1 //== ArrayBoundCheckerV2.cpp ------------------------------------*- C++ -*--==//
2 //
3 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
4 // See https://llvm.org/LICENSE.txt for license information.
5 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
6 //
7 //===----------------------------------------------------------------------===//
8 //
9 // This file defines ArrayBoundCheckerV2, which is a path-sensitive check
10 // which looks for an out-of-bound array element access.
11 //
12 //===----------------------------------------------------------------------===//
13 
14 #include "clang/AST/CharUnits.h"
15 #include "clang/StaticAnalyzer/Checkers/BuiltinCheckerRegistration.h"
16 #include "clang/StaticAnalyzer/Checkers/Taint.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/DynamicExtent.h"
23 #include "clang/StaticAnalyzer/Core/PathSensitive/ExprEngine.h"
24 #include "llvm/ADT/SmallString.h"
25 #include "llvm/Support/raw_ostream.h"
26 #include <optional>
27 
28 using namespace clang;
29 using namespace ento;
30 using namespace taint;
31 
32 namespace {
33 class ArrayBoundCheckerV2 :
34     public Checker<check::Location> {
35   mutable std::unique_ptr<BuiltinBug> BT;
36   mutable std::unique_ptr<BugType> TaintBT;
37 
38   enum OOB_Kind { OOB_Precedes, OOB_Excedes };
39 
40   void reportOOB(CheckerContext &C, ProgramStateRef errorState,
41                  OOB_Kind kind) const;
42   void reportTaintOOB(CheckerContext &C, ProgramStateRef errorState,
43                       SVal TaintedSVal) const;
44 
45   static bool isFromCtypeMacro(const Stmt *S, ASTContext &AC);
46 
47 public:
48   void checkLocation(SVal l, bool isLoad, const Stmt *S,
49                      CheckerContext &C) const;
50 };
51 
52 // FIXME: Eventually replace RegionRawOffset with this class.
53 class RegionRawOffsetV2 {
54 private:
55   const SubRegion *baseRegion;
56   NonLoc byteOffset;
57 
58 public:
59   RegionRawOffsetV2(const SubRegion *base, NonLoc offset)
60       : baseRegion(base), byteOffset(offset) { assert(base); }
61 
62   NonLoc getByteOffset() const { return byteOffset; }
63   const SubRegion *getRegion() const { return baseRegion; }
64 
65   static std::optional<RegionRawOffsetV2>
66   computeOffset(ProgramStateRef State, SValBuilder &SVB, SVal Location);
67 
68   void dump() const;
69   void dumpToStream(raw_ostream &os) const;
70 };
71 }
72 
73 // TODO: once the constraint manager is smart enough to handle non simplified
74 // symbolic expressions remove this function. Note that this can not be used in
75 // the constraint manager as is, since this does not handle overflows. It is
76 // safe to assume, however, that memory offsets will not overflow.
77 // NOTE: callers of this function need to be aware of the effects of overflows
78 // and signed<->unsigned conversions!
79 static std::pair<NonLoc, nonloc::ConcreteInt>
80 getSimplifiedOffsets(NonLoc offset, nonloc::ConcreteInt extent,
81                      SValBuilder &svalBuilder) {
82   std::optional<nonloc::SymbolVal> SymVal = offset.getAs<nonloc::SymbolVal>();
83   if (SymVal && SymVal->isExpression()) {
84     if (const SymIntExpr *SIE = dyn_cast<SymIntExpr>(SymVal->getSymbol())) {
85       llvm::APSInt constant =
86           APSIntType(extent.getValue()).convert(SIE->getRHS());
87       switch (SIE->getOpcode()) {
88       case BO_Mul:
89         // The constant should never be 0 here, since it the result of scaling
90         // based on the size of a type which is never 0.
