8/11/2019 13-5-Auth

1/97

Who are you ?

8/11/2019 13-5-Auth

2/97

2

How can you identify someone?

Certificates, Protocols: Machine to Machine Human to Machine ?

Lets have some suggestions

Be creative, not necessarily computer oriented

Classification of identification methods

What you know (e.g. password)What you are (e.g. biometrics, behaviour)

What you have (e.g. security token)

8/11/2019 13-5-Auth

3/97

Passwordswww.dilbert.com

8/11/2019 13-5-Auth

4/97

4

Your passwords

Everybody has several passwords

Did you choose them?

If so how?

Can you remember them?

Also if you do not use often?

Can no one guess them?

`Vectra bad password for known Opel fan.

8/11/2019 13-5-Auth

5/97

5

Passwords (what you know)But: How secure & secret is the secret ?

AsD5^#2a2fUHard to guess ~Hard to rememberEasyPassword

Alice

EasyPassword

BobBusterCharlie

PDf47$%2!aDilbert

*****

User:AlicePwd: EasyPassword

Recovery

AlicesMothers

Name

8/11/2019 13-5-Auth

6/97

6

Example: pin protected copier

Copier in hallway

Protected by 5 digit code

Enough entropy?

If 10 users with different codes?

Number of tries needed in practice?

*****

8/11/2019 13-5-Auth

7/97

7

Ex2: Account passwords in Unix

Usually user chosen

Passwords not stored on systemWhy?

HASHof a password stored insteadHash is one-way

Collision resistant

/etc/passwrd

World readable (for Account info; name, id, group, etc.)Hashed-password

8/11/2019 13-5-Auth

8/97

8

Theoretical Strength (ball park)

8 symbols; 128^8 = 72,000,000 G

brute force in little over a year at 1G/s (*)

If restrict to letters, digits or common symbols;

96^8: in ~ 3 months Only letters and numbers: half a day

(*) 1G/s+ easily realistic (e.g. in 2002 75G/s RC5-64passwords per seconds using distributed computing)

8/11/2019 13-5-Auth

9/97

9

Account passwords in Unix (cont.) Multiple passwords reduce effort

if any victim is fineSalt

Still significant risk

Faster computers

Weaknesses found in hash functions

Cannot simply make password longer Shadow passwords

Access only for `root, event to hashed pwd

8/11/2019 13-5-Auth

10/97

10

Example of password in Unix

Program to create Hashed passwords#!/bin/perl$salt = ab; # should randomly generate

print New Password: ;

$pwd = ; # enter pwdprint crypt( $pwd, $salt ); # lib call

Run

New Password: Hello abdF5znAEMJTk

New Password: Goodbye

abPV5atKxA04c

8/11/2019 13-5-Auth

11/97

11

Practical Strength: Password Guessing Often: dictionary words, keyboard patterns

Complexity too low even with added symbolWeak!

WHY?...

Guessing: DB with often used words.

Dictionary, common names, etc.

Add symbols, numbers.

Often only a single bad password needed

8/11/2019 13-5-Auth

12/97

12

From (Password) Crack tutorialPeople tend to pick keyboard patterns ("qwerty","!@#$%^&*', etc.) and natural language words.

Suddenly an adversary doesn't have to try 5.96E16strings.

Success rate 22% using a lists of dutch, english,french, german, italian, norwegian and swedishwords plus lists of names, jargon words, keyboardpatterns and anything else people tend to use when

picking passwords.

List of 2.2E7 "words (out of 5.96E16)

(At 1.000 tries a second: all in 6 hrs)

8/11/2019 13-5-Auth

13/97

13

Passwords pros and cons

Checking passwords

At time of entry

With password cracking tool

MyOnePwd

AssignedRandomly

generated

Reuse

Password safe(Why cannot use hash?)

