1//= CStringChecker.cpp - Checks calls to C string functions --------*- 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 defines CStringChecker, which is an assortment of checks on calls
10// to functions in <string.h>.
11//
12//===----------------------------------------------------------------------===//
13
14#include "InterCheckerAPI.h"
15#include "clang/AST/OperationKinds.h"
16#include "clang/Basic/CharInfo.h"
17#include "clang/StaticAnalyzer/Checkers/BuiltinCheckerRegistration.h"
18#include "clang/StaticAnalyzer/Core/BugReporter/BugReporterVisitors.h"
19#include "clang/StaticAnalyzer/Core/BugReporter/BugType.h"
20#include "clang/StaticAnalyzer/Core/Checker.h"
21#include "clang/StaticAnalyzer/Core/CheckerManager.h"
22#include "clang/StaticAnalyzer/Core/PathSensitive/CallDescription.h"
23#include "clang/StaticAnalyzer/Core/PathSensitive/CallEvent.h"
24#include "clang/StaticAnalyzer/Core/PathSensitive/CheckerContext.h"
25#include "clang/StaticAnalyzer/Core/PathSensitive/DynamicExtent.h"
26#include "clang/StaticAnalyzer/Core/PathSensitive/MemRegion.h"
27#include "clang/StaticAnalyzer/Core/PathSensitive/ProgramStateTrait.h"
28#include "clang/StaticAnalyzer/Core/PathSensitive/SVals.h"
29#include "llvm/ADT/APSInt.h"
30#include "llvm/ADT/STLExtras.h"
31#include "llvm/ADT/StringExtras.h"
32#include "llvm/Support/raw_ostream.h"
33#include <functional>
34#include <optional>
35
36using namespace clang;
37using namespace ento;
38using namespace std::placeholders;
39
40namespace {
41struct AnyArgExpr {
42 const Expr *Expression;
43 unsigned ArgumentIndex;
44};
45struct SourceArgExpr : AnyArgExpr {};
46struct DestinationArgExpr : AnyArgExpr {};
47struct SizeArgExpr : AnyArgExpr {};
48
49using ErrorMessage = SmallString<128>;
50enum class AccessKind { write, read };
51
52static ErrorMessage createOutOfBoundErrorMsg(StringRef FunctionDescription,
53 AccessKind Access) {
54 ErrorMessage Message;
55 llvm::raw_svector_ostream Os(Message);
56
57 // Function classification like: Memory copy function
58 Os << toUppercase(c: FunctionDescription.front())
59 << &FunctionDescription.data()[1];
60
61 if (Access == AccessKind::write) {
62 Os << " overflows the destination buffer";
63 } else { // read access
64 Os << " accesses out-of-bound array element";
65 }
66
67 return Message;
68}
69
70enum class ConcatFnKind { none = 0, strcat = 1, strlcat = 2 };
71
72enum class CharKind { Regular = 0, Wide };
73constexpr CharKind CK_Regular = CharKind::Regular;
74constexpr CharKind CK_Wide = CharKind::Wide;
75
76static QualType getCharPtrType(ASTContext &Ctx, CharKind CK) {
77 return Ctx.getPointerType(T: CK == CharKind::Regular ? Ctx.CharTy
78 : Ctx.WideCharTy);
79}
80
81class CStringChecker
82 : public CheckerFamily<eval::Call, check::PreStmt<DeclStmt>,
83 check::LiveSymbols, check::DeadSymbols,
84 check::RegionChanges> {
85 mutable const char *CurrentFunctionDescription = nullptr;
86
87public:
88 // FIXME: The bug types emitted by this checker family have confused garbage
89 // in their Description and Category fields (e.g. `categories::UnixAPI` is
90 // passed as the description in several cases and `uninitialized` is mistyped
91 // as `unitialized`). This should be cleaned up.
92 CheckerFrontendWithBugType NullArg{categories::UnixAPI};
93 CheckerFrontendWithBugType OutOfBounds{"Out-of-bound array access"};
94 CheckerFrontendWithBugType BufferOverlap{categories::UnixAPI,
95 "Improper arguments"};
96 CheckerFrontendWithBugType NotNullTerm{categories::UnixAPI};
97 CheckerFrontendWithBugType UninitializedRead{
98 "Accessing unitialized/garbage values"};
99
100 StringRef getDebugTag() const override { return "MallocChecker"; }
101
102 static void *getTag() { static int tag; return &tag; }
103
104 bool evalCall(const CallEvent &Call, CheckerContext &C) const;
105 void checkPreStmt(const DeclStmt *DS, CheckerContext &C) const;
106 void checkLiveSymbols(ProgramStateRef state, SymbolReaper &SR) const;
107 void checkDeadSymbols(SymbolReaper &SR, CheckerContext &C) const;
108
109 ProgramStateRef
110 checkRegionChanges(ProgramStateRef state,
111 const InvalidatedSymbols *,
112 ArrayRef<const MemRegion *> ExplicitRegions,
113 ArrayRef<const MemRegion *> Regions,
114 const LocationContext *LCtx,
115 const CallEvent *Call) const;
116
117 using FnCheck = std::function<void(const CStringChecker *, CheckerContext &,
118 const CallEvent &)>;
119
120 CallDescriptionMap<FnCheck> Callbacks = {
121 {{CDM::CLibraryMaybeHardened, {"memcpy"}, 3},
122 std::bind(f: &CStringChecker::evalMemcpy, args: _1, args: _2, args: _3, args: CK_Regular)},
123 {{CDM::CLibraryMaybeHardened, {"wmemcpy"}, 3},
124 std::bind(f: &CStringChecker::evalMemcpy, args: _1, args: _2, args: _3, args: CK_Wide)},
125 {{CDM::CLibraryMaybeHardened, {"mempcpy"}, 3},
126 std::bind(f: &CStringChecker::evalMempcpy, args: _1, args: _2, args: _3, args: CK_Regular)},
127 {{CDM::CLibraryMaybeHardened, {"wmempcpy"}, 3},
128 std::bind(f: &CStringChecker::evalMempcpy, args: _1, args: _2, args: _3, args: CK_Wide)},
129 {{CDM::CLibrary, {"memcmp"}, 3},
130 std::bind(f: &CStringChecker::evalMemcmp, args: _1, args: _2, args: _3, args: CK_Regular)},
131 {{CDM::CLibrary, {"wmemcmp"}, 3},
132 std::bind(f: &CStringChecker::evalMemcmp, args: _1, args: _2, args: _3, args: CK_Wide)},
133 {{CDM::CLibraryMaybeHardened, {"memmove"}, 3},
134 std::bind(f: &CStringChecker::evalMemmove, args: _1, args: _2, args: _3, args: CK_Regular)},
135 {{CDM::CLibraryMaybeHardened, {"wmemmove"}, 3},
136 std::bind(f: &CStringChecker::evalMemmove, args: _1, args: _2, args: _3, args: CK_Wide)},
137 {{CDM::CLibraryMaybeHardened, {"memset"}, 3},
138 &CStringChecker::evalMemset},
139 {{CDM::CLibrary, {"explicit_memset"}, 3}, &CStringChecker::evalMemset},
140 // FIXME: C23 introduces 'memset_explicit', maybe also model that
141 {{CDM::CLibraryMaybeHardened, {"strcpy"}, 2},
142 &CStringChecker::evalStrcpy},
143 {{CDM::CLibraryMaybeHardened, {"strncpy"}, 3},
144 &CStringChecker::evalStrncpy},
145 {{CDM::CLibraryMaybeHardened, {"stpcpy"}, 2},
146 &CStringChecker::evalStpcpy},
147 {{CDM::CLibraryMaybeHardened, {"strlcpy"}, 3},
148 &CStringChecker::evalStrlcpy},
149 {{CDM::CLibraryMaybeHardened, {"strcat"}, 2},
150 &CStringChecker::evalStrcat},
151 {{CDM::CLibraryMaybeHardened, {"strncat"}, 3},
152 &CStringChecker::evalStrncat},
153 {{CDM::CLibraryMaybeHardened, {"strlcat"}, 3},
154 &CStringChecker::evalStrlcat},
155 {{CDM::CLibraryMaybeHardened, {"strlen"}, 1},
156 &CStringChecker::evalstrLength},
157 {{CDM::CLibrary, {"wcslen"}, 1}, &CStringChecker::evalstrLength},
158 {{CDM::CLibraryMaybeHardened, {"strnlen"}, 2},
159 &CStringChecker::evalstrnLength},
160 {{CDM::CLibrary, {"wcsnlen"}, 2}, &CStringChecker::evalstrnLength},
161 {{CDM::CLibrary, {"strcmp"}, 2}, &CStringChecker::evalStrcmp},
162 {{CDM::CLibrary, {"strncmp"}, 3}, &CStringChecker::evalStrncmp},
163 {{CDM::CLibrary, {"strcasecmp"}, 2}, &CStringChecker::evalStrcasecmp},
164 {{CDM::CLibrary, {"strncasecmp"}, 3}, &CStringChecker::evalStrncasecmp},
165 {{CDM::CLibrary, {"strsep"}, 2}, &CStringChecker::evalStrsep},
166 {{CDM::CLibrary, {"strxfrm"}, 3}, &CStringChecker::evalStrxfrm},
167 {{CDM::CLibrary, {"bcopy"}, 3}, &CStringChecker::evalBcopy},
168 {{CDM::CLibrary, {"bcmp"}, 3},
169 std::bind(f: &CStringChecker::evalMemcmp, args: _1, args: _2, args: _3, args: CK_Regular)},
170 {{CDM::CLibrary, {"bzero"}, 2}, &CStringChecker::evalBzero},
171 {{CDM::CLibraryMaybeHardened, {"explicit_bzero"}, 2},
172 &CStringChecker::evalBzero},
173
174 // When recognizing calls to the following variadic functions, we accept
175 // any number of arguments in the call (std::nullopt = accept any
176 // number), but check that in the declaration there are 2 and 3
177 // parameters respectively. (Note that the parameter count does not
178 // include the "...". Calls where the number of arguments is too small
179 // will be discarded by the callback.)
180 {{CDM::CLibraryMaybeHardened, {"sprintf"}, std::nullopt, 2},
181 &CStringChecker::evalSprintf},
182 {{CDM::CLibraryMaybeHardened, {"snprintf"}, std::nullopt, 3},
183 &CStringChecker::evalSnprintf},
184 };
185
186 // These require a bit of special handling.
187 CallDescription StdCopy{CDM::SimpleFunc, {"std", "copy"}, 3},
188 StdCopyBackward{CDM::SimpleFunc, {"std", "copy_backward"}, 3};
189
190 FnCheck identifyCall(const CallEvent &Call, CheckerContext &C) const;
191 void evalMemcpy(CheckerContext &C, const CallEvent &Call, CharKind CK) const;
192 void evalMempcpy(CheckerContext &C, const CallEvent &Call, CharKind CK) const;
193 void evalMemmove(CheckerContext &C, const CallEvent &Call, CharKind CK) const;
194 void evalBcopy(CheckerContext &C, const CallEvent &Call) const;
195 void evalCopyCommon(CheckerContext &C, const CallEvent &Call,
196 ProgramStateRef state, SizeArgExpr Size,
197 DestinationArgExpr Dest, SourceArgExpr Source,
198 bool Restricted, bool IsMempcpy, CharKind CK) const;
199
200 void evalMemcmp(CheckerContext &C, const CallEvent &Call, CharKind CK) const;
201
202 void evalstrLength(CheckerContext &C, const CallEvent &Call) const;
203 void evalstrnLength(CheckerContext &C, const CallEvent &Call) const;
204 void evalstrLengthCommon(CheckerContext &C, const CallEvent &Call,
205 bool IsStrnlen = false) const;
206
207 void evalStrcpy(CheckerContext &C, const CallEvent &Call) const;
208 void evalStrncpy(CheckerContext &C, const CallEvent &Call) const;
209 void evalStpcpy(CheckerContext &C, const CallEvent &Call) const;
210 void evalStrlcpy(CheckerContext &C, const CallEvent &Call) const;
211 void evalStrcpyCommon(CheckerContext &C, const CallEvent &Call,
212 bool ReturnEnd, bool IsBounded, ConcatFnKind appendK,
213 bool returnPtr = true) const;
214
215 void evalStrxfrm(CheckerContext &C, const CallEvent &Call) const;
216
217 void evalStrcat(CheckerContext &C, const CallEvent &Call) const;
218 void evalStrncat(CheckerContext &C, const CallEvent &Call) const;
219 void evalStrlcat(CheckerContext &C, const CallEvent &Call) const;
220
221 void evalStrcmp(CheckerContext &C, const CallEvent &Call) const;
222 void evalStrncmp(CheckerContext &C, const CallEvent &Call) const;
223 void evalStrcasecmp(CheckerContext &C, const CallEvent &Call) const;
224 void evalStrncasecmp(CheckerContext &C, const CallEvent &Call) const;
225 void evalStrcmpCommon(CheckerContext &C, const CallEvent &Call,
226 bool IsBounded = false, bool IgnoreCase = false) const;
227
228 void evalStrsep(CheckerContext &C, const CallEvent &Call) const;
229
230 void evalStdCopy(CheckerContext &C, const CallEvent &Call) const;
231 void evalStdCopyBackward(CheckerContext &C, const CallEvent &Call) const;
232 void evalStdCopyCommon(CheckerContext &C, const CallEvent &Call) const;
233 void evalMemset(CheckerContext &C, const CallEvent &Call) const;
234 void evalBzero(CheckerContext &C, const CallEvent &Call) const;
235
236 void evalSprintf(CheckerContext &C, const CallEvent &Call) const;
237 void evalSnprintf(CheckerContext &C, const CallEvent &Call) const;
238 void evalSprintfCommon(CheckerContext &C, const CallEvent &Call,
239 bool IsBounded) const;
240
241 // Utility methods
242 std::pair<ProgramStateRef , ProgramStateRef >
243 static assumeZero(CheckerContext &C,
244 ProgramStateRef state, SVal V, QualType Ty);
245
246 static ProgramStateRef setCStringLength(ProgramStateRef state,
247 const MemRegion *MR,
248 SVal strLength);
249 static SVal getCStringLengthForRegion(CheckerContext &C,
250 ProgramStateRef &state,
251 const Expr *Ex,
252 const MemRegion *MR,
253 bool hypothetical);
254 static const StringLiteral *getStringLiteralFromRegion(const MemRegion *MR);
255
256 SVal getCStringLength(CheckerContext &C,
257 ProgramStateRef &state,
258 const Expr *Ex,
259 SVal Buf,
260 bool hypothetical = false) const;
261
262 const StringLiteral *getCStringLiteral(CheckerContext &C,
263 ProgramStateRef &state,
264 const Expr *expr,
265 SVal val) const;
266
267 /// Invalidate the destination buffer determined by characters copied.
268 static ProgramStateRef
269 invalidateDestinationBufferBySize(CheckerContext &C, ProgramStateRef S,
270 const Expr *BufE, ConstCFGElementRef Elem,
271 SVal BufV, SVal SizeV, QualType SizeTy);
272
273 /// Operation never overflows, do not invalidate the super region.
274 static ProgramStateRef invalidateDestinationBufferNeverOverflows(
275 CheckerContext &C, ProgramStateRef S, ConstCFGElementRef Elem, SVal BufV);
276
277 /// We do not know whether the operation can overflow (e.g. size is unknown),
278 /// invalidate the super region and escape related pointers.
279 static ProgramStateRef invalidateDestinationBufferAlwaysEscapeSuperRegion(
280 CheckerContext &C, ProgramStateRef S, ConstCFGElementRef Elem, SVal BufV);
281
282 /// Invalidate the source buffer for escaping pointers.
283 static ProgramStateRef invalidateSourceBuffer(CheckerContext &C,
284 ProgramStateRef S,
285 ConstCFGElementRef Elem,
286 SVal BufV);
287
288 /// @param InvalidationTraitOperations Determine how to invlidate the
289 /// MemRegion by setting the invalidation traits. Return true to cause pointer
290 /// escape, or false otherwise.
291 static ProgramStateRef invalidateBufferAux(
292 CheckerContext &C, ProgramStateRef State, ConstCFGElementRef Elem, SVal V,
293 llvm::function_ref<bool(RegionAndSymbolInvalidationTraits &,
294 const MemRegion *)>
295 InvalidationTraitOperations);
296
297 static bool SummarizeRegion(raw_ostream &os, ASTContext &Ctx,
298 const MemRegion *MR);
299
300 static bool memsetAux(const Expr *DstBuffer, ConstCFGElementRef Elem,
301 SVal CharE, const Expr *Size, CheckerContext &C,
302 ProgramStateRef &State);
303
304 // Re-usable checks
305 ProgramStateRef checkNonNull(CheckerContext &C, ProgramStateRef State,
306 AnyArgExpr Arg, SVal l) const;
307 // Check whether the origin region behind \p Element (like the actual array
308 // region \p Element is from) is initialized.
