Coverage Report

Created: 2022-01-22 13:19

/Users/buildslave/jenkins/workspace/coverage/llvm-project/clang/include/clang/StaticAnalyzer/Core/PathSensitive/RangedConstraintManager.h
Line
Count
Source (jump to first uncovered line)
1
//== RangedConstraintManager.h ----------------------------------*- 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
//  Ranged constraint manager, built on SimpleConstraintManager.
10
//
11
//===----------------------------------------------------------------------===//
12
13
#ifndef LLVM_CLANG_STATICANALYZER_CORE_PATHSENSITIVE_RANGEDCONSTRAINTMANAGER_H
14
#define LLVM_CLANG_STATICANALYZER_CORE_PATHSENSITIVE_RANGEDCONSTRAINTMANAGER_H
15
16
#include "clang/StaticAnalyzer/Core/PathSensitive/ProgramState.h"
17
#include "clang/StaticAnalyzer/Core/PathSensitive/ProgramStateTrait.h"
18
#include "clang/StaticAnalyzer/Core/PathSensitive/SimpleConstraintManager.h"
19
#include "llvm/ADT/APSInt.h"
20
#include "llvm/Support/Allocator.h"
21
22
namespace clang {
23
24
namespace ento {
25
26
/// A Range represents the closed range [from, to].  The caller must
27
/// guarantee that from <= to.  Note that Range is immutable, so as not
28
/// to subvert RangeSet's immutability.
29
class Range {
30
public:
31
936k
  Range(const llvm::APSInt &From, const llvm::APSInt &To) : Impl(&From, &To) {
32
936k
    assert(From <= To);
33
936k
  }
34
35
187k
  Range(const llvm::APSInt &Point) : Range(Point, Point) {}
Unexecuted instantiation: clang::ento::Range::Range(llvm::APSInt const&)
clang::ento::Range::Range(llvm::APSInt const&)
Line
Count
Source
35
187k
  Range(const llvm::APSInt &Point) : Range(Point, Point) {}
36
37
102k
  bool Includes(const llvm::APSInt &Point) const {
38
102k
    return From() <= Point && 
Point <= To()102k
;
39
102k
  }
40
4.10M
  const llvm::APSInt &From() const { return *Impl.first; }
41
3.31M
  const llvm::APSInt &To() const { return *Impl.second; }
42
797k
  const llvm::APSInt *getConcreteValue() const {
43
797k
    return &From() == &To() ? 
&From()204k
:
nullptr592k
;
44
797k
  }
45
46
1.03M
  void Profile(llvm::FoldingSetNodeID &ID) const {
47
1.03M
    ID.AddPointer(&From());
48
1.03M
    ID.AddPointer(&To());
49
1.03M
  }
50
  void dump(raw_ostream &OS) const;
51
  void dump() const;
52
53
  // In order to keep non-overlapping ranges sorted, we can compare only From
54
  // points.
55
195k
  bool operator<(const Range &RHS) const { return From() < RHS.From(); }
56
57
655k
  bool operator==(const Range &RHS) const { return Impl == RHS.Impl; }
58
0
  bool operator!=(const Range &RHS) const { return !operator==(RHS); }
59
60
private:
61
  std::pair<const llvm::APSInt *, const llvm::APSInt *> Impl;
62
};
63
64
/// @class RangeSet is a persistent set of non-overlapping ranges.
65
///
66
/// New RangeSet objects can be ONLY produced by RangeSet::Factory object, which
67
/// also supports the most common operations performed on range sets.
68
///
69
/// Empty set corresponds to an overly constrained symbol meaning that there
70
/// are no possible values for that symbol.
71
class RangeSet {
72
public:
73
  class Factory;
74
75
private:
76
  // We use llvm::SmallVector as the underlying container for the following
77
  // reasons:
78
  //
79
  //   * Range sets are usually very simple, 1 or 2 ranges.
80
  //     That's why llvm::ImmutableSet is not perfect.
81
  //
82
  //   * Ranges in sets are NOT overlapping, so it is natural to keep them
83
  //     sorted for efficient operations and queries.  For this reason,
84
  //     llvm::SmallSet doesn't fit the requirements, it is not sorted when it
85
  //     is a vector.
86
  //
87
  //   * Range set operations usually a bit harder than add/remove a range.
88
  //     Complex operations might do many of those for just one range set.
89
  //     Formerly it used to be llvm::ImmutableSet, which is inefficient for our
90
  //     purposes as we want to make these operations BOTH immutable AND
91
  //     efficient.
