Authentication Protocols

PART II

1
Chapter 10:
Real-World Protocols
The wire protocol guys don't worry about security because that's really
a network protocol problem. The network protocol guys don't
worry about it because, really, it's an application problem.
The application guys don't worry about it because, after all,
they can just use the IP address and trust the network.
 Marcus J. Ranum

In the real world, nothing happens at the right place at the right time.
It is the job of journalists and historians to correct that.
 Mark Twain

2
Real-World Protocols
 Next, we look at real protocols
o SSL  practical security on the Web
o IPSec  security at the IP layer
o Kerberos  symmetric key, single sign-on
o WEP  “Swiss cheese” of security protocols
o GSM  mobile phone (in)security

3
Secure Socket Layer

4
Socket layer

 “Socket layer” lives between application User

Socket
application and transport
“layer”
layers transport
OS
 SSL deals with web browsing
network
usually in between HTTP and
TCP link
NIC
physical

5
What is SSL?
 SSL is the protocol used for majority of secure Internet transactions today
 For example, if you want to buy a book at amazon.com…
o You want to be sure you are dealing with Amazon (authentication)
o Your credit card information must be protected in transit (confidentiality and/or
integrity)
o As long as you have money, Amazon does not really care who you are…
o …so, no need for mutual authentication

6
Simple SSL-like Protocol

I’d like to talk to you securely

Here’s my certificate
{K}Bob

Alice protected HTTP Bob

 Is Alice sure she’s talking to Bob?

 Is Bob sure he’s talking to Alice?

7
Simplified SSL Protocol

Can we talk?, cipher list, RA

certificate, cipher, RB
{S}Bob, E(h(msgs,CLNT,K),K)
h(msgs,SRVR,K)
Alice Data protected with key K Bob

 S is the so-called pre-master secret

 K = h(S,R ,R )
A B
 “msgs” means all previous messages
 CLNT and SRVR are constants

8
SSL Keys
6 “keys” derived from K = h(S,RA,RB)
o 2 encryption keys: client and server
o 2 integrity keys: client and server
o 2 IVs: client and server
o Why different keys in each direction?
 Q: Why is h(msgs,CLNT,K) encrypted?
 A: Apparently, it adds no security…

9
SSL Authentication

 Alice authenticates Bob, not vice-versa

o How does client authenticate server?
o Why would server not authenticate client?
 Mutual authentication is possible: Bob sends certificate
request in message 2
o Then client must have a valid certificate
o But, if server wants to authenticate client, server could instead
require password

10
SSL MiM Attack?
RA RA
certificateT, RB certificateB, RB
{S1}Trudy,E(X1,K1) {S2}Bob,E(X2,K2)
h(Y1,K1) h(Y2,K2)
Alice E(data,K1) Trudy E(data,K2) Bob
 Q: What prevents this MiM “attack”?
 A: Bob’s certificate must be signed by a certificate authority (CA)
 What does browser do if signature not valid?
 What does user do when browser complains?

11
SSL Sessions vs Connections
 SSL session is established as shown on previous slides
 SSL designed for use with HTTP 1.0
 HTTP 1.0 often opens multiple simultaneous (parallel) connections
o Multiple connections per session
 SSL session is costly, public key operations
 SSL has an efficient protocol for opening new connections given an
existing session

12
SSL Connection
session-ID, cipher list, RA
session-ID, cipher, RB,
h(msgs,SRVR,K)
h(msgs,CLNT,K)

Alice Protected data Bob

 Assuming SSL session exists

 So, S is already known to Alice and Bob
 Both sides must remember session-ID
 Again, K = h(S,RA,RB)
 No public key operations! (relies on known S)

13
SSL vs IPSec
 IPSec  discussed in next section
o Lives at the network layer (part of the OS)
o Encryption, integrity, authentication, etc.
o Is overly complex, has some security “issues”
 SSL (and IEEE standard known as TLS)
o Lives at socket layer (part of user space)
o Encryption, integrity, authentication, etc.
o Relatively simple and elegant specification

14
SSL vs IPSec
 IPSec: OS must be aware, but not apps
 SSL: Apps must be aware, but not OS
 SSL built into Web early-on (Netscape)
 IPSec often used in VPNs (secure tunnel)
 Reluctance to retrofit applications for SSL
 IPSec not widely deployed (complexity, etc.)
 The bottom line?
 Internet less secure than it could be!

