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/CheckerContext.h"
21 #include "clang/StaticAnalyzer/Core/PathSensitive/ExprEngine.h"
22 #include "llvm/ADT/SmallString.h"
23 #include "llvm/Support/raw_ostream.h"
25 using namespace clang;
29 class ArrayBoundCheckerV2 :
30 public Checker<check::Location> {
31 mutable std::unique_ptr<BuiltinBug> BT;
33 enum OOB_Kind { OOB_Precedes, OOB_Excedes, OOB_Tainted };
35 void reportOOB(CheckerContext &C, ProgramStateRef errorState,
39 void checkLocation(SVal l, bool isLoad, const Stmt*S,
40 CheckerContext &C) const;
43 // FIXME: Eventually replace RegionRawOffset with this class.
44 class RegionRawOffsetV2 {
46 const SubRegion *baseRegion;
50 : baseRegion(nullptr), byteOffset(UnknownVal()) {}
53 RegionRawOffsetV2(const SubRegion* base, SVal offset)
54 : baseRegion(base), byteOffset(offset) {}
56 NonLoc getByteOffset() const { return byteOffset.castAs<NonLoc>(); }
57 const SubRegion *getRegion() const { return baseRegion; }
59 static RegionRawOffsetV2 computeOffset(ProgramStateRef state,
60 SValBuilder &svalBuilder,
64 void dumpToStream(raw_ostream &os) const;
68 static SVal computeExtentBegin(SValBuilder &svalBuilder,
69 const MemRegion *region) {
71 switch (region->getKind()) {
73 return svalBuilder.makeZeroArrayIndex();
74 case MemRegion::SymbolicRegionKind:
75 // FIXME: improve this later by tracking symbolic lower bounds
76 // for symbolic regions.
78 case MemRegion::ElementRegionKind:
79 region = cast<SubRegion>(region)->getSuperRegion();
84 void ArrayBoundCheckerV2::checkLocation(SVal location, bool isLoad,
86 CheckerContext &checkerContext) const {
88 // NOTE: Instead of using ProgramState::assumeInBound(), we are prototyping
89 // 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.
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.
97 ProgramStateRef state = checkerContext.getState();
98 ProgramStateRef originalState = state;
100 SValBuilder &svalBuilder = checkerContext.getSValBuilder();
101 const RegionRawOffsetV2 &rawOffset =
102 RegionRawOffsetV2::computeOffset(state, svalBuilder, location);
104 if (!rawOffset.getRegion())
107 // CHECK LOWER BOUND: Is byteOffset < extent begin?
108 // If so, we are doing a load/store
109 // before the first valid offset in the memory region.
111 SVal extentBegin = computeExtentBegin(svalBuilder, rawOffset.getRegion());
113 if (Optional<NonLoc> NV = extentBegin.getAs<NonLoc>()) {
115 svalBuilder.evalBinOpNN(state, BO_LT, rawOffset.getByteOffset(), *NV,
116 svalBuilder.getConditionType());
118 Optional<NonLoc> lowerBoundToCheck = lowerBound.getAs<NonLoc>();
119 if (!lowerBoundToCheck)
122 ProgramStateRef state_precedesLowerBound, state_withinLowerBound;
123 std::tie(state_precedesLowerBound, state_withinLowerBound) =
124 state->assume(*lowerBoundToCheck);
126 // Are we constrained enough to definitely precede the lower bound?
127 if (state_precedesLowerBound && !state_withinLowerBound) {
128 reportOOB(checkerContext, state_precedesLowerBound, OOB_Precedes);
132 // Otherwise, assume the constraint of the lower bound.
133 assert(state_withinLowerBound);
134 state = state_withinLowerBound;
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);
142 if (!extentVal.getAs<NonLoc>())
146 = svalBuilder.evalBinOpNN(state, BO_GE, rawOffset.getByteOffset(),
147 extentVal.castAs<NonLoc>(),
148 svalBuilder.getConditionType());
150 Optional<NonLoc> upperboundToCheck = upperbound.getAs<NonLoc>();
151 if (!upperboundToCheck)
154 ProgramStateRef state_exceedsUpperBound, state_withinUpperBound;
155 std::tie(state_exceedsUpperBound, state_withinUpperBound) =
156 state->assume(*upperboundToCheck);
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()))
161 reportOOB(checkerContext, state_exceedsUpperBound, OOB_Tainted);
165 // If we are constrained enough to definitely exceed the upper bound, report.
