AUTOMATED SOFTWARE PROTECTION FOR THE MASSES AGAINST SIDE-CHANNEL - - PowerPoint PPT Presentation

▶
automated software protection for the masses against side
SMART_READER_LITE
LIVE PREVIEW

AUTOMATED SOFTWARE PROTECTION FOR THE MASSES AGAINST SIDE-CHANNEL - - PowerPoint PPT Presentation

Mar 20, 2023 174 likes •717 views

Nicolas Belleville 1 Damien Courouss 1 Karine Heydemann 2 1 Univ Grenoble Alpes, CEA, List, F-38000 Grenoble, France Henri-Pierre Charles 1 firstname.lastname@cea.fr 2 Sorbonne Universit, CNRS, LIP6, F-75005, Paris, France

slide-1
SLIDE 1

AUTOMATED SOFTWARE PROTECTION FOR THE MASSES AGAINST SIDE-CHANNEL ATTACKS

Nicolas Belleville 1 Damien Couroussé 1 Karine Heydemann 2 Henri-Pierre Charles 1

1 Univ Grenoble Alpes, CEA, List, F-38000 Grenoble, France

firstname.lastname@cea.fr

2 Sorbonne Université, CNRS, LIP6, F-75005, Paris, France

firstname.lastname@lip6.fr

slide-2
SLIDE 2

| 2

SIDE CHANNEL ATTACKS

Electromagnetic emissions Power consumption … Ciphertexts produced Try to find the key using:

  • measurements
  • ciphertexts or plaintexts
  • a consumption model
slide-3
SLIDE 3

| 3

SIDE CHANNEL ATTACKS

x1

???

Electromagnetic emissions Power consumption … Ciphertexts produced Try to find the key using:

  • measurements
  • ciphertexts or plaintexts
  • a consumption model
slide-4
SLIDE 4

| 4

SIDE CHANNEL ATTACKS

x10

???

Electromagnetic emissions Power consumption … Ciphertexts produced Try to find the key using:

  • measurements
  • ciphertexts or plaintexts
  • a consumption model
slide-5
SLIDE 5

| 5

SIDE CHANNEL ATTACKS

x100

???

Electromagnetic emissions Power consumption … Ciphertexts produced Try to find the key using:

  • measurements
  • ciphertexts or plaintexts
  • a consumption model
slide-6
SLIDE 6

| 6

SIDE CHANNEL ATTACKS

x290

฀

Electromagnetic emissions Power consumption … Ciphertexts produced Try to find the key using:

  • measurements
  • ciphertexts or plaintexts
  • a consumption model

<5kB

(less than a hello world binary file)

slide-7
SLIDE 7

| 7

SIDE CHANNEL ATTACKS

x290

฀?

Electromagnetic emissions Power consumption … Ciphertexts produced Try to find the key using:

  • measurements
  • ciphertexts or plaintexts
  • a consumption model

Make this bigger!

<5kB

(less than a hello world binary file)

slide-8
SLIDE 8

| 8

SOFTWARE COUNTERMEASURES

Hiding Masking

secret: 01001101

share 1: 10010001 share 2: 11011100 share1 xor share2 = secret secret: 01001101

add noise split secret into shares Measurements are noisy Measurements are no more directly correlated to secret Attack is harder

slide-9
SLIDE 9

| 9

SOFTWARE COUNTERMEASURES

Hiding Masking

secret: 01001101

share 1: 10010001 share 2: 11011100 share1 xor share2 = secret secret: 01001101

add noise split secret into shares Measurements are noisy Measurements are no more directly correlated to secret Attack is harder

slide-10
SLIDE 10

| 10

CODE POLYMORPHISM WITH RUNTIME CODE GENERATION

Electromagnetic emissions Power consumption Function’s result change this don’t change this

slide-11
SLIDE 11

| 11

CODE POLYMORPHISM WITH RUNTIME CODE GENERATION

Electromagnetic emissions Power consumption Executed instructions Manipulated data depends on Function’s result impacts change this don’t change this

slide-12
SLIDE 12

| 12

CODE POLYMORPHISM WITH RUNTIME CODE GENERATION

Electromagnetic emissions Power consumption impacts Function’s result impacts change this don’t change this Executed instructions Manipulated data