15
IPSec

16
IPSec and SSL
 IPSec lives at the network layer
 IPSec is transparent to applications
application User
SSL
transport
OS
IPSec network

link
NIC

physical

17
IPSec and Complexity
 IPSec is a complex protocol
 Over-engineered
o Lots of (generally useless) features
 Flawed  Some significant security issues
 Interoperability is serious challenge
o Defeats the purpose of having a standard!
 Complex
 And, did I mention, it’s complex?

18
IKE and ESP/AH
 Two parts to IPSec…
 IKE: Internet Key Exchange
o Mutual authentication
o Establish session key
o Two “phases”  like SSL session/connection
 ESP/AH
o ESP: Encapsulating Security Payload  for confidentiality and/or integrity
o AH: Authentication Header  integrity only

19
IKE

20
IKE
 IKE has 2 phases
o Phase 1  IKE security association (SA)
o Phase 2  AH/ESP security association
 Phase 1 is comparable to SSL session
 Phase 2 is comparable to SSL connection
 Not an obvious need for two phases in IKE
o In the context of IPSec, that is
 If multiple Phase 2’s do not occur, then it is more costly to have two
phases!

21
IKE Phase 1
 4 different “key options”
o Public key encryption (original version)
o Public key encryption (improved version)
o Public key signature
o Symmetric key
 For each of these, 2 different “modes”
o Main mode and aggressive mode
 There are 8 versions of IKE Phase 1!
 Need more evidence it’s over-engineered?

22
IKE Phase 1
 We discuss 6 of the 8 Phase 1 variants
o Public key signatures (main & aggressive modes)
o Symmetric key (main and aggressive modes)
o Public key encryption (main and aggressive)
 Why public key encryption and public key signatures?
o Always know your own private key
o May not (initially) know other side’s public key

23
IKE Phase 1

 Uses ephemeral Diffie-Hellman to establish session key

o Provides perfect forward secrecy (PFS)
 Let a be Alice’s Diffie-Hellman exponent
 Let b be Bob’s Diffie-Hellman exponent
 Let g be generator and p prime
 Recall that p and g are public

24
IKE Phase 1: Digital Signature (Main Mode)
IC, CP
IC,RC, CS
IC,RC, ga mod p, RA
IC,RC, gb mod p, RB
IC,RC, E(“Alice”, proofA, K)
Alice IC,RC, E(“Bob”, proofB, K) Bob

 CP = crypto proposed, CS = crypto selected

 IC = initiator “cookie”, RC = responder “cookie”
 K = h(IC,RC,gab mod p,RA,RB)
 SKEYID = h(RA, RB, gab mod p)
 proofA = [h(SKEYID,ga mod p,gb mod p,IC,RC,CP,“Alice”)]Alice
25
IKE Phase 1: Public Key Signature (Aggressive
Mode)

IC, “Alice”, ga mod p, RA, CP

IC,RC, “Bob”, RB,
gb mod p, CS, proofB
IC,RC, proofA
Alice Bob

 Main differences from main mode

o Not trying to hide identities
o Cannot negotiate g or p

26
Main vs Aggressive Modes
 Main mode MUST be implemented
 Aggressive mode SHOULD be implemented
o So, if aggressive mode is not implemented, “you should feel guilty about it”
 Might create interoperability issues
 For public key signature authentication
o Passive attacker knows identities of Alice and Bob in aggressive mode, but not in
main mode
o Active attacker can determine Alice’s and Bob’s identity in main mode