Guidelines

Generation

Use

System side

8/11/2019 13-5-Auth

14/97

14

Some Conclusions on Passwords Very commonly used system

Well known, easy to use

Cheap

A weak form of authentication

Limited complexity Badly chosen passwords

Have to be used in correct way Prevent access to encrypted passwords

Limit guess rates where possible

Remember it may be broken

8/11/2019 13-5-Auth

15/97

www.trustedreviews.com

Biometrics

8/11/2019 13-5-Auth

16/97

16

Biometrics Physical and behavioral characteristics, e.g.Fingerprints

Iris

facial characteristics

hand measurementsgrip pattern

signature

voicetyping pattern

DNA

etc.

8/11/2019 13-5-Auth

17/97

17www.cl.cam.ac.uk

www.byometric.com

Example: Priviumprogram at Schiphol

Iris recognition

Profile stored on card

Skip passport check Fallback

Regular check

At front of the line

8/11/2019 13-5-Auth

18/97

18

Typical Mode of Operation

Verification is easier than identification

8/11/2019 13-5-Auth

19/97

19

Characteristics biometric system Universal (everyone has it)

Uniqueness (different for everyone)

Permanence (same over time),... ... ... ...

8/11/2019 13-5-Auth

20/97

20

Characteristics biometric system Collectability (usability, convenience),

Performance (accurate and fast)

=

8/11/2019 13-5-Auth

21/97

21

Characteristics biometric systemAcceptance (user and societies view)

Circumvention (easy to fake)

8/11/2019 13-5-Auth

22/97

22

Some Comparisons

8/11/2019 13-5-Auth

23/97

23

Variation in Measurements

Every measurement slightly different Enrollment

Profile (e.g. average) from many measurements

Validation New measurements approximately match profile?

Threshold describes allowed distance

Trade off false acceptance rate - false reject rateQuality often specified by equal error rate

8/11/2019 13-5-Auth

24/97

24

threshold => FAR FRR trade-offAccept

imposter

Reject

valid

individual

F

alseAccept

Rate

False Reject Rate

t small

t big

8/11/2019 13-5-Auth

25/97

25

Evaluation Security of aBiometric system

8/11/2019 13-5-Auth

26/97

26

Biometrics Privacy & `key loss issues:

DNA `blueprint of a person

very privacy sensitive interesting e.g. for health insurance companies

Information does not change, cannot be replaced

Information left everywhere

Your fingerprint is on the chair, desk, lunch plate, etc. Not transferable (*)

Biometric passports electronic picture (e.g. against fraud with ID)

fingerprint (e.g. against `look alike)

8/11/2019 13-5-Auth

27/97

Template Protection

27

TemplateStorage

Securely Store templates Normal hash not possible

8/11/2019 13-5-Auth

28/97

28

A Template Protection Scheme(*)

Shielding function

G : Rk {0, 1}k{0, 1}K

K-bit secret S chosen randomly,

biometric X

create helper data W so G(X,W) = S

(*)Practical Biometric Authentication with Template Protection, P. Tyles et al.

k Features K bits secret

8/11/2019 13-5-Auth

29/97

29

Template Protection Scheme (cont.)

Noise insensitive (-contracting)

d(X, X) < => G(X,W) = G(X,W) = S

Secure (-revealing): I(W; S) W leaks less than bits on S

Template protecting (-revealing) : I(W; X) W leaks less than information on X

Shielding function G : Rk {0, 1}k{0, 1}K

helper data W

8/11/2019 13-5-Auth

30/97

30

Template Protection Scheme (cont.)

Enrolment:extract features X from Alices biometrics

choose random secret Scompute helper data WUse one-way hash function H and store

(Alice, W , H(S))

Verification of identity of Alice:

measure biometric: X load helper data W for AliceCompute S = G(X,W) and H(S).

8/11/2019 13-5-Auth

31/97

31

Design Biometric Practical

Able to do at home

Able to do in class

Keep characteristics in mind:

Choose collection method

Define Features

Enrolment Create several measurements.