slide-13
SLIDE 13

| 13

CODE POLYMORPHISM WITH RUNTIME CODE GENERATION

Electromagnetic emissions Power consumption impacts Function’s result impacts change this don’t change this change this Executed instructions Manipulated data

slide-14
SLIDE 14

| 14

CODE POLYMORPHISM WITH RUNTIME CODE GENERATION

Electromagnetic emissions Power consumption Instructions executed Data manipulated impacts Function’s result impacts change this don’t change this Electromagnetic emissions Power consumption Instructions executed Data manipulated Function’s result Electromagnetic emissions Power consumption Executed instructions Manipulated data Function’s result change this

slide-15
SLIDE 15

| 15

CODE POLYMORPHISM WITH RUNTIME CODE GENERATION

Electromagnetic emissions Power consumption Instructions executed Data manipulated impacts Function’s result impacts change this don’t change this Function’s result Function’s result change this Runtime code generation regenerate a different code regularly Electromagnetic emissions Power consumption Instructions executed Data manipulated Electromagnetic emissions Power consumption Executed instructions Manipulated data

  • nly use code

transformations that preserve program semantics

slide-16
SLIDE 16

| 16

CODE POLYMORPHISM WITH RUNTIME CODE GENERATION

Issues Our contributions Countermeasures are usually manually applied Automatic application of the countermeasure Countermeasures are usually given for particular ciphers Any code can be hardened Target a wide range of platforms (be lightweight) Use static memory allocation Allocation of a realistic size (don’t waste memory) Use specialized code generation An attacker may write on an executable memory section Use the specialization of the generator to manage memory permissions Hard to have a good trade-off between security and performance Highly configurable → possible to find a trade-off

slide-17
SLIDE 17

| 17

SIDE CHANNEL ATTACKS

x290

฀

Electromagnetic emissions Power consumption … Ciphertexts produced Try to find the key using:

  • measurements
  • ciphertexts or plaintexts
  • a consumption model

Unprotected 5kB

(less than a hello world binary file)

slide-18
SLIDE 18

| 18

SIDE CHANNEL ATTACKS

x290

Electromagnetic emissions Power consumption … Ciphertexts produced Try to find the key using:

  • measurements
  • ciphertexts or plaintexts
  • a consumption model

Protected

???

x2.5 overhead

5kB

(less than a hello world binary file)

slide-19
SLIDE 19

| 19

SIDE CHANNEL ATTACKS

x1 000

Electromagnetic emissions Power consumption … Ciphertexts produced Try to find the key using:

  • measurements
  • ciphertexts or plaintexts
  • a consumption model

Protected

???

x2.5 overhead

16kB

(one second

  • f an mp3 file)
slide-20
SLIDE 20

| 20

SIDE CHANNEL ATTACKS

x10 000

Electromagnetic emissions Power consumption … Ciphertexts produced Try to find the key using:

  • measurements
  • ciphertexts or plaintexts
  • a consumption model

Protected

???

x2.5 overhead

160kB

(10 seconds

  • f an mp3 file)
slide-21
SLIDE 21

| 21

SIDE CHANNEL ATTACKS

x100 000

Electromagnetic emissions Power consumption … Ciphertexts produced Try to find the key using:

  • measurements
  • ciphertexts or plaintexts
  • a consumption model

Protected

???

x2.5 overhead

1.6MB

(our paper)

slide-22
SLIDE 22

| 22

SIDE CHANNEL ATTACKS

x1 000 000

Electromagnetic emissions Power consumption … Ciphertexts produced Try to find the key using:

  • measurements
  • ciphertexts or plaintexts
  • a consumption model

Protected

???

x2.5 overhead

16MB

(a 20Mpixels jpeg image)

slide-23
SLIDE 23

| 23

SIDE CHANNEL ATTACKS

x3 800 000

Electromagnetic emissions Power consumption … Ciphertexts produced Try to find the key using:

  • measurements
  • ciphertexts or plaintexts
  • a consumption model

Protected

฀

x2.5 overhead

60MB

(a 3minutes long full HD video)

slide-24
SLIDE 24

| 24

SIDE CHANNEL ATTACKS

x3 800 000

Electromagnetic emissions Power consumption … Ciphertexts produced Try to find the key using:

