1. Containment and Integrity for Mobile Code End-to-end security, untrusted hosts Andrew Myers Fred Schneider Department of Computer Science Cornell University Ithaca NY 14853

2. 18 July 00End-to-end security, untrusted hosts — Andrew Myers 2 Research directions End-to-end security by program rewriting In-lined reference monitors Asynchronous proactive secret sharing Gossip protocols Mobile code integrity: –NAP protocols (primary-backup revisited) –Cryptographic-based privilege management

3. 18 July 00End-to-end security, untrusted hosts — Andrew Myers 3 Protecting confidentiality Historically: privacy protection largely a military concern (confidentiality, secrecy) Future: many commercial, end-user needs Assurance for shared information services –on-line shopping, e-mail and home page services Programs with access to private information –spreadsheets, Quicken, word processors,... Military, commercial privacy needs converging? top secret secret classified unclassified

4. 18 July 00End-to-end security, untrusted hosts — Andrew Myers 4 Privacy vs complexity Problem: complex systems, untrusted parts –both distributed and single-host computation Harder to protect confidential information ?

5. 18 July 00End-to-end security, untrusted hosts — Andrew Myers 5 Example: airplane design Boeing Air Force marketing plans, aircraft designs other customers’ info military secrets, other suppliers’ info Hosts Data Programs war simulations cost projections CAD aircraft simulations

6. 18 July 00End-to-end security, untrusted hosts — Andrew Myers 6 Policies vs. Mechanisms Problem: policy/mechanism mismatch –Conventional mechanisms (e.g., access control): control whether A is allowed to transmit to B –Privacy policy: information I can only be obtained by users U (no matter how it is transformed) –Access control is point-to-point; policy is end-to-end How to map privacy policy onto a mechanism? (we already do this by hand!) AB ? I U

7. 18 July 00End-to-end security, untrusted hosts — Andrew Myers 7 Mechanisms Discretionary access control: doesn’t control propagation AB ?... Mandatory access control: expensive, restrictive AB ? L top secret secret classified unclassified L

8. 18 July 00End-to-end security, untrusted hosts — Andrew Myers 8 Static analysis of information flow Idea: add privacy policies as annotations to programs (types) : e.g., J IF language (Java Information Flow) int {L} x; // L is an end-to-end privacy policy J IF : security-typed language Uses decentralized label model

9. 18 July 00End-to-end security, untrusted hosts — Andrew Myers 9 Static information flow Type-check information flow statically –efficient –validates all possible run-time information flows: more precise, less restrictive –allows modular composition –hybrid dynamic/static schemes possible

10. 18 July 00End-to-end security, untrusted hosts — Andrew Myers 10 Compiler architecture Program Java source J IF compiler Label annotations Java compiler Class file (Bytecode) Source-to-source translator (J IF  Java) Mostly just removes annotations Class file (Bytecode) Label annotations

11. 18 July 00End-to-end security, untrusted hosts — Andrew Myers 11 Single-machine model Source J IF compiler Bytecode Trust Host Executin g program

12. 18 July 00End-to-end security, untrusted hosts — Andrew Myers 12 Airplane design Boeing Air Force marketing plans, aircraft designs other customers’ info military secrets, other suppliers’ info Hosts Data Programs War simulations Cost projections CAD aircraft simulations

13. 18 July 00End-to-end security, untrusted hosts — Andrew Myers 13 Avoiding trusted compiler Source J IF compiler Bytecode Trust Host Executin g program verifier Java trick: substitute trusted verifier for compiler Need expressive security type system for intermediate / assembly code Trust

14. 18 July 00End-to-end security, untrusted hosts — Andrew Myers 14 Avoiding trusted hosts Security invariant: host distrusted by principal p should not see p’s confidential data Problem: multi-party computation may involve confidential data from several parties Run only on completely trusted hosts? –expensive –bottleneck Computation across available hosts

15. 18 July 00End-to-end security, untrusted hosts — Andrew Myers 15 Secure program partitioning New approach to secure distributed systems Write programs without explicit code locations or inter-host communication Automatically transform code to run securely on current hosts source compiler intermediate code Host splitter Host code partition authenticated trust declarations

16. 18 July 00End-to-end security, untrusted hosts — Andrew Myers 16 Caveats Programs annotated with security information –but: annotations are types Communication model: inter-host messages cannot be intercepted, damaged –but: private-key encryption can be used Some covert channels (e.g., timing) still exist

17. 18 July 00End-to-end security, untrusted hosts — Andrew Myers 17 Status New, expressive intermediate language with support for security types, program transformations –Next: security-typed assembly language –verifier Rewrite rules for automatic program partitioning across hosts –Next: optimizing transformations for performance –partitioning back end for J IF compiler –partitioning verifier Core technology is in place

18. 18 July 00End-to-end security, untrusted hosts — Andrew Myers 18 Conclusions Decentralized enforcement of end-to-end security policies appears surprisingly feasible Application: assurance for distributed services Other project research directions: –In-lined reference monitors –Asynchronous proactive secret sharing –Gossip protocols –Mobile code integrity