Evaluation

Universal Uniqueness

Permanence

Collectability Performance

Acceptance

Circumvention

8/11/2019 13-5-Auth

32/97

www.trustedreviews.com

Biometrics(Experimental results)

8/11/2019 13-5-Auth

33/97

33

Measure:

3 points B,C,D (A=(0,0))

A

B

C

D

A

B

C

D

8/11/2019 13-5-Auth

34/97

34

Hand features Feature 1: circumference of the middle joint

(typically thickest part of finger)

Feature 2: Length top digit From middle top to separating line

Feature 3: Length middle digit From separating line to separating line (use main;

lower line as end point).

Feature 4: Length bottom digit

Usesindex

finger

8/11/2019 13-5-Auth

35/97

35

Feature extraction

Blue line: all users

Purple line: distinctivefeature for user

Red line: weakly

distinctive feature Can help prevent false

accepts

Green: indistinctive

features Very close to average -

expect many to havesimilar results.

8/11/2019 13-5-Auth

36/97

36

8/11/2019 13-5-Auth

37/97

37

8/11/2019 13-5-Auth

38/97

38

8/11/2019 13-5-Auth

39/97

39

8/11/2019 13-5-Auth

40/97

40

Feature 1 2 3 4

1 1 0.4203 0.4285 0.1291

2 1 0.2217 -0.0403

3 1 0.6790

4 1

Feature correlation

8/11/2019 13-5-Auth

41/97

41

Options Translation measurement into features

Pre processing; rotation.Data extraction: A,B,C,D

Features should be scaling insensitive

Relative sizes

Angle insensitive?

Effect collectability

Choose features per user ?

Performance

8/11/2019 13-5-Auth

42/97

42

Biometric - Conclusions:

Varying strength of identification Can be tailored to application

Additional hardware needed

Non-replaceable

Privacy & Acceptance

8/11/2019 13-5-Auth

43/97

Security Tokens &Tamper resistant

devices

8/11/2019 13-5-Auth

44/97

44

Functional & Security Goals

Example Tokens

8/11/2019 13-5-Auth

45/97

45

Physical security

Secure processing

(image source: IBM)

8/11/2019 13-5-Auth

46/97

46

Smart Card History

Dethloff (68), Arimura (70), Moreno (74) First chip by Motorola & Bull (77)

France Telecom phone card (84)

Java Card (95) 1 Billion Java cards (2005)

Used in many SIM and ATM cards

Standards (ISO 7816, GSM, EMV, VOP, CEPS)

8/11/2019 13-5-Auth

47/97

8/11/2019 13-5-Auth

48/97

48

What makes the card smart?

CPU (8-bit, 16/32 bit) Memory (RAM, ROM, EEPROM/Flash)

I/O channel (Contact/Contact less)

Cryptographic co-processor

On card devices (Fingerprint, display)

8/11/2019 13-5-Auth

49/97

49

Applications of smartcards (1) Banking

(new) creditcards, Chipknip, internet-banking(e.g. ABN-Amro card).

Telephone cards

Toll payment`Rekening rijden

Public transportMany systems in use

OV chip card

8/11/2019 13-5-Auth

50/97

50

Applications of smartcards (2) Identification & Authorization

eNik (Electronic ID)SIM cards

Building Access cards

Loyalty cards

Secure data storage/access

Privium program schipholElectronic health record Germany

8/11/2019 13-5-Auth

51/97

51

Terminals Embedded systems

Standards (ISO 7816,PC/SC, OCF)

Communication:APDU(Application Protocol DataUnit)

Problems: connections,yield, power, thickness

8/11/2019 13-5-Auth

52/97

52

Attacks on smartcards Logical Attacks

crypto, protocols, implementation

Physical attacks

hardware Side Channel attacks

physical properties

Invasive - Non-invasive

8/11/2019 13-5-Auth

53/97

53

53

Physical AttackRemoving chip from smartcard

heat, etching, acid, etc. to remove protective covering

[Source: Oliver Kmmerling, Marcus Kuhn]

8/11/2019 13-5-Auth

54/97

54

54

Eavesdropping & Altering Physical needles

Electron beam Ion beam

Also remove/create

connection Read out Rom, etc.

