannot decrypt the packet and generate response
 Violate the mutual authentication concept
 Solutions:
 Restrict each participant to play only one role: ok for WLAN,
but inappropriate for ad hoc networks
 Each participant play both roles, but under different PMK
 4-Way Handshake Summary
 Finite-state verification
 Find subtle bugs, suitable for verifying “small” protocols
 State-space explosion is a problem
 Message 1 vulnerability
 Forge first message to block the protocol
 TPTK/PTK to prevent this, but only do partial work
 Store all states, but queue size is a problem
 Final solution adopted by Standards Committee
 Reflection attack
 Possible in ad hoc scenario
 Each entity plays only one role, or both roles under
different PMK
 Outline
 WLAN Security Overview
 IEEE 802.11i Standard
 Finite-state Verification of 4-Way Handshake
 More Insecurities and Improvements
 Poisoning and rollback attacks
 TKIP Michael Countermeasure problem
 Failure recovery
 Modular Proof of Security
 Conclusions
 RSN IE Poisoning
Supplicant Authenticator
Unauthenticated Unauthenticated
Unassociated Unassociated
802.1X Blocked 802.1X Blocked
(1) Beacon + AA RSN IE
Bogus Beacon + Modified RSN IE
(2) Probe Request
(3) Probe Response + AA RSN IE
Bogus Probe Response + Modified RSN IE

Legitimate Message Exchanges

(18) {AA, ANonce, AA RSN IE, GTK, sn+1, msg3, MIC}

RSN IE confirmation Disassociate the

failed, Disassociation Supplicant
 RSN IE Poisoning: Solutions
 Easy to launch the attack
 Legitimate participants unaware of it
 Continue message exchanges, waste resources
 Adversary have more time to repeat the attack
 Solutions
 Authenticate management frames
• Difficult to authenticate Beacon and Probe Response frame
 Confirm RSN IE as soon as possible (EAP-TLS)
• Necessary modifications on the standard
 Relax the condition of RSN IE confirmation
• Ignore insignificant bits, only confirm authentication suite
• If authentication suite modified, probably fails at the
beginning of associations and retry new APs
Security Level Rollback Attack
Supplicant Authenticator
RSNA enabled RSNA enabled
Pre-RSNA enabled Pre-RSNA enabled

Bogus Beacon (Pre-RSNA only)

Beacon + AA RSN IE
Probe Request
Bogus Probe Response (Pre-RSNA only)
Probe Response + AA RSN IE
802.11 Authentication Request
802.11 Authentication Response
Bogus Association Request (Pre-RSNA only)
Association Request + SPA RSN IE
802.11 Association Response
Pre-RSNA Connections
 Security Rollback: solutions
 Security Level Rollback Attack
 Similar to general version-rollback attack
 Destroy the security since WEP is completely insecure
 Not a real vulnerability of 802.11i standard, but an
implementation problem of TSN
 Very possible mistake if requires transparency in hybrid
configurations
 Solutions
 Allow only RSNA connections: secure, but too strict for
common network systems, where TSN is more convenient
 Adopt both, supplicant manually choose to deny or accept
a connection, authenticator restrict pre-RSNA (WEP)
connections to only insensitive data
 Michael Countermeasure
 TKIP Michael algorithm and countermeasures
 Michael algorithm provides 20-bit security for MIC
 one successful forgery / 2 min., need countermeasures
 802.11i documentation proposes:
• Cease communication for 60 sec. if two Michael MIC failures
detected in one minute, re-key & deauthentication
• Limit to one successful forgery / 6 month
• Verify in strict order: FCS < ICV < TSC < MIC
• Update TSC unless MIC is validated