27
IKE Phase 1: Symmetric Key (Main Mode)
IC, CP
IC,RC, CS
IC,RC, ga mod p, RA
IC,RC, gb mod p, RB
IC,RC, E(“Alice”, proofA, K)
Alice Bob
KAB IC,RC, E(“Bob”, proofB, K) KAB

 Same as signature mode except

o KAB = symmetric key shared in advance
o K = h(IC,RC,gab mod p,RA,RB,KAB)
o SKEYID = h(K, gab mod p)
o proofA = h(SKEYID,ga mod p,gb mod p,IC,RC,CP,“Alice”)
28
Problems with Symmetric Key (Main Mode)
 Catch-22
o Alice sends her ID in message 5
o Alice’s ID encrypted with K
o To find K Bob must know KAB
o To get KAB Bob must know he’s talking to Alice!
 Result: Alice’s IP address used as ID!
 Useless mode for the “road warrior”
 Why go to all of the trouble of trying to hide identities in 6 message
protocol?

29
IKE Phase 1: Symmetric Key (Aggressive Mode)

IC, “Alice”, ga mod p, RA, CP

IC,RC, “Bob”, RB,

gb mod p, CS, proofB
IC,RC, proofA
Alice Bob
 Same format as digital signature aggressive mode
 Not trying to hide identities…
 As a result, does not have problems of main mode
 But does not (pretend to) hide identities

30
IKE Phase 1: Public Key Encryption (Main Mode)
IC, CP
IC,RC, CS
IC,RC, ga mod p, {RA}Bob, {“Alice”}Bob

IC,RC, gb mod p, {RB}Alice, {“Bob”}Alice

IC,RC, E(proofA, K)
Alice IC,RC, E(proofB, K) Bob

 CP = crypto proposed, CS = crypto selected

 IC = initiator “cookie”, RC = responder “cookie”
 K = h(IC,RC,gab mod p,RA,RB)
 SKEYID = h(RA, RB, gab mod p)
 proofA = h(SKEYID,ga mod p,gb mod p,IC,RC,CP,“Alice”)
31
IKE Phase 1: Public Key Encryption (Aggressive
Mode)
IC, CP, ga mod p,
{“Alice”}Bob, {RA}Bob
IC,RC, CS, gb mod p,
{“Bob”}Alice, {RB}Alice, proofB

IC,RC, proofA
Alice Bob
 K, proofA, proofB computed as in main mode
 Note that identities are hidden
o The only aggressive mode to hide identities
o So, why have a main mode?

32
Public Key Encryption Issue?
 In public key encryption, aggressive mode…
 Suppose Trudy generates
o Exponents a and b
o Nonces RA and RB

 Trudy can compute “valid” keys and proofs: gab mod p, K, SKEYID,
proofA and proofB
 All of this also works in main mode

33
Public Key Encryption Issue?

IC, CP, ga mod p,

{“Alice”}Bob, {RA}Bob
IC,RC, CS, gb mod p,
{“Bob”}Alice, {RB}Alice, proofB
Trudy IC,RC, proofA Trudy
(as Alice) (as Bob)
 Trudy can create messages that appears to be between Alice and Bob
 Appears valid to any observer, including Alice and Bob!

34
Plausible Deniability
 Trudy can create fake “conversation” that appears to be between Alice and
Bob
o Appears valid, even to Alice and Bob!
 A security failure?
 In IPSec public key option, it is a feature…
o Plausible deniability: Alice and Bob can deny that any conversation took place!
 In some cases it might create a problem
o E.g., if Alice makes a purchase from Bob, she could later repudiate it (unless she
had signed)

35
IKE Phase 1 “Cookies”
 IC and RC  cookies (or “anti-clogging tokens”) supposed to prevent
DoS attacks
o No relation to Web cookies
 To reduce DoS threats, Bob wants to remain stateless as long as possible
 But Bob must remember CP from message 1 (required for proof of
identity in message 6)
 Bob must keep state from 1st message on
o So, these “cookies” offer little DoS protection