[Source: Brightsight]

blown fuse: Restore to re-enable testing mode

8/11/2019 13-5-Auth

55/97

55

Countermeasures

Smaller circuitrymakes many physical attacks harder

Obfuscate chip layout

eg hide bus lines

encrypt bus, memory content

add sensorsDetect tampering

L i l tt k

8/11/2019 13-5-Auth

56/97

56

Logical attacks

Card reader and PC (*)

Man in the middle

(*) E.g. Software from RU Nijmegen to readout chipknip:http://www.ru.nl/ds/research/smartcards/

8/11/2019 13-5-Auth

57/97

57

Countermeasures Don't invent your own cryptographic primitives

GSM: SIM card can be cloned in couple of hoursMifare Classic: ov-chipcard in seconds

Don't invent your own security protocols use standard ones

Beware of random number generationmaking good randomness is difficult

8/11/2019 13-5-Auth

58/97

58

You all know Java High-level OO language, portable

Large footprint Good tools, APIs

Java Safety

Type safety: Enforces objects are of correct typeMemory safety:

Only access to `own memory

run-time buffer size check

Helps prevent some common flaws

Program in Java or assembler?

8/11/2019 13-5-Auth

59/97

59

Java card Java card VM

Multiple applets possible Subset of java

Subset of API, exceptions Initially: No concurrency, garbage collection

Version 3: adds these, HTTP, RMI (optional)

Java card extends javaTransactions, sharable objects

8/11/2019 13-5-Auth

60/97

60

Java implementation

Parsing Type checking

Code generation

Java program Byte codeclass file

interpretercompiler

Class loading Byte code verification

Execution

8/11/2019 13-5-Auth

61/97

61

Java Card implementation

Class loading Byte code verification

CAP file generation

Digital signature

Class loading Signature verification

Ver3: On card verifier

Execution

Byte codeclass file

Byte codesCap file

installerconverter

8/11/2019 13-5-Auth

62/97

62

Architecture

8/11/2019 13-5-Auth

63/97

63

Transaction mechanism

Transaction: set of instructionsShould execute all or none

Roll back changes on failure

Non-Atomic methodsSome things should not be rolled back

Example: PIN try counter

Issues if mixed...

8/11/2019 13-5-Auth

64/97

64

Applet firewall

Context switch

Context switch

8/11/2019 13-5-Auth

65/97

65

Sharable interfaces

8/11/2019 13-5-Auth

66/97

66

Sharable Interfaces

8/11/2019 13-5-Auth

67/97

Side ChannelAttacks

8/11/2019 13-5-Auth

68/97

68

Side Channel attacks Use physical characteristics of the device

to gain extra information. Examples:Power consumption

electro-magnetic emissions (EE)Heat

Timing information

SPA, DPA, Timing attack

8/11/2019 13-5-Auth

69/97

8/11/2019 13-5-Auth

70/97

70

(*) Actually for most of current devices: Changing value causes power consumption;data with many changes consume more power.

Power AnalysisTiming attacks

Simple Power Analysis (SPA)

Power consumption is higher for a 1 than a 0(*)

Gain extra information from a single power trace:

Data with many 1s will consume more power.

Differential Power Analysis (DPA)

8/11/2019 13-5-Auth

71/97

71

Differential Power Analysis (DPA)Look at differences in average power consumption

Collect a set of power traces

Split into two groups Find difference in average power consumption:

Difference trace

InputI1 I2 In

8/11/2019 13-5-Auth

72/97

72

DPA: Bit Propagation Collect Traces (random inputs I1,I2,...In) Choose input bitX

Group traces by value bitX

Difference trace shows where bitXused

Partial diff. traces for DES input bitX= 16,17,18,19

I1 I2 ... In

TraceT1 T2 ... Tn I1 bitX= 1

T1 => G1I2 bitX= 0T2 => G0etc...

