Presentation is loading. Please wait.

Presentation is loading. Please wait.

Automated Tools for Software Reliability Suhabe Bugrara Stanford University.

Similar presentations


Presentation on theme: "Automated Tools for Software Reliability Suhabe Bugrara Stanford University."— Presentation transcript:

1 Automated Tools for Software Reliability Suhabe Bugrara suhabe@stanford.edu Stanford University

2 Problem 80% of development cost on identifying and correcting defects Software errors cost US economy $60 billion annually (0.6% of GDP)

3 Manual Testing Traditional approach to quality assurance Expensive Time consuming Not systematic Difficult to quantify effectiveness of test suite Cannot make any guarantees about reliability Insufficient for safety critical systems

4 Automated Tools Programs to find defects in programs Automated Systematic Easy to quantify effectiveness Provide guarantees about reliability Sometimes expensive (for now…) Sometimes time consuming (for now…)

5 Program Analyzers Sound Unsound CompleteIncomplete UndecidableDecidable Reports all errors May report false alarms Reports all errors Reports no false alarms May not report all errors May report false alarms Decidable May not report all errors Reports no false alarms

6 Static Driver Verifier Program analyzer for API usage rules Developed by Microsoft Research Applied to device drivers in Windows Sound: reports all possible errors Incomplete: may report false alarms

7 SDV: Overview 1.Write API usage rule specification 2.Instrument program with usage checks 3.Abstract program 4.Check abstraction for errors 5.If error found, see if error is false alarm 6.If false alarm, refine abstraction 7.If not false alarm, report error as bug

8

9

10 API Usage Rules Ex. locks are alternatingly acquired and released

11 API Usage Rules Ex. locks are alternatingly acquired and released Expressed as finite state machine –States = { locked, unlocked,error } –Transitions = { acquire(), release() }

12 API Usage Rules Ex. locks are alternatingly acquired and released Expressed as finite state machine –States = { locked, unlocked,error } –Transitions = { acquire(), release() } lockedunlocked error acquire(); release(); acquire();release();

13 state { enum { Unlocked=0; Locked=1} state = Unlocked; } KeAcquireSpinLock.return { if (state == Locked) error(); else state = Locked; } KeReleaseSpinLock.return { if (!(state == Locked)) error(); else state = Unlocked; }

14 enum {Unlocked=0, Locked=1} state = Unlocked void KeAcquireSpinLock_return() { if (state == Locked) error(); else state = Locked; } void KeReleaseSpinLock_return() { if (!(state == Locked)) error(); else state = Unlocked; }

15

16 1: void example() { 2: do { 3: KeAcquireSpinLock(); 4: 5:nPacketsOld = nPackets; 6:req = devExt->WLHV 7:if (req && req->status) { 8:devExt->WLHV = req->Next 9:KeReleaseSpinLock(); 10: 11:irp = req->irp; 12:if (req->status > 0) { 13:irp->IoS.Status = SUCCCESS; 14:irp->IoS.Info = req->Status; 15:} else { 16:irp->IoS.Status = FAIL; 17:irp->IoS.Info = req->Status; 18:} 19:SmartDevFreeBlock(req); 20:IoCompleteRequest(irp); 21:nPackets++; 22:} 23:} while (nPackets!=nPacketsOld); 24:KeReleaseSpinLock(); 25: 26: }

17

18 enum {Unlocked=0, Locked=1} state = Unlocked void KeAcquireSpinLock_return() { if (state == Locked) error(); else state = Locked; } void KeReleaseSpinLock_return() { if (!(state == Locked)) error(); else state = Unlocked; }

19 1: void example() { 2: do { 3: KeAcquireSpinLock(); 4:KeAcquireSpinLock_return(); 5:nPacketsOld = nPackets; 6:req = devExt->WLHV 7:if (req && req->status) { 8:devExt->WLHV = req->Next 9:KeReleaseSpinLock(); 10:KeReleaseSpinLock_return(); 11:irp = req->irp; 12:if (req->status > 0) { 13:irp->IoS.Status = SUCCCESS; 14:irp->IoS.Info = req->Status; 15:} else { 16:irp->IoS.Status = FAIL; 17:irp->IoS.Info = req->Status; 18:} 19:SmartDevFreeBlock(req); 20:IoCompleteRequest(irp); 21:nPackets++; 22:} 23:} while (nPackets!=nPacketsOld); 24:KeReleaseSpinLock(); 25:KeReleaseSpinLock_return(); 26: } Program A

20

21 SDV: Abstraction Construct abstraction B of original program A –Over-approximates reachability If error() is reachable in A, then it is also reachable in B –This characteristic makes SDV sound If error() is reachable in B, then it may not be reachable in A –This characteristic makes SDV incomplete Check abstraction B for any errors

