• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
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/Initialization.h"
16 #include "clang/Sema/Sema.h"
17 #include "clang/Sema/SemaInternal.h"
18 #include "clang/Sema/Initialization.h"
19 #include "clang/Sema/Lookup.h"
20 #include "clang/Sema/ScopeInfo.h"
21 #include "clang/Analysis/Analyses/FormatString.h"
22 #include "clang/AST/ASTContext.h"
23 #include "clang/AST/CharUnits.h"
24 #include "clang/AST/DeclCXX.h"
25 #include "clang/AST/DeclObjC.h"
26 #include "clang/AST/Expr.h"
27 #include "clang/AST/ExprCXX.h"
28 #include "clang/AST/ExprObjC.h"
29 #include "clang/AST/EvaluatedExprVisitor.h"
30 #include "clang/AST/DeclObjC.h"
31 #include "clang/AST/StmtCXX.h"
32 #include "clang/AST/StmtObjC.h"
33 #include "clang/Lex/Preprocessor.h"
34 #include "llvm/ADT/BitVector.h"
35 #include "llvm/ADT/SmallString.h"
36 #include "llvm/ADT/STLExtras.h"
37 #include "llvm/Support/raw_ostream.h"
38 #include "clang/Basic/TargetBuiltins.h"
39 #include "clang/Basic/TargetInfo.h"
40 #include "clang/Basic/ConvertUTF.h"
41 #include <limits>
42 using namespace clang;
43 using namespace sema;
44 
getLocationOfStringLiteralByte(const StringLiteral * SL,unsigned ByteNo) const45 SourceLocation Sema::getLocationOfStringLiteralByte(const StringLiteral *SL,
46                                                     unsigned ByteNo) const {
47   return SL->getLocationOfByte(ByteNo, PP.getSourceManager(),
48                                PP.getLangOpts(), PP.getTargetInfo());
49 }
50 
51 /// Checks that a call expression's argument count is the desired number.
52 /// This is useful when doing custom type-checking.  Returns true on error.
checkArgCount(Sema & S,CallExpr * call,unsigned desiredArgCount)53 static bool checkArgCount(Sema &S, CallExpr *call, unsigned desiredArgCount) {
54   unsigned argCount = call->getNumArgs();
55   if (argCount == desiredArgCount) return false;
56 
57   if (argCount < desiredArgCount)
58     return S.Diag(call->getLocEnd(), diag::err_typecheck_call_too_few_args)
59         << 0 /*function call*/ << desiredArgCount << argCount
60         << call->getSourceRange();
61 
62   // Highlight all the excess arguments.
63   SourceRange range(call->getArg(desiredArgCount)->getLocStart(),
64                     call->getArg(argCount - 1)->getLocEnd());
65 
66   return S.Diag(range.getBegin(), diag::err_typecheck_call_too_many_args)
67     << 0 /*function call*/ << desiredArgCount << argCount
68     << call->getArg(1)->getSourceRange();
69 }
70 
71 /// Check that the first argument to __builtin_annotation is an integer
72 /// and the second argument is a non-wide string literal.
SemaBuiltinAnnotation(Sema & S,CallExpr * TheCall)73 static bool SemaBuiltinAnnotation(Sema &S, CallExpr *TheCall) {
74   if (checkArgCount(S, TheCall, 2))
75     return true;
76 
77   // First argument should be an integer.
78   Expr *ValArg = TheCall->getArg(0);
79   QualType Ty = ValArg->getType();
80   if (!Ty->isIntegerType()) {
81     S.Diag(ValArg->getLocStart(), diag::err_builtin_annotation_first_arg)
82       << ValArg->getSourceRange();
83     return true;
84   }
85 
86   // Second argument should be a constant string.
87   Expr *StrArg = TheCall->getArg(1)->IgnoreParenCasts();
88   StringLiteral *Literal = dyn_cast<StringLiteral>(StrArg);
89   if (!Literal || !Literal->isAscii()) {
90     S.Diag(StrArg->getLocStart(), diag::err_builtin_annotation_second_arg)
91       << StrArg->getSourceRange();
92     return true;
93   }
94 
95   TheCall->setType(Ty);
96   return false;
97 }
98 
99 ExprResult
CheckBuiltinFunctionCall(unsigned BuiltinID,CallExpr * TheCall)100 Sema::CheckBuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall) {
101   ExprResult TheCallResult(Owned(TheCall));
102 
103   // Find out if any arguments are required to be integer constant expressions.
104   unsigned ICEArguments = 0;
105   ASTContext::GetBuiltinTypeError Error;
106   Context.GetBuiltinType(BuiltinID, Error, &ICEArguments);
107   if (Error != ASTContext::GE_None)
108     ICEArguments = 0;  // Don't diagnose previously diagnosed errors.
109 
110   // If any arguments are required to be ICE's, check and diagnose.
111   for (unsigned ArgNo = 0; ICEArguments != 0; ++ArgNo) {
112     // Skip arguments not required to be ICE's.
113     if ((ICEArguments & (1 << ArgNo)) == 0) continue;
114 
115     llvm::APSInt Result;
116     if (SemaBuiltinConstantArg(TheCall, ArgNo, Result))
117       return true;
118     ICEArguments &= ~(1 << ArgNo);
119   }
120 
121   switch (BuiltinID) {
122   case Builtin::BI__builtin___CFStringMakeConstantString:
123     assert(TheCall->getNumArgs() == 1 &&
124            "Wrong # arguments to builtin CFStringMakeConstantString");
125     if (CheckObjCString(TheCall->getArg(0)))
126       return ExprError();
127     break;
128   case Builtin::BI__builtin_stdarg_start:
129   case Builtin::BI__builtin_va_start:
130     if (SemaBuiltinVAStart(TheCall))
131       return ExprError();
132     break;
133   case Builtin::BI__builtin_isgreater:
134   case Builtin::BI__builtin_isgreaterequal:
135   case Builtin::BI__builtin_isless:
136   case Builtin::BI__builtin_islessequal:
137   case Builtin::BI__builtin_islessgreater:
138   case Builtin::BI__builtin_isunordered:
139     if (SemaBuiltinUnorderedCompare(TheCall))
140       return ExprError();
141     break;
142   case Builtin::BI__builtin_fpclassify:
143     if (SemaBuiltinFPClassification(TheCall, 6))
144       return ExprError();
145     break;
146   case Builtin::BI__builtin_isfinite:
147   case Builtin::BI__builtin_isinf:
148   case Builtin::BI__builtin_isinf_sign:
149   case Builtin::BI__builtin_isnan:
150   case Builtin::BI__builtin_isnormal:
151     if (SemaBuiltinFPClassification(TheCall, 1))
152       return ExprError();
153     break;
154   case Builtin::BI__builtin_shufflevector:
155     return SemaBuiltinShuffleVector(TheCall);
156     // TheCall will be freed by the smart pointer here, but that's fine, since
157     // SemaBuiltinShuffleVector guts it, but then doesn't release it.
