blob: f4de733bd7944179be7762274382d63f07bacc1b [file] [log] [blame]
Ted Kremeneke73571b2010-12-23 02:42:43 +00001//== ArrayBoundCheckerV2.cpp ------------------------------------*- C++ -*--==//
2//
3// The LLVM Compiler Infrastructure
4//
5// This file is distributed under the University of Illinois Open Source
6// License. See LICENSE.TXT for details.
7//
8//===----------------------------------------------------------------------===//
9//
10// This file defines ArrayBoundCheckerV2, which is a path-sensitive check
11// which looks for an out-of-bound array element access.
12//
13//===----------------------------------------------------------------------===//
14
Argyrios Kyrtzidis0a9ce3e2011-02-28 01:26:57 +000015#include "ClangSACheckers.h"
Chandler Carruth3a022472012-12-04 09:13:33 +000016#include "clang/AST/CharUnits.h"
17#include "clang/StaticAnalyzer/Core/BugReporter/BugType.h"
Argyrios Kyrtzidis6a5674f2011-03-01 01:16:21 +000018#include "clang/StaticAnalyzer/Core/Checker.h"
Argyrios Kyrtzidis0a9ce3e2011-02-28 01:26:57 +000019#include "clang/StaticAnalyzer/Core/CheckerManager.h"
20#include "clang/StaticAnalyzer/Core/PathSensitive/CheckerContext.h"
Ted Kremenekf8cbac42011-02-10 01:03:03 +000021#include "clang/StaticAnalyzer/Core/PathSensitive/ExprEngine.h"
Benjamin Kramer49038022012-02-04 13:45:25 +000022#include "llvm/ADT/SmallString.h"
Benjamin Kramer444a1302012-12-01 17:12:56 +000023#include "llvm/Support/raw_ostream.h"
Ted Kremeneke73571b2010-12-23 02:42:43 +000024
25using namespace clang;
Ted Kremenek98857c92010-12-23 07:20:52 +000026using namespace ento;
Ted Kremeneke73571b2010-12-23 02:42:43 +000027
28namespace {
Ted Kremenek3a0678e2015-09-08 03:50:52 +000029class ArrayBoundCheckerV2 :
Argyrios Kyrtzidis6a5674f2011-03-01 01:16:21 +000030 public Checker<check::Location> {
Ahmed Charlesb8984322014-03-07 20:03:18 +000031 mutable std::unique_ptr<BuiltinBug> BT;
32
Anna Zaksb7eac9f2012-01-21 05:07:33 +000033 enum OOB_Kind { OOB_Precedes, OOB_Excedes, OOB_Tainted };
Ted Kremenek3a0678e2015-09-08 03:50:52 +000034
Ted Kremenek49b1e382012-01-26 21:29:00 +000035 void reportOOB(CheckerContext &C, ProgramStateRef errorState,
Argyrios Kyrtzidis0a9ce3e2011-02-28 01:26:57 +000036 OOB_Kind kind) const;
Ted Kremenek3a0678e2015-09-08 03:50:52 +000037
Ted Kremeneke73571b2010-12-23 02:42:43 +000038public:
Anna Zaks3e0f4152011-10-06 00:43:15 +000039 void checkLocation(SVal l, bool isLoad, const Stmt*S,
40 CheckerContext &C) const;
Ted Kremeneke73571b2010-12-23 02:42:43 +000041};
42
43// FIXME: Eventually replace RegionRawOffset with this class.
