Detecting Similar Code Segments through Side Channel Leakage in - - PowerPoint PPT Presentation

▶
detecting similar code segments through side channel
SMART_READER_LITE
LIVE PREVIEW

Detecting Similar Code Segments through Side Channel Leakage in - - PowerPoint PPT Presentation

Feb 13, 2023 124 likes •467 views

Detecting Similar Code Segments through Side Channel Leakage in Microcontrollers Peter Samarin 1 , 2 and Kerstin Lemke-Rust 1 Bonn-Rhein-Sieg University of Applied Sciences 1 Ruhr-Universitt Bochum 2 Germany November 29, 2017 Bonn-Rhein-Sieg

slide-1
SLIDE 1

Detecting Similar Code Segments through Side Channel Leakage in Microcontrollers

Peter Samarin1,2 and Kerstin Lemke-Rust1

Bonn-Rhein-Sieg University of Applied Sciences1 Ruhr-Universität Bochum2 Germany

November 29, 2017

Bonn-Rhein-Sieg University of Applied Sciences

slide-2
SLIDE 2

Motivation: Software Plagiarism in Microcontrollers

◮ A product comes to the market with the same capabilities ◮ Does the system contain our intellectual property?

?

µC

◮ Adversary takes our binary ◮ Effective read-out protection ◮ Comparison of code binaries not possible ◮ Our solution: compare power side channel leakage of the

two implementations

Peter Samarin and Kerstin Lemke-Rust Detecting Similar Code Segments through Side Channel Leakage 1 / 22

slide-3
SLIDE 3

Observations about the Power Side Channel

Power traces of program 1 samples from all traces at time Varying inputs Power traces of program 2 samples from all traces at time

Input = Input = ◮ high correlation when same data is processed ◮ low correlation when different data is processed

Peter Samarin and Kerstin Lemke-Rust Detecting Similar Code Segments through Side Channel Leakage 2 / 22

slide-4
SLIDE 4

Our Approach

Power traces of program 1 Varying inputs Power traces of program 2

Peter Samarin and Kerstin Lemke-Rust Detecting Similar Code Segments through Side Channel Leakage 3 / 22

slide-5
SLIDE 5

Our Approach

Power traces of program 1 Varying inputs Power traces of program 2

Peter Samarin and Kerstin Lemke-Rust Detecting Similar Code Segments through Side Channel Leakage 3 / 22

slide-6
SLIDE 6

Our Approach

Power traces of program 1 Varying inputs Power traces of program 2

Peter Samarin and Kerstin Lemke-Rust Detecting Similar Code Segments through Side Channel Leakage 3 / 22

slide-7
SLIDE 7

Our Approach

Power traces of program 1 Varying inputs Power traces of program 2

Peter Samarin and Kerstin Lemke-Rust Detecting Similar Code Segments through Side Channel Leakage 3 / 22

slide-8
SLIDE 8

Our Approach

Power traces of program 1 Varying inputs Power traces of program 2

Peter Samarin and Kerstin Lemke-Rust Detecting Similar Code Segments through Side Channel Leakage 3 / 22

slide-9
SLIDE 9

Our Approach

Power traces of program 1 Varying inputs Power traces of program 2

Peter Samarin and Kerstin Lemke-Rust Detecting Similar Code Segments through Side Channel Leakage 3 / 22

slide-10
SLIDE 10

Our Approach

Power traces of program 1 Varying inputs Power traces of program 2

Peter Samarin and Kerstin Lemke-Rust Detecting Similar Code Segments through Side Channel Leakage 3 / 22

slide-11
SLIDE 11

Our Approach

Power traces of program 1 Varying inputs Power traces of program 2

Peter Samarin and Kerstin Lemke-Rust Detecting Similar Code Segments through Side Channel Leakage 3 / 22

slide-12
SLIDE 12

Our Approach: Correlate at all Times

Power traces of program 1 Varying inputs Power traces of program 2

Peter Samarin and Kerstin Lemke-Rust Detecting Similar Code Segments through Side Channel Leakage 3 / 22

slide-13
SLIDE 13

Expectations about the Similarity Matrix

◮ The similarity matrix shows at what time similar

computations happen Identical program, identical data Similar program, similar data Partially identical program, identical data Different program

