PE‐Probe: Leveraging Packer Detection and Morphological p gInformation to Detect Malicious Portable ExecutablesPortable Executables

M. Zubair Shafiq, S. Momina Tabish, Muddassar Farooq

Next Generation Intelligent Networks Research Center (nexGIN RC)National University of Computer and Emerging Sciencesy p g g

Islamabad, Pakistanhttp://www.nexginrc.org/

1

AgendaAgendaProjects’ Introduction

Motivation & Problem Statement

d l i

Motivation & Problem Statement

Proposed Solution

Results

Q/A

2

Its in your Hands, like its inyour Eyes and Face

It is believed that keystrokes of people are distinct from each other just likeare distinct from each other just like their faces, finger prints, and eyes

Doesn’t require any extra hardware for identification

3

User Authentication SystemUser Authentication System

4

IMS Security ChallengesIMS Security Challenges 

IP Multimedia Subsystem (IMS) 

&Next Generation 

Service Delivery PlatformService Delivery Platform

5

Malware DetectionMalware DetectionSignature Based:

Malware DetectionMalware Detection

Malware DetectionMalware Detection

. Detection on the basis of  known byte sequences

Signature Signature  Non‐SignatureNon‐

Signature

. Unable to detect new malware. Regular updates requiredg

BasedgBased Signature 

BasedSignature Based

required 

StaticStatic DynamicDynamic Non‐Signature Based:. Detection on the basis of  

After‐ExecutionAfter‐

Execution In‐ExecutionIn‐Execution

smarter features. Able to detect new malware

6

ExecutionExecution . Regular updates may not be necessary

Malware DetectionMalware DetectionStatic Detection:

Malware DetectionMalware DetectionStatic Detection:. Detection on the basis of  file as residing on secondary storage

Malware DetectionMalware Detection

secondary storage. Prone to techniques such as code‐obfuscationSignature Signature  Non‐

SignatureNon‐

SignaturegBasedgBased Signature 

BasedSignature Based

Dynamic Detection:. Detection on the basis of run time behavior (a

StaticStatic DynamicDynamic

run‐time behavior (a more direct look). Resilient to techniques such as code obfuscation

After‐ExecutionAfter‐

Execution In‐ExecutionIn‐Execution

7

. High processing overhead

ExecutionExecution

Malware DetectionMalware DetectionMalware DetectionMalware Detection

After‐Execution Detection:. Forensic Analysis

Malware DetectionMalware Detection

Forensic Analysis. Lower processing overhead

Signature Signature  Non‐SignatureNon‐

SignaturegBasedgBased Signature 

BasedSignature Based

In‐Execution i

StaticStatic DynamicDynamic

Detection:. End user tool. High processing After‐

ExecutionAfter‐

Execution In‐ExecutionIn‐Execution

8

overheadExecutionExecution

MotivationMotivation 

9

Motivation(2)Motivation(2) 

In Year 2008 Only [11]

•5 491 new software vulnerabilities•5,491 new software vulnerabilities •1.6 million new malware signatures •245 million new attacks 1 T illi d ll i•1 Trillion dollar in revenues

10

Motivation(3)Motivation(3) 

Norton AV Command AV McAfee AV

Chernobyl‐1.4 Not detected Not detected Not detected

F0sf0r0 Not detected Not detected Not detected

Hare Not detected Not detected Not detected

Z0mbie 6 b Not detected Not detected Not detectedZ0mbie‐6.b Not detected Not detected Not detected

11

Motivation(4)Motivation(4) 

Issues with Commercial Anti‐virus software

•Cannot detect ne mal are•Cannot detect new malware•Size of signature database cannot scaleg•Signatures are evaded by code 

bf ti t h i ( h ki )obfuscation techniques (such as packing)

12

Motivation(5)Motivation(5) 

Packing of Malware [12]

•50% ne mal are are simpl re packed•50% new malware are simply re‐packed versions of known malware

92% l ki t h i•92% malware use packing techniques

13

Motivation(6)Motivation(6) 

Non‐signature based Malware Detection SchemesSchemes

M hi l l d•Machine‐level code•Disassembled codeDisassembled code•Static calls from disassembled code•Run‐time API calls

14

Motivation(7)Motivation(7) 

Issues with Non‐signature based Schemes

•High run‐time computational complexityHi h f l l t•High false alarm rates

•Low reliability (e.g. crash, halt, evasion)y ( g , , )

15

Problem Statement

16

P bl St t tProblem Statement

•Non signature based solution•Non‐signature based solution•Low run‐time complexity•Low false alarmsR b t t P ki•Robustness to Packing

•Must not use an unpacker for detectionMust not use an unpacker for detection

17

PE File

MS D S b

PE file Header

MS Dos Stub

PE file HeaderExecutable  section

Read‐only sectionExisting non signature based

Writable section

Existing non signature based Schemes are based on this area of PE file

Read/Write section

18

MS DOS t b

PE Signature

MS DOS stub

COFF file Header

Optional HeaderStandard FieldsStandard Fields

Window Specific fieldsData directories

Section Table

Section 1

RVA /

Section 2

Section 3

/Pointe

Section n

ers

19

List of Features from PE fileList of Features from PE file

20

PE MinerPE Miner

21

Architecture of PE-Probe

22

Distribution of Number of standard sections

23

Distribution of Number of entries in Import Address Table

24

Distribution of Entropy of PE Header

25

Architecture of PE-Probe

26

Structural features for non-packed PE files

27

Distribution plot for “major linker version” feature

28

Architecture of PE-Probe

29

KL Divergence of features of packed/non-packed PE files

30

Structural features for packed PE files

31

Results

32

Dataset – Offensive Computing

33

Classification MetricsClassification Metrics

34

Accuracy of PE-Probe

35

The processing overheads in (seconds/file)

J48 NB RIPPER

SMO IBK J48 NB RIPPER

SMOR R

TRAINING TESTINGPE‐ ‐ 0.0008 0.001 0.269 0.199 0.032 0.001 0.002 0.002 0.002Miner(RFR)

PE‐Miner(PCA)

‐ 0.007 0.001 0.264 0.179 0.035 0.001 0.001 0.001 0.002

PE‐Miner(HWT)

‐ 0.007 0.001 0.252 0.147 0.032 0.001 0.002 0.001 0.002

McBoost 0 021 0 004 1 305 1 122 0 218 0 010 0 007 0 005 0 022McBoost ‐ 0.021 0.004 1.305 1.122 0.218 0.010 0.007 0.005 0.022

Strings ‐ 0.009 0.002 0.799 0.838 0.163 0.003 0.003 0.002 0.003

36

Forensic InformationForensic Information

37

PE‐Miner

38

Conficker Detected as aBackdoor

39

Evolvable Malware Framework

40

ConclusionConclusion• PE Structural Information can be leveraged to detect malware

• Packing Robustness

M hi L i Cl ifi l k d• Machine Learning Classifiers can learn packed and non‐packed models

• Robustness and Evasion analysis in accompanying PE‐Miner paper in RAID 2009accompanying PE‐Miner paper in RAID 2009.

• Zero day detection of Conficker

41

ACKNOWLEDGEMENTACKNOWLEDGEMENT

• Special thanks to National ICT R&D for funding this project.  p j

42

QUESTIONSQUESTIONS

For further information and research papers, visit http://www.nexginrc.org  

43