158   case Builtin::BI__builtin_prefetch:
159     if (SemaBuiltinPrefetch(TheCall))
160       return ExprError();
161     break;
162   case Builtin::BI__builtin_object_size:
163     if (SemaBuiltinObjectSize(TheCall))
164       return ExprError();
165     break;
166   case Builtin::BI__builtin_longjmp:
167     if (SemaBuiltinLongjmp(TheCall))
168       return ExprError();
169     break;
170 
171   case Builtin::BI__builtin_classify_type:
172     if (checkArgCount(*this, TheCall, 1)) return true;
173     TheCall->setType(Context.IntTy);
174     break;
175   case Builtin::BI__builtin_constant_p:
176     if (checkArgCount(*this, TheCall, 1)) return true;
177     TheCall->setType(Context.IntTy);
178     break;
179   case Builtin::BI__sync_fetch_and_add:
180   case Builtin::BI__sync_fetch_and_add_1:
181   case Builtin::BI__sync_fetch_and_add_2:
182   case Builtin::BI__sync_fetch_and_add_4:
183   case Builtin::BI__sync_fetch_and_add_8:
184   case Builtin::BI__sync_fetch_and_add_16:
185   case Builtin::BI__sync_fetch_and_sub:
186   case Builtin::BI__sync_fetch_and_sub_1:
187   case Builtin::BI__sync_fetch_and_sub_2:
188   case Builtin::BI__sync_fetch_and_sub_4:
189   case Builtin::BI__sync_fetch_and_sub_8:
190   case Builtin::BI__sync_fetch_and_sub_16:
191   case Builtin::BI__sync_fetch_and_or:
192   case Builtin::BI__sync_fetch_and_or_1:
193   case Builtin::BI__sync_fetch_and_or_2:
194   case Builtin::BI__sync_fetch_and_or_4:
195   case Builtin::BI__sync_fetch_and_or_8:
196   case Builtin::BI__sync_fetch_and_or_16:
197   case Builtin::BI__sync_fetch_and_and:
198   case Builtin::BI__sync_fetch_and_and_1:
199   case Builtin::BI__sync_fetch_and_and_2:
200   case Builtin::BI__sync_fetch_and_and_4:
201   case Builtin::BI__sync_fetch_and_and_8:
202   case Builtin::BI__sync_fetch_and_and_16:
203   case Builtin::BI__sync_fetch_and_xor:
204   case Builtin::BI__sync_fetch_and_xor_1:
205   case Builtin::BI__sync_fetch_and_xor_2:
206   case Builtin::BI__sync_fetch_and_xor_4:
207   case Builtin::BI__sync_fetch_and_xor_8:
208   case Builtin::BI__sync_fetch_and_xor_16:
209   case Builtin::BI__sync_add_and_fetch:
210   case Builtin::BI__sync_add_and_fetch_1:
211   case Builtin::BI__sync_add_and_fetch_2:
212   case Builtin::BI__sync_add_and_fetch_4:
213   case Builtin::BI__sync_add_and_fetch_8:
214   case Builtin::BI__sync_add_and_fetch_16:
215   case Builtin::BI__sync_sub_and_fetch:
216   case Builtin::BI__sync_sub_and_fetch_1:
217   case Builtin::BI__sync_sub_and_fetch_2:
218   case Builtin::BI__sync_sub_and_fetch_4:
219   case Builtin::BI__sync_sub_and_fetch_8:
220   case Builtin::BI__sync_sub_and_fetch_16:
221   case Builtin::BI__sync_and_and_fetch:
222   case Builtin::BI__sync_and_and_fetch_1:
223   case Builtin::BI__sync_and_and_fetch_2:
224   case Builtin::BI__sync_and_and_fetch_4:
225   case Builtin::BI__sync_and_and_fetch_8:
226   case Builtin::BI__sync_and_and_fetch_16:
227   case Builtin::BI__sync_or_and_fetch:
228   case Builtin::BI__sync_or_and_fetch_1:
229   case Builtin::BI__sync_or_and_fetch_2:
230   case Builtin::BI__sync_or_and_fetch_4:
231   case Builtin::BI__sync_or_and_fetch_8:
232   case Builtin::BI__sync_or_and_fetch_16:
233   case Builtin::BI__sync_xor_and_fetch:
234   case Builtin::BI__sync_xor_and_fetch_1:
235   case Builtin::BI__sync_xor_and_fetch_2:
236   case Builtin::BI__sync_xor_and_fetch_4:
237   case Builtin::BI__sync_xor_and_fetch_8:
238   case Builtin::BI__sync_xor_and_fetch_16:
239   case Builtin::BI__sync_val_compare_and_swap:
240   case Builtin::BI__sync_val_compare_and_swap_1:
241   case Builtin::BI__sync_val_compare_and_swap_2:
242   case Builtin::BI__sync_val_compare_and_swap_4:
243   case Builtin::BI__sync_val_compare_and_swap_8:
244   case Builtin::BI__sync_val_compare_and_swap_16:
245   case Builtin::BI__sync_bool_compare_and_swap:
246   case Builtin::BI__sync_bool_compare_and_swap_1:
247   case Builtin::BI__sync_bool_compare_and_swap_2:
248   case Builtin::BI__sync_bool_compare_and_swap_4:
249   case Builtin::BI__sync_bool_compare_and_swap_8:
250   case Builtin::BI__sync_bool_compare_and_swap_16:
251   case Builtin::BI__sync_lock_test_and_set:
252   case Builtin::BI__sync_lock_test_and_set_1:
253   case Builtin::BI__sync_lock_test_and_set_2:
254   case Builtin::BI__sync_lock_test_and_set_4:
255   case Builtin::BI__sync_lock_test_and_set_8:
256   case Builtin::BI__sync_lock_test_and_set_16:
257   case Builtin::BI__sync_lock_release:
258   case Builtin::BI__sync_lock_release_1:
259   case Builtin::BI__sync_lock_release_2:
260   case Builtin::BI__sync_lock_release_4:
261   case Builtin::BI__sync_lock_release_8:
262   case Builtin::BI__sync_lock_release_16:
263   case Builtin::BI__sync_swap:
264   case Builtin::BI__sync_swap_1:
265   case Builtin::BI__sync_swap_2:
266   case Builtin::BI__sync_swap_4:
267   case Builtin::BI__sync_swap_8:
268   case Builtin::BI__sync_swap_16:
269     return SemaBuiltinAtomicOverloaded(TheCallResult);
270 #define BUILTIN(ID, TYPE, ATTRS)
271 #define ATOMIC_BUILTIN(ID, TYPE, ATTRS) \
272   case Builtin::BI##ID: \
273     return SemaAtomicOpsOverloaded(TheCallResult, AtomicExpr::AO##ID);
274 #include "clang/Basic/Builtins.def"
275   case Builtin::BI__builtin_annotation:
276     if (SemaBuiltinAnnotation(*this, TheCall))
277       return ExprError();
278     break;
279   }
280 
281   // Since the target specific builtins for each arch overlap, only check those
282   // of the arch we are compiling for.