309 ProgramStateRef checkInit(CheckerContext &C, ProgramStateRef state,
310 AnyArgExpr Buffer, SVal Element, SVal Size) const;
311 ProgramStateRef CheckLocation(CheckerContext &C, ProgramStateRef state,
312 AnyArgExpr Buffer, SVal Element,
313 AccessKind Access,
314 CharKind CK = CharKind::Regular) const;
315 ProgramStateRef CheckBufferAccess(CheckerContext &C, ProgramStateRef State,
316 AnyArgExpr Buffer, SizeArgExpr Size,
317 AccessKind Access,
318 CharKind CK = CharKind::Regular) const;
319 ProgramStateRef CheckOverlap(CheckerContext &C, ProgramStateRef state,
320 SizeArgExpr Size, AnyArgExpr First,
321 AnyArgExpr Second,
322 CharKind CK = CharKind::Regular) const;
323 void emitOverlapBug(CheckerContext &C,
324 ProgramStateRef state,
325 const Stmt *First,
326 const Stmt *Second) const;
327
328 void emitNullArgBug(CheckerContext &C, ProgramStateRef State, const Stmt *S,
329 StringRef WarningMsg) const;
330 void emitOutOfBoundsBug(CheckerContext &C, ProgramStateRef State,
331 const Stmt *S, StringRef WarningMsg) const;
332 void emitNotCStringBug(CheckerContext &C, ProgramStateRef State,
333 const Stmt *S, StringRef WarningMsg) const;
334 void emitUninitializedReadBug(CheckerContext &C, ProgramStateRef State,
335 const Expr *E, const MemRegion *R,
336 StringRef Msg) const;
337 ProgramStateRef checkAdditionOverflow(CheckerContext &C,
338 ProgramStateRef state,
339 NonLoc left,
340 NonLoc right) const;
341
342 // Return true if the destination buffer of the copy function may be in bound.
343 // Expects SVal of Size to be positive and unsigned.
344 // Expects SVal of FirstBuf to be a FieldRegion.
345 static bool isFirstBufInBound(CheckerContext &C, ProgramStateRef State,
346 SVal BufVal, QualType BufTy, SVal LengthVal,
347 QualType LengthTy);
348};
349
350} //end anonymous namespace
351
352REGISTER_MAP_WITH_PROGRAMSTATE(CStringLength, const MemRegion *, SVal)
353
354//===----------------------------------------------------------------------===//
355// Individual checks and utility methods.
356//===----------------------------------------------------------------------===//
357
358std::pair<ProgramStateRef, ProgramStateRef>
359CStringChecker::assumeZero(CheckerContext &C, ProgramStateRef State, SVal V,
360 QualType Ty) {
361 std::optional<DefinedSVal> val = V.getAs<DefinedSVal>();
362 if (!val)
363 return std::pair<ProgramStateRef, ProgramStateRef>(State, State);
364
365 SValBuilder &svalBuilder = C.getSValBuilder();
366 DefinedOrUnknownSVal zero = svalBuilder.makeZeroVal(type: Ty);
367 return State->assume(Cond: svalBuilder.evalEQ(state: State, lhs: *val, rhs: zero));
368}
369
370ProgramStateRef CStringChecker::checkNonNull(CheckerContext &C,
371 ProgramStateRef State,
372 AnyArgExpr Arg, SVal l) const {
373 // If a previous check has failed, propagate the failure.
374 if (!State)
375 return nullptr;
376
377 ProgramStateRef stateNull, stateNonNull;
378 std::tie(args&: stateNull, args&: stateNonNull) =
379 assumeZero(C, State, V: l, Ty: Arg.Expression->getType());
380
381 if (stateNull && !stateNonNull) {
382 if (NullArg.isEnabled()) {
383 SmallString<80> buf;
384 llvm::raw_svector_ostream OS(buf);
385 assert(CurrentFunctionDescription);
386 OS << "Null pointer passed as " << (Arg.ArgumentIndex + 1)
387 << llvm::getOrdinalSuffix(Val: Arg.ArgumentIndex + 1) << " argument to "
388 << CurrentFunctionDescription;
389
390 emitNullArgBug(C, State: stateNull, S: Arg.Expression, WarningMsg: OS.str());
391 }
392 return nullptr;
393 }
394
395 // From here on, assume that the value is non-null.
396 assert(stateNonNull);
397 return stateNonNull;
398}
399
400static std::optional<NonLoc> getIndex(ProgramStateRef State,
401 const ElementRegion *ER, CharKind CK) {
402 SValBuilder &SVB = State->getStateManager().getSValBuilder();
403 ASTContext &Ctx = SVB.getContext();
404
405 if (CK == CharKind::Regular) {
406 if (ER->getValueType() != Ctx.CharTy)
407 return {};
408 return ER->getIndex();
409 }
410
411 if (ER->getValueType() != Ctx.WideCharTy)
412 return {};
413
414 QualType SizeTy = Ctx.getSizeType();
415 NonLoc WideSize =
416 SVB.makeIntVal(integer: Ctx.getTypeSizeInChars(T: Ctx.WideCharTy).getQuantity(),
417 type: SizeTy)
418 .castAs<NonLoc>();
419 SVal Offset =
420 SVB.evalBinOpNN(state: State, op: BO_Mul, lhs: ER->getIndex(), rhs: WideSize, resultTy: SizeTy);
421 if (Offset.isUnknown())
422 return {};
423 return Offset.castAs<NonLoc>();
424}
425
426// Basically 1 -> 1st, 12 -> 12th, etc.
427static void printIdxWithOrdinalSuffix(llvm::raw_ostream &Os, unsigned Idx) {
428 Os << Idx << llvm::getOrdinalSuffix(Val: Idx);
429}
430
431ProgramStateRef CStringChecker::checkInit(CheckerContext &C,
432 ProgramStateRef State,
433 AnyArgExpr Buffer, SVal Element,
434 SVal Size) const {
435
436 // If a previous check has failed, propagate the failure.
437 if (!State)
438 return nullptr;
439
440 const MemRegion *R = Element.getAsRegion();
441 const auto *ER = dyn_cast_or_null<ElementRegion>(Val: R);
442 if (!ER)
443 return State;
444
445 const auto *SuperR = ER->getSuperRegion()->getAs<TypedValueRegion>();
446 if (!SuperR)
447 return State;
448
449 // FIXME: We ought to able to check objects as well. Maybe
450 // UninitializedObjectChecker could help?
451 if (!SuperR->getValueType()->isArrayType())
452 return State;
453
454 SValBuilder &SVB = C.getSValBuilder();
455 ASTContext &Ctx = SVB.getContext();
456
457 const QualType ElemTy = Ctx.getBaseElementType(QT: SuperR->getValueType());
458 const NonLoc Zero = SVB.makeZeroArrayIndex();
459
460 std::optional<Loc> FirstElementVal =
461 State->getLValue(ElementType: ElemTy, Idx: Zero, Base: loc::MemRegionVal(SuperR)).getAs<Loc>();
462 if (!FirstElementVal)
463 return State;
464
465 // Ensure that we wouldn't read uninitialized value.
466 if (UninitializedRead.isEnabled() &&
467 State->getSVal(LV: *FirstElementVal).isUndef()) {
468 llvm::SmallString<258> Buf;
469 llvm::raw_svector_ostream OS(Buf);
470 OS << "The first element of the ";
471 printIdxWithOrdinalSuffix(Os&: OS, Idx: Buffer.ArgumentIndex + 1);
472 OS << " argument is undefined";
473 emitUninitializedReadBug(C, State, E: Buffer.Expression,
474 R: FirstElementVal->getAsRegion(), Msg: OS.str());
475 return nullptr;
476 }
477
478 // We won't check whether the entire region is fully initialized -- lets just
479 // check that the first and the last element is. So, onto checking the last
480 // element:
481 const QualType IdxTy = SVB.getArrayIndexType();
482
483 NonLoc ElemSize =
484 SVB.makeIntVal(integer: Ctx.getTypeSizeInChars(T: ElemTy).getQuantity(), type: IdxTy)
485 .castAs<NonLoc>();
486
487 // FIXME: Check that the size arg to the cstring function is divisible by
488 // size of the actual element type?
489
490 // The type of the argument to the cstring function is either char or wchar,
491 // but thats not the type of the original array (or memory region).
492 // Suppose the following:
493 // int t[5];
494 // memcpy(dst, t, sizeof(t) / sizeof(t[0]));
495 // When checking whether t is fully initialized, we see it as char array of
496 // size sizeof(int)*5. If we check the last element as a character, we read
497 // the last byte of an integer, which will be undefined. But just because
498 // that value is undefined, it doesn't mean that the element is uninitialized!
499 // For this reason, we need to retrieve the actual last element with the
500 // correct type.
501
502 // Divide the size argument to the cstring function by the actual element
503 // type. This value will be size of the array, or the index to the
504 // past-the-end element.
505 std::optional<NonLoc> Offset =
506 SVB.evalBinOpNN(state: State, op: clang::BO_Div, lhs: Size.castAs<NonLoc>(), rhs: ElemSize,
507 resultTy: IdxTy)
508 .getAs<NonLoc>();
509
510 // Retrieve the index of the last element.
511 const NonLoc One = SVB.makeIntVal(integer: 1, type: IdxTy).castAs<NonLoc>();
512 SVal LastIdx = SVB.evalBinOpNN(state: State, op: BO_Sub, lhs: *Offset, rhs: One, resultTy: IdxTy);
513
514 if (!Offset)
515 return State;
516
517 SVal LastElementVal =
518 State->getLValue(ElementType: ElemTy, Idx: LastIdx, Base: loc::MemRegionVal(SuperR));
519 if (!isa<Loc>(Val: LastElementVal))
520 return State;
521
522 if (UninitializedRead.isEnabled() &&
523 State->getSVal(LV: LastElementVal.castAs<Loc>()).isUndef()) {
524 const llvm::APSInt *IdxInt = LastIdx.getAsInteger();
525 // If we can't get emit a sensible last element index, just bail out --
526 // prefer to emit nothing in favour of emitting garbage quality reports.
527 if (!IdxInt) {
528 C.addSink();
529 return nullptr;
530 }
531 llvm::SmallString<258> Buf;
532 llvm::raw_svector_ostream OS(Buf);
533 OS << "The last accessed element (at index ";
534 OS << IdxInt->getExtValue();
535 OS << ") in the ";
536 printIdxWithOrdinalSuffix(Os&: OS, Idx: Buffer.ArgumentIndex + 1);
537 OS << " argument is undefined";
538 emitUninitializedReadBug(C, State, E: Buffer.Expression,
539 R: LastElementVal.getAsRegion(), Msg: OS.str());
540 return nullptr;
541 }
542 return State;
543}
544// FIXME: The root of this logic was copied from the old checker
545// alpha.security.ArrayBound (which is removed within this commit).
546// It should be refactored to use the different, more sophisticated bounds
547// checking logic used by the new checker ``security.ArrayBound``.
548ProgramStateRef CStringChecker::CheckLocation(CheckerContext &C,
549 ProgramStateRef state,
550 AnyArgExpr Buffer, SVal Element,
551 AccessKind Access,
552 CharKind CK) const {
553
554 // If a previous check has failed, propagate the failure.
555 if (!state)
556 return nullptr;
557
558 // Check for out of bound array element access.
559 const MemRegion *R = Element.getAsRegion();
560 if (!R)
561 return state;
562
563 const auto *ER = dyn_cast<ElementRegion>(Val: R);
564 if (!ER)
565 return state;
566
567 // Get the index of the accessed element.
568 std::optional<NonLoc> Idx = getIndex(State: state, ER, CK);
569 if (!Idx)
570 return state;
571
572 // Get the size of the array.
573 const auto *superReg = cast<SubRegion>(Val: ER->getSuperRegion());
574 DefinedOrUnknownSVal Size =
575 getDynamicExtent(State: state, MR: superReg, SVB&: C.getSValBuilder());
576
577 auto [StInBound, StOutBound] = state->assumeInBoundDual(idx: *Idx, upperBound: Size);
578 if (StOutBound && !StInBound) {
579 if (!OutOfBounds.isEnabled())
580 return nullptr;
581
582 ErrorMessage Message =
583 createOutOfBoundErrorMsg(FunctionDescription: CurrentFunctionDescription, Access);
584 emitOutOfBoundsBug(C, State: StOutBound, S: Buffer.Expression, WarningMsg: Message);
585 return nullptr;
586 }
587
588 // Array bound check succeeded. From this point forward the array bound
589 // should always succeed.
590 return StInBound;
591}
592
593ProgramStateRef
594CStringChecker::CheckBufferAccess(CheckerContext &C, ProgramStateRef State,
595 AnyArgExpr Buffer, SizeArgExpr Size,
596 AccessKind Access, CharKind CK) const {
597 // If a previous check has failed, propagate the failure.
598 if (!State)
599 return nullptr;
600
601 SValBuilder &svalBuilder = C.getSValBuilder();
602 ASTContext &Ctx = svalBuilder.getContext();
603
604 QualType SizeTy = Size.Expression->getType();
605 QualType PtrTy = getCharPtrType(Ctx, CK);
606
607 // Check that the first buffer is non-null.
608 SVal BufVal = C.getSVal(S: Buffer.Expression);
609 State = checkNonNull(C, State, Arg: Buffer, l: BufVal);
610 if (!State)
611 return nullptr;
612
613 // If out-of-bounds checking is turned off, skip the rest.
614 if (!OutOfBounds.isEnabled())
615 return State;
616
617 SVal BufStart =
618 svalBuilder.evalCast(V: BufVal, CastTy: PtrTy, OriginalTy: Buffer.Expression->getType());
619
620 // Check if the first byte of the buffer is accessible.
621 State = CheckLocation(C, state: State, Buffer, Element: BufStart, Access, CK);
622
623 if (!State)
624 return nullptr;
625
626 // Get the access length and make sure it is known.
627 // FIXME: This assumes the caller has already checked that the access length
628 // is positive. And that it's unsigned.
629 SVal LengthVal = C.getSVal(S: Size.Expression);
630 std::optional<NonLoc> Length = LengthVal.getAs<NonLoc>();
631 if (!Length)
632 return State;
633
634 // Compute the offset of the last element to be accessed: size-1.
635 NonLoc One = svalBuilder.makeIntVal(integer: 1, type: SizeTy).castAs<NonLoc>();
636 SVal Offset = svalBuilder.evalBinOpNN(state: State, op: BO_Sub, lhs: *Length, rhs: One, resultTy: SizeTy);
637 if (Offset.isUnknown())
638 return nullptr;
639 NonLoc LastOffset = Offset.castAs<NonLoc>();
640
641 // Check that the first buffer is sufficiently long.
642 if (std::optional<Loc> BufLoc = BufStart.getAs<Loc>()) {
643
644 SVal BufEnd =
645 svalBuilder.evalBinOpLN(state: State, op: BO_Add, lhs: *BufLoc, rhs: LastOffset, resultTy: PtrTy);
646 State = CheckLocation(C, state: State, Buffer, Element: BufEnd, Access, CK);
647 if (Access == AccessKind::read)
648 State = checkInit(C, State, Buffer, Element: BufEnd, Size: *Length);
649
650 // If the buffer isn't large enough, abort.
651 if (!State)
652 return nullptr;
653 }
654
655 // Large enough or not, return this state!
656 return State;
657}
658
659ProgramStateRef CStringChecker::CheckOverlap(CheckerContext &C,
660 ProgramStateRef state,
661 SizeArgExpr Size, AnyArgExpr First,
662 AnyArgExpr Second,
663 CharKind CK) const {
664 if (!BufferOverlap.isEnabled())
665 return state;
666
667 // Do a simple check for overlap: if the two arguments are from the same
668 // buffer, see if the end of the first is greater than the start of the second
669 // or vice versa.
670
671 // If a previous check has failed, propagate the failure.
672 if (!state)
673 return nullptr;
674
675 ProgramStateRef stateTrue, stateFalse;
676
677 if (!First.Expression->getType()->isAnyPointerType() ||
678 !Second.Expression->getType()->isAnyPointerType())
679 return state;
680
681 // Assume different address spaces cannot overlap.
682 if (First.Expression->getType()->getPointeeType().getAddressSpace() !=
683 Second.Expression->getType()->getPointeeType().getAddressSpace())
684 return state;
685
686 // Get the buffer values and make sure they're known locations.
687 const LocationContext *LCtx = C.getLocationContext();
688 SVal firstVal = state->getSVal(Ex: First.Expression, LCtx);
689 SVal secondVal = state->getSVal(Ex: Second.Expression, LCtx);
690
691 std::optional<Loc> firstLoc = firstVal.getAs<Loc>();
692 if (!firstLoc)
693 return state;
694
695 std::optional<Loc> secondLoc = secondVal.getAs<Loc>();
696 if (!secondLoc)
697 return state;
698
699 // Are the two values the same?
700 SValBuilder &svalBuilder = C.getSValBuilder();
701 std::tie(args&: stateTrue, args&: stateFalse) =
702 state->assume(Cond: svalBuilder.evalEQ(state, lhs: *firstLoc, rhs: *secondLoc));
703
704 if (stateTrue && !stateFalse) {
705 // If the values are known to be equal, that's automatically an overlap.
706 emitOverlapBug(C, state: stateTrue, First: First.Expression, Second: Second.Expression);
707 return nullptr;
708 }
709
710 // assume the two expressions are not equal.
711 assert(stateFalse);
712 state = stateFalse;
713
714 // Which value comes first?
715 QualType cmpTy = svalBuilder.getConditionType();
716 SVal reverse =
717 svalBuilder.evalBinOpLL(state, op: BO_GT, lhs: *firstLoc, rhs: *secondLoc, resultTy: cmpTy);
718 std::optional<DefinedOrUnknownSVal> reverseTest =
719 reverse.getAs<DefinedOrUnknownSVal>();
720 if (!reverseTest)
721 return state;
722
723 std::tie(args&: stateTrue, args&: stateFalse) = state->assume(Cond: *reverseTest);
724 if (stateTrue) {
725 if (stateFalse) {
726 // If we don't know which one comes first, we can't perform this test.
727 return state;
728 } else {
729 // Switch the values so that firstVal is before secondVal.
730 std::swap(lhs&: firstLoc, rhs&: secondLoc);
731
732 // Switch the Exprs as well, so that they still correspond.
733 std::swap(a&: First, b&: Second);
734 }
735 }
736
737 // Get the length, and make sure it too is known.