AvgAvgDiff. trace

T1

...

T2...

Boolean functionAny function (#1s)

Fundamental idea: Lower Complexity

8/11/2019 13-5-Auth

73/97

73

Fundamental idea: Lower Complexity

Goals is to check a guess for PART of the key.

Example: 64 bit key

nr of possibilities: 18,446,744,073,709,551,616

At 1G encryption per second: more than 500 years

If one can check 1 byte at a type:

nr of possibilities (256 per byte, 8 bytes): 2048 Easily doable even if millions of instructions needed

for each check.

8/11/2019 13-5-Auth

74/97

74

Extracting keys with DPA What: Validate guess for part of the key

In practice: part < 32 bits

Select intermediate result

Use guess of the key to predict Check prediction

8/11/2019 13-5-Auth

75/97

75

Extracting keys with DPA What: Validate guess for part of the key

Select intermediate resultalgorithm needs to compute

Dont care where in implementation

depends on part of the key The part to be guessed

Use guess of the key to predict

Check prediction

Input

Output

Extracting Keys with DPA

8/11/2019 13-5-Auth

76/97

What: Validate guess for part of the key

Select intermediate result

Use guess of the key to predict

calculate intermediate result using guess

Check prediction

Extracting Keys with DPA

76

InputI1 I2 ... In

R1 R2 ... Rn

PredictedIntermediate

guess guess

InputI1 I2 ... In

R1 R2 ... Rn

PredictedIntermediate

Each guess=> different function

R, R, ...

Extracting keys with DPA

8/11/2019 13-5-Auth

77/97

77

g y

What: Validate guess for part of the key

Use intermediate result

Use guess of the key to predict

Check prediction (see Bit-propagation);

function R, (R, R,...) computed?Group Ti by few 1s/many 1s in Ri, (Ri, Ri, ...)

Can see where computed

Nowhere for wrong guess/prediction

Peak in difference trace: correct guess

(Also shows when R is computed)

8/11/2019 13-5-Auth

78/97

78

Difference trace for Correct guess and incorrect guesses

Correlation power analysis (CPA)

8/11/2019 13-5-Auth

79/97

79

Compute correlation: #1s in predicatedvalue (X) and power consumption (Y)

Or actually: sample correlation coefficient

Improves efficiency

8/11/2019 13-5-Auth

80/97

80

Key Retrieval Example: DES

S-box 1:

subkey candidate: 24, peak value: 0,953 at position 66subkey candidate: 19, peak value: -0,439 at position 66

subkey candidate: 26, peak value: -0,419 at position 66

subkey candidate: 7, peak value: -0,418 at position 66

Example(*): AES (Rijndael)

8/11/2019 13-5-Auth

81/97

81

p ( ) ( j )

Symmetric cipher 128 bits key (i.e. 16 bytes)

First round starts with:

Intermediate value: input[ i ] ^ key-guess.(Need to check 256 possibilities.)

void AddRoundKey()

{

for( i = 0; i < 16; i++ )

{

inputdata[ i ] = inputdata[ i ] ^ key[ i ];

}

}

(*) for simple example: assume hamming weight leaks

8/11/2019 13-5-Auth

82/97

82

Simulation tool from the PINPAS Project

8/11/2019 13-5-Auth

83/97

83

Example: RSA Public key crypto, 512+ bit key size

Encryption: calculate mpk mod M Typical SQR-MUL-implementation:

r = 1;

for ( i = 0; i < bitlength( pk ); i++ )

{

r = SQR( r, M );

if ( bit( pk, i ) == 1 )

{r =MUL( r, m, M );

}

}

8/11/2019 13-5-Auth

84/97

84

RSA (2) Smartcard implementation:

SQR, MUL in coprocessor

Coprocessor in tool:

Add two new instructions to processorJava BigIntegerclass for functional behavior

Power consumption: leaks Hamming weights

of coprocessor input output + timing.

