1 //===--- SemaChecking.cpp - Extra Semantic Checking -----------------------===//
2 //
3 //                     The LLVM Compiler Infrastructure
4 //
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
7 //
8 //===----------------------------------------------------------------------===//
9 //
10 //  This file implements extra semantic analysis beyond what is enforced
11 //  by the C type system.
12 //
13 //===----------------------------------------------------------------------===//
14 
15 #include "clang/Sema/SemaInternal.h"
16 #include "clang/AST/ASTContext.h"
17 #include "clang/AST/CharUnits.h"
18 #include "clang/AST/DeclCXX.h"
19 #include "clang/AST/DeclObjC.h"
20 #include "clang/AST/EvaluatedExprVisitor.h"
21 #include "clang/AST/Expr.h"
22 #include "clang/AST/ExprCXX.h"
23 #include "clang/AST/ExprObjC.h"
24 #include "clang/AST/StmtCXX.h"
25 #include "clang/AST/StmtObjC.h"
26 #include "clang/Analysis/Analyses/FormatString.h"
27 #include "clang/Basic/CharInfo.h"
28 #include "clang/Basic/TargetBuiltins.h"
29 #include "clang/Basic/TargetInfo.h"
30 #include "clang/Lex/Lexer.h" // TODO: Extract static functions to fix layering.
31 #include "clang/Sema/Initialization.h"
32 #include "clang/Sema/Lookup.h"
33 #include "clang/Sema/ScopeInfo.h"
34 #include "clang/Sema/Sema.h"
35 #include "llvm/ADT/STLExtras.h"
36 #include "llvm/ADT/SmallBitVector.h"
37 #include "llvm/ADT/SmallString.h"
38 #include "llvm/Support/ConvertUTF.h"
39 #include "llvm/Support/raw_ostream.h"
40 #include <limits>
41 using namespace clang;
42 using namespace sema;
43 
getLocationOfStringLiteralByte(const StringLiteral * SL,unsigned ByteNo) const44 SourceLocation Sema::getLocationOfStringLiteralByte(const StringLiteral *SL,
45                                                     unsigned ByteNo) const {
46   return SL->getLocationOfByte(ByteNo, getSourceManager(), LangOpts,
47                                Context.getTargetInfo());
48 }
49 
50 /// Checks that a call expression's argument count is the desired number.
51 /// This is useful when doing custom type-checking.  Returns true on error.
checkArgCount(Sema & S,CallExpr * call,unsigned desiredArgCount)52 static bool checkArgCount(Sema &S, CallExpr *call, unsigned desiredArgCount) {
53   unsigned argCount = call->getNumArgs();
54   if (argCount == desiredArgCount) return false;
55 
56   if (argCount < desiredArgCount)
57     return S.Diag(call->getLocEnd(), diag::err_typecheck_call_too_few_args)
58         << 0 /*function call*/ << desiredArgCount << argCount
59         << call->getSourceRange();
60 
61   // Highlight all the excess arguments.
62   SourceRange range(call->getArg(desiredArgCount)->getLocStart(),
63                     call->getArg(argCount - 1)->getLocEnd());
64 
65   return S.Diag(range.getBegin(), diag::err_typecheck_call_too_many_args)
66     << 0 /*function call*/ << desiredArgCount << argCount
67     << call->getArg(1)->getSourceRange();
68 }
69 
70 /// Check that the first argument to __builtin_annotation is an integer
71 /// and the second argument is a non-wide string literal.
SemaBuiltinAnnotation(Sema & S,CallExpr * TheCall)72 static bool SemaBuiltinAnnotation(Sema &S, CallExpr *TheCall) {
73   if (checkArgCount(S, TheCall, 2))
74     return true;
75 
76   // First argument should be an integer.
77   Expr *ValArg = TheCall->getArg(0);
78   QualType Ty = ValArg->getType();
79   if (!Ty->isIntegerType()) {
80     S.Diag(ValArg->getLocStart(), diag::err_builtin_annotation_first_arg)
81       << ValArg->getSourceRange();
82     return true;
83   }
84 
85   // Second argument should be a constant string.
86   Expr *StrArg = TheCall->getArg(1)->IgnoreParenCasts();
87   StringLiteral *Literal = dyn_cast<StringLiteral>(StrArg);
88   if (!Literal || !Literal->isAscii()) {
89     S.Diag(StrArg->getLocStart(), diag::err_builtin_annotation_second_arg)
90       << StrArg->getSourceRange();
91     return true;
92   }
93 
94   TheCall->setType(Ty);
95   return false;
96 }
97 
98 /// Check that the argument to __builtin_addressof is a glvalue, and set the
99 /// result type to the corresponding pointer type.
SemaBuiltinAddressof(Sema & S,CallExpr * TheCall)100 static bool SemaBuiltinAddressof(Sema &S, CallExpr *TheCall) {
101   if (checkArgCount(S, TheCall, 1))
102     return true;
103 
104   ExprResult Arg(TheCall->getArg(0));
105   QualType ResultType = S.CheckAddressOfOperand(Arg, TheCall->getLocStart());
106   if (ResultType.isNull())
107     return true;
108 
109   TheCall->setArg(0, Arg.get());
110   TheCall->setType(ResultType);
111   return false;
112 }
113 
SemaBuiltinMemChkCall(Sema & S,FunctionDecl * FDecl,CallExpr * TheCall,unsigned SizeIdx,unsigned DstSizeIdx)114 static void SemaBuiltinMemChkCall(Sema &S, FunctionDecl *FDecl,
115 		                  CallExpr *TheCall, unsigned SizeIdx,
116                                   unsigned DstSizeIdx) {
117   if (TheCall->getNumArgs() <= SizeIdx ||
118       TheCall->getNumArgs() <= DstSizeIdx)
119     return;
120 
121   const Expr *SizeArg = TheCall->getArg(SizeIdx);
122   const Expr *DstSizeArg = TheCall->getArg(DstSizeIdx);
123 
124   llvm::APSInt Size, DstSize;
125 
126   // find out if both sizes are known at compile time
127   if (!SizeArg->EvaluateAsInt(Size, S.Context) ||
128       !DstSizeArg->EvaluateAsInt(DstSize, S.Context))
129     return;
130 
131   if (Size.ule(DstSize))
132     return;
133 
134   // confirmed overflow so generate the diagnostic.
135   IdentifierInfo *FnName = FDecl->getIdentifier();
136   SourceLocation SL = TheCall->getLocStart();
137   SourceRange SR = TheCall->getSourceRange();
138 
139   S.Diag(SL, diag::warn_memcpy_chk_overflow) << SR << FnName;
140 }
141 
SemaBuiltinCallWithStaticChain(Sema & S,CallExpr * BuiltinCall)142 static bool SemaBuiltinCallWithStaticChain(Sema &S, CallExpr *BuiltinCall) {
143   if (checkArgCount(S, BuiltinCall, 2))
144     return true;
145 
146   SourceLocation BuiltinLoc = BuiltinCall->getLocStart();
147   Expr *Builtin = BuiltinCall->getCallee()->IgnoreImpCasts();
148   Expr *Call = BuiltinCall->getArg(0);
149   Expr *Chain = BuiltinCall->getArg(1);
150 
151   if (Call->getStmtClass() != Stmt::CallExprClass) {
152     S.Diag(BuiltinLoc, diag::err_first_argument_to_cwsc_not_call)
153         << Call->getSourceRange();
154     return true;
155   }
156 
157   auto CE = cast<CallExpr>(Call);
158   if (CE->getCallee()->getType()->isBlockPointerType()) {
159     S.Diag(BuiltinLoc, diag::err_first_argument_to_cwsc_block_call)
160         << Call->getSourceRange();
161     return true;
162   }
163 
164   const Decl *TargetDecl = CE->getCalleeDecl();
165   if (const FunctionDecl *FD = dyn_cast_or_null<FunctionDecl>(TargetDecl))
166     if (FD->getBuiltinID()) {
167       S.Diag(BuiltinLoc, diag::err_first_argument_to_cwsc_builtin_call)
168           << Call->getSourceRange();
169       return true;
170     }
171 
172   if (isa<CXXPseudoDestructorExpr>(CE->getCallee()->IgnoreParens())) {
173     S.Diag(BuiltinLoc, diag::err_first_argument_to_cwsc_pdtor_call)
174         << Call->getSourceRange();
175     return true;
176   }
177 
178   ExprResult ChainResult = S.UsualUnaryConversions(Chain);
179   if (ChainResult.isInvalid())
180     return true;
181   if (!ChainResult.get()->getType()->isPointerType()) {
182     S.Diag(BuiltinLoc, diag::err_second_argument_to_cwsc_not_pointer)
183         << Chain->getSourceRange();
184     return true;
185   }
186 
187   QualType ReturnTy = CE->getCallReturnType();
188   QualType ArgTys[2] = { ReturnTy, ChainResult.get()->getType() };
189   QualType BuiltinTy = S.Context.getFunctionType(
190       ReturnTy, ArgTys, FunctionProtoType::ExtProtoInfo());
191   QualType BuiltinPtrTy = S.Context.getPointerType(BuiltinTy);
192 
193   Builtin =
194       S.ImpCastExprToType(Builtin, BuiltinPtrTy, CK_BuiltinFnToFnPtr).get();
195 
196   BuiltinCall->setType(CE->getType());
197   BuiltinCall->setValueKind(CE->getValueKind());
198   BuiltinCall->setObjectKind(CE->getObjectKind());
199   BuiltinCall->setCallee(Builtin);
200   BuiltinCall->setArg(1, ChainResult.get());
201 
202   return false;
203 }
204 
205 ExprResult
CheckBuiltinFunctionCall(FunctionDecl * FDecl,unsigned BuiltinID,CallExpr * TheCall)206 Sema::CheckBuiltinFunctionCall(FunctionDecl *FDecl, unsigned BuiltinID,
207                                CallExpr *TheCall) {
208   ExprResult TheCallResult(TheCall);
209 
210   // Find out if any arguments are required to be integer constant expressions.
211   unsigned ICEArguments = 0;
212   ASTContext::GetBuiltinTypeError Error;
213   Context.GetBuiltinType(BuiltinID, Error, &ICEArguments);
214   if (Error != ASTContext::GE_None)
215     ICEArguments = 0;  // Don't diagnose previously diagnosed errors.
216 
217   // If any arguments are required to be ICE's, check and diagnose.
218   for (unsigned ArgNo = 0; ICEArguments != 0; ++ArgNo) {
219     // Skip arguments not required to be ICE's.
220     if ((ICEArguments & (1 << ArgNo)) == 0) continue;
221 
222     llvm::APSInt Result;
223     if (SemaBuiltinConstantArg(TheCall, ArgNo, Result))
224       return true;
225     ICEArguments &= ~(1 << ArgNo);
226   }
227 
228   switch (BuiltinID) {
229   case Builtin::BI__builtin___CFStringMakeConstantString:
230     assert(TheCall->getNumArgs() == 1 &&
231            "Wrong # arguments to builtin CFStringMakeConstantString");
232     if (CheckObjCString(TheCall->getArg(0)))
233       return ExprError();
234     break;
235   case Builtin::BI__builtin_stdarg_start:
236   case Builtin::BI__builtin_va_start:
237     if (SemaBuiltinVAStart(TheCall))
238       return ExprError();
239     break;
240   case Builtin::BI__va_start: {
241     switch (Context.getTargetInfo().getTriple().getArch()) {
242     case llvm::Triple::arm:
243     case llvm::Triple::thumb:
244       if (SemaBuiltinVAStartARM(TheCall))
245         return ExprError();
246       break;
247     default:
248       if (SemaBuiltinVAStart(TheCall))
249         return ExprError();
250       break;
251     }
252     break;
253   }
254   case Builtin::BI__builtin_isgreater:
255   case Builtin::BI__builtin_isgreaterequal:
256   case Builtin::BI__builtin_isless:
257   case Builtin::BI__builtin_islessequal:
258   case Builtin::BI__builtin_islessgreater:
259   case Builtin::BI__builtin_isunordered:
260     if (SemaBuiltinUnorderedCompare(TheCall))
261       return ExprError();
262     break;
263   case Builtin::BI__builtin_fpclassify:
264     if (SemaBuiltinFPClassification(TheCall, 6))
265       return ExprError();
266     break;
267   case Builtin::BI__builtin_isfinite:
268   case Builtin::BI__builtin_isinf:
269   case Builtin::BI__builtin_isinf_sign:
270   case Builtin::BI__builtin_isnan:
271   case Builtin::BI__builtin_isnormal:
272     if (SemaBuiltinFPClassification(TheCall, 1))
273       return ExprError();
274     break;
275   case Builtin::BI__builtin_shufflevector:
276     return SemaBuiltinShuffleVector(TheCall);
277     // TheCall will be freed by the smart pointer here, but that's fine, since
278     // SemaBuiltinShuffleVector guts it, but then doesn't release it.
279   case Builtin::BI__builtin_prefetch:
280     if (SemaBuiltinPrefetch(TheCall))
281       return ExprError();
282     break;
283   case Builtin::BI__assume:
284   case Builtin::BI__builtin_assume:
285     if (SemaBuiltinAssume(TheCall))
286       return ExprError();
287     break;
288   case Builtin::BI__builtin_assume_aligned:
289     if (SemaBuiltinAssumeAligned(TheCall))
290       return ExprError();
291     break;
292   case Builtin::BI__builtin_object_size:
293     if (SemaBuiltinConstantArgRange(TheCall, 1, 0, 3))
294       return ExprError();
295     break;
296   case Builtin::BI__builtin_longjmp:
297     if (SemaBuiltinLongjmp(TheCall))
298       return ExprError();
299     break;
300   case Builtin::BI__builtin_setjmp:
301     if (SemaBuiltinSetjmp(TheCall))
302       return ExprError();
303     break;
304 
305   case Builtin::BI__builtin_classify_type:
306     if (checkArgCount(*this, TheCall, 1)) return true;
307     TheCall->setType(Context.IntTy);
308     break;
309   case Builtin::BI__builtin_constant_p:
310     if (checkArgCount(*this, TheCall, 1)) return true;
311     TheCall->setType(Context.IntTy);
312     break;
313   case Builtin::BI__sync_fetch_and_add:
314   case Builtin::BI__sync_fetch_and_add_1:
315   case Builtin::BI__sync_fetch_and_add_2:
316   case Builtin::BI__sync_fetch_and_add_4:
317   case Builtin::BI__sync_fetch_and_add_8:
318   case Builtin::BI__sync_fetch_and_add_16:
319   case Builtin::BI__sync_fetch_and_sub:
320   case Builtin::BI__sync_fetch_and_sub_1:
321   case Builtin::BI__sync_fetch_and_sub_2:
322   case Builtin::BI__sync_fetch_and_sub_4:
323   case Builtin::BI__sync_fetch_and_sub_8:
324   case Builtin::BI__sync_fetch_and_sub_16:
325   case Builtin::BI__sync_fetch_and_or:
326   case Builtin::BI__sync_fetch_and_or_1:
327   case Builtin::BI__sync_fetch_and_or_2:
328   case Builtin::BI__sync_fetch_and_or_4:
329   case Builtin::BI__sync_fetch_and_or_8:
330   case Builtin::BI__sync_fetch_and_or_16:
331   case Builtin::BI__sync_fetch_and_and:
332   case Builtin::BI__sync_fetch_and_and_1:
333   case Builtin::BI__sync_fetch_and_and_2:
334   case Builtin::BI__sync_fetch_and_and_4:
335   case Builtin::BI__sync_fetch_and_and_8:
336   case Builtin::BI__sync_fetch_and_and_16:
337   case Builtin::BI__sync_fetch_and_xor:
338   case Builtin::BI__sync_fetch_and_xor_1:
339   case Builtin::BI__sync_fetch_and_xor_2:
340   case Builtin::BI__sync_fetch_and_xor_4:
341   case Builtin::BI__sync_fetch_and_xor_8:
342   case Builtin::BI__sync_fetch_and_xor_16:
343   case Builtin::BI__sync_fetch_and_nand:
344   case Builtin::BI__sync_fetch_and_nand_1:
345   case Builtin::BI__sync_fetch_and_nand_2:
346   case Builtin::BI__sync_fetch_and_nand_4:
347   case Builtin::BI__sync_fetch_and_nand_8:
348   case Builtin::BI__sync_fetch_and_nand_16:
349   case Builtin::BI__sync_add_and_fetch:
350   case Builtin::BI__sync_add_and_fetch_1:
351   case Builtin::BI__sync_add_and_fetch_2:
352   case Builtin::BI__sync_add_and_fetch_4:
353   case Builtin::BI__sync_add_and_fetch_8:
354   case Builtin::BI__sync_add_and_fetch_16:
355   case Builtin::BI__sync_sub_and_fetch:
356   case Builtin::BI__sync_sub_and_fetch_1:
357   case Builtin::BI__sync_sub_and_fetch_2:
358   case Builtin::BI__sync_sub_and_fetch_4:
359   case Builtin::BI__sync_sub_and_fetch_8:
360   case Builtin::BI__sync_sub_and_fetch_16:
361   case Builtin::BI__sync_and_and_fetch:
362   case Builtin::BI__sync_and_and_fetch_1:
363   case Builtin::BI__sync_and_and_fetch_2:
364   case Builtin::BI__sync_and_and_fetch_4:
365   case Builtin::BI__sync_and_and_fetch_8:
366   case Builtin::BI__sync_and_and_fetch_16:
367   case Builtin::BI__sync_or_and_fetch:
368   case Builtin::BI__sync_or_and_fetch_1:
369   case Builtin::BI__sync_or_and_fetch_2:
370   case Builtin::BI__sync_or_and_fetch_4:
371   case Builtin::BI__sync_or_and_fetch_8:
372   case Builtin::BI__sync_or_and_fetch_16:
373   case Builtin::BI__sync_xor_and_fetch:
374   case Builtin::BI__sync_xor_and_fetch_1:
375   case Builtin::BI__sync_xor_and_fetch_2:
376   case Builtin::BI__sync_xor_and_fetch_4:
377   case Builtin::BI__sync_xor_and_fetch_8:
378   case Builtin::BI__sync_xor_and_fetch_16:
379   case Builtin::BI__sync_nand_and_fetch:
380   case Builtin::BI__sync_nand_and_fetch_1:
381   case Builtin::BI__sync_nand_and_fetch_2:
382   case Builtin::BI__sync_nand_and_fetch_4:
383   case Builtin::BI__sync_nand_and_fetch_8:
384   case Builtin::BI__sync_nand_and_fetch_16:
385   case Builtin::BI__sync_val_compare_and_swap:
386   case Builtin::BI__sync_val_compare_and_swap_1:
387   case Builtin::BI__sync_val_compare_and_swap_2:
388   case Builtin::BI__sync_val_compare_and_swap_4:
389   case Builtin::BI__sync_val_compare_and_swap_8:
390   case Builtin::BI__sync_val_compare_and_swap_16:
391   case Builtin::BI__sync_bool_compare_and_swap:
392   case Builtin::BI__sync_bool_compare_and_swap_1:
393   case Builtin::BI__sync_bool_compare_and_swap_2:
394   case Builtin::BI__sync_bool_compare_and_swap_4:
395   case Builtin::BI__sync_bool_compare_and_swap_8:
396   case Builtin::BI__sync_bool_compare_and_swap_16:
397   case Builtin::BI__sync_lock_test_and_set:
398   case Builtin::BI__sync_lock_test_and_set_1:
399   case Builtin::BI__sync_lock_test_and_set_2:
400   case Builtin::BI__sync_lock_test_and_set_4:
401   case Builtin::BI__sync_lock_test_and_set_8:
402   case Builtin::BI__sync_lock_test_and_set_16:
403   case Builtin::BI__sync_lock_release:
404   case Builtin::BI__sync_lock_release_1:
405   case Builtin::BI__sync_lock_release_2:
406   case Builtin::BI__sync_lock_release_4:
407   case Builtin::BI__sync_lock_release_8:
408   case Builtin::BI__sync_lock_release_16:
409   case Builtin::BI__sync_swap:
410   case Builtin::BI__sync_swap_1:
411   case Builtin::BI__sync_swap_2:
412   case Builtin::BI__sync_swap_4:
413   case Builtin::BI__sync_swap_8:
414   case Builtin::BI__sync_swap_16:
415     return SemaBuiltinAtomicOverloaded(TheCallResult);
416 #define BUILTIN(ID, TYPE, ATTRS)
417 #define ATOMIC_BUILTIN(ID, TYPE, ATTRS) \
418   case Builtin::BI##ID: \
419     return SemaAtomicOpsOverloaded(TheCallResult, AtomicExpr::AO##ID);
420 #include "clang/Basic/Builtins.def"
421   case Builtin::BI__builtin_annotation:
422     if (SemaBuiltinAnnotation(*this, TheCall))
423       return ExprError();
424     break;
425   case Builtin::BI__builtin_addressof:
426     if (SemaBuiltinAddressof(*this, TheCall))
427       return ExprError();
428     break;
429   case Builtin::BI__builtin_operator_new:
430   case Builtin::BI__builtin_operator_delete:
431     if (!getLangOpts().CPlusPlus) {
432       Diag(TheCall->getExprLoc(), diag::err_builtin_requires_language)
433         << (BuiltinID == Builtin::BI__builtin_operator_new
434                 ? "__builtin_operator_new"
435                 : "__builtin_operator_delete")
436         << "C++";
437       return ExprError();
438     }
439     // CodeGen assumes it can find the global new and delete to call,
440     // so ensure that they are declared.
441     DeclareGlobalNewDelete();
442     break;
443 
444   // check secure string manipulation functions where overflows
445   // are detectable at compile time
446   case Builtin::BI__builtin___memcpy_chk:
447   case Builtin::BI__builtin___memmove_chk:
448   case Builtin::BI__builtin___memset_chk:
449   case Builtin::BI__builtin___strlcat_chk:
450   case Builtin::BI__builtin___strlcpy_chk:
451   case Builtin::BI__builtin___strncat_chk:
452   case Builtin::BI__builtin___strncpy_chk:
453   case Builtin::BI__builtin___stpncpy_chk:
454     SemaBuiltinMemChkCall(*this, FDecl, TheCall, 2, 3);
455     break;
456   case Builtin::BI__builtin___memccpy_chk:
457     SemaBuiltinMemChkCall(*this, FDecl, TheCall, 3, 4);
458     break;
459   case Builtin::BI__builtin___snprintf_chk:
460   case Builtin::BI__builtin___vsnprintf_chk:
461     SemaBuiltinMemChkCall(*this, FDecl, TheCall, 1, 3);
462     break;
463 
464   case Builtin::BI__builtin_call_with_static_chain:
465     if (SemaBuiltinCallWithStaticChain(*this, TheCall))
466       return ExprError();
467     break;
468   }
469 
470   // Since the target specific builtins for each arch overlap, only check those
471   // of the arch we are compiling for.
472   if (BuiltinID >= Builtin::FirstTSBuiltin) {
473     switch (Context.getTargetInfo().getTriple().getArch()) {
474       case llvm::Triple::arm:
475       case llvm::Triple::armeb:
476       case llvm::Triple::thumb:
477       case llvm::Triple::thumbeb:
478         if (CheckARMBuiltinFunctionCall(BuiltinID, TheCall))
479           return ExprError();
480         break;
481       case llvm::Triple::aarch64:
482       case llvm::Triple::aarch64_be:
483         if (CheckAArch64BuiltinFunctionCall(BuiltinID, TheCall))
484           return ExprError();
485         break;
486       case llvm::Triple::mips:
487       case llvm::Triple::mipsel:
488       case llvm::Triple::mips64:
489       case llvm::Triple::mips64el:
490         if (CheckMipsBuiltinFunctionCall(BuiltinID, TheCall))
491           return ExprError();
492         break;
493       case llvm::Triple::x86:
494       case llvm::Triple::x86_64:
495         if (CheckX86BuiltinFunctionCall(BuiltinID, TheCall))
496           return ExprError();
497         break;
498       default:
499         break;
500     }
501   }
502 
503   return TheCallResult;
504 }
505 
506 // Get the valid immediate range for the specified NEON type code.
RFT(unsigned t,bool shift=false,bool ForceQuad=false)507 static unsigned RFT(unsigned t, bool shift = false, bool ForceQuad = false) {
508   NeonTypeFlags Type(t);
509   int IsQuad = ForceQuad ? true : Type.isQuad();
510   switch (Type.getEltType()) {
511   case NeonTypeFlags::Int8:
512   case NeonTypeFlags::Poly8:
513     return shift ? 7 : (8 << IsQuad) - 1;
514   case NeonTypeFlags::Int16:
515   case NeonTypeFlags::Poly16:
516     return shift ? 15 : (4 << IsQuad) - 1;
517   case NeonTypeFlags::Int32:
518     return shift ? 31 : (2 << IsQuad) - 1;
519   case NeonTypeFlags::Int64:
520   case NeonTypeFlags::Poly64:
521     return shift ? 63 : (1 << IsQuad) - 1;
522   case NeonTypeFlags::Poly128:
523     return shift ? 127 : (1 << IsQuad) - 1;
524   case NeonTypeFlags::Float16:
525     assert(!shift && "cannot shift float types!");
526     return (4 << IsQuad) - 1;
527   case NeonTypeFlags::Float32:
528     assert(!shift && "cannot shift float types!");
529     return (2 << IsQuad) - 1;
530   case NeonTypeFlags::Float64:
531     assert(!shift && "cannot shift float types!");
532     return (1 << IsQuad) - 1;
533   }
534   llvm_unreachable("Invalid NeonTypeFlag!");
535 }
536 
537 /// getNeonEltType - Return the QualType corresponding to the elements of
538 /// the vector type specified by the NeonTypeFlags.  This is used to check
539 /// the pointer arguments for Neon load/store intrinsics.
getNeonEltType(NeonTypeFlags Flags,ASTContext & Context,bool IsPolyUnsigned,bool IsInt64Long)540 static QualType getNeonEltType(NeonTypeFlags Flags, ASTContext &Context,
541                                bool IsPolyUnsigned, bool IsInt64Long) {
542   switch (Flags.getEltType()) {
543   case NeonTypeFlags::Int8:
544     return Flags.isUnsigned() ? Context.UnsignedCharTy : Context.SignedCharTy;
545   case NeonTypeFlags::Int16:
546     return Flags.isUnsigned() ? Context.UnsignedShortTy : Context.ShortTy;
547   case NeonTypeFlags::Int32:
548     return Flags.isUnsigned() ? Context.UnsignedIntTy : Context.IntTy;
549   case NeonTypeFlags::Int64:
550     if (IsInt64Long)
551       return Flags.isUnsigned() ? Context.UnsignedLongTy : Context.LongTy;
552     else
553       return Flags.isUnsigned() ? Context.UnsignedLongLongTy
554                                 : Context.LongLongTy;
555   case NeonTypeFlags::Poly8:
556     return IsPolyUnsigned ? Context.UnsignedCharTy : Context.SignedCharTy;
557   case NeonTypeFlags::Poly16:
558     return IsPolyUnsigned ? Context.UnsignedShortTy : Context.ShortTy;
559   case NeonTypeFlags::Poly64:
560     return Context.UnsignedLongTy;
561   case NeonTypeFlags::Poly128:
562     break;
563   case NeonTypeFlags::Float16:
564     return Context.HalfTy;
565   case NeonTypeFlags::Float32:
566     return Context.FloatTy;
567   case NeonTypeFlags::Float64:
568     return Context.DoubleTy;
569   }
570   llvm_unreachable("Invalid NeonTypeFlag!");
571 }
572 
CheckNeonBuiltinFunctionCall(unsigned BuiltinID,CallExpr * TheCall)573 bool Sema::CheckNeonBuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall) {
574   llvm::APSInt Result;
575   uint64_t mask = 0;
576   unsigned TV = 0;
577   int PtrArgNum = -1;
578   bool HasConstPtr = false;
579   switch (BuiltinID) {
580 #define GET_NEON_OVERLOAD_CHECK
581 #include "clang/Basic/arm_neon.inc"
582 #undef GET_NEON_OVERLOAD_CHECK
583   }
584 
585   // For NEON intrinsics which are overloaded on vector element type, validate
586   // the immediate which specifies which variant to emit.
587   unsigned ImmArg = TheCall->getNumArgs()-1;
588   if (mask) {
589     if (SemaBuiltinConstantArg(TheCall, ImmArg, Result))
590       return true;
591 
592     TV = Result.getLimitedValue(64);
593     if ((TV > 63) || (mask & (1ULL << TV)) == 0)
594       return Diag(TheCall->getLocStart(), diag::err_invalid_neon_type_code)
595         << TheCall->getArg(ImmArg)->getSourceRange();
596   }
597 
598   if (PtrArgNum >= 0) {
599     // Check that pointer arguments have the specified type.
600     Expr *Arg = TheCall->getArg(PtrArgNum);
601     if (ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(Arg))
602       Arg = ICE->getSubExpr();
603     ExprResult RHS = DefaultFunctionArrayLvalueConversion(Arg);
604     QualType RHSTy = RHS.get()->getType();
605 
606     llvm::Triple::ArchType Arch = Context.getTargetInfo().getTriple().getArch();
607     bool IsPolyUnsigned = Arch == llvm::Triple::aarch64;
608     bool IsInt64Long =
609         Context.getTargetInfo().getInt64Type() == TargetInfo::SignedLong;
610     QualType EltTy =
611         getNeonEltType(NeonTypeFlags(TV), Context, IsPolyUnsigned, IsInt64Long);
612     if (HasConstPtr)
613       EltTy = EltTy.withConst();
614     QualType LHSTy = Context.getPointerType(EltTy);
615     AssignConvertType ConvTy;
616     ConvTy = CheckSingleAssignmentConstraints(LHSTy, RHS);
617     if (RHS.isInvalid())
618       return true;
619     if (DiagnoseAssignmentResult(ConvTy, Arg->getLocStart(), LHSTy, RHSTy,
620                                  RHS.get(), AA_Assigning))
621       return true;
622   }
623 
624   // For NEON intrinsics which take an immediate value as part of the
625   // instruction, range check them here.
626   unsigned i = 0, l = 0, u = 0;
627   switch (BuiltinID) {
628   default:
629     return false;
630 #define GET_NEON_IMMEDIATE_CHECK
631 #include "clang/Basic/arm_neon.inc"
632 #undef GET_NEON_IMMEDIATE_CHECK
633   }
634 
635   return SemaBuiltinConstantArgRange(TheCall, i, l, u + l);
636 }
637 
CheckARMBuiltinExclusiveCall(unsigned BuiltinID,CallExpr * TheCall,unsigned MaxWidth)638 bool Sema::CheckARMBuiltinExclusiveCall(unsigned BuiltinID, CallExpr *TheCall,
639                                         unsigned MaxWidth) {
640   assert((BuiltinID == ARM::BI__builtin_arm_ldrex ||
641           BuiltinID == ARM::BI__builtin_arm_ldaex ||
642           BuiltinID == ARM::BI__builtin_arm_strex ||
643           BuiltinID == ARM::BI__builtin_arm_stlex ||
644           BuiltinID == AArch64::BI__builtin_arm_ldrex ||
645           BuiltinID == AArch64::BI__builtin_arm_ldaex ||
646           BuiltinID == AArch64::BI__builtin_arm_strex ||
647           BuiltinID == AArch64::BI__builtin_arm_stlex) &&
648          "unexpected ARM builtin");
649   bool IsLdrex = BuiltinID == ARM::BI__builtin_arm_ldrex ||
650                  BuiltinID == ARM::BI__builtin_arm_ldaex ||
651                  BuiltinID == AArch64::BI__builtin_arm_ldrex ||
652                  BuiltinID == AArch64::BI__builtin_arm_ldaex;
653 
654   DeclRefExpr *DRE =cast<DeclRefExpr>(TheCall->getCallee()->IgnoreParenCasts());
655 
656   // Ensure that we have the proper number of arguments.
657   if (checkArgCount(*this, TheCall, IsLdrex ? 1 : 2))
658     return true;
659 
660   // Inspect the pointer argument of the atomic builtin.  This should always be
661   // a pointer type, whose element is an integral scalar or pointer type.
662   // Because it is a pointer type, we don't have to worry about any implicit
663   // casts here.
664   Expr *PointerArg = TheCall->getArg(IsLdrex ? 0 : 1);
665   ExprResult PointerArgRes = DefaultFunctionArrayLvalueConversion(PointerArg);
666   if (PointerArgRes.isInvalid())
667     return true;
668   PointerArg = PointerArgRes.get();
669 
670   const PointerType *pointerType = PointerArg->getType()->getAs<PointerType>();
671   if (!pointerType) {
672     Diag(DRE->getLocStart(), diag::err_atomic_builtin_must_be_pointer)
673       << PointerArg->getType() << PointerArg->getSourceRange();
674     return true;
675   }
676 
677   // ldrex takes a "const volatile T*" and strex takes a "volatile T*". Our next
678   // task is to insert the appropriate casts into the AST. First work out just
679   // what the appropriate type is.
680   QualType ValType = pointerType->getPointeeType();
681   QualType AddrType = ValType.getUnqualifiedType().withVolatile();
682   if (IsLdrex)
683     AddrType.addConst();
684 
685   // Issue a warning if the cast is dodgy.
686   CastKind CastNeeded = CK_NoOp;
687   if (!AddrType.isAtLeastAsQualifiedAs(ValType)) {
688     CastNeeded = CK_BitCast;
689     Diag(DRE->getLocStart(), diag::ext_typecheck_convert_discards_qualifiers)
690       << PointerArg->getType()
691       << Context.getPointerType(AddrType)
692       << AA_Passing << PointerArg->getSourceRange();
693   }
694 
695   // Finally, do the cast and replace the argument with the corrected version.
696   AddrType = Context.getPointerType(AddrType);
697   PointerArgRes = ImpCastExprToType(PointerArg, AddrType, CastNeeded);
698   if (PointerArgRes.isInvalid())
699     return true;
700   PointerArg = PointerArgRes.get();
701 
702   TheCall->setArg(IsLdrex ? 0 : 1, PointerArg);
703 
704   // In general, we allow ints, floats and pointers to be loaded and stored.
705   if (!ValType->isIntegerType() && !ValType->isAnyPointerType() &&
706       !ValType->isBlockPointerType() && !ValType->isFloatingType()) {
707     Diag(DRE->getLocStart(), diag::err_atomic_builtin_must_be_pointer_intfltptr)
708       << PointerArg->getType() << PointerArg->getSourceRange();
709     return true;
710   }
711 
712   // But ARM doesn't have instructions to deal with 128-bit versions.
713   if (Context.getTypeSize(ValType) > MaxWidth) {
714     assert(MaxWidth == 64 && "Diagnostic unexpectedly inaccurate");
715     Diag(DRE->getLocStart(), diag::err_atomic_exclusive_builtin_pointer_size)
716       << PointerArg->getType() << PointerArg->getSourceRange();
717     return true;
718   }
719 
720   switch (ValType.getObjCLifetime()) {
721   case Qualifiers::OCL_None:
722   case Qualifiers::OCL_ExplicitNone:
723     // okay
724     break;
725 
726   case Qualifiers::OCL_Weak:
727   case Qualifiers::OCL_Strong:
728   case Qualifiers::OCL_Autoreleasing:
729     Diag(DRE->getLocStart(), diag::err_arc_atomic_ownership)
730       << ValType << PointerArg->getSourceRange();
731     return true;
732   }
733 
734 
735   if (IsLdrex) {
736     TheCall->setType(ValType);
737     return false;
738   }
739 
740   // Initialize the argument to be stored.
741   ExprResult ValArg = TheCall->getArg(0);
742   InitializedEntity Entity = InitializedEntity::InitializeParameter(
743       Context, ValType, /*consume*/ false);
744   ValArg = PerformCopyInitialization(Entity, SourceLocation(), ValArg);
745   if (ValArg.isInvalid())
746     return true;
747   TheCall->setArg(0, ValArg.get());
748 
749   // __builtin_arm_strex always returns an int. It's marked as such in the .def,
750   // but the custom checker bypasses all default analysis.
751   TheCall->setType(Context.IntTy);
752   return false;
753 }
754 
CheckARMBuiltinFunctionCall(unsigned BuiltinID,CallExpr * TheCall)755 bool Sema::CheckARMBuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall) {
756   llvm::APSInt Result;
757 
758   if (BuiltinID == ARM::BI__builtin_arm_ldrex ||
759       BuiltinID == ARM::BI__builtin_arm_ldaex ||
760       BuiltinID == ARM::BI__builtin_arm_strex ||
761       BuiltinID == ARM::BI__builtin_arm_stlex) {
762     return CheckARMBuiltinExclusiveCall(BuiltinID, TheCall, 64);
763   }
764 
765   if (BuiltinID == ARM::BI__builtin_arm_prefetch) {
766     return SemaBuiltinConstantArgRange(TheCall, 1, 0, 1) ||
767       SemaBuiltinConstantArgRange(TheCall, 2, 0, 1);
768   }
769 
770   if (CheckNeonBuiltinFunctionCall(BuiltinID, TheCall))
771     return true;
772 
773   // For intrinsics which take an immediate value as part of the instruction,
774   // range check them here.
775   unsigned i = 0, l = 0, u = 0;
776   switch (BuiltinID) {
777   default: return false;
778   case ARM::BI__builtin_arm_ssat: i = 1; l = 1; u = 31; break;
779   case ARM::BI__builtin_arm_usat: i = 1; u = 31; break;
780   case ARM::BI__builtin_arm_vcvtr_f:
781   case ARM::BI__builtin_arm_vcvtr_d: i = 1; u = 1; break;
782   case ARM::BI__builtin_arm_dmb:
783   case ARM::BI__builtin_arm_dsb:
784   case ARM::BI__builtin_arm_isb:
785   case ARM::BI__builtin_arm_dbg: l = 0; u = 15; break;
786   }
787 
788   // FIXME: VFP Intrinsics should error if VFP not present.
789   return SemaBuiltinConstantArgRange(TheCall, i, l, u + l);
790 }
791 
CheckAArch64BuiltinFunctionCall(unsigned BuiltinID,CallExpr * TheCall)792 bool Sema::CheckAArch64BuiltinFunctionCall(unsigned BuiltinID,
793                                          CallExpr *TheCall) {
794   llvm::APSInt Result;
795 
796   if (BuiltinID == AArch64::BI__builtin_arm_ldrex ||
797       BuiltinID == AArch64::BI__builtin_arm_ldaex ||
798       BuiltinID == AArch64::BI__builtin_arm_strex ||
799       BuiltinID == AArch64::BI__builtin_arm_stlex) {
800     return CheckARMBuiltinExclusiveCall(BuiltinID, TheCall, 128);
801   }
802 
803   if (BuiltinID == AArch64::BI__builtin_arm_prefetch) {
804     return SemaBuiltinConstantArgRange(TheCall, 1, 0, 1) ||
805       SemaBuiltinConstantArgRange(TheCall, 2, 0, 2) ||
806       SemaBuiltinConstantArgRange(TheCall, 3, 0, 1) ||
807       SemaBuiltinConstantArgRange(TheCall, 4, 0, 1);
808   }
809 
810   if (CheckNeonBuiltinFunctionCall(BuiltinID, TheCall))
811     return true;
812 
813   // For intrinsics which take an immediate value as part of the instruction,
814   // range check them here.
815   unsigned i = 0, l = 0, u = 0;
816   switch (BuiltinID) {
817   default: return false;
818   case AArch64::BI__builtin_arm_dmb:
819   case AArch64::BI__builtin_arm_dsb:
820   case AArch64::BI__builtin_arm_isb: l = 0; u = 15; break;
821   }
822 
823   return SemaBuiltinConstantArgRange(TheCall, i, l, u + l);
824 }
825 
CheckMipsBuiltinFunctionCall(unsigned BuiltinID,CallExpr * TheCall)826 bool Sema::CheckMipsBuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall) {
827   unsigned i = 0, l = 0, u = 0;
828   switch (BuiltinID) {
829   default: return false;
830   case Mips::BI__builtin_mips_wrdsp: i = 1; l = 0; u = 63; break;
831   case Mips::BI__builtin_mips_rddsp: i = 0; l = 0; u = 63; break;
832   case Mips::BI__builtin_mips_append: i = 2; l = 0; u = 31; break;
833   case Mips::BI__builtin_mips_balign: i = 2; l = 0; u = 3; break;
834   case Mips::BI__builtin_mips_precr_sra_ph_w: i = 2; l = 0; u = 31; break;
835   case Mips::BI__builtin_mips_precr_sra_r_ph_w: i = 2; l = 0; u = 31; break;
836   case Mips::BI__builtin_mips_prepend: i = 2; l = 0; u = 31; break;
837   }
838 
839   return SemaBuiltinConstantArgRange(TheCall, i, l, u);
840 }
841 
CheckX86BuiltinFunctionCall(unsigned BuiltinID,CallExpr * TheCall)842 bool Sema::CheckX86BuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall) {
843   unsigned i = 0, l = 0, u = 0;
844   switch (BuiltinID) {
845   default: return false;
846   case X86::BI_mm_prefetch: i = 1; l = 0; u = 3; break;
847   case X86::BI__builtin_ia32_cmpps:
848   case X86::BI__builtin_ia32_cmpss:
849   case X86::BI__builtin_ia32_cmppd:
850   case X86::BI__builtin_ia32_cmpsd: i = 2; l = 0; u = 31; break;
851   }
852   return SemaBuiltinConstantArgRange(TheCall, i, l, u);
853 }
854 
855 /// Given a FunctionDecl's FormatAttr, attempts to populate the FomatStringInfo
856 /// parameter with the FormatAttr's correct format_idx and firstDataArg.
857 /// Returns true when the format fits the function and the FormatStringInfo has
858 /// been populated.
getFormatStringInfo(const FormatAttr * Format,bool IsCXXMember,FormatStringInfo * FSI)859 bool Sema::getFormatStringInfo(const FormatAttr *Format, bool IsCXXMember,
860                                FormatStringInfo *FSI) {
861   FSI->HasVAListArg = Format->getFirstArg() == 0;
862   FSI->FormatIdx = Format->getFormatIdx() - 1;
863   FSI->FirstDataArg = FSI->HasVAListArg ? 0 : Format->getFirstArg() - 1;
864 
865   // The way the format attribute works in GCC, the implicit this argument
866   // of member functions is counted. However, it doesn't appear in our own
867   // lists, so decrement format_idx in that case.
868   if (IsCXXMember) {
869     if(FSI->FormatIdx == 0)
870       return false;
871     --FSI->FormatIdx;
872     if (FSI->FirstDataArg != 0)
873       --FSI->FirstDataArg;
874   }
875   return true;
876 }
877 
878 /// Checks if a the given expression evaluates to null.
879 ///
880 /// \brief Returns true if the value evaluates to null.
CheckNonNullExpr(Sema & S,const Expr * Expr)881 static bool CheckNonNullExpr(Sema &S,
882                              const Expr *Expr) {
883   // As a special case, transparent unions initialized with zero are
884   // considered null for the purposes of the nonnull attribute.
885   if (const RecordType *UT = Expr->getType()->getAsUnionType()) {
886     if (UT->getDecl()->hasAttr<TransparentUnionAttr>())
887       if (const CompoundLiteralExpr *CLE =
888           dyn_cast<CompoundLiteralExpr>(Expr))
889         if (const InitListExpr *ILE =
890             dyn_cast<InitListExpr>(CLE->getInitializer()))
891           Expr = ILE->getInit(0);
892   }
893 
894   bool Result;
895   return (!Expr->isValueDependent() &&
896           Expr->EvaluateAsBooleanCondition(Result, S.Context) &&
897           !Result);
898 }
899 
CheckNonNullArgument(Sema & S,const Expr * ArgExpr,SourceLocation CallSiteLoc)900 static void CheckNonNullArgument(Sema &S,
901                                  const Expr *ArgExpr,
902                                  SourceLocation CallSiteLoc) {
903   if (CheckNonNullExpr(S, ArgExpr))
904     S.Diag(CallSiteLoc, diag::warn_null_arg) << ArgExpr->getSourceRange();
905 }
906 
GetFormatNSStringIdx(const FormatAttr * Format,unsigned & Idx)907 bool Sema::GetFormatNSStringIdx(const FormatAttr *Format, unsigned &Idx) {
908   FormatStringInfo FSI;
909   if ((GetFormatStringType(Format) == FST_NSString) &&
910       getFormatStringInfo(Format, false, &FSI)) {
911     Idx = FSI.FormatIdx;
912     return true;
913   }
914   return false;
915 }
916 /// \brief Diagnose use of %s directive in an NSString which is being passed
917 /// as formatting string to formatting method.
918 static void
DiagnoseCStringFormatDirectiveInCFAPI(Sema & S,const NamedDecl * FDecl,Expr ** Args,unsigned NumArgs)919 DiagnoseCStringFormatDirectiveInCFAPI(Sema &S,
920                                         const NamedDecl *FDecl,
921                                         Expr **Args,
922                                         unsigned NumArgs) {
923   unsigned Idx = 0;
924   bool Format = false;
925   ObjCStringFormatFamily SFFamily = FDecl->getObjCFStringFormattingFamily();
926   if (SFFamily == ObjCStringFormatFamily::SFF_CFString) {
927     Idx = 2;
928     Format = true;
929   }
930   else
931     for (const auto *I : FDecl->specific_attrs<FormatAttr>()) {
932       if (S.GetFormatNSStringIdx(I, Idx)) {
933         Format = true;
934         break;
935       }
936     }
937   if (!Format || NumArgs <= Idx)
938     return;
939   const Expr *FormatExpr = Args[Idx];
940   if (const CStyleCastExpr *CSCE = dyn_cast<CStyleCastExpr>(FormatExpr))
941     FormatExpr = CSCE->getSubExpr();
942   const StringLiteral *FormatString;
943   if (const ObjCStringLiteral *OSL =
944       dyn_cast<ObjCStringLiteral>(FormatExpr->IgnoreParenImpCasts()))
945     FormatString = OSL->getString();
946   else
947     FormatString = dyn_cast<StringLiteral>(FormatExpr->IgnoreParenImpCasts());
948   if (!FormatString)
949     return;
950   if (S.FormatStringHasSArg(FormatString)) {
951     S.Diag(FormatExpr->getExprLoc(), diag::warn_objc_cdirective_format_string)
952       << "%s" << 1 << 1;
953     S.Diag(FDecl->getLocation(), diag::note_entity_declared_at)
954       << FDecl->getDeclName();
955   }
956 }
957 
CheckNonNullArguments(Sema & S,const NamedDecl * FDecl,ArrayRef<const Expr * > Args,SourceLocation CallSiteLoc)958 static void CheckNonNullArguments(Sema &S,
959                                   const NamedDecl *FDecl,
960                                   ArrayRef<const Expr *> Args,
961                                   SourceLocation CallSiteLoc) {
962   // Check the attributes attached to the method/function itself.
963   llvm::SmallBitVector NonNullArgs;
964   for (const auto *NonNull : FDecl->specific_attrs<NonNullAttr>()) {
965     if (!NonNull->args_size()) {
966       // Easy case: all pointer arguments are nonnull.
967       for (const auto *Arg : Args)
968         if (S.isValidPointerAttrType(Arg->getType()))
969           CheckNonNullArgument(S, Arg, CallSiteLoc);
970       return;
971     }
972 
973     for (unsigned Val : NonNull->args()) {
974       if (Val >= Args.size())
975         continue;
976       if (NonNullArgs.empty())
977         NonNullArgs.resize(Args.size());
978       NonNullArgs.set(Val);
979     }
980   }
981 
982   // Check the attributes on the parameters.
983   ArrayRef<ParmVarDecl*> parms;
984   if (const FunctionDecl *FD = dyn_cast<FunctionDecl>(FDecl))
985     parms = FD->parameters();
986   else if (const ObjCMethodDecl *MD = dyn_cast<ObjCMethodDecl>(FDecl))
987     parms = MD->parameters();
988 
989   unsigned ArgIndex = 0;
990   for (ArrayRef<ParmVarDecl*>::iterator I = parms.begin(), E = parms.end();
991        I != E; ++I, ++ArgIndex) {
992     const ParmVarDecl *PVD = *I;
993     if (PVD->hasAttr<NonNullAttr>() ||
994         (ArgIndex < NonNullArgs.size() && NonNullArgs[ArgIndex]))
995       CheckNonNullArgument(S, Args[ArgIndex], CallSiteLoc);
996   }
997 
998   // In case this is a variadic call, check any remaining arguments.
999   for (/**/; ArgIndex < NonNullArgs.size(); ++ArgIndex)
1000     if (NonNullArgs[ArgIndex])
1001       CheckNonNullArgument(S, Args[ArgIndex], CallSiteLoc);
1002 }
1003 
1004 /// Handles the checks for format strings, non-POD arguments to vararg
1005 /// functions, and NULL arguments passed to non-NULL parameters.
checkCall(NamedDecl * FDecl,ArrayRef<const Expr * > Args,unsigned NumParams,bool IsMemberFunction,SourceLocation Loc,SourceRange Range,VariadicCallType CallType)1006 void Sema::checkCall(NamedDecl *FDecl, ArrayRef<const Expr *> Args,
1007                      unsigned NumParams, bool IsMemberFunction,
1008                      SourceLocation Loc, SourceRange Range,
1009                      VariadicCallType CallType) {
1010   // FIXME: We should check as much as we can in the template definition.
1011   if (CurContext->isDependentContext())
1012     return;
1013 
1014   // Printf and scanf checking.
1015   llvm::SmallBitVector CheckedVarArgs;
1016   if (FDecl) {
1017     for (const auto *I : FDecl->specific_attrs<FormatAttr>()) {
1018       // Only create vector if there are format attributes.
1019       CheckedVarArgs.resize(Args.size());
1020 
1021       CheckFormatArguments(I, Args, IsMemberFunction, CallType, Loc, Range,
1022                            CheckedVarArgs);
1023     }
1024   }
1025 
1026   // Refuse POD arguments that weren't caught by the format string
1027   // checks above.
1028   if (CallType != VariadicDoesNotApply) {
1029     for (unsigned ArgIdx = NumParams; ArgIdx < Args.size(); ++ArgIdx) {
1030       // Args[ArgIdx] can be null in malformed code.
1031       if (const Expr *Arg = Args[ArgIdx]) {
1032         if (CheckedVarArgs.empty() || !CheckedVarArgs[ArgIdx])
1033           checkVariadicArgument(Arg, CallType);
1034       }
1035     }
1036   }
1037 
1038   if (FDecl) {
1039     CheckNonNullArguments(*this, FDecl, Args, Loc);
1040 
1041     // Type safety checking.
1042     for (const auto *I : FDecl->specific_attrs<ArgumentWithTypeTagAttr>())
1043       CheckArgumentWithTypeTag(I, Args.data());
1044   }
1045 }
1046 
1047 /// CheckConstructorCall - Check a constructor call for correctness and safety
1048 /// properties not enforced by the C type system.
CheckConstructorCall(FunctionDecl * FDecl,ArrayRef<const Expr * > Args,const FunctionProtoType * Proto,SourceLocation Loc)1049 void Sema::CheckConstructorCall(FunctionDecl *FDecl,
1050                                 ArrayRef<const Expr *> Args,
1051                                 const FunctionProtoType *Proto,
1052                                 SourceLocation Loc) {
1053   VariadicCallType CallType =
1054     Proto->isVariadic() ? VariadicConstructor : VariadicDoesNotApply;
1055   checkCall(FDecl, Args, Proto->getNumParams(),
1056             /*IsMemberFunction=*/true, Loc, SourceRange(), CallType);
1057 }
1058 
1059 /// CheckFunctionCall - Check a direct function call for various correctness
1060 /// and safety properties not strictly enforced by the C type system.
CheckFunctionCall(FunctionDecl * FDecl,CallExpr * TheCall,const FunctionProtoType * Proto)1061 bool Sema::CheckFunctionCall(FunctionDecl *FDecl, CallExpr *TheCall,
1062                              const FunctionProtoType *Proto) {
1063   bool IsMemberOperatorCall = isa<CXXOperatorCallExpr>(TheCall) &&
1064                               isa<CXXMethodDecl>(FDecl);
1065   bool IsMemberFunction = isa<CXXMemberCallExpr>(TheCall) ||
1066                           IsMemberOperatorCall;
1067   VariadicCallType CallType = getVariadicCallType(FDecl, Proto,
1068                                                   TheCall->getCallee());
1069   unsigned NumParams = Proto ? Proto->getNumParams() : 0;
1070   Expr** Args = TheCall->getArgs();
1071   unsigned NumArgs = TheCall->getNumArgs();
1072   if (IsMemberOperatorCall) {
1073     // If this is a call to a member operator, hide the first argument
1074     // from checkCall.
1075     // FIXME: Our choice of AST representation here is less than ideal.