36
IKE Phase 1 Summary
 Result of IKE phase 1 is
o Mutual authentication
o Shared symmetric key
o IKE Security Association (SA)
 But phase 1 is expensive
o Especially in public key and/or main mode
 Developers of IKE thought it would be used for lots of things  not just
IPSec
o Partly explains the over-engineering…

37
IKE Phase 2
 Phase 1 establishes IKE SA
 Phase 2 establishes IPSec SA
 Comparison to SSL…
o SSL session is comparable to IKE Phase 1
o SSL connections are like IKE Phase 2
 IKE could be used for lots of things, but in practice, it’s not!

38
IKE Phase 2
IC, RC, CP, E(hash1,SA,RA,K)

IC, RC, CS, E(hash2,SA,RB,K)

IC, RC, E(hash3,K)

Alice Bob
 Key K, IC, RC and SA known from Phase 1
 Proposal CP includes ESP and/or AH
 Hashes 1,2,3 depend on SKEYID, SA, RA and RB
 Keys derived from KEYMAT = h(SKEYID,RA,RB,junk)
 Recall SKEYID depends on phase 1 key method
 Optional PFS (ephemeral Diffie-Hellman exchange)
39
IPSec

 After IKE Phase 1, we have an IKE SA

 After IKE Phase 2, we have an IPSec SA
 Authentication completed and have a shared
symmetric key (session key)
 Now what?
o We want to protect IP datagrams
o But what is an IP datagram?
o From the perspective of IPSec…
40
IP Review
 IP datagram is of the form

IP header data

 Where IP header is

41
IP and TCP
 Consider Web traffic, for example
o IP encapsulates TCP and…
o …TCP encapsulates HTTP

IP header data

IP header TCP hdr HTTP hdr app data

 IP data includes TCP header, etc.

42
IPSec Transport Mode
 IPSec Transport Mode
IP header data

IP header IPSec header data

 Transport mode designed for host-to-host

 Transport mode is efficient
o Adds minimal amount of extra header
 The original header remains
o Passive attacker can see who is talking

43
IPSec: Host-to-Host

 IPSec transport mode used here

 There may be firewalls in between

o If so, is that a problem?

44
IPSec Tunnel Mode
 IPSec Tunnel Mode
IP header data

new IP hdr IPSec hdr IP header data

 Tunnel mode for firewall-to-firewall traffic

 Original IP packet encapsulated in IPSec
 Original IP header not visible to attacker
o New IP header from firewall to firewall
o Attacker does not know which hosts are talking
45
IPSec: Firewall-to-Firewall

 IPSec tunnel mode used here

 Note: Local networks not protected

 Is there any advantage here?

46
 Comparison of IPSec Modes
 Transport Mode  Transport Mode
o Host-to-host
IP header data
 Tunnel Mode
IP header IPSec header data
o Firewall-to-
firewall
 Tunnel Mode  Transport Mode
IP header data not necessary…
 …but it’s more
new IP hdr IPSec hdr IP header data efficient

47
IPSec Security
 What kind of protection?
o Confidentiality?
o Integrity?
o Both?
 What to protect?
o Data?
o Header?
o Both?
 ESP/AH allow some combinations of these

48
AH vs ESP
 AH  Authentication Header
o Integrity only (no confidentiality)
o Integrity-protect everything beyond IP header and some fields of header (why not
all fields?)
 ESP  Encapsulating Security Payload
o Integrity and confidentiality both required
o Protects everything beyond IP header
o Integrity-only by using NULL encryption

49
ESP NULL Encryption
 According to RFC 2410
o NULL encryption “is a block cipher the origins of which appear to be lost in
antiquity”
o “Despite rumors”, there is no evidence that NSA “suppressed publication of this
algorithm”
o Evidence suggests it was developed in Roman times as exportable version of
Caesar’s cipher
o Can make use of keys of varying length
o No IV is required
o Null(P,K) = P for any P and any key K
 Is ESP with NULL encryption same as AH ?