  • r

different data

Peter Samarin and Kerstin Lemke-Rust Detecting Similar Code Segments through Side Channel Leakage 4 / 22

slide-14
SLIDE 14

Our Approach: Similarity measure

Suspicious program Genuine program t

|Correlation|

1

Global similarity measure abs(max(col ))

Segment0 Segment1 Segment0

Local similarity measure

Peter Samarin and Kerstin Lemke-Rust Detecting Similar Code Segments through Side Channel Leakage 5 / 22

slide-15
SLIDE 15

Experimental Setup

◮ Smartcards with ATMega163 microcontroller

◮ 8-bit µC, running at 4MHz

◮ Measure using a digital oscilloscope (PicoScope 6402C)

◮ sampling rate is 375 MHz Peter Samarin and Kerstin Lemke-Rust Detecting Similar Code Segments through Side Channel Leakage 6 / 22

slide-16
SLIDE 16

Test Programs: Implementations of AES in Assembly

AES Labor AES Furious AES Fast AES Fantastic AES-0 AES Labor AES Furious AES Fast AES Fantastic AES-0

100 200 300 400 500 600 700 800 900 1000 1100 1200 1300 1400 1500 1600 1700 1800 1900 2000 2100 2200 2300

PU L AK SB MC∗ KE AK SB MC∗ KE AK SB MC∗ KE AK SB MC∗ KE AK SB MC∗ KE AK SB PU KE PU L AK SB MC PU L KE# L AK SB MC AK SB MC AK SB MC AK SB MC AK SB MC PU L KE# L AK R R R R R PU L AK KE SB MC∗ AK KE SB MC∗ AK KE SB MC∗ AK KE SB MC∗ AK KE SB MC∗ AK

PU - push registers PO - pop registers *,# - identical code KE - key expansion AK - add round key L - load key/plaintext S - store ciphertext SB - shift rows and subbytes MC - mix columns R - one AES round in Fast

Clock cycle

4100 4200 4300 4400

MC AK SB AK SPO

2200 2300 2400 2500 2600 2700 2800 2900 3000 3100 3200 3300 3400 3500 3600 3700 3800 3900 4000

MC∗ KE AK SB MC∗ KE AK SB MC∗ KE AK SB MC∗ KE AK SB KE AK SPO MC AK SB MC AK SB MC AK SB MC AK SB MC AK SB MC AK SB MC AK SB MC AK SB MC AK SB MC AK SB MC AK SB MC AK SB MC AK SB AK SPO R R R R AK SPO KE SB MC∗ AK KE SB MC∗ AK KE SB MC∗ AK KE SB MC∗ AK KE SB AK SPO

◮ 10k traces were recorded for each implementation

Peter Samarin and Kerstin Lemke-Rust Detecting Similar Code Segments through Side Channel Leakage 7 / 22

slide-17
SLIDE 17

Results: Similarity Matrix of Furious vs. Furious

Peter Samarin and Kerstin Lemke-Rust Detecting Similar Code Segments through Side Channel Leakage 8 / 22

slide-18
SLIDE 18

Results: Similarity Matrix of Fast vs. Furious

Peter Samarin and Kerstin Lemke-Rust Detecting Similar Code Segments through Side Channel Leakage 9 / 22

slide-19
SLIDE 19

Results: Maximum Projection into Furious

PU - push registers PO - pop registers KE - key expansion AK - add round key L - load key/plaintext S - store ciphertext SB - shift rows and subbytes MC - mix columns