283   if (BuiltinID >= Builtin::FirstTSBuiltin) {
284     switch (Context.getTargetInfo().getTriple().getArch()) {
285       case llvm::Triple::arm:
286       case llvm::Triple::thumb:
287         if (CheckARMBuiltinFunctionCall(BuiltinID, TheCall))
288           return ExprError();
289         break;
290       case llvm::Triple::mips:
291       case llvm::Triple::mipsel:
292       case llvm::Triple::mips64:
293       case llvm::Triple::mips64el:
294         if (CheckMipsBuiltinFunctionCall(BuiltinID, TheCall))
295           return ExprError();
296         break;
297       default:
298         break;
299     }
300   }
301 
302   return TheCallResult;
303 }
304 
305 // Get the valid immediate range for the specified NEON type code.
RFT(unsigned t,bool shift=false)306 static unsigned RFT(unsigned t, bool shift = false) {
307   NeonTypeFlags Type(t);
308   int IsQuad = Type.isQuad();
309   switch (Type.getEltType()) {
310   case NeonTypeFlags::Int8:
311   case NeonTypeFlags::Poly8:
312     return shift ? 7 : (8 << IsQuad) - 1;
313   case NeonTypeFlags::Int16:
314   case NeonTypeFlags::Poly16:
315     return shift ? 15 : (4 << IsQuad) - 1;
316   case NeonTypeFlags::Int32:
317     return shift ? 31 : (2 << IsQuad) - 1;
318   case NeonTypeFlags::Int64:
319     return shift ? 63 : (1 << IsQuad) - 1;
320   case NeonTypeFlags::Float16:
321     assert(!shift && "cannot shift float types!");
322     return (4 << IsQuad) - 1;
323   case NeonTypeFlags::Float32:
324     assert(!shift && "cannot shift float types!");
325     return (2 << IsQuad) - 1;
326   }
327   llvm_unreachable("Invalid NeonTypeFlag!");
328 }
329 
330 /// getNeonEltType - Return the QualType corresponding to the elements of
331 /// the vector type specified by the NeonTypeFlags.  This is used to check
332 /// the pointer arguments for Neon load/store intrinsics.
getNeonEltType(NeonTypeFlags Flags,ASTContext & Context)333 static QualType getNeonEltType(NeonTypeFlags Flags, ASTContext &Context) {
334   switch (Flags.getEltType()) {
335   case NeonTypeFlags::Int8:
336     return Flags.isUnsigned() ? Context.UnsignedCharTy : Context.SignedCharTy;
337   case NeonTypeFlags::Int16:
338     return Flags.isUnsigned() ? Context.UnsignedShortTy : Context.ShortTy;
339   case NeonTypeFlags::Int32:
340     return Flags.isUnsigned() ? Context.UnsignedIntTy : Context.IntTy;
341   case NeonTypeFlags::Int64:
342     return Flags.isUnsigned() ? Context.UnsignedLongLongTy : Context.LongLongTy;
343   case NeonTypeFlags::Poly8:
344     return Context.SignedCharTy;
345   case NeonTypeFlags::Poly16:
346     return Context.ShortTy;
347   case NeonTypeFlags::Float16:
348     return Context.UnsignedShortTy;
349   case NeonTypeFlags::Float32:
350     return Context.FloatTy;
351   }
352   llvm_unreachable("Invalid NeonTypeFlag!");
353 }
354 
CheckARMBuiltinFunctionCall(unsigned BuiltinID,CallExpr * TheCall)355 bool Sema::CheckARMBuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall) {
356   llvm::APSInt Result;
357 
358   uint64_t mask = 0;
359   unsigned TV = 0;
360   int PtrArgNum = -1;
361   bool HasConstPtr = false;
362   switch (BuiltinID) {
363 #define GET_NEON_OVERLOAD_CHECK
364 #include "clang/Basic/arm_neon.inc"
365 #undef GET_NEON_OVERLOAD_CHECK
366   }
367 
368   // For NEON intrinsics which are overloaded on vector element type, validate
369   // the immediate which specifies which variant to emit.
370   unsigned ImmArg = TheCall->getNumArgs()-1;
371   if (mask) {
372     if (SemaBuiltinConstantArg(TheCall, ImmArg, Result))
373       return true;
374 
375     TV = Result.getLimitedValue(64);
376     if ((TV > 63) || (mask & (1ULL << TV)) == 0)
377       return Diag(TheCall->getLocStart(), diag::err_invalid_neon_type_code)
378         << TheCall->getArg(ImmArg)->getSourceRange();
379   }
380 
381   if (PtrArgNum >= 0) {
382     // Check that pointer arguments have the specified type.
383     Expr *Arg = TheCall->getArg(PtrArgNum);
384     if (ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(Arg))
385       Arg = ICE->getSubExpr();
386     ExprResult RHS = DefaultFunctionArrayLvalueConversion(Arg);
387     QualType RHSTy = RHS.get()->getType();
388     QualType EltTy = getNeonEltType(NeonTypeFlags(TV), Context);
389     if (HasConstPtr)
390       EltTy = EltTy.withConst();
391     QualType LHSTy = Context.getPointerType(EltTy);
392     AssignConvertType ConvTy;
393     ConvTy = CheckSingleAssignmentConstraints(LHSTy, RHS);
394     if (RHS.isInvalid())
395       return true;
396     if (DiagnoseAssignmentResult(ConvTy, Arg->getLocStart(), LHSTy, RHSTy,
397                                  RHS.get(), AA_Assigning))
398       return true;
399   }
400 
401   // For NEON intrinsics which take an immediate value as part of the
402   // instruction, range check them here.