92
  //
93
  //   * Iteration over ranges is widespread and a more cache-friendly
94
  //     structure is preferred.
95
  using ImplType = llvm::SmallVector<Range, 4>;
96
97
  struct ContainerType : public ImplType, public llvm::FoldingSetNode {
98
979k
    void Profile(llvm::FoldingSetNodeID &ID) const {
99
1.03M
      for (const Range &It : *this) {
100
1.03M
        It.Profile(ID);
101
1.03M
      }
102
979k
    }
103
  };
104
  // This is a non-owning pointer to an actual container.
105
  // The memory is fully managed by the factory and is alive as long as the
106
  // factory itself is alive.
107
  // It is a pointer as opposed to a reference, so we can easily reassign
108
  // RangeSet objects.
109
  using UnderlyingType = const ContainerType *;
110
  UnderlyingType Impl;
111
112
public:
113
  using const_iterator = ImplType::const_iterator;
114
115
2.13M
  const_iterator begin() const { return Impl->begin(); }
116
553k
  const_iterator end() const { return Impl->end(); }
117
407k
  size_t size() const { return Impl->size(); }
118
119
2.21M
  bool isEmpty() const { return Impl->empty(); }
120
121
  class Factory {
122
  public:
123
14.9k
    Factory(BasicValueFactory &BV) : ValueFactory(BV) {}
124
125
    /// Create a new set with all ranges from both LHS and RHS.
126
    /// Possible intersections are not checked here.
127
    ///
128
    /// Complexity: O(N + M)
129
    ///             where N = size(LHS), M = size(RHS)
130
    RangeSet add(RangeSet LHS, RangeSet RHS);
131
    /// Create a new set with all ranges from the original set plus the new one.
132
    /// Possible intersections are not checked here.
133
    ///
134
    /// Complexity: O(N)
135
    ///             where N = size(Original)
136
    RangeSet add(RangeSet Original, Range Element);
137
    /// Create a new set with all ranges from the original set plus the point.
138
    /// Possible intersections are not checked here.
139
    ///
140
    /// Complexity: O(N)
141
    ///             where N = size(Original)
142
    RangeSet add(RangeSet Original, const llvm::APSInt &Point);
143
    /// Create a new set which is a union of two given ranges.
144
    /// Possible intersections are not checked here.
145
    ///
146
    /// Complexity: O(N + M)
147
    ///             where N = size(LHS), M = size(RHS)
148
    RangeSet unite(RangeSet LHS, RangeSet RHS);
149
    /// Create a new set by uniting given range set with the given range.
150
    /// All intersections and adjacent ranges are handled here.
151
    ///
152
    /// Complexity: O(N)
153
    ///             where N = size(Original)
154
    RangeSet unite(RangeSet Original, Range Element);
155
    /// Create a new set by uniting given range set with the given point.
156
    /// All intersections and adjacent ranges are handled here.
157
    ///
158
    /// Complexity: O(N)
159
    ///             where N = size(Original)
160
    RangeSet unite(RangeSet Original, llvm::APSInt Point);
161
    /// Create a new set by uniting given range set with the given range
162
    /// between points. All intersections and adjacent ranges are handled here.
163
    ///
164
    /// Complexity: O(N)
165
    ///             where N = size(Original)
166
    RangeSet unite(RangeSet Original, llvm::APSInt From, llvm::APSInt To);
167
168
51.8k
    RangeSet getEmptySet() { return &EmptySet; }
169
170
    /// Create a new set with just one range.
171
    /// @{
172
    RangeSet getRangeSet(Range Origin);
173
317k
    RangeSet getRangeSet(const llvm::APSInt &From, const llvm::APSInt &To) {
174
317k
      return getRangeSet(Range(From, To));
175
317k
    }
176
77.9k
    RangeSet getRangeSet(const llvm::APSInt &Origin) {
177
77.9k
      return getRangeSet(Origin, Origin);
178
77.9k
    }
179
    /// @}
180
181
    /// Intersect the given range sets.
182
    ///
183
    /// Complexity: O(N + M)
184
    ///             where N = size(LHS), M = size(RHS)
185
    RangeSet intersect(RangeSet LHS, RangeSet RHS);
186
    /// Intersect the given set with the closed range [Lower, Upper].
187
    ///
188
    /// Unlike the Range type, this range uses modular arithmetic, corresponding
189
    /// to the common treatment of C integer overflow. Thus, if the Lower bound
190
    /// is greater than the Upper bound, the range is taken to wrap around. This
191
    /// is equivalent to taking the intersection with the two ranges [Min,
192
    /// Upper] and [Lower, Max], or, alternatively, /removing/ all integers
193
    /// between Upper and Lower.