1076     ++Args;
1077     --NumArgs;
1078   }
1079   checkCall(FDecl, llvm::makeArrayRef(Args, NumArgs), NumParams,
1080             IsMemberFunction, TheCall->getRParenLoc(),
1081             TheCall->getCallee()->getSourceRange(), CallType);
1082 
1083   IdentifierInfo *FnInfo = FDecl->getIdentifier();
1084   // None of the checks below are needed for functions that don't have
1085   // simple names (e.g., C++ conversion functions).
1086   if (!FnInfo)
1087     return false;
1088 
1089   CheckAbsoluteValueFunction(TheCall, FDecl, FnInfo);
1090   if (getLangOpts().ObjC1)
1091     DiagnoseCStringFormatDirectiveInCFAPI(*this, FDecl, Args, NumArgs);
1092 
1093   unsigned CMId = FDecl->getMemoryFunctionKind();
1094   if (CMId == 0)
1095     return false;
1096 
1097   // Handle memory setting and copying functions.
1098   if (CMId == Builtin::BIstrlcpy || CMId == Builtin::BIstrlcat)
1099     CheckStrlcpycatArguments(TheCall, FnInfo);
1100   else if (CMId == Builtin::BIstrncat)
1101     CheckStrncatArguments(TheCall, FnInfo);
1102   else
1103     CheckMemaccessArguments(TheCall, CMId, FnInfo);
1104 
1105   return false;
1106 }
1107 
CheckObjCMethodCall(ObjCMethodDecl * Method,SourceLocation lbrac,ArrayRef<const Expr * > Args)1108 bool Sema::CheckObjCMethodCall(ObjCMethodDecl *Method, SourceLocation lbrac,
1109                                ArrayRef<const Expr *> Args) {
1110   VariadicCallType CallType =
1111       Method->isVariadic() ? VariadicMethod : VariadicDoesNotApply;
1112 
1113   checkCall(Method, Args, Method->param_size(),
1114             /*IsMemberFunction=*/false,
1115             lbrac, Method->getSourceRange(), CallType);
1116 
1117   return false;
1118 }
1119 
CheckPointerCall(NamedDecl * NDecl,CallExpr * TheCall,const FunctionProtoType * Proto)1120 bool Sema::CheckPointerCall(NamedDecl *NDecl, CallExpr *TheCall,
1121                             const FunctionProtoType *Proto) {
1122   const VarDecl *V = dyn_cast<VarDecl>(NDecl);
1123   if (!V)
1124     return false;
1125 
1126   QualType Ty = V->getType();
1127   if (!Ty->isBlockPointerType() && !Ty->isFunctionPointerType())
1128     return false;
1129 
1130   VariadicCallType CallType;
1131   if (!Proto || !Proto->isVariadic()) {
1132     CallType = VariadicDoesNotApply;
1133   } else if (Ty->isBlockPointerType()) {
1134     CallType = VariadicBlock;
1135   } else { // Ty->isFunctionPointerType()
1136     CallType = VariadicFunction;
1137   }
1138   unsigned NumParams = Proto ? Proto->getNumParams() : 0;
1139 
1140   checkCall(NDecl, llvm::makeArrayRef(TheCall->getArgs(),
1141                                       TheCall->getNumArgs()),
1142             NumParams, /*IsMemberFunction=*/false, TheCall->getRParenLoc(),
1143             TheCall->getCallee()->getSourceRange(), CallType);
1144 
1145   return false;
1146 }
1147 
1148 /// Checks function calls when a FunctionDecl or a NamedDecl is not available,
1149 /// such as function pointers returned from functions.
CheckOtherCall(CallExpr * TheCall,const FunctionProtoType * Proto)1150 bool Sema::CheckOtherCall(CallExpr *TheCall, const FunctionProtoType *Proto) {
1151   VariadicCallType CallType = getVariadicCallType(/*FDecl=*/nullptr, Proto,
1152                                                   TheCall->getCallee());
1153   unsigned NumParams = Proto ? Proto->getNumParams() : 0;
1154 
1155   checkCall(/*FDecl=*/nullptr,
1156             llvm::makeArrayRef(TheCall->getArgs(), TheCall->getNumArgs()),
1157             NumParams, /*IsMemberFunction=*/false, TheCall->getRParenLoc(),
1158             TheCall->getCallee()->getSourceRange(), CallType);
1159 
1160   return false;
1161 }
1162 
isValidOrderingForOp(int64_t Ordering,AtomicExpr::AtomicOp Op)1163 static bool isValidOrderingForOp(int64_t Ordering, AtomicExpr::AtomicOp Op) {
1164   if (Ordering < AtomicExpr::AO_ABI_memory_order_relaxed ||
1165       Ordering > AtomicExpr::AO_ABI_memory_order_seq_cst)
1166     return false;
1167 
1168   switch (Op) {
1169   case AtomicExpr::AO__c11_atomic_init:
1170     llvm_unreachable("There is no ordering argument for an init");
1171 
1172   case AtomicExpr::AO__c11_atomic_load:
1173   case AtomicExpr::AO__atomic_load_n:
1174   case AtomicExpr::AO__atomic_load:
1175     return Ordering != AtomicExpr::AO_ABI_memory_order_release &&
1176            Ordering != AtomicExpr::AO_ABI_memory_order_acq_rel;
1177 
1178   case AtomicExpr::AO__c11_atomic_store:
1179   case AtomicExpr::AO__atomic_store:
1180   case AtomicExpr::AO__atomic_store_n:
1181     return Ordering != AtomicExpr::AO_ABI_memory_order_consume &&
1182            Ordering != AtomicExpr::AO_ABI_memory_order_acquire &&
1183            Ordering != AtomicExpr::AO_ABI_memory_order_acq_rel;
1184 
1185   default:
1186     return true;
1187   }
1188 }
1189 
SemaAtomicOpsOverloaded(ExprResult TheCallResult,AtomicExpr::AtomicOp Op)1190 ExprResult Sema::SemaAtomicOpsOverloaded(ExprResult TheCallResult,
1191                                          AtomicExpr::AtomicOp Op) {
1192   CallExpr *TheCall = cast<CallExpr>(TheCallResult.get());
1193   DeclRefExpr *DRE =cast<DeclRefExpr>(TheCall->getCallee()->IgnoreParenCasts());
1194 
1195   // All these operations take one of the following forms:
1196   enum {
1197     // C    __c11_atomic_init(A *, C)
1198     Init,
1199     // C    __c11_atomic_load(A *, int)
1200     Load,
1201     // void __atomic_load(A *, CP, int)
1202     Copy,
1203     // C    __c11_atomic_add(A *, M, int)
1204     Arithmetic,
1205     // C    __atomic_exchange_n(A *, CP, int)
1206     Xchg,
1207     // void __atomic_exchange(A *, C *, CP, int)
1208     GNUXchg,
1209     // bool __c11_atomic_compare_exchange_strong(A *, C *, CP, int, int)
1210     C11CmpXchg,
1211     // bool __atomic_compare_exchange(A *, C *, CP, bool, int, int)
1212     GNUCmpXchg
1213   } Form = Init;
1214   const unsigned NumArgs[] = { 2, 2, 3, 3, 3, 4, 5, 6 };
1215   const unsigned NumVals[] = { 1, 0, 1, 1, 1, 2, 2, 3 };
1216   // where:
1217   //   C is an appropriate type,
1218   //   A is volatile _Atomic(C) for __c11 builtins and is C for GNU builtins,
1219   //   CP is C for __c11 builtins and GNU _n builtins and is C * otherwise,
1220   //   M is C if C is an integer, and ptrdiff_t if C is a pointer, and
1221   //   the int parameters are for orderings.
1222 
1223   assert(AtomicExpr::AO__c11_atomic_init == 0 &&
1224          AtomicExpr::AO__c11_atomic_fetch_xor + 1 == AtomicExpr::AO__atomic_load
1225          && "need to update code for modified C11 atomics");
1226   bool IsC11 = Op >= AtomicExpr::AO__c11_atomic_init &&
1227                Op <= AtomicExpr::AO__c11_atomic_fetch_xor;
1228   bool IsN = Op == AtomicExpr::AO__atomic_load_n ||
1229              Op == AtomicExpr::AO__atomic_store_n ||
1230              Op == AtomicExpr::AO__atomic_exchange_n ||
1231              Op == AtomicExpr::AO__atomic_compare_exchange_n;
1232   bool IsAddSub = false;
1233 
1234   switch (Op) {
1235   case AtomicExpr::AO__c11_atomic_init:
1236     Form = Init;
1237     break;
1238 
1239   case AtomicExpr::AO__c11_atomic_load:
1240   case AtomicExpr::AO__atomic_load_n:
1241     Form = Load;
1242     break;
1243 
1244   case AtomicExpr::AO__c11_atomic_store:
1245   case AtomicExpr::AO__atomic_load:
1246   case AtomicExpr::AO__atomic_store:
1247   case AtomicExpr::AO__atomic_store_n:
1248     Form = Copy;
1249     break;
1250 
1251   case AtomicExpr::AO__c11_atomic_fetch_add:
1252   case AtomicExpr::AO__c11_atomic_fetch_sub:
1253   case AtomicExpr::AO__atomic_fetch_add:
1254   case AtomicExpr::AO__atomic_fetch_sub:
1255   case AtomicExpr::AO__atomic_add_fetch:
1256   case AtomicExpr::AO__atomic_sub_fetch:
1257     IsAddSub = true;
1258     // Fall through.
1259   case AtomicExpr::AO__c11_atomic_fetch_and:
1260   case AtomicExpr::AO__c11_atomic_fetch_or:
1261   case AtomicExpr::AO__c11_atomic_fetch_xor:
1262   case AtomicExpr::AO__atomic_fetch_and:
1263   case AtomicExpr::AO__atomic_fetch_or:
1264   case AtomicExpr::AO__atomic_fetch_xor:
1265   case AtomicExpr::AO__atomic_fetch_nand:
1266   case AtomicExpr::AO__atomic_and_fetch:
1267   case AtomicExpr::AO__atomic_or_fetch:
1268   case AtomicExpr::AO__atomic_xor_fetch:
1269   case AtomicExpr::AO__atomic_nand_fetch:
1270     Form = Arithmetic;
1271     break;
1272 
1273   case AtomicExpr::AO__c11_atomic_exchange:
1274   case AtomicExpr::AO__atomic_exchange_n:
1275     Form = Xchg;
1276     break;
1277 
1278   case AtomicExpr::AO__atomic_exchange:
1279     Form = GNUXchg;
1280     break;
1281 
1282   case AtomicExpr::AO__c11_atomic_compare_exchange_strong:
1283   case AtomicExpr::AO__c11_atomic_compare_exchange_weak:
1284     Form = C11CmpXchg;
1285     break;
1286 
1287   case AtomicExpr::AO__atomic_compare_exchange:
1288   case AtomicExpr::AO__atomic_compare_exchange_n:
1289     Form = GNUCmpXchg;
1290     break;
1291   }
1292 
1293   // Check we have the right number of arguments.
1294   if (TheCall->getNumArgs() < NumArgs[Form]) {
1295     Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args)
1296       << 0 << NumArgs[Form] << TheCall->getNumArgs()
1297       << TheCall->getCallee()->getSourceRange();
1298     return ExprError();
1299   } else if (TheCall->getNumArgs() > NumArgs[Form]) {
1300     Diag(TheCall->getArg(NumArgs[Form])->getLocStart(),
1301          diag::err_typecheck_call_too_many_args)
1302       << 0 << NumArgs[Form] << TheCall->getNumArgs()
1303       << TheCall->getCallee()->getSourceRange();
1304     return ExprError();
1305   }
1306 
1307   // Inspect the first argument of the atomic operation.
1308   Expr *Ptr = TheCall->getArg(0);
1309   Ptr = DefaultFunctionArrayLvalueConversion(Ptr).get();
1310   const PointerType *pointerType = Ptr->getType()->getAs<PointerType>();
1311   if (!pointerType) {
1312     Diag(DRE->getLocStart(), diag::err_atomic_builtin_must_be_pointer)
1313       << Ptr->getType() << Ptr->getSourceRange();
1314     return ExprError();
1315   }
1316 
1317   // For a __c11 builtin, this should be a pointer to an _Atomic type.
1318   QualType AtomTy = pointerType->getPointeeType(); // 'A'
1319   QualType ValType = AtomTy; // 'C'
1320   if (IsC11) {
1321     if (!AtomTy->isAtomicType()) {
1322       Diag(DRE->getLocStart(), diag::err_atomic_op_needs_atomic)
1323         << Ptr->getType() << Ptr->getSourceRange();
1324       return ExprError();
1325     }
1326     if (AtomTy.isConstQualified()) {
1327       Diag(DRE->getLocStart(), diag::err_atomic_op_needs_non_const_atomic)
1328         << Ptr->getType() << Ptr->getSourceRange();
1329       return ExprError();
1330     }
1331     ValType = AtomTy->getAs<AtomicType>()->getValueType();
1332   }
1333 
1334   // For an arithmetic operation, the implied arithmetic must be well-formed.
1335   if (Form == Arithmetic) {
1336     // gcc does not enforce these rules for GNU atomics, but we do so for sanity.
1337     if (IsAddSub && !ValType->isIntegerType() && !ValType->isPointerType()) {
1338       Diag(DRE->getLocStart(), diag::err_atomic_op_needs_atomic_int_or_ptr)
1339         << IsC11 << Ptr->getType() << Ptr->getSourceRange();
1340       return ExprError();
1341     }
1342     if (!IsAddSub && !ValType->isIntegerType()) {
1343       Diag(DRE->getLocStart(), diag::err_atomic_op_bitwise_needs_atomic_int)
1344         << IsC11 << Ptr->getType() << Ptr->getSourceRange();
1345       return ExprError();
1346     }
1347     if (IsC11 && ValType->isPointerType() &&
1348         RequireCompleteType(Ptr->getLocStart(), ValType->getPointeeType(),
1349                             diag::err_incomplete_type)) {
1350       return ExprError();
1351     }
1352   } else if (IsN && !ValType->isIntegerType() && !ValType->isPointerType()) {
1353     // For __atomic_*_n operations, the value type must be a scalar integral or
1354     // pointer type which is 1, 2, 4, 8 or 16 bytes in length.
1355     Diag(DRE->getLocStart(), diag::err_atomic_op_needs_atomic_int_or_ptr)
1356       << IsC11 << Ptr->getType() << Ptr->getSourceRange();
1357     return ExprError();
1358   }
1359 
1360   if (!IsC11 && !AtomTy.isTriviallyCopyableType(Context) &&
1361       !AtomTy->isScalarType()) {
1362     // For GNU atomics, require a trivially-copyable type. This is not part of
1363     // the GNU atomics specification, but we enforce it for sanity.
1364     Diag(DRE->getLocStart(), diag::err_atomic_op_needs_trivial_copy)
1365       << Ptr->getType() << Ptr->getSourceRange();
1366     return ExprError();
1367   }
1368 
1369   // FIXME: For any builtin other than a load, the ValType must not be
1370   // const-qualified.
1371 
1372   switch (ValType.getObjCLifetime()) {
1373   case Qualifiers::OCL_None:
1374   case Qualifiers::OCL_ExplicitNone:
1375     // okay
1376     break;
1377 
1378   case Qualifiers::OCL_Weak:
1379   case Qualifiers::OCL_Strong:
1380   case Qualifiers::OCL_Autoreleasing:
1381     // FIXME: Can this happen? By this point, ValType should be known
1382     // to be trivially copyable.
1383     Diag(DRE->getLocStart(), diag::err_arc_atomic_ownership)
1384       << ValType << Ptr->getSourceRange();
1385     return ExprError();
1386   }
1387 
1388   QualType ResultType = ValType;
1389   if (Form == Copy || Form == GNUXchg || Form == Init)
1390     ResultType = Context.VoidTy;
1391   else if (Form == C11CmpXchg || Form == GNUCmpXchg)
1392     ResultType = Context.BoolTy;
1393 
1394   // The type of a parameter passed 'by value'. In the GNU atomics, such
1395   // arguments are actually passed as pointers.
1396   QualType ByValType = ValType; // 'CP'
1397   if (!IsC11 && !IsN)
1398     ByValType = Ptr->getType();
1399 
1400   // The first argument --- the pointer --- has a fixed type; we
1401   // deduce the types of the rest of the arguments accordingly.  Walk
1402   // the remaining arguments, converting them to the deduced value type.
1403   for (unsigned i = 1; i != NumArgs[Form]; ++i) {
1404     QualType Ty;
1405     if (i < NumVals[Form] + 1) {
1406       switch (i) {
1407       case 1:
1408         // The second argument is the non-atomic operand. For arithmetic, this
1409         // is always passed by value, and for a compare_exchange it is always
1410         // passed by address. For the rest, GNU uses by-address and C11 uses
1411         // by-value.
1412         assert(Form != Load);
1413         if (Form == Init || (Form == Arithmetic && ValType->isIntegerType()))
1414           Ty = ValType;
1415         else if (Form == Copy || Form == Xchg)
1416           Ty = ByValType;
1417         else if (Form == Arithmetic)
1418           Ty = Context.getPointerDiffType();
1419         else
1420           Ty = Context.getPointerType(ValType.getUnqualifiedType());
1421         break;
1422       case 2:
1423         // The third argument to compare_exchange / GNU exchange is a
1424         // (pointer to a) desired value.
1425         Ty = ByValType;
1426         break;
1427       case 3:
1428         // The fourth argument to GNU compare_exchange is a 'weak' flag.
1429         Ty = Context.BoolTy;
1430         break;
1431       }
1432     } else {
1433       // The order(s) are always converted to int.
1434       Ty = Context.IntTy;
1435     }
1436 
1437     InitializedEntity Entity =
1438         InitializedEntity::InitializeParameter(Context, Ty, false);
1439     ExprResult Arg = TheCall->getArg(i);
1440     Arg = PerformCopyInitialization(Entity, SourceLocation(), Arg);
1441     if (Arg.isInvalid())
1442       return true;
1443     TheCall->setArg(i, Arg.get());
1444   }
1445 
1446   // Permute the arguments into a 'consistent' order.
1447   SmallVector<Expr*, 5> SubExprs;
1448   SubExprs.push_back(Ptr);
1449   switch (Form) {
1450   case Init:
1451     // Note, AtomicExpr::getVal1() has a special case for this atomic.
1452     SubExprs.push_back(TheCall->getArg(1)); // Val1
1453     break;
1454   case Load:
1455     SubExprs.push_back(TheCall->getArg(1)); // Order
1456     break;
1457   case Copy:
1458   case Arithmetic:
1459   case Xchg:
1460     SubExprs.push_back(TheCall->getArg(2)); // Order
1461     SubExprs.push_back(TheCall->getArg(1)); // Val1
1462     break;
1463   case GNUXchg:
1464     // Note, AtomicExpr::getVal2() has a special case for this atomic.
1465     SubExprs.push_back(TheCall->getArg(3)); // Order
1466     SubExprs.push_back(TheCall->getArg(1)); // Val1
1467     SubExprs.push_back(TheCall->getArg(2)); // Val2
1468     break;
1469   case C11CmpXchg:
1470     SubExprs.push_back(TheCall->getArg(3)); // Order
1471     SubExprs.push_back(TheCall->getArg(1)); // Val1
1472     SubExprs.push_back(TheCall->getArg(4)); // OrderFail
1473     SubExprs.push_back(TheCall->getArg(2)); // Val2
1474     break;
1475   case GNUCmpXchg:
1476     SubExprs.push_back(TheCall->getArg(4)); // Order
1477     SubExprs.push_back(TheCall->getArg(1)); // Val1
1478     SubExprs.push_back(TheCall->getArg(5)); // OrderFail
1479     SubExprs.push_back(TheCall->getArg(2)); // Val2
1480     SubExprs.push_back(TheCall->getArg(3)); // Weak
1481     break;
1482   }
1483 
1484   if (SubExprs.size() >= 2 && Form != Init) {
1485     llvm::APSInt Result(32);
1486     if (SubExprs[1]->isIntegerConstantExpr(Result, Context) &&
1487         !isValidOrderingForOp(Result.getSExtValue(), Op))
1488       Diag(SubExprs[1]->getLocStart(),
1489            diag::warn_atomic_op_has_invalid_memory_order)
1490           << SubExprs[1]->getSourceRange();
1491   }
1492 
1493   AtomicExpr *AE = new (Context) AtomicExpr(TheCall->getCallee()->getLocStart(),
1494                                             SubExprs, ResultType, Op,
1495                                             TheCall->getRParenLoc());
1496 
1497   if ((Op == AtomicExpr::AO__c11_atomic_load ||
1498        (Op == AtomicExpr::AO__c11_atomic_store)) &&
1499       Context.AtomicUsesUnsupportedLibcall(AE))
1500     Diag(AE->getLocStart(), diag::err_atomic_load_store_uses_lib) <<
1501     ((Op == AtomicExpr::AO__c11_atomic_load) ? 0 : 1);
1502 
1503   return AE;
1504 }
1505 
1506 
1507 /// checkBuiltinArgument - Given a call to a builtin function, perform
1508 /// normal type-checking on the given argument, updating the call in
1509 /// place.  This is useful when a builtin function requires custom
1510 /// type-checking for some of its arguments but not necessarily all of
1511 /// them.
1512 ///
1513 /// Returns true on error.
checkBuiltinArgument(Sema & S,CallExpr * E,unsigned ArgIndex)1514 static bool checkBuiltinArgument(Sema &S, CallExpr *E, unsigned ArgIndex) {
1515   FunctionDecl *Fn = E->getDirectCallee();
1516   assert(Fn && "builtin call without direct callee!");
1517 
1518   ParmVarDecl *Param = Fn->getParamDecl(ArgIndex);
1519   InitializedEntity Entity =
1520     InitializedEntity::InitializeParameter(S.Context, Param);
1521 
1522   ExprResult Arg = E->getArg(0);
1523   Arg = S.PerformCopyInitialization(Entity, SourceLocation(), Arg);
1524   if (Arg.isInvalid())
1525     return true;
1526 
1527   E->setArg(ArgIndex, Arg.get());
1528   return false;
1529 }
1530 
1531 /// SemaBuiltinAtomicOverloaded - We have a call to a function like
1532 /// __sync_fetch_and_add, which is an overloaded function based on the pointer
1533 /// type of its first argument.  The main ActOnCallExpr routines have already
1534 /// promoted the types of arguments because all of these calls are prototyped as
1535 /// void(...).
1536 ///
1537 /// This function goes through and does final semantic checking for these
1538 /// builtins,
1539 ExprResult
SemaBuiltinAtomicOverloaded(ExprResult TheCallResult)1540 Sema::SemaBuiltinAtomicOverloaded(ExprResult TheCallResult) {
1541   CallExpr *TheCall = (CallExpr *)TheCallResult.get();
1542   DeclRefExpr *DRE =cast<DeclRefExpr>(TheCall->getCallee()->IgnoreParenCasts());
1543   FunctionDecl *FDecl = cast<FunctionDecl>(DRE->getDecl());
1544 
1545   // Ensure that we have at least one argument to do type inference from.
1546   if (TheCall->getNumArgs() < 1) {
1547     Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args_at_least)
1548       << 0 << 1 << TheCall->getNumArgs()
1549       << TheCall->getCallee()->getSourceRange();
1550     return ExprError();
1551   }
1552 
1553   // Inspect the first argument of the atomic builtin.  This should always be
1554   // a pointer type, whose element is an integral scalar or pointer type.
1555   // Because it is a pointer type, we don't have to worry about any implicit
1556   // casts here.
1557   // FIXME: We don't allow floating point scalars as input.
1558   Expr *FirstArg = TheCall->getArg(0);
1559   ExprResult FirstArgResult = DefaultFunctionArrayLvalueConversion(FirstArg);
1560   if (FirstArgResult.isInvalid())
1561     return ExprError();
1562   FirstArg = FirstArgResult.get();
1563   TheCall->setArg(0, FirstArg);
1564 
1565   const PointerType *pointerType = FirstArg->getType()->getAs<PointerType>();
1566   if (!pointerType) {
1567     Diag(DRE->getLocStart(), diag::err_atomic_builtin_must_be_pointer)
1568       << FirstArg->getType() << FirstArg->getSourceRange();
1569     return ExprError();
1570   }
1571 
1572   QualType ValType = pointerType->getPointeeType();
1573   if (!ValType->isIntegerType() && !ValType->isAnyPointerType() &&
1574       !ValType->isBlockPointerType()) {
1575     Diag(DRE->getLocStart(), diag::err_atomic_builtin_must_be_pointer_intptr)
1576       << FirstArg->getType() << FirstArg->getSourceRange();
1577     return ExprError();
1578   }
1579 
1580   switch (ValType.getObjCLifetime()) {
1581   case Qualifiers::OCL_None:
1582   case Qualifiers::OCL_ExplicitNone:
1583     // okay
1584     break;
1585 
1586   case Qualifiers::OCL_Weak:
1587   case Qualifiers::OCL_Strong:
1588   case Qualifiers::OCL_Autoreleasing:
1589     Diag(DRE->getLocStart(), diag::err_arc_atomic_ownership)
1590       << ValType << FirstArg->getSourceRange();
1591     return ExprError();
1592   }
1593 
1594   // Strip any qualifiers off ValType.
1595   ValType = ValType.getUnqualifiedType();
1596 
1597   // The majority of builtins return a value, but a few have special return
1598   // types, so allow them to override appropriately below.
1599   QualType ResultType = ValType;
1600 
1601   // We need to figure out which concrete builtin this maps onto.  For example,
1602   // __sync_fetch_and_add with a 2 byte object turns into
1603   // __sync_fetch_and_add_2.
1604 #define BUILTIN_ROW(x) \
1605   { Builtin::BI##x##_1, Builtin::BI##x##_2, Builtin::BI##x##_4, \
1606     Builtin::BI##x##_8, Builtin::BI##x##_16 }
1607 
1608   static const unsigned BuiltinIndices[][5] = {
1609     BUILTIN_ROW(__sync_fetch_and_add),
1610     BUILTIN_ROW(__sync_fetch_and_sub),
1611     BUILTIN_ROW(__sync_fetch_and_or),
1612     BUILTIN_ROW(__sync_fetch_and_and),
1613     BUILTIN_ROW(__sync_fetch_and_xor),
1614     BUILTIN_ROW(__sync_fetch_and_nand),
1615 
1616     BUILTIN_ROW(__sync_add_and_fetch),
1617     BUILTIN_ROW(__sync_sub_and_fetch),
1618     BUILTIN_ROW(__sync_and_and_fetch),
1619     BUILTIN_ROW(__sync_or_and_fetch),
1620     BUILTIN_ROW(__sync_xor_and_fetch),
1621     BUILTIN_ROW(__sync_nand_and_fetch),
1622 
1623     BUILTIN_ROW(__sync_val_compare_and_swap),
1624     BUILTIN_ROW(__sync_bool_compare_and_swap),
1625     BUILTIN_ROW(__sync_lock_test_and_set),
1626     BUILTIN_ROW(__sync_lock_release),
1627     BUILTIN_ROW(__sync_swap)
1628   };
1629 #undef BUILTIN_ROW
1630 
1631   // Determine the index of the size.
1632   unsigned SizeIndex;
1633   switch (Context.getTypeSizeInChars(ValType).getQuantity()) {
1634   case 1: SizeIndex = 0; break;
1635   case 2: SizeIndex = 1; break;
1636   case 4: SizeIndex = 2; break;
1637   case 8: SizeIndex = 3; break;
1638   case 16: SizeIndex = 4; break;
1639   default:
1640     Diag(DRE->getLocStart(), diag::err_atomic_builtin_pointer_size)
1641       << FirstArg->getType() << FirstArg->getSourceRange();
1642     return ExprError();
1643   }
1644 
1645   // Each of these builtins has one pointer argument, followed by some number of
1646   // values (0, 1 or 2) followed by a potentially empty varags list of stuff
1647   // that we ignore.  Find out which row of BuiltinIndices to read from as well
1648   // as the number of fixed args.
1649   unsigned BuiltinID = FDecl->getBuiltinID();
1650   unsigned BuiltinIndex, NumFixed = 1;
1651   bool WarnAboutSemanticsChange = false;
1652   switch (BuiltinID) {
1653   default: llvm_unreachable("Unknown overloaded atomic builtin!");
1654   case Builtin::BI__sync_fetch_and_add:
1655   case Builtin::BI__sync_fetch_and_add_1:
1656   case Builtin::BI__sync_fetch_and_add_2:
1657   case Builtin::BI__sync_fetch_and_add_4:
1658   case Builtin::BI__sync_fetch_and_add_8:
1659   case Builtin::BI__sync_fetch_and_add_16:
1660     BuiltinIndex = 0;
1661     break;
1662 
1663   case Builtin::BI__sync_fetch_and_sub:
1664   case Builtin::BI__sync_fetch_and_sub_1:
1665   case Builtin::BI__sync_fetch_and_sub_2:
1666   case Builtin::BI__sync_fetch_and_sub_4:
1667   case Builtin::BI__sync_fetch_and_sub_8:
1668   case Builtin::BI__sync_fetch_and_sub_16:
1669     BuiltinIndex = 1;
1670     break;
1671 
1672   case Builtin::BI__sync_fetch_and_or:
1673   case Builtin::BI__sync_fetch_and_or_1:
1674   case Builtin::BI__sync_fetch_and_or_2:
1675   case Builtin::BI__sync_fetch_and_or_4:
1676   case Builtin::BI__sync_fetch_and_or_8:
1677   case Builtin::BI__sync_fetch_and_or_16:
1678     BuiltinIndex = 2;
1679     break;
1680 
1681   case Builtin::BI__sync_fetch_and_and:
1682   case Builtin::BI__sync_fetch_and_and_1:
1683   case Builtin::BI__sync_fetch_and_and_2:
1684   case Builtin::BI__sync_fetch_and_and_4:
1685   case Builtin::BI__sync_fetch_and_and_8:
1686   case Builtin::BI__sync_fetch_and_and_16:
1687     BuiltinIndex = 3;
1688     break;
1689 
1690   case Builtin::BI__sync_fetch_and_xor:
1691   case Builtin::BI__sync_fetch_and_xor_1:
1692   case Builtin::BI__sync_fetch_and_xor_2:
1693   case Builtin::BI__sync_fetch_and_xor_4:
1694   case Builtin::BI__sync_fetch_and_xor_8:
1695   case Builtin::BI__sync_fetch_and_xor_16:
1696     BuiltinIndex = 4;
1697     break;
1698 
1699   case Builtin::BI__sync_fetch_and_nand:
1700   case Builtin::BI__sync_fetch_and_nand_1:
1701   case Builtin::BI__sync_fetch_and_nand_2:
1702   case Builtin::BI__sync_fetch_and_nand_4:
1703   case Builtin::BI__sync_fetch_and_nand_8:
1704   case Builtin::BI__sync_fetch_and_nand_16:
1705     BuiltinIndex = 5;
1706     WarnAboutSemanticsChange = true;
1707     break;
1708 
1709   case Builtin::BI__sync_add_and_fetch:
1710   case Builtin::BI__sync_add_and_fetch_1:
1711   case Builtin::BI__sync_add_and_fetch_2:
1712   case Builtin::BI__sync_add_and_fetch_4:
1713   case Builtin::BI__sync_add_and_fetch_8:
1714   case Builtin::BI__sync_add_and_fetch_16:
1715     BuiltinIndex = 6;
1716     break;
1717 
1718   case Builtin::BI__sync_sub_and_fetch:
1719   case Builtin::BI__sync_sub_and_fetch_1:
1720   case Builtin::BI__sync_sub_and_fetch_2:
1721   case Builtin::BI__sync_sub_and_fetch_4:
1722   case Builtin::BI__sync_sub_and_fetch_8:
1723   case Builtin::BI__sync_sub_and_fetch_16:
1724     BuiltinIndex = 7;
1725     break;
1726 
1727   case Builtin::BI__sync_and_and_fetch:
1728   case Builtin::BI__sync_and_and_fetch_1:
1729   case Builtin::BI__sync_and_and_fetch_2:
1730   case Builtin::BI__sync_and_and_fetch_4:
1731   case Builtin::BI__sync_and_and_fetch_8:
1732   case Builtin::BI__sync_and_and_fetch_16:
1733     BuiltinIndex = 8;
1734     break;
1735 
1736   case Builtin::BI__sync_or_and_fetch:
1737   case Builtin::BI__sync_or_and_fetch_1:
1738   case Builtin::BI__sync_or_and_fetch_2:
1739   case Builtin::BI__sync_or_and_fetch_4:
1740   case Builtin::BI__sync_or_and_fetch_8:
1741   case Builtin::BI__sync_or_and_fetch_16:
1742     BuiltinIndex = 9;
1743     break;
1744 
1745   case Builtin::BI__sync_xor_and_fetch:
1746   case Builtin::BI__sync_xor_and_fetch_1:
1747   case Builtin::BI__sync_xor_and_fetch_2:
1748   case Builtin::BI__sync_xor_and_fetch_4:
1749   case Builtin::BI__sync_xor_and_fetch_8:
1750   case Builtin::BI__sync_xor_and_fetch_16:
1751     BuiltinIndex = 10;
1752     break;
1753 
1754   case Builtin::BI__sync_nand_and_fetch:
1755   case Builtin::BI__sync_nand_and_fetch_1:
1756   case Builtin::BI__sync_nand_and_fetch_2:
1757   case Builtin::BI__sync_nand_and_fetch_4:
1758   case Builtin::BI__sync_nand_and_fetch_8:
1759   case Builtin::BI__sync_nand_and_fetch_16:
1760     BuiltinIndex = 11;
1761     WarnAboutSemanticsChange = true;
1762     break;
1763 
1764   case Builtin::BI__sync_val_compare_and_swap:
1765   case Builtin::BI__sync_val_compare_and_swap_1:
1766   case Builtin::BI__sync_val_compare_and_swap_2:
1767   case Builtin::BI__sync_val_compare_and_swap_4:
1768   case Builtin::BI__sync_val_compare_and_swap_8:
1769   case Builtin::BI__sync_val_compare_and_swap_16:
1770     BuiltinIndex = 12;
1771     NumFixed = 2;
1772     break;
1773 
1774   case Builtin::BI__sync_bool_compare_and_swap:
1775   case Builtin::BI__sync_bool_compare_and_swap_1:
1776   case Builtin::BI__sync_bool_compare_and_swap_2:
1777   case Builtin::BI__sync_bool_compare_and_swap_4:
1778   case Builtin::BI__sync_bool_compare_and_swap_8:
1779   case Builtin::BI__sync_bool_compare_and_swap_16:
1780     BuiltinIndex = 13;
1781     NumFixed = 2;
1782     ResultType = Context.BoolTy;
1783     break;
1784 
1785   case Builtin::BI__sync_lock_test_and_set:
1786   case Builtin::BI__sync_lock_test_and_set_1:
1787   case Builtin::BI__sync_lock_test_and_set_2:
1788   case Builtin::BI__sync_lock_test_and_set_4:
1789   case Builtin::BI__sync_lock_test_and_set_8:
1790   case Builtin::BI__sync_lock_test_and_set_16:
1791     BuiltinIndex = 14;
1792     break;
1793 
1794   case Builtin::BI__sync_lock_release:
1795   case Builtin::BI__sync_lock_release_1:
1796   case Builtin::BI__sync_lock_release_2:
1797   case Builtin::BI__sync_lock_release_4:
1798   case Builtin::BI__sync_lock_release_8:
1799   case Builtin::BI__sync_lock_release_16:
1800     BuiltinIndex = 15;
1801     NumFixed = 0;
1802     ResultType = Context.VoidTy;
1803     break;
1804 
1805   case Builtin::BI__sync_swap:
1806   case Builtin::BI__sync_swap_1:
1807   case Builtin::BI__sync_swap_2:
1808   case Builtin::BI__sync_swap_4:
1809   case Builtin::BI__sync_swap_8:
1810   case Builtin::BI__sync_swap_16:
1811     BuiltinIndex = 16;
1812     break;
1813   }
1814 
1815   // Now that we know how many fixed arguments we expect, first check that we
1816   // have at least that many.
1817   if (TheCall->getNumArgs() < 1+NumFixed) {
1818     Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args_at_least)
1819       << 0 << 1+NumFixed << TheCall->getNumArgs()
1820       << TheCall->getCallee()->getSourceRange();
1821     return ExprError();
1822   }
1823 
1824   if (WarnAboutSemanticsChange) {
1825     Diag(TheCall->getLocEnd(), diag::warn_sync_fetch_and_nand_semantics_change)
1826       << TheCall->getCallee()->getSourceRange();
1827   }
1828 
1829   // Get the decl for the concrete builtin from this, we can tell what the
1830   // concrete integer type we should convert to is.
1831   unsigned NewBuiltinID = BuiltinIndices[BuiltinIndex][SizeIndex];
1832   const char *NewBuiltinName = Context.BuiltinInfo.GetName(NewBuiltinID);
1833   FunctionDecl *NewBuiltinDecl;
1834   if (NewBuiltinID == BuiltinID)
1835     NewBuiltinDecl = FDecl;
1836   else {
1837     // Perform builtin lookup to avoid redeclaring it.
1838     DeclarationName DN(&Context.Idents.get(NewBuiltinName));
1839     LookupResult Res(*this, DN, DRE->getLocStart(), LookupOrdinaryName);
1840     LookupName(Res, TUScope, /*AllowBuiltinCreation=*/true);
1841     assert(Res.getFoundDecl());
1842     NewBuiltinDecl = dyn_cast<FunctionDecl>(Res.getFoundDecl());
1843     if (!NewBuiltinDecl)
1844       return ExprError();
1845   }
1846 
1847   // The first argument --- the pointer --- has a fixed type; we
1848   // deduce the types of the rest of the arguments accordingly.  Walk
1849   // the remaining arguments, converting them to the deduced value type.
1850   for (unsigned i = 0; i != NumFixed; ++i) {
1851     ExprResult Arg = TheCall->getArg(i+1);
1852 
1853     // GCC does an implicit conversion to the pointer or integer ValType.  This
1854     // can fail in some cases (1i -> int**), check for this error case now.
1855     // Initialize the argument.
1856     InitializedEntity Entity = InitializedEntity::InitializeParameter(Context,
1857                                                    ValType, /*consume*/ false);
1858     Arg = PerformCopyInitialization(Entity, SourceLocation(), Arg);
1859     if (Arg.isInvalid())
1860       return ExprError();
1861 
1862     // Okay, we have something that *can* be converted to the right type.  Check
1863     // to see if there is a potentially weird extension going on here.  This can
1864     // happen when you do an atomic operation on something like an char* and
1865     // pass in 42.  The 42 gets converted to char.  This is even more strange
1866     // for things like 45.123 -> char, etc.
1867     // FIXME: Do this check.
1868     TheCall->setArg(i+1, Arg.get());
1869   }
1870 
1871   ASTContext& Context = this->getASTContext();
1872 
1873   // Create a new DeclRefExpr to refer to the new decl.
1874   DeclRefExpr* NewDRE = DeclRefExpr::Create(
1875       Context,
1876       DRE->getQualifierLoc(),
1877       SourceLocation(),
1878       NewBuiltinDecl,
1879       /*enclosing*/ false,
1880       DRE->getLocation(),
1881       Context.BuiltinFnTy,
1882       DRE->getValueKind());
1883 
1884   // Set the callee in the CallExpr.
1885   // FIXME: This loses syntactic information.
1886   QualType CalleePtrTy = Context.getPointerType(NewBuiltinDecl->getType());
1887   ExprResult PromotedCall = ImpCastExprToType(NewDRE, CalleePtrTy,
1888                                               CK_BuiltinFnToFnPtr);
1889   TheCall->setCallee(PromotedCall.get());
1890 
1891   // Change the result type of the call to match the original value type. This
1892   // is arbitrary, but the codegen for these builtins ins design to handle it
1893   // gracefully.
1894   TheCall->setType(ResultType);
1895 
1896   return TheCallResult;
1897 }
1898 
1899 /// CheckObjCString - Checks that the argument to the builtin
1900 /// CFString constructor is correct
1901 /// Note: It might also make sense to do the UTF-16 conversion here (would
1902 /// simplify the backend).
CheckObjCString(Expr * Arg)1903 bool Sema::CheckObjCString(Expr *Arg) {
1904   Arg = Arg->IgnoreParenCasts();
1905   StringLiteral *Literal = dyn_cast<StringLiteral>(Arg);
1906 
1907   if (!Literal || !Literal->isAscii()) {
1908     Diag(Arg->getLocStart(), diag::err_cfstring_literal_not_string_constant)
1909       << Arg->getSourceRange();
1910     return true;
1911   }
1912 
1913   if (Literal->containsNonAsciiOrNull()) {
1914     StringRef String = Literal->getString();
1915     unsigned NumBytes = String.size();
1916     SmallVector<UTF16, 128> ToBuf(NumBytes);
1917     const UTF8 *FromPtr = (const UTF8 *)String.data();
1918     UTF16 *ToPtr = &ToBuf[0];
1919 
1920     ConversionResult Result = ConvertUTF8toUTF16(&FromPtr, FromPtr + NumBytes,
1921                                                  &ToPtr, ToPtr + NumBytes,
1922                                                  strictConversion);
1923     // Check for conversion failure.
1924     if (Result != conversionOK)
1925       Diag(Arg->getLocStart(),
1926            diag::warn_cfstring_truncated) << Arg->getSourceRange();
1927   }
1928   return false;
1929 }
1930 
1931 /// SemaBuiltinVAStart - Check the arguments to __builtin_va_start for validity.
1932 /// Emit an error and return true on failure, return false on success.
SemaBuiltinVAStart(CallExpr * TheCall)1933 bool Sema::SemaBuiltinVAStart(CallExpr *TheCall) {
1934   Expr *Fn = TheCall->getCallee();
1935   if (TheCall->getNumArgs() > 2) {
1936     Diag(TheCall->getArg(2)->getLocStart(),
1937          diag::err_typecheck_call_too_many_args)
1938       << 0 /*function call*/ << 2 << TheCall->getNumArgs()
1939       << Fn->getSourceRange()
1940       << SourceRange(TheCall->getArg(2)->getLocStart(),
1941                      (*(TheCall->arg_end()-1))->getLocEnd());
1942     return true;
1943   }
1944 
1945   if (TheCall->getNumArgs() < 2) {
1946     return Diag(TheCall->getLocEnd(),
1947       diag::err_typecheck_call_too_few_args_at_least)
1948       << 0 /*function call*/ << 2 << TheCall->getNumArgs();
1949   }
1950 
1951   // Type-check the first argument normally.
1952   if (checkBuiltinArgument(*this, TheCall, 0))
1953     return true;
1954 
1955   // Determine whether the current function is variadic or not.
1956   BlockScopeInfo *CurBlock = getCurBlock();
1957   bool isVariadic;
1958   if (CurBlock)
1959     isVariadic = CurBlock->TheDecl->isVariadic();
1960   else if (FunctionDecl *FD = getCurFunctionDecl())
1961     isVariadic = FD->isVariadic();
1962   else
1963     isVariadic = getCurMethodDecl()->isVariadic();
1964 
1965   if (!isVariadic) {
1966     Diag(Fn->getLocStart(), diag::err_va_start_used_in_non_variadic_function);
1967     return true;
1968   }
1969 
1970   // Verify that the second argument to the builtin is the last argument of the
1971   // current function or method.
1972   bool SecondArgIsLastNamedArgument = false;
1973   const Expr *Arg = TheCall->getArg(1)->IgnoreParenCasts();
1974 
1975   // These are valid if SecondArgIsLastNamedArgument is false after the next
1976   // block.
1977   QualType Type;
1978   SourceLocation ParamLoc;
1979 
1980   if (const DeclRefExpr *DR = dyn_cast<DeclRefExpr>(Arg)) {
1981     if (const ParmVarDecl *PV = dyn_cast<ParmVarDecl>(DR->getDecl())) {
1982       // FIXME: This isn't correct for methods (results in bogus warning).
1983       // Get the last formal in the current function.
1984       const ParmVarDecl *LastArg;
1985       if (CurBlock)
1986         LastArg = *(CurBlock->TheDecl->param_end()-1);
1987       else if (FunctionDecl *FD = getCurFunctionDecl())
1988         LastArg = *(FD->param_end()-1);
1989       else
1990         LastArg = *(getCurMethodDecl()->param_end()-1);
1991       SecondArgIsLastNamedArgument = PV == LastArg;
1992 
1993       Type = PV->getType();
1994       ParamLoc = PV->getLocation();
1995     }
1996   }
1997 
1998   if (!SecondArgIsLastNamedArgument)
1999     Diag(TheCall->getArg(1)->getLocStart(),
2000          diag::warn_second_parameter_of_va_start_not_last_named_argument);
2001   else if (Type->isReferenceType()) {
2002     Diag(Arg->getLocStart(),
2003          diag::warn_va_start_of_reference_type_is_undefined);
2004     Diag(ParamLoc, diag::note_parameter_type) << Type;
2005   }
2006 
2007   TheCall->setType(Context.VoidTy);
2008   return false;
2009 }
2010 
SemaBuiltinVAStartARM(CallExpr * Call)2011 bool Sema::SemaBuiltinVAStartARM(CallExpr *Call) {
2012   // void __va_start(va_list *ap, const char *named_addr, size_t slot_size,
2013   //                 const char *named_addr);
2014 
2015   Expr *Func = Call->getCallee();
2016 
2017   if (Call->getNumArgs() < 3)
2018     return Diag(Call->getLocEnd(),
2019                 diag::err_typecheck_call_too_few_args_at_least)
2020            << 0 /*function call*/ << 3 << Call->getNumArgs();
2021 
2022   // Determine whether the current function is variadic or not.
2023   bool IsVariadic;
2024   if (BlockScopeInfo *CurBlock = getCurBlock())
2025     IsVariadic = CurBlock->TheDecl->isVariadic();
2026   else if (FunctionDecl *FD = getCurFunctionDecl())
2027     IsVariadic = FD->isVariadic();
2028   else if (ObjCMethodDecl *MD = getCurMethodDecl())
2029     IsVariadic = MD->isVariadic();
2030   else
2031     llvm_unreachable("unexpected statement type");
2032 
2033   if (!IsVariadic) {
2034     Diag(Func->getLocStart(), diag::err_va_start_used_in_non_variadic_function);
2035     return true;
2036   }
2037 
2038   // Type-check the first argument normally.
2039   if (checkBuiltinArgument(*this, Call, 0))
2040     return true;
2041 
2042   static const struct {
2043     unsigned ArgNo;
2044     QualType Type;
2045   } ArgumentTypes[] = {
2046     { 1, Context.getPointerType(Context.CharTy.withConst()) },
2047     { 2, Context.getSizeType() },
2048   };
2049 
2050   for (const auto &AT : ArgumentTypes) {
2051     const Expr *Arg = Call->getArg(AT.ArgNo)->IgnoreParens();
2052     if (Arg->getType().getCanonicalType() == AT.Type.getCanonicalType())
2053       continue;
2054     Diag(Arg->getLocStart(), diag::err_typecheck_convert_incompatible)
2055       << Arg->getType() << AT.Type << 1 /* different class */
2056       << 0 /* qualifier difference */ << 3 /* parameter mismatch */
2057       << AT.ArgNo + 1 << Arg->getType() << AT.Type;
2058   }
2059 
2060   return false;
2061 }
2062 
2063 /// SemaBuiltinUnorderedCompare - Handle functions like __builtin_isgreater and
2064 /// friends.  This is declared to take (...), so we have to check everything.
SemaBuiltinUnorderedCompare(CallExpr * TheCall)2065 bool Sema::SemaBuiltinUnorderedCompare(CallExpr *TheCall) {
2066   if (TheCall->getNumArgs() < 2)
2067     return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args)
2068       << 0 << 2 << TheCall->getNumArgs()/*function call*/;
2069   if (TheCall->getNumArgs() > 2)
2070     return Diag(TheCall->getArg(2)->getLocStart(),
2071                 diag::err_typecheck_call_too_many_args)
2072       << 0 /*function call*/ << 2 << TheCall->getNumArgs()
2073       << SourceRange(TheCall->getArg(2)->getLocStart(),
2074                      (*(TheCall->arg_end()-1))->getLocEnd());
2075 
2076   ExprResult OrigArg0 = TheCall->getArg(0);
2077   ExprResult OrigArg1 = TheCall->getArg(1);
2078 
2079   // Do standard promotions between the two arguments, returning their common
2080   // type.
2081   QualType Res = UsualArithmeticConversions(OrigArg0, OrigArg1, false);
2082   if (OrigArg0.isInvalid() || OrigArg1.isInvalid())
2083     return true;
2084 
2085   // Make sure any conversions are pushed back into the call; this is
2086   // type safe since unordered compare builtins are declared as "_Bool
2087   // foo(...)".
2088   TheCall->setArg(0, OrigArg0.get());
2089   TheCall->setArg(1, OrigArg1.get());
2090 
2091   if (OrigArg0.get()->isTypeDependent() || OrigArg1.get()->isTypeDependent())
2092     return false;
2093 
2094   // If the common type isn't a real floating type, then the arguments were
2095   // invalid for this operation.
2096   if (Res.isNull() || !Res->isRealFloatingType())
2097     return Diag(OrigArg0.get()->getLocStart(),
2098                 diag::err_typecheck_call_invalid_ordered_compare)
2099       << OrigArg0.get()->getType() << OrigArg1.get()->getType()
2100       << SourceRange(OrigArg0.get()->getLocStart(), OrigArg1.get()->getLocEnd());
2101 
2102   return false;
2103 }
2104 
2105 /// SemaBuiltinSemaBuiltinFPClassification - Handle functions like
2106 /// __builtin_isnan and friends.  This is declared to take (...), so we have
2107 /// to check everything. We expect the last argument to be a floating point
2108 /// value.
SemaBuiltinFPClassification(CallExpr * TheCall,unsigned NumArgs)2109 bool Sema::SemaBuiltinFPClassification(CallExpr *TheCall, unsigned NumArgs) {
2110   if (TheCall->getNumArgs() < NumArgs)
2111     return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args)
2112       << 0 << NumArgs << TheCall->getNumArgs()/*function call*/;
2113   if (TheCall->getNumArgs() > NumArgs)
2114     return Diag(TheCall->getArg(NumArgs)->getLocStart(),
2115                 diag::err_typecheck_call_too_many_args)
2116       << 0 /*function call*/ << NumArgs << TheCall->getNumArgs()
2117       << SourceRange(TheCall->getArg(NumArgs)->getLocStart(),
2118                      (*(TheCall->arg_end()-1))->getLocEnd());
2119 
2120   Expr *OrigArg = TheCall->getArg(NumArgs-1);
2121 
2122   if (OrigArg->isTypeDependent())
2123     return false;
2124 
2125   // This operation requires a non-_Complex floating-point number.
2126   if (!OrigArg->getType()->isRealFloatingType())
2127     return Diag(OrigArg->getLocStart(),
2128                 diag::err_typecheck_call_invalid_unary_fp)
2129       << OrigArg->getType() << OrigArg->getSourceRange();
2130 
2131   // If this is an implicit conversion from float -> double, remove it.
2132   if (ImplicitCastExpr *Cast = dyn_cast<ImplicitCastExpr>(OrigArg)) {
2133     Expr *CastArg = Cast->getSubExpr();
2134     if (CastArg->getType()->isSpecificBuiltinType(BuiltinType::Float)) {
2135       assert(Cast->getType()->isSpecificBuiltinType(BuiltinType::Double) &&
2136              "promotion from float to double is the only expected cast here");
2137       Cast->setSubExpr(nullptr);
2138       TheCall->setArg(NumArgs-1, CastArg);
2139     }
2140   }
2141 
2142   return false;
2143 }
2144 
2145 /// SemaBuiltinShuffleVector - Handle __builtin_shufflevector.
2146 // This is declared to take (...), so we have to check everything.
SemaBuiltinShuffleVector(CallExpr * TheCall)2147 ExprResult Sema::SemaBuiltinShuffleVector(CallExpr *TheCall) {
2148   if (TheCall->getNumArgs() < 2)
2149     return ExprError(Diag(TheCall->getLocEnd(),
2150                           diag::err_typecheck_call_too_few_args_at_least)
2151                      << 0 /*function call*/ << 2 << TheCall->getNumArgs()
2152                      << TheCall->getSourceRange());
2153 
2154   // Determine which of the following types of shufflevector we're checking:
2155   // 1) unary, vector mask: (lhs, mask)
2156   // 2) binary, vector mask: (lhs, rhs, mask)
2157   // 3) binary, scalar mask: (lhs, rhs, index, ..., index)
2158   QualType resType = TheCall->getArg(0)->getType();
2159   unsigned numElements = 0;
2160 
2161   if (!TheCall->getArg(0)->isTypeDependent() &&
2162       !TheCall->getArg(1)->isTypeDependent()) {
2163     QualType LHSType = TheCall->getArg(0)->getType();
2164     QualType RHSType = TheCall->getArg(1)->getType();
2165 
2166     if (!LHSType->isVectorType() || !RHSType->isVectorType())
2167       return ExprError(Diag(TheCall->getLocStart(),
2168                             diag::err_shufflevector_non_vector)
2169                        << SourceRange(TheCall->getArg(0)->getLocStart(),
2170                                       TheCall->getArg(1)->getLocEnd()));
2171 
2172     numElements = LHSType->getAs<VectorType>()->getNumElements();
2173     unsigned numResElements = TheCall->getNumArgs() - 2;
2174 
2175     // Check to see if we have a call with 2 vector arguments, the unary shuffle
2176     // with mask.  If so, verify that RHS is an integer vector type with the
2177     // same number of elts as lhs.