403   unsigned i = 0, l = 0, u = 0;
404   switch (BuiltinID) {
405   default: return false;
406   case ARM::BI__builtin_arm_ssat: i = 1; l = 1; u = 31; break;
407   case ARM::BI__builtin_arm_usat: i = 1; u = 31; break;
408   case ARM::BI__builtin_arm_vcvtr_f:
409   case ARM::BI__builtin_arm_vcvtr_d: i = 1; u = 1; break;
410 #define GET_NEON_IMMEDIATE_CHECK
411 #include "clang/Basic/arm_neon.inc"
412 #undef GET_NEON_IMMEDIATE_CHECK
413   };
414 
415   // We can't check the value of a dependent argument.
416   if (TheCall->getArg(i)->isTypeDependent() ||
417       TheCall->getArg(i)->isValueDependent())
418     return false;
419 
420   // Check that the immediate argument is actually a constant.
421   if (SemaBuiltinConstantArg(TheCall, i, Result))
422     return true;
423 
424   // Range check against the upper/lower values for this isntruction.
425   unsigned Val = Result.getZExtValue();
426   if (Val < l || Val > (u + l))
427     return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range)
428       << l << u+l << TheCall->getArg(i)->getSourceRange();
429 
430   // FIXME: VFP Intrinsics should error if VFP not present.
431   return false;
432 }
433 
CheckMipsBuiltinFunctionCall(unsigned BuiltinID,CallExpr * TheCall)434 bool Sema::CheckMipsBuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall) {
435   unsigned i = 0, l = 0, u = 0;
436   switch (BuiltinID) {
437   default: return false;
438   case Mips::BI__builtin_mips_wrdsp: i = 1; l = 0; u = 63; break;
439   case Mips::BI__builtin_mips_rddsp: i = 0; l = 0; u = 63; break;
440   case Mips::BI__builtin_mips_append: i = 2; l = 0; u = 31; break;
441   case Mips::BI__builtin_mips_balign: i = 2; l = 0; u = 3; break;
442   case Mips::BI__builtin_mips_precr_sra_ph_w: i = 2; l = 0; u = 31; break;
443   case Mips::BI__builtin_mips_precr_sra_r_ph_w: i = 2; l = 0; u = 31; break;
444   case Mips::BI__builtin_mips_prepend: i = 2; l = 0; u = 31; break;
445   };
446 
447   // We can't check the value of a dependent argument.
448   if (TheCall->getArg(i)->isTypeDependent() ||
449       TheCall->getArg(i)->isValueDependent())
450     return false;
451 
452   // Check that the immediate argument is actually a constant.
453   llvm::APSInt Result;
454   if (SemaBuiltinConstantArg(TheCall, i, Result))
455     return true;
456 
457   // Range check against the upper/lower values for this instruction.
458   unsigned Val = Result.getZExtValue();
459   if (Val < l || Val > u)
460     return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range)
461       << l << u << TheCall->getArg(i)->getSourceRange();
462 
463   return false;
464 }
465 
466 /// Given a FunctionDecl's FormatAttr, attempts to populate the FomatStringInfo
467 /// parameter with the FormatAttr's correct format_idx and firstDataArg.
468 /// Returns true when the format fits the function and the FormatStringInfo has
469 /// been populated.
getFormatStringInfo(const FormatAttr * Format,bool IsCXXMember,FormatStringInfo * FSI)470 bool Sema::getFormatStringInfo(const FormatAttr *Format, bool IsCXXMember,
471                                FormatStringInfo *FSI) {
472   FSI->HasVAListArg = Format->getFirstArg() == 0;
473   FSI->FormatIdx = Format->getFormatIdx() - 1;
474   FSI->FirstDataArg = FSI->HasVAListArg ? 0 : Format->getFirstArg() - 1;
475 
476   // The way the format attribute works in GCC, the implicit this argument
477   // of member functions is counted. However, it doesn't appear in our own
478   // lists, so decrement format_idx in that case.
479   if (IsCXXMember) {
480     if(FSI->FormatIdx == 0)
481       return false;
482     --FSI->FormatIdx;
483     if (FSI->FirstDataArg != 0)
484       --FSI->FirstDataArg;
485   }
486   return true;
487 }
488 
489 /// Handles the checks for format strings, non-POD arguments to vararg
490 /// functions, and NULL arguments passed to non-NULL parameters.
checkCall(NamedDecl * FDecl,Expr ** Args,unsigned NumArgs,unsigned NumProtoArgs,bool IsMemberFunction,SourceLocation Loc,SourceRange Range,VariadicCallType CallType)491 void Sema::checkCall(NamedDecl *FDecl, Expr **Args,
492                      unsigned NumArgs,
493                      unsigned NumProtoArgs,
494                      bool IsMemberFunction,
495                      SourceLocation Loc,
496                      SourceRange Range,
497                      VariadicCallType CallType) {
498   // FIXME: This mechanism should be abstracted to be less fragile and
499   // more efficient. For example, just map function ids to custom
500   // handlers.
501 
502   // Printf and scanf checking.
503   bool HandledFormatString = false;
504   for (specific_attr_iterator<FormatAttr>
505          I = FDecl->specific_attr_begin<FormatAttr>(),
506          E = FDecl->specific_attr_end<FormatAttr>(); I != E ; ++I)
507     if (CheckFormatArguments(*I, Args, NumArgs, IsMemberFunction, CallType,
508                              Loc, Range))
509         HandledFormatString = true;
510 
511   // Refuse POD arguments that weren't caught by the format string
512   // checks above.
513   if (!HandledFormatString && CallType != VariadicDoesNotApply)
514     for (unsigned ArgIdx = NumProtoArgs; ArgIdx < NumArgs; ++ArgIdx)
515       variadicArgumentPODCheck(Args[ArgIdx], CallType);
516 
517   for (specific_attr_iterator<NonNullAttr>
518          I = FDecl->specific_attr_begin<NonNullAttr>(),
519          E = FDecl->specific_attr_end<NonNullAttr>(); I != E; ++I)
520     CheckNonNullArguments(*I, Args, Loc);
521 
522   // Type safety checking.
523   for (specific_attr_iterator<ArgumentWithTypeTagAttr>
524          i = FDecl->specific_attr_begin<ArgumentWithTypeTagAttr>(),
525          e = FDecl->specific_attr_end<ArgumentWithTypeTagAttr>(); i != e; ++i) {
526     CheckArgumentWithTypeTag(*i, Args);
527   }
528 }
529 
530 /// CheckConstructorCall - Check a constructor call for correctness and safety
531 /// properties not enforced by the C type system.