  • measurements
  • ciphertexts or plaintexts
  • a consumption model

Protected

฀

x13 000 improvement!

x2.5 overhead

60MB

(a 3minutes long full HD video)

slide-25
SLIDE 25

| 25

  • Background
  • Side channel attacks
  • Software countermeasures
  • Code polymorphism
  • Automated application of code polymorphism
  • Overview
  • Code transformations used
  • Memory management
  • Experimental evaluation
  • Performance evaluation
  • Security evaluation

OUTLINE

slide-26
SLIDE 26

| 26

  • Background
  • Side channel attacks
  • Software countermeasures
  • Code polymorphism
  • Automated application of code polymorphism
  • Overview
  • Code transformations used
  • Memory management
  • Experimental evaluation
  • Performance evaluation
  • Security evaluation

OUTLINE

slide-27
SLIDE 27

| 27

OVERVIEW

.c Modified compiler .c Annotated function Wrapper Specialized runtime code generator Compiler binary

STATICALLY

Main idea: The annotated function is replaced by a wrapper and a generator

slide-28
SLIDE 28

| 28

OVERVIEW

.c Modified compiler .c Annotated functions Wrappers Specialized runtime code generators Compiler binary

STATICALLY

Main idea: Each annotated function has its own generator (with shared code segments)

slide-29
SLIDE 29

| 29

OVERVIEW

.c Modified compiler .c Annotated functions Wrappers Specialized runtime code generators Compiler binary

STATICALLY

Main idea: Each annotated function has its own generator (with shared code segments)

slide-30
SLIDE 30

| 30

OVERVIEW

.c Modified compiler .c Annotated functions Wrappers Specialized runtime code generators Compiler binary

STATICALLY

Main idea: Each annotated function has its own generator (with shared code segments)

slide-31
SLIDE 31

| 31

OVERVIEW

.c Modified compiler .c Annotated function Wrapper Specialized runtime code generator Compiler binary

STATICALLY

slide-32
SLIDE 32

| 32

① calls ② generates

OVERVIEW

.c Modified compiler .c Annotated function Wrapper Specialized runtime code generator Compiler binary

RUNTIME STATICALLY

Runtime code generator Wrapper polymorphic instance

③ calls

polymorphic instance

polymorphic instance

polymorphic instance polymorphic instance polymorphic instance

polymorphic instance

Main idea: At runtime, a new polymorphic instance is generated at each call

slide-33
SLIDE 33

| 33

① calls ② generates

OVERVIEW

.c Modified compiler .c Annotated function Wrapper Specialized runtime code generator Compiler binary

RUNTIME STATICALLY

Runtime code generator Wrapper

③ calls ④ calls ⑤ calls …

How to find a good trade-off between security and performance? How to have variability in between generations? polymorphic instance polymorphic instance

polymorphic instance

polymorphic instance polymorphic instance polymorphic instance

polymorphic instance

Main idea: At runtime, a new polymorphic instance is generated at each call

  • nce in a while
slide-34
SLIDE 34

| 34

① calls ② generates

OVERVIEW

.c Modified compiler .c Annotated function Wrapper Specialized runtime code generator Compiler binary

RUNTIME STATICALLY

Runtime code generator Wrapper

③ calls ④ calls ⑤ calls …

polymorphic instance polymorphic instance

polymorphic instance

polymorphic instance polymorphic instance polymorphic instance

polymorphic instance

Main idea: The size of polymorphic instances vary How to allocate memory?

?

slide-35
SLIDE 35

| 35

OVERVIEW

.c Modified compiler .c Compiler binary

STATICALLY

front-end middle-end back-end code analysis + code emission All standard compiler optimizations are applied → the runtime generated code is statically optimized We emit C code instead of ARM Thumb assembly We perform static analysis to help the runtime code generator (memory and register)

