1. Secure Virtual Architecture John Criswell, Arushi Aggarwal, Andrew Lenharth, Dinakar Dhurjati, and Vikram Adve University of Illinois at Urbana-Champaign 1 Secure Virtual Architecture

2. Outline Background Current Work Future Work 2 Secure Virtual Architecture

3. TRANSFORMATION HARDWARE SYSTEM ARCHITECTURES SVA Binary translation and emulation Formal methods Hardware support for isolation Dealing with malicious hardware Cryptographic secure computation Data-centric security Secure browser appliance Secure servers WEB-BASED ARCHITECTURES e.g., Enforce properties on a malicious OS e.g., Prevent data exfiltration e.g., Enable complex distributed systems, with resilience to hostile OS’s Secure Virtual Architecture 3

4. Wouldn’t It Be Great? Enforce information flow policy Confidentiality Data-centric policy created by application/user Malicious OS can examine/modify any data in memory Need to control OS memory operations Keep system running when a safety violation is detected 4 Secure Virtual Architecture Process 1 Process 1 Process 2 Process 2 Operating System Memory

5. Secure Virtual Architecture Compiler-based virtual machine Uses sophisticated compiler analysis & transformation techniques Virtual instruction set Typed virtual instruction set enables sophisticated program analysis Special instructions for OS kernel support Provide safe execution environment for commodity software Supports unmodified C/C++ applications Supports commodity operating systems (e.g., Linux) 5 Commodity Applications + OS Hardware Compiler + VM Virtual ISA Native ISA Secure Virtual Architecture

6. SVA Safety Guarantees Safe LanguageSecure Virtual Architecture Control flow integrity Array indexing within bounds No uses of uninitialized variables Type safety for all objectsType safety for subset of objects No uses of dangling pointersDangling pointers are harmless Sound operational semantics Dangling pointers & non-type-safe objects do not compromise other guarantees Strongest memory safety for C sans garbage collection 6 Secure Virtual Architecture

7. What’s the Secret Sauce? Run-time Checks Load/Store Checks Bounds Checks Illegal Free Checks Indirect Call Checks Static Analysis Type Inference Points-to Analysis 7 Secure Virtual Architecture

8. Outline Background Current Work Future Work 8 Secure Virtual Architecture

9. Safe Software/Hardware Interaction OperationProblemSolution Context SwitchingKernel can load bad state on to CPU Store CPU state in SVA VM memory Stack ManagementKernel stacks are regular, mutable memory objects SVA creates new type of memory object for kernel stacks; pointers to such objects cannot be dereferenced MMU ConfigurationStatic analysis assumes virtual address space is immutable Use para-virtualization to prevent MMU configurations that violate static analysis safety guarantees 9 Secure Virtual Architecture

10. A Secure Foundation Strong memory safety enforcement Even for low level OS code! Can rely on static analysis results to hold at run-time Enforces safety properties on applications and OS kernel code 10 Secure Virtual Architecture Safety enforced despite hostile OS Code!

11. Current Work Information Flow for C Improved Type Inference Recovery from Safety Violations 11 Secure Virtual Architecture

12. CIF: C Information Flow Compiler Experimental information flow infrastructure for C/C++ Explicit information flow on memory object granularity Properly joins (meets) labels for computation results Based on SVA Memory safety errors cannot violate safety guarantees Can reuse SVA infrastructure for optimization 12 Data Memory Object Memory Object Process Meet Secure Virtual Architecture

13. SVA Controls Information Flow SVA controls Memory access MMU configuration Information Flow Uniform monitoring SVA enforces policies Not the OS 13 Process 1 Process 1 Process 2 Process 2 SVA Virtual Machine Operating System Memory Secure Virtual Architecture

14. Improving Type Safety in SVA Benefits Better pointer disambiguation due to improved field sensitivity Better safety More static type safety yields more precise run-time safety guarantees Better performance Type-safe objects do not need load/store checks 14 Secure Virtual Architecture

15. Type Safety Enhancements Tracking types at byte-offsets Permit a subset of a memory object to be type safe Supports C++ class hierarchy sub-typing Identifying C library functions and allocator wrappers Static code transformations to improve inference results Cloning of address-taken functions for use in direct calls Clone functions that take embedded structures from incompatible types 15 Secure Virtual Architecture

16. Static Type Safety SPEC 2000 Secure Virtual Architecture 16

17. Static Type Safety SPEC 2006 Secure Virtual Architecture 17

18. Outline Background Current Work Future Work 18 Secure Virtual Architecture

19. Dynamic Type Tracking in SVA Track types stored to memory at run-time Used for memory operations that cannot be proven safe statically Byte granularity tracking Fine grained tracking of fields in structures Check type of data when loading from memory 19 Secure Virtual Architecture

20. Conclusions SVA provides a secure foundation We have: Infrastructure for secure information flow Improved type inference Automated recovery from run-time safety violations In the pipeline: Secure information flow to enforce safety sans OS support Dynamic type tracking 20 Secure Virtual Architecture