CheckConstructorCall(FunctionDecl * FDecl,Expr ** Args,unsigned NumArgs,const FunctionProtoType * Proto,SourceLocation Loc)532 void Sema::CheckConstructorCall(FunctionDecl *FDecl, Expr **Args,
533                                 unsigned NumArgs,
534                                 const FunctionProtoType *Proto,
535                                 SourceLocation Loc) {
536   VariadicCallType CallType =
537     Proto->isVariadic() ? VariadicConstructor : VariadicDoesNotApply;
538   checkCall(FDecl, Args, NumArgs, Proto->getNumArgs(),
539             /*IsMemberFunction=*/true, Loc, SourceRange(), CallType);
540 }
541 
542 /// CheckFunctionCall - Check a direct function call for various correctness
543 /// and safety properties not strictly enforced by the C type system.
CheckFunctionCall(FunctionDecl * FDecl,CallExpr * TheCall,const FunctionProtoType * Proto)544 bool Sema::CheckFunctionCall(FunctionDecl *FDecl, CallExpr *TheCall,
545                              const FunctionProtoType *Proto) {
546   bool IsMemberFunction = isa<CXXMemberCallExpr>(TheCall);
547   VariadicCallType CallType = getVariadicCallType(FDecl, Proto,
548                                                   TheCall->getCallee());
549   unsigned NumProtoArgs = Proto ? Proto->getNumArgs() : 0;
550   checkCall(FDecl, TheCall->getArgs(), TheCall->getNumArgs(), NumProtoArgs,
551             IsMemberFunction, TheCall->getRParenLoc(),
552             TheCall->getCallee()->getSourceRange(), CallType);
553 
554   IdentifierInfo *FnInfo = FDecl->getIdentifier();
555   // None of the checks below are needed for functions that don't have
556   // simple names (e.g., C++ conversion functions).
557   if (!FnInfo)
558     return false;
559 
560   unsigned CMId = FDecl->getMemoryFunctionKind();
561   if (CMId == 0)
562     return false;
563 
564   // Handle memory setting and copying functions.
565   if (CMId == Builtin::BIstrlcpy || CMId == Builtin::BIstrlcat)
566     CheckStrlcpycatArguments(TheCall, FnInfo);
567   else if (CMId == Builtin::BIstrncat)
568     CheckStrncatArguments(TheCall, FnInfo);
569   else
570     CheckMemaccessArguments(TheCall, CMId, FnInfo);
571 
572   return false;
573 }
574 
CheckObjCMethodCall(ObjCMethodDecl * Method,SourceLocation lbrac,Expr ** Args,unsigned NumArgs)575 bool Sema::CheckObjCMethodCall(ObjCMethodDecl *Method, SourceLocation lbrac,
576                                Expr **Args, unsigned NumArgs) {
577   VariadicCallType CallType =
578       Method->isVariadic() ? VariadicMethod : VariadicDoesNotApply;
579 
580   checkCall(Method, Args, NumArgs, Method->param_size(),
581             /*IsMemberFunction=*/false,
582             lbrac, Method->getSourceRange(), CallType);
583 
584   return false;
585 }
586 
CheckBlockCall(NamedDecl * NDecl,CallExpr * TheCall,const FunctionProtoType * Proto)587 bool Sema::CheckBlockCall(NamedDecl *NDecl, CallExpr *TheCall,
588                           const FunctionProtoType *Proto) {
589   const VarDecl *V = dyn_cast<VarDecl>(NDecl);
590   if (!V)
591     return false;
592 
593   QualType Ty = V->getType();
594   if (!Ty->isBlockPointerType())
595     return false;
596 
597   VariadicCallType CallType =
598       Proto && Proto->isVariadic() ? VariadicBlock : VariadicDoesNotApply ;
599   unsigned NumProtoArgs = Proto ? Proto->getNumArgs() : 0;
600 
601   checkCall(NDecl, TheCall->getArgs(), TheCall->getNumArgs(),
602             NumProtoArgs, /*IsMemberFunction=*/false,
603             TheCall->getRParenLoc(),
604             TheCall->getCallee()->getSourceRange(), CallType);
605 
606   return false;
607 }
608 
SemaAtomicOpsOverloaded(ExprResult TheCallResult,AtomicExpr::AtomicOp Op)609 ExprResult Sema::SemaAtomicOpsOverloaded(ExprResult TheCallResult,
610                                          AtomicExpr::AtomicOp Op) {
611   CallExpr *TheCall = cast<CallExpr>(TheCallResult.get());
612   DeclRefExpr *DRE =cast<DeclRefExpr>(TheCall->getCallee()->IgnoreParenCasts());
613 
614   // All these operations take one of the following forms:
615   enum {
616     // C    __c11_atomic_init(A *, C)
617     Init,
618     // C    __c11_atomic_load(A *, int)
619     Load,
620     // void __atomic_load(A *, CP, int)
621     Copy,
622     // C    __c11_atomic_add(A *, M, int)
623     Arithmetic,
624     // C    __atomic_exchange_n(A *, CP, int)
625     Xchg,
626     // void __atomic_exchange(A *, C *, CP, int)
627     GNUXchg,
628     // bool __c11_atomic_compare_exchange_strong(A *, C *, CP, int, int)
629     C11CmpXchg,
630     // bool __atomic_compare_exchange(A *, C *, CP, bool, int, int)
631     GNUCmpXchg
632   } Form = Init;
633   const unsigned NumArgs[] = { 2, 2, 3, 3, 3, 4, 5, 6 };
634   const unsigned NumVals[] = { 1, 0, 1, 1, 1, 2, 2, 3 };
635   // where:
636   //   C is an appropriate type,
637   //   A is volatile _Atomic(C) for __c11 builtins and is C for GNU builtins,
638   //   CP is C for __c11 builtins and GNU _n builtins and is C * otherwise,
639   //   M is C if C is an integer, and ptrdiff_t if C is a pointer, and
640   //   the int parameters are for orderings.