738 SVal LengthVal = state->getSVal(Ex: Size.Expression, LCtx);
739 std::optional<NonLoc> Length = LengthVal.getAs<NonLoc>();
740 if (!Length)
741 return state;
742
743 // Convert the first buffer's start address to char*.
744 // Bail out if the cast fails.
745 ASTContext &Ctx = svalBuilder.getContext();
746 QualType CharPtrTy = getCharPtrType(Ctx, CK);
747 SVal FirstStart =
748 svalBuilder.evalCast(V: *firstLoc, CastTy: CharPtrTy, OriginalTy: First.Expression->getType());
749 std::optional<Loc> FirstStartLoc = FirstStart.getAs<Loc>();
750 if (!FirstStartLoc)
751 return state;
752
753 // Compute the end of the first buffer. Bail out if THAT fails.
754 SVal FirstEnd = svalBuilder.evalBinOpLN(state, op: BO_Add, lhs: *FirstStartLoc,
755 rhs: *Length, resultTy: CharPtrTy);
756 std::optional<Loc> FirstEndLoc = FirstEnd.getAs<Loc>();
757 if (!FirstEndLoc)
758 return state;
759
760 // Is the end of the first buffer past the start of the second buffer?
761 SVal Overlap =
762 svalBuilder.evalBinOpLL(state, op: BO_GT, lhs: *FirstEndLoc, rhs: *secondLoc, resultTy: cmpTy);
763 std::optional<DefinedOrUnknownSVal> OverlapTest =
764 Overlap.getAs<DefinedOrUnknownSVal>();
765 if (!OverlapTest)
766 return state;
767
768 std::tie(args&: stateTrue, args&: stateFalse) = state->assume(Cond: *OverlapTest);
769
770 if (stateTrue && !stateFalse) {
771 // Overlap!
772 emitOverlapBug(C, state: stateTrue, First: First.Expression, Second: Second.Expression);
773 return nullptr;
774 }
775
776 // assume the two expressions don't overlap.
777 assert(stateFalse);
778 return stateFalse;
779}
780
781void CStringChecker::emitOverlapBug(CheckerContext &C, ProgramStateRef state,
782 const Stmt *First, const Stmt *Second) const {
783 ExplodedNode *N = C.generateErrorNode(State: state);
784 if (!N)
785 return;
786
787 // Generate a report for this bug.
788 auto report = std::make_unique<PathSensitiveBugReport>(
789 args: BufferOverlap, args: "Arguments must not be overlapping buffers", args&: N);
790 report->addRange(R: First->getSourceRange());
791 report->addRange(R: Second->getSourceRange());
792
793 C.emitReport(R: std::move(report));
794}
795
796void CStringChecker::emitNullArgBug(CheckerContext &C, ProgramStateRef State,
797 const Stmt *S, StringRef WarningMsg) const {
798 if (ExplodedNode *N = C.generateErrorNode(State)) {
799 auto Report =
800 std::make_unique<PathSensitiveBugReport>(args: NullArg, args&: WarningMsg, args&: N);
801 Report->addRange(R: S->getSourceRange());
802 if (const auto *Ex = dyn_cast<Expr>(Val: S))
803 bugreporter::trackExpressionValue(N, E: Ex, R&: *Report);
804 C.emitReport(R: std::move(Report));
805 }
806}
807
808void CStringChecker::emitUninitializedReadBug(CheckerContext &C,
809 ProgramStateRef State,
810 const Expr *E, const MemRegion *R,
811 StringRef Msg) const {
812 if (ExplodedNode *N = C.generateErrorNode(State)) {
813 auto Report =
814 std::make_unique<PathSensitiveBugReport>(args: UninitializedRead, args&: Msg, args&: N);
815 Report->addNote(Msg: "Other elements might also be undefined",
816 Pos: Report->getLocation());
817 Report->addRange(R: E->getSourceRange());
818 bugreporter::trackExpressionValue(N, E, R&: *Report);
819 Report->addVisitor<NoStoreFuncVisitor>(ConstructorArgs: R->castAs<SubRegion>());
820 C.emitReport(R: std::move(Report));
821 }
822}
823
824void CStringChecker::emitOutOfBoundsBug(CheckerContext &C,
825 ProgramStateRef State, const Stmt *S,
826 StringRef WarningMsg) const {
827 if (ExplodedNode *N = C.generateErrorNode(State)) {
828 // FIXME: It would be nice to eventually make this diagnostic more clear,
829 // e.g., by referencing the original declaration or by saying *why* this
830 // reference is outside the range.
831 auto Report =
832 std::make_unique<PathSensitiveBugReport>(args: OutOfBounds, args&: WarningMsg, args&: N);
833 Report->addRange(R: S->getSourceRange());
834 C.emitReport(R: std::move(Report));
835 }
836}
837
838void CStringChecker::emitNotCStringBug(CheckerContext &C, ProgramStateRef State,
839 const Stmt *S,
840 StringRef WarningMsg) const {
841 if (ExplodedNode *N = C.generateNonFatalErrorNode(State)) {
842 auto Report =
843 std::make_unique<PathSensitiveBugReport>(args: NotNullTerm, args&: WarningMsg, args&: N);
844
845 Report->addRange(R: S->getSourceRange());
846 C.emitReport(R: std::move(Report));
847 }
848}
849
850ProgramStateRef CStringChecker::checkAdditionOverflow(CheckerContext &C,
851 ProgramStateRef state,
852 NonLoc left,
853 NonLoc right) const {
854 // If out-of-bounds checking is turned off, skip the rest.
855 if (!OutOfBounds.isEnabled())
856 return state;
857
858 // If a previous check has failed, propagate the failure.
859 if (!state)
860 return nullptr;
861
862 SValBuilder &svalBuilder = C.getSValBuilder();
863 BasicValueFactory &BVF = svalBuilder.getBasicValueFactory();
864
865 QualType sizeTy = svalBuilder.getContext().getSizeType();
866 const llvm::APSInt &maxValInt = BVF.getMaxValue(T: sizeTy);
867 NonLoc maxVal = svalBuilder.makeIntVal(integer: maxValInt);
868
869 SVal maxMinusRight;
870 if (isa<nonloc::ConcreteInt>(Val: right)) {
871 maxMinusRight = svalBuilder.evalBinOpNN(state, op: BO_Sub, lhs: maxVal, rhs: right,
872 resultTy: sizeTy);
873 } else {
874 // Try switching the operands. (The order of these two assignments is
875 // important!)
876 maxMinusRight = svalBuilder.evalBinOpNN(state, op: BO_Sub, lhs: maxVal, rhs: left,
877 resultTy: sizeTy);
878 left = right;
879 }
880
881 if (std::optional<NonLoc> maxMinusRightNL = maxMinusRight.getAs<NonLoc>()) {
882 QualType cmpTy = svalBuilder.getConditionType();
883 // If left > max - right, we have an overflow.
884 SVal willOverflow = svalBuilder.evalBinOpNN(state, op: BO_GT, lhs: left,
885 rhs: *maxMinusRightNL, resultTy: cmpTy);
886
887 auto [StateOverflow, StateOkay] =
888 state->assume(Cond: willOverflow.castAs<DefinedOrUnknownSVal>());
889
890 if (StateOverflow && !StateOkay) {
891 // On this path the analyzer is convinced that the addition of these two
892 // values would overflow `size_t` which must be caused by the inaccuracy
893 // of our modeling because this method is called in situations where the
894 // summands are size/length values which are much less than SIZE_MAX. To
895 // avoid false positives let's just sink this invalid path.
896 C.addSink(State: StateOverflow);
897 return nullptr;
898 }
899
900 // From now on, assume an overflow didn't occur.
901 assert(StateOkay);
902 state = StateOkay;
903 }
904
905 return state;
906}
907
908ProgramStateRef CStringChecker::setCStringLength(ProgramStateRef state,
909 const MemRegion *MR,
910 SVal strLength) {
911 assert(!strLength.isUndef() && "Attempt to set an undefined string length");
912
913 MR = MR->StripCasts();
914
915 switch (MR->getKind()) {
916 case MemRegion::StringRegionKind:
917 // FIXME: This can happen if we strcpy() into a string region. This is
918 // undefined [C99 6.4.5p6], but we should still warn about it.
919 return state;
920
921 case MemRegion::SymbolicRegionKind:
922 case MemRegion::AllocaRegionKind:
923 case MemRegion::NonParamVarRegionKind:
924 case MemRegion::ParamVarRegionKind:
925 case MemRegion::FieldRegionKind:
926 case MemRegion::ObjCIvarRegionKind:
927 // These are the types we can currently track string lengths for.
928 break;
929
930 case MemRegion::ElementRegionKind:
931 // FIXME: Handle element regions by upper-bounding the parent region's
932 // string length.
933 return state;
934
935 default:
936 // Other regions (mostly non-data) can't have a reliable C string length.
937 // For now, just ignore the change.
938 // FIXME: These are rare but not impossible. We should output some kind of
939 // warning for things like strcpy((char[]){'a', 0}, "b");
940 return state;
941 }
942
943 if (strLength.isUnknown())
944 return state->remove<CStringLength>(K: MR);
945
946 return state->set<CStringLength>(K: MR, E: strLength);
947}
948
949SVal CStringChecker::getCStringLengthForRegion(CheckerContext &C,
950 ProgramStateRef &state,
951 const Expr *Ex,
952 const MemRegion *MR,
953 bool hypothetical) {
954 if (!hypothetical) {
955 // If there's a recorded length, go ahead and return it.
956 const SVal *Recorded = state->get<CStringLength>(key: MR);
957 if (Recorded)
958 return *Recorded;
959 }
960
961 // Otherwise, get a new symbol and update the state.
962 SValBuilder &svalBuilder = C.getSValBuilder();
963 QualType sizeTy = svalBuilder.getContext().getSizeType();
964 SVal strLength = svalBuilder.getMetadataSymbolVal(symbolTag: CStringChecker::getTag(),
965 region: MR, expr: Ex, type: sizeTy,
966 LCtx: C.getLocationContext(),
967 count: C.blockCount());
968
969 if (!hypothetical) {
970 if (std::optional<NonLoc> strLn = strLength.getAs<NonLoc>()) {
971 // In case of unbounded calls strlen etc bound the range to SIZE_MAX/4
972 BasicValueFactory &BVF = svalBuilder.getBasicValueFactory();
973 const llvm::APSInt &maxValInt = BVF.getMaxValue(T: sizeTy);
974 llvm::APSInt fourInt = APSIntType(maxValInt).getValue(RawValue: 4);
975 std::optional<APSIntPtr> maxLengthInt =
976 BVF.evalAPSInt(Op: BO_Div, V1: maxValInt, V2: fourInt);
977 NonLoc maxLength = svalBuilder.makeIntVal(integer: *maxLengthInt);
978 SVal evalLength = svalBuilder.evalBinOpNN(state, op: BO_LE, lhs: *strLn, rhs: maxLength,
979 resultTy: svalBuilder.getConditionType());
980 state = state->assume(Cond: evalLength.castAs<DefinedOrUnknownSVal>(), Assumption: true);
981 }
982 state = state->set<CStringLength>(K: MR, E: strLength);
983 }
984
985 return strLength;
986}
987
988const StringLiteral *
989CStringChecker::getStringLiteralFromRegion(const MemRegion *MR) {
990 switch (MR->getKind()) {
991 case MemRegion::StringRegionKind:
992 return cast<StringRegion>(Val: MR)->getStringLiteral();
993 case MemRegion::NonParamVarRegionKind:
994 if (const VarDecl *Decl = cast<NonParamVarRegion>(Val: MR)->getDecl();
995 Decl->getType().isConstQualified() && Decl->hasGlobalStorage())
996 return dyn_cast_or_null<StringLiteral>(Val: Decl->getInit());
997 return nullptr;
998 default:
999 return nullptr;
1000 }
1001}
1002
1003SVal CStringChecker::getCStringLength(CheckerContext &C, ProgramStateRef &state,
1004 const Expr *Ex, SVal Buf,
1005 bool hypothetical) const {
1006 const MemRegion *MR = Buf.getAsRegion();
1007 if (!MR) {
1008 // If we can't get a region, see if it's something we /know/ isn't a
1009 // C string. In the context of locations, the only time we can issue such
1010 // a warning is for labels.
1011 if (std::optional<loc::GotoLabel> Label = Buf.getAs<loc::GotoLabel>()) {
1012 if (NotNullTerm.isEnabled()) {
1013 SmallString<120> buf;
1014 llvm::raw_svector_ostream os(buf);
1015 assert(CurrentFunctionDescription);
1016 os << "Argument to " << CurrentFunctionDescription
1017 << " is the address of the label '" << Label->getLabel()->getName()
1018 << "', which is not a null-terminated string";
1019
1020 emitNotCStringBug(C, State: state, S: Ex, WarningMsg: os.str());
1021 }
1022 return UndefinedVal();
1023 }
1024
1025 // If it's not a region and not a label, give up.
1026 return UnknownVal();
1027 }
1028
1029 // If we have a region, strip casts from it and see if we can figure out
1030 // its length. For anything we can't figure out, just return UnknownVal.
1031 MR = MR->StripCasts();
1032
1033 if (const StringLiteral *StrLit = getStringLiteralFromRegion(MR)) {
1034 // If we have a global constant with a string literal initializer,
1035 // compute the initializer's length.
1036 // Modifying the contents of string regions is undefined [C99 6.4.5p6],
1037 // so we can assume that the byte length is the correct C string length.
1038 // FIXME: Embedded null characters are not handled.
1039 SValBuilder &SVB = C.getSValBuilder();
1040 return SVB.makeIntVal(integer: StrLit->getLength(), type: SVB.getContext().getSizeType());
1041 }
1042
1043 switch (MR->getKind()) {
1044 case MemRegion::StringRegionKind:
1045 case MemRegion::NonParamVarRegionKind:
1046 case MemRegion::SymbolicRegionKind:
1047 case MemRegion::AllocaRegionKind:
1048 case MemRegion::ParamVarRegionKind:
1049 case MemRegion::FieldRegionKind:
1050 case MemRegion::ObjCIvarRegionKind:
1051 return getCStringLengthForRegion(C, state, Ex, MR, hypothetical);
1052 case MemRegion::CompoundLiteralRegionKind:
1053 // FIXME: Can we track this? Is it necessary?
1054 return UnknownVal();
1055 case MemRegion::ElementRegionKind: {
1056 // If an offset into the string literal is used, use the original length
1057 // minus the offset.
1058 // FIXME: Embedded null characters are not handled.
1059 const ElementRegion *ER = cast<ElementRegion>(Val: MR);
1060 const SubRegion *SuperReg =
1061 cast<SubRegion>(Val: ER->getSuperRegion()->StripCasts());
1062 const StringLiteral *StrLit = getStringLiteralFromRegion(MR: SuperReg);
1063 if (!StrLit)
1064 return UnknownVal();
1065 SValBuilder &SVB = C.getSValBuilder();
1066 NonLoc Idx = ER->getIndex();
1067 QualType SizeTy = SVB.getContext().getSizeType();
1068 NonLoc LengthVal =
1069 SVB.makeIntVal(integer: StrLit->getLength(), type: SizeTy).castAs<NonLoc>();
1070 if (state->assume(Cond: SVB.evalBinOpNN(state, op: BO_LE, lhs: Idx, rhs: LengthVal,
1071 resultTy: SVB.getConditionType())
1072 .castAs<DefinedOrUnknownSVal>(),
1073 Assumption: true))
1074 return SVB.evalBinOp(state, op: BO_Sub, lhs: LengthVal, rhs: Idx, type: SizeTy);
1075 return UnknownVal();
1076 }
1077 default:
1078 // Other regions (mostly non-data) can't have a reliable C string length.
1079 // In this case, an error is emitted and UndefinedVal is returned.
1080 // The caller should always be prepared to handle this case.
1081 if (NotNullTerm.isEnabled()) {
1082 SmallString<120> buf;
1083 llvm::raw_svector_ostream os(buf);
1084
1085 assert(CurrentFunctionDescription);
1086 os << "Argument to " << CurrentFunctionDescription << " is ";
1087
1088 if (SummarizeRegion(os, Ctx&: C.getASTContext(), MR))
1089 os << ", which is not a null-terminated string";
1090 else
1091 os << "not a null-terminated string";
1092
1093 emitNotCStringBug(C, State: state, S: Ex, WarningMsg: os.str());
1094 }
1095 return UndefinedVal();
1096 }
1097}
1098
1099const StringLiteral *CStringChecker::getCStringLiteral(CheckerContext &C,
1100 ProgramStateRef &state, const Expr *expr, SVal val) const {
1101 // FIXME: use getStringLiteralFromRegion (and remove unused parameters)?
1102
1103 // Get the memory region pointed to by the val.
1104 const MemRegion *bufRegion = val.getAsRegion();
1105 if (!bufRegion)
1106 return nullptr;
1107
1108 // Strip casts off the memory region.
1109 bufRegion = bufRegion->StripCasts();
1110
1111 // Cast the memory region to a string region.
1112 const StringRegion *strRegion= dyn_cast<StringRegion>(Val: bufRegion);
1113 if (!strRegion)
1114 return nullptr;
1115
1116 // Return the actual string in the string region.
1117 return strRegion->getStringLiteral();
1118}
1119
1120bool CStringChecker::isFirstBufInBound(CheckerContext &C, ProgramStateRef State,
1121 SVal BufVal, QualType BufTy,
1122 SVal LengthVal, QualType LengthTy) {
1123 // If we do not know that the buffer is long enough we return 'true'.
1124 // Otherwise the parent region of this field region would also get
1125 // invalidated, which would lead to warnings based on an unknown state.
1126
1127 if (LengthVal.isUnknown())
1128 return false;
1129
1130 // Originally copied from CheckBufferAccess and CheckLocation.