166 if (state_exceedsUpperBound) {
167 assert(!state_withinUpperBound);
168 reportOOB(checkerContext, state_exceedsUpperBound, OOB_Excedes);
172 assert(state_withinUpperBound);
173 state = state_withinUpperBound;
177 if (state != originalState)
178 checkerContext.addTransition(state);
181 void ArrayBoundCheckerV2::reportOOB(CheckerContext &checkerContext,
182 ProgramStateRef errorState,
183 OOB_Kind kind) const {
185 ExplodedNode *errorNode = checkerContext.generateSink(errorState);
190 BT.reset(new BuiltinBug(this, "Out-of-bound access"));
192 // FIXME: This diagnostics are preliminary. We should get far better
193 // diagnostics for explaining buffer overruns.
195 SmallString<256> buf;
196 llvm::raw_svector_ostream os(buf);
197 os << "Out of bound memory access ";
200 os << "(accessed memory precedes memory block)";
203 os << "(access exceeds upper limit of memory block)";
206 os << "(index is tainted)";
210 checkerContext.emitReport(new BugReport(*BT, os.str(), errorNode));
213 void RegionRawOffsetV2::dump() const {
214 dumpToStream(llvm::errs());
217 void RegionRawOffsetV2::dumpToStream(raw_ostream &os) const {
218 os << "raw_offset_v2{" << getRegion() << ',' << getByteOffset() << '}';
222 // Lazily computes a value to be used by 'computeOffset'. If 'val'
223 // is unknown or undefined, we lazily substitute '0'. Otherwise,
225 static inline SVal getValue(SVal val, SValBuilder &svalBuilder) {
226 return val.getAs<UndefinedVal>() ? svalBuilder.makeArrayIndex(0) : val;
229 // Scale a base value by a scaling factor, and return the scaled
230 // value as an SVal. Used by 'computeOffset'.
231 static inline SVal scaleValue(ProgramStateRef state,
232 NonLoc baseVal, CharUnits scaling,
234 return sb.evalBinOpNN(state, BO_Mul, baseVal,
235 sb.makeArrayIndex(scaling.getQuantity()),
236 sb.getArrayIndexType());
239 // Add an SVal to another, treating unknown and undefined values as
240 // summing to UnknownVal. Used by 'computeOffset'.
241 static SVal addValue(ProgramStateRef state, SVal x, SVal y,
242 SValBuilder &svalBuilder) {
243 // We treat UnknownVals and UndefinedVals the same here because we
244 // only care about computing offsets.
245 if (x.isUnknownOrUndef() || y.isUnknownOrUndef())
248 return svalBuilder.evalBinOpNN(state, BO_Add, x.castAs<NonLoc>(),
250 svalBuilder.getArrayIndexType());
253 /// Compute a raw byte offset from a base region. Used for array bounds
255 RegionRawOffsetV2 RegionRawOffsetV2::computeOffset(ProgramStateRef state,
256 SValBuilder &svalBuilder,
259 const MemRegion *region = location.getAsRegion();
260 SVal offset = UndefinedVal();
263 switch (region->getKind()) {
265 if (const SubRegion *subReg = dyn_cast<SubRegion>(region)) {
266 offset = getValue(offset, svalBuilder);
267 if (!offset.isUnknownOrUndef())
268 return RegionRawOffsetV2(subReg, offset);
270 return RegionRawOffsetV2();
272 case MemRegion::ElementRegionKind: {
273 const ElementRegion *elemReg = cast<ElementRegion>(region);
274 SVal index = elemReg->getIndex();
275 if (!index.getAs<NonLoc>())
276 return RegionRawOffsetV2();
277 QualType elemType = elemReg->getElementType();
278 // If the element is an incomplete type, go no further.
279 ASTContext &astContext = svalBuilder.getContext();
280 if (elemType->isIncompleteType())
281 return RegionRawOffsetV2();
283 // Update the offset.
284 offset = addValue(state,
285 getValue(offset, svalBuilder),
287 index.castAs<NonLoc>(),
288 astContext.getTypeSizeInChars(elemType),
292 if (offset.isUnknownOrUndef())
293 return RegionRawOffsetV2();
295 region = elemReg->getSuperRegion();
300 return RegionRawOffsetV2();
303 void ento::registerArrayBoundCheckerV2(CheckerManager &mgr) {
304 mgr.registerChecker<ArrayBoundCheckerV2>();