194
    ///
195
    /// Complexity: O(N)
196
    ///             where N = size(What)
197
    RangeSet intersect(RangeSet What, llvm::APSInt Lower, llvm::APSInt Upper);
198
    /// Intersect the given range with the given point.
199
    ///
200
    /// The result can be either an empty set or a set containing the given
201
    /// point depending on whether the point is in the range set.
202
    ///
203
    /// Complexity: O(logN)
204
    ///             where N = size(What)
205
    RangeSet intersect(RangeSet What, llvm::APSInt Point);
206
207
    /// Delete the given point from the range set.
208
    ///
209
    /// Complexity: O(N)
210
    ///             where N = size(From)
211
    RangeSet deletePoint(RangeSet From, const llvm::APSInt &Point);
212
    /// Negate the given range set.
213
    ///
214
    /// Turn all [A, B] ranges to [-B, -A], when "-" is a C-like unary minus
215
    /// operation under the values of the type.
216
    ///
217
    /// We also handle MIN because applying unary minus to MIN does not change
218
    /// it.
219
    /// Example 1:
220
    /// char x = -128;        // -128 is a MIN value in a range of 'char'
221
    /// char y = -x;          // y: -128
222
    ///
223
    /// Example 2:
224
    /// unsigned char x = 0;  // 0 is a MIN value in a range of 'unsigned char'
225
    /// unsigned char y = -x; // y: 0
226
    ///
227
    /// And it makes us to separate the range
228
    /// like [MIN, N] to [MIN, MIN] U [-N, MAX].
229
    /// For instance, whole range is {-128..127} and subrange is [-128,-126],
230
    /// thus [-128,-127,-126,...] negates to [-128,...,126,127].
231
    ///
232
    /// Negate restores disrupted ranges on bounds,
233
    /// e.g. [MIN, B] => [MIN, MIN] U [-B, MAX] => [MIN, B].
234
    ///
235
    /// Negate is a self-inverse function, i.e. negate(negate(R)) == R.
236
    ///
237
    /// Complexity: O(N)
238
    ///             where N = size(What)
239
    RangeSet negate(RangeSet What);
240
241
    /// Return associated value factory.
242
182k
    BasicValueFactory &getValueFactory() const { return ValueFactory; }
243
244
  private:
245
    /// Return a persistent version of the given container.
246
    RangeSet makePersistent(ContainerType &&From);
247
    /// Construct a new persistent version of the given container.
248
    ContainerType *construct(ContainerType &&From);
249
250
    RangeSet intersect(const ContainerType &LHS, const ContainerType &RHS);
251
    /// NOTE: This function relies on the fact that all values in the
252
    /// containers are persistent (created via BasicValueFactory::getValue).
253
    ContainerType unite(const ContainerType &LHS, const ContainerType &RHS);
254
255
    // Many operations include producing new APSInt values and that's why
256
    // we need this factory.
257
    BasicValueFactory &ValueFactory;
258
    // Allocator for all the created containers.
259
    // Containers might own their own memory and that's why it is specific
260
    // for the type, so it calls container destructors upon deletion.
261
    llvm::SpecificBumpPtrAllocator<ContainerType> Arena;
262
    // Usually we deal with the same ranges and range sets over and over.
263
    // Here we track all created containers and try not to repeat ourselves.
264
    llvm::FoldingSet<ContainerType> Cache;
265
    static ContainerType EmptySet;
266
  };
267
268
  RangeSet(const RangeSet &) = default;
269
  RangeSet &operator=(const RangeSet &) = default;
270
  RangeSet(RangeSet &&) = default;
271
  RangeSet &operator=(RangeSet &&) = default;
272
  ~RangeSet() = default;
273
274
  /// Construct a new RangeSet representing '{ [From, To] }'.
275
  RangeSet(Factory &F, const llvm::APSInt &From, const llvm::APSInt &To)
276
239k
      : RangeSet(F.getRangeSet(From, To)) {}
Unexecuted instantiation: clang::ento::RangeSet::RangeSet(clang::ento::RangeSet::Factory&, llvm::APSInt const&, llvm::APSInt const&)
clang::ento::RangeSet::RangeSet(clang::ento::RangeSet::Factory&, llvm::APSInt const&, llvm::APSInt const&)
Line
Count
Source
276
239k
      : RangeSet(F.getRangeSet(From, To)) {}
277
278
  /// Construct a new RangeSet representing the given point as a range.