MAC IV/KeyID Ext. IV Data/MSDU MIC ICV FCS

Contains TSC Encrypted

MIC: Message Integrity Code
TKIP MPDU Format
ICV: Integrity Check Value (from WEP)
FCS: Frame Check Sequence
TSC: TKIP Sequence Counter
 Michael DoS and Solutions
 Still DoS attack through MIC failures !
 Intercept a packet with valid TSC (possible)
 Modify packet and corresponding values of FCS, ICV to get
a packet with valid FCS, ICV, TSC but invalid MIC (easy)
 Send modified packet twice in one minute (easy)
 MIC always invalid, TSC always valid
 Solutions
 When MIC failure, cease communication only, no re-keying
and deauthentication
 Update TSC before MIC is validated
 What happens if modify TSC to extremely large number?
• Change TSC also change encryption key, wrong decryption
• Some confidence on TKIP key schedule algorithm
 Mitigation but not elimination
 Management Frame att. & Sol.
 DoS attacks on plain 802.11 networks
 Forge unprotected management frames, like
Deauthentication/Disassociation [Bellardo et al 03]
 Exploit virtual carrier sense mechanism by forging
unprotected control frames, like RTS/CTS etc. [Bellardo 03]
 802.11i still has these problems, solutions could be
• Authenticate management frames
• Validate virtual carrier sense in control frames
 DoS attacks on EAP messages
 Forge EAPOL-Start, EAPOL-Success, EAPOL-Logoff,
EAPOL-Failure
 802.11i can eliminate these by simply ignoring them !
 Send more than 255 association request to exhaust the EAP
identifier space (8 bits)
 Adopt separate EAP identifier counter for each association

[Arbaugh et al 01] [Lynn02] [Moore02] [Bellardo et al 03] [Chen04] …

 Failure Recovery
 Important for large protocols like 802.11i
 Not affect protocol correctness, but efficiency
 Not eliminate DoS vulnerabilities, but make DoS more difficult
 802.11i adopts a simple scheme
 Whenever failure, restart from the beginning, inefficient !
 What about restart from nearest point ?
 Tradeoffs
• Defensive DoS attack vs Captured DoS attack
• Assumptions on adversary’s capability and network scenario
 A better failure recovery for 802.11i
 If failure before 802.1X finishes, restart everything
 Otherwise restart components from nearest point
 Confidence on 802.1X authentication, mobile user?
• channel scanning time >> protocol execution time
 Summarize Insecurities
 Components insecurities
 The 4-Way Handshake: blocking and reflection
 TKIP Michael Countermeasure attacks
 Combine 802.11i as a whole
 RSN IE Poisoning Attack: waste resources
 Security Level Rollback Attack: for hybrid configuration
 Attacks on unprotected management frames
 Inefficient failure recovery
 Improved variant of 802.11i
 Mitigate all discussed vulnerabilitites
 Need more modifications on existing codes
Improved 802.11i Architecture
Stage 1: Network and Security Capability Discovery

Stage 2: 802.1X Authentication

(mutual authentication, shared secret, cipher suite)
802.1X Failure

Stage 3: Secure Association (management frames protected)

Association Failure
Stage 4: 4-Way Handshake
(PMK confirmation, PTK derivation, and GTK distribution)
4-Way Handshake Timout

Stage 5: Group Key Handshake

Group Key Handshake Timout

Stage 6: Secure Data Communications

Michael MIC Failure or Other Security Failures
 Outline
 WLAN Security Overview
 IEEE 802.11i Standard
 Finite-state Verification of 4-Way Handshake
 More Insecurities and Improvements
 Modular Proof of Security
 Methodology
 Protocol compositional logic
 Proved theorems
 Conclusions
 Motivations & Difficulties
 Proof of security is necessary and useful
 “Head-scratching” approach is inconvincible
 Finite-state verification is incomplete
 Prove security vs find bugs
 Difficulties
 How to model the protocol?
 How to define security?
 How to prove?
 There are already some formal methods to do these!
• [BAN90], [Paulson97], [Strand98], [Spi98], … …
 IEEE 802.11i
 a large protocol, complicated control flows
 Compositional Logic, originated from [DMP01]
 Methodology
 Divide-and-conquer paradigm in security
 Divide the large protocol to several components
 Use protocol logic to reason the security properties of each
subprotocol
 Study the compositionality of all components
 802.11i components
 802.11 Association: a physical connection assumed
 EAP/802.1X/RADIUS Authentication
 4-Way Handshake
 Group Key Handshake
 Secure Data Communications: CCMP believed secure
 Protocol Logic: Intuition