2178     if (TheCall->getNumArgs() == 2) {
2179       if (!RHSType->hasIntegerRepresentation() ||
2180           RHSType->getAs<VectorType>()->getNumElements() != numElements)
2181         return ExprError(Diag(TheCall->getLocStart(),
2182                               diag::err_shufflevector_incompatible_vector)
2183                          << SourceRange(TheCall->getArg(1)->getLocStart(),
2184                                         TheCall->getArg(1)->getLocEnd()));
2185     } else if (!Context.hasSameUnqualifiedType(LHSType, RHSType)) {
2186       return ExprError(Diag(TheCall->getLocStart(),
2187                             diag::err_shufflevector_incompatible_vector)
2188                        << SourceRange(TheCall->getArg(0)->getLocStart(),
2189                                       TheCall->getArg(1)->getLocEnd()));
2190     } else if (numElements != numResElements) {
2191       QualType eltType = LHSType->getAs<VectorType>()->getElementType();
2192       resType = Context.getVectorType(eltType, numResElements,
2193                                       VectorType::GenericVector);
2194     }
2195   }
2196 
2197   for (unsigned i = 2; i < TheCall->getNumArgs(); i++) {
2198     if (TheCall->getArg(i)->isTypeDependent() ||
2199         TheCall->getArg(i)->isValueDependent())
2200       continue;
2201 
2202     llvm::APSInt Result(32);
2203     if (!TheCall->getArg(i)->isIntegerConstantExpr(Result, Context))
2204       return ExprError(Diag(TheCall->getLocStart(),
2205                             diag::err_shufflevector_nonconstant_argument)
2206                        << TheCall->getArg(i)->getSourceRange());
2207 
2208     // Allow -1 which will be translated to undef in the IR.
2209     if (Result.isSigned() && Result.isAllOnesValue())
2210       continue;
2211 
2212     if (Result.getActiveBits() > 64 || Result.getZExtValue() >= numElements*2)
2213       return ExprError(Diag(TheCall->getLocStart(),
2214                             diag::err_shufflevector_argument_too_large)
2215                        << TheCall->getArg(i)->getSourceRange());
2216   }
2217 
2218   SmallVector<Expr*, 32> exprs;
2219 
2220   for (unsigned i = 0, e = TheCall->getNumArgs(); i != e; i++) {
2221     exprs.push_back(TheCall->getArg(i));
2222     TheCall->setArg(i, nullptr);
2223   }
2224 
2225   return new (Context) ShuffleVectorExpr(Context, exprs, resType,
2226                                          TheCall->getCallee()->getLocStart(),
2227                                          TheCall->getRParenLoc());
2228 }
2229 
2230 /// SemaConvertVectorExpr - Handle __builtin_convertvector
SemaConvertVectorExpr(Expr * E,TypeSourceInfo * TInfo,SourceLocation BuiltinLoc,SourceLocation RParenLoc)2231 ExprResult Sema::SemaConvertVectorExpr(Expr *E, TypeSourceInfo *TInfo,
2232                                        SourceLocation BuiltinLoc,
2233                                        SourceLocation RParenLoc) {
2234   ExprValueKind VK = VK_RValue;
2235   ExprObjectKind OK = OK_Ordinary;
2236   QualType DstTy = TInfo->getType();
2237   QualType SrcTy = E->getType();
2238 
2239   if (!SrcTy->isVectorType() && !SrcTy->isDependentType())
2240     return ExprError(Diag(BuiltinLoc,
2241                           diag::err_convertvector_non_vector)
2242                      << E->getSourceRange());
2243   if (!DstTy->isVectorType() && !DstTy->isDependentType())
2244     return ExprError(Diag(BuiltinLoc,
2245                           diag::err_convertvector_non_vector_type));
2246 
2247   if (!SrcTy->isDependentType() && !DstTy->isDependentType()) {
2248     unsigned SrcElts = SrcTy->getAs<VectorType>()->getNumElements();
2249     unsigned DstElts = DstTy->getAs<VectorType>()->getNumElements();
2250     if (SrcElts != DstElts)
2251       return ExprError(Diag(BuiltinLoc,
2252                             diag::err_convertvector_incompatible_vector)
2253                        << E->getSourceRange());
2254   }
2255 
2256   return new (Context)
2257       ConvertVectorExpr(E, TInfo, DstTy, VK, OK, BuiltinLoc, RParenLoc);
2258 }
2259 
2260 /// SemaBuiltinPrefetch - Handle __builtin_prefetch.
2261 // This is declared to take (const void*, ...) and can take two
2262 // optional constant int args.
SemaBuiltinPrefetch(CallExpr * TheCall)2263 bool Sema::SemaBuiltinPrefetch(CallExpr *TheCall) {
2264   unsigned NumArgs = TheCall->getNumArgs();
2265 
2266   if (NumArgs > 3)
2267     return Diag(TheCall->getLocEnd(),
2268              diag::err_typecheck_call_too_many_args_at_most)
2269              << 0 /*function call*/ << 3 << NumArgs
2270              << TheCall->getSourceRange();
2271 
2272   // Argument 0 is checked for us and the remaining arguments must be
2273   // constant integers.
2274   for (unsigned i = 1; i != NumArgs; ++i)
2275     if (SemaBuiltinConstantArgRange(TheCall, i, 0, i == 1 ? 1 : 3))
2276       return true;
2277 
2278   return false;
2279 }
2280 
2281 /// SemaBuiltinAssume - Handle __assume (MS Extension).
2282 // __assume does not evaluate its arguments, and should warn if its argument
2283 // has side effects.
SemaBuiltinAssume(CallExpr * TheCall)2284 bool Sema::SemaBuiltinAssume(CallExpr *TheCall) {
2285   Expr *Arg = TheCall->getArg(0);
2286   if (Arg->isInstantiationDependent()) return false;
2287 
2288   if (Arg->HasSideEffects(Context))
2289     return Diag(Arg->getLocStart(), diag::warn_assume_side_effects)
2290       << Arg->getSourceRange()
2291       << cast<FunctionDecl>(TheCall->getCalleeDecl())->getIdentifier();
2292 
2293   return false;
2294 }
2295 
2296 /// Handle __builtin_assume_aligned. This is declared
2297 /// as (const void*, size_t, ...) and can take one optional constant int arg.
SemaBuiltinAssumeAligned(CallExpr * TheCall)2298 bool Sema::SemaBuiltinAssumeAligned(CallExpr *TheCall) {
2299   unsigned NumArgs = TheCall->getNumArgs();
2300 
2301   if (NumArgs > 3)
2302     return Diag(TheCall->getLocEnd(),
2303              diag::err_typecheck_call_too_many_args_at_most)
2304              << 0 /*function call*/ << 3 << NumArgs
2305              << TheCall->getSourceRange();
2306 
2307   // The alignment must be a constant integer.
2308   Expr *Arg = TheCall->getArg(1);
2309 
2310   // We can't check the value of a dependent argument.
2311   if (!Arg->isTypeDependent() && !Arg->isValueDependent()) {
2312     llvm::APSInt Result;
2313     if (SemaBuiltinConstantArg(TheCall, 1, Result))
2314       return true;
2315 
2316     if (!Result.isPowerOf2())
2317       return Diag(TheCall->getLocStart(),
2318                   diag::err_alignment_not_power_of_two)
2319            << Arg->getSourceRange();
2320   }
2321 
2322   if (NumArgs > 2) {
2323     ExprResult Arg(TheCall->getArg(2));
2324     InitializedEntity Entity = InitializedEntity::InitializeParameter(Context,
2325       Context.getSizeType(), false);
2326     Arg = PerformCopyInitialization(Entity, SourceLocation(), Arg);
2327     if (Arg.isInvalid()) return true;
2328     TheCall->setArg(2, Arg.get());
2329   }
2330 
2331   return false;
2332 }
2333 
2334 /// SemaBuiltinConstantArg - Handle a check if argument ArgNum of CallExpr
2335 /// TheCall is a constant expression.
SemaBuiltinConstantArg(CallExpr * TheCall,int ArgNum,llvm::APSInt & Result)2336 bool Sema::SemaBuiltinConstantArg(CallExpr *TheCall, int ArgNum,
2337                                   llvm::APSInt &Result) {
2338   Expr *Arg = TheCall->getArg(ArgNum);
2339   DeclRefExpr *DRE =cast<DeclRefExpr>(TheCall->getCallee()->IgnoreParenCasts());
2340   FunctionDecl *FDecl = cast<FunctionDecl>(DRE->getDecl());
2341 
2342   if (Arg->isTypeDependent() || Arg->isValueDependent()) return false;
2343 
2344   if (!Arg->isIntegerConstantExpr(Result, Context))
2345     return Diag(TheCall->getLocStart(), diag::err_constant_integer_arg_type)
2346                 << FDecl->getDeclName() <<  Arg->getSourceRange();
2347 
2348   return false;
2349 }
2350 
2351 /// SemaBuiltinConstantArgRange - Handle a check if argument ArgNum of CallExpr
2352 /// TheCall is a constant expression in the range [Low, High].
SemaBuiltinConstantArgRange(CallExpr * TheCall,int ArgNum,int Low,int High)2353 bool Sema::SemaBuiltinConstantArgRange(CallExpr *TheCall, int ArgNum,
2354                                        int Low, int High) {
2355   llvm::APSInt Result;
2356 
2357   // We can't check the value of a dependent argument.
2358   Expr *Arg = TheCall->getArg(ArgNum);
2359   if (Arg->isTypeDependent() || Arg->isValueDependent())
2360     return false;
2361 
2362   // Check constant-ness first.
2363   if (SemaBuiltinConstantArg(TheCall, ArgNum, Result))
2364     return true;
2365 
2366   if (Result.getSExtValue() < Low || Result.getSExtValue() > High)
2367     return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range)
2368       << Low << High << Arg->getSourceRange();
2369 
2370   return false;
2371 }
2372 
2373 /// SemaBuiltinLongjmp - Handle __builtin_longjmp(void *env[5], int val).
2374 /// This checks that the target supports __builtin_longjmp and
2375 /// that val is a constant 1.
SemaBuiltinLongjmp(CallExpr * TheCall)2376 bool Sema::SemaBuiltinLongjmp(CallExpr *TheCall) {
2377   if (!Context.getTargetInfo().hasSjLjLowering())
2378     return Diag(TheCall->getLocStart(), diag::err_builtin_longjmp_unsupported)
2379              << SourceRange(TheCall->getLocStart(), TheCall->getLocEnd());
2380 
2381   Expr *Arg = TheCall->getArg(1);
2382   llvm::APSInt Result;
2383 
2384   // TODO: This is less than ideal. Overload this to take a value.
2385   if (SemaBuiltinConstantArg(TheCall, 1, Result))
2386     return true;
2387 
2388   if (Result != 1)
2389     return Diag(TheCall->getLocStart(), diag::err_builtin_longjmp_invalid_val)
2390              << SourceRange(Arg->getLocStart(), Arg->getLocEnd());
2391 
2392   return false;
2393 }
2394 
2395 
2396 /// SemaBuiltinSetjmp - Handle __builtin_setjmp(void *env[5]).
2397 /// This checks that the target supports __builtin_setjmp.
SemaBuiltinSetjmp(CallExpr * TheCall)2398 bool Sema::SemaBuiltinSetjmp(CallExpr *TheCall) {
2399   if (!Context.getTargetInfo().hasSjLjLowering())
2400     return Diag(TheCall->getLocStart(), diag::err_builtin_setjmp_unsupported)
2401              << SourceRange(TheCall->getLocStart(), TheCall->getLocEnd());
2402   return false;
2403 }
2404 
2405 namespace {
2406 enum StringLiteralCheckType {
2407   SLCT_NotALiteral,
2408   SLCT_UncheckedLiteral,
2409   SLCT_CheckedLiteral
2410 };
2411 }
2412 
2413 // Determine if an expression is a string literal or constant string.
2414 // If this function returns false on the arguments to a function expecting a
2415 // format string, we will usually need to emit a warning.
2416 // True string literals are then checked by CheckFormatString.
2417 static StringLiteralCheckType
checkFormatStringExpr(Sema & S,const Expr * E,ArrayRef<const Expr * > Args,bool HasVAListArg,unsigned format_idx,unsigned firstDataArg,Sema::FormatStringType Type,Sema::VariadicCallType CallType,bool InFunctionCall,llvm::SmallBitVector & CheckedVarArgs)2418 checkFormatStringExpr(Sema &S, const Expr *E, ArrayRef<const Expr *> Args,
2419                       bool HasVAListArg, unsigned format_idx,
2420                       unsigned firstDataArg, Sema::FormatStringType Type,
2421                       Sema::VariadicCallType CallType, bool InFunctionCall,
2422                       llvm::SmallBitVector &CheckedVarArgs) {
2423  tryAgain:
2424   if (E->isTypeDependent() || E->isValueDependent())
2425     return SLCT_NotALiteral;
2426 
2427   E = E->IgnoreParenCasts();
2428 
2429   if (E->isNullPointerConstant(S.Context, Expr::NPC_ValueDependentIsNotNull))
2430     // Technically -Wformat-nonliteral does not warn about this case.
2431     // The behavior of printf and friends in this case is implementation
2432     // dependent.  Ideally if the format string cannot be null then
2433     // it should have a 'nonnull' attribute in the function prototype.
2434     return SLCT_UncheckedLiteral;
2435 
2436   switch (E->getStmtClass()) {
2437   case Stmt::BinaryConditionalOperatorClass:
2438   case Stmt::ConditionalOperatorClass: {
2439     // The expression is a literal if both sub-expressions were, and it was
2440     // completely checked only if both sub-expressions were checked.
2441     const AbstractConditionalOperator *C =
2442         cast<AbstractConditionalOperator>(E);
2443     StringLiteralCheckType Left =
2444         checkFormatStringExpr(S, C->getTrueExpr(), Args,
2445                               HasVAListArg, format_idx, firstDataArg,
2446                               Type, CallType, InFunctionCall, CheckedVarArgs);
2447     if (Left == SLCT_NotALiteral)
2448       return SLCT_NotALiteral;
2449     StringLiteralCheckType Right =
2450         checkFormatStringExpr(S, C->getFalseExpr(), Args,
2451                               HasVAListArg, format_idx, firstDataArg,
2452                               Type, CallType, InFunctionCall, CheckedVarArgs);
2453     return Left < Right ? Left : Right;
2454   }
2455 
2456   case Stmt::ImplicitCastExprClass: {
2457     E = cast<ImplicitCastExpr>(E)->getSubExpr();
2458     goto tryAgain;
2459   }
2460 
2461   case Stmt::OpaqueValueExprClass:
2462     if (const Expr *src = cast<OpaqueValueExpr>(E)->getSourceExpr()) {
2463       E = src;
2464       goto tryAgain;
2465     }
2466     return SLCT_NotALiteral;
2467 
2468   case Stmt::PredefinedExprClass:
2469     // While __func__, etc., are technically not string literals, they
2470     // cannot contain format specifiers and thus are not a security
2471     // liability.
2472     return SLCT_UncheckedLiteral;
2473 
2474   case Stmt::DeclRefExprClass: {
2475     const DeclRefExpr *DR = cast<DeclRefExpr>(E);
2476 
2477     // As an exception, do not flag errors for variables binding to
2478     // const string literals.
2479     if (const VarDecl *VD = dyn_cast<VarDecl>(DR->getDecl())) {
2480       bool isConstant = false;
2481       QualType T = DR->getType();
2482 
2483       if (const ArrayType *AT = S.Context.getAsArrayType(T)) {
2484         isConstant = AT->getElementType().isConstant(S.Context);
2485       } else if (const PointerType *PT = T->getAs<PointerType>()) {
2486         isConstant = T.isConstant(S.Context) &&
2487                      PT->getPointeeType().isConstant(S.Context);
2488       } else if (T->isObjCObjectPointerType()) {
2489         // In ObjC, there is usually no "const ObjectPointer" type,
2490         // so don't check if the pointee type is constant.
2491         isConstant = T.isConstant(S.Context);
2492       }
2493 
2494       if (isConstant) {
2495         if (const Expr *Init = VD->getAnyInitializer()) {
2496           // Look through initializers like const char c[] = { "foo" }
2497           if (const InitListExpr *InitList = dyn_cast<InitListExpr>(Init)) {
2498             if (InitList->isStringLiteralInit())
2499               Init = InitList->getInit(0)->IgnoreParenImpCasts();
2500           }
2501           return checkFormatStringExpr(S, Init, Args,
2502                                        HasVAListArg, format_idx,
2503                                        firstDataArg, Type, CallType,
2504                                        /*InFunctionCall*/false, CheckedVarArgs);
2505         }
2506       }
2507 
2508       // For vprintf* functions (i.e., HasVAListArg==true), we add a
2509       // special check to see if the format string is a function parameter
2510       // of the function calling the printf function.  If the function
2511       // has an attribute indicating it is a printf-like function, then we
2512       // should suppress warnings concerning non-literals being used in a call
2513       // to a vprintf function.  For example:
2514       //
2515       // void
2516       // logmessage(char const *fmt __attribute__ (format (printf, 1, 2)), ...){
2517       //      va_list ap;
2518       //      va_start(ap, fmt);
2519       //      vprintf(fmt, ap);  // Do NOT emit a warning about "fmt".
2520       //      ...
2521       // }
2522       if (HasVAListArg) {
2523         if (const ParmVarDecl *PV = dyn_cast<ParmVarDecl>(VD)) {
2524           if (const NamedDecl *ND = dyn_cast<NamedDecl>(PV->getDeclContext())) {
2525             int PVIndex = PV->getFunctionScopeIndex() + 1;
2526             for (const auto *PVFormat : ND->specific_attrs<FormatAttr>()) {
2527               // adjust for implicit parameter
2528               if (const CXXMethodDecl *MD = dyn_cast<CXXMethodDecl>(ND))
2529                 if (MD->isInstance())
2530                   ++PVIndex;
2531               // We also check if the formats are compatible.
2532               // We can't pass a 'scanf' string to a 'printf' function.
2533               if (PVIndex == PVFormat->getFormatIdx() &&
2534                   Type == S.GetFormatStringType(PVFormat))
2535                 return SLCT_UncheckedLiteral;
2536             }
2537           }
2538         }
2539       }
2540     }
2541 
2542     return SLCT_NotALiteral;
2543   }
2544 
2545   case Stmt::CallExprClass:
2546   case Stmt::CXXMemberCallExprClass: {
2547     const CallExpr *CE = cast<CallExpr>(E);
2548     if (const NamedDecl *ND = dyn_cast_or_null<NamedDecl>(CE->getCalleeDecl())) {
2549       if (const FormatArgAttr *FA = ND->getAttr<FormatArgAttr>()) {
2550         unsigned ArgIndex = FA->getFormatIdx();
2551         if (const CXXMethodDecl *MD = dyn_cast<CXXMethodDecl>(ND))
2552           if (MD->isInstance())
2553             --ArgIndex;
2554         const Expr *Arg = CE->getArg(ArgIndex - 1);
2555 
2556         return checkFormatStringExpr(S, Arg, Args,
2557                                      HasVAListArg, format_idx, firstDataArg,
2558                                      Type, CallType, InFunctionCall,
2559                                      CheckedVarArgs);
2560       } else if (const FunctionDecl *FD = dyn_cast<FunctionDecl>(ND)) {
2561         unsigned BuiltinID = FD->getBuiltinID();
2562         if (BuiltinID == Builtin::BI__builtin___CFStringMakeConstantString ||
2563             BuiltinID == Builtin::BI__builtin___NSStringMakeConstantString) {
2564           const Expr *Arg = CE->getArg(0);
2565           return checkFormatStringExpr(S, Arg, Args,
2566                                        HasVAListArg, format_idx,
2567                                        firstDataArg, Type, CallType,
2568                                        InFunctionCall, CheckedVarArgs);
2569         }
2570       }
2571     }
2572 
2573     return SLCT_NotALiteral;
2574   }
2575   case Stmt::ObjCStringLiteralClass:
2576   case Stmt::StringLiteralClass: {
2577     const StringLiteral *StrE = nullptr;
2578 
2579     if (const ObjCStringLiteral *ObjCFExpr = dyn_cast<ObjCStringLiteral>(E))
2580       StrE = ObjCFExpr->getString();
2581     else
2582       StrE = cast<StringLiteral>(E);
2583 
2584     if (StrE) {
2585       S.CheckFormatString(StrE, E, Args, HasVAListArg, format_idx, firstDataArg,
2586                           Type, InFunctionCall, CallType, CheckedVarArgs);
2587       return SLCT_CheckedLiteral;
2588     }
2589 
2590     return SLCT_NotALiteral;
2591   }
2592 
2593   default:
2594     return SLCT_NotALiteral;
2595   }
2596 }
2597 
GetFormatStringType(const FormatAttr * Format)2598 Sema::FormatStringType Sema::GetFormatStringType(const FormatAttr *Format) {
2599   return llvm::StringSwitch<FormatStringType>(Format->getType()->getName())
2600   .Case("scanf", FST_Scanf)
2601   .Cases("printf", "printf0", FST_Printf)
2602   .Cases("NSString", "CFString", FST_NSString)
2603   .Case("strftime", FST_Strftime)
2604   .Case("strfmon", FST_Strfmon)
2605   .Cases("kprintf", "cmn_err", "vcmn_err", "zcmn_err", FST_Kprintf)
2606   .Default(FST_Unknown);
2607 }
2608 
2609 /// CheckFormatArguments - Check calls to printf and scanf (and similar
2610 /// functions) for correct use of format strings.
2611 /// Returns true if a format string has been fully checked.
CheckFormatArguments(const FormatAttr * Format,ArrayRef<const Expr * > Args,bool IsCXXMember,VariadicCallType CallType,SourceLocation Loc,SourceRange Range,llvm::SmallBitVector & CheckedVarArgs)2612 bool Sema::CheckFormatArguments(const FormatAttr *Format,
2613                                 ArrayRef<const Expr *> Args,
2614                                 bool IsCXXMember,
2615                                 VariadicCallType CallType,
2616                                 SourceLocation Loc, SourceRange Range,
2617                                 llvm::SmallBitVector &CheckedVarArgs) {
2618   FormatStringInfo FSI;
2619   if (getFormatStringInfo(Format, IsCXXMember, &FSI))
2620     return CheckFormatArguments(Args, FSI.HasVAListArg, FSI.FormatIdx,
2621                                 FSI.FirstDataArg, GetFormatStringType(Format),
2622                                 CallType, Loc, Range, CheckedVarArgs);
2623   return false;
2624 }
2625 
CheckFormatArguments(ArrayRef<const Expr * > Args,bool HasVAListArg,unsigned format_idx,unsigned firstDataArg,FormatStringType Type,VariadicCallType CallType,SourceLocation Loc,SourceRange Range,llvm::SmallBitVector & CheckedVarArgs)2626 bool Sema::CheckFormatArguments(ArrayRef<const Expr *> Args,
2627                                 bool HasVAListArg, unsigned format_idx,
2628                                 unsigned firstDataArg, FormatStringType Type,
2629                                 VariadicCallType CallType,
2630                                 SourceLocation Loc, SourceRange Range,
2631                                 llvm::SmallBitVector &CheckedVarArgs) {
2632   // CHECK: printf/scanf-like function is called with no format string.
2633   if (format_idx >= Args.size()) {
2634     Diag(Loc, diag::warn_missing_format_string) << Range;
2635     return false;
2636   }
2637 
2638   const Expr *OrigFormatExpr = Args[format_idx]->IgnoreParenCasts();
2639 
2640   // CHECK: format string is not a string literal.
2641   //
2642   // Dynamically generated format strings are difficult to
2643   // automatically vet at compile time.  Requiring that format strings
2644   // are string literals: (1) permits the checking of format strings by
2645   // the compiler and thereby (2) can practically remove the source of
2646   // many format string exploits.
2647 
2648   // Format string can be either ObjC string (e.g. @"%d") or
2649   // C string (e.g. "%d")
2650   // ObjC string uses the same format specifiers as C string, so we can use
2651   // the same format string checking logic for both ObjC and C strings.
2652   StringLiteralCheckType CT =
2653       checkFormatStringExpr(*this, OrigFormatExpr, Args, HasVAListArg,
2654                             format_idx, firstDataArg, Type, CallType,
2655                             /*IsFunctionCall*/true, CheckedVarArgs);
2656   if (CT != SLCT_NotALiteral)
2657     // Literal format string found, check done!
2658     return CT == SLCT_CheckedLiteral;
2659 
2660   // Strftime is particular as it always uses a single 'time' argument,
2661   // so it is safe to pass a non-literal string.
2662   if (Type == FST_Strftime)
2663     return false;
2664 
2665   // Do not emit diag when the string param is a macro expansion and the
2666   // format is either NSString or CFString. This is a hack to prevent
2667   // diag when using the NSLocalizedString and CFCopyLocalizedString macros
2668   // which are usually used in place of NS and CF string literals.
2669   if (Type == FST_NSString &&
2670       SourceMgr.isInSystemMacro(Args[format_idx]->getLocStart()))
2671     return false;
2672 
2673   // If there are no arguments specified, warn with -Wformat-security, otherwise
2674   // warn only with -Wformat-nonliteral.
2675   if (Args.size() == firstDataArg)
2676     Diag(Args[format_idx]->getLocStart(),
2677          diag::warn_format_nonliteral_noargs)
2678       << OrigFormatExpr->getSourceRange();
2679   else
2680     Diag(Args[format_idx]->getLocStart(),
2681          diag::warn_format_nonliteral)
2682            << OrigFormatExpr->getSourceRange();
2683   return false;
2684 }
2685 
2686 namespace {
2687 class CheckFormatHandler : public analyze_format_string::FormatStringHandler {
2688 protected:
2689   Sema &S;
2690   const StringLiteral *FExpr;
2691   const Expr *OrigFormatExpr;
2692   const unsigned FirstDataArg;
2693   const unsigned NumDataArgs;
2694   const char *Beg; // Start of format string.
2695   const bool HasVAListArg;
2696   ArrayRef<const Expr *> Args;
2697   unsigned FormatIdx;
2698   llvm::SmallBitVector CoveredArgs;
2699   bool usesPositionalArgs;
2700   bool atFirstArg;
2701   bool inFunctionCall;
2702   Sema::VariadicCallType CallType;
2703   llvm::SmallBitVector &CheckedVarArgs;
2704 public:
CheckFormatHandler(Sema & s,const StringLiteral * fexpr,const Expr * origFormatExpr,unsigned firstDataArg,unsigned numDataArgs,const char * beg,bool hasVAListArg,ArrayRef<const Expr * > Args,unsigned formatIdx,bool inFunctionCall,Sema::VariadicCallType callType,llvm::SmallBitVector & CheckedVarArgs)2705   CheckFormatHandler(Sema &s, const StringLiteral *fexpr,
2706                      const Expr *origFormatExpr, unsigned firstDataArg,
2707                      unsigned numDataArgs, const char *beg, bool hasVAListArg,
2708                      ArrayRef<const Expr *> Args,
2709                      unsigned formatIdx, bool inFunctionCall,
2710                      Sema::VariadicCallType callType,
2711                      llvm::SmallBitVector &CheckedVarArgs)
2712     : S(s), FExpr(fexpr), OrigFormatExpr(origFormatExpr),
2713       FirstDataArg(firstDataArg), NumDataArgs(numDataArgs),
2714       Beg(beg), HasVAListArg(hasVAListArg),
2715       Args(Args), FormatIdx(formatIdx),
2716       usesPositionalArgs(false), atFirstArg(true),
2717       inFunctionCall(inFunctionCall), CallType(callType),
2718       CheckedVarArgs(CheckedVarArgs) {
2719     CoveredArgs.resize(numDataArgs);
2720     CoveredArgs.reset();
2721   }
2722 
2723   void DoneProcessing();
2724 
2725   void HandleIncompleteSpecifier(const char *startSpecifier,
2726                                  unsigned specifierLen) override;
2727 
2728   void HandleInvalidLengthModifier(
2729                            const analyze_format_string::FormatSpecifier &FS,
2730                            const analyze_format_string::ConversionSpecifier &CS,
2731                            const char *startSpecifier, unsigned specifierLen,
2732                            unsigned DiagID);
2733 
2734   void HandleNonStandardLengthModifier(
2735                     const analyze_format_string::FormatSpecifier &FS,
2736                     const char *startSpecifier, unsigned specifierLen);
2737 
2738   void HandleNonStandardConversionSpecifier(
2739                     const analyze_format_string::ConversionSpecifier &CS,
2740                     const char *startSpecifier, unsigned specifierLen);
2741 
2742   void HandlePosition(const char *startPos, unsigned posLen) override;
2743 
2744   void HandleInvalidPosition(const char *startSpecifier,
2745                              unsigned specifierLen,
2746                              analyze_format_string::PositionContext p) override;
2747 
2748   void HandleZeroPosition(const char *startPos, unsigned posLen) override;
2749 
2750   void HandleNullChar(const char *nullCharacter) override;
2751 
2752   template <typename Range>
2753   static void EmitFormatDiagnostic(Sema &S, bool inFunctionCall,
2754                                    const Expr *ArgumentExpr,
2755                                    PartialDiagnostic PDiag,
2756                                    SourceLocation StringLoc,
2757                                    bool IsStringLocation, Range StringRange,
2758                                    ArrayRef<FixItHint> Fixit = None);
2759 
2760 protected:
2761   bool HandleInvalidConversionSpecifier(unsigned argIndex, SourceLocation Loc,
2762                                         const char *startSpec,
2763                                         unsigned specifierLen,
2764                                         const char *csStart, unsigned csLen);
2765 
2766   void HandlePositionalNonpositionalArgs(SourceLocation Loc,
2767                                          const char *startSpec,
2768                                          unsigned specifierLen);
2769 
2770   SourceRange getFormatStringRange();
2771   CharSourceRange getSpecifierRange(const char *startSpecifier,
2772                                     unsigned specifierLen);
2773   SourceLocation getLocationOfByte(const char *x);
2774 
2775   const Expr *getDataArg(unsigned i) const;
2776 
2777   bool CheckNumArgs(const analyze_format_string::FormatSpecifier &FS,
2778                     const analyze_format_string::ConversionSpecifier &CS,
2779                     const char *startSpecifier, unsigned specifierLen,
2780                     unsigned argIndex);
2781 
2782   template <typename Range>
2783   void EmitFormatDiagnostic(PartialDiagnostic PDiag, SourceLocation StringLoc,
2784                             bool IsStringLocation, Range StringRange,
2785                             ArrayRef<FixItHint> Fixit = None);
2786 };
2787 }
2788 
getFormatStringRange()2789 SourceRange CheckFormatHandler::getFormatStringRange() {
2790   return OrigFormatExpr->getSourceRange();
2791 }
2792 
2793 CharSourceRange CheckFormatHandler::
getSpecifierRange(const char * startSpecifier,unsigned specifierLen)2794 getSpecifierRange(const char *startSpecifier, unsigned specifierLen) {
2795   SourceLocation Start = getLocationOfByte(startSpecifier);
2796   SourceLocation End   = getLocationOfByte(startSpecifier + specifierLen - 1);
2797 
2798   // Advance the end SourceLocation by one due to half-open ranges.
2799   End = End.getLocWithOffset(1);
2800 
2801   return CharSourceRange::getCharRange(Start, End);
2802 }
2803 
getLocationOfByte(const char * x)2804 SourceLocation CheckFormatHandler::getLocationOfByte(const char *x) {
2805   return S.getLocationOfStringLiteralByte(FExpr, x - Beg);
2806 }
2807 
HandleIncompleteSpecifier(const char * startSpecifier,unsigned specifierLen)2808 void CheckFormatHandler::HandleIncompleteSpecifier(const char *startSpecifier,
2809                                                    unsigned specifierLen){
2810   EmitFormatDiagnostic(S.PDiag(diag::warn_printf_incomplete_specifier),
2811                        getLocationOfByte(startSpecifier),
2812                        /*IsStringLocation*/true,
2813                        getSpecifierRange(startSpecifier, specifierLen));
2814 }
2815 
HandleInvalidLengthModifier(const analyze_format_string::FormatSpecifier & FS,const analyze_format_string::ConversionSpecifier & CS,const char * startSpecifier,unsigned specifierLen,unsigned DiagID)2816 void CheckFormatHandler::HandleInvalidLengthModifier(
2817     const analyze_format_string::FormatSpecifier &FS,
2818     const analyze_format_string::ConversionSpecifier &CS,
2819     const char *startSpecifier, unsigned specifierLen, unsigned DiagID) {
2820   using namespace analyze_format_string;
2821 
2822   const LengthModifier &LM = FS.getLengthModifier();
2823   CharSourceRange LMRange = getSpecifierRange(LM.getStart(), LM.getLength());
2824 
2825   // See if we know how to fix this length modifier.
2826   Optional<LengthModifier> FixedLM = FS.getCorrectedLengthModifier();
2827   if (FixedLM) {
2828     EmitFormatDiagnostic(S.PDiag(DiagID) << LM.toString() << CS.toString(),
2829                          getLocationOfByte(LM.getStart()),
2830                          /*IsStringLocation*/true,
2831                          getSpecifierRange(startSpecifier, specifierLen));
2832 
2833     S.Diag(getLocationOfByte(LM.getStart()), diag::note_format_fix_specifier)
2834       << FixedLM->toString()
2835       << FixItHint::CreateReplacement(LMRange, FixedLM->toString());
2836 
2837   } else {
2838     FixItHint Hint;
2839     if (DiagID == diag::warn_format_nonsensical_length)
2840       Hint = FixItHint::CreateRemoval(LMRange);
2841 
2842     EmitFormatDiagnostic(S.PDiag(DiagID) << LM.toString() << CS.toString(),
2843                          getLocationOfByte(LM.getStart()),
2844                          /*IsStringLocation*/true,
2845                          getSpecifierRange(startSpecifier, specifierLen),
2846                          Hint);
2847   }
2848 }
2849 
HandleNonStandardLengthModifier(const analyze_format_string::FormatSpecifier & FS,const char * startSpecifier,unsigned specifierLen)2850 void CheckFormatHandler::HandleNonStandardLengthModifier(
2851     const analyze_format_string::FormatSpecifier &FS,
2852     const char *startSpecifier, unsigned specifierLen) {
2853   using namespace analyze_format_string;
2854 
2855   const LengthModifier &LM = FS.getLengthModifier();
2856   CharSourceRange LMRange = getSpecifierRange(LM.getStart(), LM.getLength());
2857 
2858   // See if we know how to fix this length modifier.
2859   Optional<LengthModifier> FixedLM = FS.getCorrectedLengthModifier();
2860   if (FixedLM) {
2861     EmitFormatDiagnostic(S.PDiag(diag::warn_format_non_standard)
2862                            << LM.toString() << 0,
2863                          getLocationOfByte(LM.getStart()),
2864                          /*IsStringLocation*/true,
2865                          getSpecifierRange(startSpecifier, specifierLen));
2866 
2867     S.Diag(getLocationOfByte(LM.getStart()), diag::note_format_fix_specifier)
2868       << FixedLM->toString()
2869       << FixItHint::CreateReplacement(LMRange, FixedLM->toString());
2870 
2871   } else {
2872     EmitFormatDiagnostic(S.PDiag(diag::warn_format_non_standard)
2873                            << LM.toString() << 0,
2874                          getLocationOfByte(LM.getStart()),
2875                          /*IsStringLocation*/true,
2876                          getSpecifierRange(startSpecifier, specifierLen));
2877   }
2878 }
2879 
HandleNonStandardConversionSpecifier(const analyze_format_string::ConversionSpecifier & CS,const char * startSpecifier,unsigned specifierLen)2880 void CheckFormatHandler::HandleNonStandardConversionSpecifier(
2881     const analyze_format_string::ConversionSpecifier &CS,
2882     const char *startSpecifier, unsigned specifierLen) {
2883   using namespace analyze_format_string;
2884 
2885   // See if we know how to fix this conversion specifier.
2886   Optional<ConversionSpecifier> FixedCS = CS.getStandardSpecifier();
2887   if (FixedCS) {
2888     EmitFormatDiagnostic(S.PDiag(diag::warn_format_non_standard)
2889                           << CS.toString() << /*conversion specifier*/1,
2890                          getLocationOfByte(CS.getStart()),
2891                          /*IsStringLocation*/true,
2892                          getSpecifierRange(startSpecifier, specifierLen));
2893 
2894     CharSourceRange CSRange = getSpecifierRange(CS.getStart(), CS.getLength());
2895     S.Diag(getLocationOfByte(CS.getStart()), diag::note_format_fix_specifier)
2896       << FixedCS->toString()
2897       << FixItHint::CreateReplacement(CSRange, FixedCS->toString());
2898   } else {
2899     EmitFormatDiagnostic(S.PDiag(diag::warn_format_non_standard)
2900                           << CS.toString() << /*conversion specifier*/1,
2901                          getLocationOfByte(CS.getStart()),
2902                          /*IsStringLocation*/true,
2903                          getSpecifierRange(startSpecifier, specifierLen));
2904   }
2905 }
2906 
HandlePosition(const char * startPos,unsigned posLen)2907 void CheckFormatHandler::HandlePosition(const char *startPos,
2908                                         unsigned posLen) {
2909   EmitFormatDiagnostic(S.PDiag(diag::warn_format_non_standard_positional_arg),
2910                                getLocationOfByte(startPos),
2911                                /*IsStringLocation*/true,
2912                                getSpecifierRange(startPos, posLen));
2913 }
2914 
2915 void
HandleInvalidPosition(const char * startPos,unsigned posLen,analyze_format_string::PositionContext p)2916 CheckFormatHandler::HandleInvalidPosition(const char *startPos, unsigned posLen,
2917                                      analyze_format_string::PositionContext p) {
2918   EmitFormatDiagnostic(S.PDiag(diag::warn_format_invalid_positional_specifier)
2919                          << (unsigned) p,
2920                        getLocationOfByte(startPos), /*IsStringLocation*/true,
2921                        getSpecifierRange(startPos, posLen));
2922 }
2923 
HandleZeroPosition(const char * startPos,unsigned posLen)2924 void CheckFormatHandler::HandleZeroPosition(const char *startPos,
2925                                             unsigned posLen) {
2926   EmitFormatDiagnostic(S.PDiag(diag::warn_format_zero_positional_specifier),
2927                                getLocationOfByte(startPos),
2928                                /*IsStringLocation*/true,
2929                                getSpecifierRange(startPos, posLen));
2930 }
2931 
HandleNullChar(const char * nullCharacter)2932 void CheckFormatHandler::HandleNullChar(const char *nullCharacter) {
2933   if (!isa<ObjCStringLiteral>(OrigFormatExpr)) {
2934     // The presence of a null character is likely an error.
2935     EmitFormatDiagnostic(
2936       S.PDiag(diag::warn_printf_format_string_contains_null_char),
2937       getLocationOfByte(nullCharacter), /*IsStringLocation*/true,
2938       getFormatStringRange());
2939   }
2940 }
2941 
2942 // Note that this may return NULL if there was an error parsing or building
2943 // one of the argument expressions.
getDataArg(unsigned i) const2944 const Expr *CheckFormatHandler::getDataArg(unsigned i) const {
2945   return Args[FirstDataArg + i];
2946 }
2947 
DoneProcessing()2948 void CheckFormatHandler::DoneProcessing() {
2949     // Does the number of data arguments exceed the number of
2950     // format conversions in the format string?
2951   if (!HasVAListArg) {
2952       // Find any arguments that weren't covered.
2953     CoveredArgs.flip();
2954     signed notCoveredArg = CoveredArgs.find_first();
2955     if (notCoveredArg >= 0) {
2956       assert((unsigned)notCoveredArg < NumDataArgs);
2957       if (const Expr *E = getDataArg((unsigned) notCoveredArg)) {
2958         SourceLocation Loc = E->getLocStart();
2959         if (!S.getSourceManager().isInSystemMacro(Loc)) {
2960           EmitFormatDiagnostic(S.PDiag(diag::warn_printf_data_arg_not_used),
2961                                Loc, /*IsStringLocation*/false,
2962                                getFormatStringRange());
2963         }
2964       }
2965     }
2966   }
2967 }
2968 
2969 bool
HandleInvalidConversionSpecifier(unsigned argIndex,SourceLocation Loc,const char * startSpec,unsigned specifierLen,const char * csStart,unsigned csLen)2970 CheckFormatHandler::HandleInvalidConversionSpecifier(unsigned argIndex,
2971                                                      SourceLocation Loc,
2972                                                      const char *startSpec,
2973                                                      unsigned specifierLen,
2974                                                      const char *csStart,
2975                                                      unsigned csLen) {
2976 
2977   bool keepGoing = true;
2978   if (argIndex < NumDataArgs) {
2979     // Consider the argument coverered, even though the specifier doesn't
2980     // make sense.
2981     CoveredArgs.set(argIndex);
2982   }
2983   else {
2984     // If argIndex exceeds the number of data arguments we
2985     // don't issue a warning because that is just a cascade of warnings (and
2986     // they may have intended '%%' anyway). We don't want to continue processing
2987     // the format string after this point, however, as we will like just get
2988     // gibberish when trying to match arguments.
2989     keepGoing = false;
2990   }
2991 
2992   EmitFormatDiagnostic(S.PDiag(diag::warn_format_invalid_conversion)
2993                          << StringRef(csStart, csLen),
2994                        Loc, /*IsStringLocation*/true,
2995                        getSpecifierRange(startSpec, specifierLen));
2996 
2997   return keepGoing;
2998 }
2999 
3000 void
HandlePositionalNonpositionalArgs(SourceLocation Loc,const char * startSpec,unsigned specifierLen)3001 CheckFormatHandler::HandlePositionalNonpositionalArgs(SourceLocation Loc,
3002                                                       const char *startSpec,
3003                                                       unsigned specifierLen) {
3004   EmitFormatDiagnostic(
3005     S.PDiag(diag::warn_format_mix_positional_nonpositional_args),
3006     Loc, /*isStringLoc*/true, getSpecifierRange(startSpec, specifierLen));
3007 }
3008 
3009 bool
CheckNumArgs(const analyze_format_string::FormatSpecifier & FS,const analyze_format_string::ConversionSpecifier & CS,const char * startSpecifier,unsigned specifierLen,unsigned argIndex)3010 CheckFormatHandler::CheckNumArgs(
3011   const analyze_format_string::FormatSpecifier &FS,
3012   const analyze_format_string::ConversionSpecifier &CS,
3013   const char *startSpecifier, unsigned specifierLen, unsigned argIndex) {
3014 
3015   if (argIndex >= NumDataArgs) {
3016     PartialDiagnostic PDiag = FS.usesPositionalArg()
3017       ? (S.PDiag(diag::warn_printf_positional_arg_exceeds_data_args)
3018            << (argIndex+1) << NumDataArgs)
3019       : S.PDiag(diag::warn_printf_insufficient_data_args);
3020     EmitFormatDiagnostic(
3021       PDiag, getLocationOfByte(CS.getStart()), /*IsStringLocation*/true,
3022       getSpecifierRange(startSpecifier, specifierLen));
3023     return false;
3024   }
3025   return true;
3026 }
3027 
3028 template<typename Range>
EmitFormatDiagnostic(PartialDiagnostic PDiag,SourceLocation Loc,bool IsStringLocation,Range StringRange,ArrayRef<FixItHint> FixIt)3029 void CheckFormatHandler::EmitFormatDiagnostic(PartialDiagnostic PDiag,
3030                                               SourceLocation Loc,
3031                                               bool IsStringLocation,
3032                                               Range StringRange,
3033                                               ArrayRef<FixItHint> FixIt) {
3034   EmitFormatDiagnostic(S, inFunctionCall, Args[FormatIdx], PDiag,
3035                        Loc, IsStringLocation, StringRange, FixIt);
3036 }
3037 
3038 /// \brief If the format string is not within the funcion call, emit a note
3039 /// so that the function call and string are in diagnostic messages.
3040 ///
3041 /// \param InFunctionCall if true, the format string is within the function
3042 /// call and only one diagnostic message will be produced.  Otherwise, an
3043 /// extra note will be emitted pointing to location of the format string.
3044 ///
3045 /// \param ArgumentExpr the expression that is passed as the format string
3046 /// argument in the function call.  Used for getting locations when two
3047 /// diagnostics are emitted.
3048 ///
3049 /// \param PDiag the callee should already have provided any strings for the
3050 /// diagnostic message.  This function only adds locations and fixits
3051 /// to diagnostics.
3052 ///
3053 /// \param Loc primary location for diagnostic.  If two diagnostics are
3054 /// required, one will be at Loc and a new SourceLocation will be created for
3055 /// the other one.
3056 ///
3057 /// \param IsStringLocation if true, Loc points to the format string should be
3058 /// used for the note.  Otherwise, Loc points to the argument list and will
3059 /// be used with PDiag.
3060 ///
3061 /// \param StringRange some or all of the string to highlight.  This is
3062 /// templated so it can accept either a CharSourceRange or a SourceRange.
3063 ///
3064 /// \param FixIt optional fix it hint for the format string.
3065 template<typename Range>
EmitFormatDiagnostic(Sema & S,bool InFunctionCall,const Expr * ArgumentExpr,PartialDiagnostic PDiag,SourceLocation Loc,bool IsStringLocation,Range StringRange,ArrayRef<FixItHint> FixIt)3066 void CheckFormatHandler::EmitFormatDiagnostic(Sema &S, bool InFunctionCall,
3067                                               const Expr *ArgumentExpr,
3068                                               PartialDiagnostic PDiag,
3069                                               SourceLocation Loc,
3070                                               bool IsStringLocation,
3071                                               Range StringRange,
3072                                               ArrayRef<FixItHint> FixIt) {
3073   if (InFunctionCall) {
3074     const Sema::SemaDiagnosticBuilder &D = S.Diag(Loc, PDiag);
3075     D << StringRange;
3076     for (ArrayRef<FixItHint>::iterator I = FixIt.begin(), E = FixIt.end();
3077          I != E; ++I) {
3078       D << *I;
3079     }
3080   } else {
3081     S.Diag(IsStringLocation ? ArgumentExpr->getExprLoc() : Loc, PDiag)
3082       << ArgumentExpr->getSourceRange();
3083 
3084     const Sema::SemaDiagnosticBuilder &Note =
3085       S.Diag(IsStringLocation ? Loc : StringRange.getBegin(),
3086              diag::note_format_string_defined);
3087 
3088     Note << StringRange;
3089     for (ArrayRef<FixItHint>::iterator I = FixIt.begin(), E = FixIt.end();
3090          I != E; ++I) {
3091       Note << *I;
3092     }
3093   }
3094 }
3095 
3096 //===--- CHECK: Printf format string checking ------------------------------===//
3097 
3098 namespace {
3099 class CheckPrintfHandler : public CheckFormatHandler {
3100   bool ObjCContext;
3101 public:
CheckPrintfHandler(Sema & s,const StringLiteral * fexpr,const Expr * origFormatExpr,unsigned firstDataArg,unsigned numDataArgs,bool isObjC,const char * beg,bool hasVAListArg,ArrayRef<const Expr * > Args,unsigned formatIdx,bool inFunctionCall,Sema::VariadicCallType CallType,llvm::SmallBitVector & CheckedVarArgs)3102   CheckPrintfHandler(Sema &s, const StringLiteral *fexpr,
3103                      const Expr *origFormatExpr, unsigned firstDataArg,
3104                      unsigned numDataArgs, bool isObjC,
3105                      const char *beg, bool hasVAListArg,
3106                      ArrayRef<const Expr *> Args,
3107                      unsigned formatIdx, bool inFunctionCall,
3108                      Sema::VariadicCallType CallType,
3109                      llvm::SmallBitVector &CheckedVarArgs)
3110     : CheckFormatHandler(s, fexpr, origFormatExpr, firstDataArg,
3111                          numDataArgs, beg, hasVAListArg, Args,
3112                          formatIdx, inFunctionCall, CallType, CheckedVarArgs),
3113       ObjCContext(isObjC)
3114   {}
3115 
3116 
3117   bool HandleInvalidPrintfConversionSpecifier(
3118                                       const analyze_printf::PrintfSpecifier &FS,
3119                                       const char *startSpecifier,
3120                                       unsigned specifierLen) override;
3121 
3122   bool HandlePrintfSpecifier(const analyze_printf::PrintfSpecifier &FS,
3123                              const char *startSpecifier,
3124                              unsigned specifierLen) override;
3125   bool checkFormatExpr(const analyze_printf::PrintfSpecifier &FS,
3126                        const char *StartSpecifier,
3127                        unsigned SpecifierLen,
3128                        const Expr *E);
3129 
3130   bool HandleAmount(const analyze_format_string::OptionalAmount &Amt, unsigned k,
3131                     const char *startSpecifier, unsigned specifierLen);
3132   void HandleInvalidAmount(const analyze_printf::PrintfSpecifier &FS,
3133                            const analyze_printf::OptionalAmount &Amt,
3134                            unsigned type,
3135                            const char *startSpecifier, unsigned specifierLen);
3136   void HandleFlag(const analyze_printf::PrintfSpecifier &FS,
3137                   const analyze_printf::OptionalFlag &flag,
3138                   const char *startSpecifier, unsigned specifierLen);
3139   void HandleIgnoredFlag(const analyze_printf::PrintfSpecifier &FS,
3140                          const analyze_printf::OptionalFlag &ignoredFlag,
3141                          const analyze_printf::OptionalFlag &flag,
3142                          const char *startSpecifier, unsigned specifierLen);
3143   bool checkForCStrMembers(const analyze_printf::ArgType &AT,
3144                            const Expr *E);
3145 
3146 };
3147 }
3148 
HandleInvalidPrintfConversionSpecifier(const analyze_printf::PrintfSpecifier & FS,const char * startSpecifier,unsigned specifierLen)3149 bool CheckPrintfHandler::HandleInvalidPrintfConversionSpecifier(
3150                                       const analyze_printf::PrintfSpecifier &FS,
3151                                       const char *startSpecifier,
3152                                       unsigned specifierLen) {
3153   const analyze_printf::PrintfConversionSpecifier &CS =
3154     FS.getConversionSpecifier();
3155 
3156   return HandleInvalidConversionSpecifier(FS.getArgIndex(),
3157                                           getLocationOfByte(CS.getStart()),
3158                                           startSpecifier, specifierLen,
3159                                           CS.getStart(), CS.getLength());
3160 }
3161 
HandleAmount(const analyze_format_string::OptionalAmount & Amt,unsigned k,const char * startSpecifier,unsigned specifierLen)3162 bool CheckPrintfHandler::HandleAmount(
3163                                const analyze_format_string::OptionalAmount &Amt,
3164                                unsigned k, const char *startSpecifier,
3165                                unsigned specifierLen) {
3166 
3167   if (Amt.hasDataArgument()) {
3168     if (!HasVAListArg) {
3169       unsigned argIndex = Amt.getArgIndex();
3170       if (argIndex >= NumDataArgs) {
3171         EmitFormatDiagnostic(S.PDiag(diag::warn_printf_asterisk_missing_arg)
3172                                << k,
3173                              getLocationOfByte(Amt.getStart()),
3174                              /*IsStringLocation*/true,
3175                              getSpecifierRange(startSpecifier, specifierLen));
3176         // Don't do any more checking.  We will just emit
3177         // spurious errors.
3178         return false;
3179       }
3180 
3181       // Type check the data argument.  It should be an 'int'.
3182       // Although not in conformance with C99, we also allow the argument to be
3183       // an 'unsigned int' as that is a reasonably safe case.  GCC also
3184       // doesn't emit a warning for that case.
3185       CoveredArgs.set(argIndex);
3186       const Expr *Arg = getDataArg(argIndex);
3187       if (!Arg)
3188         return false;
3189 
3190       QualType T = Arg->getType();
3191 
3192       const analyze_printf::ArgType &AT = Amt.getArgType(S.Context);
3193       assert(AT.isValid());
3194 
3195       if (!AT.matchesType(S.Context, T)) {
3196         EmitFormatDiagnostic(S.PDiag(diag::warn_printf_asterisk_wrong_type)
3197                                << k << AT.getRepresentativeTypeName(S.Context)
3198                                << T << Arg->getSourceRange(),
3199                              getLocationOfByte(Amt.getStart()),
3200                              /*IsStringLocation*/true,
3201                              getSpecifierRange(startSpecifier, specifierLen));
3202         // Don't do any more checking.  We will just emit
3203         // spurious errors.
