1 / 39

Input, Output, and Automation in x86 Proved

Input, Output, and Automation in x86 Proved. Jason Gross Hosted by Andrew Kennedy Summer 2014. Verified I/O Programs. “. When doing formal verification, come up with the simplest non-trivial example you can. Then start with something simpler. —Adam Chlipala (paraphrased). ”.

chelsa
Download Presentation

Input, Output, and Automation in x86 Proved

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Input, Output, and Automation in x86 Proved Jason Gross Hosted by Andrew Kennedy Summer 2014

  2. Verified I/O Programs “ When doing formal verification, come up with the simplest non-trivial example you can. Then start with something simpler. —Adam Chlipala (paraphrased) ” Precondition Postcondition {true}{false} while true do skip done [] I/O behavior

  3. Verified I/O Programs: Trivial Loop {true}{false} while true do skip done [] Examplesafe_loop_whileeax: ⊢ basic (EAX ≅ eax⋆OSZCP?) (while (TESTEAX, EAX) CC_Ofalseprog_skip) [] false. Proof. basic apply (while_rule_ro(I := bb = false ∧ EAX? ⋆ SF? ⋆ ZF? ⋆ CF? ⋆ PF?)) //=; rewrite /stateIsAny; specintros; basic apply . Qed.

  4. Verified I/O Programs: Trivial Loop {true}{false} while true do skip done [] I/O Spec Postcondition Precondition Code Examplesafe_loop_whileeax: ⊢ basic (EAX ≅ eax⋆OSZCP?) (while (TESTEAX, EAX) CC_Ofalseprog_skip) [] false. Proof. basic apply (while_rule_ro(I := bb = false ∧ EAX? ⋆ SF? ⋆ ZF? ⋆ CF? ⋆ PF?)) //=; rewrite /stateIsAny; specintros; basic apply . Qed. Loop invariant

  5. Verified I/O Programs: Eternal Output {true}{false} while true doout 1 done 1* Kleene star Exampleloop_forever_onechannel: ⊢ basic ( AL?) ( MOV AL, 1;; LOCAL LOOP; LOOP:;; OUT channel, AL;; JMP LOOP) 1* false. Proof. (elided) Qed.

  6. Verified I/O Programs: Eternal Output {true}{false} while true doout 1 done 1* Exampleloop_forever_onechannel: ⊢ loopy_basic ( AL?) ( MOV AL,(#1: DWORD);; LOCAL LOOP; LOOP:;; OUT channel, AL;; JMP LOOP) (starOP(outOP(zeroExtend n8channel) (#1 : BYTE))) lfalse. Proof. (still elided) Qed.

  7. Verified I/O Programs: Echo Once v, {true} {true} v ← input; out v In v; Out v Examplesafe_echo_oncein_channelout_channel: ⊢ ∀ v, basic ( AL?) ( IN in_channel, AL;; OUT out_channel, AL) [Inv; Outv] (AL v). Proof. rewrite/stateIsAny;specintrosal v. basic apply . basic apply . Qed.

  8. Digression: Hoare Rule for While Standard rule from Wikipedia What’s ? Desired output annotations The I/O doesn’t get to talk about state!

  9. Digression: Hoare Rule for While The test of had value lets us “negate” , (which is code)

  10. Digression: Hoare Rule for While Example We can drop B, which is technical info about flags

  11. Digression: Hoare Rule for While Example

  12. Digression: Hoare Rule for While Example

  13. Digression: Hoare Rule for While Example with output Worry about side conditions later

  14. Digression: Hoare Rule for While Example with output Uh-oh! Worry about side conditions later

  15. Digression: Hoare Rule for While Eliding for space “transition function”

  16. Digression: Hoare Rule for While Can’t talk about state here

  17. Digression: Hoare Rule for While Logical part of

  18. Digression: Hoare Rule for While Example with output On-the-fly demo at the end, time and interest permitting

  19. Verified I/O Programs: Echo : stream, {true} {true} while true do (v ← input; out v) done map ( In ; Out ) Examplesafe_echoeaxin_channelout_channel: ⊢ ∀ vs, basic ( AL? ⋆ EAXeax⋆ OSZCP?) ( while (TESTEAX, EAX) CC_Ofalse ( IN in_channel, AL;; OUT out_channel, AL ) (stream_Opred_map ( [Inv; Outv]) ) lfalse. Proof. (elided; 8 lines of filling in arguments to the while rule, 9 lines of automation about specs) Qed.