279
  RangeSet(Factory &F, const llvm::APSInt &Point)
280
40.1k
      : RangeSet(F.getRangeSet(Point)) {}
Unexecuted instantiation: clang::ento::RangeSet::RangeSet(clang::ento::RangeSet::Factory&, llvm::APSInt const&)
clang::ento::RangeSet::RangeSet(clang::ento::RangeSet::Factory&, llvm::APSInt const&)
Line
Count
Source
280
40.1k
      : RangeSet(F.getRangeSet(Point)) {}
281
282
587k
  static void Profile(llvm::FoldingSetNodeID &ID, const RangeSet &RS) {
283
587k
    ID.AddPointer(RS.Impl);
284
587k
  }
285
286
  /// Profile - Generates a hash profile of this RangeSet for use
287
  ///  by FoldingSet.
288
587k
  void Profile(llvm::FoldingSetNodeID &ID) const { Profile(ID, *this); }
289
290
  /// getConcreteValue - If a symbol is constrained to equal a specific integer
291
  ///  constant then this method returns that value.  Otherwise, it returns
292
  ///  NULL.
293
866k
  const llvm::APSInt *getConcreteValue() const {
294
866k
    return Impl->size() == 1 ? 
begin()->getConcreteValue()797k
:
nullptr69.0k
;
295
866k
  }
296
297
  /// Get the minimal value covered by the ranges in the set.
298
  ///
299
  /// Complexity: O(1)
300
  const llvm::APSInt &getMinValue() const;
301
  /// Get the maximal value covered by the ranges in the set.
302
  ///
303
  /// Complexity: O(1)
304
  const llvm::APSInt &getMaxValue() const;
305
306
  /// Test whether the given point is contained by any of the ranges.
307
  ///
308
  /// Complexity: O(logN)
309
  ///             where N = size(this)
310
112k
  bool contains(llvm::APSInt Point) const { return containsImpl(Point); }
311
312
3.88k
  bool containsZero() const {
313
3.88k
    APSIntType T{getMinValue()};
314
3.88k
    return contains(T.getZeroValue());
315
3.88k
  }
316
317
  /// Test if the range is the [0,0] range.
318
  ///
319
  /// Complexity: O(1)
320
161
  bool encodesFalseRange() const {
321
161
    const llvm::APSInt *Constant = getConcreteValue();
322
161
    return Constant && 
Constant->isZero()103
;
323
161
  }
324
325
  /// Test if the range doesn't contain zero.
326
  ///
327
  /// Complexity: O(logN)
328
  ///             where N = size(this)
329
123
  bool encodesTrueRange() const { return !containsZero(); }
330
331
  void dump(raw_ostream &OS) const;
332
  void dump() const;
333
334
640k
  bool operator==(const RangeSet &Other) const { return *Impl == *Other.Impl; }
335
0
  bool operator!=(const RangeSet &Other) const { return !(*this == Other); }
336
337
private:
338
541k
  /* implicit */ RangeSet(ContainerType *RawContainer) : Impl(RawContainer) {}
339
0
  /* implicit */ RangeSet(UnderlyingType Ptr) : Impl(Ptr) {}
340
341
  /// Pin given points to the type represented by the current range set.
342
  ///
343
  /// This makes parameter points to be in-out parameters.
344
  /// In order to maintain consistent types across all of the ranges in the set
345
  /// and to keep all the operations to compare ONLY points of the same type, we
346
  /// need to pin every point before any operation.
347
  ///
348
  /// @Returns true if the given points can be converted to the target type
349
  ///          without changing the values (i.e. trivially) and false otherwise.
350
  /// @{
351
  bool pin(llvm::APSInt &Lower, llvm::APSInt &Upper) const;
352
  bool pin(llvm::APSInt &Point) const;
353
  /// @}
354
355
  // This version of this function modifies its arguments (pins it).
356
  bool containsImpl(llvm::APSInt &Point) const;
357
358
  friend class Factory;
359
};
360
361
using ConstraintMap = llvm::ImmutableMap<SymbolRef, RangeSet>;
362
ConstraintMap getConstraintMap(ProgramStateRef State);
363
364
class RangedConstraintManager : public SimpleConstraintManager {
365
public:
366
  RangedConstraintManager(ExprEngine *EE, SValBuilder &SB)
367
14.8k
      : SimpleConstraintManager(EE, SB) {}
368
369
  ~RangedConstraintManager() override;
370
371
  //===------------------------------------------------------------------===//
372
  // Implementation for interface from SimpleConstraintManager.
