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, OOB_Kind kind,
37 std::unique_ptr<BugReporterVisitor> Visitor = nullptr) const;
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();
129 SValBuilder &svalBuilder = checkerContext.getSValBuilder();
130 const RegionRawOffsetV2 &rawOffset =
131 RegionRawOffsetV2::computeOffset(state, svalBuilder, location);
133 if (!rawOffset.getRegion())
136 NonLoc rawOffsetVal = rawOffset.getByteOffset();
138 // CHECK LOWER BOUND: Is byteOffset < extent begin?
139 // If so, we are doing a load/store
140 // before the first valid offset in the memory region.
142 SVal extentBegin = computeExtentBegin(svalBuilder, rawOffset.getRegion());
144 if (Optional<NonLoc> NV = extentBegin.getAs<NonLoc>()) {
145 if (NV->getAs<nonloc::ConcreteInt>()) {
146 std::pair<NonLoc, nonloc::ConcreteInt> simplifiedOffsets =
147 getSimplifiedOffsets(rawOffset.getByteOffset(),
148 NV->castAs<nonloc::ConcreteInt>(),
150 rawOffsetVal = simplifiedOffsets.first;
151 *NV = simplifiedOffsets.second;
154 SVal lowerBound = svalBuilder.evalBinOpNN(state, BO_LT, rawOffsetVal, *NV,
155 svalBuilder.getConditionType());
157 Optional<NonLoc> lowerBoundToCheck = lowerBound.getAs<NonLoc>();
158 if (!lowerBoundToCheck)
161 ProgramStateRef state_precedesLowerBound, state_withinLowerBound;
162 std::tie(state_precedesLowerBound, state_withinLowerBound) =
163 state->assume(*lowerBoundToCheck);
165 // Are we constrained enough to definitely precede the lower bound?
166 if (state_precedesLowerBound && !state_withinLowerBound) {
167 reportOOB(checkerContext, state_precedesLowerBound, OOB_Precedes);
171 // Otherwise, assume the constraint of the lower bound.
172 assert(state_withinLowerBound);
173 state = state_withinLowerBound;
177 // CHECK UPPER BOUND: Is byteOffset >= extent(baseRegion)? If so,
178 // we are doing a load/store after the last valid offset.
179 DefinedOrUnknownSVal extentVal =
180 rawOffset.getRegion()->getExtent(svalBuilder);
181 if (!extentVal.getAs<NonLoc>())
184 if (extentVal.getAs<nonloc::ConcreteInt>()) {
185 std::pair<NonLoc, nonloc::ConcreteInt> simplifiedOffsets =
186 getSimplifiedOffsets(rawOffset.getByteOffset(),
187 extentVal.castAs<nonloc::ConcreteInt>(),
189 rawOffsetVal = simplifiedOffsets.first;
190 extentVal = simplifiedOffsets.second;
193 SVal upperbound = svalBuilder.evalBinOpNN(state, BO_GE, rawOffsetVal,
194 extentVal.castAs<NonLoc>(),
195 svalBuilder.getConditionType());
197 Optional<NonLoc> upperboundToCheck = upperbound.getAs<NonLoc>();
198 if (!upperboundToCheck)
201 ProgramStateRef state_exceedsUpperBound, state_withinUpperBound;
202 std::tie(state_exceedsUpperBound, state_withinUpperBound) =
203 state->assume(*upperboundToCheck);
205 // If we are under constrained and the index variables are tainted, report.