1131 SValBuilder &SB = C.getSValBuilder();
1132 ASTContext &Ctx = C.getASTContext();
1133
1134 QualType PtrTy = Ctx.getPointerType(T: Ctx.CharTy);
1135
1136 std::optional<NonLoc> Length = LengthVal.getAs<NonLoc>();
1137 if (!Length)
1138 return true; // cf top comment.
1139
1140 // Compute the offset of the last element to be accessed: size-1.
1141 NonLoc One = SB.makeIntVal(integer: 1, type: LengthTy).castAs<NonLoc>();
1142 SVal Offset = SB.evalBinOpNN(state: State, op: BO_Sub, lhs: *Length, rhs: One, resultTy: LengthTy);
1143 if (Offset.isUnknown())
1144 return true; // cf top comment
1145 NonLoc LastOffset = Offset.castAs<NonLoc>();
1146
1147 // Check that the first buffer is sufficiently long.
1148 SVal BufStart = SB.evalCast(V: BufVal, CastTy: PtrTy, OriginalTy: BufTy);
1149 std::optional<Loc> BufLoc = BufStart.getAs<Loc>();
1150 if (!BufLoc)
1151 return true; // cf top comment.
1152
1153 SVal BufEnd = SB.evalBinOpLN(state: State, op: BO_Add, lhs: *BufLoc, rhs: LastOffset, resultTy: PtrTy);
1154
1155 // Check for out of bound array element access.
1156 const MemRegion *R = BufEnd.getAsRegion();
1157 if (!R)
1158 return true; // cf top comment.
1159
1160 const ElementRegion *ER = dyn_cast<ElementRegion>(Val: R);
1161 if (!ER)
1162 return true; // cf top comment.
1163
1164 // Support library functions defined with non-default address spaces
1165 assert(ER->getValueType()->getCanonicalTypeUnqualified() ==
1166 C.getASTContext().CharTy &&
1167 "isFirstBufInBound should only be called with char* ElementRegions");
1168
1169 // Get the size of the array.
1170 const SubRegion *superReg = cast<SubRegion>(Val: ER->getSuperRegion());
1171 DefinedOrUnknownSVal SizeDV = getDynamicExtent(State, MR: superReg, SVB&: SB);
1172
1173 // Get the index of the accessed element.
1174 DefinedOrUnknownSVal Idx = ER->getIndex().castAs<DefinedOrUnknownSVal>();
1175
1176 ProgramStateRef StInBound = State->assumeInBound(idx: Idx, upperBound: SizeDV, assumption: true);
1177
1178 return static_cast<bool>(StInBound);
1179}
1180
1181ProgramStateRef CStringChecker::invalidateDestinationBufferBySize(
1182 CheckerContext &C, ProgramStateRef S, const Expr *BufE,
1183 ConstCFGElementRef Elem, SVal BufV, SVal SizeV, QualType SizeTy) {
1184 auto InvalidationTraitOperations =
1185 [&C, S, BufTy = BufE->getType(), BufV, SizeV,
1186 SizeTy](RegionAndSymbolInvalidationTraits &ITraits, const MemRegion *R) {
1187 // If destination buffer is a field region and access is in bound, do
1188 // not invalidate its super region.
1189 if (MemRegion::FieldRegionKind == R->getKind() &&
1190 isFirstBufInBound(C, State: S, BufVal: BufV, BufTy, LengthVal: SizeV, LengthTy: SizeTy)) {
1191 ITraits.setTrait(
1192 MR: R,
1193 IK: RegionAndSymbolInvalidationTraits::TK_DoNotInvalidateSuperRegion);
1194 }
1195 return false;
1196 };
1197
1198 return invalidateBufferAux(C, State: S, Elem, V: BufV, InvalidationTraitOperations);
1199}
1200
1201ProgramStateRef
1202CStringChecker::invalidateDestinationBufferAlwaysEscapeSuperRegion(
1203 CheckerContext &C, ProgramStateRef S, ConstCFGElementRef Elem, SVal BufV) {
1204 auto InvalidationTraitOperations = [](RegionAndSymbolInvalidationTraits &,
1205 const MemRegion *R) {
1206 return isa<FieldRegion>(Val: R);
1207 };
1208
1209 return invalidateBufferAux(C, State: S, Elem, V: BufV, InvalidationTraitOperations);
1210}
1211
1212ProgramStateRef CStringChecker::invalidateDestinationBufferNeverOverflows(
1213 CheckerContext &C, ProgramStateRef S, ConstCFGElementRef Elem, SVal BufV) {
1214 auto InvalidationTraitOperations =
1215 [](RegionAndSymbolInvalidationTraits &ITraits, const MemRegion *R) {
1216 if (MemRegion::FieldRegionKind == R->getKind())
1217 ITraits.setTrait(
1218 MR: R,
1219 IK: RegionAndSymbolInvalidationTraits::TK_DoNotInvalidateSuperRegion);
1220 return false;
1221 };
1222
1223 return invalidateBufferAux(C, State: S, Elem, V: BufV, InvalidationTraitOperations);
1224}
1225
1226ProgramStateRef CStringChecker::invalidateSourceBuffer(CheckerContext &C,
1227 ProgramStateRef S,
1228 ConstCFGElementRef Elem,
1229 SVal BufV) {
1230 auto InvalidationTraitOperations =
1231 [](RegionAndSymbolInvalidationTraits &ITraits, const MemRegion *R) {
1232 ITraits.setTrait(
1233 MR: R->getBaseRegion(),
1234 IK: RegionAndSymbolInvalidationTraits::TK_PreserveContents);
1235 ITraits.setTrait(MR: R,
1236 IK: RegionAndSymbolInvalidationTraits::TK_SuppressEscape);
1237 return true;
1238 };
1239
1240 return invalidateBufferAux(C, State: S, Elem, V: BufV, InvalidationTraitOperations);
1241}
1242
1243ProgramStateRef CStringChecker::invalidateBufferAux(
1244 CheckerContext &C, ProgramStateRef State, ConstCFGElementRef Elem, SVal V,
1245 llvm::function_ref<bool(RegionAndSymbolInvalidationTraits &,
1246 const MemRegion *)>
1247 InvalidationTraitOperations) {
1248 std::optional<Loc> L = V.getAs<Loc>();
1249 if (!L)
1250 return State;
1251
1252 // FIXME: This is a simplified version of what's in CFRefCount.cpp -- it makes
1253 // some assumptions about the value that CFRefCount can't. Even so, it should
1254 // probably be refactored.
1255 if (std::optional<loc::MemRegionVal> MR = L->getAs<loc::MemRegionVal>()) {
1256 const MemRegion *R = MR->getRegion()->StripCasts();
1257
1258 // Are we dealing with an ElementRegion? If so, we should be invalidating
1259 // the super-region.
1260 if (const ElementRegion *ER = dyn_cast<ElementRegion>(Val: R)) {
1261 R = ER->getSuperRegion();
1262 // FIXME: What about layers of ElementRegions?
1263 }
1264
1265 // Invalidate this region.
1266 const LocationContext *LCtx = C.getPredecessor()->getLocationContext();
1267 RegionAndSymbolInvalidationTraits ITraits;
1268 bool CausesPointerEscape = InvalidationTraitOperations(ITraits, R);
1269
1270 return State->invalidateRegions(Regions: R, Elem, BlockCount: C.blockCount(), LCtx,
1271 CausesPointerEscape, IS: nullptr, Call: nullptr,
1272 ITraits: &ITraits);
1273 }
1274
1275 // If we have a non-region value by chance, just remove the binding.
1276 // FIXME: is this necessary or correct? This handles the non-Region
1277 // cases. Is it ever valid to store to these?
1278 return State->killBinding(LV: *L);
1279}
1280
1281bool CStringChecker::SummarizeRegion(raw_ostream &os, ASTContext &Ctx,
1282 const MemRegion *MR) {
1283 switch (MR->getKind()) {
1284 case MemRegion::FunctionCodeRegionKind: {
1285 if (const auto *FD = cast<FunctionCodeRegion>(Val: MR)->getDecl())
1286 os << "the address of the function '" << *FD << '\'';
1287 else
1288 os << "the address of a function";
1289 return true;
1290 }
1291 case MemRegion::BlockCodeRegionKind:
1292 os << "block text";
1293 return true;
1294 case MemRegion::BlockDataRegionKind:
1295 os << "a block";
1296 return true;
1297 case MemRegion::CXXThisRegionKind:
1298 case MemRegion::CXXTempObjectRegionKind:
1299 os << "a C++ temp object of type "
1300 << cast<TypedValueRegion>(Val: MR)->getValueType();
1301 return true;
1302 case MemRegion::NonParamVarRegionKind:
1303 os << "a variable of type" << cast<TypedValueRegion>(Val: MR)->getValueType();
1304 return true;
1305 case MemRegion::ParamVarRegionKind:
1306 os << "a parameter of type" << cast<TypedValueRegion>(Val: MR)->getValueType();
1307 return true;
1308 case MemRegion::FieldRegionKind:
1309 os << "a field of type " << cast<TypedValueRegion>(Val: MR)->getValueType();
1310 return true;
1311 case MemRegion::ObjCIvarRegionKind:
1312 os << "an instance variable of type "
1313 << cast<TypedValueRegion>(Val: MR)->getValueType();
1314 return true;
1315 default:
1316 return false;
1317 }
1318}
1319
1320bool CStringChecker::memsetAux(const Expr *DstBuffer, ConstCFGElementRef Elem,
1321 SVal CharVal, const Expr *Size,
1322 CheckerContext &C, ProgramStateRef &State) {
1323 SVal MemVal = C.getSVal(S: DstBuffer);
1324 SVal SizeVal = C.getSVal(S: Size);
1325 const MemRegion *MR = MemVal.getAsRegion();
1326 if (!MR)
1327 return false;
1328
1329 // We're about to model memset by producing a "default binding" in the Store.
1330 // Our current implementation - RegionStore - doesn't support default bindings
1331 // that don't cover the whole base region. So we should first get the offset
1332 // and the base region to figure out whether the offset of buffer is 0.
1333 RegionOffset Offset = MR->getAsOffset();
1334 const MemRegion *BR = Offset.getRegion();
1335
1336 std::optional<NonLoc> SizeNL = SizeVal.getAs<NonLoc>();
1337 if (!SizeNL)
1338 return false;
1339
1340 SValBuilder &svalBuilder = C.getSValBuilder();
1341 ASTContext &Ctx = C.getASTContext();
1342
1343 // void *memset(void *dest, int ch, size_t count);
1344 // For now we can only handle the case of offset is 0 and concrete char value.
1345 if (Offset.isValid() && !Offset.hasSymbolicOffset() &&
1346 Offset.getOffset() == 0) {
1347 // Get the base region's size.
1348 DefinedOrUnknownSVal SizeDV = getDynamicExtent(State, MR: BR, SVB&: svalBuilder);
1349
1350 ProgramStateRef StateWholeReg, StateNotWholeReg;
1351 std::tie(args&: StateWholeReg, args&: StateNotWholeReg) =
1352 State->assume(Cond: svalBuilder.evalEQ(state: State, lhs: SizeDV, rhs: *SizeNL));
1353
1354 // With the semantic of 'memset()', we should convert the CharVal to
1355 // unsigned char.
1356 CharVal = svalBuilder.evalCast(V: CharVal, CastTy: Ctx.UnsignedCharTy, OriginalTy: Ctx.IntTy);
1357
1358 ProgramStateRef StateNullChar, StateNonNullChar;
1359 std::tie(args&: StateNullChar, args&: StateNonNullChar) =
1360 assumeZero(C, State, V: CharVal, Ty: Ctx.UnsignedCharTy);
1361
1362 if (StateWholeReg && !StateNotWholeReg && StateNullChar &&
1363 !StateNonNullChar) {
1364 // If the 'memset()' acts on the whole region of destination buffer and
1365 // the value of the second argument of 'memset()' is zero, bind the second
1366 // argument's value to the destination buffer with 'default binding'.
1367 // FIXME: Since there is no perfect way to bind the non-zero character, we
1368 // can only deal with zero value here. In the future, we need to deal with
1369 // the binding of non-zero value in the case of whole region.
1370 State = State->bindDefaultZero(loc: svalBuilder.makeLoc(region: BR),
1371 LCtx: C.getLocationContext());
1372 } else {
1373 // If the destination buffer's extent is not equal to the value of
1374 // third argument, just invalidate buffer.
1375 State = invalidateDestinationBufferBySize(
1376 C, S: State, BufE: DstBuffer, Elem, BufV: MemVal, SizeV: SizeVal, SizeTy: Size->getType());
1377 }
1378
1379 if (StateNullChar && !StateNonNullChar) {
1380 // If the value of the second argument of 'memset()' is zero, set the
1381 // string length of destination buffer to 0 directly.
1382 State = setCStringLength(state: State, MR,
1383 strLength: svalBuilder.makeZeroVal(type: Ctx.getSizeType()));
1384 } else if (!StateNullChar && StateNonNullChar) {
1385 SVal NewStrLen = svalBuilder.getMetadataSymbolVal(
1386 symbolTag: CStringChecker::getTag(), region: MR, expr: DstBuffer, type: Ctx.getSizeType(),
1387 LCtx: C.getLocationContext(), count: C.blockCount());
1388
1389 // If the value of second argument is not zero, then the string length
1390 // is at least the size argument.
1391 SVal NewStrLenGESize = svalBuilder.evalBinOp(
1392 state: State, op: BO_GE, lhs: NewStrLen, rhs: SizeVal, type: svalBuilder.getConditionType());
1393
1394 State = setCStringLength(
1395 state: State->assume(Cond: NewStrLenGESize.castAs<DefinedOrUnknownSVal>(), Assumption: true),
1396 MR, strLength: NewStrLen);
1397 }
1398 } else {
1399 // If the offset is not zero and char value is not concrete, we can do
1400 // nothing but invalidate the buffer.
1401 State = invalidateDestinationBufferBySize(C, S: State, BufE: DstBuffer, Elem, BufV: MemVal,
1402 SizeV: SizeVal, SizeTy: Size->getType());
1403 }
1404 return true;
1405}
1406
1407//===----------------------------------------------------------------------===//
1408// evaluation of individual function calls.
1409//===----------------------------------------------------------------------===//
1410
1411void CStringChecker::evalCopyCommon(CheckerContext &C, const CallEvent &Call,
1412 ProgramStateRef state, SizeArgExpr Size,
1413 DestinationArgExpr Dest,
1414 SourceArgExpr Source, bool Restricted,
1415 bool IsMempcpy, CharKind CK) const {
1416 CurrentFunctionDescription = "memory copy function";
1417
1418 // See if the size argument is zero.
1419 const LocationContext *LCtx = C.getLocationContext();
1420 SVal sizeVal = state->getSVal(Ex: Size.Expression, LCtx);
1421 QualType sizeTy = Size.Expression->getType();
1422
1423 ProgramStateRef stateZeroSize, stateNonZeroSize;
1424 std::tie(args&: stateZeroSize, args&: stateNonZeroSize) =
1425 assumeZero(C, State: state, V: sizeVal, Ty: sizeTy);
1426
1427 // Get the value of the Dest.
1428 SVal destVal = state->getSVal(Ex: Dest.Expression, LCtx);
1429
1430 // If the size is zero, there won't be any actual memory access, so
1431 // just bind the return value to the destination buffer and return.
1432 if (stateZeroSize && !stateNonZeroSize) {
1433 stateZeroSize =
1434 stateZeroSize->BindExpr(S: Call.getOriginExpr(), LCtx, V: destVal);
1435 C.addTransition(State: stateZeroSize);
1436 return;
1437 }
1438
1439 // If the size can be nonzero, we have to check the other arguments.
1440 if (stateNonZeroSize) {
1441 // TODO: If Size is tainted and we cannot prove that it is smaller or equal
1442 // to the size of the destination buffer, then emit a warning
1443 // that an attacker may provoke a buffer overflow error.
1444 state = stateNonZeroSize;
1445
1446 // Ensure the destination is not null. If it is NULL there will be a
1447 // NULL pointer dereference.
1448 state = checkNonNull(C, State: state, Arg: Dest, l: destVal);
1449 if (!state)
1450 return;
1451
1452 // Get the value of the Src.
1453 SVal srcVal = state->getSVal(Ex: Source.Expression, LCtx);
1454
1455 // Ensure the source is not null. If it is NULL there will be a
1456 // NULL pointer dereference.
1457 state = checkNonNull(C, State: state, Arg: Source, l: srcVal);
1458 if (!state)
1459 return;
1460
1461 // Ensure the accesses are valid and that the buffers do not overlap.
1462 state = CheckBufferAccess(C, State: state, Buffer: Dest, Size, Access: AccessKind::write, CK);
1463 state = CheckBufferAccess(C, State: state, Buffer: Source, Size, Access: AccessKind::read, CK);
1464
1465 if (Restricted)
1466 state = CheckOverlap(C, state, Size, First: Dest, Second: Source, CK);
1467
1468 if (!state)
1469 return;
1470
1471 // If this is mempcpy, get the byte after the last byte copied and
1472 // bind the expr.
1473 if (IsMempcpy) {
1474 // Get the byte after the last byte copied.
1475 SValBuilder &SvalBuilder = C.getSValBuilder();
1476 ASTContext &Ctx = SvalBuilder.getContext();
1477 QualType CharPtrTy = getCharPtrType(Ctx, CK);
1478 SVal DestRegCharVal =
1479 SvalBuilder.evalCast(V: destVal, CastTy: CharPtrTy, OriginalTy: Dest.Expression->getType());
1480 SVal lastElement = C.getSValBuilder().evalBinOp(
1481 state, op: BO_Add, lhs: DestRegCharVal, rhs: sizeVal, type: Dest.Expression->getType());
1482 // If we don't know how much we copied, we can at least
1483 // conjure a return value for later.