22 Reachable States Original A error Abstraction B Sound: If A has error, then B has error real bug!

23 Reachable States Original A Abstraction B error Incomplete: If B has error, then A may not have error false alarm!

24 bool b1; b1 = false; Abstract state == Locked with b1 void KeAcquireSpinLock_return() { if (b1) error(); else b1 = true; } void KeReleaseSpinLock_return() { if (!(b1)) error(); else b1 = false; }

25 1: void example() { 2: do { 3: ; 4:KeAcquireSpinLock_return(); 5:; 6:; 7:if (SdvMakeChoice()) { 8:; 9:; 10:KeReleaseSpinLock_return(); 11:; 12:if (SdvMakeChoice()) { 13:; 14:; 15:} else { 16:; 17:; 18:} 19:; 20:; 21:; 22:} 23:} while (SdvMakeChoice()); 24:; 25:KeReleaseSpinLock_return(); 26: } Program B

26

27 1: void example() { 2: do { 3: ; 4:KeAcquireSpinLock_return(); 5:; 6:; 7:if (SdvMakeChoice()) { 8:; 9:; 10:KeReleaseSpinLock_return(); 11:; 12:if (SdvMakeChoice()) { 13:; 14:; 15:} else { 16:; 17:; 18:} 19:; 20:; 21:; 22:} 23:} while (SdvMakeChoice()); 24:; 25:KeReleaseSpinLock_return(); 26: } Error trace found!

28 1: void example() { 2: do { 3: KeAcquireSpinLock(); 4:KeAcquireSpinLock_return(); 5:nPacketsOld = nPackets; 6:req = devExt->WLHV 7:if (req && req->status) { 8:devExt->WLHV = req->Next 9:KeReleaseSpinLock(); 10:KeReleaseSpinLock_return(); 11:irp = req->irp; 12:if (req->status > 0) { 13:irp->IoS.Status = SUCCCESS; 14:irp->IoS.Info = req->Status; 15:} else { 16:irp->IoS.Status = FAIL; 17:irp->IoS.Info = req->Status; 18:} 19:SmartDevFreeBlock(req); 20:IoCompleteRequest(irp); 21:nPackets++; 22:} 23:} while (nPackets!=nPacketsOld); 24:KeReleaseSpinLock(); 25:KeReleaseSpinLock_return(); 26: } But, no bug in original program!

29

30 1: void example() { 2: do { 3: ; 4:KeAcquireSpinLock_return(); 5:b2 = false; 6:; 7:if (SdvMakeChoice()) { 8:; 9:; 10:KeReleaseSpinLock_return(); 11:; 12:if (SdvMakeChoice()) { 13:; 14:; 15:} else { 16:; 17:; 18:} 19:; 20:; 21:b2 = !b2 ? true : SdvMakeChoice(); 22:} 23:} while (b2); 24:; 25:KeReleaseSpinLock_return(); 26: } Program C

31 Reachable States Original A Abstraction B error Refined C false alarm no longer reported!

32 SDV: Summary 1.Write API usage rule specification 2.Instrument program with usage checks 3.Abstract program 4.Check abstraction for errors 5.If error found, see if error is false alarm 6.If false alarm, refine abstraction 7.If not false alarm, report error as bug

33 Soundness Assume memory safety –No buffer/integer overflows –Safe memory management –No null pointer dereferences Oversimplified harness –Use stubs to model calls into OS procedures –Stubs may not represent all behavior

34 Research Challenges in Verification Eliminate assumption of memory safety Eliminate false alarms Scale to the entire operating system Verify more complicated properties –prove consistency of file system data structures

35 Program Analyzers Sound Unsound CompleteIncomplete UndecidableDecidable Reports all errors May report false alarms Reports all errors Reports no false alarms May not report all errors May report false alarms Decidable May not report all errors Reports no false alarms

36 EXE Automatically generate test cases that explore important program paths Developed by Dawson Engler’s group Bug finding tool Unsound: may not report all errors Complete: never reports false alarms

37 int bad_abs (int x) { if (x < 0) return –x; if (x == 12345678) return –x; return x; }

38 int bad_abs (int x) { if (x < 0) return –x; if (x == 12345678) return –x; return x; }

39 int bad_abs (int x) { if (x < 0) return –x; if (x == 12345678) return –x; return x; } (x >= INT_MIN) && (x <= INT_MAX) && (x < 0) && (ret = -x) find a solution using an automatic constraint solver… x = -1

40 int bad_abs (int x) { if (x < 0) return –x; if (x == 12345678) return –x; return x; } (x >= INT_MIN) && (x = 0) && (x = 12345678) && (ret = -x) find a solution using an automatic constraint solver… x = 12345678