3204         return false;
3205       }
3206     }
3207   }
3208   return true;
3209 }
3210 
HandleInvalidAmount(const analyze_printf::PrintfSpecifier & FS,const analyze_printf::OptionalAmount & Amt,unsigned type,const char * startSpecifier,unsigned specifierLen)3211 void CheckPrintfHandler::HandleInvalidAmount(
3212                                       const analyze_printf::PrintfSpecifier &FS,
3213                                       const analyze_printf::OptionalAmount &Amt,
3214                                       unsigned type,
3215                                       const char *startSpecifier,
3216                                       unsigned specifierLen) {
3217   const analyze_printf::PrintfConversionSpecifier &CS =
3218     FS.getConversionSpecifier();
3219 
3220   FixItHint fixit =
3221     Amt.getHowSpecified() == analyze_printf::OptionalAmount::Constant
3222       ? FixItHint::CreateRemoval(getSpecifierRange(Amt.getStart(),
3223                                  Amt.getConstantLength()))
3224       : FixItHint();
3225 
3226   EmitFormatDiagnostic(S.PDiag(diag::warn_printf_nonsensical_optional_amount)
3227                          << type << CS.toString(),
3228                        getLocationOfByte(Amt.getStart()),
3229                        /*IsStringLocation*/true,
3230                        getSpecifierRange(startSpecifier, specifierLen),
3231                        fixit);
3232 }
3233 
HandleFlag(const analyze_printf::PrintfSpecifier & FS,const analyze_printf::OptionalFlag & flag,const char * startSpecifier,unsigned specifierLen)3234 void CheckPrintfHandler::HandleFlag(const analyze_printf::PrintfSpecifier &FS,
3235                                     const analyze_printf::OptionalFlag &flag,
3236                                     const char *startSpecifier,
3237                                     unsigned specifierLen) {
3238   // Warn about pointless flag with a fixit removal.
3239   const analyze_printf::PrintfConversionSpecifier &CS =
3240     FS.getConversionSpecifier();
3241   EmitFormatDiagnostic(S.PDiag(diag::warn_printf_nonsensical_flag)
3242                          << flag.toString() << CS.toString(),
3243                        getLocationOfByte(flag.getPosition()),
3244                        /*IsStringLocation*/true,
3245                        getSpecifierRange(startSpecifier, specifierLen),
3246                        FixItHint::CreateRemoval(
3247                          getSpecifierRange(flag.getPosition(), 1)));
3248 }
3249 
HandleIgnoredFlag(const analyze_printf::PrintfSpecifier & FS,const analyze_printf::OptionalFlag & ignoredFlag,const analyze_printf::OptionalFlag & flag,const char * startSpecifier,unsigned specifierLen)3250 void CheckPrintfHandler::HandleIgnoredFlag(
3251                                 const analyze_printf::PrintfSpecifier &FS,
3252                                 const analyze_printf::OptionalFlag &ignoredFlag,
3253                                 const analyze_printf::OptionalFlag &flag,
3254                                 const char *startSpecifier,
3255                                 unsigned specifierLen) {
3256   // Warn about ignored flag with a fixit removal.
3257   EmitFormatDiagnostic(S.PDiag(diag::warn_printf_ignored_flag)
3258                          << ignoredFlag.toString() << flag.toString(),
3259                        getLocationOfByte(ignoredFlag.getPosition()),
3260                        /*IsStringLocation*/true,
3261                        getSpecifierRange(startSpecifier, specifierLen),
3262                        FixItHint::CreateRemoval(
3263                          getSpecifierRange(ignoredFlag.getPosition(), 1)));
3264 }
3265 
3266 // Determines if the specified is a C++ class or struct containing
3267 // a member with the specified name and kind (e.g. a CXXMethodDecl named
3268 // "c_str()").
3269 template<typename MemberKind>
3270 static llvm::SmallPtrSet<MemberKind*, 1>
CXXRecordMembersNamed(StringRef Name,Sema & S,QualType Ty)3271 CXXRecordMembersNamed(StringRef Name, Sema &S, QualType Ty) {
3272   const RecordType *RT = Ty->getAs<RecordType>();
3273   llvm::SmallPtrSet<MemberKind*, 1> Results;
3274 
3275   if (!RT)
3276     return Results;
3277   const CXXRecordDecl *RD = dyn_cast<CXXRecordDecl>(RT->getDecl());
3278   if (!RD || !RD->getDefinition())
3279     return Results;
3280 
3281   LookupResult R(S, &S.Context.Idents.get(Name), SourceLocation(),
3282                  Sema::LookupMemberName);
3283   R.suppressDiagnostics();
3284 
3285   // We just need to include all members of the right kind turned up by the
3286   // filter, at this point.
3287   if (S.LookupQualifiedName(R, RT->getDecl()))
3288     for (LookupResult::iterator I = R.begin(), E = R.end(); I != E; ++I) {
3289       NamedDecl *decl = (*I)->getUnderlyingDecl();
3290       if (MemberKind *FK = dyn_cast<MemberKind>(decl))
3291         Results.insert(FK);
3292     }
3293   return Results;
3294 }
3295 
3296 /// Check if we could call '.c_str()' on an object.
3297 ///
3298 /// FIXME: This returns the wrong results in some cases (if cv-qualifiers don't
3299 /// allow the call, or if it would be ambiguous).
hasCStrMethod(const Expr * E)3300 bool Sema::hasCStrMethod(const Expr *E) {
3301   typedef llvm::SmallPtrSet<CXXMethodDecl*, 1> MethodSet;
3302   MethodSet Results =
3303       CXXRecordMembersNamed<CXXMethodDecl>("c_str", *this, E->getType());
3304   for (MethodSet::iterator MI = Results.begin(), ME = Results.end();
3305        MI != ME; ++MI)
3306     if ((*MI)->getMinRequiredArguments() == 0)
3307       return true;
3308   return false;
3309 }
3310 
3311 // Check if a (w)string was passed when a (w)char* was needed, and offer a
3312 // better diagnostic if so. AT is assumed to be valid.
3313 // Returns true when a c_str() conversion method is found.
checkForCStrMembers(const analyze_printf::ArgType & AT,const Expr * E)3314 bool CheckPrintfHandler::checkForCStrMembers(
3315     const analyze_printf::ArgType &AT, const Expr *E) {
3316   typedef llvm::SmallPtrSet<CXXMethodDecl*, 1> MethodSet;
3317 
3318   MethodSet Results =
3319       CXXRecordMembersNamed<CXXMethodDecl>("c_str", S, E->getType());
3320 
3321   for (MethodSet::iterator MI = Results.begin(), ME = Results.end();
3322        MI != ME; ++MI) {
3323     const CXXMethodDecl *Method = *MI;
3324     if (Method->getMinRequiredArguments() == 0 &&
3325         AT.matchesType(S.Context, Method->getReturnType())) {
3326       // FIXME: Suggest parens if the expression needs them.
3327       SourceLocation EndLoc = S.getLocForEndOfToken(E->getLocEnd());
3328       S.Diag(E->getLocStart(), diag::note_printf_c_str)
3329           << "c_str()"
3330           << FixItHint::CreateInsertion(EndLoc, ".c_str()");
3331       return true;
3332     }
3333   }
3334 
3335   return false;
3336 }
3337 
3338 bool
HandlePrintfSpecifier(const analyze_printf::PrintfSpecifier & FS,const char * startSpecifier,unsigned specifierLen)3339 CheckPrintfHandler::HandlePrintfSpecifier(const analyze_printf::PrintfSpecifier
3340                                             &FS,
3341                                           const char *startSpecifier,
3342                                           unsigned specifierLen) {
3343 
3344   using namespace analyze_format_string;
3345   using namespace analyze_printf;
3346   const PrintfConversionSpecifier &CS = FS.getConversionSpecifier();
3347 
3348   if (FS.consumesDataArgument()) {
3349     if (atFirstArg) {
3350         atFirstArg = false;
3351         usesPositionalArgs = FS.usesPositionalArg();
3352     }
3353     else if (usesPositionalArgs != FS.usesPositionalArg()) {
3354       HandlePositionalNonpositionalArgs(getLocationOfByte(CS.getStart()),
3355                                         startSpecifier, specifierLen);
3356       return false;
3357     }
3358   }
3359 
3360   // First check if the field width, precision, and conversion specifier
3361   // have matching data arguments.
3362   if (!HandleAmount(FS.getFieldWidth(), /* field width */ 0,
3363                     startSpecifier, specifierLen)) {
3364     return false;
3365   }
3366 
3367   if (!HandleAmount(FS.getPrecision(), /* precision */ 1,
3368                     startSpecifier, specifierLen)) {
3369     return false;
3370   }
3371 
3372   if (!CS.consumesDataArgument()) {
3373     // FIXME: Technically specifying a precision or field width here
3374     // makes no sense.  Worth issuing a warning at some point.
3375     return true;
3376   }
3377 
3378   // Consume the argument.
3379   unsigned argIndex = FS.getArgIndex();
3380   if (argIndex < NumDataArgs) {
3381     // The check to see if the argIndex is valid will come later.
3382     // We set the bit here because we may exit early from this
3383     // function if we encounter some other error.
3384     CoveredArgs.set(argIndex);
3385   }
3386 
3387   // Check for using an Objective-C specific conversion specifier
3388   // in a non-ObjC literal.
3389   if (!ObjCContext && CS.isObjCArg()) {
3390     return HandleInvalidPrintfConversionSpecifier(FS, startSpecifier,
3391                                                   specifierLen);
3392   }
3393 
3394   // Check for invalid use of field width
3395   if (!FS.hasValidFieldWidth()) {
3396     HandleInvalidAmount(FS, FS.getFieldWidth(), /* field width */ 0,
3397         startSpecifier, specifierLen);
3398   }
3399 
3400   // Check for invalid use of precision
3401   if (!FS.hasValidPrecision()) {
3402     HandleInvalidAmount(FS, FS.getPrecision(), /* precision */ 1,
3403         startSpecifier, specifierLen);
3404   }
3405 
3406   // Check each flag does not conflict with any other component.
3407   if (!FS.hasValidThousandsGroupingPrefix())
3408     HandleFlag(FS, FS.hasThousandsGrouping(), startSpecifier, specifierLen);
3409   if (!FS.hasValidLeadingZeros())
3410     HandleFlag(FS, FS.hasLeadingZeros(), startSpecifier, specifierLen);
3411   if (!FS.hasValidPlusPrefix())
3412     HandleFlag(FS, FS.hasPlusPrefix(), startSpecifier, specifierLen);
3413   if (!FS.hasValidSpacePrefix())
3414     HandleFlag(FS, FS.hasSpacePrefix(), startSpecifier, specifierLen);
3415   if (!FS.hasValidAlternativeForm())
3416     HandleFlag(FS, FS.hasAlternativeForm(), startSpecifier, specifierLen);
3417   if (!FS.hasValidLeftJustified())
3418     HandleFlag(FS, FS.isLeftJustified(), startSpecifier, specifierLen);
3419 
3420   // Check that flags are not ignored by another flag
3421   if (FS.hasSpacePrefix() && FS.hasPlusPrefix()) // ' ' ignored by '+'
3422     HandleIgnoredFlag(FS, FS.hasSpacePrefix(), FS.hasPlusPrefix(),
3423         startSpecifier, specifierLen);
3424   if (FS.hasLeadingZeros() && FS.isLeftJustified()) // '0' ignored by '-'
3425     HandleIgnoredFlag(FS, FS.hasLeadingZeros(), FS.isLeftJustified(),
3426             startSpecifier, specifierLen);
3427 
3428   // Check the length modifier is valid with the given conversion specifier.
3429   if (!FS.hasValidLengthModifier(S.getASTContext().getTargetInfo()))
3430     HandleInvalidLengthModifier(FS, CS, startSpecifier, specifierLen,
3431                                 diag::warn_format_nonsensical_length);
3432   else if (!FS.hasStandardLengthModifier())
3433     HandleNonStandardLengthModifier(FS, startSpecifier, specifierLen);
3434   else if (!FS.hasStandardLengthConversionCombination())
3435     HandleInvalidLengthModifier(FS, CS, startSpecifier, specifierLen,
3436                                 diag::warn_format_non_standard_conversion_spec);
3437 
3438   if (!FS.hasStandardConversionSpecifier(S.getLangOpts()))
3439     HandleNonStandardConversionSpecifier(CS, startSpecifier, specifierLen);
3440 
3441   // The remaining checks depend on the data arguments.
3442   if (HasVAListArg)
3443     return true;
3444 
3445   if (!CheckNumArgs(FS, CS, startSpecifier, specifierLen, argIndex))
3446     return false;
3447 
3448   const Expr *Arg = getDataArg(argIndex);
3449   if (!Arg)
3450     return true;
3451 
3452   return checkFormatExpr(FS, startSpecifier, specifierLen, Arg);
3453 }
3454 
requiresParensToAddCast(const Expr * E)3455 static bool requiresParensToAddCast(const Expr *E) {
3456   // FIXME: We should have a general way to reason about operator
3457   // precedence and whether parens are actually needed here.
3458   // Take care of a few common cases where they aren't.
3459   const Expr *Inside = E->IgnoreImpCasts();
3460   if (const PseudoObjectExpr *POE = dyn_cast<PseudoObjectExpr>(Inside))
3461     Inside = POE->getSyntacticForm()->IgnoreImpCasts();
3462 
3463   switch (Inside->getStmtClass()) {
3464   case Stmt::ArraySubscriptExprClass:
3465   case Stmt::CallExprClass:
3466   case Stmt::CharacterLiteralClass:
3467   case Stmt::CXXBoolLiteralExprClass:
3468   case Stmt::DeclRefExprClass:
3469   case Stmt::FloatingLiteralClass:
3470   case Stmt::IntegerLiteralClass:
3471   case Stmt::MemberExprClass:
3472   case Stmt::ObjCArrayLiteralClass:
3473   case Stmt::ObjCBoolLiteralExprClass:
3474   case Stmt::ObjCBoxedExprClass:
3475   case Stmt::ObjCDictionaryLiteralClass:
3476   case Stmt::ObjCEncodeExprClass:
3477   case Stmt::ObjCIvarRefExprClass:
3478   case Stmt::ObjCMessageExprClass:
3479   case Stmt::ObjCPropertyRefExprClass:
3480   case Stmt::ObjCStringLiteralClass:
3481   case Stmt::ObjCSubscriptRefExprClass:
3482   case Stmt::ParenExprClass:
3483   case Stmt::StringLiteralClass:
3484   case Stmt::UnaryOperatorClass:
3485     return false;
3486   default:
3487     return true;
3488   }
3489 }
3490 
3491 static std::pair<QualType, StringRef>
shouldNotPrintDirectly(const ASTContext & Context,QualType IntendedTy,const Expr * E)3492 shouldNotPrintDirectly(const ASTContext &Context,
3493                        QualType IntendedTy,
3494                        const Expr *E) {
3495   // Use a 'while' to peel off layers of typedefs.
3496   QualType TyTy = IntendedTy;
3497   while (const TypedefType *UserTy = TyTy->getAs<TypedefType>()) {
3498     StringRef Name = UserTy->getDecl()->getName();
3499     QualType CastTy = llvm::StringSwitch<QualType>(Name)
3500       .Case("NSInteger", Context.LongTy)
3501       .Case("NSUInteger", Context.UnsignedLongTy)
3502       .Case("SInt32", Context.IntTy)
3503       .Case("UInt32", Context.UnsignedIntTy)
3504       .Default(QualType());
3505 
3506     if (!CastTy.isNull())
3507       return std::make_pair(CastTy, Name);
3508 
3509     TyTy = UserTy->desugar();
3510   }
3511 
3512   // Strip parens if necessary.
3513   if (const ParenExpr *PE = dyn_cast<ParenExpr>(E))
3514     return shouldNotPrintDirectly(Context,
3515                                   PE->getSubExpr()->getType(),
3516                                   PE->getSubExpr());
3517 
3518   // If this is a conditional expression, then its result type is constructed
3519   // via usual arithmetic conversions and thus there might be no necessary
3520   // typedef sugar there.  Recurse to operands to check for NSInteger &
3521   // Co. usage condition.
3522   if (const ConditionalOperator *CO = dyn_cast<ConditionalOperator>(E)) {
3523     QualType TrueTy, FalseTy;
3524     StringRef TrueName, FalseName;
3525 
3526     std::tie(TrueTy, TrueName) =
3527       shouldNotPrintDirectly(Context,
3528                              CO->getTrueExpr()->getType(),
3529                              CO->getTrueExpr());
3530     std::tie(FalseTy, FalseName) =
3531       shouldNotPrintDirectly(Context,
3532                              CO->getFalseExpr()->getType(),
3533                              CO->getFalseExpr());
3534 
3535     if (TrueTy == FalseTy)
3536       return std::make_pair(TrueTy, TrueName);
3537     else if (TrueTy.isNull())
3538       return std::make_pair(FalseTy, FalseName);
3539     else if (FalseTy.isNull())
3540       return std::make_pair(TrueTy, TrueName);
3541   }
3542 
3543   return std::make_pair(QualType(), StringRef());
3544 }
3545 
3546 bool
checkFormatExpr(const analyze_printf::PrintfSpecifier & FS,const char * StartSpecifier,unsigned SpecifierLen,const Expr * E)3547 CheckPrintfHandler::checkFormatExpr(const analyze_printf::PrintfSpecifier &FS,
3548                                     const char *StartSpecifier,
3549                                     unsigned SpecifierLen,
3550                                     const Expr *E) {
3551   using namespace analyze_format_string;
3552   using namespace analyze_printf;
3553   // Now type check the data expression that matches the
3554   // format specifier.
3555   const analyze_printf::ArgType &AT = FS.getArgType(S.Context,
3556                                                     ObjCContext);
3557   if (!AT.isValid())
3558     return true;
3559 
3560   QualType ExprTy = E->getType();
3561   while (const TypeOfExprType *TET = dyn_cast<TypeOfExprType>(ExprTy)) {
3562     ExprTy = TET->getUnderlyingExpr()->getType();
3563   }
3564 
3565   if (AT.matchesType(S.Context, ExprTy))
3566     return true;
3567 
3568   // Look through argument promotions for our error message's reported type.
3569   // This includes the integral and floating promotions, but excludes array
3570   // and function pointer decay; seeing that an argument intended to be a
3571   // string has type 'char [6]' is probably more confusing than 'char *'.
3572   if (const ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(E)) {
3573     if (ICE->getCastKind() == CK_IntegralCast ||
3574         ICE->getCastKind() == CK_FloatingCast) {
3575       E = ICE->getSubExpr();
3576       ExprTy = E->getType();
3577 
3578       // Check if we didn't match because of an implicit cast from a 'char'
3579       // or 'short' to an 'int'.  This is done because printf is a varargs
3580       // function.
3581       if (ICE->getType() == S.Context.IntTy ||
3582           ICE->getType() == S.Context.UnsignedIntTy) {
3583         // All further checking is done on the subexpression.
3584         if (AT.matchesType(S.Context, ExprTy))
3585           return true;
3586       }
3587     }
3588   } else if (const CharacterLiteral *CL = dyn_cast<CharacterLiteral>(E)) {
3589     // Special case for 'a', which has type 'int' in C.
3590     // Note, however, that we do /not/ want to treat multibyte constants like
3591     // 'MooV' as characters! This form is deprecated but still exists.
3592     if (ExprTy == S.Context.IntTy)
3593       if (llvm::isUIntN(S.Context.getCharWidth(), CL->getValue()))
3594         ExprTy = S.Context.CharTy;
3595   }
3596 
3597   // Look through enums to their underlying type.
3598   bool IsEnum = false;
3599   if (auto EnumTy = ExprTy->getAs<EnumType>()) {
3600     ExprTy = EnumTy->getDecl()->getIntegerType();
3601     IsEnum = true;
3602   }
3603 
3604   // %C in an Objective-C context prints a unichar, not a wchar_t.
3605   // If the argument is an integer of some kind, believe the %C and suggest
3606   // a cast instead of changing the conversion specifier.
3607   QualType IntendedTy = ExprTy;
3608   if (ObjCContext &&
3609       FS.getConversionSpecifier().getKind() == ConversionSpecifier::CArg) {
3610     if (ExprTy->isIntegralOrUnscopedEnumerationType() &&
3611         !ExprTy->isCharType()) {
3612       // 'unichar' is defined as a typedef of unsigned short, but we should
3613       // prefer using the typedef if it is visible.
3614       IntendedTy = S.Context.UnsignedShortTy;
3615 
3616       // While we are here, check if the value is an IntegerLiteral that happens
3617       // to be within the valid range.
3618       if (const IntegerLiteral *IL = dyn_cast<IntegerLiteral>(E)) {
3619         const llvm::APInt &V = IL->getValue();
3620         if (V.getActiveBits() <= S.Context.getTypeSize(IntendedTy))
3621           return true;
3622       }
3623 
3624       LookupResult Result(S, &S.Context.Idents.get("unichar"), E->getLocStart(),
3625                           Sema::LookupOrdinaryName);
3626       if (S.LookupName(Result, S.getCurScope())) {
3627         NamedDecl *ND = Result.getFoundDecl();
3628         if (TypedefNameDecl *TD = dyn_cast<TypedefNameDecl>(ND))
3629           if (TD->getUnderlyingType() == IntendedTy)
3630             IntendedTy = S.Context.getTypedefType(TD);
3631       }
3632     }
3633   }
3634 
3635   // Special-case some of Darwin's platform-independence types by suggesting
3636   // casts to primitive types that are known to be large enough.
3637   bool ShouldNotPrintDirectly = false; StringRef CastTyName;
3638   if (S.Context.getTargetInfo().getTriple().isOSDarwin()) {
3639     QualType CastTy;
3640     std::tie(CastTy, CastTyName) = shouldNotPrintDirectly(S.Context, IntendedTy, E);
3641     if (!CastTy.isNull()) {
3642       IntendedTy = CastTy;
3643       ShouldNotPrintDirectly = true;
3644     }
3645   }
3646 
3647   // We may be able to offer a FixItHint if it is a supported type.
3648   PrintfSpecifier fixedFS = FS;
3649   bool success = fixedFS.fixType(IntendedTy, S.getLangOpts(),
3650                                  S.Context, ObjCContext);
3651 
3652   if (success) {
3653     // Get the fix string from the fixed format specifier
3654     SmallString<16> buf;
3655     llvm::raw_svector_ostream os(buf);
3656     fixedFS.toString(os);
3657 
3658     CharSourceRange SpecRange = getSpecifierRange(StartSpecifier, SpecifierLen);
3659 
3660     if (IntendedTy == ExprTy && !ShouldNotPrintDirectly) {
3661       // In this case, the specifier is wrong and should be changed to match
3662       // the argument.
3663       EmitFormatDiagnostic(
3664         S.PDiag(diag::warn_format_conversion_argument_type_mismatch)
3665           << AT.getRepresentativeTypeName(S.Context) << IntendedTy << IsEnum
3666           << E->getSourceRange(),
3667         E->getLocStart(),
3668         /*IsStringLocation*/false,
3669         SpecRange,
3670         FixItHint::CreateReplacement(SpecRange, os.str()));
3671 
3672     } else {
3673       // The canonical type for formatting this value is different from the
3674       // actual type of the expression. (This occurs, for example, with Darwin's
3675       // NSInteger on 32-bit platforms, where it is typedef'd as 'int', but
3676       // should be printed as 'long' for 64-bit compatibility.)
3677       // Rather than emitting a normal format/argument mismatch, we want to
3678       // add a cast to the recommended type (and correct the format string
3679       // if necessary).
3680       SmallString<16> CastBuf;
3681       llvm::raw_svector_ostream CastFix(CastBuf);
3682       CastFix << "(";
3683       IntendedTy.print(CastFix, S.Context.getPrintingPolicy());
3684       CastFix << ")";
3685 
3686       SmallVector<FixItHint,4> Hints;
3687       if (!AT.matchesType(S.Context, IntendedTy))
3688         Hints.push_back(FixItHint::CreateReplacement(SpecRange, os.str()));
3689 
3690       if (const CStyleCastExpr *CCast = dyn_cast<CStyleCastExpr>(E)) {
3691         // If there's already a cast present, just replace it.
3692         SourceRange CastRange(CCast->getLParenLoc(), CCast->getRParenLoc());
3693         Hints.push_back(FixItHint::CreateReplacement(CastRange, CastFix.str()));
3694 
3695       } else if (!requiresParensToAddCast(E)) {
3696         // If the expression has high enough precedence,
3697         // just write the C-style cast.
3698         Hints.push_back(FixItHint::CreateInsertion(E->getLocStart(),
3699                                                    CastFix.str()));
3700       } else {
3701         // Otherwise, add parens around the expression as well as the cast.
3702         CastFix << "(";
3703         Hints.push_back(FixItHint::CreateInsertion(E->getLocStart(),
3704                                                    CastFix.str()));
3705 
3706         SourceLocation After = S.getLocForEndOfToken(E->getLocEnd());
3707         Hints.push_back(FixItHint::CreateInsertion(After, ")"));
3708       }
3709 
3710       if (ShouldNotPrintDirectly) {
3711         // The expression has a type that should not be printed directly.
3712         // We extract the name from the typedef because we don't want to show
3713         // the underlying type in the diagnostic.
3714         StringRef Name;
3715         if (const TypedefType *TypedefTy = dyn_cast<TypedefType>(ExprTy))
3716           Name = TypedefTy->getDecl()->getName();
3717         else
3718           Name = CastTyName;
3719         EmitFormatDiagnostic(S.PDiag(diag::warn_format_argument_needs_cast)
3720                                << Name << IntendedTy << IsEnum
3721                                << E->getSourceRange(),
3722                              E->getLocStart(), /*IsStringLocation=*/false,
3723                              SpecRange, Hints);
3724       } else {
3725         // In this case, the expression could be printed using a different
3726         // specifier, but we've decided that the specifier is probably correct
3727         // and we should cast instead. Just use the normal warning message.
3728         EmitFormatDiagnostic(
3729           S.PDiag(diag::warn_format_conversion_argument_type_mismatch)
3730             << AT.getRepresentativeTypeName(S.Context) << ExprTy << IsEnum
3731             << E->getSourceRange(),
3732           E->getLocStart(), /*IsStringLocation*/false,
3733           SpecRange, Hints);
3734       }
3735     }
3736   } else {
3737     const CharSourceRange &CSR = getSpecifierRange(StartSpecifier,
3738                                                    SpecifierLen);
3739     // Since the warning for passing non-POD types to variadic functions
3740     // was deferred until now, we emit a warning for non-POD
3741     // arguments here.
3742     switch (S.isValidVarArgType(ExprTy)) {
3743     case Sema::VAK_Valid:
3744     case Sema::VAK_ValidInCXX11:
3745       EmitFormatDiagnostic(
3746         S.PDiag(diag::warn_format_conversion_argument_type_mismatch)
3747           << AT.getRepresentativeTypeName(S.Context) << ExprTy << IsEnum
3748           << CSR
3749           << E->getSourceRange(),
3750         E->getLocStart(), /*IsStringLocation*/false, CSR);
3751       break;
3752 
3753     case Sema::VAK_Undefined:
3754     case Sema::VAK_MSVCUndefined:
3755       EmitFormatDiagnostic(
3756         S.PDiag(diag::warn_non_pod_vararg_with_format_string)
3757           << S.getLangOpts().CPlusPlus11
3758           << ExprTy
3759           << CallType
3760           << AT.getRepresentativeTypeName(S.Context)
3761           << CSR
3762           << E->getSourceRange(),
3763         E->getLocStart(), /*IsStringLocation*/false, CSR);
3764       checkForCStrMembers(AT, E);
3765       break;
3766 
3767     case Sema::VAK_Invalid:
3768       if (ExprTy->isObjCObjectType())
3769         EmitFormatDiagnostic(
3770           S.PDiag(diag::err_cannot_pass_objc_interface_to_vararg_format)
3771             << S.getLangOpts().CPlusPlus11
3772             << ExprTy
3773             << CallType
3774             << AT.getRepresentativeTypeName(S.Context)
3775             << CSR
3776             << E->getSourceRange(),
3777           E->getLocStart(), /*IsStringLocation*/false, CSR);
3778       else
3779         // FIXME: If this is an initializer list, suggest removing the braces
3780         // or inserting a cast to the target type.
3781         S.Diag(E->getLocStart(), diag::err_cannot_pass_to_vararg_format)
3782           << isa<InitListExpr>(E) << ExprTy << CallType
3783           << AT.getRepresentativeTypeName(S.Context)
3784           << E->getSourceRange();
3785       break;
3786     }
3787 
3788     assert(FirstDataArg + FS.getArgIndex() < CheckedVarArgs.size() &&
3789            "format string specifier index out of range");
3790     CheckedVarArgs[FirstDataArg + FS.getArgIndex()] = true;
3791   }
3792 
3793   return true;
3794 }
3795 
3796 //===--- CHECK: Scanf format string checking ------------------------------===//
3797 
3798 namespace {
3799 class CheckScanfHandler : public CheckFormatHandler {
3800 public:
CheckScanfHandler(Sema & s,const StringLiteral * fexpr,const Expr * origFormatExpr,unsigned firstDataArg,unsigned numDataArgs,const char * beg,bool hasVAListArg,ArrayRef<const Expr * > Args,unsigned formatIdx,bool inFunctionCall,Sema::VariadicCallType CallType,llvm::SmallBitVector & CheckedVarArgs)3801   CheckScanfHandler(Sema &s, const StringLiteral *fexpr,
3802                     const Expr *origFormatExpr, unsigned firstDataArg,
3803                     unsigned numDataArgs, const char *beg, bool hasVAListArg,
3804                     ArrayRef<const Expr *> Args,
3805                     unsigned formatIdx, bool inFunctionCall,
3806                     Sema::VariadicCallType CallType,
3807                     llvm::SmallBitVector &CheckedVarArgs)
3808     : CheckFormatHandler(s, fexpr, origFormatExpr, firstDataArg,
3809                          numDataArgs, beg, hasVAListArg,
3810                          Args, formatIdx, inFunctionCall, CallType,
3811                          CheckedVarArgs)
3812   {}
3813 
3814   bool HandleScanfSpecifier(const analyze_scanf::ScanfSpecifier &FS,
3815                             const char *startSpecifier,
3816                             unsigned specifierLen) override;
3817 
3818   bool HandleInvalidScanfConversionSpecifier(
3819           const analyze_scanf::ScanfSpecifier &FS,
3820           const char *startSpecifier,
3821           unsigned specifierLen) override;
3822 
3823   void HandleIncompleteScanList(const char *start, const char *end) override;
3824 };
3825 }
3826 
HandleIncompleteScanList(const char * start,const char * end)3827 void CheckScanfHandler::HandleIncompleteScanList(const char *start,
3828                                                  const char *end) {
3829   EmitFormatDiagnostic(S.PDiag(diag::warn_scanf_scanlist_incomplete),
3830                        getLocationOfByte(end), /*IsStringLocation*/true,
3831                        getSpecifierRange(start, end - start));
3832 }
3833 
HandleInvalidScanfConversionSpecifier(const analyze_scanf::ScanfSpecifier & FS,const char * startSpecifier,unsigned specifierLen)3834 bool CheckScanfHandler::HandleInvalidScanfConversionSpecifier(
3835                                         const analyze_scanf::ScanfSpecifier &FS,
3836                                         const char *startSpecifier,
3837                                         unsigned specifierLen) {
3838 
3839   const analyze_scanf::ScanfConversionSpecifier &CS =
3840     FS.getConversionSpecifier();
3841 
3842   return HandleInvalidConversionSpecifier(FS.getArgIndex(),
3843                                           getLocationOfByte(CS.getStart()),
3844                                           startSpecifier, specifierLen,
3845                                           CS.getStart(), CS.getLength());
3846 }
3847 
HandleScanfSpecifier(const analyze_scanf::ScanfSpecifier & FS,const char * startSpecifier,unsigned specifierLen)3848 bool CheckScanfHandler::HandleScanfSpecifier(
3849                                        const analyze_scanf::ScanfSpecifier &FS,
3850                                        const char *startSpecifier,
3851                                        unsigned specifierLen) {
3852 
3853   using namespace analyze_scanf;
3854   using namespace analyze_format_string;
3855 
3856   const ScanfConversionSpecifier &CS = FS.getConversionSpecifier();
3857 
3858   // Handle case where '%' and '*' don't consume an argument.  These shouldn't
3859   // be used to decide if we are using positional arguments consistently.
3860   if (FS.consumesDataArgument()) {
3861     if (atFirstArg) {
3862       atFirstArg = false;
3863       usesPositionalArgs = FS.usesPositionalArg();
3864     }
3865     else if (usesPositionalArgs != FS.usesPositionalArg()) {
3866       HandlePositionalNonpositionalArgs(getLocationOfByte(CS.getStart()),
3867                                         startSpecifier, specifierLen);
3868       return false;
3869     }
3870   }
3871 
3872   // Check if the field with is non-zero.
3873   const OptionalAmount &Amt = FS.getFieldWidth();
3874   if (Amt.getHowSpecified() == OptionalAmount::Constant) {
3875     if (Amt.getConstantAmount() == 0) {
3876       const CharSourceRange &R = getSpecifierRange(Amt.getStart(),
3877                                                    Amt.getConstantLength());
3878       EmitFormatDiagnostic(S.PDiag(diag::warn_scanf_nonzero_width),
3879                            getLocationOfByte(Amt.getStart()),
3880                            /*IsStringLocation*/true, R,
3881                            FixItHint::CreateRemoval(R));
3882     }
3883   }
3884 
3885   if (!FS.consumesDataArgument()) {
3886     // FIXME: Technically specifying a precision or field width here
3887     // makes no sense.  Worth issuing a warning at some point.
3888     return true;
3889   }
3890 
3891   // Consume the argument.
3892   unsigned argIndex = FS.getArgIndex();
3893   if (argIndex < NumDataArgs) {
3894       // The check to see if the argIndex is valid will come later.
3895       // We set the bit here because we may exit early from this
3896       // function if we encounter some other error.
3897     CoveredArgs.set(argIndex);
3898   }
3899 
3900   // Check the length modifier is valid with the given conversion specifier.
3901   if (!FS.hasValidLengthModifier(S.getASTContext().getTargetInfo()))
3902     HandleInvalidLengthModifier(FS, CS, startSpecifier, specifierLen,
3903                                 diag::warn_format_nonsensical_length);
3904   else if (!FS.hasStandardLengthModifier())
3905     HandleNonStandardLengthModifier(FS, startSpecifier, specifierLen);
3906   else if (!FS.hasStandardLengthConversionCombination())
3907     HandleInvalidLengthModifier(FS, CS, startSpecifier, specifierLen,
3908                                 diag::warn_format_non_standard_conversion_spec);
3909 
3910   if (!FS.hasStandardConversionSpecifier(S.getLangOpts()))
3911     HandleNonStandardConversionSpecifier(CS, startSpecifier, specifierLen);
3912 
3913   // The remaining checks depend on the data arguments.
3914   if (HasVAListArg)
3915     return true;
3916 
3917   if (!CheckNumArgs(FS, CS, startSpecifier, specifierLen, argIndex))
3918     return false;
3919 
3920   // Check that the argument type matches the format specifier.
3921   const Expr *Ex = getDataArg(argIndex);
3922   if (!Ex)
3923     return true;
3924 
3925   const analyze_format_string::ArgType &AT = FS.getArgType(S.Context);
3926   if (AT.isValid() && !AT.matchesType(S.Context, Ex->getType())) {
3927     ScanfSpecifier fixedFS = FS;
3928     bool success = fixedFS.fixType(Ex->getType(),
3929                                    Ex->IgnoreImpCasts()->getType(),
3930                                    S.getLangOpts(), S.Context);
3931 
3932     if (success) {
3933       // Get the fix string from the fixed format specifier.
3934       SmallString<128> buf;
3935       llvm::raw_svector_ostream os(buf);
3936       fixedFS.toString(os);
3937 
3938       EmitFormatDiagnostic(
3939         S.PDiag(diag::warn_format_conversion_argument_type_mismatch)
3940           << AT.getRepresentativeTypeName(S.Context) << Ex->getType() << false
3941           << Ex->getSourceRange(),
3942         Ex->getLocStart(),
3943         /*IsStringLocation*/false,
3944         getSpecifierRange(startSpecifier, specifierLen),
3945         FixItHint::CreateReplacement(
3946           getSpecifierRange(startSpecifier, specifierLen),
3947           os.str()));
3948     } else {
3949       EmitFormatDiagnostic(
3950         S.PDiag(diag::warn_format_conversion_argument_type_mismatch)
3951           << AT.getRepresentativeTypeName(S.Context) << Ex->getType() << false
3952           << Ex->getSourceRange(),
3953         Ex->getLocStart(),
3954         /*IsStringLocation*/false,
3955         getSpecifierRange(startSpecifier, specifierLen));
3956     }
3957   }
3958 
3959   return true;
3960 }
3961 
CheckFormatString(const StringLiteral * FExpr,const Expr * OrigFormatExpr,ArrayRef<const Expr * > Args,bool HasVAListArg,unsigned format_idx,unsigned firstDataArg,FormatStringType Type,bool inFunctionCall,VariadicCallType CallType,llvm::SmallBitVector & CheckedVarArgs)3962 void Sema::CheckFormatString(const StringLiteral *FExpr,
3963                              const Expr *OrigFormatExpr,
3964                              ArrayRef<const Expr *> Args,
3965                              bool HasVAListArg, unsigned format_idx,
3966                              unsigned firstDataArg, FormatStringType Type,
3967                              bool inFunctionCall, VariadicCallType CallType,
3968                              llvm::SmallBitVector &CheckedVarArgs) {
3969 
3970   // CHECK: is the format string a wide literal?
3971   if (!FExpr->isAscii() && !FExpr->isUTF8()) {
3972     CheckFormatHandler::EmitFormatDiagnostic(
3973       *this, inFunctionCall, Args[format_idx],
3974       PDiag(diag::warn_format_string_is_wide_literal), FExpr->getLocStart(),
3975       /*IsStringLocation*/true, OrigFormatExpr->getSourceRange());
3976     return;
3977   }
3978 
3979   // Str - The format string.  NOTE: this is NOT null-terminated!
3980   StringRef StrRef = FExpr->getString();
3981   const char *Str = StrRef.data();
3982   // Account for cases where the string literal is truncated in a declaration.
3983   const ConstantArrayType *T = Context.getAsConstantArrayType(FExpr->getType());
3984   assert(T && "String literal not of constant array type!");
3985   size_t TypeSize = T->getSize().getZExtValue();
3986   size_t StrLen = std::min(std::max(TypeSize, size_t(1)) - 1, StrRef.size());
3987   const unsigned numDataArgs = Args.size() - firstDataArg;
3988 
3989   // Emit a warning if the string literal is truncated and does not contain an
3990   // embedded null character.
3991   if (TypeSize <= StrRef.size() &&
3992       StrRef.substr(0, TypeSize).find('\0') == StringRef::npos) {
3993     CheckFormatHandler::EmitFormatDiagnostic(
3994         *this, inFunctionCall, Args[format_idx],
3995         PDiag(diag::warn_printf_format_string_not_null_terminated),
3996         FExpr->getLocStart(),
3997         /*IsStringLocation=*/true, OrigFormatExpr->getSourceRange());
3998     return;
3999   }
4000 
4001   // CHECK: empty format string?
4002   if (StrLen == 0 && numDataArgs > 0) {
4003     CheckFormatHandler::EmitFormatDiagnostic(
4004       *this, inFunctionCall, Args[format_idx],
4005       PDiag(diag::warn_empty_format_string), FExpr->getLocStart(),
4006       /*IsStringLocation*/true, OrigFormatExpr->getSourceRange());
4007     return;
4008   }
4009 
4010   if (Type == FST_Printf || Type == FST_NSString) {
4011     CheckPrintfHandler H(*this, FExpr, OrigFormatExpr, firstDataArg,
4012                          numDataArgs, (Type == FST_NSString),
4013                          Str, HasVAListArg, Args, format_idx,
4014                          inFunctionCall, CallType, CheckedVarArgs);
4015 
4016     if (!analyze_format_string::ParsePrintfString(H, Str, Str + StrLen,
4017                                                   getLangOpts(),
4018                                                   Context.getTargetInfo()))
4019       H.DoneProcessing();
4020   } else if (Type == FST_Scanf) {
4021     CheckScanfHandler H(*this, FExpr, OrigFormatExpr, firstDataArg, numDataArgs,
4022                         Str, HasVAListArg, Args, format_idx,
4023                         inFunctionCall, CallType, CheckedVarArgs);
4024 
4025     if (!analyze_format_string::ParseScanfString(H, Str, Str + StrLen,
4026                                                  getLangOpts(),
4027                                                  Context.getTargetInfo()))
4028       H.DoneProcessing();
4029   } // TODO: handle other formats
4030 }
4031 
FormatStringHasSArg(const StringLiteral * FExpr)4032 bool Sema::FormatStringHasSArg(const StringLiteral *FExpr) {
4033   // Str - The format string.  NOTE: this is NOT null-terminated!
4034   StringRef StrRef = FExpr->getString();
4035   const char *Str = StrRef.data();
4036   // Account for cases where the string literal is truncated in a declaration.
4037   const ConstantArrayType *T = Context.getAsConstantArrayType(FExpr->getType());
4038   assert(T && "String literal not of constant array type!");
4039   size_t TypeSize = T->getSize().getZExtValue();
4040   size_t StrLen = std::min(std::max(TypeSize, size_t(1)) - 1, StrRef.size());
4041   return analyze_format_string::ParseFormatStringHasSArg(Str, Str + StrLen,
4042                                                          getLangOpts(),
4043                                                          Context.getTargetInfo());
4044 }
4045 
4046 //===--- CHECK: Warn on use of wrong absolute value function. -------------===//
4047 
4048 // Returns the related absolute value function that is larger, of 0 if one
4049 // does not exist.
getLargerAbsoluteValueFunction(unsigned AbsFunction)4050 static unsigned getLargerAbsoluteValueFunction(unsigned AbsFunction) {
4051   switch (AbsFunction) {
4052   default:
4053     return 0;
4054 
4055   case Builtin::BI__builtin_abs:
4056     return Builtin::BI__builtin_labs;
4057   case Builtin::BI__builtin_labs:
4058     return Builtin::BI__builtin_llabs;
4059   case Builtin::BI__builtin_llabs:
4060     return 0;
4061 
4062   case Builtin::BI__builtin_fabsf:
4063     return Builtin::BI__builtin_fabs;
4064   case Builtin::BI__builtin_fabs:
4065     return Builtin::BI__builtin_fabsl;
4066   case Builtin::BI__builtin_fabsl:
4067     return 0;
4068 
4069   case Builtin::BI__builtin_cabsf:
4070     return Builtin::BI__builtin_cabs;
4071   case Builtin::BI__builtin_cabs:
4072     return Builtin::BI__builtin_cabsl;
4073   case Builtin::BI__builtin_cabsl:
4074     return 0;
4075 
4076   case Builtin::BIabs:
4077     return Builtin::BIlabs;
4078   case Builtin::BIlabs:
4079     return Builtin::BIllabs;
4080   case Builtin::BIllabs:
4081     return 0;
4082 
4083   case Builtin::BIfabsf:
4084     return Builtin::BIfabs;
4085   case Builtin::BIfabs:
4086     return Builtin::BIfabsl;
4087   case Builtin::BIfabsl:
4088     return 0;
4089 
4090   case Builtin::BIcabsf:
4091    return Builtin::BIcabs;
4092   case Builtin::BIcabs:
4093     return Builtin::BIcabsl;
4094   case Builtin::BIcabsl:
4095     return 0;
4096   }
4097 }
4098 
4099 // Returns the argument type of the absolute value function.
getAbsoluteValueArgumentType(ASTContext & Context,unsigned AbsType)4100 static QualType getAbsoluteValueArgumentType(ASTContext &Context,
4101                                              unsigned AbsType) {
4102   if (AbsType == 0)
4103     return QualType();
4104 
4105   ASTContext::GetBuiltinTypeError Error = ASTContext::GE_None;
4106   QualType BuiltinType = Context.GetBuiltinType(AbsType, Error);
4107   if (Error != ASTContext::GE_None)
4108     return QualType();
4109 
4110   const FunctionProtoType *FT = BuiltinType->getAs<FunctionProtoType>();
4111   if (!FT)
4112     return QualType();
4113 
4114   if (FT->getNumParams() != 1)
4115     return QualType();
4116 
4117   return FT->getParamType(0);
4118 }
4119 
4120 // Returns the best absolute value function, or zero, based on type and
4121 // current absolute value function.
getBestAbsFunction(ASTContext & Context,QualType ArgType,unsigned AbsFunctionKind)4122 static unsigned getBestAbsFunction(ASTContext &Context, QualType ArgType,
4123                                    unsigned AbsFunctionKind) {
4124   unsigned BestKind = 0;
4125   uint64_t ArgSize = Context.getTypeSize(ArgType);
4126   for (unsigned Kind = AbsFunctionKind; Kind != 0;
4127        Kind = getLargerAbsoluteValueFunction(Kind)) {
4128     QualType ParamType = getAbsoluteValueArgumentType(Context, Kind);
4129     if (Context.getTypeSize(ParamType) >= ArgSize) {
4130       if (BestKind == 0)
4131         BestKind = Kind;
4132       else if (Context.hasSameType(ParamType, ArgType)) {
4133         BestKind = Kind;
4134         break;
4135       }
4136     }
4137   }
4138   return BestKind;
4139 }
4140 
4141 enum AbsoluteValueKind {
4142   AVK_Integer,
4143   AVK_Floating,
4144   AVK_Complex
4145 };
4146 
getAbsoluteValueKind(QualType T)4147 static AbsoluteValueKind getAbsoluteValueKind(QualType T) {
4148   if (T->isIntegralOrEnumerationType())
4149     return AVK_Integer;
4150   if (T->isRealFloatingType())
4151     return AVK_Floating;
4152   if (T->isAnyComplexType())
4153     return AVK_Complex;
4154 
4155   llvm_unreachable("Type not integer, floating, or complex");
4156 }
4157 
4158 // Changes the absolute value function to a different type.  Preserves whether
4159 // the function is a builtin.
changeAbsFunction(unsigned AbsKind,AbsoluteValueKind ValueKind)4160 static unsigned changeAbsFunction(unsigned AbsKind,
4161                                   AbsoluteValueKind ValueKind) {
4162   switch (ValueKind) {
4163   case AVK_Integer:
4164     switch (AbsKind) {
4165     default:
4166       return 0;
4167     case Builtin::BI__builtin_fabsf:
4168     case Builtin::BI__builtin_fabs:
4169     case Builtin::BI__builtin_fabsl:
4170     case Builtin::BI__builtin_cabsf:
4171     case Builtin::BI__builtin_cabs:
4172     case Builtin::BI__builtin_cabsl:
4173       return Builtin::BI__builtin_abs;
4174     case Builtin::BIfabsf:
4175     case Builtin::BIfabs:
4176     case Builtin::BIfabsl:
4177     case Builtin::BIcabsf:
4178     case Builtin::BIcabs:
4179     case Builtin::BIcabsl:
4180       return Builtin::BIabs;
4181     }
4182   case AVK_Floating:
4183     switch (AbsKind) {
4184     default:
4185       return 0;
4186     case Builtin::BI__builtin_abs:
4187     case Builtin::BI__builtin_labs:
4188     case Builtin::BI__builtin_llabs:
4189     case Builtin::BI__builtin_cabsf:
4190     case Builtin::BI__builtin_cabs:
4191     case Builtin::BI__builtin_cabsl:
4192       return Builtin::BI__builtin_fabsf;
4193     case Builtin::BIabs:
4194     case Builtin::BIlabs:
4195     case Builtin::BIllabs:
4196     case Builtin::BIcabsf:
4197     case Builtin::BIcabs:
4198     case Builtin::BIcabsl:
4199       return Builtin::BIfabsf;
4200     }
4201   case AVK_Complex:
4202     switch (AbsKind) {
4203     default:
4204       return 0;
4205     case Builtin::BI__builtin_abs:
4206     case Builtin::BI__builtin_labs:
4207     case Builtin::BI__builtin_llabs:
4208     case Builtin::BI__builtin_fabsf:
4209     case Builtin::BI__builtin_fabs:
4210     case Builtin::BI__builtin_fabsl:
4211       return Builtin::BI__builtin_cabsf;
4212     case Builtin::BIabs:
4213     case Builtin::BIlabs:
4214     case Builtin::BIllabs:
4215     case Builtin::BIfabsf:
4216     case Builtin::BIfabs:
4217     case Builtin::BIfabsl:
4218       return Builtin::BIcabsf;
4219     }
4220   }
4221   llvm_unreachable("Unable to convert function");
4222 }
4223 
getAbsoluteValueFunctionKind(const FunctionDecl * FDecl)4224 static unsigned getAbsoluteValueFunctionKind(const FunctionDecl *FDecl) {
4225   const IdentifierInfo *FnInfo = FDecl->getIdentifier();
4226   if (!FnInfo)
4227     return 0;
4228 
4229   switch (FDecl->getBuiltinID()) {
4230   default:
4231     return 0;
4232   case Builtin::BI__builtin_abs:
4233   case Builtin::BI__builtin_fabs:
4234   case Builtin::BI__builtin_fabsf:
4235   case Builtin::BI__builtin_fabsl:
4236   case Builtin::BI__builtin_labs:
4237   case Builtin::BI__builtin_llabs:
4238   case Builtin::BI__builtin_cabs:
4239   case Builtin::BI__builtin_cabsf:
4240   case Builtin::BI__builtin_cabsl:
4241   case Builtin::BIabs:
4242   case Builtin::BIlabs:
4243   case Builtin::BIllabs:
4244   case Builtin::BIfabs:
4245   case Builtin::BIfabsf:
4246   case Builtin::BIfabsl:
4247   case Builtin::BIcabs:
4248   case Builtin::BIcabsf:
4249   case Builtin::BIcabsl:
4250     return FDecl->getBuiltinID();
4251   }
4252   llvm_unreachable("Unknown Builtin type");
4253 }
4254 
4255 // If the replacement is valid, emit a note with replacement function.
4256 // Additionally, suggest including the proper header if not already included.
emitReplacement(Sema & S,SourceLocation Loc,SourceRange Range,unsigned AbsKind,QualType ArgType)4257 static void emitReplacement(Sema &S, SourceLocation Loc, SourceRange Range,
4258                             unsigned AbsKind, QualType ArgType) {
4259   bool EmitHeaderHint = true;
4260   const char *HeaderName = nullptr;
4261   const char *FunctionName = nullptr;
4262   if (S.getLangOpts().CPlusPlus && !ArgType->isAnyComplexType()) {
4263     FunctionName = "std::abs";
4264     if (ArgType->isIntegralOrEnumerationType()) {
4265       HeaderName = "cstdlib";
4266     } else if (ArgType->isRealFloatingType()) {
4267       HeaderName = "cmath";
4268     } else {
4269       llvm_unreachable("Invalid Type");
4270     }
4271 
4272     // Lookup all std::abs
4273     if (NamespaceDecl *Std = S.getStdNamespace()) {
4274       LookupResult R(S, &S.Context.Idents.get("abs"), Loc, Sema::LookupAnyName);
4275       R.suppressDiagnostics();
4276       S.LookupQualifiedName(R, Std);
4277 
4278       for (const auto *I : R) {
4279         const FunctionDecl *FDecl = nullptr;
4280         if (const UsingShadowDecl *UsingD = dyn_cast<UsingShadowDecl>(I)) {
4281           FDecl = dyn_cast<FunctionDecl>(UsingD->getTargetDecl());
4282         } else {
4283           FDecl = dyn_cast<FunctionDecl>(I);
4284         }
4285         if (!FDecl)
4286           continue;
4287 
4288         // Found std::abs(), check that they are the right ones.
4289         if (FDecl->getNumParams() != 1)
4290           continue;
4291 
4292         // Check that the parameter type can handle the argument.
4293         QualType ParamType = FDecl->getParamDecl(0)->getType();
4294         if (getAbsoluteValueKind(ArgType) == getAbsoluteValueKind(ParamType) &&
4295             S.Context.getTypeSize(ArgType) <=
4296                 S.Context.getTypeSize(ParamType)) {
4297           // Found a function, don't need the header hint.