1484 if (lastElement.isUnknown())
1485 lastElement = C.getSValBuilder().conjureSymbolVal(call: Call, visitCount: C.blockCount());
1486
1487 // The byte after the last byte copied is the return value.
1488 state = state->BindExpr(S: Call.getOriginExpr(), LCtx, V: lastElement);
1489 } else {
1490 // All other copies return the destination buffer.
1491 // (Well, bcopy() has a void return type, but this won't hurt.)
1492 state = state->BindExpr(S: Call.getOriginExpr(), LCtx, V: destVal);
1493 }
1494
1495 // Invalidate the destination (regular invalidation without pointer-escaping
1496 // the address of the top-level region).
1497 // FIXME: Even if we can't perfectly model the copy, we should see if we
1498 // can use LazyCompoundVals to copy the source values into the destination.
1499 // This would probably remove any existing bindings past the end of the
1500 // copied region, but that's still an improvement over blank invalidation.
1501 state = invalidateDestinationBufferBySize(
1502 C, S: state, BufE: Dest.Expression, Elem: Call.getCFGElementRef(),
1503 BufV: C.getSVal(S: Dest.Expression), SizeV: sizeVal, SizeTy: Size.Expression->getType());
1504
1505 // Invalidate the source (const-invalidation without const-pointer-escaping
1506 // the address of the top-level region).
1507 state = invalidateSourceBuffer(C, S: state, Elem: Call.getCFGElementRef(),
1508 BufV: C.getSVal(S: Source.Expression));
1509
1510 C.addTransition(State: state);
1511 }
1512}
1513
1514void CStringChecker::evalMemcpy(CheckerContext &C, const CallEvent &Call,
1515 CharKind CK) const {
1516 // void *memcpy(void *restrict dst, const void *restrict src, size_t n);
1517 // The return value is the address of the destination buffer.
1518 DestinationArgExpr Dest = {{.Expression: Call.getArgExpr(Index: 0), .ArgumentIndex: 0}};
1519 SourceArgExpr Src = {{.Expression: Call.getArgExpr(Index: 1), .ArgumentIndex: 1}};
1520 SizeArgExpr Size = {{.Expression: Call.getArgExpr(Index: 2), .ArgumentIndex: 2}};
1521
1522 ProgramStateRef State = C.getState();
1523
1524 constexpr bool IsRestricted = true;
1525 constexpr bool IsMempcpy = false;
1526 evalCopyCommon(C, Call, state: State, Size, Dest, Source: Src, Restricted: IsRestricted, IsMempcpy, CK);
1527}
1528
1529void CStringChecker::evalMempcpy(CheckerContext &C, const CallEvent &Call,
1530 CharKind CK) const {
1531 // void *mempcpy(void *restrict dst, const void *restrict src, size_t n);
1532 // The return value is a pointer to the byte following the last written byte.
1533 DestinationArgExpr Dest = {{.Expression: Call.getArgExpr(Index: 0), .ArgumentIndex: 0}};
1534 SourceArgExpr Src = {{.Expression: Call.getArgExpr(Index: 1), .ArgumentIndex: 1}};
1535 SizeArgExpr Size = {{.Expression: Call.getArgExpr(Index: 2), .ArgumentIndex: 2}};
1536
1537 constexpr bool IsRestricted = true;
1538 constexpr bool IsMempcpy = true;
1539 evalCopyCommon(C, Call, state: C.getState(), Size, Dest, Source: Src, Restricted: IsRestricted,
1540 IsMempcpy, CK);
1541}
1542
1543void CStringChecker::evalMemmove(CheckerContext &C, const CallEvent &Call,
1544 CharKind CK) const {
1545 // void *memmove(void *dst, const void *src, size_t n);
1546 // The return value is the address of the destination buffer.
1547 DestinationArgExpr Dest = {{.Expression: Call.getArgExpr(Index: 0), .ArgumentIndex: 0}};
1548 SourceArgExpr Src = {{.Expression: Call.getArgExpr(Index: 1), .ArgumentIndex: 1}};
1549 SizeArgExpr Size = {{.Expression: Call.getArgExpr(Index: 2), .ArgumentIndex: 2}};
1550
1551 constexpr bool IsRestricted = false;
1552 constexpr bool IsMempcpy = false;
1553 evalCopyCommon(C, Call, state: C.getState(), Size, Dest, Source: Src, Restricted: IsRestricted,
1554 IsMempcpy, CK);
1555}
1556
1557void CStringChecker::evalBcopy(CheckerContext &C, const CallEvent &Call) const {
1558 // void bcopy(const void *src, void *dst, size_t n);
1559 SourceArgExpr Src{{.Expression: Call.getArgExpr(Index: 0), .ArgumentIndex: 0}};
1560 DestinationArgExpr Dest = {{.Expression: Call.getArgExpr(Index: 1), .ArgumentIndex: 1}};
1561 SizeArgExpr Size = {{.Expression: Call.getArgExpr(Index: 2), .ArgumentIndex: 2}};
1562
1563 constexpr bool IsRestricted = false;
1564 constexpr bool IsMempcpy = false;
1565 evalCopyCommon(C, Call, state: C.getState(), Size, Dest, Source: Src, Restricted: IsRestricted,
1566 IsMempcpy, CK: CharKind::Regular);
1567}
1568
1569void CStringChecker::evalMemcmp(CheckerContext &C, const CallEvent &Call,
1570 CharKind CK) const {
1571 // int memcmp(const void *s1, const void *s2, size_t n);
1572 CurrentFunctionDescription = "memory comparison function";
1573
1574 AnyArgExpr Left = {.Expression: Call.getArgExpr(Index: 0), .ArgumentIndex: 0};
1575 AnyArgExpr Right = {.Expression: Call.getArgExpr(Index: 1), .ArgumentIndex: 1};
1576 SizeArgExpr Size = {{.Expression: Call.getArgExpr(Index: 2), .ArgumentIndex: 2}};
1577
1578 ProgramStateRef State = C.getState();
1579 SValBuilder &Builder = C.getSValBuilder();
1580 const LocationContext *LCtx = C.getLocationContext();
1581
1582 // See if the size argument is zero.
1583 SVal sizeVal = State->getSVal(Ex: Size.Expression, LCtx);
1584 QualType sizeTy = Size.Expression->getType();
1585
1586 ProgramStateRef stateZeroSize, stateNonZeroSize;
1587 std::tie(args&: stateZeroSize, args&: stateNonZeroSize) =
1588 assumeZero(C, State, V: sizeVal, Ty: sizeTy);
1589
1590 // If the size can be zero, the result will be 0 in that case, and we don't
1591 // have to check either of the buffers.
1592 if (stateZeroSize) {
1593 State = stateZeroSize;
1594 State = State->BindExpr(S: Call.getOriginExpr(), LCtx,
1595 V: Builder.makeZeroVal(type: Call.getResultType()));
1596 C.addTransition(State);
1597 }
1598
1599 // If the size can be nonzero, we have to check the other arguments.
1600 if (stateNonZeroSize) {
1601 State = stateNonZeroSize;
1602 // If we know the two buffers are the same, we know the result is 0.
1603 // First, get the two buffers' addresses. Another checker will have already
1604 // made sure they're not undefined.
1605 DefinedOrUnknownSVal LV =
1606 State->getSVal(Ex: Left.Expression, LCtx).castAs<DefinedOrUnknownSVal>();
1607 DefinedOrUnknownSVal RV =
1608 State->getSVal(Ex: Right.Expression, LCtx).castAs<DefinedOrUnknownSVal>();
1609
1610 // See if they are the same.
1611 ProgramStateRef SameBuffer, NotSameBuffer;
1612 std::tie(args&: SameBuffer, args&: NotSameBuffer) =
1613 State->assume(Cond: Builder.evalEQ(state: State, lhs: LV, rhs: RV));
1614
1615 // If the two arguments are the same buffer, we know the result is 0,
1616 // and we only need to check one size.
1617 if (SameBuffer && !NotSameBuffer) {
1618 State = SameBuffer;
1619 State = CheckBufferAccess(C, State, Buffer: Left, Size, Access: AccessKind::read);
1620 if (State) {
1621 State = SameBuffer->BindExpr(S: Call.getOriginExpr(), LCtx,
1622 V: Builder.makeZeroVal(type: Call.getResultType()));
1623 C.addTransition(State);
1624 }
1625 return;
1626 }
1627
1628 // If the two arguments might be different buffers, we have to check
1629 // the size of both of them.
1630 assert(NotSameBuffer);
1631 State = CheckBufferAccess(C, State, Buffer: Right, Size, Access: AccessKind::read, CK);
1632 State = CheckBufferAccess(C, State, Buffer: Left, Size, Access: AccessKind::read, CK);
1633 if (State) {
1634 // The return value is the comparison result, which we don't know.
1635 SVal CmpV = Builder.conjureSymbolVal(call: Call, visitCount: C.blockCount());
1636 State = State->BindExpr(S: Call.getOriginExpr(), LCtx, V: CmpV);
1637 C.addTransition(State);
1638 }
1639 }
1640}
1641
1642void CStringChecker::evalstrLength(CheckerContext &C,
1643 const CallEvent &Call) const {
1644 // size_t strlen(const char *s);
1645 evalstrLengthCommon(C, Call, /* IsStrnlen = */ false);
1646}
1647
1648void CStringChecker::evalstrnLength(CheckerContext &C,
1649 const CallEvent &Call) const {
1650 // size_t strnlen(const char *s, size_t maxlen);
1651 evalstrLengthCommon(C, Call, /* IsStrnlen = */ true);
1652}
1653
1654void CStringChecker::evalstrLengthCommon(CheckerContext &C,
1655 const CallEvent &Call,
1656 bool IsStrnlen) const {
1657 CurrentFunctionDescription = "string length function";
1658 ProgramStateRef state = C.getState();
1659 const LocationContext *LCtx = C.getLocationContext();
1660
1661 if (IsStrnlen) {
1662 const Expr *maxlenExpr = Call.getArgExpr(Index: 1);
1663 SVal maxlenVal = state->getSVal(Ex: maxlenExpr, LCtx);
1664
1665 ProgramStateRef stateZeroSize, stateNonZeroSize;
1666 std::tie(args&: stateZeroSize, args&: stateNonZeroSize) =
1667 assumeZero(C, State: state, V: maxlenVal, Ty: maxlenExpr->getType());
1668
1669 // If the size can be zero, the result will be 0 in that case, and we don't
1670 // have to check the string itself.
1671 if (stateZeroSize) {
1672 SVal zero = C.getSValBuilder().makeZeroVal(type: Call.getResultType());
1673 stateZeroSize = stateZeroSize->BindExpr(S: Call.getOriginExpr(), LCtx, V: zero);
1674 C.addTransition(State: stateZeroSize);
1675 }
1676
1677 // If the size is GUARANTEED to be zero, we're done!
1678 if (!stateNonZeroSize)
1679 return;
1680
1681 // Otherwise, record the assumption that the size is nonzero.
1682 state = stateNonZeroSize;
1683 }
1684
1685 // Check that the string argument is non-null.
1686 AnyArgExpr Arg = {.Expression: Call.getArgExpr(Index: 0), .ArgumentIndex: 0};
1687 SVal ArgVal = state->getSVal(Ex: Arg.Expression, LCtx);
1688 state = checkNonNull(C, State: state, Arg, l: ArgVal);
1689
1690 if (!state)
1691 return;
1692
1693 SVal strLength = getCStringLength(C, state, Ex: Arg.Expression, Buf: ArgVal);
1694
1695 // If the argument isn't a valid C string, there's no valid state to
1696 // transition to.
1697 if (strLength.isUndef())
1698 return;
1699
1700 DefinedOrUnknownSVal result = UnknownVal();
1701
1702 // If the check is for strnlen() then bind the return value to no more than
1703 // the maxlen value.
1704 if (IsStrnlen) {
1705 QualType cmpTy = C.getSValBuilder().getConditionType();
1706
1707 // It's a little unfortunate to be getting this again,
1708 // but it's not that expensive...
1709 const Expr *maxlenExpr = Call.getArgExpr(Index: 1);
1710 SVal maxlenVal = state->getSVal(Ex: maxlenExpr, LCtx);
1711
1712 std::optional<NonLoc> strLengthNL = strLength.getAs<NonLoc>();
1713 std::optional<NonLoc> maxlenValNL = maxlenVal.getAs<NonLoc>();
1714
1715 if (strLengthNL && maxlenValNL) {
1716 ProgramStateRef stateStringTooLong, stateStringNotTooLong;
1717
1718 // Check if the strLength is greater than the maxlen.
1719 std::tie(args&: stateStringTooLong, args&: stateStringNotTooLong) = state->assume(
1720 Cond: C.getSValBuilder()
1721 .evalBinOpNN(state, op: BO_GT, lhs: *strLengthNL, rhs: *maxlenValNL, resultTy: cmpTy)
1722 .castAs<DefinedOrUnknownSVal>());
1723
1724 if (stateStringTooLong && !stateStringNotTooLong) {
1725 // If the string is longer than maxlen, return maxlen.
1726 result = *maxlenValNL;
1727 } else if (stateStringNotTooLong && !stateStringTooLong) {
1728 // If the string is shorter than maxlen, return its length.
1729 result = *strLengthNL;
1730 }
1731 }
1732
1733 if (result.isUnknown()) {
1734 // If we don't have enough information for a comparison, there's
1735 // no guarantee the full string length will actually be returned.
1736 // All we know is the return value is the min of the string length
1737 // and the limit. This is better than nothing.
1738 result = C.getSValBuilder().conjureSymbolVal(call: Call, visitCount: C.blockCount());
1739 NonLoc resultNL = result.castAs<NonLoc>();
1740
1741 if (strLengthNL) {
1742 state = state->assume(Cond: C.getSValBuilder().evalBinOpNN(
1743 state, op: BO_LE, lhs: resultNL, rhs: *strLengthNL, resultTy: cmpTy)
1744 .castAs<DefinedOrUnknownSVal>(), Assumption: true);
1745 }
1746
1747 if (maxlenValNL) {
1748 state = state->assume(Cond: C.getSValBuilder().evalBinOpNN(
1749 state, op: BO_LE, lhs: resultNL, rhs: *maxlenValNL, resultTy: cmpTy)
1750 .castAs<DefinedOrUnknownSVal>(), Assumption: true);
1751 }
1752 }
1753
1754 } else {
1755 // This is a plain strlen(), not strnlen().
1756 result = strLength.castAs<DefinedOrUnknownSVal>();
1757
1758 // If we don't know the length of the string, conjure a return
1759 // value, so it can be used in constraints, at least.
1760 if (result.isUnknown()) {
1761 result = C.getSValBuilder().conjureSymbolVal(call: Call, visitCount: C.blockCount());
1762 }
1763 }
1764
1765 // Bind the return value.
1766 assert(!result.isUnknown() && "Should have conjured a value by now");
1767 state = state->BindExpr(S: Call.getOriginExpr(), LCtx, V: result);
1768 C.addTransition(State: state);
1769}
1770
1771void CStringChecker::evalStrcpy(CheckerContext &C,
1772 const CallEvent &Call) const {
1773 // char *strcpy(char *restrict dst, const char *restrict src);
1774 evalStrcpyCommon(C, Call,
1775 /* ReturnEnd = */ false,
1776 /* IsBounded = */ false,
1777 /* appendK = */ ConcatFnKind::none);
1778}
1779
1780void CStringChecker::evalStrncpy(CheckerContext &C,
1781 const CallEvent &Call) const {
1782 // char *strncpy(char *restrict dst, const char *restrict src, size_t n);
1783 evalStrcpyCommon(C, Call,
1784 /* ReturnEnd = */ false,
1785 /* IsBounded = */ true,
1786 /* appendK = */ ConcatFnKind::none);
1787}
1788
1789void CStringChecker::evalStpcpy(CheckerContext &C,
1790 const CallEvent &Call) const {
1791 // char *stpcpy(char *restrict dst, const char *restrict src);
1792 evalStrcpyCommon(C, Call,
1793 /* ReturnEnd = */ true,
1794 /* IsBounded = */ false,
1795 /* appendK = */ ConcatFnKind::none);
1796}
1797
1798void CStringChecker::evalStrlcpy(CheckerContext &C,
1799 const CallEvent &Call) const {
1800 // size_t strlcpy(char *dest, const char *src, size_t size);
1801 evalStrcpyCommon(C, Call,
1802 /* ReturnEnd = */ true,
1803 /* IsBounded = */ true,
1804 /* appendK = */ ConcatFnKind::none,
1805 /* returnPtr = */ false);
1806}
1807
1808void CStringChecker::evalStrcat(CheckerContext &C,
1809 const CallEvent &Call) const {
1810 // char *strcat(char *restrict s1, const char *restrict s2);
1811 evalStrcpyCommon(C, Call,
1812 /* ReturnEnd = */ false,
1813 /* IsBounded = */ false,
1814 /* appendK = */ ConcatFnKind::strcat);
1815}
1816
1817void CStringChecker::evalStrncat(CheckerContext &C,
1818 const CallEvent &Call) const {
1819 // char *strncat(char *restrict s1, const char *restrict s2, size_t n);
1820 evalStrcpyCommon(C, Call,
1821 /* ReturnEnd = */ false,
1822 /* IsBounded = */ true,
1823 /* appendK = */ ConcatFnKind::strcat);
1824}
1825
1826void CStringChecker::evalStrlcat(CheckerContext &C,
1827 const CallEvent &Call) const {
1828 // size_t strlcat(char *dst, const char *src, size_t size);
1829 // It will append at most size - strlen(dst) - 1 bytes,
1830 // NULL-terminating the result.