50
Why Does AH Exist? (1)
 Cannot encrypt IP header
o Routers must look at the IP header
o IP addresses, TTL, etc.
o IP header exists to route packets!
 AH protects immutable fields in IP header
o Cannot integrity protect all header fields
o TTL, for example, will change
 ESP does not protect IP header at all

51
Why Does AH Exist? (2)
 ESP encrypts everything beyond the IP header (if non-null encryption)
 If ESP-encrypted, firewall cannot look at TCP header in host-to-host case
 Why not use ESP with NULL encryption?
o Firewall sees ESP header, but does not know whether null encryption is used
o End systems know, but not the firewalls

52
Why Does AH Exist? (3)
 The real reason why AH exists:
o At one IETF meeting “someone from Microsoft gave an impassioned
speech about how AH was useless…”
o “…everyone in the room looked around and said `Hmm. He’s right,
and we hate AH also, but if it annoys Microsoft let’s leave it in since
we hate Microsoft more than we hate AH.’ ”

53
Kerberos

54
Kerberos
 In Greek mythology, Kerberos is 3-headed dog that guards entrance to
Hades
o “Wouldn’t it make more sense to guard the exit?”
 In security, Kerberos is an authentication protocol based on symmetric
key crypto
o Originated at MIT
o Based on Needham and Schroeder protocol
o Relies on a Trusted Third Party (TTP)

55
Motivation for Kerberos
 Authentication using public keys
o N users  N key pairs
 Authentication using symmetric keys
o N users requires (on the order of) N2 keys
 Symmetric key case does not scale
 Kerberos based on symmetric keys but only requires N keys for N users
- Security depends on TTP
+ No PKI is needed

56
Kerberos KDC
 Kerberos Key Distribution Center or KDC
o KDC acts as the TTP
o TTP is trusted, so it must not be compromised
 KDC shares symmetric key KA with Alice, key KB with Bob, key KC with
Carol, etc.
 And a master key KKDC known only to KDC
 KDC enables authentication, session keys
o Session key for confidentiality and integrity
 In practice, crypto algorithm is DES

57
Kerberos Tickets
 KDC issue tickets containing info needed to access network resources
 KDC also issues Ticket-Granting Tickets or TGTs that are used to
obtain tickets
 Each TGT contains
o Session key
o User’s ID
o Expiration time
 Every TGT is encrypted with KKDC
o So, TGT can only be read by the KDC

58
Kerberized Login

 Alice enters her password

 Then Alice’s computer does following:
o Derives KA from Alice’s password
o Uses KA to get TGT for Alice from KDC
 Alice then uses her TGT (credentials) to securely access
network resources
 Plus: Security is transparent to Alice
 Minus: KDC must be secure  it’s trusted!

59
 Kerberized Login
Alice wants
Alice’s a TGT
password
E(SA,TGT,KA)

Alice Computer KDC

 Key KA = h(Alice’s password)
 KDC creates session key SA
 Alice’s computer decrypts SA and TGT
o Then it forgets K
A

 TGT = E(“Alice”, SA, KKDC)

60
Alice Requests “Ticket to Bob”
I want to
talk to Bob
Talk to Bob REQUEST

REPLY

Alice Computer KDC

 REQUEST = (TGT, authenticator)
o authenticator = E(timestamp, SA)

 REPLY = E(“Bob”, KAB, ticket to Bob, SA)

o ticket to Bob = E(“Alice”, KAB, KB)
 KDC gets SA from TGT to verify timestamp
61
Alice Uses Ticket to Bob

ticket to Bob, authenticator

E(timestamp + 1, KAB)

Alice’s Bob
Computer

 ticket to Bob = E(“Alice”, KAB, KB)

 authenticator = E(timestamp, KAB)
 Bob decrypts “ticket to Bob” to get KAB which he
then uses to verify timestamp
62
Kerberos
 Key SA used in authentication
o For confidentiality/integrity
 Timestamps for authentication and replay protection
 Recall, that timestamps…
o Reduce the number of messages  like a nonce that is known in
advance
o But, “time” is a security-critical parameter