41 int bad_abs (int x) { if (x < 0) return –x; if (x == 12345678) return –x; return x; } (x >= INT_MIN) && (x = 0) && (x != 12345678) && (ret = x) find a solution using an automatic constraint solver… x = 4

42 int bad_abs (int x) { if (x < 0) return –x; if (x == 12345678) return –x; return x; } EXE automatically generated test cases for each path… x = -1 x = 12345678 x = 4

43 int bad_abs (int x) { if (x < 0) return –x; if (x == 12345678) return –x; return x; }

44 1: int symbolic_bad_abs (int x) { 2:add_constraints(x >= INT_MIN, x <= INT_MAX); 3: ret = new symbol; 4: 5:if (fork() == child) { 6:add_constraints(x < 0, ret = -x); 7:return ret; 8: //(x >= INT_MIN) && (x <= INT_MAX) && (x < 0) && (ret = -x) 9:} else 10:add_constraints(x >= 0); 11: 12:if (fork() == child) { 13:add_constraints(x = 12345678, ret = -x); 14:return ret; 15: //(x >= INT_MIN) && (x = 0) && (x = 12345678) 16: //&& (ret = -x) 17:} else 18:add_constraints(x != 12345678); 19: 20:add_constraints(ret = x); 21:return ret; 22: //(x >= INT_MIN) && (x = 0) && (x != 12345678) 23: && (ret = x) 24:}

45 1: int main (void) { 2:unsigned i, t, a[4] = { 1, 3, 5, 2}; 3:make_symbolic(&i); 4: 5:if (i >= 4) 6:exit(0); 7: 8:char *p = (char *) a + i * 4; 9:*p = *p – 1; 10: 11:t = a[*p]; 12: 13:t = t / a[i]; 14: 15:if (t == 2) 16:assert(i == 1); 17:else 18:assert(i == 3); 19: }

46

47 Review Why does SDV produce false alarms and EXE doesn’t? Why use SDV, then?

48 Saturn Large-scale program verification Developed by Alex Aiken’s group Sound: reports all errors Incomplete: may report false alarms Gives guarantees of reliability on systems as large as the Linux kernel with over 6.2 million lines of code

49 Program Analyzers Sound Unsound CompleteIncomplete UndecidableDecidable Reports all errors May report false alarms Reports all errors Reports no false alarms May not report all errors May report false alarms Decidable May not report all errors Reports no false alarms

50 Unchecked User Pointer Dereferences Security property of operating systems Two types of pointers in operating systems –kernel pointer: pointer created by the operating system –user pointer: pointer created by a user application and passed to the operating system via an entry point such as a system call Must check that a user pointer points into userspace before dereferencing it

51 Unchecked User Pointer Dereferences 1: static ssize_t read_port(…, char * __user buf, …) { 2:unsigned long i = *ppos; 3:char * __user tmp = buf; 4: 5:if (!access_ok(..,buf,...)) //check 6:return -EFAULT; 7: 8: while (count-- > 0 && i < 65536) { 9: if (__put_user(inb(i),tmp) < 0) //deref 10:return -EFAULT; 11:i++; 12:tmp++; 13:} 14: 15:*ppos = i; 16:return tmp-buf; 17: }

52 Security Vulnerability Malicious user could –Take control of the operating system –Overwrite kernel data structures –Read sensitive data out of kernel memory –Crash machine by corrupting data

53 Verifying the Security Property Eliminate the need for annotations Eliminate false positives Provide guarantee that no security vulnerabilities of this kind are present

54 Security Verifier Design a sound and incomplete verifier to prove statically that no unchecked user pointer dereferences exist

55 Security Verifier Compute set of facts at each program point States = { user, checked, error } Facts are pairs of locations and states –( *v,user) signifies that v is a user pointer Verify that program never in error state

56 Security Verifier Pointer is in user state if created by user application Pointer is in checked state if access_ok applied Pointer is in error state if dereferenced when 1.Pointer is in user state, AND 2.Pointer is NOT in checked state

57 Example 1: int sys_call (int *u, int cmd) {//u is user pointer 2: int x; 3: 4:if (cmd == 1) { 5:if (!access_ok(u)) { //check u 6:return –ERR; 7:} 8:} 9:… 10:if (cmd == 1) 11:x = *u;//dereference u 12:…

58 One Possible Approach …, but, procedure does not contain any vulnerabilities! (*u,user) (*u,error) emit warning! (*u,user) lost precision! (*u,user) (*u,checked) 1: int sys_call (int *u, int cmd) { 2: int x; 3: 4:if (cmd == 1) { 5:if (!access_ok(u)) { 6:return –ERR; 7:} 8:} 9:… 10:if (cmd == 1) 11:x = *u; 12:… (*u,user)