91         if ((extent.getValue() % constant) != 0)
92           return std::pair<NonLoc, nonloc::ConcreteInt>(offset, extent);
93         else
94           return getSimplifiedOffsets(
95               nonloc::SymbolVal(SIE->getLHS()),
96               svalBuilder.makeIntVal(extent.getValue() / constant),
97               svalBuilder);
98       case BO_Add:
99         return getSimplifiedOffsets(
100             nonloc::SymbolVal(SIE->getLHS()),
101             svalBuilder.makeIntVal(extent.getValue() - constant), svalBuilder);
102       default:
103         break;
104       }
105     }
106   }
107 
108   return std::pair<NonLoc, nonloc::ConcreteInt>(offset, extent);
109 }
110 
111 // Evaluate the comparison Value < Threshold with the help of the custom
112 // simplification algorithm defined for this checker. Return a pair of states,
113 // where the first one corresponds to "value below threshold" and the second
114 // corresponds to "value at or above threshold". Returns {nullptr, nullptr} in
115 // the case when the evaluation fails.
116 static std::pair<ProgramStateRef, ProgramStateRef>
117 compareValueToThreshold(ProgramStateRef State, NonLoc Value, NonLoc Threshold,
118                         SValBuilder &SVB) {
119   if (auto ConcreteThreshold = Threshold.getAs<nonloc::ConcreteInt>()) {
120     std::tie(Value, Threshold) = getSimplifiedOffsets(Value, *ConcreteThreshold, SVB);
121   }
122   if (auto ConcreteThreshold = Threshold.getAs<nonloc::ConcreteInt>()) {
123     QualType T = Value.getType(SVB.getContext());
124     if (T->isUnsignedIntegerType() && ConcreteThreshold->getValue().isNegative()) {
125       // In this case we reduced the bound check to a comparison of the form
126       //   (symbol or value with unsigned type) < (negative number)
127       // which is always false. We are handling these cases separately because
128       // evalBinOpNN can perform a signed->unsigned conversion that turns the
129       // negative number into a huge positive value and leads to wildly
130       // inaccurate conclusions.
131       return {nullptr, State};
132     }
133   }
134   auto BelowThreshold =
135       SVB.evalBinOpNN(State, BO_LT, Value, Threshold, SVB.getConditionType()).getAs<NonLoc>();
136 
137   if (BelowThreshold)
138     return State->assume(*BelowThreshold);
139 
140   return {nullptr, nullptr};
141 }
142 
143 void ArrayBoundCheckerV2::checkLocation(SVal location, bool isLoad,
144                                         const Stmt* LoadS,
145                                         CheckerContext &checkerContext) const {
146 
147   // NOTE: Instead of using ProgramState::assumeInBound(), we are prototyping
148   // some new logic here that reasons directly about memory region extents.
149   // Once that logic is more mature, we can bring it back to assumeInBound()
150   // for all clients to use.
151   //
152   // The algorithm we are using here for bounds checking is to see if the
153   // memory access is within the extent of the base region.  Since we
154   // have some flexibility in defining the base region, we can achieve
155   // various levels of conservatism in our buffer overflow checking.
156 
157   // The header ctype.h (from e.g. glibc) implements the isXXXXX() macros as
158   //   #define isXXXXX(arg) (LOOKUP_TABLE[arg] & BITMASK_FOR_XXXXX)
159   // and incomplete analysis of these leads to false positives. As even
160   // accurate reports would be confusing for the users, just disable reports
161   // from these macros:
162   if (isFromCtypeMacro(LoadS, checkerContext.getASTContext()))
163     return;
164 
165   ProgramStateRef state = checkerContext.getState();
166 
167   SValBuilder &svalBuilder = checkerContext.getSValBuilder();
168   const std::optional<RegionRawOffsetV2> &RawOffset =
169       RegionRawOffsetV2::computeOffset(state, svalBuilder, location);
170 
171   if (!RawOffset)
172     return;
173 
174   NonLoc ByteOffset = RawOffset->getByteOffset();
175 
176   // CHECK LOWER BOUND
177   const MemSpaceRegion *SR = RawOffset->getRegion()->getMemorySpace();
178   if (!llvm::isa<UnknownSpaceRegion>(SR)) {
179     // A pointer to UnknownSpaceRegion may point to the middle of
180     // an allocated region.