63
Questions about Kerberos
 When Alice logs in, KDC sends E(SA, TGT, KA) where TGT = E(“Alice”, SA,
KKDC)
Q: Why is TGT encrypted with KA?
A: Enables Alice to be anonymous when she later uses her TGT to request a ticket
 In Alice’s “Kerberized” login to Bob, why can Alice remain anonymous?
 Why is “ticket to Bob” sent to Alice?
o Why doesn’t KDC send it directly to Bob?

64
Kerberos Alternatives
 Could have Alice’s computer remember password and use that for
authentication
o Then no KDC required
o But hard to protect passwords
o Also, does not scale
 Could have KDC remember session key instead of putting it in a TGT
o Then no need for TGT
o But stateless KDC is major feature of Kerberos

65
Kerberos Keys
 In Kerberos, KA = h(Alice’s password)
 Could instead generate random KA
o Compute Kh = h(Alice’s password)
o And Alice’s computer stores E(KA, Kh)
 Then KA need not change when Alice changes her password
o But E(KA, Kh) must be stored on computer
 This alternative approach is often used
o But not in Kerberos

66
WEP

67
WEP
 WEP  Wired Equivalent Privacy
 The stated goal of WEP is to make wireless LAN as secure as a wired
LAN
 According to Tanenbaum:
o “The 802.11 standard prescribes a data link-level security protocol called WEP
(Wired Equivalent Privacy), which is designed to make the security of a wireless
LAN as good as that of a wired LAN. Since the default for a wired LAN is no
security at all, this goal is easy to achieve, and WEP achieves it as we shall see.”

68
WEP Authentication
Authentication Request
R

E(R, K)
Alice, K Bob, K

 Bob is wireless access point

 Key K shared by access point and all users
o Key K seldom (if ever) changes
 WEP has many, many, many security flaws
69
WEP Issues
 WEP uses RC4 cipher for confidentiality
o RC4 can be a strong cipher
o But WEP introduces a subtle flaw…
o …making cryptanalytic attacks feasible
 WEP uses CRC for “integrity”
o Should have used a MAC, HMAC, or similar
o CRC is for error detection, not crypto integrity
o Everyone should know NOT to use CRC here…

70
WEP Integrity Problems
 WEP “integrity” gives no crypto integrity
o CRC is linear, so is stream cipher (XOR)
o Trudy can change ciphertext and CRC so that checksum on plaintext remains
valid
o Then Trudy’s introduced changes go undetected
o Requires no knowledge of the plaintext!
 CRC does not provide a cryptographic integrity check
o CRC designed to detect random errors
o Not to detect intelligent changes

71
More WEP Integrity Issues
 Suppose Trudy knows destination IP
 Then Trudy also knows keystream used to encrypt IP address, since
C = destination IP address  keystream
 Then Trudy can replace C with
C = Trudy’s IP address  keystream
 And change the CRC so no error detected
o Then what happens??
 Moral: Big problems when integrity fails

72
WEP Key
 Recall WEP uses a long-term key K
 RC4 is a stream cipher, so each packet must be encrypted using a different
key
o Initialization Vector (IV) sent with packet
o Sent in the clear, that is, IV is not secret
o Note: IV similar to MI in WWII ciphers
 Actual RC4 key for packet is (IV,K)
o That is, IV is pre-pended to long-term key K

73
WEP Encryption

IV, E(packet,KIV)

Alice, K Bob, K

 KIV = (IV,K)
o That is, RC4 key is K with 3-byte IV pre-pended
 The IV is known to Trudy

74
WEP IV Issues
 WEP uses 24-bit (3 byte) IV
o Each packet gets its own IV
o Key: IV pre-pended to long-term key, K
 Long term key K seldom changes
 If long-term key and IV are same, then same keystream is used
o This is bad, bad, really really bad!
o Why?

