1 //== ArrayBoundCheckerV2.cpp ------------------------------------*- C++ -*--==//
3 // The LLVM Compiler Infrastructure
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
8 //===----------------------------------------------------------------------===//
10 // This file defines ArrayBoundCheckerV2, which is a path-sensitive check
11 // which looks for an out-of-bound array element access.
13 //===----------------------------------------------------------------------===//
15 #include "ClangSACheckers.h"
16 #include "clang/AST/CharUnits.h"
17 #include "clang/StaticAnalyzer/Core/BugReporter/BugType.h"
18 #include "clang/StaticAnalyzer/Core/Checker.h"
19 #include "clang/StaticAnalyzer/Core/CheckerManager.h"
20 #include "clang/StaticAnalyzer/Core/PathSensitive/APSIntType.h"
21 #include "clang/StaticAnalyzer/Core/PathSensitive/CheckerContext.h"
22 #include "clang/StaticAnalyzer/Core/PathSensitive/ExprEngine.h"
23 #include "llvm/ADT/SmallString.h"
24 #include "llvm/Support/raw_ostream.h"
26 using namespace clang;
30 class ArrayBoundCheckerV2 :
31 public Checker<check::Location> {
32 mutable std::unique_ptr<BuiltinBug> BT;
34 enum OOB_Kind { OOB_Precedes, OOB_Excedes, OOB_Tainted };
36 void reportOOB(CheckerContext &C, ProgramStateRef errorState,
40 void checkLocation(SVal l, bool isLoad, const Stmt*S,
41 CheckerContext &C) const;
44 // FIXME: Eventually replace RegionRawOffset with this class.
45 class RegionRawOffsetV2 {
47 const SubRegion *baseRegion;
51 : baseRegion(nullptr), byteOffset(UnknownVal()) {}
54 RegionRawOffsetV2(const SubRegion* base, SVal offset)
55 : baseRegion(base), byteOffset(offset) {}
57 NonLoc getByteOffset() const { return byteOffset.castAs<NonLoc>(); }
58 const SubRegion *getRegion() const { return baseRegion; }
60 static RegionRawOffsetV2 computeOffset(ProgramStateRef state,
61 SValBuilder &svalBuilder,
65 void dumpToStream(raw_ostream &os) const;
69 static SVal computeExtentBegin(SValBuilder &svalBuilder,
70 const MemRegion *region) {
71 const MemSpaceRegion *SR = region->getMemorySpace();
72 if (SR->getKind() == MemRegion::UnknownSpaceRegionKind)
75 return svalBuilder.makeZeroArrayIndex();
78 // TODO: once the constraint manager is smart enough to handle non simplified
79 // symbolic expressions remove this function. Note that this can not be used in
80 // the constraint manager as is, since this does not handle overflows. It is
81 // safe to assume, however, that memory offsets will not overflow.
82 static std::pair<NonLoc, nonloc::ConcreteInt>
83 getSimplifiedOffsets(NonLoc offset, nonloc::ConcreteInt extent,
84 SValBuilder &svalBuilder) {
85 Optional<nonloc::SymbolVal> SymVal = offset.getAs<nonloc::SymbolVal>();
86 if (SymVal && SymVal->isExpression()) {
87 if (const SymIntExpr *SIE = dyn_cast<SymIntExpr>(SymVal->getSymbol())) {
88 llvm::APSInt constant =
89 APSIntType(extent.getValue()).convert(SIE->getRHS());
90 switch (SIE->getOpcode()) {
92 // The constant should never be 0 here, since it the result of scaling
93 // based on the size of a type which is never 0.
94 if ((extent.getValue() % constant) != 0)
95 return std::pair<NonLoc, nonloc::ConcreteInt>(offset, extent);
97 return getSimplifiedOffsets(
98 nonloc::SymbolVal(SIE->getLHS()),
99 svalBuilder.makeIntVal(extent.getValue() / constant),
102 return getSimplifiedOffsets(
103 nonloc::SymbolVal(SIE->getLHS()),
104 svalBuilder.makeIntVal(extent.getValue() - constant), svalBuilder);
111 return std::pair<NonLoc, nonloc::ConcreteInt>(offset, extent);
114 void ArrayBoundCheckerV2::checkLocation(SVal location, bool isLoad,
116 CheckerContext &checkerContext) const {
118 // NOTE: Instead of using ProgramState::assumeInBound(), we are prototyping
119 // some new logic here that reasons directly about memory region extents.