  20. Verified I/O Programs: Accumulator : list (non-zero BYTE), {acc=x} {acc = fold accumulate x vs} while ((v ← input) 0) do (acc = accumulate acc v) done map ( In ) ExampleaddB_until_zero_prog_safech o s z c p S al : S(initial (x : BYTE) (xs: seq BYTE) (pf1 : only_last(t : BYTE⇒ t == #0) x xs), (loopy_basic(AH≅ initial ⋆ ALal ⋆ OSZCPo s z c p) ( IN ch, AL;; while (CMP AL,#0) CC_Zfalse (ADDAH, AL;; INch, AL)) (foldrcatOPempOP(map (inOP(zeroExtend8 ch)) (x::xs))) ((AH ≅ (foldladdBinitial (drop_lastx xs))) ⋆ AL ≅ #0 ⋆ OF? ⋆ SF? ⋆ ZF ≅ true ⋆ CF? ⋆ PF?))). Proof. specintros⇒ ∗. basic apply (@accumulate_until_zero_prog_safe_ (𝜆 x ⇒ AH≅ x)) ⇒ ∗; first assumption. basic apply ∗. Qed.

  21. Verified I/O Programs: Next Steps • readline (via accumulator template) • prime number printer • text adventure? • use memory-mapped I/O rather than IN and OUT

  22. Automation

  23. Instruction Automation: Ideal • Define the instruction in the model • State the high-level (e.g., Hoare) rule • Push-button verification Maybe even omit 2, if the inference is good enough. Image from https://flic.kr/p/3KYpfJ, “State of Mind”, tshein

  24. Instruction Automation: Reality • Define the instruction in the model DefinitionevalBinOp{n} op : BITSn→ BITSn→ ST (BITSn) := matchopwith | OP_XOR ⇒ evalLogicalOpxorB … end. Definition evalLogicalOp{n} (f : BITSn → BITSn → BITSn) arg1 arg2 := letresult := f arg1 arg2in do!updateFlagInProcStateCFfalse; do!updateFlagInProcStateOF false; do!updateZPSresult; retnresult.

  25. Instruction Automation: Reality • State the high-level (e.g., Hoare) rule LemmaXOR_RR_rules (r1 r2:VRegs) v1 (v2:VWORDs): basic (VRegIsr1 v1⋆ VRegIsr2 v2⋆ OSZCP?) (XORr1, r2) [] (VRegIsr1 (xorBv1 v2) ⋆ VRegIsr2 v2 ⋆ OSZCP false (msb(xorBv1 v2)) (xorBv1 v2 == #0) false (lsb(xorBv1 v2))).

  26. Instruction Automation: Reality • Push-button verification LemmaXOR_RR_rules (r1 r2:VRegs) v1 (v2:VWORDs): basic (VRegIsr1 v1⋆ VRegIsr2 v2⋆ OSZCP?) (XORr1, r2) [] (VRegIsr1 (xorBv1 v2) ⋆ VRegIsr2 v2 ⋆ OSZCP false (msb(xorBv1 v2)) (xorBv1 v2 == #0) false (lsb(xorBv1 v2))). Proof. destruct s; do_instrrule_triple. Qed.

  27. Instruction Automation: Old Reality • Push-button verification

  28. Instruction Automation: Old Reality

  29. Instruction Automation: Mechanics • Lookup • Application • Heuristics LemmaXOR_RR_rules (r1 r2:VRegs) v1 (v2:VWORDs): basic (VRegIsr1 v1⋆ VRegIsr2 v2⋆ OSZCP?) (XORr1, r2) [] (VRegIsr1 (xorBv1 v2) ⋆ VRegIsr2 v2 ⋆ OSZCP false (msb(xorBv1 v2)) (xorBv1 v2 == #0) false (lsb(xorBv1 v2))). Proof. destruct s; do_instrrule_triple. Qed.