slide-36
SLIDE 36

| 36

OVERVIEW

.c Modified compiler .c Annotated function Compiler binary

STATICALLY

Idée: Un wrapper et un générateur #pragma odo_polymorphic int f_critical(int a, int b) { int c = a^b; a = a+b; a = a % c; return a; } void SGPC_f_critical() { raise_interrupt_rm_X_add_W(code_f); reg_t r[] = {0,1,2,3,4,5,6,...,12,13,14,15}; push_T2_callee_saved_registers(); eor_T2(r[4], r[1], r[0]); add_T2(r[0], r[1], r[0]); sdiv_T2(r[1], r[0], r[4]); mls_T2(r[0], r[1], r[4], r[0]); pop_T2_callee_saved_registers(); raise_interrupt_rm_W_add_X(code_f); } int f_critical(int a, int b) { if (SHOULD_BE_REGENERATED()) SGPC_f_critical(); return code_f(a, b); }

slide-37
SLIDE 37

| 37

  • Background
  • Side channel attacks
  • Software countermeasures
  • Code polymorphism
  • Automated application of code polymorphism
  • Overview
  • Code transformations used
  • Memory management
  • Experimental evaluation
  • Performance evaluation
  • Security evaluation

OUTLINE

slide-38
SLIDE 38

| 38

xor r6, r5, r8 add r4, r4, r5

CODE TRANSFORMATIONS USED AT RUNTIME

Register shuffling

RANDOM general purpose

register permutation

Instruction shuffling

independent instructions are emitted in a RANDOM

  • rder

Semantic variants

replacement of an instruction by a RANDOMLY selected semantic variant

Noise instructions

insertion of a RANDOM number of RANDOMLY chosen noise instructions

Dynamic noise

RANDOM insertion of noise

instructions with a RANDOM jump add r4, r4, r5 xor r6, r5, r8 add r11, r11, r7 xor r8, r7, r5 r4

r11

… add r4, r4, r5 xor r6, r5, #12348 xor r6, r6, r8 xor r6, r6, #12348 add r4, r4, r5 sub r7, r6, r2 load r3, r10, #53 xor r6, r5, r8

useless instructions

add r4, r4, r5 jump 0, 1 or 2 instructions sub r7, r6, r2 load r3, r10, #53 xor r6, r5, r8

slide-39
SLIDE 39

| 39

CONFIGURABILITY

Period of regeneration

ℕ

Register shuffling

{𝟏, 𝟐}

Instruction shuffling

{𝟏, 𝟐}

Semantic variants

{𝟏, 𝟐, 𝟑}

Noise instructions

{𝟏, 𝟐, 𝟑} × ℝ × ℕ

Dynamic noise

ℕ

Total configuration space: {𝟏, 𝟐}𝟑× {𝟏, 𝟐, 𝟑}𝟑× ℝ × ℕ𝟒

slide-40
SLIDE 40

| 40

  • Background
  • Side channel attacks
  • Software countermeasures
  • Code polymorphism
  • Automated application of code polymorphism
  • Overview
  • Code transformations used
  • Memory management
  • Memory allocation & overflow prevention
  • Memory permissions
  • Experimental evaluation
  • Performance evaluation
  • Security evaluation

OUTLINE

slide-41
SLIDE 41

| 41

① calls ② generates

OVERVIEW

.c Modified compiler .c Annotated function Wrapper Specialized runtime code generator Compiler binary

RUNTIME STATICALLY

Runtime code generator Wrapper

③ calls ④ calls ⑤ calls …

polymorphic instance polymorphic instance

polymorphic instance

polymorphic instance polymorphic instance polymorphic instance

polymorphic instance

Main idea: The size of polymorphic instances vary How to allocate memory?

?

slide-42
SLIDE 42

| 42

MEMORY STATIC ALLOCATION

(type of curve one can obtain) Probability Amount of used memory Worst case is terrible

Distribution of generated codes’ size

slide-43
SLIDE 43

| 43

MEMORY STATIC ALLOCATION

Probability Worst case is terrible

Compute a more realistic size using a threshold

threshold: probability of having an overflow 10-6 by default Amount to allocate Range where an

  • verflow is possible

For a 100 instructions code, allocated size is 5x smaller than worst case!

