A Translation from Typed Assembly Language to Certified Assembly Programming Zhong Shao Yale University Joint work with Zhaozhong Ni Paper URL: .

A Translation from Typed Assembly Language to Certified Assembly

Programming

Zhong Shao Yale University

Joint work with Zhaozhong NiPaper URL: http://flint.cs.yale.edu/flint/publications/talcap.html

August 11, 2006

Research objective (of the FLINT group)

To build a certified software platform with real guarantee of reliability & security !

Hardware

certified L1 software

legacy SW layer1

legacy SW layer2

legacy SW layer3

legacy SW layer4

certified L2 SW

certified L3 SW

certified L4 SW

certified L5 SW

The lowest SW layer is the key!

A buggy L1 software can take over the machine easily!

Hardware

buggy L1 software (or VM)

legacy SW layer1

legacy SW layer2

legacy SW layer3

legacy SW layer4

certified L2 SW

certified L3 SW

certified L4 SW

certified L5 SW

infected L2 SW

infected L3 SW

infected L4 SW

infected L5 SW

Must be Trusted!

Structure of our certified framework

• certified code (proof + machine code)

• machine model• safety policy• mechanized meta-logic• proof checker

Proof Checker Yes

CPU

Safety Policy

Proof

machine code

No

What makes a good mechanized meta logic?

You’d better be very paranoid!

• The logic must be “rock-solid”, i.e., consistent!

• The logic must be expressive to express everything a hacker wants to say

• Support explicit, machine-checkable proof objects

• The logic must be simple so that the proof checker can be hand-verified

• Can serve as logical framework and meta-logical framework to allow one to prove using specialized logics

• Compatible with “automated proof construction”

How to scale?

Modularity, modularity, modularity!

specification S1

binary code C1

formal proof P1

specification S2

binary code C2

formal proof P2

specification S3

binary code C3

formal proof P3

specification S4

binary code C4

formal proof P4

specification S6

binary code C6

formal proof P6

specification S5

binary code C5

formal proof P5

specification S

binary code C

formal proof P

Linking

Another form of modularity

Software is often organized as a vertical stack of abstractions!

Not everything is certified at the assembly level!

Hardware

certified L1 software

certified L2 SW

certified L3 SW

certified L4 SW

certified L5 SW

Must accurately specify & certify all these interfaces!

A really “juicy” research area …Many interesting & exciting problems:

• How to certify each standard language and OS abstraction?– general code pointers– procedure call/return – general stack-based control abstraction– mutable data structures (& malloc/free …)– self-modifying code (& OS boot loader …)– interrupt/signal handling– device drivers and IO managers– thread libraries and synchronization – multiprocessor and memory model– OS kernel/user abstraction – …………

• How to combine proof assistant with general-purpose programming?

• Other exciting interplays btw machine-checked proofs & computation

Related research projects at Yale

Certifying different language & OS abstractions:

– certified assembly programming [CAP ESOP’02]– embedded code pointers [XCAP POPL’06]– non-preemptive threads [CCAP ICFP’04 & CMAP ICFP’05]– stack-based control abstractions [SCAP PLDI’06]– self-modifying code & local reasoning [Cai et al GCAP on-going]– thread libraries and synchronizations [Ni et al on-going]– interrupts & multiprocessors [Ferreira et al on-going]– open framework for interoperability [Feng et al OCAP on-going]– boot-loaders & preemptive threads [Feng et al on-going]– memory management using malloc/free [CAP ESOP’02]– garbage collector & mutator [McCreight et al on-going]

Features of a CAP-style system

• All built on a mechanized meta logic (e.g., Coq)

• Both the machine-level program and the property are specified by formulas in the meta logic

• Like TLA except our meta logic is mechanized

• Hoare-style assertions & inference rules enforce both the correctness & type safety properties

• No need of a separate type system; not a “refinement”

• Assertion languages can vary:• Borrow those from Coq (shallow embedding) --- CAP• Hybrid: Coq assertions + a thin layer of syntax --- XCAP

TAL vs. CAP

• Type-based Approach– TAL [Morrisett98]

– Touchstone PCC [Colby00]

– Syntactic FPCC [Hamid02]

– FTAL [Crary03]

– LTAL [Chen03]

– …

– Modular– Generate proof easily– Type safety

• Logic-based Approach– Original PCC [Necula98]

– CAP [Yu03]

– CCAP/CMAP [Yu04, Feng05] – XCAP [Ni & Shao 06]

– SCAP [Feng et al 06]

– …

– Expressive– Advanced properties– Good interoperability

This talk! We also show how to embed TAL into our new XCAP!

SCAP [Feng et al PLDI’06]

XCAP [Ni & Shao POPL’06]

Can we have the best of both worlds?

• Can a Hoare-style CAP system support:– embedded code pointers?– closures?– exceptions?– runtime stacks?– general references w. weak update?– recursive data structures?

Syntax of target machine TM

Operational semantics of TM

Mechanized meta logic

All implemented in the Coq proof assistant!

XCAP assertion language

Validity rules for PropX [P] = ¢ ` P

Soundness of interpretation

XCAP inference rules: top level

XCAP inference rules: instructions

Soundness of XCAP

Memory mutation in XCAP?

• Strong update!– special conjunction (p * q) in separation logic

– directly definable in Prop and PropX– explicit alias control, popular in system level

• Weak update (general reference)?– mutable reference (int ref) in ML– managed data pointers (int __gc*) in .NET

– rely on GC to recycle memory

XCAP extension: general reference

XCAP ext.: supporting weak update

Recursive specification in XCAP?

• Simple recursive data structures!– linked list, queue, stack, tree, etc.

– supported via inductive definition of Prop

• Complex recursive structures with ECP?– object (self refers to the entire object)

– threading invariant (each thread assumes others)

XCAP ext.: recursive predicates

TAL: type definitions

TAL typing rules: top level

TAL typing rules: instructions

TAL typing rules: instructions (cont’d)

TAL uses “syntactic” subtyping!

Well-formed state in TAL

TAL-to-XCAP translation

Step #1: we build a “logic-based” TAL that uses semantic subtyping

Step #2: translating regular TAL into “logic-based” TAL

(this is fairly straight-forward!)

Step #3: translating “logic-based” TAL into XCAP

TAL-to-XCAP: translating value types

• translation of preconditions

• translation of code heap types

• translation of data heap types

TAL-to-XCAP: other translations

TAL-to-XCAP: typing preservation

Conclusion

user application

libraryTAL

device driver

OS kernelfirmwareXCAP

XCAP can be extended to support general reference, weak update, and recursive specification !

We give a direct embedding of TAL into XCAP.