clang  6.0.0svn
ArrayBoundCheckerV2.cpp
Go to the documentation of this file.
1 //== ArrayBoundCheckerV2.cpp ------------------------------------*- C++ -*--==//
2 //
3 // The LLVM Compiler Infrastructure
4 //
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
7 //
8 //===----------------------------------------------------------------------===//
9 //
10 // This file defines ArrayBoundCheckerV2, which is a path-sensitive check
11 // which looks for an out-of-bound array element access.
12 //
13 //===----------------------------------------------------------------------===//
14 
15 #include "ClangSACheckers.h"
16 #include "clang/AST/CharUnits.h"
23 #include "llvm/ADT/SmallString.h"
24 #include "llvm/Support/raw_ostream.h"
25 
26 using namespace clang;
27 using namespace ento;
28 
29 namespace {
30 class ArrayBoundCheckerV2 :
31  public Checker<check::Location> {
32  mutable std::unique_ptr<BuiltinBug> BT;
33 
34  enum OOB_Kind { OOB_Precedes, OOB_Excedes, OOB_Tainted };
35 
36  void reportOOB(CheckerContext &C, ProgramStateRef errorState,
37  OOB_Kind kind) const;
38 
39 public:
40  void checkLocation(SVal l, bool isLoad, const Stmt*S,
41  CheckerContext &C) const;
42 };
43 
44 // FIXME: Eventually replace RegionRawOffset with this class.
45 class RegionRawOffsetV2 {
46 private:
47  const SubRegion *baseRegion;
48  SVal byteOffset;
49 
50  RegionRawOffsetV2()
51  : baseRegion(nullptr), byteOffset(UnknownVal()) {}
52 
53 public:
54  RegionRawOffsetV2(const SubRegion* base, SVal offset)
55  : baseRegion(base), byteOffset(offset) {}
56 
57  NonLoc getByteOffset() const { return byteOffset.castAs<NonLoc>(); }
58  const SubRegion *getRegion() const { return baseRegion; }
59 
60  static RegionRawOffsetV2 computeOffset(ProgramStateRef state,
61  SValBuilder &svalBuilder,
62  SVal location);
63 
64  void dump() const;
65  void dumpToStream(raw_ostream &os) const;
66 };
67 }
68 
69 static SVal computeExtentBegin(SValBuilder &svalBuilder,
70  const MemRegion *region) {
71  const MemSpaceRegion *SR = region->getMemorySpace();
72  if (SR->getKind() == MemRegion::UnknownSpaceRegionKind)
73  return UnknownVal();
74  else
75  return svalBuilder.makeZeroArrayIndex();
76 }
77 
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>
84  SValBuilder &svalBuilder) {
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()) {
91  case BO_Mul:
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);
96  else
97  return getSimplifiedOffsets(
98  nonloc::SymbolVal(SIE->getLHS()),
99  svalBuilder.makeIntVal(extent.getValue() / constant),
100  svalBuilder);
101  case BO_Add:
102  return getSimplifiedOffsets(
103  nonloc::SymbolVal(SIE->getLHS()),
104  svalBuilder.makeIntVal(extent.getValue() - constant), svalBuilder);
105  default:
106  break;
107  }
108  }
109  }
110 
111  return std::pair<NonLoc, nonloc::ConcreteInt>(offset, extent);
112 }
113 
114 void ArrayBoundCheckerV2::checkLocation(SVal location, bool isLoad,
115  const Stmt* LoadS,
116  CheckerContext &checkerContext) const {
117 
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.
122  //
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;
129 
130  SValBuilder &svalBuilder = checkerContext.getSValBuilder();
131  const RegionRawOffsetV2 &rawOffset =
132  RegionRawOffsetV2::computeOffset(state, svalBuilder, location);
133 
134  if (!rawOffset.getRegion())
135  return;
136 
137  NonLoc rawOffsetVal = rawOffset.getByteOffset();
138 
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.