44class RegionRawOffsetV2 {
45private:
46 const SubRegion *baseRegion;
47 SVal byteOffset;
Craig Topper0dbb7832014-05-27 02:45:47 +000048
Ted Kremeneke73571b2010-12-23 02:42:43 +000049 RegionRawOffsetV2()
Craig Topper0dbb7832014-05-27 02:45:47 +000050 : baseRegion(nullptr), byteOffset(UnknownVal()) {}
Ted Kremeneke73571b2010-12-23 02:42:43 +000051
52public:
53 RegionRawOffsetV2(const SubRegion* base, SVal offset)
54 : baseRegion(base), byteOffset(offset) {}
55
David Blaikie2fdacbc2013-02-20 05:52:05 +000056 NonLoc getByteOffset() const { return byteOffset.castAs<NonLoc>(); }
Ted Kremeneke73571b2010-12-23 02:42:43 +000057 const SubRegion *getRegion() const { return baseRegion; }
Ted Kremenek3a0678e2015-09-08 03:50:52 +000058
Ted Kremenek49b1e382012-01-26 21:29:00 +000059 static RegionRawOffsetV2 computeOffset(ProgramStateRef state,
Ted Kremeneke73571b2010-12-23 02:42:43 +000060 SValBuilder &svalBuilder,
61 SVal location);
62
63 void dump() const;
Ted Kremenek5ef32db2011-08-12 23:37:29 +000064 void dumpToStream(raw_ostream &os) const;
Ted Kremeneke73571b2010-12-23 02:42:43 +000065};
Alexander Kornienkoab9db512015-06-22 23:07:51 +000066}
Ted Kremeneke73571b2010-12-23 02:42:43 +000067
Ted Kremenek3a0678e2015-09-08 03:50:52 +000068static SVal computeExtentBegin(SValBuilder &svalBuilder,
Ted Kremenek8a4c7602011-04-12 17:21:33 +000069 const MemRegion *region) {
70 while (true)
71 switch (region->getKind()) {
72 default:
Ted Kremenek3a0678e2015-09-08 03:50:52 +000073 return svalBuilder.makeZeroArrayIndex();
Ted Kremenek8a4c7602011-04-12 17:21:33 +000074 case MemRegion::SymbolicRegionKind:
75 // FIXME: improve this later by tracking symbolic lower bounds
76 // for symbolic regions.
77 return UnknownVal();
78 case MemRegion::ElementRegionKind:
79 region = cast<SubRegion>(region)->getSuperRegion();
80 continue;
81 }
82}
83
Argyrios Kyrtzidis0a9ce3e2011-02-28 01:26:57 +000084void ArrayBoundCheckerV2::checkLocation(SVal location, bool isLoad,
Anna Zaks3e0f4152011-10-06 00:43:15 +000085 const Stmt* LoadS,
Argyrios Kyrtzidis0a9ce3e2011-02-28 01:26:57 +000086 CheckerContext &checkerContext) const {
Ted Kremeneke73571b2010-12-23 02:42:43 +000087
Ted Kremenek001fd5b2011-08-15 22:09:50 +000088 // NOTE: Instead of using ProgramState::assumeInBound(), we are prototyping
Ted Kremeneke73571b2010-12-23 02:42:43 +000089 // some new logic here that reasons directly about memory region extents.
90 // Once that logic is more mature, we can bring it back to assumeInBound()
91 // for all clients to use.
92 //
93 // The algorithm we are using here for bounds checking is to see if the
94 // memory access is within the extent of the base region. Since we
95 // have some flexibility in defining the base region, we can achieve
96 // various levels of conservatism in our buffer overflow checking.
Ted Kremenek3a0678e2015-09-08 03:50:52 +000097 ProgramStateRef state = checkerContext.getState();
Ted Kremenek49b1e382012-01-26 21:29:00 +000098 ProgramStateRef originalState = state;
Ted Kremeneke73571b2010-12-23 02:42:43 +000099
100 SValBuilder &svalBuilder = checkerContext.getSValBuilder();
Ted Kremenek3a0678e2015-09-08 03:50:52 +0000101 const RegionRawOffsetV2 &rawOffset =
Ted Kremeneke73571b2010-12-23 02:42:43 +0000102 RegionRawOffsetV2::computeOffset(state, svalBuilder, location);
103
104 if (!rawOffset.getRegion())
105 return;
106
Ted Kremenek3a0678e2015-09-08 03:50:52 +0000107 // CHECK LOWER BOUND: Is byteOffset < extent begin?
Ted Kremenek8a4c7602011-04-12 17:21:33 +0000108 // If so, we are doing a load/store
Ted Kremeneke73571b2010-12-23 02:42:43 +0000109 // before the first valid offset in the memory region.