4298           EmitHeaderHint = false;
4299           break;
4300         }
4301       }
4302     }
4303   } else {
4304     FunctionName = S.Context.BuiltinInfo.GetName(AbsKind);
4305     HeaderName = S.Context.BuiltinInfo.getHeaderName(AbsKind);
4306 
4307     if (HeaderName) {
4308       DeclarationName DN(&S.Context.Idents.get(FunctionName));
4309       LookupResult R(S, DN, Loc, Sema::LookupAnyName);
4310       R.suppressDiagnostics();
4311       S.LookupName(R, S.getCurScope());
4312 
4313       if (R.isSingleResult()) {
4314         FunctionDecl *FD = dyn_cast<FunctionDecl>(R.getFoundDecl());
4315         if (FD && FD->getBuiltinID() == AbsKind) {
4316           EmitHeaderHint = false;
4317         } else {
4318           return;
4319         }
4320       } else if (!R.empty()) {
4321         return;
4322       }
4323     }
4324   }
4325 
4326   S.Diag(Loc, diag::note_replace_abs_function)
4327       << FunctionName << FixItHint::CreateReplacement(Range, FunctionName);
4328 
4329   if (!HeaderName)
4330     return;
4331 
4332   if (!EmitHeaderHint)
4333     return;
4334 
4335   S.Diag(Loc, diag::note_include_header_or_declare) << HeaderName
4336                                                     << FunctionName;
4337 }
4338 
IsFunctionStdAbs(const FunctionDecl * FDecl)4339 static bool IsFunctionStdAbs(const FunctionDecl *FDecl) {
4340   if (!FDecl)
4341     return false;
4342 
4343   if (!FDecl->getIdentifier() || !FDecl->getIdentifier()->isStr("abs"))
4344     return false;
4345 
4346   const NamespaceDecl *ND = dyn_cast<NamespaceDecl>(FDecl->getDeclContext());
4347 
4348   while (ND && ND->isInlineNamespace()) {
4349     ND = dyn_cast<NamespaceDecl>(ND->getDeclContext());
4350   }
4351 
4352   if (!ND || !ND->getIdentifier() || !ND->getIdentifier()->isStr("std"))
4353     return false;
4354 
4355   if (!isa<TranslationUnitDecl>(ND->getDeclContext()))
4356     return false;
4357 
4358   return true;
4359 }
4360 
4361 // Warn when using the wrong abs() function.
CheckAbsoluteValueFunction(const CallExpr * Call,const FunctionDecl * FDecl,IdentifierInfo * FnInfo)4362 void Sema::CheckAbsoluteValueFunction(const CallExpr *Call,
4363                                       const FunctionDecl *FDecl,
4364                                       IdentifierInfo *FnInfo) {
4365   if (Call->getNumArgs() != 1)
4366     return;
4367 
4368   unsigned AbsKind = getAbsoluteValueFunctionKind(FDecl);
4369   bool IsStdAbs = IsFunctionStdAbs(FDecl);
4370   if (AbsKind == 0 && !IsStdAbs)
4371     return;
4372 
4373   QualType ArgType = Call->getArg(0)->IgnoreParenImpCasts()->getType();
4374   QualType ParamType = Call->getArg(0)->getType();
4375 
4376   // Unsigned types cannot be negative.  Suggest removing the absolute value
4377   // function call.
4378   if (ArgType->isUnsignedIntegerType()) {
4379     const char *FunctionName =
4380         IsStdAbs ? "std::abs" : Context.BuiltinInfo.GetName(AbsKind);
4381     Diag(Call->getExprLoc(), diag::warn_unsigned_abs) << ArgType << ParamType;
4382     Diag(Call->getExprLoc(), diag::note_remove_abs)
4383         << FunctionName
4384         << FixItHint::CreateRemoval(Call->getCallee()->getSourceRange());
4385     return;
4386   }
4387 
4388   // std::abs has overloads which prevent most of the absolute value problems
4389   // from occurring.
4390   if (IsStdAbs)
4391     return;
4392 
4393   AbsoluteValueKind ArgValueKind = getAbsoluteValueKind(ArgType);
4394   AbsoluteValueKind ParamValueKind = getAbsoluteValueKind(ParamType);
4395 
4396   // The argument and parameter are the same kind.  Check if they are the right
4397   // size.
4398   if (ArgValueKind == ParamValueKind) {
4399     if (Context.getTypeSize(ArgType) <= Context.getTypeSize(ParamType))
4400       return;
4401 
4402     unsigned NewAbsKind = getBestAbsFunction(Context, ArgType, AbsKind);
4403     Diag(Call->getExprLoc(), diag::warn_abs_too_small)
4404         << FDecl << ArgType << ParamType;
4405 
4406     if (NewAbsKind == 0)
4407       return;
4408 
4409     emitReplacement(*this, Call->getExprLoc(),
4410                     Call->getCallee()->getSourceRange(), NewAbsKind, ArgType);
4411     return;
4412   }
4413 
4414   // ArgValueKind != ParamValueKind
4415   // The wrong type of absolute value function was used.  Attempt to find the
4416   // proper one.
4417   unsigned NewAbsKind = changeAbsFunction(AbsKind, ArgValueKind);
4418   NewAbsKind = getBestAbsFunction(Context, ArgType, NewAbsKind);
4419   if (NewAbsKind == 0)
4420     return;
4421 
4422   Diag(Call->getExprLoc(), diag::warn_wrong_absolute_value_type)
4423       << FDecl << ParamValueKind << ArgValueKind;
4424 
4425   emitReplacement(*this, Call->getExprLoc(),
4426                   Call->getCallee()->getSourceRange(), NewAbsKind, ArgType);
4427   return;
4428 }
4429 
4430 //===--- CHECK: Standard memory functions ---------------------------------===//
4431 
4432 /// \brief Takes the expression passed to the size_t parameter of functions
4433 /// such as memcmp, strncat, etc and warns if it's a comparison.
4434 ///
4435 /// This is to catch typos like `if (memcmp(&a, &b, sizeof(a) > 0))`.
CheckMemorySizeofForComparison(Sema & S,const Expr * E,IdentifierInfo * FnName,SourceLocation FnLoc,SourceLocation RParenLoc)4436 static bool CheckMemorySizeofForComparison(Sema &S, const Expr *E,
4437                                            IdentifierInfo *FnName,
4438                                            SourceLocation FnLoc,
4439                                            SourceLocation RParenLoc) {
4440   const BinaryOperator *Size = dyn_cast<BinaryOperator>(E);
4441   if (!Size)
4442     return false;
4443 
4444   // if E is binop and op is >, <, >=, <=, ==, &&, ||:
4445   if (!Size->isComparisonOp() && !Size->isEqualityOp() && !Size->isLogicalOp())
4446     return false;
4447 
4448   SourceRange SizeRange = Size->getSourceRange();
4449   S.Diag(Size->getOperatorLoc(), diag::warn_memsize_comparison)
4450       << SizeRange << FnName;
4451   S.Diag(FnLoc, diag::note_memsize_comparison_paren)
4452       << FnName << FixItHint::CreateInsertion(
4453                        S.getLocForEndOfToken(Size->getLHS()->getLocEnd()), ")")
4454       << FixItHint::CreateRemoval(RParenLoc);
4455   S.Diag(SizeRange.getBegin(), diag::note_memsize_comparison_cast_silence)
4456       << FixItHint::CreateInsertion(SizeRange.getBegin(), "(size_t)(")
4457       << FixItHint::CreateInsertion(S.getLocForEndOfToken(SizeRange.getEnd()),
4458                                     ")");
4459 
4460   return true;
4461 }
4462 
4463 /// \brief Determine whether the given type is or contains a dynamic class type
4464 /// (e.g., whether it has a vtable).
getContainedDynamicClass(QualType T,bool & IsContained)4465 static const CXXRecordDecl *getContainedDynamicClass(QualType T,
4466                                                      bool &IsContained) {
4467   // Look through array types while ignoring qualifiers.
4468   const Type *Ty = T->getBaseElementTypeUnsafe();
4469   IsContained = false;
4470 
4471   const CXXRecordDecl *RD = Ty->getAsCXXRecordDecl();
4472   RD = RD ? RD->getDefinition() : nullptr;
4473   if (!RD)
4474     return nullptr;
4475 
4476   if (RD->isDynamicClass())
4477     return RD;
4478 
4479   // Check all the fields.  If any bases were dynamic, the class is dynamic.
4480   // It's impossible for a class to transitively contain itself by value, so
4481   // infinite recursion is impossible.
4482   for (auto *FD : RD->fields()) {
4483     bool SubContained;
4484     if (const CXXRecordDecl *ContainedRD =
4485             getContainedDynamicClass(FD->getType(), SubContained)) {
4486       IsContained = true;
4487       return ContainedRD;
4488     }
4489   }
4490 
4491   return nullptr;
4492 }
4493 
4494 /// \brief If E is a sizeof expression, returns its argument expression,
4495 /// otherwise returns NULL.
getSizeOfExprArg(const Expr * E)4496 static const Expr *getSizeOfExprArg(const Expr* E) {
4497   if (const UnaryExprOrTypeTraitExpr *SizeOf =
4498       dyn_cast<UnaryExprOrTypeTraitExpr>(E))
4499     if (SizeOf->getKind() == clang::UETT_SizeOf && !SizeOf->isArgumentType())
4500       return SizeOf->getArgumentExpr()->IgnoreParenImpCasts();
4501 
4502   return nullptr;
4503 }
4504 
4505 /// \brief If E is a sizeof expression, returns its argument type.
getSizeOfArgType(const Expr * E)4506 static QualType getSizeOfArgType(const Expr* E) {
4507   if (const UnaryExprOrTypeTraitExpr *SizeOf =
4508       dyn_cast<UnaryExprOrTypeTraitExpr>(E))
4509     if (SizeOf->getKind() == clang::UETT_SizeOf)
4510       return SizeOf->getTypeOfArgument();
4511 
4512   return QualType();
4513 }
4514 
4515 /// \brief Check for dangerous or invalid arguments to memset().
4516 ///
4517 /// This issues warnings on known problematic, dangerous or unspecified
4518 /// arguments to the standard 'memset', 'memcpy', 'memmove', and 'memcmp'
4519 /// function calls.
4520 ///
4521 /// \param Call The call expression to diagnose.
CheckMemaccessArguments(const CallExpr * Call,unsigned BId,IdentifierInfo * FnName)4522 void Sema::CheckMemaccessArguments(const CallExpr *Call,
4523                                    unsigned BId,
4524                                    IdentifierInfo *FnName) {
4525   assert(BId != 0);
4526 
4527   // It is possible to have a non-standard definition of memset.  Validate
4528   // we have enough arguments, and if not, abort further checking.
4529   unsigned ExpectedNumArgs = (BId == Builtin::BIstrndup ? 2 : 3);
4530   if (Call->getNumArgs() < ExpectedNumArgs)
4531     return;
4532 
4533   unsigned LastArg = (BId == Builtin::BImemset ||
4534                       BId == Builtin::BIstrndup ? 1 : 2);
4535   unsigned LenArg = (BId == Builtin::BIstrndup ? 1 : 2);
4536   const Expr *LenExpr = Call->getArg(LenArg)->IgnoreParenImpCasts();
4537 
4538   if (CheckMemorySizeofForComparison(*this, LenExpr, FnName,
4539                                      Call->getLocStart(), Call->getRParenLoc()))
4540     return;
4541 
4542   // We have special checking when the length is a sizeof expression.
4543   QualType SizeOfArgTy = getSizeOfArgType(LenExpr);
4544   const Expr *SizeOfArg = getSizeOfExprArg(LenExpr);
4545   llvm::FoldingSetNodeID SizeOfArgID;
4546 
4547   for (unsigned ArgIdx = 0; ArgIdx != LastArg; ++ArgIdx) {
4548     const Expr *Dest = Call->getArg(ArgIdx)->IgnoreParenImpCasts();
4549     SourceRange ArgRange = Call->getArg(ArgIdx)->getSourceRange();
4550 
4551     QualType DestTy = Dest->getType();
4552     if (const PointerType *DestPtrTy = DestTy->getAs<PointerType>()) {
4553       QualType PointeeTy = DestPtrTy->getPointeeType();
4554 
4555       // Never warn about void type pointers. This can be used to suppress
4556       // false positives.
4557       if (PointeeTy->isVoidType())
4558         continue;
4559 
4560       // Catch "memset(p, 0, sizeof(p))" -- needs to be sizeof(*p). Do this by
4561       // actually comparing the expressions for equality. Because computing the
4562       // expression IDs can be expensive, we only do this if the diagnostic is
4563       // enabled.
4564       if (SizeOfArg &&
4565           !Diags.isIgnored(diag::warn_sizeof_pointer_expr_memaccess,
4566                            SizeOfArg->getExprLoc())) {
4567         // We only compute IDs for expressions if the warning is enabled, and
4568         // cache the sizeof arg's ID.
4569         if (SizeOfArgID == llvm::FoldingSetNodeID())
4570           SizeOfArg->Profile(SizeOfArgID, Context, true);
4571         llvm::FoldingSetNodeID DestID;
4572         Dest->Profile(DestID, Context, true);
4573         if (DestID == SizeOfArgID) {
4574           // TODO: For strncpy() and friends, this could suggest sizeof(dst)
4575           //       over sizeof(src) as well.
4576           unsigned ActionIdx = 0; // Default is to suggest dereferencing.
4577           StringRef ReadableName = FnName->getName();
4578 
4579           if (const UnaryOperator *UnaryOp = dyn_cast<UnaryOperator>(Dest))
4580             if (UnaryOp->getOpcode() == UO_AddrOf)
4581               ActionIdx = 1; // If its an address-of operator, just remove it.
4582           if (!PointeeTy->isIncompleteType() &&
4583               (Context.getTypeSize(PointeeTy) == Context.getCharWidth()))
4584             ActionIdx = 2; // If the pointee's size is sizeof(char),
4585                            // suggest an explicit length.
4586 
4587           // If the function is defined as a builtin macro, do not show macro
4588           // expansion.
4589           SourceLocation SL = SizeOfArg->getExprLoc();
4590           SourceRange DSR = Dest->getSourceRange();
4591           SourceRange SSR = SizeOfArg->getSourceRange();
4592           SourceManager &SM = getSourceManager();
4593 
4594           if (SM.isMacroArgExpansion(SL)) {
4595             ReadableName = Lexer::getImmediateMacroName(SL, SM, LangOpts);
4596             SL = SM.getSpellingLoc(SL);
4597             DSR = SourceRange(SM.getSpellingLoc(DSR.getBegin()),
4598                              SM.getSpellingLoc(DSR.getEnd()));
4599             SSR = SourceRange(SM.getSpellingLoc(SSR.getBegin()),
4600                              SM.getSpellingLoc(SSR.getEnd()));
4601           }
4602 
4603           DiagRuntimeBehavior(SL, SizeOfArg,
4604                               PDiag(diag::warn_sizeof_pointer_expr_memaccess)
4605                                 << ReadableName
4606                                 << PointeeTy
4607                                 << DestTy
4608                                 << DSR
4609                                 << SSR);
4610           DiagRuntimeBehavior(SL, SizeOfArg,
4611                          PDiag(diag::warn_sizeof_pointer_expr_memaccess_note)
4612                                 << ActionIdx
4613                                 << SSR);
4614 
4615           break;
4616         }
4617       }
4618 
4619       // Also check for cases where the sizeof argument is the exact same
4620       // type as the memory argument, and where it points to a user-defined
4621       // record type.
4622       if (SizeOfArgTy != QualType()) {
4623         if (PointeeTy->isRecordType() &&
4624             Context.typesAreCompatible(SizeOfArgTy, DestTy)) {
4625           DiagRuntimeBehavior(LenExpr->getExprLoc(), Dest,
4626                               PDiag(diag::warn_sizeof_pointer_type_memaccess)
4627                                 << FnName << SizeOfArgTy << ArgIdx
4628                                 << PointeeTy << Dest->getSourceRange()
4629                                 << LenExpr->getSourceRange());
4630           break;
4631         }
4632       }
4633 
4634       // Always complain about dynamic classes.
4635       bool IsContained;
4636       if (const CXXRecordDecl *ContainedRD =
4637               getContainedDynamicClass(PointeeTy, IsContained)) {
4638 
4639         unsigned OperationType = 0;
4640         // "overwritten" if we're warning about the destination for any call
4641         // but memcmp; otherwise a verb appropriate to the call.
4642         if (ArgIdx != 0 || BId == Builtin::BImemcmp) {
4643           if (BId == Builtin::BImemcpy)
4644             OperationType = 1;
4645           else if(BId == Builtin::BImemmove)
4646             OperationType = 2;
4647           else if (BId == Builtin::BImemcmp)
4648             OperationType = 3;
4649         }
4650 
4651         DiagRuntimeBehavior(
4652           Dest->getExprLoc(), Dest,
4653           PDiag(diag::warn_dyn_class_memaccess)
4654             << (BId == Builtin::BImemcmp ? ArgIdx + 2 : ArgIdx)
4655             << FnName << IsContained << ContainedRD << OperationType
4656             << Call->getCallee()->getSourceRange());
4657       } else if (PointeeTy.hasNonTrivialObjCLifetime() &&
4658                BId != Builtin::BImemset)
4659         DiagRuntimeBehavior(
4660           Dest->getExprLoc(), Dest,
4661           PDiag(diag::warn_arc_object_memaccess)
4662             << ArgIdx << FnName << PointeeTy
4663             << Call->getCallee()->getSourceRange());
4664       else
4665         continue;
4666 
4667       DiagRuntimeBehavior(
4668         Dest->getExprLoc(), Dest,
4669         PDiag(diag::note_bad_memaccess_silence)
4670           << FixItHint::CreateInsertion(ArgRange.getBegin(), "(void*)"));
4671       break;
4672     }
4673   }
4674 }
4675 
4676 // A little helper routine: ignore addition and subtraction of integer literals.
4677 // This intentionally does not ignore all integer constant expressions because
4678 // we don't want to remove sizeof().
ignoreLiteralAdditions(const Expr * Ex,ASTContext & Ctx)4679 static const Expr *ignoreLiteralAdditions(const Expr *Ex, ASTContext &Ctx) {
4680   Ex = Ex->IgnoreParenCasts();
4681 
4682   for (;;) {
4683     const BinaryOperator * BO = dyn_cast<BinaryOperator>(Ex);
4684     if (!BO || !BO->isAdditiveOp())
4685       break;
4686 
4687     const Expr *RHS = BO->getRHS()->IgnoreParenCasts();
4688     const Expr *LHS = BO->getLHS()->IgnoreParenCasts();
4689 
4690     if (isa<IntegerLiteral>(RHS))
4691       Ex = LHS;
4692     else if (isa<IntegerLiteral>(LHS))
4693       Ex = RHS;
4694     else
4695       break;
4696   }
4697 
4698   return Ex;
4699 }
4700 
isConstantSizeArrayWithMoreThanOneElement(QualType Ty,ASTContext & Context)4701 static bool isConstantSizeArrayWithMoreThanOneElement(QualType Ty,
4702                                                       ASTContext &Context) {
4703   // Only handle constant-sized or VLAs, but not flexible members.
4704   if (const ConstantArrayType *CAT = Context.getAsConstantArrayType(Ty)) {
4705     // Only issue the FIXIT for arrays of size > 1.
4706     if (CAT->getSize().getSExtValue() <= 1)
4707       return false;
4708   } else if (!Ty->isVariableArrayType()) {
4709     return false;
4710   }
4711   return true;
4712 }
4713 
4714 // Warn if the user has made the 'size' argument to strlcpy or strlcat
4715 // be the size of the source, instead of the destination.
CheckStrlcpycatArguments(const CallExpr * Call,IdentifierInfo * FnName)4716 void Sema::CheckStrlcpycatArguments(const CallExpr *Call,
4717                                     IdentifierInfo *FnName) {
4718 
4719   // Don't crash if the user has the wrong number of arguments
4720   unsigned NumArgs = Call->getNumArgs();
4721   if ((NumArgs != 3) && (NumArgs != 4))
4722     return;
4723 
4724   const Expr *SrcArg = ignoreLiteralAdditions(Call->getArg(1), Context);
4725   const Expr *SizeArg = ignoreLiteralAdditions(Call->getArg(2), Context);
4726   const Expr *CompareWithSrc = nullptr;
4727 
4728   if (CheckMemorySizeofForComparison(*this, SizeArg, FnName,
4729                                      Call->getLocStart(), Call->getRParenLoc()))
4730     return;
4731 
4732   // Look for 'strlcpy(dst, x, sizeof(x))'
4733   if (const Expr *Ex = getSizeOfExprArg(SizeArg))
4734     CompareWithSrc = Ex;
4735   else {
4736     // Look for 'strlcpy(dst, x, strlen(x))'
4737     if (const CallExpr *SizeCall = dyn_cast<CallExpr>(SizeArg)) {
4738       if (SizeCall->getBuiltinCallee() == Builtin::BIstrlen &&
4739           SizeCall->getNumArgs() == 1)
4740         CompareWithSrc = ignoreLiteralAdditions(SizeCall->getArg(0), Context);
4741     }
4742   }
4743 
4744   if (!CompareWithSrc)
4745     return;
4746 
4747   // Determine if the argument to sizeof/strlen is equal to the source
4748   // argument.  In principle there's all kinds of things you could do
4749   // here, for instance creating an == expression and evaluating it with
4750   // EvaluateAsBooleanCondition, but this uses a more direct technique:
4751   const DeclRefExpr *SrcArgDRE = dyn_cast<DeclRefExpr>(SrcArg);
4752   if (!SrcArgDRE)
4753     return;
4754 
4755   const DeclRefExpr *CompareWithSrcDRE = dyn_cast<DeclRefExpr>(CompareWithSrc);
4756   if (!CompareWithSrcDRE ||
4757       SrcArgDRE->getDecl() != CompareWithSrcDRE->getDecl())
4758     return;
4759 
4760   const Expr *OriginalSizeArg = Call->getArg(2);
4761   Diag(CompareWithSrcDRE->getLocStart(), diag::warn_strlcpycat_wrong_size)
4762     << OriginalSizeArg->getSourceRange() << FnName;
4763 
4764   // Output a FIXIT hint if the destination is an array (rather than a
4765   // pointer to an array).  This could be enhanced to handle some
4766   // pointers if we know the actual size, like if DstArg is 'array+2'
4767   // we could say 'sizeof(array)-2'.
4768   const Expr *DstArg = Call->getArg(0)->IgnoreParenImpCasts();
4769   if (!isConstantSizeArrayWithMoreThanOneElement(DstArg->getType(), Context))
4770     return;
4771 
4772   SmallString<128> sizeString;
4773   llvm::raw_svector_ostream OS(sizeString);
4774   OS << "sizeof(";
4775   DstArg->printPretty(OS, nullptr, getPrintingPolicy());
4776   OS << ")";
4777 
4778   Diag(OriginalSizeArg->getLocStart(), diag::note_strlcpycat_wrong_size)
4779     << FixItHint::CreateReplacement(OriginalSizeArg->getSourceRange(),
4780                                     OS.str());
4781 }
4782 
4783 /// Check if two expressions refer to the same declaration.
referToTheSameDecl(const Expr * E1,const Expr * E2)4784 static bool referToTheSameDecl(const Expr *E1, const Expr *E2) {
4785   if (const DeclRefExpr *D1 = dyn_cast_or_null<DeclRefExpr>(E1))
4786     if (const DeclRefExpr *D2 = dyn_cast_or_null<DeclRefExpr>(E2))
4787       return D1->getDecl() == D2->getDecl();
4788   return false;
4789 }
4790 
getStrlenExprArg(const Expr * E)4791 static const Expr *getStrlenExprArg(const Expr *E) {
4792   if (const CallExpr *CE = dyn_cast<CallExpr>(E)) {
4793     const FunctionDecl *FD = CE->getDirectCallee();
4794     if (!FD || FD->getMemoryFunctionKind() != Builtin::BIstrlen)
4795       return nullptr;
4796     return CE->getArg(0)->IgnoreParenCasts();
4797   }
4798   return nullptr;
4799 }
4800 
4801 // Warn on anti-patterns as the 'size' argument to strncat.
4802 // The correct size argument should look like following:
4803 //   strncat(dst, src, sizeof(dst) - strlen(dest) - 1);
CheckStrncatArguments(const CallExpr * CE,IdentifierInfo * FnName)4804 void Sema::CheckStrncatArguments(const CallExpr *CE,
4805                                  IdentifierInfo *FnName) {
4806   // Don't crash if the user has the wrong number of arguments.
4807   if (CE->getNumArgs() < 3)
4808     return;
4809   const Expr *DstArg = CE->getArg(0)->IgnoreParenCasts();
4810   const Expr *SrcArg = CE->getArg(1)->IgnoreParenCasts();
4811   const Expr *LenArg = CE->getArg(2)->IgnoreParenCasts();
4812 
4813   if (CheckMemorySizeofForComparison(*this, LenArg, FnName, CE->getLocStart(),
4814                                      CE->getRParenLoc()))
4815     return;
4816 
4817   // Identify common expressions, which are wrongly used as the size argument
4818   // to strncat and may lead to buffer overflows.
4819   unsigned PatternType = 0;
4820   if (const Expr *SizeOfArg = getSizeOfExprArg(LenArg)) {
4821     // - sizeof(dst)
4822     if (referToTheSameDecl(SizeOfArg, DstArg))
4823       PatternType = 1;
4824     // - sizeof(src)
4825     else if (referToTheSameDecl(SizeOfArg, SrcArg))
4826       PatternType = 2;
4827   } else if (const BinaryOperator *BE = dyn_cast<BinaryOperator>(LenArg)) {
4828     if (BE->getOpcode() == BO_Sub) {
4829       const Expr *L = BE->getLHS()->IgnoreParenCasts();
4830       const Expr *R = BE->getRHS()->IgnoreParenCasts();
4831       // - sizeof(dst) - strlen(dst)
4832       if (referToTheSameDecl(DstArg, getSizeOfExprArg(L)) &&
4833           referToTheSameDecl(DstArg, getStrlenExprArg(R)))
4834         PatternType = 1;
4835       // - sizeof(src) - (anything)
4836       else if (referToTheSameDecl(SrcArg, getSizeOfExprArg(L)))
4837         PatternType = 2;
4838     }
4839   }
4840 
4841   if (PatternType == 0)
4842     return;
4843 
4844   // Generate the diagnostic.
4845   SourceLocation SL = LenArg->getLocStart();
4846   SourceRange SR = LenArg->getSourceRange();
4847   SourceManager &SM = getSourceManager();
4848 
4849   // If the function is defined as a builtin macro, do not show macro expansion.
4850   if (SM.isMacroArgExpansion(SL)) {
4851     SL = SM.getSpellingLoc(SL);
4852     SR = SourceRange(SM.getSpellingLoc(SR.getBegin()),
4853                      SM.getSpellingLoc(SR.getEnd()));
4854   }
4855 
4856   // Check if the destination is an array (rather than a pointer to an array).
4857   QualType DstTy = DstArg->getType();
4858   bool isKnownSizeArray = isConstantSizeArrayWithMoreThanOneElement(DstTy,
4859                                                                     Context);
4860   if (!isKnownSizeArray) {
4861     if (PatternType == 1)
4862       Diag(SL, diag::warn_strncat_wrong_size) << SR;
4863     else
4864       Diag(SL, diag::warn_strncat_src_size) << SR;
4865     return;
4866   }
4867 
4868   if (PatternType == 1)
4869     Diag(SL, diag::warn_strncat_large_size) << SR;
4870   else
4871     Diag(SL, diag::warn_strncat_src_size) << SR;
4872 
4873   SmallString<128> sizeString;
4874   llvm::raw_svector_ostream OS(sizeString);
4875   OS << "sizeof(";
4876   DstArg->printPretty(OS, nullptr, getPrintingPolicy());
4877   OS << ") - ";
4878   OS << "strlen(";
4879   DstArg->printPretty(OS, nullptr, getPrintingPolicy());
4880   OS << ") - 1";
4881 
4882   Diag(SL, diag::note_strncat_wrong_size)
4883     << FixItHint::CreateReplacement(SR, OS.str());
4884 }
4885 
4886 //===--- CHECK: Return Address of Stack Variable --------------------------===//
4887 
4888 static Expr *EvalVal(Expr *E, SmallVectorImpl<DeclRefExpr *> &refVars,
4889                      Decl *ParentDecl);
4890 static Expr *EvalAddr(Expr* E, SmallVectorImpl<DeclRefExpr *> &refVars,
4891                       Decl *ParentDecl);
4892 
4893 /// CheckReturnStackAddr - Check if a return statement returns the address
4894 ///   of a stack variable.
4895 static void
CheckReturnStackAddr(Sema & S,Expr * RetValExp,QualType lhsType,SourceLocation ReturnLoc)4896 CheckReturnStackAddr(Sema &S, Expr *RetValExp, QualType lhsType,
4897                      SourceLocation ReturnLoc) {
4898 
4899   Expr *stackE = nullptr;
4900   SmallVector<DeclRefExpr *, 8> refVars;
4901 
4902   // Perform checking for returned stack addresses, local blocks,
4903   // label addresses or references to temporaries.
4904   if (lhsType->isPointerType() ||
4905       (!S.getLangOpts().ObjCAutoRefCount && lhsType->isBlockPointerType())) {
4906     stackE = EvalAddr(RetValExp, refVars, /*ParentDecl=*/nullptr);
4907   } else if (lhsType->isReferenceType()) {
4908     stackE = EvalVal(RetValExp, refVars, /*ParentDecl=*/nullptr);
4909   }
4910 
4911   if (!stackE)
4912     return; // Nothing suspicious was found.
4913 
4914   SourceLocation diagLoc;
4915   SourceRange diagRange;
4916   if (refVars.empty()) {
4917     diagLoc = stackE->getLocStart();
4918     diagRange = stackE->getSourceRange();
4919   } else {
4920     // We followed through a reference variable. 'stackE' contains the
4921     // problematic expression but we will warn at the return statement pointing
4922     // at the reference variable. We will later display the "trail" of
4923     // reference variables using notes.
4924     diagLoc = refVars[0]->getLocStart();
4925     diagRange = refVars[0]->getSourceRange();
4926   }
4927 
4928   if (DeclRefExpr *DR = dyn_cast<DeclRefExpr>(stackE)) { //address of local var.
4929     S.Diag(diagLoc, lhsType->isReferenceType() ? diag::warn_ret_stack_ref
4930                                              : diag::warn_ret_stack_addr)
4931      << DR->getDecl()->getDeclName() << diagRange;
4932   } else if (isa<BlockExpr>(stackE)) { // local block.
4933     S.Diag(diagLoc, diag::err_ret_local_block) << diagRange;
4934   } else if (isa<AddrLabelExpr>(stackE)) { // address of label.
4935     S.Diag(diagLoc, diag::warn_ret_addr_label) << diagRange;
4936   } else { // local temporary.
4937     S.Diag(diagLoc, lhsType->isReferenceType() ? diag::warn_ret_local_temp_ref
4938                                                : diag::warn_ret_local_temp_addr)
4939      << diagRange;
4940   }
4941 
4942   // Display the "trail" of reference variables that we followed until we
4943   // found the problematic expression using notes.
4944   for (unsigned i = 0, e = refVars.size(); i != e; ++i) {
4945     VarDecl *VD = cast<VarDecl>(refVars[i]->getDecl());
4946     // If this var binds to another reference var, show the range of the next
4947     // var, otherwise the var binds to the problematic expression, in which case
4948     // show the range of the expression.
4949     SourceRange range = (i < e-1) ? refVars[i+1]->getSourceRange()
4950                                   : stackE->getSourceRange();
4951     S.Diag(VD->getLocation(), diag::note_ref_var_local_bind)
4952         << VD->getDeclName() << range;
4953   }
4954 }
4955 
4956 /// EvalAddr - EvalAddr and EvalVal are mutually recursive functions that
4957 ///  check if the expression in a return statement evaluates to an address
4958 ///  to a location on the stack, a local block, an address of a label, or a
4959 ///  reference to local temporary. The recursion is used to traverse the
4960 ///  AST of the return expression, with recursion backtracking when we
4961 ///  encounter a subexpression that (1) clearly does not lead to one of the
4962 ///  above problematic expressions (2) is something we cannot determine leads to
4963 ///  a problematic expression based on such local checking.
4964 ///
4965 ///  Both EvalAddr and EvalVal follow through reference variables to evaluate
4966 ///  the expression that they point to. Such variables are added to the
4967 ///  'refVars' vector so that we know what the reference variable "trail" was.
4968 ///
4969 ///  EvalAddr processes expressions that are pointers that are used as
4970 ///  references (and not L-values).  EvalVal handles all other values.
4971 ///  At the base case of the recursion is a check for the above problematic
4972 ///  expressions.
4973 ///
4974 ///  This implementation handles:
4975 ///
4976 ///   * pointer-to-pointer casts
4977 ///   * implicit conversions from array references to pointers
4978 ///   * taking the address of fields
4979 ///   * arbitrary interplay between "&" and "*" operators
4980 ///   * pointer arithmetic from an address of a stack variable
4981 ///   * taking the address of an array element where the array is on the stack
EvalAddr(Expr * E,SmallVectorImpl<DeclRefExpr * > & refVars,Decl * ParentDecl)4982 static Expr *EvalAddr(Expr *E, SmallVectorImpl<DeclRefExpr *> &refVars,
4983                       Decl *ParentDecl) {
4984   if (E->isTypeDependent())
4985     return nullptr;
4986 
4987   // We should only be called for evaluating pointer expressions.
4988   assert((E->getType()->isAnyPointerType() ||
4989           E->getType()->isBlockPointerType() ||
4990           E->getType()->isObjCQualifiedIdType()) &&
4991          "EvalAddr only works on pointers");
4992 
4993   E = E->IgnoreParens();
4994 
4995   // Our "symbolic interpreter" is just a dispatch off the currently
4996   // viewed AST node.  We then recursively traverse the AST by calling
4997   // EvalAddr and EvalVal appropriately.
4998   switch (E->getStmtClass()) {
4999   case Stmt::DeclRefExprClass: {
5000     DeclRefExpr *DR = cast<DeclRefExpr>(E);
5001 
5002     // If we leave the immediate function, the lifetime isn't about to end.
5003     if (DR->refersToEnclosingVariableOrCapture())
5004       return nullptr;
5005 
5006     if (VarDecl *V = dyn_cast<VarDecl>(DR->getDecl()))
5007       // If this is a reference variable, follow through to the expression that
5008       // it points to.
5009       if (V->hasLocalStorage() &&
5010           V->getType()->isReferenceType() && V->hasInit()) {
5011         // Add the reference variable to the "trail".
5012         refVars.push_back(DR);
5013         return EvalAddr(V->getInit(), refVars, ParentDecl);
5014       }
5015 
5016     return nullptr;
5017   }
5018 
5019   case Stmt::UnaryOperatorClass: {
5020     // The only unary operator that make sense to handle here
5021     // is AddrOf.  All others don't make sense as pointers.
5022     UnaryOperator *U = cast<UnaryOperator>(E);
5023 
5024     if (U->getOpcode() == UO_AddrOf)
5025       return EvalVal(U->getSubExpr(), refVars, ParentDecl);
5026     else
5027       return nullptr;
5028   }
5029 
5030   case Stmt::BinaryOperatorClass: {
5031     // Handle pointer arithmetic.  All other binary operators are not valid
5032     // in this context.
5033     BinaryOperator *B = cast<BinaryOperator>(E);
5034     BinaryOperatorKind op = B->getOpcode();
5035 
5036     if (op != BO_Add && op != BO_Sub)
5037       return nullptr;
5038 
5039     Expr *Base = B->getLHS();
5040 
5041     // Determine which argument is the real pointer base.  It could be
5042     // the RHS argument instead of the LHS.
5043     if (!Base->getType()->isPointerType()) Base = B->getRHS();
5044 
5045     assert (Base->getType()->isPointerType());
5046     return EvalAddr(Base, refVars, ParentDecl);
5047   }
5048 
5049   // For conditional operators we need to see if either the LHS or RHS are
5050   // valid DeclRefExpr*s.  If one of them is valid, we return it.
5051   case Stmt::ConditionalOperatorClass: {
5052     ConditionalOperator *C = cast<ConditionalOperator>(E);
5053 
5054     // Handle the GNU extension for missing LHS.
5055     // FIXME: That isn't a ConditionalOperator, so doesn't get here.
5056     if (Expr *LHSExpr = C->getLHS()) {
5057       // In C++, we can have a throw-expression, which has 'void' type.
5058       if (!LHSExpr->getType()->isVoidType())
5059         if (Expr *LHS = EvalAddr(LHSExpr, refVars, ParentDecl))
5060           return LHS;
5061     }
5062 
5063     // In C++, we can have a throw-expression, which has 'void' type.
5064     if (C->getRHS()->getType()->isVoidType())
5065       return nullptr;
5066 
5067     return EvalAddr(C->getRHS(), refVars, ParentDecl);
5068   }
5069 
5070   case Stmt::BlockExprClass:
5071     if (cast<BlockExpr>(E)->getBlockDecl()->hasCaptures())
5072       return E; // local block.
5073     return nullptr;
5074 
5075   case Stmt::AddrLabelExprClass:
5076     return E; // address of label.
5077 
5078   case Stmt::ExprWithCleanupsClass:
5079     return EvalAddr(cast<ExprWithCleanups>(E)->getSubExpr(), refVars,
5080                     ParentDecl);
5081 
5082   // For casts, we need to handle conversions from arrays to
5083   // pointer values, and pointer-to-pointer conversions.
5084   case Stmt::ImplicitCastExprClass:
5085   case Stmt::CStyleCastExprClass:
5086   case Stmt::CXXFunctionalCastExprClass:
5087   case Stmt::ObjCBridgedCastExprClass:
5088   case Stmt::CXXStaticCastExprClass:
5089   case Stmt::CXXDynamicCastExprClass:
5090   case Stmt::CXXConstCastExprClass:
5091   case Stmt::CXXReinterpretCastExprClass: {
5092     Expr* SubExpr = cast<CastExpr>(E)->getSubExpr();
5093     switch (cast<CastExpr>(E)->getCastKind()) {
5094     case CK_LValueToRValue:
5095     case CK_NoOp:
5096     case CK_BaseToDerived:
5097     case CK_DerivedToBase:
5098     case CK_UncheckedDerivedToBase:
5099     case CK_Dynamic:
5100     case CK_CPointerToObjCPointerCast:
5101     case CK_BlockPointerToObjCPointerCast:
5102     case CK_AnyPointerToBlockPointerCast:
5103       return EvalAddr(SubExpr, refVars, ParentDecl);
5104 
5105     case CK_ArrayToPointerDecay:
5106       return EvalVal(SubExpr, refVars, ParentDecl);
5107 
5108     case CK_BitCast:
5109       if (SubExpr->getType()->isAnyPointerType() ||
5110           SubExpr->getType()->isBlockPointerType() ||
5111           SubExpr->getType()->isObjCQualifiedIdType())
5112         return EvalAddr(SubExpr, refVars, ParentDecl);
5113       else
5114         return nullptr;
5115 
5116     default:
5117       return nullptr;
5118     }
5119   }
5120 
5121   case Stmt::MaterializeTemporaryExprClass:
5122     if (Expr *Result = EvalAddr(
5123                          cast<MaterializeTemporaryExpr>(E)->GetTemporaryExpr(),
5124                                 refVars, ParentDecl))
5125       return Result;
5126 
5127     return E;
5128 
5129   // Everything else: we simply don't reason about them.
5130   default:
5131     return nullptr;
5132   }
5133 }
5134 
5135 
5136 ///  EvalVal - This function is complements EvalAddr in the mutual recursion.
5137 ///   See the comments for EvalAddr for more details.
EvalVal(Expr * E,SmallVectorImpl<DeclRefExpr * > & refVars,Decl * ParentDecl)5138 static Expr *EvalVal(Expr *E, SmallVectorImpl<DeclRefExpr *> &refVars,
5139                      Decl *ParentDecl) {
5140 do {
5141   // We should only be called for evaluating non-pointer expressions, or
5142   // expressions with a pointer type that are not used as references but instead
5143   // are l-values (e.g., DeclRefExpr with a pointer type).
5144 
5145   // Our "symbolic interpreter" is just a dispatch off the currently
5146   // viewed AST node.  We then recursively traverse the AST by calling
5147   // EvalAddr and EvalVal appropriately.
5148 
5149   E = E->IgnoreParens();
5150   switch (E->getStmtClass()) {
5151   case Stmt::ImplicitCastExprClass: {
5152     ImplicitCastExpr *IE = cast<ImplicitCastExpr>(E);
5153     if (IE->getValueKind() == VK_LValue) {
5154       E = IE->getSubExpr();
5155       continue;
5156     }
5157     return nullptr;
5158   }
5159 
5160   case Stmt::ExprWithCleanupsClass:
5161     return EvalVal(cast<ExprWithCleanups>(E)->getSubExpr(), refVars,ParentDecl);
5162 
5163   case Stmt::DeclRefExprClass: {
5164     // When we hit a DeclRefExpr we are looking at code that refers to a
5165     // variable's name. If it's not a reference variable we check if it has
5166     // local storage within the function, and if so, return the expression.
5167     DeclRefExpr *DR = cast<DeclRefExpr>(E);
5168 
5169     // If we leave the immediate function, the lifetime isn't about to end.
5170     if (DR->refersToEnclosingVariableOrCapture())
5171       return nullptr;
5172 
5173     if (VarDecl *V = dyn_cast<VarDecl>(DR->getDecl())) {
5174       // Check if it refers to itself, e.g. "int& i = i;".
5175       if (V == ParentDecl)
5176         return DR;
5177 
5178       if (V->hasLocalStorage()) {
5179         if (!V->getType()->isReferenceType())
5180           return DR;
5181 
5182         // Reference variable, follow through to the expression that
5183         // it points to.
5184         if (V->hasInit()) {
5185           // Add the reference variable to the "trail".
5186           refVars.push_back(DR);
5187           return EvalVal(V->getInit(), refVars, V);
5188         }
5189       }
5190     }
5191 
5192     return nullptr;
5193   }
5194 
5195   case Stmt::UnaryOperatorClass: {
5196     // The only unary operator that make sense to handle here
5197     // is Deref.  All others don't resolve to a "name."  This includes
5198     // handling all sorts of rvalues passed to a unary operator.
5199     UnaryOperator *U = cast<UnaryOperator>(E);
5200 
5201     if (U->getOpcode() == UO_Deref)
5202       return EvalAddr(U->getSubExpr(), refVars, ParentDecl);
5203 
5204     return nullptr;
5205   }
5206 
5207   case Stmt::ArraySubscriptExprClass: {
5208     // Array subscripts are potential references to data on the stack.  We
5209     // retrieve the DeclRefExpr* for the array variable if it indeed
5210     // has local storage.
5211     return EvalAddr(cast<ArraySubscriptExpr>(E)->getBase(), refVars,ParentDecl);
5212   }
5213 
5214   case Stmt::ConditionalOperatorClass: {
5215     // For conditional operators we need to see if either the LHS or RHS are
5216     // non-NULL Expr's.  If one is non-NULL, we return it.
5217     ConditionalOperator *C = cast<ConditionalOperator>(E);
5218 
5219     // Handle the GNU extension for missing LHS.
5220     if (Expr *LHSExpr = C->getLHS()) {
5221       // In C++, we can have a throw-expression, which has 'void' type.
5222       if (!LHSExpr->getType()->isVoidType())
5223         if (Expr *LHS = EvalVal(LHSExpr, refVars, ParentDecl))
5224           return LHS;
5225     }
5226 
5227     // In C++, we can have a throw-expression, which has 'void' type.
5228     if (C->getRHS()->getType()->isVoidType())
5229       return nullptr;
5230 
5231     return EvalVal(C->getRHS(), refVars, ParentDecl);
5232   }
5233 
5234   // Accesses to members are potential references to data on the stack.
5235   case Stmt::MemberExprClass: {
5236     MemberExpr *M = cast<MemberExpr>(E);
5237 
5238     // Check for indirect access.  We only want direct field accesses.
5239     if (M->isArrow())
5240       return nullptr;
5241 
5242     // Check whether the member type is itself a reference, in which case
5243     // we're not going to refer to the member, but to what the member refers to.
5244     if (M->getMemberDecl()->getType()->isReferenceType())
5245       return nullptr;
5246 
5247     return EvalVal(M->getBase(), refVars, ParentDecl);
5248   }
5249 
5250   case Stmt::MaterializeTemporaryExprClass:
5251     if (Expr *Result = EvalVal(
5252                           cast<MaterializeTemporaryExpr>(E)->GetTemporaryExpr(),
5253                                refVars, ParentDecl))
5254       return Result;
5255 
5256     return E;
5257 
5258   default:
5259     // Check that we don't return or take the address of a reference to a
5260     // temporary. This is only useful in C++.
5261     if (!E->isTypeDependent() && E->isRValue())
5262       return E;
5263 
5264     // Everything else: we simply don't reason about them.
5265     return nullptr;
5266   }
5267 } while (true);
5268 }
5269 
5270 void
CheckReturnValExpr(Expr * RetValExp,QualType lhsType,SourceLocation ReturnLoc,bool isObjCMethod,const AttrVec * Attrs,const FunctionDecl * FD)5271 Sema::CheckReturnValExpr(Expr *RetValExp, QualType lhsType,
5272                          SourceLocation ReturnLoc,
5273                          bool isObjCMethod,
5274                          const AttrVec *Attrs,
5275                          const FunctionDecl *FD) {
5276   CheckReturnStackAddr(*this, RetValExp, lhsType, ReturnLoc);
5277 
5278   // Check if the return value is null but should not be.
5279   if (Attrs && hasSpecificAttr<ReturnsNonNullAttr>(*Attrs) &&
5280       CheckNonNullExpr(*this, RetValExp))
5281     Diag(ReturnLoc, diag::warn_null_ret)
5282       << (isObjCMethod ? 1 : 0) << RetValExp->getSourceRange();
5283 
5284   // C++11 [basic.stc.dynamic.allocation]p4:
5285   //   If an allocation function declared with a non-throwing
5286   //   exception-specification fails to allocate storage, it shall return
5287   //   a null pointer. Any other allocation function that fails to allocate
5288   //   storage shall indicate failure only by throwing an exception [...]
5289   if (FD) {
5290     OverloadedOperatorKind Op = FD->getOverloadedOperator();
5291     if (Op == OO_New || Op == OO_Array_New) {
5292       const FunctionProtoType *Proto
5293         = FD->getType()->castAs<FunctionProtoType>();
5294       if (!Proto->isNothrow(Context, /*ResultIfDependent*/true) &&
5295           CheckNonNullExpr(*this, RetValExp))
5296         Diag(ReturnLoc, diag::warn_operator_new_returns_null)
5297           << FD << getLangOpts().CPlusPlus11;
5298     }
5299   }
5300 }
5301 
5302 //===--- CHECK: Floating-Point comparisons (-Wfloat-equal) ---------------===//
5303 
5304 /// Check for comparisons of floating point operands using != and ==.
5305 /// Issue a warning if these are no self-comparisons, as they are not likely
5306 /// to do what the programmer intended.
CheckFloatComparison(SourceLocation Loc,Expr * LHS,Expr * RHS)5307 void Sema::CheckFloatComparison(SourceLocation Loc, Expr* LHS, Expr *RHS) {
5308   Expr* LeftExprSansParen = LHS->IgnoreParenImpCasts();
5309   Expr* RightExprSansParen = RHS->IgnoreParenImpCasts();
5310 
5311   // Special case: check for x == x (which is OK).
5312   // Do not emit warnings for such cases.
5313   if (DeclRefExpr* DRL = dyn_cast<DeclRefExpr>(LeftExprSansParen))
5314     if (DeclRefExpr* DRR = dyn_cast<DeclRefExpr>(RightExprSansParen))
5315       if (DRL->getDecl() == DRR->getDecl())
5316         return;
5317 
5318 
5319   // Special case: check for comparisons against literals that can be exactly
5320   //  represented by APFloat.  In such cases, do not emit a warning.  This
5321   //  is a heuristic: often comparison against such literals are used to
5322   //  detect if a value in a variable has not changed.  This clearly can
5323   //  lead to false negatives.
5324   if (FloatingLiteral* FLL = dyn_cast<FloatingLiteral>(LeftExprSansParen)) {
5325     if (FLL->isExact())
5326       return;
5327   } else
5328     if (FloatingLiteral* FLR = dyn_cast<FloatingLiteral>(RightExprSansParen))
5329       if (FLR->isExact())
5330         return;
5331 
5332   // Check for comparisons with builtin types.
5333   if (CallExpr* CL = dyn_cast<CallExpr>(LeftExprSansParen))
5334     if (CL->getBuiltinCallee())
5335       return;
5336 
5337   if (CallExpr* CR = dyn_cast<CallExpr>(RightExprSansParen))
5338     if (CR->getBuiltinCallee())
5339       return;
5340 
5341   // Emit the diagnostic.
5342   Diag(Loc, diag::warn_floatingpoint_eq)
5343     << LHS->getSourceRange() << RHS->getSourceRange();
5344 }
5345 
5346 //===--- CHECK: Integer mixed-sign comparisons (-Wsign-compare) --------===//
5347 //===--- CHECK: Lossy implicit conversions (-Wconversion) --------------===//
5348 
5349 namespace {
5350 
5351 /// Structure recording the 'active' range of an integer-valued
5352 /// expression.
5353 struct IntRange {
5354   /// The number of bits active in the int.
5355   unsigned Width;
5356 
5357   /// True if the int is known not to have negative values.
5358   bool NonNegative;
5359 
IntRange__anonb40dc79c0711::IntRange5360   IntRange(unsigned Width, bool NonNegative)
5361     : Width(Width), NonNegative(NonNegative)
5362   {}
5363 
5364   /// Returns the range of the bool type.
forBoolType__anonb40dc79c0711::IntRange5365   static IntRange forBoolType() {
5366     return IntRange(1, true);
5367   }
5368 
5369   /// Returns the range of an opaque value of the given integral type.
forValueOfType__anonb40dc79c0711::IntRange5370   static IntRange forValueOfType(ASTContext &C, QualType T) {
5371     return forValueOfCanonicalType(C,
5372                           T->getCanonicalTypeInternal().getTypePtr());
5373   }
5374 
5375   /// Returns the range of an opaque value of a canonical integral type.
forValueOfCanonicalType__anonb40dc79c0711::IntRange5376   static IntRange forValueOfCanonicalType(ASTContext &C, const Type *T) {
5377     assert(T->isCanonicalUnqualified());
5378 
5379     if (const VectorType *VT = dyn_cast<VectorType>(T))
5380       T = VT->getElementType().getTypePtr();
5381     if (const ComplexType *CT = dyn_cast<ComplexType>(T))
5382       T = CT->getElementType().getTypePtr();
5383     if (const AtomicType *AT = dyn_cast<AtomicType>(T))
5384       T = AT->getValueType().getTypePtr();
5385 
5386     // For enum types, use the known bit width of the enumerators.
5387     if (const EnumType *ET = dyn_cast<EnumType>(T)) {
5388       EnumDecl *Enum = ET->getDecl();
5389       if (!Enum->isCompleteDefinition())
5390         return IntRange(C.getIntWidth(QualType(T, 0)), false);
5391 
5392       unsigned NumPositive = Enum->getNumPositiveBits();
5393       unsigned NumNegative = Enum->getNumNegativeBits();
5394 
5395       if (NumNegative == 0)
5396         return IntRange(NumPositive, true/*NonNegative*/);
5397       else
5398         return IntRange(std::max(NumPositive + 1, NumNegative),
5399                         false/*NonNegative*/);
5400     }
5401 
5402     const BuiltinType *BT = cast<BuiltinType>(T);
5403     assert(BT->isInteger());
5404 
5405     return IntRange(C.getIntWidth(QualType(T, 0)), BT->isUnsignedInteger());
5406   }
5407 
5408   /// Returns the "target" range of a canonical integral type, i.e.
5409   /// the range of values expressible in the type.