641 
642   assert(AtomicExpr::AO__c11_atomic_init == 0 &&
643          AtomicExpr::AO__c11_atomic_fetch_xor + 1 == AtomicExpr::AO__atomic_load
644          && "need to update code for modified C11 atomics");
645   bool IsC11 = Op >= AtomicExpr::AO__c11_atomic_init &&
646                Op <= AtomicExpr::AO__c11_atomic_fetch_xor;
647   bool IsN = Op == AtomicExpr::AO__atomic_load_n ||
648              Op == AtomicExpr::AO__atomic_store_n ||
649              Op == AtomicExpr::AO__atomic_exchange_n ||
650              Op == AtomicExpr::AO__atomic_compare_exchange_n;
651   bool IsAddSub = false;
652 
653   switch (Op) {
654   case AtomicExpr::AO__c11_atomic_init:
655     Form = Init;
656     break;
657 
658   case AtomicExpr::AO__c11_atomic_load:
659   case AtomicExpr::AO__atomic_load_n:
660     Form = Load;
661     break;
662 
663   case AtomicExpr::AO__c11_atomic_store:
664   case AtomicExpr::AO__atomic_load:
665   case AtomicExpr::AO__atomic_store:
666   case AtomicExpr::AO__atomic_store_n:
667     Form = Copy;
668     break;
669 
670   case AtomicExpr::AO__c11_atomic_fetch_add:
671   case AtomicExpr::AO__c11_atomic_fetch_sub:
672   case AtomicExpr::AO__atomic_fetch_add:
673   case AtomicExpr::AO__atomic_fetch_sub:
674   case AtomicExpr::AO__atomic_add_fetch:
675   case AtomicExpr::AO__atomic_sub_fetch:
676     IsAddSub = true;
677     // Fall through.
678   case AtomicExpr::AO__c11_atomic_fetch_and:
679   case AtomicExpr::AO__c11_atomic_fetch_or:
680   case AtomicExpr::AO__c11_atomic_fetch_xor:
681   case AtomicExpr::AO__atomic_fetch_and:
682   case AtomicExpr::AO__atomic_fetch_or:
683   case AtomicExpr::AO__atomic_fetch_xor:
684   case AtomicExpr::AO__atomic_fetch_nand:
685   case AtomicExpr::AO__atomic_and_fetch:
686   case AtomicExpr::AO__atomic_or_fetch:
687   case AtomicExpr::AO__atomic_xor_fetch:
688   case AtomicExpr::AO__atomic_nand_fetch:
689     Form = Arithmetic;
690     break;
691 
692   case AtomicExpr::AO__c11_atomic_exchange:
693   case AtomicExpr::AO__atomic_exchange_n:
694     Form = Xchg;
695     break;
696 
697   case AtomicExpr::AO__atomic_exchange:
698     Form = GNUXchg;
699     break;
700 
701   case AtomicExpr::AO__c11_atomic_compare_exchange_strong:
702   case AtomicExpr::AO__c11_atomic_compare_exchange_weak:
703     Form = C11CmpXchg;
704     break;
705 
706   case AtomicExpr::AO__atomic_compare_exchange:
707   case AtomicExpr::AO__atomic_compare_exchange_n:
708     Form = GNUCmpXchg;
709     break;
710   }
711 
712   // Check we have the right number of arguments.
713   if (TheCall->getNumArgs() < NumArgs[Form]) {
714     Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args)
715       << 0 << NumArgs[Form] << TheCall->getNumArgs()
716       << TheCall->getCallee()->getSourceRange();
717     return ExprError();
718   } else if (TheCall->getNumArgs() > NumArgs[Form]) {
719     Diag(TheCall->getArg(NumArgs[Form])->getLocStart(),
720          diag::err_typecheck_call_too_many_args)
721       << 0 << NumArgs[Form] << TheCall->getNumArgs()
722       << TheCall->getCallee()->getSourceRange();
723     return ExprError();
724   }
725 
726   // Inspect the first argument of the atomic operation.
727   Expr *Ptr = TheCall->getArg(0);
728   Ptr = DefaultFunctionArrayLvalueConversion(Ptr).get();
729   const PointerType *pointerType = Ptr->getType()->getAs<PointerType>();
730   if (!pointerType) {
731     Diag(DRE->getLocStart(), diag::err_atomic_builtin_must_be_pointer)
732       << Ptr->getType() << Ptr->getSourceRange();
733     return ExprError();
734   }
735 
736   // For a __c11 builtin, this should be a pointer to an _Atomic type.
737   QualType AtomTy = pointerType->getPointeeType(); // 'A'
738   QualType ValType = AtomTy; // 'C'
739   if (IsC11) {
740     if (!AtomTy->isAtomicType()) {
741       Diag(DRE->getLocStart(), diag::err_atomic_op_needs_atomic)
742         << Ptr->getType() << Ptr->getSourceRange();
743       return ExprError();
744     }
745     ValType = AtomTy->getAs<AtomicType>()->getValueType();
746   }
747 
748   // For an arithmetic operation, the implied arithmetic must be well-formed.
749   if (Form == Arithmetic) {
750     // gcc does not enforce these rules for GNU atomics, but we do so for sanity.
751     if (IsAddSub && !ValType->isIntegerType() && !ValType->isPointerType()) {
752       Diag(DRE->getLocStart(), diag::err_atomic_op_needs_atomic_int_or_ptr)
753         << IsC11 << Ptr->getType() << Ptr->getSourceRange();
754       return ExprError();
755     }
756     if (!IsAddSub && !ValType->isIntegerType()) {
757       Diag(DRE->getLocStart(), diag::err_atomic_op_bitwise_needs_atomic_int)
758         << IsC11 << Ptr->getType() << Ptr->getSourceRange();
759       return ExprError();
760     }
761   } else if (IsN && !ValType->isIntegerType() && !ValType->isPointerType()) {
762     // For __atomic_*_n operations, the value type must be a scalar integral or
763     // pointer type which is 1, 2, 4, 8 or 16 bytes in length.
764     Diag(DRE->getLocStart(), diag::err_atomic_op_needs_atomic_int_or_ptr)
765       << IsC11 << Ptr->getType() << Ptr->getSourceRange();
766     return ExprError();
767   }
768 
769   if (!IsC11 && !AtomTy.isTriviallyCopyableType(Context)) {
770     // For GNU atomics, require a trivially-copyable type. This is not part of
771     // the GNU atomics specification, but we enforce it for sanity.
772     Diag(DRE->getLocStart(), diag::err_atomic_op_needs_trivial_copy)
773       << Ptr->getType() << Ptr->getSourceRange();
774     return ExprError();
775   }
776 
777   // FIXME: For any builtin other than a load, the ValType must not be
778   // const-qualified.
779 
780   switch (ValType.getObjCLifetime()) {
781   case Qualifiers::OCL_None:
782   case Qualifiers::OCL_ExplicitNone:
783     // okay
784     break;
785 
786   case Qualifiers::OCL_Weak:
787   case Qualifiers::OCL_Strong:
788   case Qualifiers::OCL_Autoreleasing:
789     // FIXME: Can this happen? By this point, ValType should be known
790     // to be trivially copyable.