(configuration low (defined later))

(type of curve one can obtain)

Distribution of generated codes’ size

Amount of used memory

slide-44
SLIDE 44

| 44

OVERFLOW PREVENTION

.c Modified compiler .c Compiler binary

RUNTIME STATICALLY

Runtime code generator Wrapper polymorphic instance polymorphic instance

polymorphic instance

polymorphic instance polymorphic instance polymorphic instance

polymorphic instance

always keep space for useful instructions (limit polymorphism if necessary) computes the size of useful instructions puts the information directly in runtime code generator’s code ① ② ③ ① ② ③

slide-45
SLIDE 45

| 45

  • Background
  • Side channel attacks
  • Software countermeasures
  • Code polymorphism
  • Automated application of code polymorphism
  • Overview
  • Code transformations used
  • Memory management
  • Memory allocation & overflow prevention
  • Memory permissions
  • Experimental evaluation
  • Performance evaluation
  • Security evaluation

OUTLINE

slide-46
SLIDE 46

| 46

MANAGEMENT OF MEMORY PERMISSIONS Objective: Guarantee W ⊕ X and that only the generator can write into the buffer

Generation begins Specialized runtime code generator Interrupt handler Instance buffer (memory allocated) Check address of interrupt good bad X only to W only ERROR ① ② Generate code X only W only W only to X only ERROR ① X only Generation ends resume execution resume execution Check address of interrupt ② bad good raise interrupt raise interrupt

slide-47
SLIDE 47

| 47

  • Background
  • Side channel attacks
  • Software countermeasures
  • Code polymorphism
  • Automated application of code polymorphism
  • Overview
  • Code transformations used
  • Memory management
  • Experimental evaluation
  • Performance evaluation
  • Security evaluation

OUTLINE

slide-48
SLIDE 48

| 48

  • 15 different test cases
  • 4 different selected configurations
  • none: no polymorphism
  • low: only noise instructions, generation is done every 250 executions
  • Theoretical number of variants is already very high!

>6×1022 variants for a 10 instructions code >10704 variants for the 278 instructions AES we use

  • medium: all transformations activated, generation is done every execution
  • high: all transformations activated, different probability model for noise instructions

insertion, generation is done every execution

  • STM32 board (ARM cortex M3 – 24 MHz – 8kB of RAM)

EXPERIMENTAL SETUP

slide-49
SLIDE 49

| 49

PERFORMANCE EVALUATION

Configuration Execution time overhead (geometric mean) Size overhead (geometric mean) none x1.40 x1.70 low x2.31 x2.87 medium x2.45 x3.44 high x4.03 x3.81 more results in our paper

  • verheads depend
  • n configuration

→ trade-off to find generation done in linear complexity

slide-50
SLIDE 50

| 50

  • Background
  • Side channel attacks
  • Software countermeasures
  • Code polymorphism
  • Automated application of code polymorphism
  • Overview
  • Code transformations used
  • Memory management
  • Experimental evaluation
  • Performance evaluation
  • Security evaluation

OUTLINE

slide-51
SLIDE 51

| 51

  • CPA on Sbox output with HW
  • Success rate at 0.8 in
  • 290 traces for unprotected AES
  • 3 800 000 traces for configuration low

SECURITY EVALUATION

Technical details: PicoScope 2208A, EM probe RF-U 5-2 (Langer), PA 303 preamplifier (Langer) Sampling at 500 Msample/s with 8bits resolution, 24500 samples per trace

13000x more traces needed! Execution time

  • verhead: x2.5

including generation cost!

more results in our paper

slide-52
SLIDE 52

| 52

  • Automatic ฀
  • Configurable ฀
  • Efficient ฀
  • With static memory

allocation of a realistic size ฀

  • With memory permission

management ฀

  • Usable on constrained devices ฀
  • Open question: interest of code polymorphism

against fault injection attacks?

CONCLUSION

slide-53
SLIDE 53

Centre de Saclay Nano-Innov PC 172 91191 Gif sur Yvette Cedex Centre de Grenoble 17 rue des Martyrs 38054 Grenoble Cedex

Thank you for your attention Questions?

contact: nicolas.belleville@cea.fr

Automated software protection for the masses against side-channel attacks

Nicolas Belleville Damien Couroussé Karine Heydemann Henri-Pierre Charles

This work was partially funded by the French National Research Agency (ANR) as part of the projects COGITO and PROSECCO, respectively funded by the programs INS-2013 under grant agreement ANR-13-INSE-0006-01 and AAP-2015 under grant agreement ANR-15-CE39.