5410   ///
5411   /// This matches forValueOfCanonicalType except that enums have the
5412   /// full range of their type, not the range of their enumerators.
forTargetOfCanonicalType__anonb40dc79c0711::IntRange5413   static IntRange forTargetOfCanonicalType(ASTContext &C, const Type *T) {
5414     assert(T->isCanonicalUnqualified());
5415 
5416     if (const VectorType *VT = dyn_cast<VectorType>(T))
5417       T = VT->getElementType().getTypePtr();
5418     if (const ComplexType *CT = dyn_cast<ComplexType>(T))
5419       T = CT->getElementType().getTypePtr();
5420     if (const AtomicType *AT = dyn_cast<AtomicType>(T))
5421       T = AT->getValueType().getTypePtr();
5422     if (const EnumType *ET = dyn_cast<EnumType>(T))
5423       T = C.getCanonicalType(ET->getDecl()->getIntegerType()).getTypePtr();
5424 
5425     const BuiltinType *BT = cast<BuiltinType>(T);
5426     assert(BT->isInteger());
5427 
5428     return IntRange(C.getIntWidth(QualType(T, 0)), BT->isUnsignedInteger());
5429   }
5430 
5431   /// Returns the supremum of two ranges: i.e. their conservative merge.
join__anonb40dc79c0711::IntRange5432   static IntRange join(IntRange L, IntRange R) {
5433     return IntRange(std::max(L.Width, R.Width),
5434                     L.NonNegative && R.NonNegative);
5435   }
5436 
5437   /// Returns the infinum of two ranges: i.e. their aggressive merge.
meet__anonb40dc79c0711::IntRange5438   static IntRange meet(IntRange L, IntRange R) {
5439     return IntRange(std::min(L.Width, R.Width),
5440                     L.NonNegative || R.NonNegative);
5441   }
5442 };
5443 
GetValueRange(ASTContext & C,llvm::APSInt & value,unsigned MaxWidth)5444 static IntRange GetValueRange(ASTContext &C, llvm::APSInt &value,
5445                               unsigned MaxWidth) {
5446   if (value.isSigned() && value.isNegative())
5447     return IntRange(value.getMinSignedBits(), false);
5448 
5449   if (value.getBitWidth() > MaxWidth)
5450     value = value.trunc(MaxWidth);
5451 
5452   // isNonNegative() just checks the sign bit without considering
5453   // signedness.
5454   return IntRange(value.getActiveBits(), true);
5455 }
5456 
GetValueRange(ASTContext & C,APValue & result,QualType Ty,unsigned MaxWidth)5457 static IntRange GetValueRange(ASTContext &C, APValue &result, QualType Ty,
5458                               unsigned MaxWidth) {
5459   if (result.isInt())
5460     return GetValueRange(C, result.getInt(), MaxWidth);
5461 
5462   if (result.isVector()) {
5463     IntRange R = GetValueRange(C, result.getVectorElt(0), Ty, MaxWidth);
5464     for (unsigned i = 1, e = result.getVectorLength(); i != e; ++i) {
5465       IntRange El = GetValueRange(C, result.getVectorElt(i), Ty, MaxWidth);
5466       R = IntRange::join(R, El);
5467     }
5468     return R;
5469   }
5470 
5471   if (result.isComplexInt()) {
5472     IntRange R = GetValueRange(C, result.getComplexIntReal(), MaxWidth);
5473     IntRange I = GetValueRange(C, result.getComplexIntImag(), MaxWidth);
5474     return IntRange::join(R, I);
5475   }
5476 
5477   // This can happen with lossless casts to intptr_t of "based" lvalues.
5478   // Assume it might use arbitrary bits.
5479   // FIXME: The only reason we need to pass the type in here is to get
5480   // the sign right on this one case.  It would be nice if APValue
5481   // preserved this.
5482   assert(result.isLValue() || result.isAddrLabelDiff());
5483   return IntRange(MaxWidth, Ty->isUnsignedIntegerOrEnumerationType());
5484 }
5485 
GetExprType(Expr * E)5486 static QualType GetExprType(Expr *E) {
5487   QualType Ty = E->getType();
5488   if (const AtomicType *AtomicRHS = Ty->getAs<AtomicType>())
5489     Ty = AtomicRHS->getValueType();
5490   return Ty;
5491 }
5492 
5493 /// Pseudo-evaluate the given integer expression, estimating the
5494 /// range of values it might take.
5495 ///
5496 /// \param MaxWidth - the width to which the value will be truncated
GetExprRange(ASTContext & C,Expr * E,unsigned MaxWidth)5497 static IntRange GetExprRange(ASTContext &C, Expr *E, unsigned MaxWidth) {
5498   E = E->IgnoreParens();
5499 
5500   // Try a full evaluation first.
5501   Expr::EvalResult result;
5502   if (E->EvaluateAsRValue(result, C))
5503     return GetValueRange(C, result.Val, GetExprType(E), MaxWidth);
5504 
5505   // I think we only want to look through implicit casts here; if the
5506   // user has an explicit widening cast, we should treat the value as
5507   // being of the new, wider type.
5508   if (ImplicitCastExpr *CE = dyn_cast<ImplicitCastExpr>(E)) {
5509     if (CE->getCastKind() == CK_NoOp || CE->getCastKind() == CK_LValueToRValue)
5510       return GetExprRange(C, CE->getSubExpr(), MaxWidth);
5511 
5512     IntRange OutputTypeRange = IntRange::forValueOfType(C, GetExprType(CE));
5513 
5514     bool isIntegerCast = (CE->getCastKind() == CK_IntegralCast);
5515 
5516     // Assume that non-integer casts can span the full range of the type.
5517     if (!isIntegerCast)
5518       return OutputTypeRange;
5519 
5520     IntRange SubRange
5521       = GetExprRange(C, CE->getSubExpr(),
5522                      std::min(MaxWidth, OutputTypeRange.Width));
5523 
5524     // Bail out if the subexpr's range is as wide as the cast type.
5525     if (SubRange.Width >= OutputTypeRange.Width)
5526       return OutputTypeRange;
5527 
5528     // Otherwise, we take the smaller width, and we're non-negative if
5529     // either the output type or the subexpr is.
5530     return IntRange(SubRange.Width,
5531                     SubRange.NonNegative || OutputTypeRange.NonNegative);
5532   }
5533 
5534   if (ConditionalOperator *CO = dyn_cast<ConditionalOperator>(E)) {
5535     // If we can fold the condition, just take that operand.
5536     bool CondResult;
5537     if (CO->getCond()->EvaluateAsBooleanCondition(CondResult, C))
5538       return GetExprRange(C, CondResult ? CO->getTrueExpr()
5539                                         : CO->getFalseExpr(),
5540                           MaxWidth);
5541 
5542     // Otherwise, conservatively merge.
5543     IntRange L = GetExprRange(C, CO->getTrueExpr(), MaxWidth);
5544     IntRange R = GetExprRange(C, CO->getFalseExpr(), MaxWidth);
5545     return IntRange::join(L, R);
5546   }
5547 
5548   if (BinaryOperator *BO = dyn_cast<BinaryOperator>(E)) {
5549     switch (BO->getOpcode()) {
5550 
5551     // Boolean-valued operations are single-bit and positive.
5552     case BO_LAnd:
5553     case BO_LOr:
5554     case BO_LT:
5555     case BO_GT:
5556     case BO_LE:
5557     case BO_GE:
5558     case BO_EQ:
5559     case BO_NE:
5560       return IntRange::forBoolType();
5561 
5562     // The type of the assignments is the type of the LHS, so the RHS
5563     // is not necessarily the same type.
5564     case BO_MulAssign:
5565     case BO_DivAssign:
5566     case BO_RemAssign:
5567     case BO_AddAssign:
5568     case BO_SubAssign:
5569     case BO_XorAssign:
5570     case BO_OrAssign:
5571       // TODO: bitfields?
5572       return IntRange::forValueOfType(C, GetExprType(E));
5573 
5574     // Simple assignments just pass through the RHS, which will have
5575     // been coerced to the LHS type.
5576     case BO_Assign:
5577       // TODO: bitfields?
5578       return GetExprRange(C, BO->getRHS(), MaxWidth);
5579 
5580     // Operations with opaque sources are black-listed.
5581     case BO_PtrMemD:
5582     case BO_PtrMemI:
5583       return IntRange::forValueOfType(C, GetExprType(E));
5584 
5585     // Bitwise-and uses the *infinum* of the two source ranges.
5586     case BO_And:
5587     case BO_AndAssign:
5588       return IntRange::meet(GetExprRange(C, BO->getLHS(), MaxWidth),
5589                             GetExprRange(C, BO->getRHS(), MaxWidth));
5590 
5591     // Left shift gets black-listed based on a judgement call.
5592     case BO_Shl:
5593       // ...except that we want to treat '1 << (blah)' as logically
5594       // positive.  It's an important idiom.
5595       if (IntegerLiteral *I
5596             = dyn_cast<IntegerLiteral>(BO->getLHS()->IgnoreParenCasts())) {
5597         if (I->getValue() == 1) {
5598           IntRange R = IntRange::forValueOfType(C, GetExprType(E));
5599           return IntRange(R.Width, /*NonNegative*/ true);
5600         }
5601       }
5602       // fallthrough
5603 
5604     case BO_ShlAssign:
5605       return IntRange::forValueOfType(C, GetExprType(E));
5606 
5607     // Right shift by a constant can narrow its left argument.
5608     case BO_Shr:
5609     case BO_ShrAssign: {
5610       IntRange L = GetExprRange(C, BO->getLHS(), MaxWidth);
5611 
5612       // If the shift amount is a positive constant, drop the width by
5613       // that much.
5614       llvm::APSInt shift;
5615       if (BO->getRHS()->isIntegerConstantExpr(shift, C) &&
5616           shift.isNonNegative()) {
5617         unsigned zext = shift.getZExtValue();
5618         if (zext >= L.Width)
5619           L.Width = (L.NonNegative ? 0 : 1);
5620         else
5621           L.Width -= zext;
5622       }
5623 
5624       return L;
5625     }
5626 
5627     // Comma acts as its right operand.
5628     case BO_Comma:
5629       return GetExprRange(C, BO->getRHS(), MaxWidth);
5630 
5631     // Black-list pointer subtractions.
5632     case BO_Sub:
5633       if (BO->getLHS()->getType()->isPointerType())
5634         return IntRange::forValueOfType(C, GetExprType(E));
5635       break;
5636 
5637     // The width of a division result is mostly determined by the size
5638     // of the LHS.
5639     case BO_Div: {
5640       // Don't 'pre-truncate' the operands.
5641       unsigned opWidth = C.getIntWidth(GetExprType(E));
5642       IntRange L = GetExprRange(C, BO->getLHS(), opWidth);
5643 
5644       // If the divisor is constant, use that.
5645       llvm::APSInt divisor;
5646       if (BO->getRHS()->isIntegerConstantExpr(divisor, C)) {
5647         unsigned log2 = divisor.logBase2(); // floor(log_2(divisor))
5648         if (log2 >= L.Width)
5649           L.Width = (L.NonNegative ? 0 : 1);
5650         else
5651           L.Width = std::min(L.Width - log2, MaxWidth);
5652         return L;
5653       }
5654 
5655       // Otherwise, just use the LHS's width.
5656       IntRange R = GetExprRange(C, BO->getRHS(), opWidth);
5657       return IntRange(L.Width, L.NonNegative && R.NonNegative);
5658     }
5659 
5660     // The result of a remainder can't be larger than the result of
5661     // either side.
5662     case BO_Rem: {
5663       // Don't 'pre-truncate' the operands.
5664       unsigned opWidth = C.getIntWidth(GetExprType(E));
5665       IntRange L = GetExprRange(C, BO->getLHS(), opWidth);
5666       IntRange R = GetExprRange(C, BO->getRHS(), opWidth);
5667 
5668       IntRange meet = IntRange::meet(L, R);
5669       meet.Width = std::min(meet.Width, MaxWidth);
5670       return meet;
5671     }
5672 
5673     // The default behavior is okay for these.
5674     case BO_Mul:
5675     case BO_Add:
5676     case BO_Xor:
5677     case BO_Or:
5678       break;
5679     }
5680 
5681     // The default case is to treat the operation as if it were closed
5682     // on the narrowest type that encompasses both operands.
5683     IntRange L = GetExprRange(C, BO->getLHS(), MaxWidth);
5684     IntRange R = GetExprRange(C, BO->getRHS(), MaxWidth);
5685     return IntRange::join(L, R);
5686   }
5687 
5688   if (UnaryOperator *UO = dyn_cast<UnaryOperator>(E)) {
5689     switch (UO->getOpcode()) {
5690     // Boolean-valued operations are white-listed.
5691     case UO_LNot:
5692       return IntRange::forBoolType();
5693 
5694     // Operations with opaque sources are black-listed.
5695     case UO_Deref:
5696     case UO_AddrOf: // should be impossible
5697       return IntRange::forValueOfType(C, GetExprType(E));
5698 
5699     default:
5700       return GetExprRange(C, UO->getSubExpr(), MaxWidth);
5701     }
5702   }
5703 
5704   if (OpaqueValueExpr *OVE = dyn_cast<OpaqueValueExpr>(E))
5705     return GetExprRange(C, OVE->getSourceExpr(), MaxWidth);
5706 
5707   if (FieldDecl *BitField = E->getSourceBitField())
5708     return IntRange(BitField->getBitWidthValue(C),
5709                     BitField->getType()->isUnsignedIntegerOrEnumerationType());
5710 
5711   return IntRange::forValueOfType(C, GetExprType(E));
5712 }
5713 
GetExprRange(ASTContext & C,Expr * E)5714 static IntRange GetExprRange(ASTContext &C, Expr *E) {
5715   return GetExprRange(C, E, C.getIntWidth(GetExprType(E)));
5716 }
5717 
5718 /// Checks whether the given value, which currently has the given
5719 /// source semantics, has the same value when coerced through the
5720 /// target semantics.
IsSameFloatAfterCast(const llvm::APFloat & value,const llvm::fltSemantics & Src,const llvm::fltSemantics & Tgt)5721 static bool IsSameFloatAfterCast(const llvm::APFloat &value,
5722                                  const llvm::fltSemantics &Src,
5723                                  const llvm::fltSemantics &Tgt) {
5724   llvm::APFloat truncated = value;
5725 
5726   bool ignored;
5727   truncated.convert(Src, llvm::APFloat::rmNearestTiesToEven, &ignored);
5728   truncated.convert(Tgt, llvm::APFloat::rmNearestTiesToEven, &ignored);
5729 
5730   return truncated.bitwiseIsEqual(value);
5731 }
5732 
5733 /// Checks whether the given value, which currently has the given
5734 /// source semantics, has the same value when coerced through the
5735 /// target semantics.
5736 ///
5737 /// The value might be a vector of floats (or a complex number).
IsSameFloatAfterCast(const APValue & value,const llvm::fltSemantics & Src,const llvm::fltSemantics & Tgt)5738 static bool IsSameFloatAfterCast(const APValue &value,
5739                                  const llvm::fltSemantics &Src,
5740                                  const llvm::fltSemantics &Tgt) {
5741   if (value.isFloat())
5742     return IsSameFloatAfterCast(value.getFloat(), Src, Tgt);
5743 
5744   if (value.isVector()) {
5745     for (unsigned i = 0, e = value.getVectorLength(); i != e; ++i)
5746       if (!IsSameFloatAfterCast(value.getVectorElt(i), Src, Tgt))
5747         return false;
5748     return true;
5749   }
5750 
5751   assert(value.isComplexFloat());
5752   return (IsSameFloatAfterCast(value.getComplexFloatReal(), Src, Tgt) &&
5753           IsSameFloatAfterCast(value.getComplexFloatImag(), Src, Tgt));
5754 }
5755 
5756 static void AnalyzeImplicitConversions(Sema &S, Expr *E, SourceLocation CC);
5757 
IsZero(Sema & S,Expr * E)5758 static bool IsZero(Sema &S, Expr *E) {
5759   // Suppress cases where we are comparing against an enum constant.
5760   if (const DeclRefExpr *DR =
5761       dyn_cast<DeclRefExpr>(E->IgnoreParenImpCasts()))
5762     if (isa<EnumConstantDecl>(DR->getDecl()))
5763       return false;
5764 
5765   // Suppress cases where the '0' value is expanded from a macro.
5766   if (E->getLocStart().isMacroID())
5767     return false;
5768 
5769   llvm::APSInt Value;
5770   return E->isIntegerConstantExpr(Value, S.Context) && Value == 0;
5771 }
5772 
HasEnumType(Expr * E)5773 static bool HasEnumType(Expr *E) {
5774   // Strip off implicit integral promotions.
5775   while (ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(E)) {
5776     if (ICE->getCastKind() != CK_IntegralCast &&
5777         ICE->getCastKind() != CK_NoOp)
5778       break;
5779     E = ICE->getSubExpr();
5780   }
5781 
5782   return E->getType()->isEnumeralType();
5783 }
5784 
CheckTrivialUnsignedComparison(Sema & S,BinaryOperator * E)5785 static void CheckTrivialUnsignedComparison(Sema &S, BinaryOperator *E) {
5786   // Disable warning in template instantiations.
5787   if (!S.ActiveTemplateInstantiations.empty())
5788     return;
5789 
5790   BinaryOperatorKind op = E->getOpcode();
5791   if (E->isValueDependent())
5792     return;
5793 
5794   if (op == BO_LT && IsZero(S, E->getRHS())) {
5795     S.Diag(E->getOperatorLoc(), diag::warn_lunsigned_always_true_comparison)
5796       << "< 0" << "false" << HasEnumType(E->getLHS())
5797       << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange();
5798   } else if (op == BO_GE && IsZero(S, E->getRHS())) {
5799     S.Diag(E->getOperatorLoc(), diag::warn_lunsigned_always_true_comparison)
5800       << ">= 0" << "true" << HasEnumType(E->getLHS())
5801       << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange();
5802   } else if (op == BO_GT && IsZero(S, E->getLHS())) {
5803     S.Diag(E->getOperatorLoc(), diag::warn_runsigned_always_true_comparison)
5804       << "0 >" << "false" << HasEnumType(E->getRHS())
5805       << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange();
5806   } else if (op == BO_LE && IsZero(S, E->getLHS())) {
5807     S.Diag(E->getOperatorLoc(), diag::warn_runsigned_always_true_comparison)
5808       << "0 <=" << "true" << HasEnumType(E->getRHS())
5809       << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange();
5810   }
5811 }
5812 
DiagnoseOutOfRangeComparison(Sema & S,BinaryOperator * E,Expr * Constant,Expr * Other,llvm::APSInt Value,bool RhsConstant)5813 static void DiagnoseOutOfRangeComparison(Sema &S, BinaryOperator *E,
5814                                          Expr *Constant, Expr *Other,
5815                                          llvm::APSInt Value,
5816                                          bool RhsConstant) {
5817   // Disable warning in template instantiations.
5818   if (!S.ActiveTemplateInstantiations.empty())
5819     return;
5820 
5821   // TODO: Investigate using GetExprRange() to get tighter bounds
5822   // on the bit ranges.
5823   QualType OtherT = Other->getType();
5824   if (const AtomicType *AT = dyn_cast<AtomicType>(OtherT))
5825     OtherT = AT->getValueType();
5826   IntRange OtherRange = IntRange::forValueOfType(S.Context, OtherT);
5827   unsigned OtherWidth = OtherRange.Width;
5828 
5829   bool OtherIsBooleanType = Other->isKnownToHaveBooleanValue();
5830 
5831   // 0 values are handled later by CheckTrivialUnsignedComparison().
5832   if ((Value == 0) && (!OtherIsBooleanType))
5833     return;
5834 
5835   BinaryOperatorKind op = E->getOpcode();
5836   bool IsTrue = true;
5837 
5838   // Used for diagnostic printout.
5839   enum {
5840     LiteralConstant = 0,
5841     CXXBoolLiteralTrue,
5842     CXXBoolLiteralFalse
5843   } LiteralOrBoolConstant = LiteralConstant;
5844 
5845   if (!OtherIsBooleanType) {
5846     QualType ConstantT = Constant->getType();
5847     QualType CommonT = E->getLHS()->getType();
5848 
5849     if (S.Context.hasSameUnqualifiedType(OtherT, ConstantT))
5850       return;
5851     assert((OtherT->isIntegerType() && ConstantT->isIntegerType()) &&
5852            "comparison with non-integer type");
5853 
5854     bool ConstantSigned = ConstantT->isSignedIntegerType();
5855     bool CommonSigned = CommonT->isSignedIntegerType();
5856 
5857     bool EqualityOnly = false;
5858 
5859     if (CommonSigned) {
5860       // The common type is signed, therefore no signed to unsigned conversion.
5861       if (!OtherRange.NonNegative) {
5862         // Check that the constant is representable in type OtherT.
5863         if (ConstantSigned) {
5864           if (OtherWidth >= Value.getMinSignedBits())
5865             return;
5866         } else { // !ConstantSigned
5867           if (OtherWidth >= Value.getActiveBits() + 1)
5868             return;
5869         }
5870       } else { // !OtherSigned
5871                // Check that the constant is representable in type OtherT.
5872         // Negative values are out of range.
5873         if (ConstantSigned) {
5874           if (Value.isNonNegative() && OtherWidth >= Value.getActiveBits())
5875             return;
5876         } else { // !ConstantSigned
5877           if (OtherWidth >= Value.getActiveBits())
5878             return;
5879         }
5880       }
5881     } else { // !CommonSigned
5882       if (OtherRange.NonNegative) {
5883         if (OtherWidth >= Value.getActiveBits())
5884           return;
5885       } else { // OtherSigned
5886         assert(!ConstantSigned &&
5887                "Two signed types converted to unsigned types.");
5888         // Check to see if the constant is representable in OtherT.
5889         if (OtherWidth > Value.getActiveBits())
5890           return;
5891         // Check to see if the constant is equivalent to a negative value
5892         // cast to CommonT.
5893         if (S.Context.getIntWidth(ConstantT) ==
5894                 S.Context.getIntWidth(CommonT) &&
5895             Value.isNegative() && Value.getMinSignedBits() <= OtherWidth)
5896           return;
5897         // The constant value rests between values that OtherT can represent
5898         // after conversion.  Relational comparison still works, but equality
5899         // comparisons will be tautological.
5900         EqualityOnly = true;
5901       }
5902     }
5903 
5904     bool PositiveConstant = !ConstantSigned || Value.isNonNegative();
5905 
5906     if (op == BO_EQ || op == BO_NE) {
5907       IsTrue = op == BO_NE;
5908     } else if (EqualityOnly) {
5909       return;
5910     } else if (RhsConstant) {
5911       if (op == BO_GT || op == BO_GE)
5912         IsTrue = !PositiveConstant;
5913       else // op == BO_LT || op == BO_LE
5914         IsTrue = PositiveConstant;
5915     } else {
5916       if (op == BO_LT || op == BO_LE)
5917         IsTrue = !PositiveConstant;
5918       else // op == BO_GT || op == BO_GE
5919         IsTrue = PositiveConstant;
5920     }
5921   } else {
5922     // Other isKnownToHaveBooleanValue
5923     enum CompareBoolWithConstantResult { AFals, ATrue, Unkwn };
5924     enum ConstantValue { LT_Zero, Zero, One, GT_One, SizeOfConstVal };
5925     enum ConstantSide { Lhs, Rhs, SizeOfConstSides };
5926 
5927     static const struct LinkedConditions {
5928       CompareBoolWithConstantResult BO_LT_OP[SizeOfConstSides][SizeOfConstVal];
5929       CompareBoolWithConstantResult BO_GT_OP[SizeOfConstSides][SizeOfConstVal];
5930       CompareBoolWithConstantResult BO_LE_OP[SizeOfConstSides][SizeOfConstVal];
5931       CompareBoolWithConstantResult BO_GE_OP[SizeOfConstSides][SizeOfConstVal];
5932       CompareBoolWithConstantResult BO_EQ_OP[SizeOfConstSides][SizeOfConstVal];
5933       CompareBoolWithConstantResult BO_NE_OP[SizeOfConstSides][SizeOfConstVal];
5934 
5935     } TruthTable = {
5936         // Constant on LHS.              | Constant on RHS.              |
5937         // LT_Zero| Zero  | One   |GT_One| LT_Zero| Zero  | One   |GT_One|
5938         { { ATrue, Unkwn, AFals, AFals }, { AFals, AFals, Unkwn, ATrue } },
5939         { { AFals, AFals, Unkwn, ATrue }, { ATrue, Unkwn, AFals, AFals } },
5940         { { ATrue, ATrue, Unkwn, AFals }, { AFals, Unkwn, ATrue, ATrue } },
5941         { { AFals, Unkwn, ATrue, ATrue }, { ATrue, ATrue, Unkwn, AFals } },
5942         { { AFals, Unkwn, Unkwn, AFals }, { AFals, Unkwn, Unkwn, AFals } },
5943         { { ATrue, Unkwn, Unkwn, ATrue }, { ATrue, Unkwn, Unkwn, ATrue } }
5944       };
5945 
5946     bool ConstantIsBoolLiteral = isa<CXXBoolLiteralExpr>(Constant);
5947 
5948     enum ConstantValue ConstVal = Zero;
5949     if (Value.isUnsigned() || Value.isNonNegative()) {
5950       if (Value == 0) {
5951         LiteralOrBoolConstant =
5952             ConstantIsBoolLiteral ? CXXBoolLiteralFalse : LiteralConstant;
5953         ConstVal = Zero;
5954       } else if (Value == 1) {
5955         LiteralOrBoolConstant =
5956             ConstantIsBoolLiteral ? CXXBoolLiteralTrue : LiteralConstant;
5957         ConstVal = One;
5958       } else {
5959         LiteralOrBoolConstant = LiteralConstant;
5960         ConstVal = GT_One;
5961       }
5962     } else {
5963       ConstVal = LT_Zero;
5964     }
5965 
5966     CompareBoolWithConstantResult CmpRes;
5967 
5968     switch (op) {
5969     case BO_LT:
5970       CmpRes = TruthTable.BO_LT_OP[RhsConstant][ConstVal];
5971       break;
5972     case BO_GT:
5973       CmpRes = TruthTable.BO_GT_OP[RhsConstant][ConstVal];
5974       break;
5975     case BO_LE:
5976       CmpRes = TruthTable.BO_LE_OP[RhsConstant][ConstVal];
5977       break;
5978     case BO_GE:
5979       CmpRes = TruthTable.BO_GE_OP[RhsConstant][ConstVal];
5980       break;
5981     case BO_EQ:
5982       CmpRes = TruthTable.BO_EQ_OP[RhsConstant][ConstVal];
5983       break;
5984     case BO_NE:
5985       CmpRes = TruthTable.BO_NE_OP[RhsConstant][ConstVal];
5986       break;
5987     default:
5988       CmpRes = Unkwn;
5989       break;
5990     }
5991 
5992     if (CmpRes == AFals) {
5993       IsTrue = false;
5994     } else if (CmpRes == ATrue) {
5995       IsTrue = true;
5996     } else {
5997       return;
5998     }
5999   }
6000 
6001   // If this is a comparison to an enum constant, include that
6002   // constant in the diagnostic.
6003   const EnumConstantDecl *ED = nullptr;
6004   if (const DeclRefExpr *DR = dyn_cast<DeclRefExpr>(Constant))
6005     ED = dyn_cast<EnumConstantDecl>(DR->getDecl());
6006 
6007   SmallString<64> PrettySourceValue;
6008   llvm::raw_svector_ostream OS(PrettySourceValue);
6009   if (ED)
6010     OS << '\'' << *ED << "' (" << Value << ")";
6011   else
6012     OS << Value;
6013 
6014   S.DiagRuntimeBehavior(
6015     E->getOperatorLoc(), E,
6016     S.PDiag(diag::warn_out_of_range_compare)
6017         << OS.str() << LiteralOrBoolConstant
6018         << OtherT << (OtherIsBooleanType && !OtherT->isBooleanType()) << IsTrue
6019         << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange());
6020 }
6021 
6022 /// Analyze the operands of the given comparison.  Implements the
6023 /// fallback case from AnalyzeComparison.
AnalyzeImpConvsInComparison(Sema & S,BinaryOperator * E)6024 static void AnalyzeImpConvsInComparison(Sema &S, BinaryOperator *E) {
6025   AnalyzeImplicitConversions(S, E->getLHS(), E->getOperatorLoc());
6026   AnalyzeImplicitConversions(S, E->getRHS(), E->getOperatorLoc());
6027 }
6028 
6029 /// \brief Implements -Wsign-compare.
6030 ///
6031 /// \param E the binary operator to check for warnings
AnalyzeComparison(Sema & S,BinaryOperator * E)6032 static void AnalyzeComparison(Sema &S, BinaryOperator *E) {
6033   // The type the comparison is being performed in.
6034   QualType T = E->getLHS()->getType();
6035 
6036   // Only analyze comparison operators where both sides have been converted to
6037   // the same type.
6038   if (!S.Context.hasSameUnqualifiedType(T, E->getRHS()->getType()))
6039     return AnalyzeImpConvsInComparison(S, E);
6040 
6041   // Don't analyze value-dependent comparisons directly.
6042   if (E->isValueDependent())
6043     return AnalyzeImpConvsInComparison(S, E);
6044 
6045   Expr *LHS = E->getLHS()->IgnoreParenImpCasts();
6046   Expr *RHS = E->getRHS()->IgnoreParenImpCasts();
6047 
6048   bool IsComparisonConstant = false;
6049 
6050   // Check whether an integer constant comparison results in a value
6051   // of 'true' or 'false'.
6052   if (T->isIntegralType(S.Context)) {
6053     llvm::APSInt RHSValue;
6054     bool IsRHSIntegralLiteral =
6055       RHS->isIntegerConstantExpr(RHSValue, S.Context);
6056     llvm::APSInt LHSValue;
6057     bool IsLHSIntegralLiteral =
6058       LHS->isIntegerConstantExpr(LHSValue, S.Context);
6059     if (IsRHSIntegralLiteral && !IsLHSIntegralLiteral)
6060         DiagnoseOutOfRangeComparison(S, E, RHS, LHS, RHSValue, true);
6061     else if (!IsRHSIntegralLiteral && IsLHSIntegralLiteral)
6062       DiagnoseOutOfRangeComparison(S, E, LHS, RHS, LHSValue, false);
6063     else
6064       IsComparisonConstant =
6065         (IsRHSIntegralLiteral && IsLHSIntegralLiteral);
6066   } else if (!T->hasUnsignedIntegerRepresentation())
6067       IsComparisonConstant = E->isIntegerConstantExpr(S.Context);
6068 
6069   // We don't do anything special if this isn't an unsigned integral
6070   // comparison:  we're only interested in integral comparisons, and
6071   // signed comparisons only happen in cases we don't care to warn about.
6072   //
6073   // We also don't care about value-dependent expressions or expressions
6074   // whose result is a constant.
6075   if (!T->hasUnsignedIntegerRepresentation() || IsComparisonConstant)
6076     return AnalyzeImpConvsInComparison(S, E);
6077 
6078   // Check to see if one of the (unmodified) operands is of different
6079   // signedness.
6080   Expr *signedOperand, *unsignedOperand;
6081   if (LHS->getType()->hasSignedIntegerRepresentation()) {
6082     assert(!RHS->getType()->hasSignedIntegerRepresentation() &&
6083            "unsigned comparison between two signed integer expressions?");
6084     signedOperand = LHS;
6085     unsignedOperand = RHS;
6086   } else if (RHS->getType()->hasSignedIntegerRepresentation()) {
6087     signedOperand = RHS;
6088     unsignedOperand = LHS;
6089   } else {
6090     CheckTrivialUnsignedComparison(S, E);
6091     return AnalyzeImpConvsInComparison(S, E);
6092   }
6093 
6094   // Otherwise, calculate the effective range of the signed operand.
6095   IntRange signedRange = GetExprRange(S.Context, signedOperand);
6096 
6097   // Go ahead and analyze implicit conversions in the operands.  Note
6098   // that we skip the implicit conversions on both sides.
6099   AnalyzeImplicitConversions(S, LHS, E->getOperatorLoc());
6100   AnalyzeImplicitConversions(S, RHS, E->getOperatorLoc());
6101 
6102   // If the signed range is non-negative, -Wsign-compare won't fire,
6103   // but we should still check for comparisons which are always true
6104   // or false.
6105   if (signedRange.NonNegative)
6106     return CheckTrivialUnsignedComparison(S, E);
6107 
6108   // For (in)equality comparisons, if the unsigned operand is a
6109   // constant which cannot collide with a overflowed signed operand,
6110   // then reinterpreting the signed operand as unsigned will not
6111   // change the result of the comparison.
6112   if (E->isEqualityOp()) {
6113     unsigned comparisonWidth = S.Context.getIntWidth(T);
6114     IntRange unsignedRange = GetExprRange(S.Context, unsignedOperand);
6115 
6116     // We should never be unable to prove that the unsigned operand is
6117     // non-negative.
6118     assert(unsignedRange.NonNegative && "unsigned range includes negative?");
6119 
6120     if (unsignedRange.Width < comparisonWidth)
6121       return;
6122   }
6123 
6124   S.DiagRuntimeBehavior(E->getOperatorLoc(), E,
6125     S.PDiag(diag::warn_mixed_sign_comparison)
6126       << LHS->getType() << RHS->getType()
6127       << LHS->getSourceRange() << RHS->getSourceRange());
6128 }
6129 
6130 /// Analyzes an attempt to assign the given value to a bitfield.
6131 ///
6132 /// Returns true if there was something fishy about the attempt.
AnalyzeBitFieldAssignment(Sema & S,FieldDecl * Bitfield,Expr * Init,SourceLocation InitLoc)6133 static bool AnalyzeBitFieldAssignment(Sema &S, FieldDecl *Bitfield, Expr *Init,
6134                                       SourceLocation InitLoc) {
6135   assert(Bitfield->isBitField());
6136   if (Bitfield->isInvalidDecl())
6137     return false;
6138 
6139   // White-list bool bitfields.
6140   if (Bitfield->getType()->isBooleanType())
6141     return false;
6142 
6143   // Ignore value- or type-dependent expressions.
6144   if (Bitfield->getBitWidth()->isValueDependent() ||
6145       Bitfield->getBitWidth()->isTypeDependent() ||
6146       Init->isValueDependent() ||
6147       Init->isTypeDependent())
6148     return false;
6149 
6150   Expr *OriginalInit = Init->IgnoreParenImpCasts();
6151 
6152   llvm::APSInt Value;
6153   if (!OriginalInit->EvaluateAsInt(Value, S.Context, Expr::SE_AllowSideEffects))
6154     return false;
6155 
6156   unsigned OriginalWidth = Value.getBitWidth();
6157   unsigned FieldWidth = Bitfield->getBitWidthValue(S.Context);
6158 
6159   if (OriginalWidth <= FieldWidth)
6160     return false;
6161 
6162   // Compute the value which the bitfield will contain.
6163   llvm::APSInt TruncatedValue = Value.trunc(FieldWidth);
6164   TruncatedValue.setIsSigned(Bitfield->getType()->isSignedIntegerType());
6165 
6166   // Check whether the stored value is equal to the original value.
6167   TruncatedValue = TruncatedValue.extend(OriginalWidth);
6168   if (llvm::APSInt::isSameValue(Value, TruncatedValue))
6169     return false;
6170 
6171   // Special-case bitfields of width 1: booleans are naturally 0/1, and
6172   // therefore don't strictly fit into a signed bitfield of width 1.
6173   if (FieldWidth == 1 && Value == 1)
6174     return false;
6175 
6176   std::string PrettyValue = Value.toString(10);
6177   std::string PrettyTrunc = TruncatedValue.toString(10);
6178 
6179   S.Diag(InitLoc, diag::warn_impcast_bitfield_precision_constant)
6180     << PrettyValue << PrettyTrunc << OriginalInit->getType()
6181     << Init->getSourceRange();
6182 
6183   return true;
6184 }
6185 
6186 /// Analyze the given simple or compound assignment for warning-worthy
6187 /// operations.
AnalyzeAssignment(Sema & S,BinaryOperator * E)6188 static void AnalyzeAssignment(Sema &S, BinaryOperator *E) {
6189   // Just recurse on the LHS.
6190   AnalyzeImplicitConversions(S, E->getLHS(), E->getOperatorLoc());
6191 
6192   // We want to recurse on the RHS as normal unless we're assigning to
6193   // a bitfield.
6194   if (FieldDecl *Bitfield = E->getLHS()->getSourceBitField()) {
6195     if (AnalyzeBitFieldAssignment(S, Bitfield, E->getRHS(),
6196                                   E->getOperatorLoc())) {
6197       // Recurse, ignoring any implicit conversions on the RHS.
6198       return AnalyzeImplicitConversions(S, E->getRHS()->IgnoreParenImpCasts(),
6199                                         E->getOperatorLoc());
6200     }
6201   }
6202 
6203   AnalyzeImplicitConversions(S, E->getRHS(), E->getOperatorLoc());
6204 }
6205 
6206 /// Diagnose an implicit cast;  purely a helper for CheckImplicitConversion.
DiagnoseImpCast(Sema & S,Expr * E,QualType SourceType,QualType T,SourceLocation CContext,unsigned diag,bool pruneControlFlow=false)6207 static void DiagnoseImpCast(Sema &S, Expr *E, QualType SourceType, QualType T,
6208                             SourceLocation CContext, unsigned diag,
6209                             bool pruneControlFlow = false) {
6210   if (pruneControlFlow) {
6211     S.DiagRuntimeBehavior(E->getExprLoc(), E,
6212                           S.PDiag(diag)
6213                             << SourceType << T << E->getSourceRange()
6214                             << SourceRange(CContext));
6215     return;
6216   }
6217   S.Diag(E->getExprLoc(), diag)
6218     << SourceType << T << E->getSourceRange() << SourceRange(CContext);
6219 }
6220 
6221 /// Diagnose an implicit cast;  purely a helper for CheckImplicitConversion.
DiagnoseImpCast(Sema & S,Expr * E,QualType T,SourceLocation CContext,unsigned diag,bool pruneControlFlow=false)6222 static void DiagnoseImpCast(Sema &S, Expr *E, QualType T,
6223                             SourceLocation CContext, unsigned diag,
6224                             bool pruneControlFlow = false) {
6225   DiagnoseImpCast(S, E, E->getType(), T, CContext, diag, pruneControlFlow);
6226 }
6227 
6228 /// Diagnose an implicit cast from a literal expression. Does not warn when the
6229 /// cast wouldn't lose information.
DiagnoseFloatingLiteralImpCast(Sema & S,FloatingLiteral * FL,QualType T,SourceLocation CContext)6230 void DiagnoseFloatingLiteralImpCast(Sema &S, FloatingLiteral *FL, QualType T,
6231                                     SourceLocation CContext) {
6232   // Try to convert the literal exactly to an integer. If we can, don't warn.
6233   bool isExact = false;
6234   const llvm::APFloat &Value = FL->getValue();
6235   llvm::APSInt IntegerValue(S.Context.getIntWidth(T),
6236                             T->hasUnsignedIntegerRepresentation());
6237   if (Value.convertToInteger(IntegerValue,
6238                              llvm::APFloat::rmTowardZero, &isExact)
6239       == llvm::APFloat::opOK && isExact)
6240     return;
6241 
6242   // FIXME: Force the precision of the source value down so we don't print
6243   // digits which are usually useless (we don't really care here if we
6244   // truncate a digit by accident in edge cases).  Ideally, APFloat::toString
6245   // would automatically print the shortest representation, but it's a bit
6246   // tricky to implement.
6247   SmallString<16> PrettySourceValue;
6248   unsigned precision = llvm::APFloat::semanticsPrecision(Value.getSemantics());
6249   precision = (precision * 59 + 195) / 196;
6250   Value.toString(PrettySourceValue, precision);
6251 
6252   SmallString<16> PrettyTargetValue;
6253   if (T->isSpecificBuiltinType(BuiltinType::Bool))
6254     PrettyTargetValue = IntegerValue == 0 ? "false" : "true";
6255   else
6256     IntegerValue.toString(PrettyTargetValue);
6257 
6258   S.Diag(FL->getExprLoc(), diag::warn_impcast_literal_float_to_integer)
6259     << FL->getType() << T.getUnqualifiedType() << PrettySourceValue
6260     << PrettyTargetValue << FL->getSourceRange() << SourceRange(CContext);
6261 }
6262 
PrettyPrintInRange(const llvm::APSInt & Value,IntRange Range)6263 std::string PrettyPrintInRange(const llvm::APSInt &Value, IntRange Range) {
6264   if (!Range.Width) return "0";
6265 
6266   llvm::APSInt ValueInRange = Value;
6267   ValueInRange.setIsSigned(!Range.NonNegative);
6268   ValueInRange = ValueInRange.trunc(Range.Width);
6269   return ValueInRange.toString(10);
6270 }
6271 
IsImplicitBoolFloatConversion(Sema & S,Expr * Ex,bool ToBool)6272 static bool IsImplicitBoolFloatConversion(Sema &S, Expr *Ex, bool ToBool) {
6273   if (!isa<ImplicitCastExpr>(Ex))
6274     return false;
6275 
6276   Expr *InnerE = Ex->IgnoreParenImpCasts();
6277   const Type *Target = S.Context.getCanonicalType(Ex->getType()).getTypePtr();
6278   const Type *Source =
6279     S.Context.getCanonicalType(InnerE->getType()).getTypePtr();
6280   if (Target->isDependentType())
6281     return false;
6282 
6283   const BuiltinType *FloatCandidateBT =
6284     dyn_cast<BuiltinType>(ToBool ? Source : Target);
6285   const Type *BoolCandidateType = ToBool ? Target : Source;
6286 
6287   return (BoolCandidateType->isSpecificBuiltinType(BuiltinType::Bool) &&
6288           FloatCandidateBT && (FloatCandidateBT->isFloatingPoint()));
6289 }
6290 
CheckImplicitArgumentConversions(Sema & S,CallExpr * TheCall,SourceLocation CC)6291 void CheckImplicitArgumentConversions(Sema &S, CallExpr *TheCall,
6292                                       SourceLocation CC) {
6293   unsigned NumArgs = TheCall->getNumArgs();
6294   for (unsigned i = 0; i < NumArgs; ++i) {
6295     Expr *CurrA = TheCall->getArg(i);
6296     if (!IsImplicitBoolFloatConversion(S, CurrA, true))
6297       continue;
6298 
6299     bool IsSwapped = ((i > 0) &&
6300         IsImplicitBoolFloatConversion(S, TheCall->getArg(i - 1), false));
6301     IsSwapped |= ((i < (NumArgs - 1)) &&
6302         IsImplicitBoolFloatConversion(S, TheCall->getArg(i + 1), false));
6303     if (IsSwapped) {
6304       // Warn on this floating-point to bool conversion.
6305       DiagnoseImpCast(S, CurrA->IgnoreParenImpCasts(),
6306                       CurrA->getType(), CC,
6307                       diag::warn_impcast_floating_point_to_bool);
6308     }
6309   }
6310 }
6311 
DiagnoseNullConversion(Sema & S,Expr * E,QualType T,SourceLocation CC)6312 static void DiagnoseNullConversion(Sema &S, Expr *E, QualType T,
6313                                    SourceLocation CC) {
6314   if (S.Diags.isIgnored(diag::warn_impcast_null_pointer_to_integer,
6315                         E->getExprLoc()))
6316     return;
6317 
6318   // Check for NULL (GNUNull) or nullptr (CXX11_nullptr).
6319   const Expr::NullPointerConstantKind NullKind =
6320       E->isNullPointerConstant(S.Context, Expr::NPC_ValueDependentIsNotNull);
6321   if (NullKind != Expr::NPCK_GNUNull && NullKind != Expr::NPCK_CXX11_nullptr)
6322     return;
6323 
6324   // Return if target type is a safe conversion.
6325   if (T->isAnyPointerType() || T->isBlockPointerType() ||
6326       T->isMemberPointerType() || !T->isScalarType() || T->isNullPtrType())
6327     return;
6328 
6329   SourceLocation Loc = E->getSourceRange().getBegin();
6330 
6331   // __null is usually wrapped in a macro.  Go up a macro if that is the case.
6332   if (NullKind == Expr::NPCK_GNUNull) {
6333     if (Loc.isMacroID())
6334       Loc = S.SourceMgr.getImmediateExpansionRange(Loc).first;
6335   }
6336 
6337   // Only warn if the null and context location are in the same macro expansion.
6338   if (S.SourceMgr.getFileID(Loc) != S.SourceMgr.getFileID(CC))
6339     return;
6340 
6341   S.Diag(Loc, diag::warn_impcast_null_pointer_to_integer)
6342       << (NullKind == Expr::NPCK_CXX11_nullptr) << T << clang::SourceRange(CC)
6343       << FixItHint::CreateReplacement(Loc,
6344                                       S.getFixItZeroLiteralForType(T, Loc));
6345 }
6346 
CheckImplicitConversion(Sema & S,Expr * E,QualType T,SourceLocation CC,bool * ICContext=nullptr)6347 void CheckImplicitConversion(Sema &S, Expr *E, QualType T,
6348                              SourceLocation CC, bool *ICContext = nullptr) {
6349   if (E->isTypeDependent() || E->isValueDependent()) return;
6350 
6351   const Type *Source = S.Context.getCanonicalType(E->getType()).getTypePtr();
6352   const Type *Target = S.Context.getCanonicalType(T).getTypePtr();
6353   if (Source == Target) return;
6354   if (Target->isDependentType()) return;
6355 
6356   // If the conversion context location is invalid don't complain. We also
6357   // don't want to emit a warning if the issue occurs from the expansion of
6358   // a system macro. The problem is that 'getSpellingLoc()' is slow, so we
6359   // delay this check as long as possible. Once we detect we are in that
6360   // scenario, we just return.
6361   if (CC.isInvalid())
6362     return;
6363 
6364   // Diagnose implicit casts to bool.
6365   if (Target->isSpecificBuiltinType(BuiltinType::Bool)) {
6366     if (isa<StringLiteral>(E))
6367       // Warn on string literal to bool.  Checks for string literals in logical
6368       // and expressions, for instance, assert(0 && "error here"), are
6369       // prevented by a check in AnalyzeImplicitConversions().
6370       return DiagnoseImpCast(S, E, T, CC,
6371                              diag::warn_impcast_string_literal_to_bool);
6372     if (isa<ObjCStringLiteral>(E) || isa<ObjCArrayLiteral>(E) ||
6373         isa<ObjCDictionaryLiteral>(E) || isa<ObjCBoxedExpr>(E)) {
6374       // This covers the literal expressions that evaluate to Objective-C
6375       // objects.
6376       return DiagnoseImpCast(S, E, T, CC,
6377                              diag::warn_impcast_objective_c_literal_to_bool);
6378     }
6379     if (Source->isPointerType() || Source->canDecayToPointerType()) {
6380       // Warn on pointer to bool conversion that is always true.
6381       S.DiagnoseAlwaysNonNullPointer(E, Expr::NPCK_NotNull, /*IsEqual*/ false,
6382                                      SourceRange(CC));
6383     }
6384   }
6385 
6386   // Strip vector types.
6387   if (isa<VectorType>(Source)) {
6388     if (!isa<VectorType>(Target)) {
6389       if (S.SourceMgr.isInSystemMacro(CC))
6390         return;
6391       return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_vector_scalar);
6392     }
6393 
6394     // If the vector cast is cast between two vectors of the same size, it is
6395     // a bitcast, not a conversion.
6396     if (S.Context.getTypeSize(Source) == S.Context.getTypeSize(Target))
6397       return;
6398 
6399     Source = cast<VectorType>(Source)->getElementType().getTypePtr();
6400     Target = cast<VectorType>(Target)->getElementType().getTypePtr();
6401   }
6402   if (auto VecTy = dyn_cast<VectorType>(Target))
6403     Target = VecTy->getElementType().getTypePtr();
6404 
6405   // Strip complex types.
6406   if (isa<ComplexType>(Source)) {
6407     if (!isa<ComplexType>(Target)) {
6408       if (S.SourceMgr.isInSystemMacro(CC))
6409         return;
6410 
6411       return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_complex_scalar);
6412     }
6413 
6414     Source = cast<ComplexType>(Source)->getElementType().getTypePtr();
6415     Target = cast<ComplexType>(Target)->getElementType().getTypePtr();
6416   }
6417 
6418   const BuiltinType *SourceBT = dyn_cast<BuiltinType>(Source);
6419   const BuiltinType *TargetBT = dyn_cast<BuiltinType>(Target);
6420 
6421   // If the source is floating point...
6422   if (SourceBT && SourceBT->isFloatingPoint()) {
6423     // ...and the target is floating point...
6424     if (TargetBT && TargetBT->isFloatingPoint()) {
6425       // ...then warn if we're dropping FP rank.
6426 
6427       // Builtin FP kinds are ordered by increasing FP rank.
6428       if (SourceBT->getKind() > TargetBT->getKind()) {
6429         // Don't warn about float constants that are precisely
6430         // representable in the target type.
6431         Expr::EvalResult result;
6432         if (E->EvaluateAsRValue(result, S.Context)) {
6433           // Value might be a float, a float vector, or a float complex.
6434           if (IsSameFloatAfterCast(result.Val,
6435                    S.Context.getFloatTypeSemantics(QualType(TargetBT, 0)),
6436                    S.Context.getFloatTypeSemantics(QualType(SourceBT, 0))))
6437             return;
6438         }
6439 
6440         if (S.SourceMgr.isInSystemMacro(CC))
6441           return;
6442 
6443         DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_float_precision);
6444       }
6445       return;
6446     }
6447 
6448     // If the target is integral, always warn.
6449     if (TargetBT && TargetBT->isInteger()) {
6450       if (S.SourceMgr.isInSystemMacro(CC))
6451         return;
6452 
6453       Expr *InnerE = E->IgnoreParenImpCasts();
6454       // We also want to warn on, e.g., "int i = -1.234"
6455       if (UnaryOperator *UOp = dyn_cast<UnaryOperator>(InnerE))
6456         if (UOp->getOpcode() == UO_Minus || UOp->getOpcode() == UO_Plus)
6457           InnerE = UOp->getSubExpr()->IgnoreParenImpCasts();
6458 
6459       if (FloatingLiteral *FL = dyn_cast<FloatingLiteral>(InnerE)) {
6460         DiagnoseFloatingLiteralImpCast(S, FL, T, CC);
6461       } else {
6462         DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_float_integer);
6463       }
6464     }
6465 
6466     // If the target is bool, warn if expr is a function or method call.
6467     if (Target->isSpecificBuiltinType(BuiltinType::Bool) &&
6468         isa<CallExpr>(E)) {
6469       // Check last argument of function call to see if it is an
6470       // implicit cast from a type matching the type the result
6471       // is being cast to.
6472       CallExpr *CEx = cast<CallExpr>(E);
6473       unsigned NumArgs = CEx->getNumArgs();
6474       if (NumArgs > 0) {
6475         Expr *LastA = CEx->getArg(NumArgs - 1);
6476         Expr *InnerE = LastA->IgnoreParenImpCasts();
6477         const Type *InnerType =
6478           S.Context.getCanonicalType(InnerE->getType()).getTypePtr();
6479         if (isa<ImplicitCastExpr>(LastA) && (InnerType == Target)) {
6480           // Warn on this floating-point to bool conversion
6481           DiagnoseImpCast(S, E, T, CC,
6482                           diag::warn_impcast_floating_point_to_bool);
6483         }
6484       }
6485     }
6486     return;
6487   }
6488 
6489   DiagnoseNullConversion(S, E, T, CC);
6490 
6491   if (!Source->isIntegerType() || !Target->isIntegerType())
6492     return;
6493 
6494   // TODO: remove this early return once the false positives for constant->bool
6495   // in templates, macros, etc, are reduced or removed.
6496   if (Target->isSpecificBuiltinType(BuiltinType::Bool))
6497     return;
6498 
6499   IntRange SourceRange = GetExprRange(S.Context, E);
6500   IntRange TargetRange = IntRange::forTargetOfCanonicalType(S.Context, Target);
6501 
6502   if (SourceRange.Width > TargetRange.Width) {
6503     // If the source is a constant, use a default-on diagnostic.
6504     // TODO: this should happen for bitfield stores, too.
6505     llvm::APSInt Value(32);
6506     if (E->isIntegerConstantExpr(Value, S.Context)) {
6507       if (S.SourceMgr.isInSystemMacro(CC))
6508         return;
6509 
6510       std::string PrettySourceValue = Value.toString(10);
6511       std::string PrettyTargetValue = PrettyPrintInRange(Value, TargetRange);
6512 
6513       S.DiagRuntimeBehavior(E->getExprLoc(), E,
6514         S.PDiag(diag::warn_impcast_integer_precision_constant)
6515             << PrettySourceValue << PrettyTargetValue
6516             << E->getType() << T << E->getSourceRange()
6517             << clang::SourceRange(CC));
6518       return;
6519     }
6520 
6521     // People want to build with -Wshorten-64-to-32 and not -Wconversion.
6522     if (S.SourceMgr.isInSystemMacro(CC))
6523       return;
6524 
6525     if (TargetRange.Width == 32 && S.Context.getIntWidth(E->getType()) == 64)
6526       return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_integer_64_32,
6527                              /* pruneControlFlow */ true);
6528     return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_integer_precision);
6529   }
6530 
6531   if ((TargetRange.NonNegative && !SourceRange.NonNegative) ||
6532       (!TargetRange.NonNegative && SourceRange.NonNegative &&
6533        SourceRange.Width == TargetRange.Width)) {
6534 
6535     if (S.SourceMgr.isInSystemMacro(CC))
6536       return;
6537 
6538     unsigned DiagID = diag::warn_impcast_integer_sign;
6539 
6540     // Traditionally, gcc has warned about this under -Wsign-compare.
