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/StaticAnalyzer/Core/Checker.h"
17 #include "clang/StaticAnalyzer/Core/CheckerManager.h"
18 #include "clang/StaticAnalyzer/Core/PathSensitive/CheckerContext.h"
19 #include "clang/StaticAnalyzer/Core/BugReporter/BugType.h"
20 #include "clang/StaticAnalyzer/Core/PathSensitive/ExprEngine.h"
21 #include "clang/AST/CharUnits.h"
22 #include "llvm/ADT/SmallString.h"
23 #include "llvm/ADT/STLExtras.h"
25 using namespace clang;
29 class ArrayBoundCheckerV2 :
30 public Checker<check::Location> {
31 mutable OwningPtr<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(0), byteOffset(UnknownVal()) {}
53 RegionRawOffsetV2(const SubRegion* base, SVal offset)
54 : baseRegion(base), byteOffset(offset) {}
56 NonLoc getByteOffset() const { return cast<NonLoc>(byteOffset); }
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 (isa<NonLoc>(extentBegin)) {
115 = svalBuilder.evalBinOpNN(state, BO_LT, rawOffset.getByteOffset(),
116 cast<NonLoc>(extentBegin),
117 svalBuilder.getConditionType());
119 NonLoc *lowerBoundToCheck = dyn_cast<NonLoc>(&lowerBound);
120 if (!lowerBoundToCheck)
123 ProgramStateRef state_precedesLowerBound, state_withinLowerBound;
124 llvm::tie(state_precedesLowerBound, state_withinLowerBound) =
125 state->assume(*lowerBoundToCheck);
127 // Are we constrained enough to definitely precede the lower bound?
128 if (state_precedesLowerBound && !state_withinLowerBound) {
129 reportOOB(checkerContext, state_precedesLowerBound, OOB_Precedes);
133 // Otherwise, assume the constraint of the lower bound.
134 assert(state_withinLowerBound);
135 state = state_withinLowerBound;
139 // CHECK UPPER BOUND: Is byteOffset >= extent(baseRegion)? If so,
140 // we are doing a load/store after the last valid offset.
141 DefinedOrUnknownSVal extentVal =
142 rawOffset.getRegion()->getExtent(svalBuilder);
143 if (!isa<NonLoc>(extentVal))
147 = svalBuilder.evalBinOpNN(state, BO_GE, rawOffset.getByteOffset(),
148 cast<NonLoc>(extentVal),
149 svalBuilder.getConditionType());
151 NonLoc *upperboundToCheck = dyn_cast<NonLoc>(&upperbound);
152 if (!upperboundToCheck)
155 ProgramStateRef state_exceedsUpperBound, state_withinUpperBound;
156 llvm::tie(state_exceedsUpperBound, state_withinUpperBound) =
157 state->assume(*upperboundToCheck);
159 // If we are under constrained and the index variables are tainted, report.
160 if (state_exceedsUpperBound && state_withinUpperBound) {
161 if (state->isTainted(rawOffset.getByteOffset()))
162 reportOOB(checkerContext, state_exceedsUpperBound, OOB_Tainted);
166 // If we are constrained enough to definitely exceed the upper bound, report.
167 if (state_exceedsUpperBound) {
168 assert(!state_withinUpperBound);
169 reportOOB(checkerContext, state_exceedsUpperBound, OOB_Excedes);
173 assert(state_withinUpperBound);
174 state = state_withinUpperBound;
178 if (state != originalState)
179 checkerContext.addTransition(state);
182 void ArrayBoundCheckerV2::reportOOB(CheckerContext &checkerContext,
183 ProgramStateRef errorState,
184 OOB_Kind kind) const {
186 ExplodedNode *errorNode = checkerContext.generateSink(errorState);
191 BT.reset(new BuiltinBug("Out-of-bound access"));
193 // FIXME: This diagnostics are preliminary. We should get far better
194 // diagnostics for explaining buffer overruns.