0.2 0.4 0.6 0.8 1 500 1000 1500 2000 2500 3000 3500 |Correlation| Clock cycle

AES-0 in Furious

PU L KE L AK SB MC AK SB MC AK SB MC AK SB MC AK SB MC AK SB MC AK SB MC AK SB MC AK SB MC AK SB AK S PO

0.2 0.4 0.6 0.8 1 500 1000 1500 2000 2500 3000 3500 |Correlation| Clock cycle

AES Labor in Furious

PU L KE L AK SB MC AK SB MC AK SB MC AK SB MC AK SB MC AK SB MC AK SB MC AK SB MC AK SB MC AK SB AK S PO

0.2 0.4 0.6 0.8 1 500 1000 1500 2000 2500 3000 3500 |Correlation| Clock cycle

Furious in Furious

PU L KE L AK SB MC AK SB MC AK SB MC AK SB MC AK SB MC AK SB MC AK SB MC AK SB MC AK SB MC AK SB AK S PO

0.2 0.4 0.6 0.8 1 500 1000 1500 2000 2500 3000 3500 |Correlation| Clock cycle

Fast in Furious

PU L KE L AK SB MC AK SB MC AK SB MC AK SB MC AK SB MC AK SB MC AK SB MC AK SB MC AK SB MC AK SB AK S PO

0.2 0.4 0.6 0.8 1 500 1000 1500 2000 2500 3000 3500 |Correlation| Clock cycle

Fantastic in Furious

PU L KE L AK SB MC AK SB MC AK SB MC AK SB MC AK SB MC AK SB MC AK SB MC AK SB MC AK SB MC AK SB AK S PO

Peter Samarin and Kerstin Lemke-Rust Detecting Similar Code Segments through Side Channel Leakage 10 / 22

slide-20
SLIDE 20

Results: Maximum Projection, Global Similarity

AES-0 AES Labor Furious Fast Fantastic AES-0 0.97 0.41 0.63 0.33 0.53 AES Labor 0.42 0.91 0.46 0.29 0.39 Furious 0.61 0.44 0.96 0.45 0.54 Fast 0.35 0.32 0.46 0.96 0.29 Fantastic 0.58 0.40 0.62 0.30 0.93

Peter Samarin and Kerstin Lemke-Rust Detecting Similar Code Segments through Side Channel Leakage 11 / 22

slide-21
SLIDE 21

Results: Maximum Projection of Code Segments

AK SB MC KE AES-0 0.96 0.97 0.98 0.97 AES Labor 0.64 0.33 0.36 0.43 Furious 0.68 0.65 0.73 0.46 Fast 0.45 0.31 0.26 0.44 Fantastic 0.64 0.58 0.75 0.41

(a) →AES-0

AK SB MC KE 0.68 0.31 0.38 0.40 0.96 0.97 0.96 0.88 0.73 0.38 0.40 0.41 0.48 0.24 0.19 0.39 0.62 0.31 0.37 0.43

(b) →AES Labor

AK SB MC KE 0.71 0.65 0.71 0.46 0.75 0.40 0.37 0.45 0.95 0.98 0.98 0.96 0.47 0.31 0.27 0.95 0.65 0.72 0.68 0.41

(c) →Furious

AK KE R AES-0 0.69 0.46 0.28 AES Labor 0.73 0.45 0.23 Furious 0.85 0.95 0.27 Fast 0.97 0.95 0.98 Fantastic 0.64 0.40 0.25

(d) →Fast

AK SB MC KE 0.66 0.57 0.75 0.33 0.62 0.32 0.35 0.40 0.62 0.71 0.70 0.32 0.43 0.27 0.25 0.31 0.96 0.96 0.97 0.90

(e) →Fantastic

Peter Samarin and Kerstin Lemke-Rust Detecting Similar Code Segments through Side Channel Leakage 12 / 22

slide-22
SLIDE 22

Experiment Set #2: Furious vs. Modified Furious

◮ addr: change register and data addresses ◮ swap: change the order of instruction execution ◮ addr+swap ◮ dummy: add 792 NOP instruction randomly ◮ dummy smart: add 792 leakage-generating instructions ◮ dummy smart+addr+swap