6541     // We also want to warn about it in -Wconversion.
6542     // So if -Wconversion is off, use a completely identical diagnostic
6543     // in the sign-compare group.
6544     // The conditional-checking code will
6545     if (ICContext) {
6546       DiagID = diag::warn_impcast_integer_sign_conditional;
6547       *ICContext = true;
6548     }
6549 
6550     return DiagnoseImpCast(S, E, T, CC, DiagID);
6551   }
6552 
6553   // Diagnose conversions between different enumeration types.
6554   // In C, we pretend that the type of an EnumConstantDecl is its enumeration
6555   // type, to give us better diagnostics.
6556   QualType SourceType = E->getType();
6557   if (!S.getLangOpts().CPlusPlus) {
6558     if (DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(E))
6559       if (EnumConstantDecl *ECD = dyn_cast<EnumConstantDecl>(DRE->getDecl())) {
6560         EnumDecl *Enum = cast<EnumDecl>(ECD->getDeclContext());
6561         SourceType = S.Context.getTypeDeclType(Enum);
6562         Source = S.Context.getCanonicalType(SourceType).getTypePtr();
6563       }
6564   }
6565 
6566   if (const EnumType *SourceEnum = Source->getAs<EnumType>())
6567     if (const EnumType *TargetEnum = Target->getAs<EnumType>())
6568       if (SourceEnum->getDecl()->hasNameForLinkage() &&
6569           TargetEnum->getDecl()->hasNameForLinkage() &&
6570           SourceEnum != TargetEnum) {
6571         if (S.SourceMgr.isInSystemMacro(CC))
6572           return;
6573 
6574         return DiagnoseImpCast(S, E, SourceType, T, CC,
6575                                diag::warn_impcast_different_enum_types);
6576       }
6577 
6578   return;
6579 }
6580 
6581 void CheckConditionalOperator(Sema &S, ConditionalOperator *E,
6582                               SourceLocation CC, QualType T);
6583 
CheckConditionalOperand(Sema & S,Expr * E,QualType T,SourceLocation CC,bool & ICContext)6584 void CheckConditionalOperand(Sema &S, Expr *E, QualType T,
6585                              SourceLocation CC, bool &ICContext) {
6586   E = E->IgnoreParenImpCasts();
6587 
6588   if (isa<ConditionalOperator>(E))
6589     return CheckConditionalOperator(S, cast<ConditionalOperator>(E), CC, T);
6590 
6591   AnalyzeImplicitConversions(S, E, CC);
6592   if (E->getType() != T)
6593     return CheckImplicitConversion(S, E, T, CC, &ICContext);
6594   return;
6595 }
6596 
CheckConditionalOperator(Sema & S,ConditionalOperator * E,SourceLocation CC,QualType T)6597 void CheckConditionalOperator(Sema &S, ConditionalOperator *E,
6598                               SourceLocation CC, QualType T) {
6599   AnalyzeImplicitConversions(S, E->getCond(), E->getQuestionLoc());
6600 
6601   bool Suspicious = false;
6602   CheckConditionalOperand(S, E->getTrueExpr(), T, CC, Suspicious);
6603   CheckConditionalOperand(S, E->getFalseExpr(), T, CC, Suspicious);
6604 
6605   // If -Wconversion would have warned about either of the candidates
6606   // for a signedness conversion to the context type...
6607   if (!Suspicious) return;
6608 
6609   // ...but it's currently ignored...
6610   if (!S.Diags.isIgnored(diag::warn_impcast_integer_sign_conditional, CC))
6611     return;
6612 
6613   // ...then check whether it would have warned about either of the
6614   // candidates for a signedness conversion to the condition type.
6615   if (E->getType() == T) return;
6616 
6617   Suspicious = false;
6618   CheckImplicitConversion(S, E->getTrueExpr()->IgnoreParenImpCasts(),
6619                           E->getType(), CC, &Suspicious);
6620   if (!Suspicious)
6621     CheckImplicitConversion(S, E->getFalseExpr()->IgnoreParenImpCasts(),
6622                             E->getType(), CC, &Suspicious);
6623 }
6624 
6625 /// CheckBoolLikeConversion - Check conversion of given expression to boolean.
6626 /// Input argument E is a logical expression.
CheckBoolLikeConversion(Sema & S,Expr * E,SourceLocation CC)6627 static void CheckBoolLikeConversion(Sema &S, Expr *E, SourceLocation CC) {
6628   if (S.getLangOpts().Bool)
6629     return;
6630   CheckImplicitConversion(S, E->IgnoreParenImpCasts(), S.Context.BoolTy, CC);
6631 }
6632 
6633 /// AnalyzeImplicitConversions - Find and report any interesting
6634 /// implicit conversions in the given expression.  There are a couple
6635 /// of competing diagnostics here, -Wconversion and -Wsign-compare.
AnalyzeImplicitConversions(Sema & S,Expr * OrigE,SourceLocation CC)6636 void AnalyzeImplicitConversions(Sema &S, Expr *OrigE, SourceLocation CC) {
6637   QualType T = OrigE->getType();
6638   Expr *E = OrigE->IgnoreParenImpCasts();
6639 
6640   if (E->isTypeDependent() || E->isValueDependent())
6641     return;
6642 
6643   // For conditional operators, we analyze the arguments as if they
6644   // were being fed directly into the output.
6645   if (isa<ConditionalOperator>(E)) {
6646     ConditionalOperator *CO = cast<ConditionalOperator>(E);
6647     CheckConditionalOperator(S, CO, CC, T);
6648     return;
6649   }
6650 
6651   // Check implicit argument conversions for function calls.
6652   if (CallExpr *Call = dyn_cast<CallExpr>(E))
6653     CheckImplicitArgumentConversions(S, Call, CC);
6654 
6655   // Go ahead and check any implicit conversions we might have skipped.
6656   // The non-canonical typecheck is just an optimization;
6657   // CheckImplicitConversion will filter out dead implicit conversions.
6658   if (E->getType() != T)
6659     CheckImplicitConversion(S, E, T, CC);
6660 
6661   // Now continue drilling into this expression.
6662 
6663   if (PseudoObjectExpr * POE = dyn_cast<PseudoObjectExpr>(E)) {
6664     if (POE->getResultExpr())
6665       E = POE->getResultExpr();
6666   }
6667 
6668   if (const OpaqueValueExpr *OVE = dyn_cast<OpaqueValueExpr>(E))
6669     return AnalyzeImplicitConversions(S, OVE->getSourceExpr(), CC);
6670 
6671   // Skip past explicit casts.
6672   if (isa<ExplicitCastExpr>(E)) {
6673     E = cast<ExplicitCastExpr>(E)->getSubExpr()->IgnoreParenImpCasts();
6674     return AnalyzeImplicitConversions(S, E, CC);
6675   }
6676 
6677   if (BinaryOperator *BO = dyn_cast<BinaryOperator>(E)) {
6678     // Do a somewhat different check with comparison operators.
6679     if (BO->isComparisonOp())
6680       return AnalyzeComparison(S, BO);
6681 
6682     // And with simple assignments.
6683     if (BO->getOpcode() == BO_Assign)
6684       return AnalyzeAssignment(S, BO);
6685   }
6686 
6687   // These break the otherwise-useful invariant below.  Fortunately,
6688   // we don't really need to recurse into them, because any internal
6689   // expressions should have been analyzed already when they were
6690   // built into statements.
6691   if (isa<StmtExpr>(E)) return;
6692 
6693   // Don't descend into unevaluated contexts.
6694   if (isa<UnaryExprOrTypeTraitExpr>(E)) return;
6695 
6696   // Now just recurse over the expression's children.
6697   CC = E->getExprLoc();
6698   BinaryOperator *BO = dyn_cast<BinaryOperator>(E);
6699   bool IsLogicalAndOperator = BO && BO->getOpcode() == BO_LAnd;
6700   for (Stmt::child_range I = E->children(); I; ++I) {
6701     Expr *ChildExpr = dyn_cast_or_null<Expr>(*I);
6702     if (!ChildExpr)
6703       continue;
6704 
6705     if (IsLogicalAndOperator &&
6706         isa<StringLiteral>(ChildExpr->IgnoreParenImpCasts()))
6707       // Ignore checking string literals that are in logical and operators.
6708       // This is a common pattern for asserts.
6709       continue;
6710     AnalyzeImplicitConversions(S, ChildExpr, CC);
6711   }
6712 
6713   if (BO && BO->isLogicalOp()) {
6714     Expr *SubExpr = BO->getLHS()->IgnoreParenImpCasts();
6715     if (!IsLogicalAndOperator || !isa<StringLiteral>(SubExpr))
6716       ::CheckBoolLikeConversion(S, SubExpr, BO->getExprLoc());
6717 
6718     SubExpr = BO->getRHS()->IgnoreParenImpCasts();
6719     if (!IsLogicalAndOperator || !isa<StringLiteral>(SubExpr))
6720       ::CheckBoolLikeConversion(S, SubExpr, BO->getExprLoc());
6721   }
6722 
6723   if (const UnaryOperator *U = dyn_cast<UnaryOperator>(E))
6724     if (U->getOpcode() == UO_LNot)
6725       ::CheckBoolLikeConversion(S, U->getSubExpr(), CC);
6726 }
6727 
6728 } // end anonymous namespace
6729 
6730 enum {
6731   AddressOf,
6732   FunctionPointer,
6733   ArrayPointer
6734 };
6735 
6736 // Helper function for Sema::DiagnoseAlwaysNonNullPointer.
6737 // Returns true when emitting a warning about taking the address of a reference.
CheckForReference(Sema & SemaRef,const Expr * E,PartialDiagnostic PD)6738 static bool CheckForReference(Sema &SemaRef, const Expr *E,
6739                               PartialDiagnostic PD) {
6740   E = E->IgnoreParenImpCasts();
6741 
6742   const FunctionDecl *FD = nullptr;
6743 
6744   if (const DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(E)) {
6745     if (!DRE->getDecl()->getType()->isReferenceType())
6746       return false;
6747   } else if (const MemberExpr *M = dyn_cast<MemberExpr>(E)) {
6748     if (!M->getMemberDecl()->getType()->isReferenceType())
6749       return false;
6750   } else if (const CallExpr *Call = dyn_cast<CallExpr>(E)) {
6751     if (!Call->getCallReturnType()->isReferenceType())
6752       return false;
6753     FD = Call->getDirectCallee();
6754   } else {
6755     return false;
6756   }
6757 
6758   SemaRef.Diag(E->getExprLoc(), PD);
6759 
6760   // If possible, point to location of function.
6761   if (FD) {
6762     SemaRef.Diag(FD->getLocation(), diag::note_reference_is_return_value) << FD;
6763   }
6764 
6765   return true;
6766 }
6767 
6768 // Returns true if the SourceLocation is expanded from any macro body.
6769 // Returns false if the SourceLocation is invalid, is from not in a macro
6770 // expansion, or is from expanded from a top-level macro argument.
IsInAnyMacroBody(const SourceManager & SM,SourceLocation Loc)6771 static bool IsInAnyMacroBody(const SourceManager &SM, SourceLocation Loc) {
6772   if (Loc.isInvalid())
6773     return false;
6774 
6775   while (Loc.isMacroID()) {
6776     if (SM.isMacroBodyExpansion(Loc))
6777       return true;
6778     Loc = SM.getImmediateMacroCallerLoc(Loc);
6779   }
6780 
6781   return false;
6782 }
6783 
6784 /// \brief Diagnose pointers that are always non-null.
6785 /// \param E the expression containing the pointer
6786 /// \param NullKind NPCK_NotNull if E is a cast to bool, otherwise, E is
6787 /// compared to a null pointer
6788 /// \param IsEqual True when the comparison is equal to a null pointer
6789 /// \param Range Extra SourceRange to highlight in the diagnostic
DiagnoseAlwaysNonNullPointer(Expr * E,Expr::NullPointerConstantKind NullKind,bool IsEqual,SourceRange Range)6790 void Sema::DiagnoseAlwaysNonNullPointer(Expr *E,
6791                                         Expr::NullPointerConstantKind NullKind,
6792                                         bool IsEqual, SourceRange Range) {
6793   if (!E)
6794     return;
6795 
6796   // Don't warn inside macros.
6797   if (E->getExprLoc().isMacroID()) {
6798     const SourceManager &SM = getSourceManager();
6799     if (IsInAnyMacroBody(SM, E->getExprLoc()) ||
6800         IsInAnyMacroBody(SM, Range.getBegin()))
6801       return;
6802   }
6803   E = E->IgnoreImpCasts();
6804 
6805   const bool IsCompare = NullKind != Expr::NPCK_NotNull;
6806 
6807   if (isa<CXXThisExpr>(E)) {
6808     unsigned DiagID = IsCompare ? diag::warn_this_null_compare
6809                                 : diag::warn_this_bool_conversion;
6810     Diag(E->getExprLoc(), DiagID) << E->getSourceRange() << Range << IsEqual;
6811     return;
6812   }
6813 
6814   bool IsAddressOf = false;
6815 
6816   if (UnaryOperator *UO = dyn_cast<UnaryOperator>(E)) {
6817     if (UO->getOpcode() != UO_AddrOf)
6818       return;
6819     IsAddressOf = true;
6820     E = UO->getSubExpr();
6821   }
6822 
6823   if (IsAddressOf) {
6824     unsigned DiagID = IsCompare
6825                           ? diag::warn_address_of_reference_null_compare
6826                           : diag::warn_address_of_reference_bool_conversion;
6827     PartialDiagnostic PD = PDiag(DiagID) << E->getSourceRange() << Range
6828                                          << IsEqual;
6829     if (CheckForReference(*this, E, PD)) {
6830       return;
6831     }
6832   }
6833 
6834   // Expect to find a single Decl.  Skip anything more complicated.
6835   ValueDecl *D = nullptr;
6836   if (DeclRefExpr *R = dyn_cast<DeclRefExpr>(E)) {
6837     D = R->getDecl();
6838   } else if (MemberExpr *M = dyn_cast<MemberExpr>(E)) {
6839     D = M->getMemberDecl();
6840   }
6841 
6842   // Weak Decls can be null.
6843   if (!D || D->isWeak())
6844     return;
6845 
6846   // Check for parameter decl with nonnull attribute
6847   if (const ParmVarDecl* PV = dyn_cast<ParmVarDecl>(D)) {
6848     if (getCurFunction() && !getCurFunction()->ModifiedNonNullParams.count(PV))
6849       if (const FunctionDecl* FD = dyn_cast<FunctionDecl>(PV->getDeclContext())) {
6850         unsigned NumArgs = FD->getNumParams();
6851         llvm::SmallBitVector AttrNonNull(NumArgs);
6852         for (const auto *NonNull : FD->specific_attrs<NonNullAttr>()) {
6853           if (!NonNull->args_size()) {
6854             AttrNonNull.set(0, NumArgs);
6855             break;
6856           }
6857           for (unsigned Val : NonNull->args()) {
6858             if (Val >= NumArgs)
6859               continue;
6860             AttrNonNull.set(Val);
6861           }
6862         }
6863         if (!AttrNonNull.empty())
6864           for (unsigned i = 0; i < NumArgs; ++i)
6865             if (FD->getParamDecl(i) == PV &&
6866                 (AttrNonNull[i] || PV->hasAttr<NonNullAttr>())) {
6867               std::string Str;
6868               llvm::raw_string_ostream S(Str);
6869               E->printPretty(S, nullptr, getPrintingPolicy());
6870               unsigned DiagID = IsCompare ? diag::warn_nonnull_parameter_compare
6871                                           : diag::warn_cast_nonnull_to_bool;
6872               Diag(E->getExprLoc(), DiagID) << S.str() << E->getSourceRange()
6873                 << Range << IsEqual;
6874               return;
6875             }
6876       }
6877     }
6878 
6879   QualType T = D->getType();
6880   const bool IsArray = T->isArrayType();
6881   const bool IsFunction = T->isFunctionType();
6882 
6883   // Address of function is used to silence the function warning.
6884   if (IsAddressOf && IsFunction) {
6885     return;
6886   }
6887 
6888   // Found nothing.
6889   if (!IsAddressOf && !IsFunction && !IsArray)
6890     return;
6891 
6892   // Pretty print the expression for the diagnostic.
6893   std::string Str;
6894   llvm::raw_string_ostream S(Str);
6895   E->printPretty(S, nullptr, getPrintingPolicy());
6896 
6897   unsigned DiagID = IsCompare ? diag::warn_null_pointer_compare
6898                               : diag::warn_impcast_pointer_to_bool;
6899   unsigned DiagType;
6900   if (IsAddressOf)
6901     DiagType = AddressOf;
6902   else if (IsFunction)
6903     DiagType = FunctionPointer;
6904   else if (IsArray)
6905     DiagType = ArrayPointer;
6906   else
6907     llvm_unreachable("Could not determine diagnostic.");
6908   Diag(E->getExprLoc(), DiagID) << DiagType << S.str() << E->getSourceRange()
6909                                 << Range << IsEqual;
6910 
6911   if (!IsFunction)
6912     return;
6913 
6914   // Suggest '&' to silence the function warning.
6915   Diag(E->getExprLoc(), diag::note_function_warning_silence)
6916       << FixItHint::CreateInsertion(E->getLocStart(), "&");
6917 
6918   // Check to see if '()' fixit should be emitted.
6919   QualType ReturnType;
6920   UnresolvedSet<4> NonTemplateOverloads;
6921   tryExprAsCall(*E, ReturnType, NonTemplateOverloads);
6922   if (ReturnType.isNull())
6923     return;
6924 
6925   if (IsCompare) {
6926     // There are two cases here.  If there is null constant, the only suggest
6927     // for a pointer return type.  If the null is 0, then suggest if the return
6928     // type is a pointer or an integer type.
6929     if (!ReturnType->isPointerType()) {
6930       if (NullKind == Expr::NPCK_ZeroExpression ||
6931           NullKind == Expr::NPCK_ZeroLiteral) {
6932         if (!ReturnType->isIntegerType())
6933           return;
6934       } else {
6935         return;
6936       }
6937     }
6938   } else { // !IsCompare
6939     // For function to bool, only suggest if the function pointer has bool
6940     // return type.
6941     if (!ReturnType->isSpecificBuiltinType(BuiltinType::Bool))
6942       return;
6943   }
6944   Diag(E->getExprLoc(), diag::note_function_to_function_call)
6945       << FixItHint::CreateInsertion(getLocForEndOfToken(E->getLocEnd()), "()");
6946 }
6947 
6948 
6949 /// Diagnoses "dangerous" implicit conversions within the given
6950 /// expression (which is a full expression).  Implements -Wconversion
6951 /// and -Wsign-compare.
6952 ///
6953 /// \param CC the "context" location of the implicit conversion, i.e.
6954 ///   the most location of the syntactic entity requiring the implicit
6955 ///   conversion
CheckImplicitConversions(Expr * E,SourceLocation CC)6956 void Sema::CheckImplicitConversions(Expr *E, SourceLocation CC) {
6957   // Don't diagnose in unevaluated contexts.
6958   if (isUnevaluatedContext())
6959     return;
6960 
6961   // Don't diagnose for value- or type-dependent expressions.
6962   if (E->isTypeDependent() || E->isValueDependent())
6963     return;
6964 
6965   // Check for array bounds violations in cases where the check isn't triggered
6966   // elsewhere for other Expr types (like BinaryOperators), e.g. when an
6967   // ArraySubscriptExpr is on the RHS of a variable initialization.
6968   CheckArrayAccess(E);
6969 
6970   // This is not the right CC for (e.g.) a variable initialization.
6971   AnalyzeImplicitConversions(*this, E, CC);
6972 }
6973 
6974 /// CheckBoolLikeConversion - Check conversion of given expression to boolean.
6975 /// Input argument E is a logical expression.
CheckBoolLikeConversion(Expr * E,SourceLocation CC)6976 void Sema::CheckBoolLikeConversion(Expr *E, SourceLocation CC) {
6977   ::CheckBoolLikeConversion(*this, E, CC);
6978 }
6979 
6980 /// Diagnose when expression is an integer constant expression and its evaluation
6981 /// results in integer overflow
CheckForIntOverflow(Expr * E)6982 void Sema::CheckForIntOverflow (Expr *E) {
6983   if (isa<BinaryOperator>(E->IgnoreParenCasts()))
6984     E->IgnoreParenCasts()->EvaluateForOverflow(Context);
6985 }
6986 
6987 namespace {
6988 /// \brief Visitor for expressions which looks for unsequenced operations on the
6989 /// same object.
6990 class SequenceChecker : public EvaluatedExprVisitor<SequenceChecker> {
6991   typedef EvaluatedExprVisitor<SequenceChecker> Base;
6992 
6993   /// \brief A tree of sequenced regions within an expression. Two regions are
6994   /// unsequenced if one is an ancestor or a descendent of the other. When we
6995   /// finish processing an expression with sequencing, such as a comma
6996   /// expression, we fold its tree nodes into its parent, since they are
6997   /// unsequenced with respect to nodes we will visit later.
6998   class SequenceTree {
6999     struct Value {
Value__anonb40dc79c0a11::SequenceChecker::SequenceTree::Value7000       explicit Value(unsigned Parent) : Parent(Parent), Merged(false) {}
7001       unsigned Parent : 31;
7002       bool Merged : 1;
7003     };
7004     SmallVector<Value, 8> Values;
7005 
7006   public:
7007     /// \brief A region within an expression which may be sequenced with respect
7008     /// to some other region.
7009     class Seq {
Seq(unsigned N)7010       explicit Seq(unsigned N) : Index(N) {}
7011       unsigned Index;
7012       friend class SequenceTree;
7013     public:
Seq()7014       Seq() : Index(0) {}
7015     };
7016 
SequenceTree()7017     SequenceTree() { Values.push_back(Value(0)); }
root() const7018     Seq root() const { return Seq(0); }
7019 
7020     /// \brief Create a new sequence of operations, which is an unsequenced
7021     /// subset of \p Parent. This sequence of operations is sequenced with
7022     /// respect to other children of \p Parent.
allocate(Seq Parent)7023     Seq allocate(Seq Parent) {
7024       Values.push_back(Value(Parent.Index));
7025       return Seq(Values.size() - 1);
7026     }
7027 
7028     /// \brief Merge a sequence of operations into its parent.
merge(Seq S)7029     void merge(Seq S) {
7030       Values[S.Index].Merged = true;
7031     }
7032 
7033     /// \brief Determine whether two operations are unsequenced. This operation
7034     /// is asymmetric: \p Cur should be the more recent sequence, and \p Old
7035     /// should have been merged into its parent as appropriate.
isUnsequenced(Seq Cur,Seq Old)7036     bool isUnsequenced(Seq Cur, Seq Old) {
7037       unsigned C = representative(Cur.Index);
7038       unsigned Target = representative(Old.Index);
7039       while (C >= Target) {
7040         if (C == Target)
7041           return true;
7042         C = Values[C].Parent;
7043       }
7044       return false;
7045     }
7046 
7047   private:
7048     /// \brief Pick a representative for a sequence.
representative(unsigned K)7049     unsigned representative(unsigned K) {
7050       if (Values[K].Merged)
7051         // Perform path compression as we go.
7052         return Values[K].Parent = representative(Values[K].Parent);
7053       return K;
7054     }
7055   };
7056 
7057   /// An object for which we can track unsequenced uses.
7058   typedef NamedDecl *Object;
7059 
7060   /// Different flavors of object usage which we track. We only track the
7061   /// least-sequenced usage of each kind.
7062   enum UsageKind {
7063     /// A read of an object. Multiple unsequenced reads are OK.
7064     UK_Use,
7065     /// A modification of an object which is sequenced before the value
7066     /// computation of the expression, such as ++n in C++.
7067     UK_ModAsValue,
7068     /// A modification of an object which is not sequenced before the value
7069     /// computation of the expression, such as n++.
7070     UK_ModAsSideEffect,
7071 
7072     UK_Count = UK_ModAsSideEffect + 1
7073   };
7074 
7075   struct Usage {
Usage__anonb40dc79c0a11::SequenceChecker::Usage7076     Usage() : Use(nullptr), Seq() {}
7077     Expr *Use;
7078     SequenceTree::Seq Seq;
7079   };
7080 
7081   struct UsageInfo {
UsageInfo__anonb40dc79c0a11::SequenceChecker::UsageInfo7082     UsageInfo() : Diagnosed(false) {}
7083     Usage Uses[UK_Count];
7084     /// Have we issued a diagnostic for this variable already?
7085     bool Diagnosed;
7086   };
7087   typedef llvm::SmallDenseMap<Object, UsageInfo, 16> UsageInfoMap;
7088 
7089   Sema &SemaRef;
7090   /// Sequenced regions within the expression.
7091   SequenceTree Tree;
7092   /// Declaration modifications and references which we have seen.
7093   UsageInfoMap UsageMap;
7094   /// The region we are currently within.
7095   SequenceTree::Seq Region;
7096   /// Filled in with declarations which were modified as a side-effect
7097   /// (that is, post-increment operations).
7098   SmallVectorImpl<std::pair<Object, Usage> > *ModAsSideEffect;
7099   /// Expressions to check later. We defer checking these to reduce
7100   /// stack usage.
7101   SmallVectorImpl<Expr *> &WorkList;
7102 
7103   /// RAII object wrapping the visitation of a sequenced subexpression of an
7104   /// expression. At the end of this process, the side-effects of the evaluation
7105   /// become sequenced with respect to the value computation of the result, so
7106   /// we downgrade any UK_ModAsSideEffect within the evaluation to
7107   /// UK_ModAsValue.
7108   struct SequencedSubexpression {
SequencedSubexpression__anonb40dc79c0a11::SequenceChecker::SequencedSubexpression7109     SequencedSubexpression(SequenceChecker &Self)
7110       : Self(Self), OldModAsSideEffect(Self.ModAsSideEffect) {
7111       Self.ModAsSideEffect = &ModAsSideEffect;
7112     }
~SequencedSubexpression__anonb40dc79c0a11::SequenceChecker::SequencedSubexpression7113     ~SequencedSubexpression() {
7114       for (auto MI = ModAsSideEffect.rbegin(), ME = ModAsSideEffect.rend();
7115            MI != ME; ++MI) {
7116         UsageInfo &U = Self.UsageMap[MI->first];
7117         auto &SideEffectUsage = U.Uses[UK_ModAsSideEffect];
7118         Self.addUsage(U, MI->first, SideEffectUsage.Use, UK_ModAsValue);
7119         SideEffectUsage = MI->second;
7120       }
7121       Self.ModAsSideEffect = OldModAsSideEffect;
7122     }
7123 
7124     SequenceChecker &Self;
7125     SmallVector<std::pair<Object, Usage>, 4> ModAsSideEffect;
7126     SmallVectorImpl<std::pair<Object, Usage> > *OldModAsSideEffect;
7127   };
7128 
7129   /// RAII object wrapping the visitation of a subexpression which we might
7130   /// choose to evaluate as a constant. If any subexpression is evaluated and
7131   /// found to be non-constant, this allows us to suppress the evaluation of
7132   /// the outer expression.
7133   class EvaluationTracker {
7134   public:
EvaluationTracker(SequenceChecker & Self)7135     EvaluationTracker(SequenceChecker &Self)
7136         : Self(Self), Prev(Self.EvalTracker), EvalOK(true) {
7137       Self.EvalTracker = this;
7138     }
~EvaluationTracker()7139     ~EvaluationTracker() {
7140       Self.EvalTracker = Prev;
7141       if (Prev)
7142         Prev->EvalOK &= EvalOK;
7143     }
7144 
evaluate(const Expr * E,bool & Result)7145     bool evaluate(const Expr *E, bool &Result) {
7146       if (!EvalOK || E->isValueDependent())
7147         return false;
7148       EvalOK = E->EvaluateAsBooleanCondition(Result, Self.SemaRef.Context);
7149       return EvalOK;
7150     }
7151 
7152   private:
7153     SequenceChecker &Self;
7154     EvaluationTracker *Prev;
7155     bool EvalOK;
7156   } *EvalTracker;
7157 
7158   /// \brief Find the object which is produced by the specified expression,
7159   /// if any.
getObject(Expr * E,bool Mod) const7160   Object getObject(Expr *E, bool Mod) const {
7161     E = E->IgnoreParenCasts();
7162     if (UnaryOperator *UO = dyn_cast<UnaryOperator>(E)) {
7163       if (Mod && (UO->getOpcode() == UO_PreInc || UO->getOpcode() == UO_PreDec))
7164         return getObject(UO->getSubExpr(), Mod);
7165     } else if (BinaryOperator *BO = dyn_cast<BinaryOperator>(E)) {
7166       if (BO->getOpcode() == BO_Comma)
7167         return getObject(BO->getRHS(), Mod);
7168       if (Mod && BO->isAssignmentOp())
7169         return getObject(BO->getLHS(), Mod);
7170     } else if (MemberExpr *ME = dyn_cast<MemberExpr>(E)) {
7171       // FIXME: Check for more interesting cases, like "x.n = ++x.n".
7172       if (isa<CXXThisExpr>(ME->getBase()->IgnoreParenCasts()))
7173         return ME->getMemberDecl();
7174     } else if (DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(E))
7175       // FIXME: If this is a reference, map through to its value.
7176       return DRE->getDecl();
7177     return nullptr;
7178   }
7179 
7180   /// \brief Note that an object was modified or used by an expression.
addUsage(UsageInfo & UI,Object O,Expr * Ref,UsageKind UK)7181   void addUsage(UsageInfo &UI, Object O, Expr *Ref, UsageKind UK) {
7182     Usage &U = UI.Uses[UK];
7183     if (!U.Use || !Tree.isUnsequenced(Region, U.Seq)) {
7184       if (UK == UK_ModAsSideEffect && ModAsSideEffect)
7185         ModAsSideEffect->push_back(std::make_pair(O, U));
7186       U.Use = Ref;
7187       U.Seq = Region;
7188     }
7189   }
7190   /// \brief Check whether a modification or use conflicts with a prior usage.
checkUsage(Object O,UsageInfo & UI,Expr * Ref,UsageKind OtherKind,bool IsModMod)7191   void checkUsage(Object O, UsageInfo &UI, Expr *Ref, UsageKind OtherKind,
7192                   bool IsModMod) {
7193     if (UI.Diagnosed)
7194       return;
7195 
7196     const Usage &U = UI.Uses[OtherKind];
7197     if (!U.Use || !Tree.isUnsequenced(Region, U.Seq))
7198       return;
7199 
7200     Expr *Mod = U.Use;
7201     Expr *ModOrUse = Ref;
7202     if (OtherKind == UK_Use)
7203       std::swap(Mod, ModOrUse);
7204 
7205     SemaRef.Diag(Mod->getExprLoc(),
7206                  IsModMod ? diag::warn_unsequenced_mod_mod
7207                           : diag::warn_unsequenced_mod_use)
7208       << O << SourceRange(ModOrUse->getExprLoc());
7209     UI.Diagnosed = true;
7210   }
7211 
notePreUse(Object O,Expr * Use)7212   void notePreUse(Object O, Expr *Use) {
7213     UsageInfo &U = UsageMap[O];
7214     // Uses conflict with other modifications.
7215     checkUsage(O, U, Use, UK_ModAsValue, false);
7216   }
notePostUse(Object O,Expr * Use)7217   void notePostUse(Object O, Expr *Use) {
7218     UsageInfo &U = UsageMap[O];
7219     checkUsage(O, U, Use, UK_ModAsSideEffect, false);
7220     addUsage(U, O, Use, UK_Use);
7221   }
7222 
notePreMod(Object O,Expr * Mod)7223   void notePreMod(Object O, Expr *Mod) {
7224     UsageInfo &U = UsageMap[O];
7225     // Modifications conflict with other modifications and with uses.
7226     checkUsage(O, U, Mod, UK_ModAsValue, true);
7227     checkUsage(O, U, Mod, UK_Use, false);
7228   }
notePostMod(Object O,Expr * Use,UsageKind UK)7229   void notePostMod(Object O, Expr *Use, UsageKind UK) {
7230     UsageInfo &U = UsageMap[O];
7231     checkUsage(O, U, Use, UK_ModAsSideEffect, true);
7232     addUsage(U, O, Use, UK);
7233   }
7234 
7235 public:
SequenceChecker(Sema & S,Expr * E,SmallVectorImpl<Expr * > & WorkList)7236   SequenceChecker(Sema &S, Expr *E, SmallVectorImpl<Expr *> &WorkList)
7237       : Base(S.Context), SemaRef(S), Region(Tree.root()),
7238         ModAsSideEffect(nullptr), WorkList(WorkList), EvalTracker(nullptr) {
7239     Visit(E);
7240   }
7241 
VisitStmt(Stmt * S)7242   void VisitStmt(Stmt *S) {
7243     // Skip all statements which aren't expressions for now.
7244   }
7245 
VisitExpr(Expr * E)7246   void VisitExpr(Expr *E) {
7247     // By default, just recurse to evaluated subexpressions.
7248     Base::VisitStmt(E);
7249   }
7250 
VisitCastExpr(CastExpr * E)7251   void VisitCastExpr(CastExpr *E) {
7252     Object O = Object();
7253     if (E->getCastKind() == CK_LValueToRValue)
7254       O = getObject(E->getSubExpr(), false);
7255 
7256     if (O)
7257       notePreUse(O, E);
7258     VisitExpr(E);
7259     if (O)
7260       notePostUse(O, E);
7261   }
7262 
VisitBinComma(BinaryOperator * BO)7263   void VisitBinComma(BinaryOperator *BO) {
7264     // C++11 [expr.comma]p1:
7265     //   Every value computation and side effect associated with the left
7266     //   expression is sequenced before every value computation and side
7267     //   effect associated with the right expression.
7268     SequenceTree::Seq LHS = Tree.allocate(Region);
7269     SequenceTree::Seq RHS = Tree.allocate(Region);
7270     SequenceTree::Seq OldRegion = Region;
7271 
7272     {
7273       SequencedSubexpression SeqLHS(*this);
7274       Region = LHS;
7275       Visit(BO->getLHS());
7276     }
7277 
7278     Region = RHS;
7279     Visit(BO->getRHS());
7280 
7281     Region = OldRegion;
7282 
7283     // Forget that LHS and RHS are sequenced. They are both unsequenced
7284     // with respect to other stuff.
7285     Tree.merge(LHS);
7286     Tree.merge(RHS);
7287   }
7288 
VisitBinAssign(BinaryOperator * BO)7289   void VisitBinAssign(BinaryOperator *BO) {
7290     // The modification is sequenced after the value computation of the LHS
7291     // and RHS, so check it before inspecting the operands and update the
7292     // map afterwards.
7293     Object O = getObject(BO->getLHS(), true);
7294     if (!O)
7295       return VisitExpr(BO);
7296 
7297     notePreMod(O, BO);
7298 
7299     // C++11 [expr.ass]p7:
7300     //   E1 op= E2 is equivalent to E1 = E1 op E2, except that E1 is evaluated
7301     //   only once.
7302     //
7303     // Therefore, for a compound assignment operator, O is considered used
7304     // everywhere except within the evaluation of E1 itself.
7305     if (isa<CompoundAssignOperator>(BO))
7306       notePreUse(O, BO);
7307 
7308     Visit(BO->getLHS());
7309 
7310     if (isa<CompoundAssignOperator>(BO))
7311       notePostUse(O, BO);
7312 
7313     Visit(BO->getRHS());
7314 
7315     // C++11 [expr.ass]p1:
7316     //   the assignment is sequenced [...] before the value computation of the
7317     //   assignment expression.
7318     // C11 6.5.16/3 has no such rule.
7319     notePostMod(O, BO, SemaRef.getLangOpts().CPlusPlus ? UK_ModAsValue
7320                                                        : UK_ModAsSideEffect);
7321   }
VisitCompoundAssignOperator(CompoundAssignOperator * CAO)7322   void VisitCompoundAssignOperator(CompoundAssignOperator *CAO) {
7323     VisitBinAssign(CAO);
7324   }
7325 
VisitUnaryPreInc(UnaryOperator * UO)7326   void VisitUnaryPreInc(UnaryOperator *UO) { VisitUnaryPreIncDec(UO); }
VisitUnaryPreDec(UnaryOperator * UO)7327   void VisitUnaryPreDec(UnaryOperator *UO) { VisitUnaryPreIncDec(UO); }
VisitUnaryPreIncDec(UnaryOperator * UO)7328   void VisitUnaryPreIncDec(UnaryOperator *UO) {
7329     Object O = getObject(UO->getSubExpr(), true);
7330     if (!O)
7331       return VisitExpr(UO);
7332 
7333     notePreMod(O, UO);
7334     Visit(UO->getSubExpr());
7335     // C++11 [expr.pre.incr]p1:
7336     //   the expression ++x is equivalent to x+=1
7337     notePostMod(O, UO, SemaRef.getLangOpts().CPlusPlus ? UK_ModAsValue
7338                                                        : UK_ModAsSideEffect);
7339   }
7340 
VisitUnaryPostInc(UnaryOperator * UO)7341   void VisitUnaryPostInc(UnaryOperator *UO) { VisitUnaryPostIncDec(UO); }
VisitUnaryPostDec(UnaryOperator * UO)7342   void VisitUnaryPostDec(UnaryOperator *UO) { VisitUnaryPostIncDec(UO); }
VisitUnaryPostIncDec(UnaryOperator * UO)7343   void VisitUnaryPostIncDec(UnaryOperator *UO) {
7344     Object O = getObject(UO->getSubExpr(), true);
7345     if (!O)
7346       return VisitExpr(UO);
7347 
7348     notePreMod(O, UO);
7349     Visit(UO->getSubExpr());
7350     notePostMod(O, UO, UK_ModAsSideEffect);
7351   }
7352 
7353   /// Don't visit the RHS of '&&' or '||' if it might not be evaluated.
VisitBinLOr(BinaryOperator * BO)7354   void VisitBinLOr(BinaryOperator *BO) {
7355     // The side-effects of the LHS of an '&&' are sequenced before the
7356     // value computation of the RHS, and hence before the value computation
7357     // of the '&&' itself, unless the LHS evaluates to zero. We treat them
7358     // as if they were unconditionally sequenced.
7359     EvaluationTracker Eval(*this);
7360     {
7361       SequencedSubexpression Sequenced(*this);
7362       Visit(BO->getLHS());
7363     }
7364 
7365     bool Result;
7366     if (Eval.evaluate(BO->getLHS(), Result)) {
7367       if (!Result)
7368         Visit(BO->getRHS());
7369     } else {
7370       // Check for unsequenced operations in the RHS, treating it as an
7371       // entirely separate evaluation.
7372       //
7373       // FIXME: If there are operations in the RHS which are unsequenced
7374       // with respect to operations outside the RHS, and those operations
7375       // are unconditionally evaluated, diagnose them.
7376       WorkList.push_back(BO->getRHS());
7377     }
7378   }
VisitBinLAnd(BinaryOperator * BO)7379   void VisitBinLAnd(BinaryOperator *BO) {
7380     EvaluationTracker Eval(*this);
7381     {
7382       SequencedSubexpression Sequenced(*this);
7383       Visit(BO->getLHS());
7384     }
7385 
7386     bool Result;
7387     if (Eval.evaluate(BO->getLHS(), Result)) {
7388       if (Result)
7389         Visit(BO->getRHS());
7390     } else {
7391       WorkList.push_back(BO->getRHS());
7392     }
7393   }
7394 
7395   // Only visit the condition, unless we can be sure which subexpression will
7396   // be chosen.
VisitAbstractConditionalOperator(AbstractConditionalOperator * CO)7397   void VisitAbstractConditionalOperator(AbstractConditionalOperator *CO) {
7398     EvaluationTracker Eval(*this);
7399     {
7400       SequencedSubexpression Sequenced(*this);
7401       Visit(CO->getCond());
7402     }
7403 
7404     bool Result;
7405     if (Eval.evaluate(CO->getCond(), Result))
7406       Visit(Result ? CO->getTrueExpr() : CO->getFalseExpr());
7407     else {
7408       WorkList.push_back(CO->getTrueExpr());
7409       WorkList.push_back(CO->getFalseExpr());
7410     }
7411   }
7412 
VisitCallExpr(CallExpr * CE)7413   void VisitCallExpr(CallExpr *CE) {
7414     // C++11 [intro.execution]p15:
7415     //   When calling a function [...], every value computation and side effect
7416     //   associated with any argument expression, or with the postfix expression
7417     //   designating the called function, is sequenced before execution of every
7418     //   expression or statement in the body of the function [and thus before
7419     //   the value computation of its result].
7420     SequencedSubexpression Sequenced(*this);
7421     Base::VisitCallExpr(CE);
7422 
7423     // FIXME: CXXNewExpr and CXXDeleteExpr implicitly call functions.
7424   }
7425 
VisitCXXConstructExpr(CXXConstructExpr * CCE)7426   void VisitCXXConstructExpr(CXXConstructExpr *CCE) {
7427     // This is a call, so all subexpressions are sequenced before the result.
7428     SequencedSubexpression Sequenced(*this);
7429 
7430     if (!CCE->isListInitialization())
7431       return VisitExpr(CCE);
7432 
7433     // In C++11, list initializations are sequenced.
7434     SmallVector<SequenceTree::Seq, 32> Elts;
7435     SequenceTree::Seq Parent = Region;
7436     for (CXXConstructExpr::arg_iterator I = CCE->arg_begin(),
7437                                         E = CCE->arg_end();
7438          I != E; ++I) {
7439       Region = Tree.allocate(Parent);
7440       Elts.push_back(Region);
7441       Visit(*I);
7442     }
7443 
7444     // Forget that the initializers are sequenced.
7445     Region = Parent;
7446     for (unsigned I = 0; I < Elts.size(); ++I)
7447       Tree.merge(Elts[I]);
7448   }
7449 
VisitInitListExpr(InitListExpr * ILE)7450   void VisitInitListExpr(InitListExpr *ILE) {
7451     if (!SemaRef.getLangOpts().CPlusPlus11)
7452       return VisitExpr(ILE);
7453 
7454     // In C++11, list initializations are sequenced.
7455     SmallVector<SequenceTree::Seq, 32> Elts;
7456     SequenceTree::Seq Parent = Region;
7457     for (unsigned I = 0; I < ILE->getNumInits(); ++I) {
7458       Expr *E = ILE->getInit(I);
7459       if (!E) continue;
7460       Region = Tree.allocate(Parent);
7461       Elts.push_back(Region);
7462       Visit(E);
7463     }
7464 
7465     // Forget that the initializers are sequenced.
7466     Region = Parent;
7467     for (unsigned I = 0; I < Elts.size(); ++I)
7468       Tree.merge(Elts[I]);
7469   }
7470 };
7471 }
7472 
CheckUnsequencedOperations(Expr * E)7473 void Sema::CheckUnsequencedOperations(Expr *E) {
7474   SmallVector<Expr *, 8> WorkList;
7475   WorkList.push_back(E);
7476   while (!WorkList.empty()) {
7477     Expr *Item = WorkList.pop_back_val();
7478     SequenceChecker(*this, Item, WorkList);
7479   }
7480 }
7481 
CheckCompletedExpr(Expr * E,SourceLocation CheckLoc,bool IsConstexpr)7482 void Sema::CheckCompletedExpr(Expr *E, SourceLocation CheckLoc,
7483                               bool IsConstexpr) {
7484   CheckImplicitConversions(E, CheckLoc);
7485   CheckUnsequencedOperations(E);
7486   if (!IsConstexpr && !E->isValueDependent())
7487     CheckForIntOverflow(E);
7488 }
7489 
CheckBitFieldInitialization(SourceLocation InitLoc,FieldDecl * BitField,Expr * Init)7490 void Sema::CheckBitFieldInitialization(SourceLocation InitLoc,
7491                                        FieldDecl *BitField,
7492                                        Expr *Init) {
7493   (void) AnalyzeBitFieldAssignment(*this, BitField, Init, InitLoc);
7494 }
7495 
7496 /// CheckParmsForFunctionDef - Check that the parameters of the given
7497 /// function are appropriate for the definition of a function. This
7498 /// takes care of any checks that cannot be performed on the
7499 /// declaration itself, e.g., that the types of each of the function
7500 /// parameters are complete.
CheckParmsForFunctionDef(ParmVarDecl * const * P,ParmVarDecl * const * PEnd,bool CheckParameterNames)7501 bool Sema::CheckParmsForFunctionDef(ParmVarDecl *const *P,
7502                                     ParmVarDecl *const *PEnd,
7503                                     bool CheckParameterNames) {
7504   bool HasInvalidParm = false;
7505   for (; P != PEnd; ++P) {
7506     ParmVarDecl *Param = *P;
7507 
7508     // C99 6.7.5.3p4: the parameters in a parameter type list in a
7509     // function declarator that is part of a function definition of
7510     // that function shall not have incomplete type.
7511     //
7512     // This is also C++ [dcl.fct]p6.
7513     if (!Param->isInvalidDecl() &&
7514         RequireCompleteType(Param->getLocation(), Param->getType(),
7515                             diag::err_typecheck_decl_incomplete_type)) {
7516       Param->setInvalidDecl();
7517       HasInvalidParm = true;
7518     }
7519 
7520     // C99 6.9.1p5: If the declarator includes a parameter type list, the
7521     // declaration of each parameter shall include an identifier.
7522     if (CheckParameterNames &&
7523         Param->getIdentifier() == nullptr &&
7524         !Param->isImplicit() &&
7525         !getLangOpts().CPlusPlus)
7526       Diag(Param->getLocation(), diag::err_parameter_name_omitted);
7527 
7528     // C99 6.7.5.3p12:
7529     //   If the function declarator is not part of a definition of that
7530     //   function, parameters may have incomplete type and may use the [*]
7531     //   notation in their sequences of declarator specifiers to specify
7532     //   variable length array types.
7533     QualType PType = Param->getOriginalType();
7534     while (const ArrayType *AT = Context.getAsArrayType(PType)) {
7535       if (AT->getSizeModifier() == ArrayType::Star) {
7536         // FIXME: This diagnostic should point the '[*]' if source-location
7537         // information is added for it.
7538         Diag(Param->getLocation(), diag::err_array_star_in_function_definition);
7539         break;
7540       }
7541       PType= AT->getElementType();
7542     }
7543 
7544     // MSVC destroys objects passed by value in the callee.  Therefore a
7545     // function definition which takes such a parameter must be able to call the
7546     // object's destructor.  However, we don't perform any direct access check
7547     // on the dtor.
7548     if (getLangOpts().CPlusPlus && Context.getTargetInfo()
7549                                        .getCXXABI()
7550                                        .areArgsDestroyedLeftToRightInCallee()) {
7551       if (!Param->isInvalidDecl()) {
7552         if (const RecordType *RT = Param->getType()->getAs<RecordType>()) {
7553           CXXRecordDecl *ClassDecl = cast<CXXRecordDecl>(RT->getDecl());
7554           if (!ClassDecl->isInvalidDecl() &&
7555               !ClassDecl->hasIrrelevantDestructor() &&
7556               !ClassDecl->isDependentContext()) {
7557             CXXDestructorDecl *Destructor = LookupDestructor(ClassDecl);
7558             MarkFunctionReferenced(Param->getLocation(), Destructor);
7559             DiagnoseUseOfDecl(Destructor, Param->getLocation());
7560           }
7561         }
7562       }
7563     }
7564   }
7565 
7566   return HasInvalidParm;
7567 }
7568 
7569 /// CheckCastAlign - Implements -Wcast-align, which warns when a
7570 /// pointer cast increases the alignment requirements.
CheckCastAlign(Expr * Op,QualType T,SourceRange TRange)7571 void Sema::CheckCastAlign(Expr *Op, QualType T, SourceRange TRange) {
7572   // This is actually a lot of work to potentially be doing on every
7573   // cast; don't do it if we're ignoring -Wcast_align (as is the default).
7574   if (getDiagnostics().isIgnored(diag::warn_cast_align, TRange.getBegin()))
7575     return;
7576 
7577   // Ignore dependent types.
7578   if (T->isDependentType() || Op->getType()->isDependentType())
7579     return;
7580 
7581   // Require that the destination be a pointer type.
7582   const PointerType *DestPtr = T->getAs<PointerType>();
7583   if (!DestPtr) return;
7584 
7585   // If the destination has alignment 1, we're done.
7586   QualType DestPointee = DestPtr->getPointeeType();
7587   if (DestPointee->isIncompleteType()) return;
7588   CharUnits DestAlign = Context.getTypeAlignInChars(DestPointee);
7589   if (DestAlign.isOne()) return;
7590 
7591   // Require that the source be a pointer type.
7592   const PointerType *SrcPtr = Op->getType()->getAs<PointerType>();
7593   if (!SrcPtr) return;
7594   QualType SrcPointee = SrcPtr->getPointeeType();
7595 
7596   // Whitelist casts from cv void*.  We already implicitly
7597   // whitelisted casts to cv void*, since they have alignment 1.
7598   // Also whitelist casts involving incomplete types, which implicitly
7599   // includes 'void'.
7600   if (SrcPointee->isIncompleteType()) return;
7601 
7602   CharUnits SrcAlign = Context.getTypeAlignInChars(SrcPointee);
7603   if (SrcAlign >= DestAlign) return;
7604 
7605   Diag(TRange.getBegin(), diag::warn_cast_align)
7606     << Op->getType() << T
7607     << static_cast<unsigned>(SrcAlign.getQuantity())
7608     << static_cast<unsigned>(DestAlign.getQuantity())
7609     << TRange << Op->getSourceRange();
7610 }
7611 
getElementType(const Expr * BaseExpr)7612 static const Type* getElementType(const Expr *BaseExpr) {
7613   const Type* EltType = BaseExpr->getType().getTypePtr();
7614   if (EltType->isAnyPointerType())
7615     return EltType->getPointeeType().getTypePtr();
7616   else if (EltType->isArrayType())
7617     return EltType->getBaseElementTypeUnsafe();
7618   return EltType;
7619 }
7620 
7621 /// \brief Check whether this array fits the idiom of a size-one tail padded
7622 /// array member of a struct.
7623 ///
7624 /// We avoid emitting out-of-bounds access warnings for such arrays as they are
7625 /// commonly used to emulate flexible arrays in C89 code.
IsTailPaddedMemberArray(Sema & S,llvm::APInt Size,const NamedDecl * ND)7626 static bool IsTailPaddedMemberArray(Sema &S, llvm::APInt Size,
7627                                     const NamedDecl *ND) {
7628   if (Size != 1 || !ND) return false;
7629 
7630   const FieldDecl *FD = dyn_cast<FieldDecl>(ND);
7631   if (!FD) return false;
7632 
7633   // Don't consider sizes resulting from macro expansions or template argument
7634   // substitution to form C89 tail-padded arrays.
7635 
7636   TypeSourceInfo *TInfo = FD->getTypeSourceInfo();
7637   while (TInfo) {
7638     TypeLoc TL = TInfo->getTypeLoc();
7639     // Look through typedefs.
7640     if (TypedefTypeLoc TTL = TL.getAs<TypedefTypeLoc>()) {
7641       const TypedefNameDecl *TDL = TTL.getTypedefNameDecl();
7642       TInfo = TDL->getTypeSourceInfo();
7643       continue;
7644     }
7645     if (ConstantArrayTypeLoc CTL = TL.getAs<ConstantArrayTypeLoc>()) {
7646       const Expr *SizeExpr = dyn_cast<IntegerLiteral>(CTL.getSizeExpr());
7647       if (!SizeExpr || SizeExpr->getExprLoc().isMacroID())
7648         return false;
7649     }
7650     break;
7651   }
7652 
7653   const RecordDecl *RD = dyn_cast<RecordDecl>(FD->getDeclContext());
7654   if (!RD) return false;
7655   if (RD->isUnion()) return false;
7656   if (const CXXRecordDecl *CRD = dyn_cast<CXXRecordDecl>(RD)) {
7657     if (!CRD->isStandardLayout()) return false;
7658   }
7659 
7660   // See if this is the last field decl in the record.