110
Ted Kremenek8a4c7602011-04-12 17:21:33 +0000111 SVal extentBegin = computeExtentBegin(svalBuilder, rawOffset.getRegion());
Ted Kremenek3a0678e2015-09-08 03:50:52 +0000112
David Blaikie05785d12013-02-20 22:23:23 +0000113 if (Optional<NonLoc> NV = extentBegin.getAs<NonLoc>()) {
David Blaikie2fdacbc2013-02-20 05:52:05 +0000114 SVal lowerBound =
115 svalBuilder.evalBinOpNN(state, BO_LT, rawOffset.getByteOffset(), *NV,
Ted Kremenek8a4c7602011-04-12 17:21:33 +0000116 svalBuilder.getConditionType());
Ted Kremeneke73571b2010-12-23 02:42:43 +0000117
David Blaikie05785d12013-02-20 22:23:23 +0000118 Optional<NonLoc> lowerBoundToCheck = lowerBound.getAs<NonLoc>();
Ted Kremenek8a4c7602011-04-12 17:21:33 +0000119 if (!lowerBoundToCheck)
120 return;
Ted Kremenek3a0678e2015-09-08 03:50:52 +0000121
Ted Kremenek49b1e382012-01-26 21:29:00 +0000122 ProgramStateRef state_precedesLowerBound, state_withinLowerBound;
Benjamin Kramer867ea1d2014-03-02 13:01:17 +0000123 std::tie(state_precedesLowerBound, state_withinLowerBound) =
Ted Kremeneke73571b2010-12-23 02:42:43 +0000124 state->assume(*lowerBoundToCheck);
125
Ted Kremenek8a4c7602011-04-12 17:21:33 +0000126 // Are we constrained enough to definitely precede the lower bound?
127 if (state_precedesLowerBound && !state_withinLowerBound) {
128 reportOOB(checkerContext, state_precedesLowerBound, OOB_Precedes);
129 return;
130 }
Ted Kremenek3a0678e2015-09-08 03:50:52 +0000131
Ted Kremenek8a4c7602011-04-12 17:21:33 +0000132 // Otherwise, assume the constraint of the lower bound.
133 assert(state_withinLowerBound);
134 state = state_withinLowerBound;
135 }
Ted Kremenek3a0678e2015-09-08 03:50:52 +0000136
Ted Kremeneke73571b2010-12-23 02:42:43 +0000137 do {
138 // CHECK UPPER BOUND: Is byteOffset >= extent(baseRegion)? If so,
139 // we are doing a load/store after the last valid offset.
140 DefinedOrUnknownSVal extentVal =
141 rawOffset.getRegion()->getExtent(svalBuilder);
David Blaikie2fdacbc2013-02-20 05:52:05 +0000142 if (!extentVal.getAs<NonLoc>())
Ted Kremeneke73571b2010-12-23 02:42:43 +0000143 break;
144
145 SVal upperbound
146 = svalBuilder.evalBinOpNN(state, BO_GE, rawOffset.getByteOffset(),
David Blaikie2fdacbc2013-02-20 05:52:05 +0000147 extentVal.castAs<NonLoc>(),
Ted Kremeneke73571b2010-12-23 02:42:43 +0000148 svalBuilder.getConditionType());
Ted Kremenek3a0678e2015-09-08 03:50:52 +0000149
David Blaikie05785d12013-02-20 22:23:23 +0000150 Optional<NonLoc> upperboundToCheck = upperbound.getAs<NonLoc>();
Ted Kremeneke73571b2010-12-23 02:42:43 +0000151 if (!upperboundToCheck)
152 break;
Ted Kremenek3a0678e2015-09-08 03:50:52 +0000153
Ted Kremenek49b1e382012-01-26 21:29:00 +0000154 ProgramStateRef state_exceedsUpperBound, state_withinUpperBound;
Benjamin Kramer867ea1d2014-03-02 13:01:17 +0000155 std::tie(state_exceedsUpperBound, state_withinUpperBound) =
Ted Kremeneke73571b2010-12-23 02:42:43 +0000156 state->assume(*upperboundToCheck);
Anna Zaks20829c92011-11-16 19:58:17 +0000157
158 // If we are under constrained and the index variables are tainted, report.
159 if (state_exceedsUpperBound && state_withinUpperBound) {
160 if (state->isTainted(rawOffset.getByteOffset()))
Anna Zaksb7eac9f2012-01-21 05:07:33 +0000161 reportOOB(checkerContext, state_exceedsUpperBound, OOB_Tainted);
Anna Zaks20829c92011-11-16 19:58:17 +0000162 return;
163 }
Ted Kremenek3a0678e2015-09-08 03:50:52 +0000164
Anna Zaks20829c92011-11-16 19:58:17 +0000165 // If we are constrained enough to definitely exceed the upper bound, report.