791     Diag(DRE->getLocStart(), diag::err_arc_atomic_ownership)
792       << ValType << Ptr->getSourceRange();
793     return ExprError();
794   }
795 
796   QualType ResultType = ValType;
797   if (Form == Copy || Form == GNUXchg || Form == Init)
798     ResultType = Context.VoidTy;
799   else if (Form == C11CmpXchg || Form == GNUCmpXchg)
800     ResultType = Context.BoolTy;
801 
802   // The type of a parameter passed 'by value'. In the GNU atomics, such
803   // arguments are actually passed as pointers.
804   QualType ByValType = ValType; // 'CP'
805   if (!IsC11 && !IsN)
806     ByValType = Ptr->getType();
807 
808   // The first argument --- the pointer --- has a fixed type; we
809   // deduce the types of the rest of the arguments accordingly.  Walk
810   // the remaining arguments, converting them to the deduced value type.
811   for (unsigned i = 1; i != NumArgs[Form]; ++i) {
812     QualType Ty;
813     if (i < NumVals[Form] + 1) {
814       switch (i) {
815       case 1:
816         // The second argument is the non-atomic operand. For arithmetic, this
817         // is always passed by value, and for a compare_exchange it is always
818         // passed by address. For the rest, GNU uses by-address and C11 uses
819         // by-value.
820         assert(Form != Load);
821         if (Form == Init || (Form == Arithmetic && ValType->isIntegerType()))
822           Ty = ValType;
823         else if (Form == Copy || Form == Xchg)
824           Ty = ByValType;
825         else if (Form == Arithmetic)
826           Ty = Context.getPointerDiffType();
827         else
828           Ty = Context.getPointerType(ValType.getUnqualifiedType());
829         break;
830       case 2:
831         // The third argument to compare_exchange / GNU exchange is a
832         // (pointer to a) desired value.
833         Ty = ByValType;
834         break;
835       case 3:
836         // The fourth argument to GNU compare_exchange is a 'weak' flag.
837         Ty = Context.BoolTy;
838         break;
839       }
840     } else {
841       // The order(s) are always converted to int.
842       Ty = Context.IntTy;
843     }
844 
845     InitializedEntity Entity =
846         InitializedEntity::InitializeParameter(Context, Ty, false);
847     ExprResult Arg = TheCall->getArg(i);
848     Arg = PerformCopyInitialization(Entity, SourceLocation(), Arg);
849     if (Arg.isInvalid())
850       return true;
851     TheCall->setArg(i, Arg.get());
852   }
853 
854   // Permute the arguments into a 'consistent' order.
855   SmallVector<Expr*, 5> SubExprs;
856   SubExprs.push_back(Ptr);
857   switch (Form) {
858   case Init:
859     // Note, AtomicExpr::getVal1() has a special case for this atomic.
860     SubExprs.push_back(TheCall->getArg(1)); // Val1
861     break;
862   case Load:
863     SubExprs.push_back(TheCall->getArg(1)); // Order
864     break;
865   case Copy:
866   case Arithmetic:
867   case Xchg:
868     SubExprs.push_back(TheCall->getArg(2)); // Order
869     SubExprs.push_back(TheCall->getArg(1)); // Val1
870     break;
871   case GNUXchg:
872     // Note, AtomicExpr::getVal2() has a special case for this atomic.
873     SubExprs.push_back(TheCall->getArg(3)); // Order
874     SubExprs.push_back(TheCall->getArg(1)); // Val1
875     SubExprs.push_back(TheCall->getArg(2)); // Val2
876     break;
877   case C11CmpXchg:
878     SubExprs.push_back(TheCall->getArg(3)); // Order
879     SubExprs.push_back(TheCall->getArg(1)); // Val1
880     SubExprs.push_back(TheCall->getArg(4)); // OrderFail
881     SubExprs.push_back(TheCall->getArg(2)); // Val2
882     break;
883   case GNUCmpXchg:
884     SubExprs.push_back(TheCall->getArg(4)); // Order
885     SubExprs.push_back(TheCall->getArg(1)); // Val1
886     SubExprs.push_back(TheCall->getArg(5)); // OrderFail
887     SubExprs.push_back(TheCall->getArg(2)); // Val2
888     SubExprs.push_back(TheCall->getArg(3)); // Weak
889     break;
890   }
891 
892   return Owned(new (Context) AtomicExpr(TheCall->getCallee()->getLocStart(),
893                                         SubExprs, ResultType, Op,
894                                         TheCall->getRParenLoc()));
895 }
896 
897 
898 /// checkBuiltinArgument - Given a call to a builtin function, perform
899 /// normal type-checking on the given argument, updating the call in
900 /// place.  This is useful when a builtin function requires custom
901 /// type-checking for some of its arguments but not necessarily all of
902 /// them.
903 ///
904 /// Returns true on error.
checkBuiltinArgument(Sema & S,CallExpr * E,unsigned ArgIndex)905 static bool checkBuiltinArgument(Sema &S, CallExpr *E, unsigned ArgIndex) {
906   FunctionDecl *Fn = E->getDirectCallee();
907   assert(Fn && "builtin call without direct callee!");
908 
909   ParmVarDecl *Param = Fn->getParamDecl(ArgIndex);
910   InitializedEntity Entity =
911     InitializedEntity::InitializeParameter(S.Context, Param);
912 
913   ExprResult Arg = E->getArg(0);
914   Arg = S.PerformCopyInitialization(Entity, SourceLocation(), Arg);
915   if (Arg.isInvalid())
916     return true;
917 
918   E->setArg(ArgIndex, Arg.take());
919   return false;
920 }
921 
922 /// SemaBuiltinAtomicOverloaded - We have a call to a function like
923 /// __sync_fetch_and_add, which is an overloaded function based on the pointer
924 /// type of its first argument.  The main ActOnCallExpr routines have already
925 /// promoted the types of arguments because all of these calls are prototyped as
926 /// void(...).
927 ///
928 /// This function goes through and does final semantic checking for these
929 /// builtins,
930 ExprResult
SemaBuiltinAtomicOverloaded(ExprResult TheCallResult)931 Sema::SemaBuiltinAtomicOverloaded(ExprResult TheCallResult) {
932   CallExpr *TheCall = (CallExpr *)TheCallResult.get();
933   DeclRefExpr *DRE =cast<DeclRefExpr>(TheCall->getCallee()->IgnoreParenCasts());
934   FunctionDecl *FDecl = cast<FunctionDecl>(DRE->getDecl());
935 
936   // Ensure that we have at least one argument to do type inference from.