59 Path Sensitivity Ability to reason about branch correlations Important for reducing false positive rate Programs use substantial amount of branch correlation in practice

60 Example 1: int sys_call (int *u, int cmd) {//u is user pointer 2: int x; 3: 4:if (cmd == 1) { 5:if (!access_ok(u)) { //check u 6:return –ERR; 7:} 8:} 9:… 10:if (cmd == 1) 11:x = *u;//dereference u 12:…

61 Path Sensitivity Valid Path 1: int sys_call (int *u, int cmd) {//u is user pointer 2: int x; 3: 4:if (cmd == 1) { 5:if (!access_ok(u)) { //check u 6:return –ERR; 7:} 8:} 9:… 10:if (cmd == 1) 11:x = *u;//dereference u 12:…

62 Path Sensitivity Valid Path 1: int sys_call (int *u, int cmd) {//u is user pointer 2: int x; 3: 4:if (cmd == 1) { 5:if (!access_ok(u)) { //check u 6:return –ERR; 7:} 8:} 9:… 10:if (cmd == 1) 11:x = *u;//dereference u 12:…

63 Path Sensitivity Valid Path 1: int sys_call (int *u, int cmd) {//u is user pointer 2: int x; 3: 4:if (cmd == 1) { 5:if (!access_ok(u)) { //check u 6:return –ERR; 7:} 8:} 9:… 10:if (cmd == 1) 11:x = *u;//dereference u 12:…

64 Path Sensitivity Invalid Path! 1: int sys_call (int *u, int cmd) {//u is user pointer 2: int x; 3: 4:if (cmd == 1) { 5:if (!access_ok(u)) { //check u 6:return –ERR; 7:} 8:} 9:… 10:if (cmd == 1) 11:x = *u;//dereference u 12:…

65 Path Sensitive Analysis (*u,user)  true (*u,checked)  cmd == 1 (*u,error)  cmd == 1 && !(cmd == 1) && true  false (*u,user)  true (*u,checked)  cmd == 1 (*u,user)  true (*u,checked)  cmd == 1 (*u,user)  true 1: int sys_call (int *u, int cmd) { 2: int x; 3: 4:if (cmd == 1) { 5:if (!access_ok(u)) { 6:return –ERR; 7:} 8:} 9:… 10:if (cmd == 1) 11:x = *u; 12:… (*u,user)  true

66 Design of Saturn Security Verifier Generate summary of behavior for each procedure with respect to calling context Apply summary of callee at call site in caller Repeatedly generate and apply summaries until a fixed point is reached

67 Experimental Setup Implemented verifier for unchecked user pointer dereferences Applied verifier to Linux 2.6.17.1 built for x86 architecture 6.4 million lines of code Analyzed in 6 hours over 50 node cluster

68 Results 91,543 procedures 154 (.17%) of procedures time out 627 system call parameters 867,544 dereferences 15,452 (1.8%) of dereferences time out

69 Results Verified automatically –620 out of 627 system call arguments (99%) –851,914 out of 852,092 dereferences (99.96%) Warnings –7 warnings on system call arguments –278 warnings on dereferences –20 annotations required to verify

70 Saturn: Other Analyses Null pointer dereferences bug finder –Found hundreds of bugs in systems code –Isil Dillig, Thomas Dillig, and Alex Aiken. Static Error Detection Using Semantic Inconsistency Inference, PLDI 2007 Buffer overflow Safe casting Integer overflow Locking Safe memory management

71 Other Tools BLAST CQual Metal Daikon Vault ESP ESPX MOPS DART Prefast Failure Oblivious Computing CSSV Alloy eXplode Chord TVLA CCured Clouseau STeP Prefix

72 References A. Aiken et al. An Overview of the Saturn Project. PASTE 2007 T. Ball et al. Thorough Static Analysis of Device Drivers. EuroSys 2006 C. Cadar et al. EXE: Automatically Generating Inputs of Death. CCS 2006 C. Cadar et al. Execution Generated Test Cases: How to Make Systems Code Crash Itself. SPIN 2006 B. Hackett et al. Modular Checking for Buffer Overflows in the Large. ICSE 2006. J. Yang et al. Automatically generating malicious disks using symbolic execution. IEEE Security and Privacy 2006 Software Errors Cost U.S. Economy $59.5 Billion Annually. NIST 2002. http://www.nist.gov/public_affairs/releases/n02-10.htm


Download ppt "Automated Tools for Software Reliability Suhabe Bugrara Stanford University."

Similar presentations


Ads by Google