166 if (state_exceedsUpperBound) {
167 assert(!state_withinUpperBound);
Ted Kremeneke73571b2010-12-23 02:42:43 +0000168 reportOOB(checkerContext, state_exceedsUpperBound, OOB_Excedes);
169 return;
170 }
Ted Kremenek3a0678e2015-09-08 03:50:52 +0000171
Ted Kremeneke73571b2010-12-23 02:42:43 +0000172 assert(state_withinUpperBound);
173 state = state_withinUpperBound;
174 }
175 while (false);
Ted Kremenek3a0678e2015-09-08 03:50:52 +0000176
Ted Kremeneke73571b2010-12-23 02:42:43 +0000177 if (state != originalState)
Anna Zaksda4c8d62011-10-26 21:06:34 +0000178 checkerContext.addTransition(state);
Ted Kremeneke73571b2010-12-23 02:42:43 +0000179}
180
181void ArrayBoundCheckerV2::reportOOB(CheckerContext &checkerContext,
Ted Kremenek49b1e382012-01-26 21:29:00 +0000182 ProgramStateRef errorState,
Argyrios Kyrtzidis0a9ce3e2011-02-28 01:26:57 +0000183 OOB_Kind kind) const {
Ted Kremenek3a0678e2015-09-08 03:50:52 +0000184
Devin Coughline39bd402015-09-16 22:03:05 +0000185 ExplodedNode *errorNode = checkerContext.generateErrorNode(errorState);
Ted Kremeneke73571b2010-12-23 02:42:43 +0000186 if (!errorNode)
187 return;
188
189 if (!BT)
Alexander Kornienko4aca9b12014-02-11 21:49:21 +0000190 BT.reset(new BuiltinBug(this, "Out-of-bound access"));
Ted Kremeneke73571b2010-12-23 02:42:43 +0000191
192 // FIXME: This diagnostics are preliminary. We should get far better
193 // diagnostics for explaining buffer overruns.
194
Dylan Noblesmith2c1dd272012-02-05 02:13:05 +0000195 SmallString<256> buf;
Ted Kremeneke73571b2010-12-23 02:42:43 +0000196 llvm::raw_svector_ostream os(buf);
Anna Zaksb7eac9f2012-01-21 05:07:33 +0000197 os << "Out of bound memory access ";
198 switch (kind) {
199 case OOB_Precedes:
200 os << "(accessed memory precedes memory block)";
201 break;
202 case OOB_Excedes:
203 os << "(access exceeds upper limit of memory block)";
204 break;
205 case OOB_Tainted:
206 os << "(index is tainted)";
207 break;
208 }
Ted Kremeneke73571b2010-12-23 02:42:43 +0000209
Aaron Ballman8d3a7a52015-06-23 13:15:32 +0000210 checkerContext.emitReport(
211 llvm::make_unique<BugReport>(*BT, os.str(), errorNode));
Ted Kremeneke73571b2010-12-23 02:42:43 +0000212}
213
214void RegionRawOffsetV2::dump() const {
215 dumpToStream(llvm::errs());
216}
217
Ted Kremenek5ef32db2011-08-12 23:37:29 +0000218void RegionRawOffsetV2::dumpToStream(raw_ostream &os) const {
Ted Kremeneke73571b2010-12-23 02:42:43 +0000219 os << "raw_offset_v2{" << getRegion() << ',' << getByteOffset() << '}';
220}
221
Ted Kremeneke73571b2010-12-23 02:42:43 +0000222
223// Lazily computes a value to be used by 'computeOffset'. If 'val'
224// is unknown or undefined, we lazily substitute '0'. Otherwise,
225// return 'val'.
226static inline SVal getValue(SVal val, SValBuilder &svalBuilder) {
David Blaikie2fdacbc2013-02-20 05:52:05 +0000227 return val.getAs<UndefinedVal>() ? svalBuilder.makeArrayIndex(0) : val;
Ted Kremeneke73571b2010-12-23 02:42:43 +0000228}
229
230// Scale a base value by a scaling factor, and return the scaled
231// value as an SVal. Used by 'computeOffset'.