142 
143  SVal extentBegin = computeExtentBegin(svalBuilder, rawOffset.getRegion());
144 
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>(),
150  svalBuilder);
151  rawOffsetVal = simplifiedOffsets.first;
152  *NV = simplifiedOffsets.second;
153  }
154 
155  SVal lowerBound = svalBuilder.evalBinOpNN(state, BO_LT, rawOffsetVal, *NV,
156  svalBuilder.getConditionType());
157 
158  Optional<NonLoc> lowerBoundToCheck = lowerBound.getAs<NonLoc>();
159  if (!lowerBoundToCheck)
160  return;
161 
162  ProgramStateRef state_precedesLowerBound, state_withinLowerBound;
163  std::tie(state_precedesLowerBound, state_withinLowerBound) =
164  state->assume(*lowerBoundToCheck);
165 
166  // Are we constrained enough to definitely precede the lower bound?
167  if (state_precedesLowerBound && !state_withinLowerBound) {
168  reportOOB(checkerContext, state_precedesLowerBound, OOB_Precedes);
169  return;
170  }
171 
172  // Otherwise, assume the constraint of the lower bound.
173  assert(state_withinLowerBound);
174  state = state_withinLowerBound;
175  }
176 
177  do {
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>())
183  break;
184 
185  if (extentVal.getAs<nonloc::ConcreteInt>()) {
186  std::pair<NonLoc, nonloc::ConcreteInt> simplifiedOffsets =
187  getSimplifiedOffsets(rawOffset.getByteOffset(),
188  extentVal.castAs<nonloc::ConcreteInt>(),
189  svalBuilder);
190  rawOffsetVal = simplifiedOffsets.first;
191  extentVal = simplifiedOffsets.second;
192  }
193 
194  SVal upperbound = svalBuilder.evalBinOpNN(state, BO_GE, rawOffsetVal,
195  extentVal.castAs<NonLoc>(),
196  svalBuilder.getConditionType());
197 
198  Optional<NonLoc> upperboundToCheck = upperbound.getAs<NonLoc>();
199  if (!upperboundToCheck)
200  break;
201 
202  ProgramStateRef state_exceedsUpperBound, state_withinUpperBound;
203  std::tie(state_exceedsUpperBound, state_withinUpperBound) =
204  state->assume(*upperboundToCheck);
205 
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);
210  return;
211  }
212  } else if (state_exceedsUpperBound) {
213  // If we are constrained enough to definitely exceed the upper bound,
214  // report.
215  assert(!state_withinUpperBound);
216  reportOOB(checkerContext, state_exceedsUpperBound, OOB_Excedes);
217  return;
218  }
219 
220  assert(state_withinUpperBound);
221  state = state_withinUpperBound;
222  }
223  while (false);
224 
225  if (state != originalState)
226  checkerContext.addTransition(state);
227 }
228 
229 void ArrayBoundCheckerV2::reportOOB(CheckerContext &checkerContext,
230  ProgramStateRef errorState,
231  OOB_Kind kind) const {
232 
233  ExplodedNode *errorNode = checkerContext.generateErrorNode(errorState);
234  if (!errorNode)
235  return;
236 
237  if (!BT)
238  BT.reset(new BuiltinBug(this, "Out-of-bound access"));
239 
240  // FIXME: This diagnostics are preliminary. We should get far better
241  // diagnostics for explaining buffer overruns.