  30. Instruction Automation: Mechanics Automated timing scripts helped ensure that the automation didn’t slow things down unreasonably. Example diff After | File Name | Before || Change ------------------------------------------------------------ 17m33.61s | Total | 18m51.54s || -1m17.92s ------------------------------------------------------------ 0m35.85s | examples/mulc | 0m42.18s || -0m06.32s 0m33.42s | examples/specexamples | 0m27.19s || +0m06.23s 0m24.29s | x86/lifeimp | 0m30.41s || -0m06.12s 0m17.60s | x86/inlinealloc | 0m21.79s || -0m04.18s 0m27.75s | x86/imp | 0m31.41s || -0m03.66s 0m17.56s | x86/instrrules/mov | 0m20.79s || -0m03.23s 0m15.76s | x86/call | 0m19.29s || -0m03.52s 0m51.82s | x86/instrrules/addsub | 0m54.54s || -0m02.71s 0m33.68s | x86/instrrules/cmp | 0m35.90s || -0m02.21s 0m27.69s | x86/cfuncspec | 0m29.79s || -0m02.09s 0m25.84s | x86/screenproof | 0m28.47s || -0m02.62s 0m19.90s | x86/instrcodec | 0m22.39s || -0m02.49s 0m26.98s | triple/set | 0m28.91s || -0m01.92s

  31. Program Automation: Ideal • Write a program • State the spec • Sprinkle annotations • Push-button verification Maybe even omit 3, if the inference is good enough. Image from https://flic.kr/p/3KYpfJ, “State of Mind”, tshein

  32. Program Automation: Reality specapply∗ takes care of essentially all of the unstructured code reasoning Examplesafe_echoeaxin_cout_c: ⊢ ∀ vs, basic ( AL? ⋆ EAXeax⋆ OSZCP?) ( while (TESTEAX, EAX) CC_Ofalse ( IN in_c, AL;; OUT out_c, AL ) (eq_opred_stream(stream_to_in_outin_cout_c)) lfalse. Proof. eapply@while_rule_ind with (I_logic := 𝜆 _ b⇒ false == b) (Otest := 𝜆 _ ⇒ empOP) (Obody := 𝜆 s ⇒ echo_once_OP_specin_cout_c(hds)) (I_state := 𝜆 s_ ⇒ EAXeax⋆ AL? ⋆ SF? ⋆ ZF? ⋆ CF? ⋆ PF?) (transition_body := @tl _) (O_after_test := 𝜆 s ⇒ default_PointedOPred (catOP(echo_once_OP_specin_cout_c(hds)) (eq_opred_stream(stream_to_in_outin_cout_c(tls))))); do ![ progress rewrite → ?empOPL, → ?eq_opred_stream__echo_once | specintros⇒ ∗ | done | byssimpl | basic apply ∗ | progress simplOPred_pred | progress move ⇒ ∗ | progress rewrite /stateIsAny | reflexivity ]. Qed. basic apply ∗ takes care of all of the code reasoning here

  33. Program Automation: Mechanics • Lookup (via typeclasses) • Application (via helper lemmas) • Heuristics (for side conditions)

  34. Open Issues

  35. Open Issues: Pointy Predicates DefinitionbasicP (c : T) (O : OPred) Q : spec := (i j : DWORD) (O': OPred), (obsO' @ (EIPjQ) obs(O + O') @ (EIPiP)) <@ (i -- jc). Definitionloopy_basicP (c : T) (O : OPred) Q : spec := (i j : DWORD) (O' : PointedOPred), (obsO' @ (EIP jQ) obs(O + O') @ (EIPiP)) <@ (i -- jc). Image from https://flic.kr/p/ab1bwS, “Spiky balls” Tom Magliery

  36. Open Issues: Quantifier Location Program Definition obs(O: OPred) := mkspec(k P (s: ProcState), (Pltrue) so, O orunsForWithPrefixOfk s o) _ _. Can we hide the quantifiers in the I/O predicate, so that the specification for echo doesn’t have to quantify over streams? Currently, both (* and (* say the wrong thing.

  37. Take Home Messages • We can verify the I/O behavior of simple assembly programs. • Putting effort into Coq automation results in tactics which are: • comparable in speed to manual proofs • capable of push-button verification in well-specified domains

  38. Acknowledgements • Thanks to: • Andrew Kennedy, my host here at MSR, and Nick Benton, who is also on this project. • Georges Gonthier, for help with Coq and SSReflect • Jonas Jensen, Jesper Bengtson, Gregory Malecha, and Edward Z. Yang for discussions about various aspects of I/O specifications, among other things.

  39. Thanks!Questions?Requests for verification demo?

More Related