Ted Kremenek49b1e382012-01-26 21:29:00 +0000232static inline SVal scaleValue(ProgramStateRef state,
Ted Kremeneke73571b2010-12-23 02:42:43 +0000233 NonLoc baseVal, CharUnits scaling,
234 SValBuilder &sb) {
235 return sb.evalBinOpNN(state, BO_Mul, baseVal,
236 sb.makeArrayIndex(scaling.getQuantity()),
237 sb.getArrayIndexType());
238}
239
240// Add an SVal to another, treating unknown and undefined values as
241// summing to UnknownVal. Used by 'computeOffset'.
Ted Kremenek49b1e382012-01-26 21:29:00 +0000242static SVal addValue(ProgramStateRef state, SVal x, SVal y,
Ted Kremeneke73571b2010-12-23 02:42:43 +0000243 SValBuilder &svalBuilder) {
244 // We treat UnknownVals and UndefinedVals the same here because we
245 // only care about computing offsets.
246 if (x.isUnknownOrUndef() || y.isUnknownOrUndef())
247 return UnknownVal();
David Blaikie2fdacbc2013-02-20 05:52:05 +0000248
249 return svalBuilder.evalBinOpNN(state, BO_Add, x.castAs<NonLoc>(),
250 y.castAs<NonLoc>(),
Ted Kremeneke73571b2010-12-23 02:42:43 +0000251 svalBuilder.getArrayIndexType());
252}
253
254/// Compute a raw byte offset from a base region. Used for array bounds
255/// checking.
Ted Kremenek49b1e382012-01-26 21:29:00 +0000256RegionRawOffsetV2 RegionRawOffsetV2::computeOffset(ProgramStateRef state,
Ted Kremeneke73571b2010-12-23 02:42:43 +0000257 SValBuilder &svalBuilder,
258 SVal location)
259{
260 const MemRegion *region = location.getAsRegion();
261 SVal offset = UndefinedVal();
Ted Kremenek3a0678e2015-09-08 03:50:52 +0000262
Ted Kremeneke73571b2010-12-23 02:42:43 +0000263 while (region) {
264 switch (region->getKind()) {
265 default: {
Ted Kremenek8a4c7602011-04-12 17:21:33 +0000266 if (const SubRegion *subReg = dyn_cast<SubRegion>(region)) {
267 offset = getValue(offset, svalBuilder);
Ted Kremeneke73571b2010-12-23 02:42:43 +0000268 if (!offset.isUnknownOrUndef())
269 return RegionRawOffsetV2(subReg, offset);
Ted Kremenek8a4c7602011-04-12 17:21:33 +0000270 }
Ted Kremeneke73571b2010-12-23 02:42:43 +0000271 return RegionRawOffsetV2();
272 }
273 case MemRegion::ElementRegionKind: {
274 const ElementRegion *elemReg = cast<ElementRegion>(region);
275 SVal index = elemReg->getIndex();
David Blaikie2fdacbc2013-02-20 05:52:05 +0000276 if (!index.getAs<NonLoc>())
Ted Kremeneke73571b2010-12-23 02:42:43 +0000277 return RegionRawOffsetV2();
278 QualType elemType = elemReg->getElementType();
279 // If the element is an incomplete type, go no further.
280 ASTContext &astContext = svalBuilder.getContext();
Anna Zaks0820e132014-10-03 21:49:03 +0000281 if (elemType->isIncompleteType())
Ted Kremeneke73571b2010-12-23 02:42:43 +0000282 return RegionRawOffsetV2();
Ted Kremenek3a0678e2015-09-08 03:50:52 +0000283
Ted Kremeneke73571b2010-12-23 02:42:43 +0000284 // Update the offset.
285 offset = addValue(state,
286 getValue(offset, svalBuilder),
287 scaleValue(state,
David Blaikie2fdacbc2013-02-20 05:52:05 +0000288 index.castAs<NonLoc>(),
Anna Zaks20829c92011-11-16 19:58:17 +0000289 astContext.getTypeSizeInChars(elemType),
290 svalBuilder),
Ted Kremeneke73571b2010-12-23 02:42:43 +0000291 svalBuilder);
292
293 if (offset.isUnknownOrUndef())
294 return RegionRawOffsetV2();
295
296 region = elemReg->getSuperRegion();
297 continue;
298 }
299 }
300 }
301 return RegionRawOffsetV2();
302}
303
Argyrios Kyrtzidis0a9ce3e2011-02-28 01:26:57 +0000304void ento::registerArrayBoundCheckerV2(CheckerManager &mgr) {
305 mgr.registerChecker<ArrayBoundCheckerV2>();
306}