181 
182     auto [state_precedesLowerBound, state_withinLowerBound] =
183         compareValueToThreshold(state, ByteOffset,
184                                 svalBuilder.makeZeroArrayIndex(), svalBuilder);
185 
186     if (state_precedesLowerBound && !state_withinLowerBound) {
187       // We know that the index definitely precedes the lower bound.
188       reportOOB(checkerContext, state_precedesLowerBound, OOB_Precedes);
189       return;
190     }
191 
192     if (state_withinLowerBound)
193       state = state_withinLowerBound;
194   }
195 
196   // CHECK UPPER BOUND
197   DefinedOrUnknownSVal Size =
198       getDynamicExtent(state, RawOffset->getRegion(), svalBuilder);
199   if (auto KnownSize = Size.getAs<NonLoc>()) {
200     auto [state_withinUpperBound, state_exceedsUpperBound] =
201         compareValueToThreshold(state, ByteOffset, *KnownSize, svalBuilder);
202 
203     if (state_exceedsUpperBound) {
204       if (!state_withinUpperBound) {
205         // We know that the index definitely exceeds the upper bound.
206         reportOOB(checkerContext, state_exceedsUpperBound, OOB_Excedes);
207         return;
208       }
209       if (isTainted(state, ByteOffset)) {
210         // Both cases are possible, but the index is tainted, so report.
211         reportTaintOOB(checkerContext, state_exceedsUpperBound, ByteOffset);
212         return;
213       }
214     }
215 
216     if (state_withinUpperBound)
217       state = state_withinUpperBound;
218   }
219 
220   checkerContext.addTransition(state);
221 }
222 
223 void ArrayBoundCheckerV2::reportTaintOOB(CheckerContext &checkerContext,
224                                          ProgramStateRef errorState,
225                                          SVal TaintedSVal) const {
226   ExplodedNode *errorNode = checkerContext.generateErrorNode(errorState);
227   if (!errorNode)
228     return;
229 
230   if (!TaintBT)
231     TaintBT.reset(
232         new BugType(this, "Out-of-bound access", categories::TaintedData));
233 
234   SmallString<256> buf;
235   llvm::raw_svector_ostream os(buf);
236   os << "Out of bound memory access (index is tainted)";
237   auto BR =
238       std::make_unique<PathSensitiveBugReport>(*TaintBT, os.str(), errorNode);
239 
240   // Track back the propagation of taintedness.
241   for (SymbolRef Sym : getTaintedSymbols(errorState, TaintedSVal)) {
242     BR->markInteresting(Sym);
243   }
244 
245   checkerContext.emitReport(std::move(BR));
246 }
247 
248 void ArrayBoundCheckerV2::reportOOB(CheckerContext &checkerContext,
249                                     ProgramStateRef errorState,
250                                     OOB_Kind kind) const {
251 
252   ExplodedNode *errorNode = checkerContext.generateErrorNode(errorState);
253   if (!errorNode)
254     return;
255 
256   if (!BT)
257     BT.reset(new BuiltinBug(this, "Out-of-bound access"));
258 
259   // FIXME: This diagnostics are preliminary.  We should get far better
260   // diagnostics for explaining buffer overruns.