1831 evalStrcpyCommon(C, Call,
1832 /* ReturnEnd = */ false,
1833 /* IsBounded = */ true,
1834 /* appendK = */ ConcatFnKind::strlcat,
1835 /* returnPtr = */ false);
1836}
1837
1838void CStringChecker::evalStrcpyCommon(CheckerContext &C, const CallEvent &Call,
1839 bool ReturnEnd, bool IsBounded,
1840 ConcatFnKind appendK,
1841 bool returnPtr) const {
1842 if (appendK == ConcatFnKind::none)
1843 CurrentFunctionDescription = "string copy function";
1844 else
1845 CurrentFunctionDescription = "string concatenation function";
1846
1847 ProgramStateRef state = C.getState();
1848 const LocationContext *LCtx = C.getLocationContext();
1849
1850 // Check that the destination is non-null.
1851 DestinationArgExpr Dst = {{.Expression: Call.getArgExpr(Index: 0), .ArgumentIndex: 0}};
1852 SVal DstVal = state->getSVal(Ex: Dst.Expression, LCtx);
1853 state = checkNonNull(C, State: state, Arg: Dst, l: DstVal);
1854 if (!state)
1855 return;
1856
1857 // Check that the source is non-null.
1858 SourceArgExpr srcExpr = {{.Expression: Call.getArgExpr(Index: 1), .ArgumentIndex: 1}};
1859 SVal srcVal = state->getSVal(Ex: srcExpr.Expression, LCtx);
1860 state = checkNonNull(C, State: state, Arg: srcExpr, l: srcVal);
1861 if (!state)
1862 return;
1863
1864 // Get the string length of the source.
1865 SVal strLength = getCStringLength(C, state, Ex: srcExpr.Expression, Buf: srcVal);
1866 std::optional<NonLoc> strLengthNL = strLength.getAs<NonLoc>();
1867
1868 // Get the string length of the destination buffer.
1869 SVal dstStrLength = getCStringLength(C, state, Ex: Dst.Expression, Buf: DstVal);
1870 std::optional<NonLoc> dstStrLengthNL = dstStrLength.getAs<NonLoc>();
1871
1872 // If the source isn't a valid C string, give up.
1873 if (strLength.isUndef())
1874 return;
1875
1876 SValBuilder &svalBuilder = C.getSValBuilder();
1877 QualType cmpTy = svalBuilder.getConditionType();
1878 QualType sizeTy = svalBuilder.getContext().getSizeType();
1879
1880 // These two values allow checking two kinds of errors:
1881 // - actual overflows caused by a source that doesn't fit in the destination
1882 // - potential overflows caused by a bound that could exceed the destination
1883 SVal amountCopied = UnknownVal();
1884 SVal maxLastElementIndex = UnknownVal();
1885 const char *boundWarning = nullptr;
1886
1887 // FIXME: Why do we choose the srcExpr if the access has no size?
1888 // Note that the 3rd argument of the call would be the size parameter.
1889 SizeArgExpr SrcExprAsSizeDummy = {
1890 {.Expression: srcExpr.Expression, .ArgumentIndex: srcExpr.ArgumentIndex}};
1891 state = CheckOverlap(
1892 C, state,
1893 Size: (IsBounded ? SizeArgExpr{{.Expression: Call.getArgExpr(Index: 2), .ArgumentIndex: 2}} : SrcExprAsSizeDummy),
1894 First: Dst, Second: srcExpr);
1895
1896 if (!state)
1897 return;
1898
1899 // If the function is strncpy, strncat, etc... it is bounded.
1900 if (IsBounded) {
1901 // Get the max number of characters to copy.
1902 SizeArgExpr lenExpr = {{.Expression: Call.getArgExpr(Index: 2), .ArgumentIndex: 2}};
1903 SVal lenVal = state->getSVal(Ex: lenExpr.Expression, LCtx);
1904
1905 // Protect against misdeclared strncpy().
1906 lenVal =
1907 svalBuilder.evalCast(V: lenVal, CastTy: sizeTy, OriginalTy: lenExpr.Expression->getType());
1908
1909 std::optional<NonLoc> lenValNL = lenVal.getAs<NonLoc>();
1910
1911 // If we know both values, we might be able to figure out how much
1912 // we're copying.
1913 if (strLengthNL && lenValNL) {
1914 switch (appendK) {
1915 case ConcatFnKind::none:
1916 case ConcatFnKind::strcat: {
1917 ProgramStateRef stateSourceTooLong, stateSourceNotTooLong;
1918 // Check if the max number to copy is less than the length of the src.
1919 // If the bound is equal to the source length, strncpy won't null-
1920 // terminate the result!
1921 std::tie(args&: stateSourceTooLong, args&: stateSourceNotTooLong) = state->assume(
1922 Cond: svalBuilder
1923 .evalBinOpNN(state, op: BO_GE, lhs: *strLengthNL, rhs: *lenValNL, resultTy: cmpTy)
1924 .castAs<DefinedOrUnknownSVal>());
1925
1926 if (stateSourceTooLong && !stateSourceNotTooLong) {
1927 // Max number to copy is less than the length of the src, so the
1928 // actual strLength copied is the max number arg.
1929 state = stateSourceTooLong;
1930 amountCopied = lenVal;
1931
1932 } else if (!stateSourceTooLong && stateSourceNotTooLong) {
1933 // The source buffer entirely fits in the bound.
1934 state = stateSourceNotTooLong;
1935 amountCopied = strLength;
1936 }
1937 break;
1938 }
1939 case ConcatFnKind::strlcat:
1940 if (!dstStrLengthNL)
1941 return;
1942
1943 // amountCopied = min (size - dstLen - 1 , srcLen)
1944 SVal freeSpace = svalBuilder.evalBinOpNN(state, op: BO_Sub, lhs: *lenValNL,
1945 rhs: *dstStrLengthNL, resultTy: sizeTy);
1946 if (!isa<NonLoc>(Val: freeSpace))
1947 return;
1948 freeSpace =
1949 svalBuilder.evalBinOp(state, op: BO_Sub, lhs: freeSpace,
1950 rhs: svalBuilder.makeIntVal(integer: 1, type: sizeTy), type: sizeTy);
1951 std::optional<NonLoc> freeSpaceNL = freeSpace.getAs<NonLoc>();
1952
1953 // While unlikely, it is possible that the subtraction is
1954 // too complex to compute, let's check whether it succeeded.
1955 if (!freeSpaceNL)
1956 return;
1957 SVal hasEnoughSpace = svalBuilder.evalBinOpNN(
1958 state, op: BO_LE, lhs: *strLengthNL, rhs: *freeSpaceNL, resultTy: cmpTy);
1959
1960 ProgramStateRef TrueState, FalseState;
1961 std::tie(args&: TrueState, args&: FalseState) =
1962 state->assume(Cond: hasEnoughSpace.castAs<DefinedOrUnknownSVal>());
1963
1964 // srcStrLength <= size - dstStrLength -1
1965 if (TrueState && !FalseState) {
1966 amountCopied = strLength;
1967 }
1968
1969 // srcStrLength > size - dstStrLength -1
1970 if (!TrueState && FalseState) {
1971 amountCopied = freeSpace;
1972 }
1973
1974 if (TrueState && FalseState)
1975 amountCopied = UnknownVal();
1976 break;
1977 }
1978 }
1979 // We still want to know if the bound is known to be too large.
1980 if (lenValNL) {
1981 switch (appendK) {
1982 case ConcatFnKind::strcat:
1983 // For strncat, the check is strlen(dst) + lenVal < sizeof(dst)
1984
1985 // Get the string length of the destination. If the destination is
1986 // memory that can't have a string length, we shouldn't be copying
1987 // into it anyway.
1988 if (dstStrLength.isUndef())
1989 return;
1990
1991 if (dstStrLengthNL) {
1992 maxLastElementIndex = svalBuilder.evalBinOpNN(
1993 state, op: BO_Add, lhs: *lenValNL, rhs: *dstStrLengthNL, resultTy: sizeTy);
1994
1995 boundWarning = "Size argument is greater than the free space in the "
1996 "destination buffer";
1997 }
1998 break;
1999 case ConcatFnKind::none:
2000 case ConcatFnKind::strlcat:
2001 // For strncpy and strlcat, this is just checking
2002 // that lenVal <= sizeof(dst).
2003 // (Yes, strncpy and strncat differ in how they treat termination.
2004 // strncat ALWAYS terminates, but strncpy doesn't.)
2005
2006 // We need a special case for when the copy size is zero, in which
2007 // case strncpy will do no work at all. Our bounds check uses n-1
2008 // as the last element accessed, so n == 0 is problematic.
2009 ProgramStateRef StateZeroSize, StateNonZeroSize;
2010 std::tie(args&: StateZeroSize, args&: StateNonZeroSize) =
2011 assumeZero(C, State: state, V: *lenValNL, Ty: sizeTy);
2012
2013 // If the size is known to be zero, we're done.
2014 if (StateZeroSize && !StateNonZeroSize) {
2015 if (returnPtr) {
2016 StateZeroSize =
2017 StateZeroSize->BindExpr(S: Call.getOriginExpr(), LCtx, V: DstVal);
2018 } else {
2019 if (appendK == ConcatFnKind::none) {
2020 // strlcpy returns strlen(src)
2021 StateZeroSize = StateZeroSize->BindExpr(S: Call.getOriginExpr(),
2022 LCtx, V: strLength);
2023 } else {
2024 // strlcat returns strlen(src) + strlen(dst)
2025 SVal retSize = svalBuilder.evalBinOp(
2026 state, op: BO_Add, lhs: strLength, rhs: dstStrLength, type: sizeTy);
2027 StateZeroSize =
2028 StateZeroSize->BindExpr(S: Call.getOriginExpr(), LCtx, V: retSize);
2029 }
2030 }
2031 C.addTransition(State: StateZeroSize);
2032 return;
2033 }
2034
2035 // Otherwise, go ahead and figure out the last element we'll touch.
2036 // We don't record the non-zero assumption here because we can't
2037 // be sure. We won't warn on a possible zero.
2038 NonLoc one = svalBuilder.makeIntVal(integer: 1, type: sizeTy).castAs<NonLoc>();
2039 maxLastElementIndex =
2040 svalBuilder.evalBinOpNN(state, op: BO_Sub, lhs: *lenValNL, rhs: one, resultTy: sizeTy);
2041 boundWarning = "Size argument is greater than the length of the "
2042 "destination buffer";
2043 break;
2044 }
2045 }
2046 } else {
2047 // The function isn't bounded. The amount copied should match the length
2048 // of the source buffer.
2049 amountCopied = strLength;
2050 }
2051
2052 assert(state);
2053
2054 // This represents the number of characters copied into the destination
2055 // buffer. (It may not actually be the strlen if the destination buffer
2056 // is not terminated.)
2057 SVal finalStrLength = UnknownVal();
2058 SVal strlRetVal = UnknownVal();
2059
2060 if (appendK == ConcatFnKind::none && !returnPtr) {
2061 // strlcpy returns the sizeof(src)
2062 strlRetVal = strLength;
2063 }
2064
2065 // If this is an appending function (strcat, strncat...) then set the
2066 // string length to strlen(src) + strlen(dst) since the buffer will
2067 // ultimately contain both.
2068 if (appendK != ConcatFnKind::none) {
2069 // Get the string length of the destination. If the destination is memory
2070 // that can't have a string length, we shouldn't be copying into it anyway.
2071 if (dstStrLength.isUndef())
2072 return;
2073
2074 if (appendK == ConcatFnKind::strlcat && dstStrLengthNL && strLengthNL) {
2075 strlRetVal = svalBuilder.evalBinOpNN(state, op: BO_Add, lhs: *strLengthNL,
2076 rhs: *dstStrLengthNL, resultTy: sizeTy);
2077 }
2078
2079 std::optional<NonLoc> amountCopiedNL = amountCopied.getAs<NonLoc>();
2080
2081 // If we know both string lengths, we might know the final string length.
2082 if (amountCopiedNL && dstStrLengthNL) {
2083 // Make sure the two lengths together don't overflow a size_t.
2084 state = checkAdditionOverflow(C, state, left: *amountCopiedNL, right: *dstStrLengthNL);
2085 if (!state)
2086 return;
2087
2088 finalStrLength = svalBuilder.evalBinOpNN(state, op: BO_Add, lhs: *amountCopiedNL,
2089 rhs: *dstStrLengthNL, resultTy: sizeTy);
2090 }
2091
2092 // If we couldn't get a single value for the final string length,
2093 // we can at least bound it by the individual lengths.
2094 if (finalStrLength.isUnknown()) {
2095 // Try to get a "hypothetical" string length symbol, which we can later
2096 // set as a real value if that turns out to be the case.
2097 finalStrLength =
2098 getCStringLength(C, state, Ex: Call.getOriginExpr(), Buf: DstVal, hypothetical: true);
2099 assert(!finalStrLength.isUndef());
2100
2101 if (std::optional<NonLoc> finalStrLengthNL =
2102 finalStrLength.getAs<NonLoc>()) {
2103 if (amountCopiedNL && appendK == ConcatFnKind::none) {
2104 // we overwrite dst string with the src
2105 // finalStrLength >= srcStrLength
2106 SVal sourceInResult = svalBuilder.evalBinOpNN(
2107 state, op: BO_GE, lhs: *finalStrLengthNL, rhs: *amountCopiedNL, resultTy: cmpTy);
2108 state = state->assume(Cond: sourceInResult.castAs<DefinedOrUnknownSVal>(),
2109 Assumption: true);
2110 if (!state)
2111 return;
2112 }
2113
2114 if (dstStrLengthNL && appendK != ConcatFnKind::none) {
2115 // we extend the dst string with the src
2116 // finalStrLength >= dstStrLength
2117 SVal destInResult = svalBuilder.evalBinOpNN(state, op: BO_GE,
2118 lhs: *finalStrLengthNL,
2119 rhs: *dstStrLengthNL,
2120 resultTy: cmpTy);
2121 state =
2122 state->assume(Cond: destInResult.castAs<DefinedOrUnknownSVal>(), Assumption: true);
2123 if (!state)
2124 return;
2125 }
2126 }
2127 }
2128
2129 } else {
2130 // Otherwise, this is a copy-over function (strcpy, strncpy, ...), and
2131 // the final string length will match the input string length.
2132 finalStrLength = amountCopied;
2133 }
2134
2135 SVal Result;
2136
2137 if (returnPtr) {
2138 // The final result of the function will either be a pointer past the last
2139 // copied element, or a pointer to the start of the destination buffer.
2140 Result = (ReturnEnd ? UnknownVal() : DstVal);
2141 } else {
2142 if (appendK == ConcatFnKind::strlcat || appendK == ConcatFnKind::none)
2143 //strlcpy, strlcat
2144 Result = strlRetVal;
2145 else
2146 Result = finalStrLength;
2147 }
2148
2149 assert(state);
2150
2151 // If the destination is a MemRegion, try to check for a buffer overflow and
2152 // record the new string length.
2153 if (std::optional<loc::MemRegionVal> dstRegVal =
2154 DstVal.getAs<loc::MemRegionVal>()) {
2155 QualType ptrTy = Dst.Expression->getType();
2156
2157 // If we have an exact value on a bounded copy, use that to check for
2158 // overflows, rather than our estimate about how much is actually copied.
2159 if (std::optional<NonLoc> maxLastNL = maxLastElementIndex.getAs<NonLoc>()) {
2160 SVal maxLastElement =
2161 svalBuilder.evalBinOpLN(state, op: BO_Add, lhs: *dstRegVal, rhs: *maxLastNL, resultTy: ptrTy);
2162
2163 // Check if the first byte of the destination is writable.
2164 state = CheckLocation(C, state, Buffer: Dst, Element: DstVal, Access: AccessKind::write);
2165 if (!state)
2166 return;
2167 // Check if the last byte of the destination is writable.
2168 state = CheckLocation(C, state, Buffer: Dst, Element: maxLastElement, Access: AccessKind::write);
2169 if (!state)
2170 return;
2171 }
2172
2173 // Then, if the final length is known...
2174 if (std::optional<NonLoc> knownStrLength = finalStrLength.getAs<NonLoc>()) {
2175 SVal lastElement = svalBuilder.evalBinOpLN(state, op: BO_Add, lhs: *dstRegVal,
2176 rhs: *knownStrLength, resultTy: ptrTy);
2177
2178 // ...and we haven't checked the bound, we'll check the actual copy.
2179 if (!boundWarning) {
2180 // Check if the first byte of the destination is writable.
2181 state = CheckLocation(C, state, Buffer: Dst, Element: DstVal, Access: AccessKind::write);
2182 if (!state)
2183 return;
2184 // Check if the last byte of the destination is writable.
2185 state = CheckLocation(C, state, Buffer: Dst, Element: lastElement, Access: AccessKind::write);
2186 if (!state)
2187 return;
2188 }
2189
2190 // If this is a stpcpy-style copy, the last element is the return value.
2191 if (returnPtr && ReturnEnd)
2192 Result = lastElement;
2193 }
2194
2195 // For bounded method, amountCopied take the minimum of two values,
2196 // for ConcatFnKind::strlcat:
2197 // amountCopied = min (size - dstLen - 1 , srcLen)
2198 // for others:
2199 // amountCopied = min (srcLen, size)
2200 // So even if we don't know about amountCopied, as long as one of them will
2201 // not cause an out-of-bound access, the whole function's operation will not
2202 // too, that will avoid invalidating the superRegion of data member in that
2203 // situation.