120 // Once that logic is more mature, we can bring it back to assumeInBound()
121 // for all clients to use.
123 // The algorithm we are using here for bounds checking is to see if the
124 // memory access is within the extent of the base region. Since we
125 // have some flexibility in defining the base region, we can achieve
126 // various levels of conservatism in our buffer overflow checking.
127 ProgramStateRef state = checkerContext.getState();
128 ProgramStateRef originalState = state;
130 SValBuilder &svalBuilder = checkerContext.getSValBuilder();
131 const RegionRawOffsetV2 &rawOffset =
132 RegionRawOffsetV2::computeOffset(state, svalBuilder, location);
134 if (!rawOffset.getRegion())
137 NonLoc rawOffsetVal = rawOffset.getByteOffset();
139 // CHECK LOWER BOUND: Is byteOffset < extent begin?
140 // If so, we are doing a load/store
141 // before the first valid offset in the memory region.
143 SVal extentBegin = computeExtentBegin(svalBuilder, rawOffset.getRegion());
145 if (Optional<NonLoc> NV = extentBegin.getAs<NonLoc>()) {
146 if (NV->getAs<nonloc::ConcreteInt>()) {
147 std::pair<NonLoc, nonloc::ConcreteInt> simplifiedOffsets =
148 getSimplifiedOffsets(rawOffset.getByteOffset(),
149 NV->castAs<nonloc::ConcreteInt>(),
151 rawOffsetVal = simplifiedOffsets.first;
152 *NV = simplifiedOffsets.second;
155 SVal lowerBound = svalBuilder.evalBinOpNN(state, BO_LT, rawOffsetVal, *NV,
156 svalBuilder.getConditionType());
158 Optional<NonLoc> lowerBoundToCheck = lowerBound.getAs<NonLoc>();
159 if (!lowerBoundToCheck)
162 ProgramStateRef state_precedesLowerBound, state_withinLowerBound;
163 std::tie(state_precedesLowerBound, state_withinLowerBound) =
164 state->assume(*lowerBoundToCheck);
166 // Are we constrained enough to definitely precede the lower bound?
167 if (state_precedesLowerBound && !state_withinLowerBound) {
168 reportOOB(checkerContext, state_precedesLowerBound, OOB_Precedes);
172 // Otherwise, assume the constraint of the lower bound.
173 assert(state_withinLowerBound);
174 state = state_withinLowerBound;
178 // CHECK UPPER BOUND: Is byteOffset >= extent(baseRegion)? If so,
179 // we are doing a load/store after the last valid offset.
180 DefinedOrUnknownSVal extentVal =
181 rawOffset.getRegion()->getExtent(svalBuilder);
182 if (!extentVal.getAs<NonLoc>())
185 if (extentVal.getAs<nonloc::ConcreteInt>()) {
186 std::pair<NonLoc, nonloc::ConcreteInt> simplifiedOffsets =
187 getSimplifiedOffsets(rawOffset.getByteOffset(),
188 extentVal.castAs<nonloc::ConcreteInt>(),
190 rawOffsetVal = simplifiedOffsets.first;
191 extentVal = simplifiedOffsets.second;
194 SVal upperbound = svalBuilder.evalBinOpNN(state, BO_GE, rawOffsetVal,
195 extentVal.castAs<NonLoc>(),
196 svalBuilder.getConditionType());
198 Optional<NonLoc> upperboundToCheck = upperbound.getAs<NonLoc>();
199 if (!upperboundToCheck)
202 ProgramStateRef state_exceedsUpperBound, state_withinUpperBound;
203 std::tie(state_exceedsUpperBound, state_withinUpperBound) =
204 state->assume(*upperboundToCheck);
206 // If we are under constrained and the index variables are tainted, report.