261 
262   SmallString<256> buf;
263   llvm::raw_svector_ostream os(buf);
264   os << "Out of bound memory access ";
265   switch (kind) {
266   case OOB_Precedes:
267     os << "(accessed memory precedes memory block)";
268     break;
269   case OOB_Excedes:
270     os << "(access exceeds upper limit of memory block)";
271     break;
272   }
273   auto BR = std::make_unique<PathSensitiveBugReport>(*BT, os.str(), errorNode);
274   checkerContext.emitReport(std::move(BR));
275 }
276 
277 bool ArrayBoundCheckerV2::isFromCtypeMacro(const Stmt *S, ASTContext &ACtx) {
278   SourceLocation Loc = S->getBeginLoc();
279   if (!Loc.isMacroID())
280     return false;
281 
282   StringRef MacroName = Lexer::getImmediateMacroName(
283       Loc, ACtx.getSourceManager(), ACtx.getLangOpts());
284 
285   if (MacroName.size() < 7 || MacroName[0] != 'i' || MacroName[1] != 's')
286     return false;
287 
288   return ((MacroName == "isalnum") || (MacroName == "isalpha") ||
289           (MacroName == "isblank") || (MacroName == "isdigit") ||
290           (MacroName == "isgraph") || (MacroName == "islower") ||
291           (MacroName == "isnctrl") || (MacroName == "isprint") ||
292           (MacroName == "ispunct") || (MacroName == "isspace") ||
293           (MacroName == "isupper") || (MacroName == "isxdigit"));
294 }
295 
296 #ifndef NDEBUG
297 LLVM_DUMP_METHOD void RegionRawOffsetV2::dump() const {
298   dumpToStream(llvm::errs());
299 }
300 
301 void RegionRawOffsetV2::dumpToStream(raw_ostream &os) const {
302   os << "raw_offset_v2{" << getRegion() << ',' << getByteOffset() << '}';
303 }
304 #endif
305 
306 /// For a given Location that can be represented as a symbolic expression
307 /// Arr[Idx] (or perhaps Arr[Idx1][Idx2] etc.), return the parent memory block
308 /// Arr and the distance of Location from the beginning of Arr (expressed in a
309 /// NonLoc that specifies the number of CharUnits). Returns nullopt when these
310 /// cannot be determined.
311 std::optional<RegionRawOffsetV2>
312 RegionRawOffsetV2::computeOffset(ProgramStateRef State, SValBuilder &SVB,
313                                  SVal Location) {
314   QualType T = SVB.getArrayIndexType();
315   auto Calc = [&SVB, State, T](BinaryOperatorKind Op, NonLoc LHS, NonLoc RHS) {
316     // We will use this utility to add and multiply values.
317     return SVB.evalBinOpNN(State, Op, LHS, RHS, T).getAs<NonLoc>();
318   };
319 
320   const MemRegion *Region = Location.getAsRegion();
321   NonLoc Offset = SVB.makeZeroArrayIndex();
322 
323   while (Region) {
324     if (const auto *ERegion = dyn_cast<ElementRegion>(Region)) {
325       if (const auto Index = ERegion->getIndex().getAs<NonLoc>()) {
326         QualType ElemType = ERegion->getElementType();
327         // If the element is an incomplete type, go no further.
328         if (ElemType->isIncompleteType())
329           return std::nullopt;
330 
331         // Perform Offset += Index * sizeof(ElemType); then continue the offset
332         // calculations with SuperRegion:
333         NonLoc Size = SVB.makeArrayIndex(
334             SVB.getContext().getTypeSizeInChars(ElemType).getQuantity());
335         if (auto Delta = Calc(BO_Mul, *Index, Size)) {
336           if (auto NewOffset = Calc(BO_Add, Offset, *Delta)) {
337             Offset = *NewOffset;
338             Region = ERegion->getSuperRegion();
339             continue;
340           }
341         }
342       }
343     } else if (const auto *SRegion = dyn_cast<SubRegion>(Region)) {
344       // NOTE: The dyn_cast<>() is expected to succeed, it'd be very surprising
345       // to see a MemSpaceRegion at this point.
346       // FIXME: We may return with {<Region>, 0} even if we didn't handle any
347       // ElementRegion layers. I think that this behavior was introduced
348       // accidentally by 8a4c760c204546aba566e302f299f7ed2e00e287 in 2011, so
349       // it may be useful to review it in the future.
350       return RegionRawOffsetV2(SRegion, Offset);
351     }
352     return std::nullopt;
353   }
354   return std::nullopt;
355 }
356 
357 void ento::registerArrayBoundCheckerV2(CheckerManager &mgr) {
358   mgr.registerChecker<ArrayBoundCheckerV2>();
359 }
360 
361 bool ento::shouldRegisterArrayBoundCheckerV2(const CheckerManager &mgr) {
362   return true;
363 }
364