M. Alexander Helen J. Wang Yunxin Liu Microsoft Research 1 Presented by Zhaoliang Duan.

32
M. Alexander Helen J. Wang Yunxin Liu Microsoft Research 1 Presented by Zhaoliang Duan

Transcript of M. Alexander Helen J. Wang Yunxin Liu Microsoft Research 1 Presented by Zhaoliang Duan.

Page 1: M. Alexander Helen J. Wang Yunxin Liu Microsoft Research 1 Presented by Zhaoliang Duan.

M. Alexander Helen J. Wang Yunxin Liu

Microsoft Research

1

Presented by Zhaoliang Duan

Page 2: M. Alexander Helen J. Wang Yunxin Liu Microsoft Research 1 Presented by Zhaoliang Duan.

2

Introduction Design goal and contribution Defining principals Enforcing principal definitions Implementation Evaluation Conclusion and future work

Page 3: M. Alexander Helen J. Wang Yunxin Liu Microsoft Research 1 Presented by Zhaoliang Duan.

3

Page 4: M. Alexander Helen J. Wang Yunxin Liu Microsoft Research 1 Presented by Zhaoliang Duan.

4

Sandbox runs programs in an isolated space which prevents them from making permanent changes to other programs and data in computer.

From google image on website: http://www.sandboxie.com/

Page 5: M. Alexander Helen J. Wang Yunxin Liu Microsoft Research 1 Presented by Zhaoliang Duan.

5

• It is not enough !• Example: Mutually distrusting content

Page 6: M. Alexander Helen J. Wang Yunxin Liu Microsoft Research 1 Presented by Zhaoliang Duan.

6

• Each application handle content protection has drawbacks

Page 7: M. Alexander Helen J. Wang Yunxin Liu Microsoft Research 1 Presented by Zhaoliang Duan.

7

Content isolation from application is not good!• Security of a users' cloud data is duplicated and entrusted to all of

the user's applications

• Security logic in application is often mixed with error prone content processing logic

Page 8: M. Alexander Helen J. Wang Yunxin Liu Microsoft Research 1 Presented by Zhaoliang Duan.

8

Introduction Design goal and contribution Defining principals Enforcing principal definitions Implementation Evaluation Conclusion and future work

Page 9: M. Alexander Helen J. Wang Yunxin Liu Microsoft Research 1 Presented by Zhaoliang Duan.

9

Contribution:Contribution:

Flexible isolation

Compatibility with browser's isolation policy

Advocate a content-based principal model in which the OS treats content owners as its principals and isolate content of different owners from one another

Generalize the content-based principal model from web browsers to all applications

Easy adaptation of traditional application

Page 10: M. Alexander Helen J. Wang Yunxin Liu Microsoft Research 1 Presented by Zhaoliang Duan.

10

• No sharing across principals or isolation containers

Page 11: M. Alexander Helen J. Wang Yunxin Liu Microsoft Research 1 Presented by Zhaoliang Duan.

11

Introduction Design goal and contribution Defining principals Enforcing principal definitions Implementation Evaluation Conclusion and future work

Page 12: M. Alexander Helen J. Wang Yunxin Liu Microsoft Research 1 Presented by Zhaoliang Duan.

12

Page 13: M. Alexander Helen J. Wang Yunxin Liu Microsoft Research 1 Presented by Zhaoliang Duan.

13

Principal labeling:

Separate content owning

Trust list mechanism

Page 14: M. Alexander Helen J. Wang Yunxin Liu Microsoft Research 1 Presented by Zhaoliang Duan.

14

http://blog.com/alice/index.html

Trust:list=http://blog.com/alice/*

http://youtube.com

http://blog.com/

Page 15: M. Alexander Helen J. Wang Yunxin Liu Microsoft Research 1 Presented by Zhaoliang Duan.

15

Introduction Design goal and contribution Defining principals Enforcing principal definitions Implementation Evaluation Conclusion and future work

Page 16: M. Alexander Helen J. Wang Yunxin Liu Microsoft Research 1 Presented by Zhaoliang Duan.

16

Same principal fetching: Check with IsSamePrincipal algorithm• Owner public key• Trust list

Cross-principal fetching: Data communication; spawning a new principal

Page 17: M. Alexander Helen J. Wang Yunxin Liu Microsoft Research 1 Presented by Zhaoliang Duan.