373
  //===------------------------------------------------------------------===//
374
375
  ProgramStateRef assumeSym(ProgramStateRef State, SymbolRef Sym,
376
                            bool Assumption) override;
377
378
  ProgramStateRef assumeSymInclusiveRange(ProgramStateRef State, SymbolRef Sym,
379
                                          const llvm::APSInt &From,
380
                                          const llvm::APSInt &To,
381
                                          bool InRange) override;
382
383
  ProgramStateRef assumeSymUnsupported(ProgramStateRef State, SymbolRef Sym,
384
                                       bool Assumption) override;
385
386
protected:
387
  /// Assume a constraint between a symbolic expression and a concrete integer.
388
  virtual ProgramStateRef assumeSymRel(ProgramStateRef State, SymbolRef Sym,
389
                                       BinaryOperator::Opcode op,
390
                                       const llvm::APSInt &Int);
391
392
  //===------------------------------------------------------------------===//
393
  // Interface that subclasses must implement.
394
  //===------------------------------------------------------------------===//
395
396
  // Each of these is of the form "$Sym+Adj <> V", where "<>" is the comparison
397
  // operation for the method being invoked.
398
399
  virtual ProgramStateRef assumeSymNE(ProgramStateRef State, SymbolRef Sym,
400
                                      const llvm::APSInt &V,
401
                                      const llvm::APSInt &Adjustment) = 0;
402
403
  virtual ProgramStateRef assumeSymEQ(ProgramStateRef State, SymbolRef Sym,
404
                                      const llvm::APSInt &V,
405
                                      const llvm::APSInt &Adjustment) = 0;
406
407
  virtual ProgramStateRef assumeSymLT(ProgramStateRef State, SymbolRef Sym,
408
                                      const llvm::APSInt &V,
409
                                      const llvm::APSInt &Adjustment) = 0;
410
411
  virtual ProgramStateRef assumeSymGT(ProgramStateRef State, SymbolRef Sym,
412
                                      const llvm::APSInt &V,
413
                                      const llvm::APSInt &Adjustment) = 0;
414
415
  virtual ProgramStateRef assumeSymLE(ProgramStateRef State, SymbolRef Sym,
416
                                      const llvm::APSInt &V,
417
                                      const llvm::APSInt &Adjustment) = 0;
418
419
  virtual ProgramStateRef assumeSymGE(ProgramStateRef State, SymbolRef Sym,
420
                                      const llvm::APSInt &V,
421
                                      const llvm::APSInt &Adjustment) = 0;
422
423
  virtual ProgramStateRef assumeSymWithinInclusiveRange(
424
      ProgramStateRef State, SymbolRef Sym, const llvm::APSInt &From,
425
      const llvm::APSInt &To, const llvm::APSInt &Adjustment) = 0;
426
427
  virtual ProgramStateRef assumeSymOutsideInclusiveRange(
428
      ProgramStateRef State, SymbolRef Sym, const llvm::APSInt &From,
429
      const llvm::APSInt &To, const llvm::APSInt &Adjustment) = 0;
430
431
  //===------------------------------------------------------------------===//
432
  // Internal implementation.
433
  //===------------------------------------------------------------------===//
434
private:
435
  static void computeAdjustment(SymbolRef &Sym, llvm::APSInt &Adjustment);
436
};
437
438
/// Try to simplify a given symbolic expression based on the constraints in
439
/// State. This is needed because the Environment bindings are not getting
440
/// updated when a new constraint is added to the State. If the symbol is
441
/// simplified to a non-symbol (e.g. to a constant) then the original symbol
442
/// is returned. We use this function in the family of assumeSymNE/EQ/LT/../GE
443
/// functions where we can work only with symbols. Use the other function
444
/// (simplifyToSVal) if you are interested in a simplification that may yield
445
/// a concrete constant value.
446
SymbolRef simplify(ProgramStateRef State, SymbolRef Sym);
447
448
/// Try to simplify a given symbolic expression's associated `SVal` based on the
449
/// constraints in State. This is very similar to `simplify`, but this function
450
/// always returns the simplified SVal. The simplified SVal might be a single
451
/// constant (i.e. `ConcreteInt`).
452
SVal simplifyToSVal(ProgramStateRef State, SymbolRef Sym);
453
454
} // namespace ento
455
} // namespace clang
456
457
REGISTER_FACTORY_WITH_PROGRAMSTATE(ConstraintMap)
458
459
#endif