This document discusses cache-based side channel attacks and proposes new cache designs to prevent them. It begins with an introduction to side channel attacks and two examples: an RSA attack using cache interference and an AES attack analyzing computation times. It then proposes two new cache architectures: a partitioned locked cache that isolates sensitive processes and a randomly permuting cache that obscures which cache lines are evicted. The models are evaluated using an OpenSSL AES implementation, finding the proposed designs prevent attacks with minimal performance overhead.

1. LOGO
New Cache Designs for Thwarting
Software Cache-based Side Channel
Attacks - Z. Wang & R. B. Lee
Anestis Bechtsoudis
mpechtsoud@ceid.upatras.gr
Patra 2010

2. Cache Based Side Channel Attacks
Contents
1
Introduction
2
Threat Model and Attacks
3
Proposed Models
4
Evaluation
5
Conclusions
2
COMPANY LOGO

3. Cache Based Side Channel Attacks
1.
Introduction
3
COMPANY LOGO

4. Cache Based Side Channel Attacks
Introduction 1/4
 Information intensive society – imperative
need for security
 Design of cryptographic systems to ensure
the data protection
 Extensive test to cryptosystems over time
 Cryptanalysis: the study of techniques to
reveal the secret parameters of a security
system
4
COMPANY LOGO

5. Cache Based Side Channel Attacks
Introduction 2/4
 Classical cryptanalysis approach
 Weaknesses in the algorithm – mathematical model
 Attacks based on: ciphertext-only, known plaintext,
chosen plaintext/ciphertext …
 Black box approach of the cryptosystem
 The cryptographic primitive is actually
implemented in hardware
 Modern cryptanalysis: attacker knows much
more for the device – side channel leakage
5
COMPANY LOGO

6. Cache Based Side Channel Attacks
Introduction 3/4
6
COMPANY LOGO

7. Cache Based Side Channel Attacks
Introduction 4/4
7
COMPANY LOGO

8. Cache Based Side Channel Attacks
2.
Threat Model and Attacks
8
COMPANY LOGO

9. Cache Based Side Channel Attacks
Threat Model and Attacks 1/6
 Goal of the adversary is to learn information
that he has no legitimate access to
 Adversary: one or more unprivileged user
processes, including remote clients, in the
server where the secrets are processed
 No physical access to the device
 Goal achieved by performing legitimate
operations – normal process
 Victim and adversary are isolated processes
9
COMPANY LOGO

10. Cache Based Side Channel Attacks
Threat Model and Attacks 2/6
Percival’s attack on OpenSSL implementation
of RSA algorithm in a SMT CPU
 RSA core operation: modulo exponentiation –
implemented with a series of ^2 and *
 The encryption key is divided into segments
 For each *, a multiplier is selected from precomputed constants stored in a LUT
 Segment of key is used to index the LUT
10
COMPANY LOGO

11. Cache Based Side Channel Attacks
Threat Model and Attacks 3/6
 Attacker manages to run simultaneously
 Attack process sequentially and repeatedly
accesses an array, thus loading data to
occupy all cache lines
 At the same time he measures the delay for
each access to detect cache misses (ex. rdtsc
timer in intel x86)
 Victim’s cache accesses evict attacker’s data,
enabling detection from the attacker
11
COMPANY LOGO

12. Cache Based Side Channel Attacks
Threat Model and Attacks 4/6
Cache
RAM
RSA
Attacker
 The attacker can identify which table entry is
accessed -> the index used -> segment of
the key
12
COMPANY LOGO

13. Cache Based Side Channel Attacks
Threat Model and Attacks 5/6
Bernstein’s Attack on AES
 AES - “Black Box” software module
 Give inputs and measure computation time
 The execution time is input dependant and
can be exploited to recover secret key
 Attack consists of three phases: Learning,
Attacking and Key Recovery
 Statistical correlation analysis
13
COMPANY LOGO

14. Cache Based Side Channel Attacks
Threat Model and Attacks 6/6
14
COMPANY LOGO

15. Cache Based Side Channel Attacks
3.
Proposed Models
15
COMPANY LOGO

16. Cache Based Side Channel Attacks
Proposed Models 1/4
 Problem -> Directly or indirectly cache
interference
 Learn from attacks and rewrite software
 Solutions are attack specific and performance
degradation (2x, 4x slower)
 Authors attempt to eliminate the root cause
with minimum impact and low cost
 Ideas -> Partitioning - Randomization
16
COMPANY LOGO

17. Cache Based Side Channel Attacks
Proposed Models 2/4
Partition-Locked Cache (PLCache)
L
ID
Original Cache Line
17
COMPANY LOGO

18. Cache Based Side Channel Attacks
Proposed Models 3/4
Random Permutation Cache (RPCache)
 Introduce randomization factor – no useful
information about which cache lines evicted
 Memory-to-cache mappings
18
COMPANY LOGO

19. Cache Based Side Channel Attacks
Proposed Models 4/4
19
COMPANY LOGO

20. Cache Based Side Channel Attacks
4.
Evaluation
20
COMPANY LOGO

21. Cache Based Side Channel Attacks
Evaluation 1/
 OpenSSL 0.9.7a AES implementation
 Traditional cache, L1 PLCache and L1 RPCache
 5KByte AES protected data
 L2 large enough – no performance impact
21
COMPANY LOGO

22. Cache Based Side Channel Attacks
Evaluation 1/
 PLCache & RPCache implemented in M-Sim v2.0
22
COMPANY LOGO

23. Cache Based Side Channel Attacks
5.
Conclusions
23
COMPANY LOGO

24. Cache Based Side Channel Attacks
Conclusions
 Cache-based side channel attacks can harm
general purpose cache based systems
 Software solution -> attack specific
 Hardware solutions -> general purpose
 PLCache: minimal hardware cost – software
developer must use different API
 RPCache: area & complexity in hardware – no
special treatment from software developers
24
COMPANY LOGO

25. LOGO
Anestis Bechtsoudis
mpechtsoud@ceid.upatras.gr
Patra 2010