75
WEP IV Issues
 Assume 1500 byte packets, 11 Mbps link
 Suppose IVs generated in sequence
o Since 1500  8/(11  106)  224 = 18,000 seconds, an IV repeat in about 5 hours of
traffic
 Suppose IVs generated at random
o By birthday problem, some IV repeats in seconds
 Again, repeated IV (with same K) is bad

76
Another Active Attack
 Suppose Trudy can insert traffic and observe corresponding ciphertext
o Then she knows the keystream for some IV
o She can decrypt any packet that uses that IV
 If Trudy does this many times, she can then decrypt data for lots of IVs
o Remember, IV is sent in the clear
 Is such an attack feasible?

77
Cryptanalytic Attack
 WEP data encrypted using RC4
o Packet key is IV with long-term key K
o 3-byte IV is pre-pended to K
o Packet key is (IV,K)
 Recall IV is sent in the clear (not secret)
o New IV sent with every packet
o Long-term key K seldom changes (maybe never)
 So Trudy always knows IV and ciphertext
o Trudy wants to find the key K

78
Cryptanalytic Attack
 3-byte IV pre-pended to key
 Denote the RC4 key bytes …
o … as K ,K ,K ,K ,K ,K , …
0 1 2 3 4 5
o Where IV = (K ,K ,K ) , which Trudy knows
0 1 2
o Trudy wants to find K = (K ,K ,K , …)
3 4 5
 Given enough IVs, Trudy can easily find key K
o Regardless of the length of the key
o Provided Trudy knows first keystream byte
o Known plaintext attack (1st byte of each packet)
o Prevent by discarding first 256 keystream bytes

79
WEP Conclusions
 Many attacks are practical
 Attacks have been used to recover keys and break real WEP traffic
 How to prevent these attacks?
o Don’t use WEP
o Good alternatives: WPA, WPA2, etc.
 How to make WEP a little better?
o Restrict MAC addresses, don’t broadcast ID, …

80
GSM (In)Security

81
Cell Phones
 First generation cell phones
o Brick-sized, analog, few standards
o Little or no security
o Susceptible to cloning
 Second generation cell phones: GSM
o Began in 1982 as “Groupe Speciale Mobile”
o Now, Global System for Mobile Communications
 Third generation?
o 3rd Generation Partnership Project (3GPP)

82
 GSM System Overview

air
interface

Mobile
Base AuC
VLR
Station
“land line”
HLR
PSTN
Base Internet
etc. Home
Visited Station Network
Network Controller

83
GSM System Components
 Mobile phone
o Contains SIM (Subscriber Identity Module)
 SIM is the security module
o IMSI (International Mobile Subscriber ID)
o User key: Ki (128 bits)
o Tamper resistant (smart card)
o PIN activated (often not used)
SIM

84
GSM System Components
 Visited network  network where mobile is currently located
o Base station  one “cell”
o Base station controller  manages many cells
o VLR (Visitor Location Register)  info on all visiting mobiles currently in the
network
 Home network  “home” of the mobile
o HLR (Home Location Register)  keeps track of most recent location of mobile
o AuC (Authentication Center)  has IMSI and Ki

85
GSM Security Goals
 Primary design goals
o Make GSM as secure as ordinary telephone
o Prevent phone cloning
 Not designed to resist an active attacks
o At the time this seemed infeasible
o Today such an attacks are clearly feasible…
 Designers considered biggest threats to be
o Insecure billing
o Corruption
o Other low-tech attacks

86
GSM Security Features
 Anonymity
o Intercepted traffic does not identify user
o Not so important to phone company
 Authentication
o Necessary for proper billing
o Very, very important to phone company!
 Confidentiality
o Confidentiality of calls over the air interface
o Not important to phone company…
o …except for marketing