242 
243  SmallString<256> buf;
244  llvm::raw_svector_ostream os(buf);
245  os << "Out of bound memory access ";
246  switch (kind) {
247  case OOB_Precedes:
248  os << "(accessed memory precedes memory block)";
249  break;
250  case OOB_Excedes:
251  os << "(access exceeds upper limit of memory block)";
252  break;
253  case OOB_Tainted:
254  os << "(index is tainted)";
255  break;
256  }
257 
258  checkerContext.emitReport(
259  llvm::make_unique<BugReport>(*BT, os.str(), errorNode));
260 }
261 
262 LLVM_DUMP_METHOD void RegionRawOffsetV2::dump() const {
263  dumpToStream(llvm::errs());
264 }
265 
266 void RegionRawOffsetV2::dumpToStream(raw_ostream &os) const {
267  os << "raw_offset_v2{" << getRegion() << ',' << getByteOffset() << '}';
268 }
269 
270 
271 // Lazily computes a value to be used by 'computeOffset'. If 'val'
272 // is unknown or undefined, we lazily substitute '0'. Otherwise,
273 // return 'val'.
274 static inline SVal getValue(SVal val, SValBuilder &svalBuilder) {
275  return val.getAs<UndefinedVal>() ? svalBuilder.makeArrayIndex(0) : val;
276 }
277 
278 // Scale a base value by a scaling factor, and return the scaled
279 // value as an SVal. Used by 'computeOffset'.
281  NonLoc baseVal, CharUnits scaling,
282  SValBuilder &sb) {
283  return sb.evalBinOpNN(state, BO_Mul, baseVal,
284  sb.makeArrayIndex(scaling.getQuantity()),
285  sb.getArrayIndexType());
286 }
287 
288 // Add an SVal to another, treating unknown and undefined values as
289 // summing to UnknownVal. Used by 'computeOffset'.
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())
295  return UnknownVal();
296 
297  return svalBuilder.evalBinOpNN(state, BO_Add, x.castAs<NonLoc>(),
298  y.castAs<NonLoc>(),
299  svalBuilder.getArrayIndexType());
300 }
301 
302 /// Compute a raw byte offset from a base region. Used for array bounds
303 /// checking.
304 RegionRawOffsetV2 RegionRawOffsetV2::computeOffset(ProgramStateRef state,
305  SValBuilder &svalBuilder,
306  SVal location)
307 {
308  const MemRegion *region = location.getAsRegion();
309  SVal offset = UndefinedVal();
310 
311  while (region) {
312  switch (region->getKind()) {
313  default: {
314  if (const SubRegion *subReg = dyn_cast<SubRegion>(region)) {
315  offset = getValue(offset, svalBuilder);
316  if (!offset.isUnknownOrUndef())
317  return RegionRawOffsetV2(subReg, offset);
318  }
319  return RegionRawOffsetV2();
320  }
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();
331 
332  // Update the offset.
333  offset = addValue(state,
334  getValue(offset, svalBuilder),
335  scaleValue(state,
336  index.castAs<NonLoc>(),
337  astContext.getTypeSizeInChars(elemType),
338  svalBuilder),
339  svalBuilder);
340 
341  if (offset.isUnknownOrUndef())
342  return RegionRawOffsetV2();
343 
344  region = elemReg->getSuperRegion();
345  continue;
346  }
347  }
348  }
349  return RegionRawOffsetV2();
350 }
351 
352 void ento::registerArrayBoundCheckerV2(CheckerManager &mgr) {
353  mgr.registerChecker<ArrayBoundCheckerV2>();
354 }
nonloc::ConcreteInt makeIntVal(const IntegerLiteral *integer)
Definition: SValBuilder.h:254
QualType getArrayIndexType() const
Definition: SValBuilder.h:140
A (possibly-)qualified type.
Definition: Type.h:614
MemRegion - The root abstract class for all memory regions.
Definition: MemRegion.h:79
static SVal addValue(ProgramStateRef state, SVal x, SVal y, SValBuilder &svalBuilder)
ExplodedNode * generateErrorNode(ProgramStateRef State=nullptr, const ProgramPointTag *Tag=nullptr)
Generate a transition to a node that will be used to report an error.
Stmt - This represents one statement.