Peter Samarin and Kerstin Lemke-Rust Detecting Similar Code Segments through Side Channel Leakage 13 / 22

slide-23
SLIDE 23

Dummy Smart Explanation

◮ Assembly language macros applied to state registers

randomly throughout the code

LDI ZL, 0x00 LPM \tmp, Z EOR \tmp, \tmp PUSH \tmp LDI \tmp, \c EOR \reg, \tmp POP \tmp INC \reg DEC \reg ROL \reg ROR \reg NEG \reg NEG \reg 1 2 3 4 5 6 MOV \tmp, \reg ;; save register LDI ZH, hi8(hd_temp) LDI ZL, lo8(hd_temp) LD \reg, z MOV \reg, \tmp ;; restore register 8 PUSH \reg1 PUSH \reg2 PUSH \reg3 EOR \reg1, \reg2 EOR \reg2, \reg3 EOR \reg3, \reg1 POP \reg3 POP \reg2 POP \reg1 7 Peter Samarin and Kerstin Lemke-Rust Detecting Similar Code Segments through Side Channel Leakage 14 / 22

slide-24
SLIDE 24

Results: Maximum Projection

0.2 0.4 0.6 0.8 1 500 1000 1500 2000 2500 3000 3500 |Correlation| Clock cycle

genuine in genuine

PU L KE L AK SB MC AK SB MC AK SB MC AK SB MC AK SB MC AK SB MC AK SB MC AK SB MC AK SB MC AK SB AK S PO

0.2 0.4 0.6 0.8 1 500 1000 1500 2000 2500 3000 3500 |Correlation| Clock cycle

addr in genuine

PU L KE L AK SB MC AK SB MC AK SB MC AK SB MC AK SB MC AK SB MC AK SB MC AK SB MC AK SB MC AK SB AK S PO

0.2 0.4 0.6 0.8 1 500 1000 1500 2000 2500 3000 3500 |Correlation| Clock cycle

swap in genuine

PU L KE L AK SB MC AK SB MC AK SB MC AK SB MC AK SB MC AK SB MC AK SB MC AK SB MC AK SB MC AK SB AK S PO

0.2 0.4 0.6 0.8 1 500 1000 1500 2000 2500 3000 3500 |Correlation| Clock cycle

addr+swap in genuine

PU L KE L AK SB MC AK SB MC AK SB MC AK SB MC AK SB MC AK SB MC AK SB MC AK SB MC AK SB MC AK SB AK S PO

PU - push registers PO - pop registers KE - key expansion AK - add round key L - load key/plaintext S - store ciphertext SB - shift rows and subbytes MC - mix columns Peter Samarin and Kerstin Lemke-Rust Detecting Similar Code Segments through Side Channel Leakage 15 / 22

slide-25
SLIDE 25

Results: Maximum Projection Contd.

0.2 0.4 0.6 0.8 1 500 1000 1500 2000 2500 3000 3500 |Correlation| Clock cycle

dummy NOPs in genuine

PU L KE L AK SB MC AK SB MC AK SB MC AK SB MC AK SB MC AK SB MC AK SB MC AK SB MC AK SB MC AK SB AK S PO

0.2 0.4 0.6 0.8 1 500 1000 1500 2000 2500 3000 3500 |Correlation| Clock cycle

dummy smart in genuine

PU L KE L AK SB MC AK SB MC AK SB MC AK SB MC AK SB MC AK SB MC AK SB MC AK SB MC AK SB MC AK SB AK S PO

0.2 0.4 0.6 0.8 1 500 1000 1500 2000 2500 3000 3500 |Correlation| Clock cycle

dummy smart+addr+swap in genuine

PU L KE L AK SB MC AK SB MC AK SB MC AK SB MC AK SB MC AK SB MC AK SB MC AK SB MC AK SB MC AK SB AK S PO

PU - push registers PO - pop registers KE - key expansion AK - add round key L - load key/plaintext S - store ciphertext SB - shift rows and subbytes MC - mix columns Peter Samarin and Kerstin Lemke-Rust Detecting Similar Code Segments through Side Channel Leakage 16 / 22

slide-26
SLIDE 26

Results: Detection of Similar Code Segments

genuine genuine 0.96 addr 0.64 swap 0.73 addr+swap 0.52 dummy NOPs 0.84 dummy smart 0.83 dummy smart+addr+swap 0.51