7661   const Decl *D = FD;
7662   while ((D = D->getNextDeclInContext()))
7663     if (isa<FieldDecl>(D))
7664       return false;
7665   return true;
7666 }
7667 
CheckArrayAccess(const Expr * BaseExpr,const Expr * IndexExpr,const ArraySubscriptExpr * ASE,bool AllowOnePastEnd,bool IndexNegated)7668 void Sema::CheckArrayAccess(const Expr *BaseExpr, const Expr *IndexExpr,
7669                             const ArraySubscriptExpr *ASE,
7670                             bool AllowOnePastEnd, bool IndexNegated) {
7671   IndexExpr = IndexExpr->IgnoreParenImpCasts();
7672   if (IndexExpr->isValueDependent())
7673     return;
7674 
7675   const Type *EffectiveType = getElementType(BaseExpr);
7676   BaseExpr = BaseExpr->IgnoreParenCasts();
7677   const ConstantArrayType *ArrayTy =
7678     Context.getAsConstantArrayType(BaseExpr->getType());
7679   if (!ArrayTy)
7680     return;
7681 
7682   llvm::APSInt index;
7683   if (!IndexExpr->EvaluateAsInt(index, Context))
7684     return;
7685   if (IndexNegated)
7686     index = -index;
7687 
7688   const NamedDecl *ND = nullptr;
7689   if (const DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(BaseExpr))
7690     ND = dyn_cast<NamedDecl>(DRE->getDecl());
7691   if (const MemberExpr *ME = dyn_cast<MemberExpr>(BaseExpr))
7692     ND = dyn_cast<NamedDecl>(ME->getMemberDecl());
7693 
7694   if (index.isUnsigned() || !index.isNegative()) {
7695     llvm::APInt size = ArrayTy->getSize();
7696     if (!size.isStrictlyPositive())
7697       return;
7698 
7699     const Type* BaseType = getElementType(BaseExpr);
7700     if (BaseType != EffectiveType) {
7701       // Make sure we're comparing apples to apples when comparing index to size
7702       uint64_t ptrarith_typesize = Context.getTypeSize(EffectiveType);
7703       uint64_t array_typesize = Context.getTypeSize(BaseType);
7704       // Handle ptrarith_typesize being zero, such as when casting to void*
7705       if (!ptrarith_typesize) ptrarith_typesize = 1;
7706       if (ptrarith_typesize != array_typesize) {
7707         // There's a cast to a different size type involved
7708         uint64_t ratio = array_typesize / ptrarith_typesize;
7709         // TODO: Be smarter about handling cases where array_typesize is not a
7710         // multiple of ptrarith_typesize
7711         if (ptrarith_typesize * ratio == array_typesize)
7712           size *= llvm::APInt(size.getBitWidth(), ratio);
7713       }
7714     }
7715 
7716     if (size.getBitWidth() > index.getBitWidth())
7717       index = index.zext(size.getBitWidth());
7718     else if (size.getBitWidth() < index.getBitWidth())
7719       size = size.zext(index.getBitWidth());
7720 
7721     // For array subscripting the index must be less than size, but for pointer
7722     // arithmetic also allow the index (offset) to be equal to size since
7723     // computing the next address after the end of the array is legal and
7724     // commonly done e.g. in C++ iterators and range-based for loops.
7725     if (AllowOnePastEnd ? index.ule(size) : index.ult(size))
7726       return;
7727 
7728     // Also don't warn for arrays of size 1 which are members of some
7729     // structure. These are often used to approximate flexible arrays in C89
7730     // code.
7731     if (IsTailPaddedMemberArray(*this, size, ND))
7732       return;
7733 
7734     // Suppress the warning if the subscript expression (as identified by the
7735     // ']' location) and the index expression are both from macro expansions
7736     // within a system header.
7737     if (ASE) {
7738       SourceLocation RBracketLoc = SourceMgr.getSpellingLoc(
7739           ASE->getRBracketLoc());
7740       if (SourceMgr.isInSystemHeader(RBracketLoc)) {
7741         SourceLocation IndexLoc = SourceMgr.getSpellingLoc(
7742             IndexExpr->getLocStart());
7743         if (SourceMgr.isWrittenInSameFile(RBracketLoc, IndexLoc))
7744           return;
7745       }
7746     }
7747 
7748     unsigned DiagID = diag::warn_ptr_arith_exceeds_bounds;
7749     if (ASE)
7750       DiagID = diag::warn_array_index_exceeds_bounds;
7751 
7752     DiagRuntimeBehavior(BaseExpr->getLocStart(), BaseExpr,
7753                         PDiag(DiagID) << index.toString(10, true)
7754                           << size.toString(10, true)
7755                           << (unsigned)size.getLimitedValue(~0U)
7756                           << IndexExpr->getSourceRange());
7757   } else {
7758     unsigned DiagID = diag::warn_array_index_precedes_bounds;
7759     if (!ASE) {
7760       DiagID = diag::warn_ptr_arith_precedes_bounds;
7761       if (index.isNegative()) index = -index;
7762     }
7763 
7764     DiagRuntimeBehavior(BaseExpr->getLocStart(), BaseExpr,
7765                         PDiag(DiagID) << index.toString(10, true)
7766                           << IndexExpr->getSourceRange());
7767   }
7768 
7769   if (!ND) {
7770     // Try harder to find a NamedDecl to point at in the note.
7771     while (const ArraySubscriptExpr *ASE =
7772            dyn_cast<ArraySubscriptExpr>(BaseExpr))
7773       BaseExpr = ASE->getBase()->IgnoreParenCasts();
7774     if (const DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(BaseExpr))
7775       ND = dyn_cast<NamedDecl>(DRE->getDecl());
7776     if (const MemberExpr *ME = dyn_cast<MemberExpr>(BaseExpr))
7777       ND = dyn_cast<NamedDecl>(ME->getMemberDecl());
7778   }
7779 
7780   if (ND)
7781     DiagRuntimeBehavior(ND->getLocStart(), BaseExpr,
7782                         PDiag(diag::note_array_index_out_of_bounds)
7783                           << ND->getDeclName());
7784 }
7785 
CheckArrayAccess(const Expr * expr)7786 void Sema::CheckArrayAccess(const Expr *expr) {
7787   int AllowOnePastEnd = 0;
7788   while (expr) {
7789     expr = expr->IgnoreParenImpCasts();
7790     switch (expr->getStmtClass()) {
7791       case Stmt::ArraySubscriptExprClass: {
7792         const ArraySubscriptExpr *ASE = cast<ArraySubscriptExpr>(expr);
7793         CheckArrayAccess(ASE->getBase(), ASE->getIdx(), ASE,
7794                          AllowOnePastEnd > 0);
7795         return;
7796       }
7797       case Stmt::UnaryOperatorClass: {
7798         // Only unwrap the * and & unary operators
7799         const UnaryOperator *UO = cast<UnaryOperator>(expr);
7800         expr = UO->getSubExpr();
7801         switch (UO->getOpcode()) {
7802           case UO_AddrOf:
7803             AllowOnePastEnd++;
7804             break;
7805           case UO_Deref:
7806             AllowOnePastEnd--;
7807             break;
7808           default:
7809             return;
7810         }
7811         break;
7812       }
7813       case Stmt::ConditionalOperatorClass: {
7814         const ConditionalOperator *cond = cast<ConditionalOperator>(expr);
7815         if (const Expr *lhs = cond->getLHS())
7816           CheckArrayAccess(lhs);
7817         if (const Expr *rhs = cond->getRHS())
7818           CheckArrayAccess(rhs);
7819         return;
7820       }
7821       default:
7822         return;
7823     }
7824   }
7825 }
7826 
7827 //===--- CHECK: Objective-C retain cycles ----------------------------------//
7828 
7829 namespace {
7830   struct RetainCycleOwner {
RetainCycleOwner__anonb40dc79c0b11::RetainCycleOwner7831     RetainCycleOwner() : Variable(nullptr), Indirect(false) {}
7832     VarDecl *Variable;
7833     SourceRange Range;
7834     SourceLocation Loc;
7835     bool Indirect;
7836 
setLocsFrom__anonb40dc79c0b11::RetainCycleOwner7837     void setLocsFrom(Expr *e) {
7838       Loc = e->getExprLoc();
7839       Range = e->getSourceRange();
7840     }
7841   };
7842 }
7843 
7844 /// Consider whether capturing the given variable can possibly lead to
7845 /// a retain cycle.
considerVariable(VarDecl * var,Expr * ref,RetainCycleOwner & owner)7846 static bool considerVariable(VarDecl *var, Expr *ref, RetainCycleOwner &owner) {
7847   // In ARC, it's captured strongly iff the variable has __strong
7848   // lifetime.  In MRR, it's captured strongly if the variable is
7849   // __block and has an appropriate type.
7850   if (var->getType().getObjCLifetime() != Qualifiers::OCL_Strong)
7851     return false;
7852 
7853   owner.Variable = var;
7854   if (ref)
7855     owner.setLocsFrom(ref);
7856   return true;
7857 }
7858 
findRetainCycleOwner(Sema & S,Expr * e,RetainCycleOwner & owner)7859 static bool findRetainCycleOwner(Sema &S, Expr *e, RetainCycleOwner &owner) {
7860   while (true) {
7861     e = e->IgnoreParens();
7862     if (CastExpr *cast = dyn_cast<CastExpr>(e)) {
7863       switch (cast->getCastKind()) {
7864       case CK_BitCast:
7865       case CK_LValueBitCast:
7866       case CK_LValueToRValue:
7867       case CK_ARCReclaimReturnedObject:
7868         e = cast->getSubExpr();
7869         continue;
7870 
7871       default:
7872         return false;
7873       }
7874     }
7875 
7876     if (ObjCIvarRefExpr *ref = dyn_cast<ObjCIvarRefExpr>(e)) {
7877       ObjCIvarDecl *ivar = ref->getDecl();
7878       if (ivar->getType().getObjCLifetime() != Qualifiers::OCL_Strong)
7879         return false;
7880 
7881       // Try to find a retain cycle in the base.
7882       if (!findRetainCycleOwner(S, ref->getBase(), owner))
7883         return false;
7884 
7885       if (ref->isFreeIvar()) owner.setLocsFrom(ref);
7886       owner.Indirect = true;
7887       return true;
7888     }
7889 
7890     if (DeclRefExpr *ref = dyn_cast<DeclRefExpr>(e)) {
7891       VarDecl *var = dyn_cast<VarDecl>(ref->getDecl());
7892       if (!var) return false;
7893       return considerVariable(var, ref, owner);
7894     }
7895 
7896     if (MemberExpr *member = dyn_cast<MemberExpr>(e)) {
7897       if (member->isArrow()) return false;
7898 
7899       // Don't count this as an indirect ownership.
7900       e = member->getBase();
7901       continue;
7902     }
7903 
7904     if (PseudoObjectExpr *pseudo = dyn_cast<PseudoObjectExpr>(e)) {
7905       // Only pay attention to pseudo-objects on property references.
7906       ObjCPropertyRefExpr *pre
7907         = dyn_cast<ObjCPropertyRefExpr>(pseudo->getSyntacticForm()
7908                                               ->IgnoreParens());
7909       if (!pre) return false;
7910       if (pre->isImplicitProperty()) return false;
7911       ObjCPropertyDecl *property = pre->getExplicitProperty();
7912       if (!property->isRetaining() &&
7913           !(property->getPropertyIvarDecl() &&
7914             property->getPropertyIvarDecl()->getType()
7915               .getObjCLifetime() == Qualifiers::OCL_Strong))
7916           return false;
7917 
7918       owner.Indirect = true;
7919       if (pre->isSuperReceiver()) {
7920         owner.Variable = S.getCurMethodDecl()->getSelfDecl();
7921         if (!owner.Variable)
7922           return false;
7923         owner.Loc = pre->getLocation();
7924         owner.Range = pre->getSourceRange();
7925         return true;
7926       }
7927       e = const_cast<Expr*>(cast<OpaqueValueExpr>(pre->getBase())
7928                               ->getSourceExpr());
7929       continue;
7930     }
7931 
7932     // Array ivars?
7933 
7934     return false;
7935   }
7936 }
7937 
7938 namespace {
7939   struct FindCaptureVisitor : EvaluatedExprVisitor<FindCaptureVisitor> {
FindCaptureVisitor__anonb40dc79c0c11::FindCaptureVisitor7940     FindCaptureVisitor(ASTContext &Context, VarDecl *variable)
7941       : EvaluatedExprVisitor<FindCaptureVisitor>(Context),
7942         Context(Context), Variable(variable), Capturer(nullptr),
7943         VarWillBeReased(false) {}
7944     ASTContext &Context;
7945     VarDecl *Variable;
7946     Expr *Capturer;
7947     bool VarWillBeReased;
7948 
VisitDeclRefExpr__anonb40dc79c0c11::FindCaptureVisitor7949     void VisitDeclRefExpr(DeclRefExpr *ref) {
7950       if (ref->getDecl() == Variable && !Capturer)
7951         Capturer = ref;
7952     }
7953 
VisitObjCIvarRefExpr__anonb40dc79c0c11::FindCaptureVisitor7954     void VisitObjCIvarRefExpr(ObjCIvarRefExpr *ref) {
7955       if (Capturer) return;
7956       Visit(ref->getBase());
7957       if (Capturer && ref->isFreeIvar())
7958         Capturer = ref;
7959     }
7960 
VisitBlockExpr__anonb40dc79c0c11::FindCaptureVisitor7961     void VisitBlockExpr(BlockExpr *block) {
7962       // Look inside nested blocks
7963       if (block->getBlockDecl()->capturesVariable(Variable))
7964         Visit(block->getBlockDecl()->getBody());
7965     }
7966 
VisitOpaqueValueExpr__anonb40dc79c0c11::FindCaptureVisitor7967     void VisitOpaqueValueExpr(OpaqueValueExpr *OVE) {
7968       if (Capturer) return;
7969       if (OVE->getSourceExpr())
7970         Visit(OVE->getSourceExpr());
7971     }
VisitBinaryOperator__anonb40dc79c0c11::FindCaptureVisitor7972     void VisitBinaryOperator(BinaryOperator *BinOp) {
7973       if (!Variable || VarWillBeReased || BinOp->getOpcode() != BO_Assign)
7974         return;
7975       Expr *LHS = BinOp->getLHS();
7976       if (const DeclRefExpr *DRE = dyn_cast_or_null<DeclRefExpr>(LHS)) {
7977         if (DRE->getDecl() != Variable)
7978           return;
7979         if (Expr *RHS = BinOp->getRHS()) {
7980           RHS = RHS->IgnoreParenCasts();
7981           llvm::APSInt Value;
7982           VarWillBeReased =
7983             (RHS && RHS->isIntegerConstantExpr(Value, Context) && Value == 0);
7984         }
7985       }
7986     }
7987   };
7988 }
7989 
7990 /// Check whether the given argument is a block which captures a
7991 /// variable.
findCapturingExpr(Sema & S,Expr * e,RetainCycleOwner & owner)7992 static Expr *findCapturingExpr(Sema &S, Expr *e, RetainCycleOwner &owner) {
7993   assert(owner.Variable && owner.Loc.isValid());
7994 
7995   e = e->IgnoreParenCasts();
7996 
7997   // Look through [^{...} copy] and Block_copy(^{...}).
7998   if (ObjCMessageExpr *ME = dyn_cast<ObjCMessageExpr>(e)) {
7999     Selector Cmd = ME->getSelector();
8000     if (Cmd.isUnarySelector() && Cmd.getNameForSlot(0) == "copy") {
8001       e = ME->getInstanceReceiver();
8002       if (!e)
8003         return nullptr;
8004       e = e->IgnoreParenCasts();
8005     }
8006   } else if (CallExpr *CE = dyn_cast<CallExpr>(e)) {
8007     if (CE->getNumArgs() == 1) {
8008       FunctionDecl *Fn = dyn_cast_or_null<FunctionDecl>(CE->getCalleeDecl());
8009       if (Fn) {
8010         const IdentifierInfo *FnI = Fn->getIdentifier();
8011         if (FnI && FnI->isStr("_Block_copy")) {
8012           e = CE->getArg(0)->IgnoreParenCasts();
8013         }
8014       }
8015     }
8016   }
8017 
8018   BlockExpr *block = dyn_cast<BlockExpr>(e);
8019   if (!block || !block->getBlockDecl()->capturesVariable(owner.Variable))
8020     return nullptr;
8021 
8022   FindCaptureVisitor visitor(S.Context, owner.Variable);
8023   visitor.Visit(block->getBlockDecl()->getBody());
8024   return visitor.VarWillBeReased ? nullptr : visitor.Capturer;
8025 }
8026 
diagnoseRetainCycle(Sema & S,Expr * capturer,RetainCycleOwner & owner)8027 static void diagnoseRetainCycle(Sema &S, Expr *capturer,
8028                                 RetainCycleOwner &owner) {
8029   assert(capturer);
8030   assert(owner.Variable && owner.Loc.isValid());
8031 
8032   S.Diag(capturer->getExprLoc(), diag::warn_arc_retain_cycle)
8033     << owner.Variable << capturer->getSourceRange();
8034   S.Diag(owner.Loc, diag::note_arc_retain_cycle_owner)
8035     << owner.Indirect << owner.Range;
8036 }
8037 
8038 /// Check for a keyword selector that starts with the word 'add' or
8039 /// 'set'.
isSetterLikeSelector(Selector sel)8040 static bool isSetterLikeSelector(Selector sel) {
8041   if (sel.isUnarySelector()) return false;
8042 
8043   StringRef str = sel.getNameForSlot(0);
8044   while (!str.empty() && str.front() == '_') str = str.substr(1);
8045   if (str.startswith("set"))
8046     str = str.substr(3);
8047   else if (str.startswith("add")) {
8048     // Specially whitelist 'addOperationWithBlock:'.
8049     if (sel.getNumArgs() == 1 && str.startswith("addOperationWithBlock"))
8050       return false;
8051     str = str.substr(3);
8052   }
8053   else
8054     return false;
8055 
8056   if (str.empty()) return true;
8057   return !isLowercase(str.front());
8058 }
8059 
8060 /// Check a message send to see if it's likely to cause a retain cycle.
checkRetainCycles(ObjCMessageExpr * msg)8061 void Sema::checkRetainCycles(ObjCMessageExpr *msg) {
8062   // Only check instance methods whose selector looks like a setter.
8063   if (!msg->isInstanceMessage() || !isSetterLikeSelector(msg->getSelector()))
8064     return;
8065 
8066   // Try to find a variable that the receiver is strongly owned by.
8067   RetainCycleOwner owner;
8068   if (msg->getReceiverKind() == ObjCMessageExpr::Instance) {
8069     if (!findRetainCycleOwner(*this, msg->getInstanceReceiver(), owner))
8070       return;
8071   } else {
8072     assert(msg->getReceiverKind() == ObjCMessageExpr::SuperInstance);
8073     owner.Variable = getCurMethodDecl()->getSelfDecl();
8074     owner.Loc = msg->getSuperLoc();
8075     owner.Range = msg->getSuperLoc();
8076   }
8077 
8078   // Check whether the receiver is captured by any of the arguments.
8079   for (unsigned i = 0, e = msg->getNumArgs(); i != e; ++i)
8080     if (Expr *capturer = findCapturingExpr(*this, msg->getArg(i), owner))
8081       return diagnoseRetainCycle(*this, capturer, owner);
8082 }
8083 
8084 /// Check a property assign to see if it's likely to cause a retain cycle.
checkRetainCycles(Expr * receiver,Expr * argument)8085 void Sema::checkRetainCycles(Expr *receiver, Expr *argument) {
8086   RetainCycleOwner owner;
8087   if (!findRetainCycleOwner(*this, receiver, owner))
8088     return;
8089 
8090   if (Expr *capturer = findCapturingExpr(*this, argument, owner))
8091     diagnoseRetainCycle(*this, capturer, owner);
8092 }
8093 
checkRetainCycles(VarDecl * Var,Expr * Init)8094 void Sema::checkRetainCycles(VarDecl *Var, Expr *Init) {
8095   RetainCycleOwner Owner;
8096   if (!considerVariable(Var, /*DeclRefExpr=*/nullptr, Owner))
8097     return;
8098 
8099   // Because we don't have an expression for the variable, we have to set the
8100   // location explicitly here.
8101   Owner.Loc = Var->getLocation();
8102   Owner.Range = Var->getSourceRange();
8103 
8104   if (Expr *Capturer = findCapturingExpr(*this, Init, Owner))
8105     diagnoseRetainCycle(*this, Capturer, Owner);
8106 }
8107 
checkUnsafeAssignLiteral(Sema & S,SourceLocation Loc,Expr * RHS,bool isProperty)8108 static bool checkUnsafeAssignLiteral(Sema &S, SourceLocation Loc,
8109                                      Expr *RHS, bool isProperty) {
8110   // Check if RHS is an Objective-C object literal, which also can get
8111   // immediately zapped in a weak reference.  Note that we explicitly
8112   // allow ObjCStringLiterals, since those are designed to never really die.
8113   RHS = RHS->IgnoreParenImpCasts();
8114 
8115   // This enum needs to match with the 'select' in
8116   // warn_objc_arc_literal_assign (off-by-1).
8117   Sema::ObjCLiteralKind Kind = S.CheckLiteralKind(RHS);
8118   if (Kind == Sema::LK_String || Kind == Sema::LK_None)
8119     return false;
8120 
8121   S.Diag(Loc, diag::warn_arc_literal_assign)
8122     << (unsigned) Kind
8123     << (isProperty ? 0 : 1)
8124     << RHS->getSourceRange();
8125 
8126   return true;
8127 }
8128 
checkUnsafeAssignObject(Sema & S,SourceLocation Loc,Qualifiers::ObjCLifetime LT,Expr * RHS,bool isProperty)8129 static bool checkUnsafeAssignObject(Sema &S, SourceLocation Loc,
8130                                     Qualifiers::ObjCLifetime LT,
8131                                     Expr *RHS, bool isProperty) {
8132   // Strip off any implicit cast added to get to the one ARC-specific.
8133   while (ImplicitCastExpr *cast = dyn_cast<ImplicitCastExpr>(RHS)) {
8134     if (cast->getCastKind() == CK_ARCConsumeObject) {
8135       S.Diag(Loc, diag::warn_arc_retained_assign)
8136         << (LT == Qualifiers::OCL_ExplicitNone)
8137         << (isProperty ? 0 : 1)
8138         << RHS->getSourceRange();
8139       return true;
8140     }
8141     RHS = cast->getSubExpr();
8142   }
8143 
8144   if (LT == Qualifiers::OCL_Weak &&
8145       checkUnsafeAssignLiteral(S, Loc, RHS, isProperty))
8146     return true;
8147 
8148   return false;
8149 }
8150 
checkUnsafeAssigns(SourceLocation Loc,QualType LHS,Expr * RHS)8151 bool Sema::checkUnsafeAssigns(SourceLocation Loc,
8152                               QualType LHS, Expr *RHS) {
8153   Qualifiers::ObjCLifetime LT = LHS.getObjCLifetime();
8154 
8155   if (LT != Qualifiers::OCL_Weak && LT != Qualifiers::OCL_ExplicitNone)
8156     return false;
8157 
8158   if (checkUnsafeAssignObject(*this, Loc, LT, RHS, false))
8159     return true;
8160 
8161   return false;
8162 }
8163 
checkUnsafeExprAssigns(SourceLocation Loc,Expr * LHS,Expr * RHS)8164 void Sema::checkUnsafeExprAssigns(SourceLocation Loc,
8165                               Expr *LHS, Expr *RHS) {
8166   QualType LHSType;
8167   // PropertyRef on LHS type need be directly obtained from
8168   // its declaration as it has a PseudoType.
8169   ObjCPropertyRefExpr *PRE
8170     = dyn_cast<ObjCPropertyRefExpr>(LHS->IgnoreParens());
8171   if (PRE && !PRE->isImplicitProperty()) {
8172     const ObjCPropertyDecl *PD = PRE->getExplicitProperty();
8173     if (PD)
8174       LHSType = PD->getType();
8175   }
8176 
8177   if (LHSType.isNull())
8178     LHSType = LHS->getType();
8179 
8180   Qualifiers::ObjCLifetime LT = LHSType.getObjCLifetime();
8181 
8182   if (LT == Qualifiers::OCL_Weak) {
8183     if (!Diags.isIgnored(diag::warn_arc_repeated_use_of_weak, Loc))
8184       getCurFunction()->markSafeWeakUse(LHS);
8185   }
8186 
8187   if (checkUnsafeAssigns(Loc, LHSType, RHS))
8188     return;
8189 
8190   // FIXME. Check for other life times.
8191   if (LT != Qualifiers::OCL_None)
8192     return;
8193 
8194   if (PRE) {
8195     if (PRE->isImplicitProperty())
8196       return;
8197     const ObjCPropertyDecl *PD = PRE->getExplicitProperty();
8198     if (!PD)
8199       return;
8200 
8201     unsigned Attributes = PD->getPropertyAttributes();
8202     if (Attributes & ObjCPropertyDecl::OBJC_PR_assign) {
8203       // when 'assign' attribute was not explicitly specified
8204       // by user, ignore it and rely on property type itself
8205       // for lifetime info.
8206       unsigned AsWrittenAttr = PD->getPropertyAttributesAsWritten();
8207       if (!(AsWrittenAttr & ObjCPropertyDecl::OBJC_PR_assign) &&
8208           LHSType->isObjCRetainableType())
8209         return;
8210 
8211       while (ImplicitCastExpr *cast = dyn_cast<ImplicitCastExpr>(RHS)) {
8212         if (cast->getCastKind() == CK_ARCConsumeObject) {
8213           Diag(Loc, diag::warn_arc_retained_property_assign)
8214           << RHS->getSourceRange();
8215           return;
8216         }
8217         RHS = cast->getSubExpr();
8218       }
8219     }
8220     else if (Attributes & ObjCPropertyDecl::OBJC_PR_weak) {
8221       if (checkUnsafeAssignObject(*this, Loc, Qualifiers::OCL_Weak, RHS, true))
8222         return;
8223     }
8224   }
8225 }
8226 
8227 //===--- CHECK: Empty statement body (-Wempty-body) ---------------------===//
8228 
8229 namespace {
ShouldDiagnoseEmptyStmtBody(const SourceManager & SourceMgr,SourceLocation StmtLoc,const NullStmt * Body)8230 bool ShouldDiagnoseEmptyStmtBody(const SourceManager &SourceMgr,
8231                                  SourceLocation StmtLoc,
8232                                  const NullStmt *Body) {
8233   // Do not warn if the body is a macro that expands to nothing, e.g:
8234   //
8235   // #define CALL(x)
8236   // if (condition)
8237   //   CALL(0);
8238   //
8239   if (Body->hasLeadingEmptyMacro())
8240     return false;
8241 
8242   // Get line numbers of statement and body.
8243   bool StmtLineInvalid;
8244   unsigned StmtLine = SourceMgr.getSpellingLineNumber(StmtLoc,
8245                                                       &StmtLineInvalid);
8246   if (StmtLineInvalid)
8247     return false;
8248 
8249   bool BodyLineInvalid;
8250   unsigned BodyLine = SourceMgr.getSpellingLineNumber(Body->getSemiLoc(),
8251                                                       &BodyLineInvalid);
8252   if (BodyLineInvalid)
8253     return false;
8254 
8255   // Warn if null statement and body are on the same line.
8256   if (StmtLine != BodyLine)
8257     return false;
8258 
8259   return true;
8260 }
8261 } // Unnamed namespace
8262 
DiagnoseEmptyStmtBody(SourceLocation StmtLoc,const Stmt * Body,unsigned DiagID)8263 void Sema::DiagnoseEmptyStmtBody(SourceLocation StmtLoc,
8264                                  const Stmt *Body,
8265                                  unsigned DiagID) {
8266   // Since this is a syntactic check, don't emit diagnostic for template
8267   // instantiations, this just adds noise.
8268   if (CurrentInstantiationScope)
8269     return;
8270 
8271   // The body should be a null statement.
8272   const NullStmt *NBody = dyn_cast<NullStmt>(Body);
8273   if (!NBody)
8274     return;
8275 
8276   // Do the usual checks.
8277   if (!ShouldDiagnoseEmptyStmtBody(SourceMgr, StmtLoc, NBody))
8278     return;
8279 
8280   Diag(NBody->getSemiLoc(), DiagID);
8281   Diag(NBody->getSemiLoc(), diag::note_empty_body_on_separate_line);
8282 }
8283 
DiagnoseEmptyLoopBody(const Stmt * S,const Stmt * PossibleBody)8284 void Sema::DiagnoseEmptyLoopBody(const Stmt *S,
8285                                  const Stmt *PossibleBody) {
8286   assert(!CurrentInstantiationScope); // Ensured by caller
8287 
8288   SourceLocation StmtLoc;
8289   const Stmt *Body;
8290   unsigned DiagID;
8291   if (const ForStmt *FS = dyn_cast<ForStmt>(S)) {
8292     StmtLoc = FS->getRParenLoc();
8293     Body = FS->getBody();
8294     DiagID = diag::warn_empty_for_body;
8295   } else if (const WhileStmt *WS = dyn_cast<WhileStmt>(S)) {
8296     StmtLoc = WS->getCond()->getSourceRange().getEnd();
8297     Body = WS->getBody();
8298     DiagID = diag::warn_empty_while_body;
8299   } else
8300     return; // Neither `for' nor `while'.
8301 
8302   // The body should be a null statement.
8303   const NullStmt *NBody = dyn_cast<NullStmt>(Body);
8304   if (!NBody)
8305     return;
8306 
8307   // Skip expensive checks if diagnostic is disabled.
8308   if (Diags.isIgnored(DiagID, NBody->getSemiLoc()))
8309     return;
8310 
8311   // Do the usual checks.
8312   if (!ShouldDiagnoseEmptyStmtBody(SourceMgr, StmtLoc, NBody))
8313     return;
8314 
8315   // `for(...);' and `while(...);' are popular idioms, so in order to keep
8316   // noise level low, emit diagnostics only if for/while is followed by a
8317   // CompoundStmt, e.g.:
8318   //    for (int i = 0; i < n; i++);
8319   //    {
8320   //      a(i);
8321   //    }
8322   // or if for/while is followed by a statement with more indentation
8323   // than for/while itself:
8324   //    for (int i = 0; i < n; i++);
8325   //      a(i);
8326   bool ProbableTypo = isa<CompoundStmt>(PossibleBody);
8327   if (!ProbableTypo) {
8328     bool BodyColInvalid;
8329     unsigned BodyCol = SourceMgr.getPresumedColumnNumber(
8330                              PossibleBody->getLocStart(),
8331                              &BodyColInvalid);
8332     if (BodyColInvalid)
8333       return;
8334 
8335     bool StmtColInvalid;
8336     unsigned StmtCol = SourceMgr.getPresumedColumnNumber(
8337                              S->getLocStart(),
8338                              &StmtColInvalid);
8339     if (StmtColInvalid)
8340       return;
8341 
8342     if (BodyCol > StmtCol)
8343       ProbableTypo = true;
8344   }
8345 
8346   if (ProbableTypo) {
8347     Diag(NBody->getSemiLoc(), DiagID);
8348     Diag(NBody->getSemiLoc(), diag::note_empty_body_on_separate_line);
8349   }
8350 }
8351 
8352 //===--- CHECK: Warn on self move with std::move. -------------------------===//
8353 
8354 /// DiagnoseSelfMove - Emits a warning if a value is moved to itself.
DiagnoseSelfMove(const Expr * LHSExpr,const Expr * RHSExpr,SourceLocation OpLoc)8355 void Sema::DiagnoseSelfMove(const Expr *LHSExpr, const Expr *RHSExpr,
8356                              SourceLocation OpLoc) {
8357 
8358   if (Diags.isIgnored(diag::warn_sizeof_pointer_expr_memaccess, OpLoc))
8359     return;
8360 
8361   if (!ActiveTemplateInstantiations.empty())
8362     return;
8363 
8364   // Strip parens and casts away.
8365   LHSExpr = LHSExpr->IgnoreParenImpCasts();
8366   RHSExpr = RHSExpr->IgnoreParenImpCasts();
8367 
8368   // Check for a call expression
8369   const CallExpr *CE = dyn_cast<CallExpr>(RHSExpr);
8370   if (!CE || CE->getNumArgs() != 1)
8371     return;
8372 
8373   // Check for a call to std::move
8374   const FunctionDecl *FD = CE->getDirectCallee();
8375   if (!FD || !FD->isInStdNamespace() || !FD->getIdentifier() ||
8376       !FD->getIdentifier()->isStr("move"))
8377     return;
8378 
8379   // Get argument from std::move
8380   RHSExpr = CE->getArg(0);
8381 
8382   const DeclRefExpr *LHSDeclRef = dyn_cast<DeclRefExpr>(LHSExpr);
8383   const DeclRefExpr *RHSDeclRef = dyn_cast<DeclRefExpr>(RHSExpr);
8384 
8385   // Two DeclRefExpr's, check that the decls are the same.
8386   if (LHSDeclRef && RHSDeclRef) {
8387     if (!LHSDeclRef->getDecl() || !RHSDeclRef->getDecl())
8388       return;
8389     if (LHSDeclRef->getDecl()->getCanonicalDecl() !=
8390         RHSDeclRef->getDecl()->getCanonicalDecl())
8391       return;
8392 
8393     Diag(OpLoc, diag::warn_self_move) << LHSExpr->getType()
8394                                         << LHSExpr->getSourceRange()
8395                                         << RHSExpr->getSourceRange();
8396     return;
8397   }
8398 
8399   // Member variables require a different approach to check for self moves.
8400   // MemberExpr's are the same if every nested MemberExpr refers to the same
8401   // Decl and that the base Expr's are DeclRefExpr's with the same Decl or
8402   // the base Expr's are CXXThisExpr's.
8403   const Expr *LHSBase = LHSExpr;
8404   const Expr *RHSBase = RHSExpr;
8405   const MemberExpr *LHSME = dyn_cast<MemberExpr>(LHSExpr);
8406   const MemberExpr *RHSME = dyn_cast<MemberExpr>(RHSExpr);
8407   if (!LHSME || !RHSME)
8408     return;
8409 
8410   while (LHSME && RHSME) {
8411     if (LHSME->getMemberDecl()->getCanonicalDecl() !=
8412         RHSME->getMemberDecl()->getCanonicalDecl())
8413       return;
8414 
8415     LHSBase = LHSME->getBase();
8416     RHSBase = RHSME->getBase();
8417     LHSME = dyn_cast<MemberExpr>(LHSBase);
8418     RHSME = dyn_cast<MemberExpr>(RHSBase);
8419   }
8420 
8421   LHSDeclRef = dyn_cast<DeclRefExpr>(LHSBase);
8422   RHSDeclRef = dyn_cast<DeclRefExpr>(RHSBase);
8423   if (LHSDeclRef && RHSDeclRef) {
8424     if (!LHSDeclRef->getDecl() || !RHSDeclRef->getDecl())
8425       return;
8426     if (LHSDeclRef->getDecl()->getCanonicalDecl() !=
8427         RHSDeclRef->getDecl()->getCanonicalDecl())
8428       return;
8429 
8430     Diag(OpLoc, diag::warn_self_move) << LHSExpr->getType()
8431                                         << LHSExpr->getSourceRange()
8432                                         << RHSExpr->getSourceRange();
8433     return;
8434   }
8435 
8436   if (isa<CXXThisExpr>(LHSBase) && isa<CXXThisExpr>(RHSBase))
8437     Diag(OpLoc, diag::warn_self_move) << LHSExpr->getType()
8438                                         << LHSExpr->getSourceRange()
8439                                         << RHSExpr->getSourceRange();
8440 }
8441 
8442 //===--- Layout compatibility ----------------------------------------------//
8443 
8444 namespace {
8445 
8446 bool isLayoutCompatible(ASTContext &C, QualType T1, QualType T2);
8447 
8448 /// \brief Check if two enumeration types are layout-compatible.
isLayoutCompatible(ASTContext & C,EnumDecl * ED1,EnumDecl * ED2)8449 bool isLayoutCompatible(ASTContext &C, EnumDecl *ED1, EnumDecl *ED2) {
8450   // C++11 [dcl.enum] p8:
8451   // Two enumeration types are layout-compatible if they have the same
8452   // underlying type.
8453   return ED1->isComplete() && ED2->isComplete() &&
8454          C.hasSameType(ED1->getIntegerType(), ED2->getIntegerType());
8455 }
8456 
8457 /// \brief Check if two fields are layout-compatible.
isLayoutCompatible(ASTContext & C,FieldDecl * Field1,FieldDecl * Field2)8458 bool isLayoutCompatible(ASTContext &C, FieldDecl *Field1, FieldDecl *Field2) {
8459   if (!isLayoutCompatible(C, Field1->getType(), Field2->getType()))
8460     return false;
8461 
8462   if (Field1->isBitField() != Field2->isBitField())
8463     return false;
8464 
8465   if (Field1->isBitField()) {
8466     // Make sure that the bit-fields are the same length.
8467     unsigned Bits1 = Field1->getBitWidthValue(C);
8468     unsigned Bits2 = Field2->getBitWidthValue(C);
8469 
8470     if (Bits1 != Bits2)
8471       return false;
8472   }
8473 
8474   return true;
8475 }
8476 
8477 /// \brief Check if two standard-layout structs are layout-compatible.
8478 /// (C++11 [class.mem] p17)
isLayoutCompatibleStruct(ASTContext & C,RecordDecl * RD1,RecordDecl * RD2)8479 bool isLayoutCompatibleStruct(ASTContext &C,
8480                               RecordDecl *RD1,
8481                               RecordDecl *RD2) {
8482   // If both records are C++ classes, check that base classes match.
8483   if (const CXXRecordDecl *D1CXX = dyn_cast<CXXRecordDecl>(RD1)) {
8484     // If one of records is a CXXRecordDecl we are in C++ mode,
8485     // thus the other one is a CXXRecordDecl, too.
8486     const CXXRecordDecl *D2CXX = cast<CXXRecordDecl>(RD2);
8487     // Check number of base classes.
8488     if (D1CXX->getNumBases() != D2CXX->getNumBases())
8489       return false;
8490 
8491     // Check the base classes.
8492     for (CXXRecordDecl::base_class_const_iterator
8493                Base1 = D1CXX->bases_begin(),
8494            BaseEnd1 = D1CXX->bases_end(),
8495               Base2 = D2CXX->bases_begin();
8496          Base1 != BaseEnd1;
8497          ++Base1, ++Base2) {
8498       if (!isLayoutCompatible(C, Base1->getType(), Base2->getType()))
8499         return false;
8500     }
8501   } else if (const CXXRecordDecl *D2CXX = dyn_cast<CXXRecordDecl>(RD2)) {
8502     // If only RD2 is a C++ class, it should have zero base classes.
8503     if (D2CXX->getNumBases() > 0)
8504       return false;
8505   }
8506 
8507   // Check the fields.
8508   RecordDecl::field_iterator Field2 = RD2->field_begin(),
8509                              Field2End = RD2->field_end(),
8510                              Field1 = RD1->field_begin(),
8511                              Field1End = RD1->field_end();
8512   for ( ; Field1 != Field1End && Field2 != Field2End; ++Field1, ++Field2) {
8513     if (!isLayoutCompatible(C, *Field1, *Field2))
8514       return false;
8515   }
8516   if (Field1 != Field1End || Field2 != Field2End)
8517     return false;
8518 
8519   return true;
8520 }
8521 
8522 /// \brief Check if two standard-layout unions are layout-compatible.
8523 /// (C++11 [class.mem] p18)
isLayoutCompatibleUnion(ASTContext & C,RecordDecl * RD1,RecordDecl * RD2)8524 bool isLayoutCompatibleUnion(ASTContext &C,
8525                              RecordDecl *RD1,
8526                              RecordDecl *RD2) {
8527   llvm::SmallPtrSet<FieldDecl *, 8> UnmatchedFields;
8528   for (auto *Field2 : RD2->fields())
8529     UnmatchedFields.insert(Field2);
8530 
8531   for (auto *Field1 : RD1->fields()) {
8532     llvm::SmallPtrSet<FieldDecl *, 8>::iterator
8533         I = UnmatchedFields.begin(),
8534         E = UnmatchedFields.end();
8535 
8536     for ( ; I != E; ++I) {
8537       if (isLayoutCompatible(C, Field1, *I)) {
8538         bool Result = UnmatchedFields.erase(*I);
8539         (void) Result;
8540         assert(Result);
8541         break;
8542       }
8543     }
8544     if (I == E)
8545       return false;
8546   }
8547 
8548   return UnmatchedFields.empty();
8549 }
8550 
isLayoutCompatible(ASTContext & C,RecordDecl * RD1,RecordDecl * RD2)8551 bool isLayoutCompatible(ASTContext &C, RecordDecl *RD1, RecordDecl *RD2) {
8552   if (RD1->isUnion() != RD2->isUnion())
8553     return false;
8554 
8555   if (RD1->isUnion())
8556     return isLayoutCompatibleUnion(C, RD1, RD2);
8557   else
8558     return isLayoutCompatibleStruct(C, RD1, RD2);
8559 }
8560 
8561 /// \brief Check if two types are layout-compatible in C++11 sense.
isLayoutCompatible(ASTContext & C,QualType T1,QualType T2)8562 bool isLayoutCompatible(ASTContext &C, QualType T1, QualType T2) {
8563   if (T1.isNull() || T2.isNull())
8564     return false;
8565 
8566   // C++11 [basic.types] p11:
8567   // If two types T1 and T2 are the same type, then T1 and T2 are
8568   // layout-compatible types.
8569   if (C.hasSameType(T1, T2))
8570     return true;
8571 
8572   T1 = T1.getCanonicalType().getUnqualifiedType();
8573   T2 = T2.getCanonicalType().getUnqualifiedType();
8574 
8575   const Type::TypeClass TC1 = T1->getTypeClass();
8576   const Type::TypeClass TC2 = T2->getTypeClass();
8577 
8578   if (TC1 != TC2)
8579     return false;
8580 
8581   if (TC1 == Type::Enum) {
8582     return isLayoutCompatible(C,
8583                               cast<EnumType>(T1)->getDecl(),
8584                               cast<EnumType>(T2)->getDecl());
8585   } else if (TC1 == Type::Record) {
8586     if (!T1->isStandardLayoutType() || !T2->isStandardLayoutType())
8587       return false;
8588 
8589     return isLayoutCompatible(C,
8590                               cast<RecordType>(T1)->getDecl(),
8591                               cast<RecordType>(T2)->getDecl());
8592   }
8593 
8594   return false;
8595 }
8596 }
8597 
8598 //===--- CHECK: pointer_with_type_tag attribute: datatypes should match ----//
8599 
8600 namespace {
8601 /// \brief Given a type tag expression find the type tag itself.
8602 ///
8603 /// \param TypeExpr Type tag expression, as it appears in user's code.
8604 ///
8605 /// \param VD Declaration of an identifier that appears in a type tag.
8606 ///
8607 /// \param MagicValue Type tag magic value.
FindTypeTagExpr(const Expr * TypeExpr,const ASTContext & Ctx,const ValueDecl ** VD,uint64_t * MagicValue)8608 bool FindTypeTagExpr(const Expr *TypeExpr, const ASTContext &Ctx,
8609                      const ValueDecl **VD, uint64_t *MagicValue) {
8610   while(true) {
8611     if (!TypeExpr)
8612       return false;
8613 
8614     TypeExpr = TypeExpr->IgnoreParenImpCasts()->IgnoreParenCasts();
8615 
8616     switch (TypeExpr->getStmtClass()) {
8617     case Stmt::UnaryOperatorClass: {
8618       const UnaryOperator *UO = cast<UnaryOperator>(TypeExpr);
8619       if (UO->getOpcode() == UO_AddrOf || UO->getOpcode() == UO_Deref) {
8620         TypeExpr = UO->getSubExpr();
8621         continue;
8622       }
8623       return false;
8624     }
8625 
8626     case Stmt::DeclRefExprClass: {
8627       const DeclRefExpr *DRE = cast<DeclRefExpr>(TypeExpr);
8628       *VD = DRE->getDecl();
8629       return true;
8630     }
8631 
8632     case Stmt::IntegerLiteralClass: {
8633       const IntegerLiteral *IL = cast<IntegerLiteral>(TypeExpr);
8634       llvm::APInt MagicValueAPInt = IL->getValue();
8635       if (MagicValueAPInt.getActiveBits() <= 64) {
8636         *MagicValue = MagicValueAPInt.getZExtValue();
8637         return true;
8638       } else
8639         return false;
8640     }
8641 
8642     case Stmt::BinaryConditionalOperatorClass:
8643     case Stmt::ConditionalOperatorClass: {
8644       const AbstractConditionalOperator *ACO =
8645           cast<AbstractConditionalOperator>(TypeExpr);
8646       bool Result;
8647       if (ACO->getCond()->EvaluateAsBooleanCondition(Result, Ctx)) {
8648         if (Result)
8649           TypeExpr = ACO->getTrueExpr();
8650         else
8651           TypeExpr = ACO->getFalseExpr();
8652         continue;
8653       }
8654       return false;
8655     }
8656 
8657     case Stmt::BinaryOperatorClass: {
8658       const BinaryOperator *BO = cast<BinaryOperator>(TypeExpr);
8659       if (BO->getOpcode() == BO_Comma) {
8660         TypeExpr = BO->getRHS();
8661         continue;
8662       }
8663       return false;
8664     }
8665 
8666     default:
8667       return false;
8668     }
8669   }
8670 }
8671 
8672 /// \brief Retrieve the C type corresponding to type tag TypeExpr.
8673 ///
8674 /// \param TypeExpr Expression that specifies a type tag.
8675 ///
8676 /// \param MagicValues Registered magic values.
8677 ///
8678 /// \param FoundWrongKind Set to true if a type tag was found, but of a wrong
8679 ///        kind.
8680 ///
8681 /// \param TypeInfo Information about the corresponding C type.
8682 ///
8683 /// \returns true if the corresponding C type was found.
GetMatchingCType(const IdentifierInfo * ArgumentKind,const Expr * TypeExpr,const ASTContext & Ctx,const llvm::DenseMap<Sema::TypeTagMagicValue,Sema::TypeTagData> * MagicValues,bool & FoundWrongKind,Sema::TypeTagData & TypeInfo)8684 bool GetMatchingCType(
8685         const IdentifierInfo *ArgumentKind,
8686         const Expr *TypeExpr, const ASTContext &Ctx,
8687         const llvm::DenseMap<Sema::TypeTagMagicValue,
8688                              Sema::TypeTagData> *MagicValues,
8689         bool &FoundWrongKind,
8690         Sema::TypeTagData &TypeInfo) {
8691   FoundWrongKind = false;
8692 
8693   // Variable declaration that has type_tag_for_datatype attribute.
8694   const ValueDecl *VD = nullptr;
8695 
8696   uint64_t MagicValue;
8697 
8698   if (!FindTypeTagExpr(TypeExpr, Ctx, &VD, &MagicValue))
8699     return false;
8700 
8701   if (VD) {
8702     if (TypeTagForDatatypeAttr *I = VD->getAttr<TypeTagForDatatypeAttr>()) {
8703       if (I->getArgumentKind() != ArgumentKind) {
8704         FoundWrongKind = true;
8705         return false;
8706       }
8707       TypeInfo.Type = I->getMatchingCType();
8708       TypeInfo.LayoutCompatible = I->getLayoutCompatible();
8709       TypeInfo.MustBeNull = I->getMustBeNull();
8710       return true;
8711     }
8712     return false;
8713   }
8714 
8715   if (!MagicValues)
8716     return false;
8717 
8718   llvm::DenseMap<Sema::TypeTagMagicValue,
8719                  Sema::TypeTagData>::const_iterator I =
8720       MagicValues->find(std::make_pair(ArgumentKind, MagicValue));
8721   if (I == MagicValues->end())
8722     return false;
8723 
8724   TypeInfo = I->second;
8725   return true;
8726 }
8727 } // unnamed namespace
8728 
RegisterTypeTagForDatatype(const IdentifierInfo * ArgumentKind,uint64_t MagicValue,QualType Type,bool LayoutCompatible,bool MustBeNull)8729 void Sema::RegisterTypeTagForDatatype(const IdentifierInfo *ArgumentKind,
8730                                       uint64_t MagicValue, QualType Type,
8731                                       bool LayoutCompatible,
8732                                       bool MustBeNull) {
8733   if (!TypeTagForDatatypeMagicValues)
8734     TypeTagForDatatypeMagicValues.reset(
8735         new llvm::DenseMap<TypeTagMagicValue, TypeTagData>);
8736 
8737   TypeTagMagicValue Magic(ArgumentKind, MagicValue);
8738   (*TypeTagForDatatypeMagicValues)[Magic] =
8739       TypeTagData(Type, LayoutCompatible, MustBeNull);
8740 }
8741 
8742 namespace {
IsSameCharType(QualType T1,QualType T2)8743 bool IsSameCharType(QualType T1, QualType T2) {
8744   const BuiltinType *BT1 = T1->getAs<BuiltinType>();
8745   if (!BT1)
8746     return false;
8747 
8748   const BuiltinType *BT2 = T2->getAs<BuiltinType>();
8749   if (!BT2)
8750     return false;
8751 
8752   BuiltinType::Kind T1Kind = BT1->getKind();
8753   BuiltinType::Kind T2Kind = BT2->getKind();
8754 
8755   return (T1Kind == BuiltinType::SChar  && T2Kind == BuiltinType::Char_S) ||
8756          (T1Kind == BuiltinType::UChar  && T2Kind == BuiltinType::Char_U) ||
8757          (T1Kind == BuiltinType::Char_U && T2Kind == BuiltinType::UChar) ||
8758          (T1Kind == BuiltinType::Char_S && T2Kind == BuiltinType::SChar);
8759 }
8760 } // unnamed namespace
8761 
CheckArgumentWithTypeTag(const ArgumentWithTypeTagAttr * Attr,const Expr * const * ExprArgs)8762 void Sema::CheckArgumentWithTypeTag(const ArgumentWithTypeTagAttr *Attr,
8763                                     const Expr * const *ExprArgs) {
8764   const IdentifierInfo *ArgumentKind = Attr->getArgumentKind();
8765   bool IsPointerAttr = Attr->getIsPointer();
8766 
8767   const Expr *TypeTagExpr = ExprArgs[Attr->getTypeTagIdx()];
8768   bool FoundWrongKind;
8769   TypeTagData TypeInfo;
8770   if (!GetMatchingCType(ArgumentKind, TypeTagExpr, Context,
8771                         TypeTagForDatatypeMagicValues.get(),
8772                         FoundWrongKind, TypeInfo)) {
8773     if (FoundWrongKind)
8774       Diag(TypeTagExpr->getExprLoc(),
8775            diag::warn_type_tag_for_datatype_wrong_kind)
8776         << TypeTagExpr->getSourceRange();
8777     return;
8778   }
8779 
8780   const Expr *ArgumentExpr = ExprArgs[Attr->getArgumentIdx()];
8781   if (IsPointerAttr) {
8782     // Skip implicit cast of pointer to `void *' (as a function argument).
8783     if (const ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(ArgumentExpr))
8784       if (ICE->getType()->isVoidPointerType() &&
8785           ICE->getCastKind() == CK_BitCast)
8786         ArgumentExpr = ICE->getSubExpr();
8787   }
8788   QualType ArgumentType = ArgumentExpr->getType();
8789 
8790   // Passing a `void*' pointer shouldn't trigger a warning.
8791   if (IsPointerAttr && ArgumentType->isVoidPointerType())
8792     return;
8793 
8794   if (TypeInfo.MustBeNull) {
8795     // Type tag with matching void type requires a null pointer.
8796     if (!ArgumentExpr->isNullPointerConstant(Context,
8797                                              Expr::NPC_ValueDependentIsNotNull)) {
8798       Diag(ArgumentExpr->getExprLoc(),
8799            diag::warn_type_safety_null_pointer_required)
8800           << ArgumentKind->getName()
8801           << ArgumentExpr->getSourceRange()
8802           << TypeTagExpr->getSourceRange();
8803     }
8804     return;
8805   }
8806 
8807   QualType RequiredType = TypeInfo.Type;
8808   if (IsPointerAttr)
8809     RequiredType = Context.getPointerType(RequiredType);
8810 
8811   bool mismatch = false;
8812   if (!TypeInfo.LayoutCompatible) {
8813     mismatch = !Context.hasSameType(ArgumentType, RequiredType);
8814 
8815     // C++11 [basic.fundamental] p1:
8816     // Plain char, signed char, and unsigned char are three distinct types.
8817     //
8818     // But we treat plain `char' as equivalent to `signed char' or `unsigned
8819     // char' depending on the current char signedness mode.
8820     if (mismatch)
8821       if ((IsPointerAttr && IsSameCharType(ArgumentType->getPointeeType(),
8822                                            RequiredType->getPointeeType())) ||
8823           (!IsPointerAttr && IsSameCharType(ArgumentType, RequiredType)))
8824         mismatch = false;
8825   } else
8826     if (IsPointerAttr)
8827       mismatch = !isLayoutCompatible(Context,
8828                                      ArgumentType->getPointeeType(),
8829                                      RequiredType->getPointeeType());
8830     else
8831       mismatch = !isLayoutCompatible(Context, ArgumentType, RequiredType);
8832 
8833   if (mismatch)
8834     Diag(ArgumentExpr->getExprLoc(), diag::warn_type_safety_type_mismatch)
8835         << ArgumentType << ArgumentKind
8836         << TypeInfo.LayoutCompatible << RequiredType
8837         << ArgumentExpr->getSourceRange()
8838         << TypeTagExpr->getSourceRange();
8839 }
8840 
8841