17

Bit live in the response.

Bit live in the request.

Page 18: M. Alexander Helen J. Wang Yunxin Liu Microsoft Research 1 Presented by Zhaoliang Duan.

18

Introduction Design goal and contribution Defining principals Enforcing principal definitions Implementation Evaluation Conclusion and future work

Page 19: M. Alexander Helen J. Wang Yunxin Liu Microsoft Research 1 Presented by Zhaoliang Duan.

19

Page 20: M. Alexander Helen J. Wang Yunxin Liu Microsoft Research 1 Presented by Zhaoliang Duan.

20

Warping operation: Wininet library of HTTP communication, which remaps its HTTP calls to invoke Service OS fetch call

Application have plug-in interface: Write add-in code to achieve CreatePI() and Embed() functions

Application does not provide plug-in interfaces: Modified the UI code to make room for embedded content

Adaptation onto ServiceOS is feasible.

Page 21: M. Alexander Helen J. Wang Yunxin Liu Microsoft Research 1 Presented by Zhaoliang Duan.

21

Introduction Design goal and contribution Defining principals Enforcing principal definitions Implementation Evaluation Conclusion and future work

Page 22: M. Alexander Helen J. Wang Yunxin Liu Microsoft Research 1 Presented by Zhaoliang Duan.

22

Content processing errors are widespread.

ServiceOS does not rely on large applications to enforce remote content security.

Page 23: M. Alexander Helen J. Wang Yunxin Liu Microsoft Research 1 Presented by Zhaoliang Duan.

23

Test case 1: Uses a RTF Header stack overflow vulnerability to construct a malicious document

Test case 2: Malicious document that uses macros to perform the same attack

The application-based isolation would not be able to stop these two exploits, but serviceOS stopped both exploits

Page 24: M. Alexander Helen J. Wang Yunxin Liu Microsoft Research 1 Presented by Zhaoliang Duan.

24

Startup latencies: • Compare with the startup time of applications' native versions on

Windows.

Overheads on memory usage: • Drawbridge• Loading Excel’s add-in libraries

Performance of content fetch APIs:• Overhead increase following the size of document

Page 25: M. Alexander Helen J. Wang Yunxin Liu Microsoft Research 1 Presented by Zhaoliang Duan.

25

Result 1: In all tests ServiceOS adds less than 200ms to connect to the monitor and initialize

Page 26: M. Alexander Helen J. Wang Yunxin Liu Microsoft Research 1 Presented by Zhaoliang Duan.

26

Result1: Both applications carry a very small memory overhead

Result12: No significant penalty for opening documents from the same owner, but for different owners carry a sizable memory overhead

Page 27: M. Alexander Helen J. Wang Yunxin Liu Microsoft Research 1 Presented by Zhaoliang Duan.

27

Result1: ServiceOS introduces some latency for passing content to renderers

Result 2: Overhead is amortized for larger document sizes

Page 28: M. Alexander Helen J. Wang Yunxin Liu Microsoft Research 1 Presented by Zhaoliang Duan.

28

Introduction Design goal and contribution Defining principals Enforcing principal definitions Implementation Evaluation Conclusion and future work

Page 29: M. Alexander Helen J. Wang Yunxin Liu Microsoft Research 1 Presented by Zhaoliang Duan.

Generalize web browsers’ same-origin policy into an isolation policy suitable for all applications

Advocate a content-based principal model by minimizing the impact of any content including malicious content

Built a substantial prototype system and adapted to it a number of real-world applications

Page 30: M. Alexander Helen J. Wang Yunxin Liu Microsoft Research 1 Presented by Zhaoliang Duan.
Page 31: M. Alexander Helen J. Wang Yunxin Liu Microsoft Research 1 Presented by Zhaoliang Duan.

It need to modify or add plug-in code for each applications

how to partition the system into other meaningful pieces and how to set permissions for each piece

If we take a contemporary OS, simple bug in any of the kernel components allows to bypass of the isolation mechanisms

Page 32: M. Alexander Helen J. Wang Yunxin Liu Microsoft Research 1 Presented by Zhaoliang Duan.

32