(a) Global similarity

AK SB MC KE 0.95 0.98 0.98 0.96 0.61 0.52 0.76 0.60 0.84 0.62 0.78 0.80 0.59 0.37 0.64 0.45 0.92 0.72 0.87 0.86 0.82 0.75 0.85 0.85 0.54 0.36 0.63 0.44

(b) Local similarity

Peter Samarin and Kerstin Lemke-Rust Detecting Similar Code Segments through Side Channel Leakage 17 / 22

slide-27
SLIDE 27

Results: Similarity Matrix of genuine vs. genuine

Peter Samarin and Kerstin Lemke-Rust Detecting Similar Code Segments through Side Channel Leakage 18 / 22

slide-28
SLIDE 28

Results: Similarity Matrix of addr vs. genuine

Peter Samarin and Kerstin Lemke-Rust Detecting Similar Code Segments through Side Channel Leakage 19 / 22

slide-29
SLIDE 29

Related Work

◮ (Becker et al. 2011)

◮ Detect Hamming weight of the instructions ◮ Embed watermarks detectable in the side channel ◮ Problem: not all microcontrollers leak the Hamming weight

  • f the instruction

◮ (Strobel et al. 2015)

◮ Side channel disassembler ◮ Use electromagnetic emanation ◮ Detect individual instructions ◮ Problem: Only tested on one microcontroller

◮ (Durvaux et al. 2012)

◮ Use power consumption as its own watermark ◮ Horizontal correlation one two traces ◮ Problem: sensitive to the dummy cycles Peter Samarin and Kerstin Lemke-Rust Detecting Similar Code Segments through Side Channel Leakage 20 / 22

slide-30
SLIDE 30

Conclusions and Future Work

◮ Method for detecting similarity of programs using side

channels

◮ We can detect identical code segments in the power

consumption of a microcontroller

◮ Our method also works well with cases where many

dummy cycles have been inserted

◮ Interesting application: detecting unlicensed

implementations of patented technology Future Work

◮ Combination of horizontal and vertical approaches ◮ Non-linear programs

◮ dissect into data-dependent code paths ◮ compute similarity for each code path

◮ Evaluation using different microcontrollers ◮ Dealing with random data

Peter Samarin and Kerstin Lemke-Rust Detecting Similar Code Segments through Side Channel Leakage 21 / 22

slide-31
SLIDE 31

Questions?

Peter Samarin and Kerstin Lemke-Rust Detecting Similar Code Segments through Side Channel Leakage 22 / 22

slide-32
SLIDE 32

Backup: Furious vs Furious Wrong Data

0.2 0.4 0.6 0.8 1 500 1000 1500 2000 2500 3000 3500

|Correlation| Clock cycle

0.2 0.4 0.6 0.8 1 500 1000 1500 2000 2500 3000 3500

|Correlation| Clock cycle Peter Samarin and Kerstin Lemke-Rust Detecting Similar Code Segments through Side Channel Leakage 22 / 22

slide-33
SLIDE 33

Backup: Visual Inspection

AES-0 AES Labor Furious Fast Fantastic AES-0 AES Labor Furious Fast Fantastic AES-0 AES Labor Furious Fast Fantastic AES-0 AES Labor Furious Fast Fantastic AES-0 AES Labor Furious Fast Fantastic ExpandKey ShiftRows and SubBytes AddRoundKey MixColumns 1 2 2 2

Peter Samarin and Kerstin Lemke-Rust Detecting Similar Code Segments through Side Channel Leakage 22 / 22