2204 bool CouldAccessOutOfBound = true;
2205 if (IsBounded && amountCopied.isUnknown()) {
2206 auto CouldAccessOutOfBoundForSVal =
2207 [&](std::optional<NonLoc> Val) -> bool {
2208 if (!Val)
2209 return true;
2210 return !isFirstBufInBound(C, State: state, BufVal: C.getSVal(S: Dst.Expression),
2211 BufTy: Dst.Expression->getType(), LengthVal: *Val,
2212 LengthTy: C.getASTContext().getSizeType());
2213 };
2214
2215 CouldAccessOutOfBound = CouldAccessOutOfBoundForSVal(strLengthNL);
2216
2217 if (CouldAccessOutOfBound) {
2218 // Get the max number of characters to copy.
2219 const Expr *LenExpr = Call.getArgExpr(Index: 2);
2220 SVal LenVal = state->getSVal(Ex: LenExpr, LCtx);
2221
2222 // Protect against misdeclared strncpy().
2223 LenVal = svalBuilder.evalCast(V: LenVal, CastTy: sizeTy, OriginalTy: LenExpr->getType());
2224
2225 // Because analyzer doesn't handle expressions like `size -
2226 // dstLen - 1` very well, we roughly use `size` for
2227 // ConcatFnKind::strlcat here, same with other concat kinds.
2228 CouldAccessOutOfBound =
2229 CouldAccessOutOfBoundForSVal(LenVal.getAs<NonLoc>());
2230 }
2231 }
2232
2233 // Invalidate the destination (regular invalidation without pointer-escaping
2234 // the address of the top-level region). This must happen before we set the
2235 // C string length because invalidation will clear the length.
2236 // FIXME: Even if we can't perfectly model the copy, we should see if we
2237 // can use LazyCompoundVals to copy the source values into the destination.
2238 // This would probably remove any existing bindings past the end of the
2239 // string, but that's still an improvement over blank invalidation.
2240 if (CouldAccessOutOfBound)
2241 state = invalidateDestinationBufferBySize(
2242 C, S: state, BufE: Dst.Expression, Elem: Call.getCFGElementRef(), BufV: *dstRegVal,
2243 SizeV: amountCopied, SizeTy: C.getASTContext().getSizeType());
2244 else
2245 state = invalidateDestinationBufferNeverOverflows(
2246 C, S: state, Elem: Call.getCFGElementRef(), BufV: *dstRegVal);
2247
2248 // Invalidate the source (const-invalidation without const-pointer-escaping
2249 // the address of the top-level region).
2250 state = invalidateSourceBuffer(C, S: state, Elem: Call.getCFGElementRef(), BufV: srcVal);
2251
2252 // Set the C string length of the destination, if we know it.
2253 if (IsBounded && (appendK == ConcatFnKind::none)) {
2254 // strncpy is annoying in that it doesn't guarantee to null-terminate
2255 // the result string. If the original string didn't fit entirely inside
2256 // the bound (including the null-terminator), we don't know how long the
2257 // result is.
2258 if (amountCopied != strLength)
2259 finalStrLength = UnknownVal();
2260 }
2261 state = setCStringLength(state, MR: dstRegVal->getRegion(), strLength: finalStrLength);
2262 }
2263
2264 assert(state);
2265
2266 if (returnPtr) {
2267 // If this is a stpcpy-style copy, but we were unable to check for a buffer
2268 // overflow, we still need a result. Conjure a return value.
2269 if (ReturnEnd && Result.isUnknown()) {
2270 Result = svalBuilder.conjureSymbolVal(call: Call, visitCount: C.blockCount());
2271 }
2272 }
2273 // Set the return value.
2274 state = state->BindExpr(S: Call.getOriginExpr(), LCtx, V: Result);
2275 C.addTransition(State: state);
2276}
2277
2278void CStringChecker::evalStrxfrm(CheckerContext &C,
2279 const CallEvent &Call) const {
2280 // size_t strxfrm(char *dest, const char *src, size_t n);
2281 CurrentFunctionDescription = "locale transformation function";
2282
2283 ProgramStateRef State = C.getState();
2284 const LocationContext *LCtx = C.getLocationContext();
2285 SValBuilder &SVB = C.getSValBuilder();
2286
2287 // Get arguments
2288 DestinationArgExpr Dest = {{.Expression: Call.getArgExpr(Index: 0), .ArgumentIndex: 0}};
2289 SourceArgExpr Source = {{.Expression: Call.getArgExpr(Index: 1), .ArgumentIndex: 1}};
2290 SizeArgExpr Size = {{.Expression: Call.getArgExpr(Index: 2), .ArgumentIndex: 2}};
2291
2292 // `src` can never be null
2293 SVal SrcVal = State->getSVal(Ex: Source.Expression, LCtx);
2294 State = checkNonNull(C, State, Arg: Source, l: SrcVal);
2295 if (!State)
2296 return;
2297
2298 // Buffer must not overlap
2299 State = CheckOverlap(C, state: State, Size, First: Dest, Second: Source, CK: CK_Regular);
2300 if (!State)
2301 return;
2302
2303 // The function returns an implementation-defined length needed for
2304 // transformation
2305 SVal RetVal = SVB.conjureSymbolVal(call: Call, visitCount: C.blockCount());
2306
2307 auto BindReturnAndTransition = [&RetVal, &Call, LCtx,
2308 &C](ProgramStateRef State) {
2309 if (State) {
2310 State = State->BindExpr(S: Call.getOriginExpr(), LCtx, V: RetVal);
2311 C.addTransition(State);
2312 }
2313 };
2314
2315 // Check if size is zero
2316 SVal SizeVal = State->getSVal(Ex: Size.Expression, LCtx);
2317 QualType SizeTy = Size.Expression->getType();
2318
2319 auto [StateZeroSize, StateSizeNonZero] =
2320 assumeZero(C, State, V: SizeVal, Ty: SizeTy);
2321
2322 // We can't assume anything about size, just bind the return value and be done
2323 if (!StateZeroSize && !StateSizeNonZero)
2324 return BindReturnAndTransition(State);
2325
2326 // If `n` is 0, we just return the implementation defined length
2327 if (StateZeroSize && !StateSizeNonZero)
2328 return BindReturnAndTransition(StateZeroSize);
2329
2330 // If `n` is not 0, `dest` can not be null.
2331 SVal DestVal = StateSizeNonZero->getSVal(Ex: Dest.Expression, LCtx);
2332 StateSizeNonZero = checkNonNull(C, State: StateSizeNonZero, Arg: Dest, l: DestVal);
2333 if (!StateSizeNonZero)
2334 return;
2335
2336 // Check that we can write to the destination buffer
2337 StateSizeNonZero = CheckBufferAccess(C, State: StateSizeNonZero, Buffer: Dest, Size,
2338 Access: AccessKind::write, CK: CK_Regular);
2339 if (!StateSizeNonZero)
2340 return;
2341
2342 // Success: return value < `n`
2343 // Failure: return value >= `n`
2344 auto ComparisonVal = SVB.evalBinOp(state: StateSizeNonZero, op: BO_LT, lhs: RetVal, rhs: SizeVal,
2345 type: SVB.getConditionType())
2346 .getAs<DefinedOrUnknownSVal>();
2347 if (!ComparisonVal) {
2348 // Fallback: invalidate the buffer.
2349 StateSizeNonZero = invalidateDestinationBufferBySize(
2350 C, S: StateSizeNonZero, BufE: Dest.Expression, Elem: Call.getCFGElementRef(), BufV: DestVal,
2351 SizeV: SizeVal, SizeTy: Size.Expression->getType());
2352 return BindReturnAndTransition(StateSizeNonZero);
2353 }
2354
2355 auto [StateSuccess, StateFailure] = StateSizeNonZero->assume(Cond: *ComparisonVal);
2356
2357 if (StateSuccess) {
2358 // The transformation invalidated the buffer.
2359 StateSuccess = invalidateDestinationBufferBySize(
2360 C, S: StateSuccess, BufE: Dest.Expression, Elem: Call.getCFGElementRef(), BufV: DestVal,
2361 SizeV: SizeVal, SizeTy: Size.Expression->getType());
2362 BindReturnAndTransition(StateSuccess);
2363 // Fallthrough: We also want to add a transition to the failure state below.
2364 }
2365
2366 if (StateFailure) {
2367 // `dest` buffer content is undefined
2368 if (auto DestLoc = DestVal.getAs<loc::MemRegionVal>()) {
2369 StateFailure = StateFailure->killBinding(LV: *DestLoc);
2370 StateFailure =
2371 StateFailure->bindDefaultInitial(loc: *DestLoc, V: UndefinedVal{}, LCtx);
2372 }
2373
2374 BindReturnAndTransition(StateFailure);
2375 }
2376}
2377
2378void CStringChecker::evalStrcmp(CheckerContext &C,
2379 const CallEvent &Call) const {
2380 //int strcmp(const char *s1, const char *s2);
2381 evalStrcmpCommon(C, Call, /* IsBounded = */ false, /* IgnoreCase = */ false);
2382}
2383
2384void CStringChecker::evalStrncmp(CheckerContext &C,
2385 const CallEvent &Call) const {
2386 //int strncmp(const char *s1, const char *s2, size_t n);
2387 evalStrcmpCommon(C, Call, /* IsBounded = */ true, /* IgnoreCase = */ false);
2388}
2389
2390void CStringChecker::evalStrcasecmp(CheckerContext &C,
2391 const CallEvent &Call) const {
2392 //int strcasecmp(const char *s1, const char *s2);
2393 evalStrcmpCommon(C, Call, /* IsBounded = */ false, /* IgnoreCase = */ true);
2394}
2395
2396void CStringChecker::evalStrncasecmp(CheckerContext &C,
2397 const CallEvent &Call) const {
2398 //int strncasecmp(const char *s1, const char *s2, size_t n);
2399 evalStrcmpCommon(C, Call, /* IsBounded = */ true, /* IgnoreCase = */ true);
2400}
2401
2402void CStringChecker::evalStrcmpCommon(CheckerContext &C, const CallEvent &Call,
2403 bool IsBounded, bool IgnoreCase) const {
2404 CurrentFunctionDescription = "string comparison function";
2405 ProgramStateRef state = C.getState();
2406 const LocationContext *LCtx = C.getLocationContext();
2407
2408 // Check that the first string is non-null
2409 AnyArgExpr Left = {.Expression: Call.getArgExpr(Index: 0), .ArgumentIndex: 0};
2410 SVal LeftVal = state->getSVal(Ex: Left.Expression, LCtx);
2411 state = checkNonNull(C, State: state, Arg: Left, l: LeftVal);
2412 if (!state)
2413 return;
2414
2415 // Check that the second string is non-null.
2416 AnyArgExpr Right = {.Expression: Call.getArgExpr(Index: 1), .ArgumentIndex: 1};
2417 SVal RightVal = state->getSVal(Ex: Right.Expression, LCtx);
2418 state = checkNonNull(C, State: state, Arg: Right, l: RightVal);
2419 if (!state)
2420 return;
2421
2422 // Get the string length of the first string or give up.
2423 SVal LeftLength = getCStringLength(C, state, Ex: Left.Expression, Buf: LeftVal);
2424 if (LeftLength.isUndef())
2425 return;
2426
2427 // Get the string length of the second string or give up.
2428 SVal RightLength = getCStringLength(C, state, Ex: Right.Expression, Buf: RightVal);
2429 if (RightLength.isUndef())
2430 return;
2431
2432 // If we know the two buffers are the same, we know the result is 0.
2433 // First, get the two buffers' addresses. Another checker will have already
2434 // made sure they're not undefined.
2435 DefinedOrUnknownSVal LV = LeftVal.castAs<DefinedOrUnknownSVal>();
2436 DefinedOrUnknownSVal RV = RightVal.castAs<DefinedOrUnknownSVal>();
2437
2438 // See if they are the same.
2439 SValBuilder &svalBuilder = C.getSValBuilder();
2440 DefinedOrUnknownSVal SameBuf = svalBuilder.evalEQ(state, lhs: LV, rhs: RV);
2441 ProgramStateRef StSameBuf, StNotSameBuf;
2442 std::tie(args&: StSameBuf, args&: StNotSameBuf) = state->assume(Cond: SameBuf);
2443
2444 // If the two arguments might be the same buffer, we know the result is 0,
2445 // and we only need to check one size.
2446 if (StSameBuf) {
2447 StSameBuf =
2448 StSameBuf->BindExpr(S: Call.getOriginExpr(), LCtx,
2449 V: svalBuilder.makeZeroVal(type: Call.getResultType()));
2450 C.addTransition(State: StSameBuf);
2451
2452 // If the two arguments are GUARANTEED to be the same, we're done!
2453 if (!StNotSameBuf)
2454 return;
2455 }
2456
2457 assert(StNotSameBuf);
2458 state = StNotSameBuf;
2459
2460 // At this point we can go about comparing the two buffers.
2461 // For now, we only do this if they're both known string literals.
2462
2463 // Attempt to extract string literals from both expressions.
2464 const StringLiteral *LeftStrLiteral =
2465 getCStringLiteral(C, state, expr: Left.Expression, val: LeftVal);
2466 const StringLiteral *RightStrLiteral =
2467 getCStringLiteral(C, state, expr: Right.Expression, val: RightVal);
2468 bool canComputeResult = false;
2469 SVal resultVal = svalBuilder.conjureSymbolVal(call: Call, visitCount: C.blockCount());
2470
2471 if (LeftStrLiteral && RightStrLiteral) {
2472 StringRef LeftStrRef = LeftStrLiteral->getString();
2473 StringRef RightStrRef = RightStrLiteral->getString();
2474
2475 if (IsBounded) {
2476 // Get the max number of characters to compare.
2477 const Expr *lenExpr = Call.getArgExpr(Index: 2);
2478 SVal lenVal = state->getSVal(Ex: lenExpr, LCtx);
2479
2480 // If the length is known, we can get the right substrings.
2481 if (const llvm::APSInt *len = svalBuilder.getKnownValue(state, val: lenVal)) {
2482 // Create substrings of each to compare the prefix.
2483 LeftStrRef = LeftStrRef.substr(Start: 0, N: (size_t)len->getZExtValue());
2484 RightStrRef = RightStrRef.substr(Start: 0, N: (size_t)len->getZExtValue());
2485 canComputeResult = true;
2486 }
2487 } else {
2488 // This is a normal, unbounded strcmp.
2489 canComputeResult = true;
2490 }
2491
2492 if (canComputeResult) {
2493 // Real strcmp stops at null characters.
2494 size_t s1Term = LeftStrRef.find(C: '\0');
2495 if (s1Term != StringRef::npos)
2496 LeftStrRef = LeftStrRef.substr(Start: 0, N: s1Term);
2497
2498 size_t s2Term = RightStrRef.find(C: '\0');
2499 if (s2Term != StringRef::npos)
2500 RightStrRef = RightStrRef.substr(Start: 0, N: s2Term);
2501
2502 // Use StringRef's comparison methods to compute the actual result.
2503 int compareRes = IgnoreCase ? LeftStrRef.compare_insensitive(RHS: RightStrRef)
2504 : LeftStrRef.compare(RHS: RightStrRef);
2505
2506 // The strcmp function returns an integer greater than, equal to, or less
2507 // than zero, [c11, p7.24.4.2].
2508 if (compareRes == 0) {
2509 resultVal = svalBuilder.makeIntVal(integer: compareRes, type: Call.getResultType());
2510 }
2511 else {
2512 DefinedSVal zeroVal = svalBuilder.makeIntVal(integer: 0, type: Call.getResultType());
2513 // Constrain strcmp's result range based on the result of StringRef's
2514 // comparison methods.
2515 BinaryOperatorKind op = (compareRes > 0) ? BO_GT : BO_LT;
2516 SVal compareWithZero =
2517 svalBuilder.evalBinOp(state, op, lhs: resultVal, rhs: zeroVal,
2518 type: svalBuilder.getConditionType());
2519 DefinedSVal compareWithZeroVal = compareWithZero.castAs<DefinedSVal>();
2520 state = state->assume(Cond: compareWithZeroVal, Assumption: true);
2521 }
2522 }
2523 }
2524
2525 state = state->BindExpr(S: Call.getOriginExpr(), LCtx, V: resultVal);
2526
2527 // Record this as a possible path.
2528 C.addTransition(State: state);
2529}
2530
2531void CStringChecker::evalStrsep(CheckerContext &C,
2532 const CallEvent &Call) const {
2533 // char *strsep(char **stringp, const char *delim);
2534 // Verify whether the search string parameter matches the return type.
2535 SourceArgExpr SearchStrPtr = {{.Expression: Call.getArgExpr(Index: 0), .ArgumentIndex: 0}};
2536
2537 QualType CharPtrTy = SearchStrPtr.Expression->getType()->getPointeeType();
2538 if (CharPtrTy.isNull() || Call.getResultType().getUnqualifiedType() !=
2539 CharPtrTy.getUnqualifiedType())
2540 return;
2541
2542 CurrentFunctionDescription = "strsep()";
2543 ProgramStateRef State = C.getState();
2544 const LocationContext *LCtx = C.getLocationContext();
2545
2546 // Check that the search string pointer is non-null (though it may point to
2547 // a null string).
2548 SVal SearchStrVal = State->getSVal(Ex: SearchStrPtr.Expression, LCtx);
2549 State = checkNonNull(C, State, Arg: SearchStrPtr, l: SearchStrVal);
2550 if (!State)
2551 return;
2552
2553 // Check that the delimiter string is non-null.
2554 AnyArgExpr DelimStr = {.Expression: Call.getArgExpr(Index: 1), .ArgumentIndex: 1};
2555 SVal DelimStrVal = State->getSVal(Ex: DelimStr.Expression, LCtx);
2556 State = checkNonNull(C, State, Arg: DelimStr, l: DelimStrVal);
2557 if (!State)
2558 return;
2559
2560 SValBuilder &SVB = C.getSValBuilder();
2561 SVal Result;
2562 if (std::optional<Loc> SearchStrLoc = SearchStrVal.getAs<Loc>()) {
2563 // Get the current value of the search string pointer, as a char*.
