WEP2 Security Analysis

1. May 2001 Bernard Aboba, Microsoft Slide 1 WEP2 Security Analysis Bernard Aboba Microsoft

2. May 2001 Bernard Aboba, Microsoft Slide 2 Goals To (briefly) summarize security weaknesses discovered in WEP v1.0 To analyze security vulnerabilities of WEP2 To recommend potential improvements

3. May 2001 Bernard Aboba, Microsoft Slide 3 Classes of Attacks Against WEP v1.0 IV (key) reuse [Walker, Berkeley team, Arbaugh] Made possible by small IV space in WEPv1.0, lack of IV replay protection Enables statistical attack against ciphertexts w/replayed IVs Known plaintext attack [Walker, Berkeley team, Arbaugh] Lots of known plaintext in IP traffic: ICMP, ARP, TCP ACK, etc. Can send pings from Internet through AP to snooping attacker Enables recovery of key stream of length N for a given IV Can forge packets of size N by reusing IV in absence of a keyed MIC Partial known plaintext [Berkeley team, Arbaugh] May only know a portion of the plaintext (e.g. IP header) Possible to recover M octets of the keystream, M < N Via repeated probing, can extend keystream from M to N [Arbaugh] Possible to flip bits in realtime, adjust CRC32, divert traffic to attacker Enabled by linearity of CRC32, absence of keyed MIC

4. May 2001 Bernard Aboba, Microsoft Slide 4 Classes of Attacks (cont�d) Authentication forging [Berkeley team] WEP v1.0 encrypts challenge using IV chosen by client Recovery of key stream for a given IV enables re-use of that IV for forging WEP v1.0 authentication Does not provide key, so can�t join LAN Denial of service Disassociate, reassociate messages not authenticated Dictionary attack Possible where WEP keys derived from passwords Realtime decryption [Berkeley team, Arbaugh] Repeated IV reuse, probing enables building of a dictionary of IVs, key streams Enables decryption of traffic in realtime Possible to store dictionary due to small IV space Need 1500 octets of key stream for each IV 2^24 * 1500 octets = 24 GB

5. May 2001 Bernard Aboba, Microsoft Slide 5 WEP2 Increases size of IV space to 128 bits Key may be changed periodically via IEEE 802.1X re-authentication to avoid staleness No keyed MIC No authentication for reassociate, disassociate No IV replay protection Use of Kerberos for authentication within IEEE 802.1X

6. May 2001 Bernard Aboba, Microsoft Slide 6 WEP2 Security Analysis IV (key) reuse Larger IV, re-key support makes unintentional reuse much less likely Without IV replay protection, intentional reuse still possible Known/Partial plaintext attacks Not affected by larger IV Probing, key stream extension still possible in absence of keyed MIC Still possible to recover key streams via ping from Internet Can still forge packets by reusing IV, key stream Can still divert traffic in absence of non-linear, keyed MIC Authentication forging attack Not affected by larger IV, since intentional IV replay still possible Dictionary attack New vulnerabilities introduced by mandatory KerberosV authentication Realtime decryption Much more difficult due to larger IV 2^128 * 1500 octets = 5.1E32 GB

7. May 2001 Bernard Aboba, Microsoft Slide 7 KerberosV Dictionary Attack Vulnerabilities References Bellovin & Meritt �Limitations of the Kerberos authentication system�, USENIX 1991 Wu, T. �A Real-World Analysis of Kerberos Password Security�, 1998 http://theory.stanford.edu/~tjw/krbpass.html Scenario Attacker snoops AS_REQ/AS_REP exchange, recovers passwords offline In popular 802.11 networks (�hot spots�), may be possible to collect many such exchanges in a single attempt Vulnerabilities PADATA or TGT encrypted with client Key derived from password via STRING-TO-KEY(P) Results [Wu, 1998] Password checkers not successful in significantly increasing password entropy Structure of TGT (service name = krbtgt) enables verification of key guess by decrypting only 14 octets; similar issues with PADATA Use of DES to encrypt TGT enables use of parallel DES cracking techniques Of 25,000 sample TGTs, 2045 could be decrypted in two weeks using a cluster of 3 UltraSPARC-2 (200 Mhz) and 5 UltraSPARC-1 (167 Mhz) machines Today, 15 off-the-shelf PCs could accomplish the same thing in 1 day at a cost of < $15K

8. May 2001 Bernard Aboba, Microsoft Slide 8 Solutions Machine versus user authentication Machine keys typically have full entropy Use of alterative ciphers in Kerberos Draft-raeburn-krb-gssapi-krb5-3des-01.txt Draft-raeburn-krb-rijndael-krb-00.txt Revision to Kerberos [Wu] SRP used for Kerberos pre-authentication Derived key used to encrypt TGT

9. May 2001 Bernard Aboba, Microsoft Slide 9 Reassociate, Disassociate & Beacon Security Currently, reassociate, disassociate messages are not secure Enables denial of service attacks Proposal Add an authenticator to reassociate and disassociate messages Replay counter, HMAC-SHA1 (replay counter || SourceMAC || destMAC || transmit key) On disassociate: ignore if HMAC is not valid On reassociate: validate authenticator via move-request to old AP; if invalid, old AP ignores move-request Beacon security Currently, beacon messages are not authenticated Enables station to roam to a rogue AP Proposal: validate beacon before reassociating Replay Counter, HMAC-SHA1 (replay counter || sourceMAC || multicast key) Any station can forge this, but better than nothing

10. May 2001 Bernard Aboba, Microsoft Slide 10 Summary � Vulnerabilities Thwarted

11. May 2001 Bernard Aboba, Microsoft Slide 11 Conclusions WEP2 not significantly more secure than WEPv1.0 Small IV only part of the problem; absence of a keyed MIC remains a major deficiency Denial of service attacks not addressed WEP2 should not be treated as a significant security enhancement (should state this explicitly in security considerations section) Kerberos V vulnerable to dictionary attack Most important in �Hot spot� scenarios where many exchanges could be recovered Expect at least 10 percent of passwords to be crackable in 24 hours Downside greater than WEP v1.0 vulnerabilities: not only can traffic be decrypted, but attacker can assume user identity and access other services! Protocol modifications required to address the vulnerability Support for 3DES, AES ciphers Support for SRP pre-authentication Backward compatibility issues AP with built-in KDC AP would need software upgrade to support new ciphers, pre-auth types AP in �pass-through� mode (IAKERB or RADIUS) AP does not need to understand AS_REQ/AS_REP, so no issue

12. May 2001 Bernard Aboba, Microsoft Slide 12 Recommendations Examine feasibility of adding keyed MIC to WEP2 Without keyed MIC, downplay security value of WEP2 Make this clear up front Choose a mandatory-to-implement authentication method resistant to dictionary attack Example: SRP: RFC 2945 EAP-SRP: draft-ietf-pppext-eap-srp-01.txt