207 if (state_exceedsUpperBound && state_withinUpperBound) {
208 if (state->isTainted(rawOffset.getByteOffset())) {
209 reportOOB(checkerContext, state_exceedsUpperBound, OOB_Tainted);
212 } else if (state_exceedsUpperBound) {
213 // If we are constrained enough to definitely exceed the upper bound,
215 assert(!state_withinUpperBound);
216 reportOOB(checkerContext, state_exceedsUpperBound, OOB_Excedes);
220 assert(state_withinUpperBound);
221 state = state_withinUpperBound;
225 if (state != originalState)
226 checkerContext.addTransition(state);
229 void ArrayBoundCheckerV2::reportOOB(CheckerContext &checkerContext,
230 ProgramStateRef errorState,
231 OOB_Kind kind) const {
233 ExplodedNode *errorNode = checkerContext.generateErrorNode(errorState);
238 BT.reset(new BuiltinBug(this, "Out-of-bound access"));
240 // FIXME: This diagnostics are preliminary. We should get far better
241 // diagnostics for explaining buffer overruns.
243 SmallString<256> buf;
244 llvm::raw_svector_ostream os(buf);
245 os << "Out of bound memory access ";
248 os << "(accessed memory precedes memory block)";
251 os << "(access exceeds upper limit of memory block)";
254 os << "(index is tainted)";
258 checkerContext.emitReport(
259 llvm::make_unique<BugReport>(*BT, os.str(), errorNode));
262 LLVM_DUMP_METHOD void RegionRawOffsetV2::dump() const {
263 dumpToStream(llvm::errs());
266 void RegionRawOffsetV2::dumpToStream(raw_ostream &os) const {
267 os << "raw_offset_v2{" << getRegion() << ',' << getByteOffset() << '}';
271 // Lazily computes a value to be used by 'computeOffset'. If 'val'
272 // is unknown or undefined, we lazily substitute '0'. Otherwise,
274 static inline SVal getValue(SVal val, SValBuilder &svalBuilder) {
275 return val.getAs<UndefinedVal>() ? svalBuilder.makeArrayIndex(0) : val;
278 // Scale a base value by a scaling factor, and return the scaled
279 // value as an SVal. Used by 'computeOffset'.
280 static inline SVal scaleValue(ProgramStateRef state,
281 NonLoc baseVal, CharUnits scaling,
283 return sb.evalBinOpNN(state, BO_Mul, baseVal,
284 sb.makeArrayIndex(scaling.getQuantity()),
285 sb.getArrayIndexType());
288 // Add an SVal to another, treating unknown and undefined values as
289 // summing to UnknownVal. Used by 'computeOffset'.
290 static SVal addValue(ProgramStateRef state, SVal x, SVal y,
291 SValBuilder &svalBuilder) {
292 // We treat UnknownVals and UndefinedVals the same here because we
293 // only care about computing offsets.
294 if (x.isUnknownOrUndef() || y.isUnknownOrUndef())
297 return svalBuilder.evalBinOpNN(state, BO_Add, x.castAs<NonLoc>(),
299 svalBuilder.getArrayIndexType());
302 /// Compute a raw byte offset from a base region. Used for array bounds
304 RegionRawOffsetV2 RegionRawOffsetV2::computeOffset(ProgramStateRef state,
305 SValBuilder &svalBuilder,
308 const MemRegion *region = location.getAsRegion();
309 SVal offset = UndefinedVal();
312 switch (region->getKind()) {
314 if (const SubRegion *subReg = dyn_cast<SubRegion>(region)) {
315 offset = getValue(offset, svalBuilder);
316 if (!offset.isUnknownOrUndef())
317 return RegionRawOffsetV2(subReg, offset);
319 return RegionRawOffsetV2();
321 case MemRegion::ElementRegionKind: {
322 const ElementRegion *elemReg = cast<ElementRegion>(region);
323 SVal index = elemReg->getIndex();
324 if (!index.getAs<NonLoc>())
325 return RegionRawOffsetV2();
326 QualType elemType = elemReg->getElementType();
327 // If the element is an incomplete type, go no further.
328 ASTContext &astContext = svalBuilder.getContext();
329 if (elemType->isIncompleteType())
330 return RegionRawOffsetV2();
332 // Update the offset.
333 offset = addValue(state,
334 getValue(offset, svalBuilder),
336 index.castAs<NonLoc>(),
337 astContext.getTypeSizeInChars(elemType),
341 if (offset.isUnknownOrUndef())
342 return RegionRawOffsetV2();
344 region = elemReg->getSuperRegion();
349 return RegionRawOffsetV2();
352 void ento::registerArrayBoundCheckerV2(CheckerManager &mgr) {
353 mgr.registerChecker<ArrayBoundCheckerV2>();