206 if (state_exceedsUpperBound && state_withinUpperBound) {
207 SVal ByteOffset = rawOffset.getByteOffset();
208 if (state->isTainted(ByteOffset)) {
209 reportOOB(checkerContext, state_exceedsUpperBound, OOB_Tainted,
210 llvm::make_unique<TaintBugVisitor>(ByteOffset));
213 } else if (state_exceedsUpperBound) {
214 // If we are constrained enough to definitely exceed the upper bound,
216 assert(!state_withinUpperBound);
217 reportOOB(checkerContext, state_exceedsUpperBound, OOB_Excedes);
221 assert(state_withinUpperBound);
222 state = state_withinUpperBound;
226 checkerContext.addTransition(state);
229 void ArrayBoundCheckerV2::reportOOB(
230 CheckerContext &checkerContext, ProgramStateRef errorState, OOB_Kind kind,
231 std::unique_ptr<BugReporterVisitor> Visitor) 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 auto BR = llvm::make_unique<BugReport>(*BT, os.str(), errorNode);
259 BR->addVisitor(std::move(Visitor));
260 checkerContext.emitReport(std::move(BR));
264 LLVM_DUMP_METHOD void RegionRawOffsetV2::dump() const {
265 dumpToStream(llvm::errs());
268 void RegionRawOffsetV2::dumpToStream(raw_ostream &os) const {
269 os << "raw_offset_v2{" << getRegion() << ',' << getByteOffset() << '}';
273 // Lazily computes a value to be used by 'computeOffset'. If 'val'
274 // is unknown or undefined, we lazily substitute '0'. Otherwise,
276 static inline SVal getValue(SVal val, SValBuilder &svalBuilder) {
277 return val.getAs<UndefinedVal>() ? svalBuilder.makeArrayIndex(0) : val;
280 // Scale a base value by a scaling factor, and return the scaled
281 // value as an SVal. Used by 'computeOffset'.
282 static inline SVal scaleValue(ProgramStateRef state,
283 NonLoc baseVal, CharUnits scaling,
285 return sb.evalBinOpNN(state, BO_Mul, baseVal,
286 sb.makeArrayIndex(scaling.getQuantity()),
287 sb.getArrayIndexType());
290 // Add an SVal to another, treating unknown and undefined values as
291 // summing to UnknownVal. Used by 'computeOffset'.
292 static SVal addValue(ProgramStateRef state, SVal x, SVal y,
293 SValBuilder &svalBuilder) {
294 // We treat UnknownVals and UndefinedVals the same here because we
295 // only care about computing offsets.
296 if (x.isUnknownOrUndef() || y.isUnknownOrUndef())
299 return svalBuilder.evalBinOpNN(state, BO_Add, x.castAs<NonLoc>(),
301 svalBuilder.getArrayIndexType());
304 /// Compute a raw byte offset from a base region. Used for array bounds
306 RegionRawOffsetV2 RegionRawOffsetV2::computeOffset(ProgramStateRef state,
307 SValBuilder &svalBuilder,
310 const MemRegion *region = location.getAsRegion();
311 SVal offset = UndefinedVal();
314 switch (region->getKind()) {
316 if (const SubRegion *subReg = dyn_cast<SubRegion>(region)) {
317 offset = getValue(offset, svalBuilder);
318 if (!offset.isUnknownOrUndef())
319 return RegionRawOffsetV2(subReg, offset);
321 return RegionRawOffsetV2();
323 case MemRegion::ElementRegionKind: {
324 const ElementRegion *elemReg = cast<ElementRegion>(region);
325 SVal index = elemReg->getIndex();
326 if (!index.getAs<NonLoc>())
327 return RegionRawOffsetV2();
328 QualType elemType = elemReg->getElementType();
329 // If the element is an incomplete type, go no further.
330 ASTContext &astContext = svalBuilder.getContext();
331 if (elemType->isIncompleteType())
332 return RegionRawOffsetV2();
334 // Update the offset.
335 offset = addValue(state,
336 getValue(offset, svalBuilder),
338 index.castAs<NonLoc>(),
339 astContext.getTypeSizeInChars(elemType),
343 if (offset.isUnknownOrUndef())
344 return RegionRawOffsetV2();
346 region = elemReg->getSuperRegion();
351 return RegionRawOffsetV2();
354 void ento::registerArrayBoundCheckerV2(CheckerManager &mgr) {
355 mgr.registerChecker<ArrayBoundCheckerV2>();