8/11/2019 13-5-Auth

85/97

85

RSA (3) Timing attack possible:

0 0 0 1 1 0 0 0

...

if ( bit( pk, i ) == 1 )

{

r = MUL( r, m, M );

}

else{

dummy = MUL( r, m, M );

}

...

8/11/2019 13-5-Auth

86/97

86

RSA (4) Defense against timing attack lines up traces

No more timing information However: potential for DPA attacks

r = 1;

for ( i = 0; i < bitlength( pk ); i++ )

{

r = SQR( r, M );

if ( bit( pk, i ) == 1 )

r =MUL( r, m, M );else

dummy = MUL( r, m, M );

}

8/11/2019 13-5-Auth

87/97

87

RSA Countermeasures (1)

Randomization

Prevent traces from lining up.

Add dummy operations

Randomize order real operations

void AddRoundKey()

{ order = random_permutation( 0, 15 );

for( i = 0; i < 16; i++ )

{ inputdata[ order[ i ] ] =inputdata[ order[ i ] ] ^ key[ order[ i ] ];

}

}

8/11/2019 13-5-Auth

88/97

88

RSA Countermeasures (2) Randomization

Dummy operations must appear realCorrelation reduced not removed

Nr traces needed increases ~ square of probability.

MaskingMask intermediate values with random mask

am = a XOR m

confusion: confuse mask also

diffusion: `masked version operation.

Sid Ch l C t

8/11/2019 13-5-Auth

89/97

89

Side-Channel Countermeasures Don't make your own implementation

Unless know state-of-the art in side-channel attacks

Architectural defense:

Prevent collection of set of traces: Try counter

Regularly change keys.

May be defeated by `template attacks.

Limit vulnerability of key loss

Do not use single key Have alternate levels of defense; e.g. revocation possibilities

Sid Ch l C t

8/11/2019 13-5-Auth

90/97

90

Side-Channel Countermeasures Software defense:

Try to change implementation

Implement countermeasures above

Choose less vulnerable encryption schema

Hardware defense, try to:remove correlation data power consumption

Random activity, noise, reduce power consumption

Prevent traces from lining up e.g. variable clock cycles

Provably resistant against DPA

8/11/2019 13-5-Auth

91/97

Sbox:

Compute for every possible input

in random order

return the result for the correct value

Masked computation of S-box Input x + r ( x masked with random r )

Output S(x) + s S(x) masked with new mask s

91

T i t t

8/11/2019 13-5-Auth

92/97

92

Tamper resistant

... but not tamper proof

make something foolproof and

somebody will invent a better fool

Design moral: breaking card should notbreak whole system

A ti /I i tt k

8/11/2019 13-5-Auth

93/97

93

Active/Invasive attacks

Probing attacks:

Tap specific parts of a chip (a bus)

Change specific parts

reconnect test circuits ...both require special equipment

Map entire chip logic

If algorithm/implementation unknown...

Active/Invasive attacks

8/11/2019 13-5-Auth

94/97

94

Fault attacks:Purposefully cause errors in computation;

prevent unwanted updates (e.g. pincounter) use faulty computation to derive information.

check for dummy instructions

How: Change voltage

Bombard with light

Shake it

Drop a hammer on it...

Mifare (also OV chip) hacking

8/11/2019 13-5-Auth

95/97

95

Mifare (also: OV-chip) hacking

Hack of Mifare:e.g. http://www.youtube.com/watch?v=Y8VVKnUdECg

Work by researchers in Nijmegen

`Dagkaart broken. Later: All cards broken, cloneable

Recently: Discussion traceability

SCard breakable however

8/11/2019 13-5-Auth

96/97

96

SCard breakable, however, ...

Smartcard often not the weakest point

Weaknesses in issuing / use / ...

(missing security goals? e.g. OV-chip)

8/11/2019 13-5-Auth

97/97