Protocol Honest Principals,

Attacker

Private
Data
 Alice’s information
 Protocol
 Private data
 Sends and receives
 Formalizing the Approach
 Abstraction based on Dolev-Yao Model [1983]
 “Black-box” cryptography
 No partial knowledge, no statistical tests
 Language for protocol description
 Arrows-and-messages are informal
 “Cord” program for each protocol role: terms & actions
 Protocol Semantics
 How does the protocol execute?
 Protocol, initial configuration, run
 Protocol logic
 Stating security properties
 Proof system
 Formally proving security properties
 Cords: 4-Way Handshake
Authenticator = Supplicant =
(X, Y, PMKXY)[ (Y, PMKXY)[
new x; receive X, Y, z;
send X, Y, x, m1; match z/x, m1;
receive Y, X, z; new y;
match z/y,m2,Hash(PTKXY,y,m2); send Y, X, y, m2,Hash(PTKXY,y,m2);
send X,Y,x,m3,Hash(PTKXY,x,m3); receive X, Y, z;
receive Y, X, z; match z/x,m3,Hash(PTKXY,x,m3);
match z/m4, Hash(PTKXY, m4) send Y, X, m4, Hash(PTKXY, m4)
]X ]Y
Authenticator: 4-Way Handshake
Pre-condition
Has(X, PMKXY)  Has(Y, PMKXY)  (Has(Z, PMKXY)  (Z = X  Z = Y))
Secret Key Agreement
Honest(X)  Honest(Y) 
Has(X, PTKXY)  Has(Y, PTKXY)  (Has(Z, PMKXY)  (Z = X  Z = Y))
Session Authentication
Y. Honest(X)  Honest(Y) 
Send(X, X, Y, x, m1)  Receive(Y, X, Y, x, m1) 
Send(Y, Y, X, y, m2, Hash(PTKXY,y,m2)) 
Receive(X, Y, X, y, m2, Hash(PTKXY,y,m2)) 
Send(X, X, Y, x, m3, Hash(PTKXY,x,m3)) 
Receive(Y, X, Y, x, m3, Hash(PTKXY,x,m3)) 
Send(Y, Y, X, m4, Hash(PTKXY, m4)) 
Receive(X, Y, X, m4, Hash(PTKXY, m4))
 Protocol Composition
 Composition
 Convenient for analyzing large protocols and systems
 Not only for proving things, but also for building systems
 Non-destructive combination
 Ensure that the combined parts do not degrade each
other’s security
 Assumptions about the environment: invariance assertions
 Additive combination:
 Accumulate security properties of combined parts,
assuming they do not interfere
 Properties achieved by individual protocol roles
 Invariants of one component must be
satisfied by any other components
 802.11i Proved Properties

 EAP-TLS
 Mutual authentication: actions matched in order
 Secret key achieved: known and only known to peers
 The 4-Way Handshake
 Session authentication: actions matched in order
 Secret key agreement: fresh PTK derived
 Group Key Handshake
 Session authentication: actions matched in order
 Secret group key: group key distributed
 802.11i components compose safely
 Answering the question

 Can we build a secure WLAN? (with 802.11i)

 Yes !
 Limitations
 “Security” means “mutual authentication” and “secret key”
 If CCMP secure, also have data confidentiality and integrity
 Assume perfect “black-box” cryptography
 Assume legitimate entities are honest
 Only from perspective of the protocol
 No consideration on implementations
• Software bug, buffer overflow …
 No consideration on the whole system
• IPsec or firewall implemented?
• SSL application?
 Conclusions
 Finite-state verification of 802.11i components
 Very useful methodology to find bugs
 Vulnerabilities found in the 4-Way Handshake
 Attacks and solutions on 802.11i
 On the 4-Way Handshake: blocking and reflection
 On RSN IE: poisoning and rollback
 On the TKIP Michael Algorithm
 Failure recovery
 Improved 802.11i
 It is really secure
 Prove “mutual authentication” and “secret key” properties
 We CAN build a secure WLAN !
Questions?

[http://www.pacwireless.com]
Temporal Key Integrity Protocol
 Re-use legacy hardware: RC4 for encryption
 Michael algorithm for MIC
 “Dolev-Yao” Model

 Inspired by their 1983 paper

 D. Dolev and A. Yao. “On the security of public key protocols”. IEEE
Transactions on Information Theory, 29(2):198-208.
 Adversary is a nondeterministic process
 Can read any message, decompose it into parts and re-assemble
 Cannot gain partial knowledge, perform statistical tests, …
 “Black-box” cryptography
 Adversary can decrypt if and only if he knows the correct key
 Assumes that cryptographic functions have no special properties
 Most mechanized formal methods for security
analysis use some version of this model
 Typical Dolev-Yao Term Algebra
Attacker’s term algebra is a set of derivation rules

vT Tu Tv Tu Tv

if u=v for some 
Tu T[u,v] Tcryptu[v]

T[u,v] T[u,v] Tcryptu[v] Tu

Tu Tv Tv

In the real world, there is no guarantee

that attacker is restricted to these
operations! He may perform probabilistic
operations, learn partial information, etc.
Cords: 4-Way Handshake
Security properties
Sample Proof