87
GSM: Anonymity

 IMSI used to initially identify caller

 Then TMSI (Temporary Mobile Subscriber ID) used
o TMSI changed frequently
o TMSI’s encrypted when sent
 Not a strong form of anonymity
 But probably useful in many cases

88
GSM: Authentication
 Caller is authenticated to base station
 Authentication is not mutual
 Authentication via challenge-response
o Home network generates RAND and computes XRES = A3(RAND, Ki) where A3
is a hash
o Then (RAND,XRES) sent to base station
o Base station sends challenge RAND to mobile
o Mobile’s response is SRES = A3(RAND, Ki)
o Base station verifies SRES = XRES
 Note: Ki never leaves home network
89
GSM: Confidentiality
 Data encrypted with stream cipher
 Error rate estimated at about 1/1000
o Error rate is high for a block cipher
 Encryption key Kc
o Home network computes Kc = A8(RAND, Ki) where A8 is a hash
o Then Kc sent to base station with (RAND,XRES)
o Mobile computes Kc = A8(RAND, Ki)
o Keystream generated from A5(Kc)
 Note: Ki never leaves home network

90
GSM Security
1. IMSI
2. IMSI
4. RAND
3. (RAND,XRES,Kc)
5. SRES
Mobile Base Home
6. Encrypt with Kc Station Network

 SRES and Kc must be uncorrelated

o Even though both are derived from RAND and Ki
 Must not be possible to deduce Ki from known RAND/SRES pairs (known plaintext
attack)
 Must not be possible to deduce Ki from chosen RAND/SRES pairs (chosen plaintext
attack)
o With possession of SIM, attacker can choose RAND’s
91
GSM Insecurity (1)
 Hash used for A3/A8 is COMP128
o Broken by 160,000 chosen plaintexts
o With SIM, can get Ki in 2 to 10 hours
Base
 Encryption between mobile and base station but no Station
encryption from base station to base station controller
o Often transmitted over microwave link
VLR
 Encryption algorithm A5/1
o Broken with 2 seconds of known plaintext
Base
Station
Controller

92
GSM Insecurity (2)

 Attacks on SIM card

o Optical Fault Induction  could attack SIM with a flashbulb to
recover Ki
o Partitioning Attacks  using timing and power consumption, could
recover Ki with only 8 adaptively chosen “plaintexts”
 With possession of SIM, attacker could recover Ki in
seconds

93
GSM Insecurity (3)
 Fake base station exploits two flaws
1. Encryption not automatic
2. Base station not authenticated

RAND
SRES Call to
destination
No
Mobile Fake
encryption Base Station Base Station

 Note: GSM bill goes to fake base station!

94
GSM Insecurity (4)
 Denial of service is possible
o Jamming (always an issue in wireless)
 Can replay triple: (RAND,XRES,Kc)
o One compromised triple gives attacker a key Kc that is valid forever
o No replay protection here

95
GSM Conclusion
 Did GSM achieve its goals?
o Eliminate cloning? Yes, as a practical matter
o Make air interface as secure as PSTN? Perhaps…
 But design goals were clearly too limited
 GSM insecurities  weak crypto, SIM issues, fake base station, replay,
etc.
 PSTN insecurities  tapping, active attack, passive attack (e.g., cordless
phones), etc.
 GSM a (modest) security success?

96
3rd Generation Partnership Project (3GPP)
 3G security built on GSM (in)security
 3G fixed known GSM security problems
o Mutual authentication
o Integrity-protect signaling (such as “start encryption” command)
o Keys (encryption/integrity) cannot be reused
o Triples cannot be replayed
o Strong encryption algorithm (KASUMI)
o Encryption extended to base station controller

97
Protocols Summary

 Generic authentication protocols

o Protocols are subtle!
 SSH
 SSL
 IPSec
 Kerberos
 Wireless: GSM and WEP

98
Coming Attractions…
 Software and security
o Software flaws  buffer overflow, etc.
o Malware  viruses, worms, etc.
o Software reverse engineering
o Digital rights management
o OS and security/NGSCB

99