Mimesis Aegis: A Mimicry Privacy Shield

1
Mimesis Aegis A Mimicry Privacy Shield A
System's Approach to Data Privacy on Public
Cloud Billy Lau, Simon Chung, Chengyu Song,
Yeongjin Jang, Wenke Lee, and Alexandra
Boldyreva USENIX 14 Presenter Shasha Wen
Some figures are from the author's slides
2
Outline

• Motivation
• Related work
• M-Aegis
• System design
• implementation
• Evaluation
• Limitation
• Conclusion

3
Motivation

• Public cloud services(PCS)
• Trust server to secure our data
• Conflict of interest for data privacy between
  users and PCS providers
• Text communication service
• Smart phone

4
Change the status

• End to end encryption
• Hard in practical
• Good solution
• Users' ease-of-use
• Developers' effort to maintain support
• Feasibility and deployability of solution on a
  mobile system

5
Design goals
For a solution to be secure, it must be properly
isolated from untrusted entities
1
For a solution to be adoptable, it must preserve
the user experience
2
For a solution to be sustainable, it must be
easy to maintain and scalable
3
6
Outline

• Motivation
• Related work
• M-Aegis
• System design
• implementation
• Evaluation
• Limitation
• Conclusion

7
Related work

• Standalone solutions
• Protect data confidentiality
• Good isolation from untrusted entities
• e.g. Gibberbot, TextSecure, SafeSlinger......
• Problem
• Not preserve user experience

1
8
Related work

• Browser Plugin/Extension Solutions
• e.g. Cryptocat, Scramble, NOYB......
• Provides transparent integration
• Problem
• Not applicable to mobile platform
• App repackaging/Rewriting solution
• e.g. Aurasium, Dr. Android......
• Problem
• Breaks app updates
• Isolation model is unclear

2
2
9
Outline

• Motivation
• Related work
• M-Aegis
• System design
• Implementation
• Evaluation
• Limitation
• Conclusion

10
Mimesis Aegis

• Apply end-to-end encryption while preserving user
  experience
• Mimicking GUIs of app of interest
• Interacting with app on behalf of user
• Good isolation model
• Generalizable across different apps
• Resilient to app updates

2
1
3
11
Mimesis Aegis
12
M-Aegis System Design-Threat Model

• In-scope
• Untrusted parties
• Public cloud service providers
• Client side apps
• Middle boxes between a PCS and client side app
• Trusted components
• Hardware, OS, keyboard, M-Aegis components, the
  user
• Out of scope
• availability(denial-of-service)
• Attacks against our TCB

13
M-Aegis System Design-Architecture

• Layer 7.5
• User interface automation/accessibility(UIA)
• Secure
• No developer attention
• Users' sense

14
M-Aegis System Design-Architecture

• Layer 7.5
• Interactions without data confidentiality
• click-through
• Interactions with data confidentiality
• Place opaque GUIs in different color
• Interactions with control GUIs
• Semi-transparent mimic GUIs

15
M-Aegis System Design-Architecture

• UI Automation Manager(UIAM)
• Give M-Aegis the context of the screen
• TCA, GUI tree, content of the GUI
• Provide information to correctly render GUIs
• Per-Target Client App(TCA) Logic
• Handle direct user input
• Decides suitable actions
• Pass or encode
• Encryption and encoding scheme

16
M-Aegis System Design-Architecture

• Cryptographic Module
• Key manager
• Per TCA
• Password based(default) to more sophisticated
  one
• Searchable Encryption Scheme
• Easily-deployable efficiently-searchable
  symmetric encryption(EDESE)

17
Searchable Encryption Scheme

• Normal encryption
• ab, abc
• Bellare et al1
• keywords
• HMACs of unique keywords
• Add dummy keywords
• Bloom filter(BF)
• Efficient set-inclusion tests
• Encodes the positions of on-bits in a BF

1 BELLARE, M., BOLDYREVA, A., AND ONEILL, A.
Determin- istic and efficiently searchable
encryption. In CRYPTO (2007), A. Menezes, Ed.,
vol. 4622 of Lecture Notes in Computer Sci- ence,
Springer, pp. 535552.
18
Searchable Encryption Scheme

• Bloom filter(BF)

• Keyword128bit
• BF 224
• k 10

From http//en.wikipedia.org/wiki/Bloom_filter
19
M-Aegis System Design- User Workflow
User
L7.5
UIAM
Per-TCA logic
App
launch the Gmail
initialize
detect
Detect state
composing
Mimic GUI
send
Mimic GUI
Send to app Click send
Get content Get the key encode
UI
20
M-Aegis Implementation

• UIAM based on UIA libraries
• Monitor events
• WINDOW_CONTEST_CHANGED
• WINDOW_STATE_CHANGED
• VIEW_SCROLLED
• Query for UI node
• Resource ID ? UI Automator Viewer
• performAction()

• Layer 7.5 always-on-top
• Creation of various system windows
• TYPE_SYSTEM_OVERLAY
• TYPE_SYSTEM_ERROR

21
M-Aegis Implementation

• Per-TCA Logic
• Identify signatures for each TCA state
• Event handler
• Type of input data
• Encrypt data into CJK unicode

• Cryptographic Schemes
• AES-GCM-256
• PBKDF2 SHA-1 as HMAC

22
Outline

• Motivation
• Related work
• M-Aegis
• System design
• Implementation
• Evaluation
• Limitation
• Conclusion

23
M-Aegis Evaluations-Performance

• Experimental Setup
• Stock Android phone(LG Nexus 4)
• Android 4.4.2(Kit Kat, API level 19)
• Repeat 10 times and take average

• Preview Encrypted Email
• 76 ms to render plaintext on L-7.5
• Well with expected response time(50-150ms)
• Composing and Sending Encrypted Email
• With longest email
• 953 words, of which 362 are unique
• 205 ms to encrypt, build the search index, and
  encode

24
M-Aegis Evaluations-User Study

• Users college students
• Tasks previewing, reading, composing, sending
  and searching
• Difference and experience

• Report from 15 participants
• No one notice major difference
• One L-7.5 didn't catch up smoothly when scrolled
• One L-7.5 lag when reading emails
• All would use M-Aegis to protect the privacy of
  their data

25
Outline

• Motivation
• Related work
• M-Aegis
• System design
• Implementation
• Evaluation
• Limitation
• Conclusion

26
Limitation

• Data format
• text(Unicode-compatible)
• Not support other types(e.g. Image, audio, video)
• Unavailability of transformation functions
• Other process steps, like compression
• Typographical error during search
• Rare happen
• Encryption schemes that tolerate typographical
  error search without server modification

27
Outline

• Motivation
• Related work
• M-Aegis
• System design
• Implementation
• Evaluation
• Limitation
• Conclusion

28
Conclusion

• Users can control private data using Mimesis
  Aegis
• Layer 7.5
• Strong isolation
• Preserve original user experience
• Scalar to more apps and resilient to app updates
• Prototype on Android
• Minimal overhead
• User study shows acceptable

29
Questions?