196 SmallString<256> buf;
197 llvm::raw_svector_ostream os(buf);
198 os << "Out of bound memory access ";
201 os << "(accessed memory precedes memory block)";
204 os << "(access exceeds upper limit of memory block)";
207 os << "(index is tainted)";
211 checkerContext.EmitReport(new BugReport(*BT, os.str(), errorNode));
214 void RegionRawOffsetV2::dump() const {
215 dumpToStream(llvm::errs());
218 void RegionRawOffsetV2::dumpToStream(raw_ostream &os) const {
219 os << "raw_offset_v2{" << getRegion() << ',' << getByteOffset() << '}';
222 // FIXME: Merge with the implementation of the same method in Store.cpp
223 static bool IsCompleteType(ASTContext &Ctx, QualType Ty) {
224 if (const RecordType *RT = Ty->getAs<RecordType>()) {
225 const RecordDecl *D = RT->getDecl();
226 if (!D->getDefinition())
234 // Lazily computes a value to be used by 'computeOffset'. If 'val'
235 // is unknown or undefined, we lazily substitute '0'. Otherwise,
237 static inline SVal getValue(SVal val, SValBuilder &svalBuilder) {
238 return isa<UndefinedVal>(val) ? svalBuilder.makeArrayIndex(0) : val;
241 // Scale a base value by a scaling factor, and return the scaled
242 // value as an SVal. Used by 'computeOffset'.
243 static inline SVal scaleValue(ProgramStateRef state,
244 NonLoc baseVal, CharUnits scaling,
246 return sb.evalBinOpNN(state, BO_Mul, baseVal,
247 sb.makeArrayIndex(scaling.getQuantity()),
248 sb.getArrayIndexType());
251 // Add an SVal to another, treating unknown and undefined values as
252 // summing to UnknownVal. Used by 'computeOffset'.
253 static SVal addValue(ProgramStateRef state, SVal x, SVal y,
254 SValBuilder &svalBuilder) {
255 // We treat UnknownVals and UndefinedVals the same here because we
256 // only care about computing offsets.
257 if (x.isUnknownOrUndef() || y.isUnknownOrUndef())
260 return svalBuilder.evalBinOpNN(state, BO_Add,
261 cast<NonLoc>(x), cast<NonLoc>(y),
262 svalBuilder.getArrayIndexType());
265 /// Compute a raw byte offset from a base region. Used for array bounds
267 RegionRawOffsetV2 RegionRawOffsetV2::computeOffset(ProgramStateRef state,
268 SValBuilder &svalBuilder,
271 const MemRegion *region = location.getAsRegion();
272 SVal offset = UndefinedVal();
275 switch (region->getKind()) {
277 if (const SubRegion *subReg = dyn_cast<SubRegion>(region)) {
278 offset = getValue(offset, svalBuilder);
279 if (!offset.isUnknownOrUndef())
280 return RegionRawOffsetV2(subReg, offset);
282 return RegionRawOffsetV2();
284 case MemRegion::ElementRegionKind: {
285 const ElementRegion *elemReg = cast<ElementRegion>(region);
286 SVal index = elemReg->getIndex();
287 if (!isa<NonLoc>(index))
288 return RegionRawOffsetV2();
289 QualType elemType = elemReg->getElementType();
290 // If the element is an incomplete type, go no further.
291 ASTContext &astContext = svalBuilder.getContext();
292 if (!IsCompleteType(astContext, elemType))
293 return RegionRawOffsetV2();
295 // Update the offset.
296 offset = addValue(state,
297 getValue(offset, svalBuilder),
300 astContext.getTypeSizeInChars(elemType),
304 if (offset.isUnknownOrUndef())
305 return RegionRawOffsetV2();
307 region = elemReg->getSuperRegion();
312 return RegionRawOffsetV2();
316 void ento::registerArrayBoundCheckerV2(CheckerManager &mgr) {
317 mgr.registerChecker<ArrayBoundCheckerV2>();