Mojitos: A Distributed Steganographic File System

1
Mojitos A Distributed Steganographic File System

• Charles Giefer
• Julia Letchner

2
Presentation Outline

• Motivation
• Steganography
• Distribution
• Previous Work
• Mnemosyne
• StegFS
• Mojitos
• Architecture
• Server Design
• Client Design
• Network Security
• Reliability Scalability
• Simulation Results

3
Motivation - Steganography

• Hides existence of files
• Allows for plausible deniability
• Examples
• User is tortured to give password to encrypted
  data
• Court order demands access to encrypted files
• Bottom line No one can prove the files even
  exist!

4
Motivation - Distribution

• Undesirable for all files to be in one location
• Access hidden files from anywhere
• Expanded storage capacity

5
Previous Work - Mnemosyne

• Peer-to-Peer Steganographic File System
• Positives
• Both clients and servers are untrusted members of
  system
• Locations of files are determined only by name
  and secret key
• Per-file granularity of steganography
• Block-level storage
• Negatives
• File persistence issues result from collisions
• Impossible to avoid collisions
• Any new file can overwrite older files

6
Previous Work - Mnemosyne
Server 1
Server 2
Server 3

• A file is mapped to deterministic locations
• As are its replicas
• A new file is introduced
• It has no knowledge of the old file
• Overwrites cannot be avoided

File (on client)
7
Previous Work - StegFS

• Local Steganographic File System
• Positives
• Maintains block allocation table (BAT) to avoid
  many collisions
• Offers linear levels of steganography
• Guarantees persistence of lowest level of
  steganography
• Negatives
• Not distributed
• Files stored in contiguous blocks (fragmentation)
• No deterministic location of files
• Lower-level files collide with higher-level files

8
Previous Work - StegFS
BAT
Disk
File

• Level 3 key reveals information about all files
  at lower levels
• Level 1 key reveals no information about files at
  higher levels

Level 1
Level 2
Level 3
9
Mojitos

• Combines desirable traits from both StegFS and
  Mnemosyne
• Distributed like Mnemosyne
• Block level storage like Mnemosyne
• Block allocation tables like StegFS
• Levels of steganography like StegFS
• File persistence better than Mnemosyne, and at
  least as good as StegFS

10
Mojitos
Server 1
Server 2
Server 3
Level 1
Level 2
Level 3
File (on client)
11
Mojitos - Architecture

• Servers
• Export block-level storage interface
• Perform security verification
• Autonomous and stateless (contain no knowledge of
  other servers or clients)
• Clients
• Make details of system transparent to user
• Contain list of active and authenticated servers
• Coordinate calls to servers
• Distribute or reassemble files and replicas

12
Mojitos Server Design

• Block allocation table (BAT) keeps track of usage
  for each block
• BAT entries are encrypted with a security-level
  key
• Avoids overwriting blocks of other files at or
  below this level

BAT Entry
Server Interface
13
Mojitos Client Design

• Provides read, write, and update abstractions to
  user
• Locates servers that should contain files inode
  (server set)
• Uses inode to collect or distribute the blocks
  replicas amongst all servers

14
Mojitos - Network and Security

• RSA public/private key encryption
• Servers require authentication to allow
  decryption keys to be passed in good faith
• Cover traffic hides patterns
• Key hierarchy can be used to contain security
  breaches

15
Mojitos - Reliability Scalability

• Design can be scaled to very large numbers of
  servers and clients
• No broadcasts necessary
• Replication serves dual purpose
• Reliability
• Persistence
• If number of servers grows or shrinks quickly,
  persistence problems are exaggerated

16
Mojitos - Simulation

• C
• Single, local machine (LPC, not RPC)
• Arrays instead of actual disks
• No encryption
• Simplified BAT entries (no checksums)

17
Mojitos Results (Persistence)

• 80 retrieval with full disk (4x)
• 99 retrieval with 20 disk use (4x)

18
Mojitos Results (Persistence)
(3x Replication)

• Persistence degrades with level disk use
• 100 retrieval at Level 1

19
Mojitos Results (Persistence)

• Persistence degrades with down servers
• 5x replication tolerates loss of 10 servers

20
Mojitos Results (Scalability)

• Redundant requests increase with security
• Maximum request/block ratio 1.15

21
Mojitos Results (Scalability)

• Even load distribution
• Similar for other security levels

22
Future Work Conclusion

• Direct comparison with StegFS and Mnemosyne would
  be beneficial
• Designing for persistence and security is complex
• Need better understanding of steganographic
  systems requirements (usage model)
• Mojitos has good persistence scalability

23
Questions?
?
?
?
?
?
?
?
?
?
?
?
?
?
?
?
?
?
?
?
?
?
?
?
?
?
?