2564 Result = State->getSVal(LV: *SearchStrLoc, T: CharPtrTy);
2565
2566 // Invalidate the search string, representing the change of one delimiter
2567 // character to NUL.
2568 // As the replacement never overflows, do not invalidate its super region.
2569 State = invalidateDestinationBufferNeverOverflows(
2570 C, S: State, Elem: Call.getCFGElementRef(), BufV: Result);
2571
2572 // Overwrite the search string pointer. The new value is either an address
2573 // further along in the same string, or NULL if there are no more tokens.
2574 State = State->bindLoc(location: *SearchStrLoc,
2575 V: SVB.conjureSymbolVal(call: Call, visitCount: C.blockCount(), symbolTag: getTag()),
2576 LCtx);
2577 } else {
2578 assert(SearchStrVal.isUnknown());
2579 // Conjure a symbolic value. It's the best we can do.
2580 Result = SVB.conjureSymbolVal(call: Call, visitCount: C.blockCount());
2581 }
2582
2583 // Set the return value, and finish.
2584 State = State->BindExpr(S: Call.getOriginExpr(), LCtx, V: Result);
2585 C.addTransition(State);
2586}
2587
2588// These should probably be moved into a C++ standard library checker.
2589void CStringChecker::evalStdCopy(CheckerContext &C,
2590 const CallEvent &Call) const {
2591 evalStdCopyCommon(C, Call);
2592}
2593
2594void CStringChecker::evalStdCopyBackward(CheckerContext &C,
2595 const CallEvent &Call) const {
2596 evalStdCopyCommon(C, Call);
2597}
2598
2599void CStringChecker::evalStdCopyCommon(CheckerContext &C,
2600 const CallEvent &Call) const {
2601 if (!Call.getArgExpr(Index: 2)->getType()->isPointerType())
2602 return;
2603
2604 ProgramStateRef State = C.getState();
2605
2606 const LocationContext *LCtx = C.getLocationContext();
2607
2608 // template <class _InputIterator, class _OutputIterator>
2609 // _OutputIterator
2610 // copy(_InputIterator __first, _InputIterator __last,
2611 // _OutputIterator __result)
2612
2613 // Invalidate the destination buffer
2614 const Expr *Dst = Call.getArgExpr(Index: 2);
2615 SVal DstVal = State->getSVal(Ex: Dst, LCtx);
2616 // FIXME: As we do not know how many items are copied, we also invalidate the
2617 // super region containing the target location.
2618 State = invalidateDestinationBufferAlwaysEscapeSuperRegion(
2619 C, S: State, Elem: Call.getCFGElementRef(), BufV: DstVal);
2620
2621 SValBuilder &SVB = C.getSValBuilder();
2622
2623 SVal ResultVal = SVB.conjureSymbolVal(call: Call, visitCount: C.blockCount());
2624 State = State->BindExpr(S: Call.getOriginExpr(), LCtx, V: ResultVal);
2625
2626 C.addTransition(State);
2627}
2628
2629void CStringChecker::evalMemset(CheckerContext &C,
2630 const CallEvent &Call) const {
2631 // void *memset(void *s, int c, size_t n);
2632 CurrentFunctionDescription = "memory set function";
2633
2634 DestinationArgExpr Buffer = {{.Expression: Call.getArgExpr(Index: 0), .ArgumentIndex: 0}};
2635 AnyArgExpr CharE = {.Expression: Call.getArgExpr(Index: 1), .ArgumentIndex: 1};
2636 SizeArgExpr Size = {{.Expression: Call.getArgExpr(Index: 2), .ArgumentIndex: 2}};
2637
2638 ProgramStateRef State = C.getState();
2639
2640 // See if the size argument is zero.
2641 const LocationContext *LCtx = C.getLocationContext();
2642 SVal SizeVal = C.getSVal(S: Size.Expression);
2643 QualType SizeTy = Size.Expression->getType();
2644
2645 ProgramStateRef ZeroSize, NonZeroSize;
2646 std::tie(args&: ZeroSize, args&: NonZeroSize) = assumeZero(C, State, V: SizeVal, Ty: SizeTy);
2647
2648 // Get the value of the memory area.
2649 SVal BufferPtrVal = C.getSVal(S: Buffer.Expression);
2650
2651 // If the size is zero, there won't be any actual memory access, so
2652 // just bind the return value to the buffer and return.
2653 if (ZeroSize && !NonZeroSize) {
2654 ZeroSize = ZeroSize->BindExpr(S: Call.getOriginExpr(), LCtx, V: BufferPtrVal);
2655 C.addTransition(State: ZeroSize);
2656 return;
2657 }
2658
2659 // Ensure the memory area is not null.
2660 // If it is NULL there will be a NULL pointer dereference.
2661 State = checkNonNull(C, State: NonZeroSize, Arg: Buffer, l: BufferPtrVal);
2662 if (!State)
2663 return;
2664
2665 State = CheckBufferAccess(C, State, Buffer, Size, Access: AccessKind::write);
2666 if (!State)
2667 return;
2668
2669 // According to the values of the arguments, bind the value of the second
2670 // argument to the destination buffer and set string length, or just
2671 // invalidate the destination buffer.
2672 if (!memsetAux(DstBuffer: Buffer.Expression, Elem: Call.getCFGElementRef(),
2673 CharVal: C.getSVal(S: CharE.Expression), Size: Size.Expression, C, State))
2674 return;
2675
2676 State = State->BindExpr(S: Call.getOriginExpr(), LCtx, V: BufferPtrVal);
2677 C.addTransition(State);
2678}
2679
2680void CStringChecker::evalBzero(CheckerContext &C, const CallEvent &Call) const {
2681 CurrentFunctionDescription = "memory clearance function";
2682
2683 DestinationArgExpr Buffer = {{.Expression: Call.getArgExpr(Index: 0), .ArgumentIndex: 0}};
2684 SizeArgExpr Size = {{.Expression: Call.getArgExpr(Index: 1), .ArgumentIndex: 1}};
2685 SVal Zero = C.getSValBuilder().makeZeroVal(type: C.getASTContext().IntTy);
2686
2687 ProgramStateRef State = C.getState();
2688
2689 // See if the size argument is zero.
2690 SVal SizeVal = C.getSVal(S: Size.Expression);
2691 QualType SizeTy = Size.Expression->getType();
2692
2693 ProgramStateRef StateZeroSize, StateNonZeroSize;
2694 std::tie(args&: StateZeroSize, args&: StateNonZeroSize) =
2695 assumeZero(C, State, V: SizeVal, Ty: SizeTy);
2696
2697 // If the size is zero, there won't be any actual memory access,
2698 // In this case we just return.
2699 if (StateZeroSize && !StateNonZeroSize) {
2700 C.addTransition(State: StateZeroSize);
2701 return;
2702 }
2703
2704 // Get the value of the memory area.
2705 SVal MemVal = C.getSVal(S: Buffer.Expression);
2706
2707 // Ensure the memory area is not null.
2708 // If it is NULL there will be a NULL pointer dereference.
2709 State = checkNonNull(C, State: StateNonZeroSize, Arg: Buffer, l: MemVal);
2710 if (!State)
2711 return;
2712
2713 State = CheckBufferAccess(C, State, Buffer, Size, Access: AccessKind::write);
2714 if (!State)
2715 return;
2716
2717 if (!memsetAux(DstBuffer: Buffer.Expression, Elem: Call.getCFGElementRef(), CharVal: Zero,
2718 Size: Size.Expression, C, State))
2719 return;
2720
2721 C.addTransition(State);
2722}
2723
2724void CStringChecker::evalSprintf(CheckerContext &C,
2725 const CallEvent &Call) const {
2726 CurrentFunctionDescription = "'sprintf'";
2727 evalSprintfCommon(C, Call, /* IsBounded = */ false);
2728}
2729
2730void CStringChecker::evalSnprintf(CheckerContext &C,
2731 const CallEvent &Call) const {
2732 CurrentFunctionDescription = "'snprintf'";
2733 evalSprintfCommon(C, Call, /* IsBounded = */ true);
2734}
2735
2736void CStringChecker::evalSprintfCommon(CheckerContext &C, const CallEvent &Call,
2737 bool IsBounded) const {
2738 ProgramStateRef State = C.getState();
2739 const auto *CE = cast<CallExpr>(Val: Call.getOriginExpr());
2740 DestinationArgExpr Dest = {{.Expression: Call.getArgExpr(Index: 0), .ArgumentIndex: 0}};
2741
2742 const auto NumParams = Call.parameters().size();
2743 if (CE->getNumArgs() < NumParams) {
2744 // This is an invalid call, let's just ignore it.
2745 return;
2746 }
2747
2748 const auto AllArguments =
2749 llvm::make_range(x: CE->getArgs(), y: CE->getArgs() + CE->getNumArgs());
2750 const auto VariadicArguments = drop_begin(RangeOrContainer: enumerate(First: AllArguments), N: NumParams);
2751
2752 for (const auto &[ArgIdx, ArgExpr] : VariadicArguments) {
2753 // We consider only string buffers
2754 if (const QualType type = ArgExpr->getType();
2755 !type->isAnyPointerType() ||
2756 !type->getPointeeType()->isAnyCharacterType())
2757 continue;
2758 SourceArgExpr Source = {{.Expression: ArgExpr, .ArgumentIndex: unsigned(ArgIdx)}};
2759
2760 // Ensure the buffers do not overlap.
2761 SizeArgExpr SrcExprAsSizeDummy = {
2762 {.Expression: Source.Expression, .ArgumentIndex: Source.ArgumentIndex}};
2763 State = CheckOverlap(
2764 C, state: State,
2765 Size: (IsBounded ? SizeArgExpr{{.Expression: Call.getArgExpr(Index: 1), .ArgumentIndex: 1}} : SrcExprAsSizeDummy),
2766 First: Dest, Second: Source);
2767 if (!State)
2768 return;
2769 }
2770
2771 C.addTransition(State);
2772}
2773
2774//===----------------------------------------------------------------------===//
2775// The driver method, and other Checker callbacks.
2776//===----------------------------------------------------------------------===//
2777
2778CStringChecker::FnCheck CStringChecker::identifyCall(const CallEvent &Call,
2779 CheckerContext &C) const {
2780 const auto *CE = dyn_cast_or_null<CallExpr>(Val: Call.getOriginExpr());
2781 if (!CE)
2782 return nullptr;
2783
2784 const FunctionDecl *FD = dyn_cast_or_null<FunctionDecl>(Val: Call.getDecl());
2785 if (!FD)
2786 return nullptr;
2787
2788 if (StdCopy.matches(Call))
2789 return &CStringChecker::evalStdCopy;
2790 if (StdCopyBackward.matches(Call))
2791 return &CStringChecker::evalStdCopyBackward;
2792
2793 // Pro-actively check that argument types are safe to do arithmetic upon.
2794 // We do not want to crash if someone accidentally passes a structure
2795 // into, say, a C++ overload of any of these functions. We could not check
2796 // that for std::copy because they may have arguments of other types.
2797 for (auto I : CE->arguments()) {
2798 QualType T = I->getType();
2799 if (!T->isIntegralOrEnumerationType() && !T->isPointerType())
2800 return nullptr;
2801 }
2802
2803 const FnCheck *Callback = Callbacks.lookup(Call);
2804 if (Callback)
2805 return *Callback;
2806
2807 return nullptr;
2808}
2809
2810bool CStringChecker::evalCall(const CallEvent &Call, CheckerContext &C) const {
2811 FnCheck Callback = identifyCall(Call, C);
2812
2813 // If the callee isn't a string function, let another checker handle it.
2814 if (!Callback)
2815 return false;
2816
2817 // Check and evaluate the call.
2818 assert(isa<CallExpr>(Call.getOriginExpr()));
2819 Callback(this, C, Call);
2820
2821 // If the evaluate call resulted in no change, chain to the next eval call
2822 // handler.
2823 // Note, the custom CString evaluation calls assume that basic safety
2824 // properties are held. However, if the user chooses to turn off some of these
2825 // checks, we ignore the issues and leave the call evaluation to a generic
2826 // handler.
2827 return C.isDifferent();
2828}
2829
2830void CStringChecker::checkPreStmt(const DeclStmt *DS, CheckerContext &C) const {
2831 // Record string length for char a[] = "abc";
2832 ProgramStateRef state = C.getState();
2833
2834 for (const auto *I : DS->decls()) {
2835 const VarDecl *D = dyn_cast<VarDecl>(Val: I);
2836 if (!D)
2837 continue;
2838
2839 // FIXME: Handle array fields of structs.
2840 if (!D->getType()->isArrayType())
2841 continue;
2842
2843 const Expr *Init = D->getInit();
2844 if (!Init)
2845 continue;
2846 if (!isa<StringLiteral>(Val: Init))
2847 continue;
2848
2849 Loc VarLoc = state->getLValue(VD: D, LC: C.getLocationContext());
2850 const MemRegion *MR = VarLoc.getAsRegion();
2851 if (!MR)
2852 continue;
2853
2854 SVal StrVal = C.getSVal(S: Init);
2855 assert(StrVal.isValid() && "Initializer string is unknown or undefined");
2856 DefinedOrUnknownSVal strLength =
2857 getCStringLength(C, state, Ex: Init, Buf: StrVal).castAs<DefinedOrUnknownSVal>();
2858
2859 state = state->set<CStringLength>(K: MR, E: strLength);
2860 }
2861
2862 C.addTransition(State: state);
2863}
2864
2865ProgramStateRef
2866CStringChecker::checkRegionChanges(ProgramStateRef state,
2867 const InvalidatedSymbols *,
2868 ArrayRef<const MemRegion *> ExplicitRegions,
2869 ArrayRef<const MemRegion *> Regions,
2870 const LocationContext *LCtx,
2871 const CallEvent *Call) const {
2872 CStringLengthTy Entries = state->get<CStringLength>();
2873 if (Entries.isEmpty())
2874 return state;
2875
2876 llvm::SmallPtrSet<const MemRegion *, 8> Invalidated;
2877 llvm::SmallPtrSet<const MemRegion *, 32> SuperRegions;
2878
2879 // First build sets for the changed regions and their super-regions.
2880 for (const MemRegion *MR : Regions) {
2881 Invalidated.insert(Ptr: MR);
2882
2883 SuperRegions.insert(Ptr: MR);
2884 while (const SubRegion *SR = dyn_cast<SubRegion>(Val: MR)) {
2885 MR = SR->getSuperRegion();
2886 SuperRegions.insert(Ptr: MR);
2887 }
2888 }
2889
2890 CStringLengthTy::Factory &F = state->get_context<CStringLength>();
2891
2892 // Then loop over the entries in the current state.
2893 for (const MemRegion *MR : llvm::make_first_range(c&: Entries)) {
2894 // Is this entry for a super-region of a changed region?
2895 if (SuperRegions.count(Ptr: MR)) {
2896 Entries = F.remove(Old: Entries, K: MR);
2897 continue;
2898 }
2899
2900 // Is this entry for a sub-region of a changed region?
2901 const MemRegion *Super = MR;
2902 while (const SubRegion *SR = dyn_cast<SubRegion>(Val: Super)) {
2903 Super = SR->getSuperRegion();
2904 if (Invalidated.count(Ptr: Super)) {
2905 Entries = F.remove(Old: Entries, K: MR);
2906 break;
2907 }
2908 }
2909 }
2910
2911 return state->set<CStringLength>(Entries);
2912}
2913
2914void CStringChecker::checkLiveSymbols(ProgramStateRef state,
2915 SymbolReaper &SR) const {
2916 // Mark all symbols in our string length map as valid.
2917 CStringLengthTy Entries = state->get<CStringLength>();
2918
2919 for (SVal Len : llvm::make_second_range(c&: Entries)) {
2920 for (SymbolRef Sym : Len.symbols())
2921 SR.markInUse(sym: Sym);
2922 }
2923}
2924
2925void CStringChecker::checkDeadSymbols(SymbolReaper &SR,
2926 CheckerContext &C) const {
2927 ProgramStateRef state = C.getState();
2928 CStringLengthTy Entries = state->get<CStringLength>();
2929 if (Entries.isEmpty())
2930 return;
2931
2932 CStringLengthTy::Factory &F = state->get_context<CStringLength>();
2933 for (auto [Reg, Len] : Entries) {
2934 if (SymbolRef Sym = Len.getAsSymbol()) {
2935 if (SR.isDead(sym: Sym))
2936 Entries = F.remove(Old: Entries, K: Reg);
2937 }
2938 }
2939
2940 state = state->set<CStringLength>(Entries);
2941 C.addTransition(State: state);
2942}
2943
2944void ento::registerCStringModeling(CheckerManager &Mgr) {
2945 // Other checker relies on the modeling implemented in this checker family,
2946 // so this "modeling checker" can register the 'CStringChecker' backend for
2947 // its callbacks without enabling any of its frontends.
2948 Mgr.getChecker<CStringChecker>();
2949}
2950
2951bool ento::shouldRegisterCStringModeling(const CheckerManager &) {
2952 return true;
2953}
2954
2955#define REGISTER_CHECKER(NAME) \
2956 void ento::registerCString##NAME(CheckerManager &Mgr) { \
2957 Mgr.getChecker<CStringChecker>()->NAME.enable(Mgr); \
2958 } \
2959 \
2960 bool ento::shouldRegisterCString##NAME(const CheckerManager &) { \
2961 return true; \
2962 }
2963
2964REGISTER_CHECKER(NullArg)
2965REGISTER_CHECKER(OutOfBounds)
2966REGISTER_CHECKER(BufferOverlap)
2967REGISTER_CHECKER(NotNullTerm)
2968REGISTER_CHECKER(UninitializedRead)
2969