Definition: Stmt.h:60
NonLoc getIndex() const
Definition: MemRegion.h:1085
ExplodedNode * addTransition(ProgramStateRef State=nullptr, const ProgramPointTag *Tag=nullptr)
Generates a new transition in the program state graph (ExplodedGraph).
MemSpaceRegion - A memory region that represents a "memory space"; for example, the set of global var...
Definition: MemRegion.h:179
Value representing integer constant.
Definition: SVals.h:352
NonLoc makeArrayIndex(uint64_t idx)
Definition: SValBuilder.h:248
QualType getElementType() const
Definition: MemRegion.h:1091
const MemRegion * getSuperRegion() const
Definition: MemRegion.h:430
Holds long-lived AST nodes (such as types and decls) that can be referred to throughout the semantic ...
Definition: ASTContext.h:128
i32 captured_struct **param SharedsTy A type which contains references the shared variables *param Shareds Context with the list of shared variables from the p *TaskFunction *param Data Additional data for task generation like final * state
CharUnits - This is an opaque type for sizes expressed in character units.
Definition: CharUnits.h:38
static void dump(llvm::raw_ostream &OS, StringRef FunctionName, ArrayRef< CounterExpression > Expressions, ArrayRef< CounterMappingRegion > Regions)
A record of the "type" of an APSInt, used for conversions.
Definition: APSIntType.h:20
Represents a symbolic expression like &#39;x&#39; + 3.
const MemSpaceRegion * getMemorySpace() const
Definition: MemRegion.cpp:1059
static SVal computeExtentBegin(SValBuilder &svalBuilder, const MemRegion *region)
QuantityType getQuantity() const
getQuantity - Get the raw integer representation of this quantity.
Definition: CharUnits.h:179
QualType getConditionType() const
Definition: SValBuilder.h:136
static SVal getValue(SVal val, SValBuilder &svalBuilder)
Optional< T > getAs() const
Convert to the specified SVal type, returning None if this SVal is not of the desired type...
Definition: SVals.h:100
void emitReport(std::unique_ptr< BugReport > R)
Emit the diagnostics report.
Kind getKind() const
Definition: MemRegion.h:148
CHECKER * registerChecker()
Used to register checkers.
static SVal scaleValue(ProgramStateRef state, NonLoc baseVal, CharUnits scaling, SValBuilder &sb)
const MemRegion * getAsRegion() const
Definition: SVals.cpp:140
ASTContext & getContext()
Definition: SValBuilder.h:131
SVal - This represents a symbolic expression, which can be either an L-value or an R-value...
Definition: SVals.h:63
virtual SVal evalBinOpNN(ProgramStateRef state, BinaryOperator::Opcode op, NonLoc lhs, NonLoc rhs, QualType resultTy)=0
Create a new value which represents a binary expression with two non- location operands.
Dataflow Directional Tag Classes.
Represents symbolic expression.
Definition: SVals.h:326
T castAs() const
Convert to the specified SVal type, asserting that this SVal is of the desired type.
Definition: SVals.h:92
SubRegion - A region that subsets another larger region.
Definition: MemRegion.h:419
const ProgramStateRef & getState() const
bool isIncompleteType(NamedDecl **Def=nullptr) const
Types are partitioned into 3 broad categories (C99 6.2.5p1): object types, function types...
Definition: Type.cpp:1931
unsigned kind
All of the diagnostics that can be emitted by the frontend.
Definition: DiagnosticIDs.h:59
SValBuilder & getSValBuilder()
ElementRegin is used to represent both array elements and casts.
Definition: MemRegion.h:1066
static std::pair< NonLoc, nonloc::ConcreteInt > getSimplifiedOffsets(NonLoc offset, nonloc::ConcreteInt extent, SValBuilder &svalBuilder)
CharUnits getTypeSizeInChars(QualType T) const
Return the size of the specified (complete) type T, in characters.
const llvm::APSInt & getValue() const
Definition: SVals.h:356
bool isUnknownOrUndef() const
Definition: SVals.h:136