937   if (TheCall->getNumArgs() < 1) {
938     Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args_at_least)
939       << 0 << 1 << TheCall->getNumArgs()
940       << TheCall->getCallee()->getSourceRange();
941     return ExprError();
942   }
943 
944   // Inspect the first argument of the atomic builtin.  This should always be
945   // a pointer type, whose element is an integral scalar or pointer type.
946   // Because it is a pointer type, we don't have to worry about any implicit
947   // casts here.
948   // FIXME: We don't allow floating point scalars as input.
949   Expr *FirstArg = TheCall->getArg(0);
950   ExprResult FirstArgResult = DefaultFunctionArrayLvalueConversion(FirstArg);
951   if (FirstArgResult.isInvalid())
952     return ExprError();
953   FirstArg = FirstArgResult.take();
954   TheCall->setArg(0, FirstArg);
955 
956   const PointerType *pointerType = FirstArg->getType()->getAs<PointerType>();
957   if (!pointerType) {
958     Diag(DRE->getLocStart(), diag::err_atomic_builtin_must_be_pointer)
959       << FirstArg->getType() << FirstArg->getSourceRange();
960     return ExprError();
961   }
962 
963   QualType ValType = pointerType->getPointeeType();
964   if (!ValType->isIntegerType() && !ValType->isAnyPointerType() &&
965       !ValType->isBlockPointerType()) {
966     Diag(DRE->getLocStart(), diag::err_atomic_builtin_must_be_pointer_intptr)
967       << FirstArg->getType() << FirstArg->getSourceRange();
968     return ExprError();
969   }
970 
971   switch (ValType.getObjCLifetime()) {
972   case Qualifiers::OCL_None:
973   case Qualifiers::OCL_ExplicitNone:
974     // okay
975     break;
976 
977   case Qualifiers::OCL_Weak:
978   case Qualifiers::OCL_Strong:
979   case Qualifiers::OCL_Autoreleasing:
980     Diag(DRE->getLocStart(), diag::err_arc_atomic_ownership)
981       << ValType << FirstArg->getSourceRange();
982     return ExprError();
983   }
984 
985   // Strip any qualifiers off ValType.
986   ValType = ValType.getUnqualifiedType();
987 
988   // The majority of builtins return a value, but a few have special return
989   // types, so allow them to override appropriately below.
990   QualType ResultType = ValType;
991 
992   // We need to figure out which concrete builtin this maps onto.  For example,
993   // __sync_fetch_and_add with a 2 byte object turns into
994   // __sync_fetch_and_add_2.
995 #define BUILTIN_ROW(x) \
996   { Builtin::BI##x##_1, Builtin::BI##x##_2, Builtin::BI##x##_4, \
997     Builtin::BI##x##_8, Builtin::BI##x##_16 }
998 
999   static const unsigned BuiltinIndices[][5] = {
1000     BUILTIN_ROW(__sync_fetch_and_add),
1001     BUILTIN_ROW(__sync_fetch_and_sub),
1002     BUILTIN_ROW(__sync_fetch_and_or),
1003     BUILTIN_ROW(__sync_fetch_and_and),
1004     BUILTIN_ROW(__sync_fetch_and_xor),
1005 
1006     BUILTIN_ROW(__sync_add_and_fetch),
1007     BUILTIN_ROW(__sync_sub_and_fetch),
1008     BUILTIN_ROW(__sync_and_and_fetch),
1009     BUILTIN_ROW(__sync_or_and_fetch),
1010     BUILTIN_ROW(__sync_xor_and_fetch),
1011 
1012     BUILTIN_ROW(__sync_val_compare_and_swap),
1013     BUILTIN_ROW(__sync_bool_compare_and_swap),
1014     BUILTIN_ROW(__sync_lock_test_and_set),
1015     BUILTIN_ROW(__sync_lock_release),
1016     BUILTIN_ROW(__sync_swap)
1017   };
1018 #undef BUILTIN_ROW
1019 
1020   // Determine the index of the size.
1021   unsigned SizeIndex;
1022   switch (Context.getTypeSizeInChars(ValType).getQuantity()) {
1023   case 1: SizeIndex = 0; break;
1024   case 2: SizeIndex = 1; break;
1025   case 4: SizeIndex = 2; break;
1026   case 8: SizeIndex = 3; break;
1027   case 16: SizeIndex = 4; break;
1028   default:
1029     Diag(DRE->getLocStart(), diag::err_atomic_builtin_pointer_size)
1030       << FirstArg->getType() << FirstArg->getSourceRange();
1031     return ExprError();
1032   }
1033 
1034   // Each of these builtins has one pointer argument, followed by some number of
1035   // values (0, 1 or 2) followed by a potentially empty varags list of stuff
1036   // that we ignore.  Find out which row of BuiltinIndices to read from as well
1037   // as the number of fixed args.
1038   unsigned BuiltinID = FDecl->getBuiltinID();
1039   unsigned BuiltinIndex, NumFixed = 1;
1040   switch (BuiltinID) {
1041   default: llvm_unreachable("Unknown overloaded atomic builtin!");
1042   case Builtin::BI__sync_fetch_and_add:
1043   case Builtin::BI__sync_fetch_and_add_1:
1044   case Builtin::BI__sync_fetch_and_add_2:
1045   case Builtin::BI__sync_fetch_and_add_4:
1046   case Builtin::BI__sync_fetch_and_add_8:
1047   case Builtin::BI__sync_fetch_and_add_16:
1048     BuiltinIndex = 0;
1049     break;
1050 
1051   case Builtin::BI__sync_fetch_and_sub:
1052   case Builtin::BI__sync_fetch_and_sub_1:
1053   case Builtin::BI__sync_fetch_and_sub_2:
1054   case Builtin::BI__sync_fetch_and_sub_4:
1055   case Builtin::BI__sync_fetch_and_sub_8:
1056   case Builtin::BI__sync_fetch_and_sub_16:
1057     BuiltinIndex = 1;
1058     break;
1059 
1060   case Builtin::BI__sync_fetch_and_or:
1061   case Builtin::BI__sync_fetch_and_or_1:
1062   case Builtin::BI__sync_fetch_and_or_2:
1063   case Builtin::BI__sync_fetch_and_or_4:
1064   case Builtin::BI__sync_fetch_and_or_8:
1065   case Builtin::BI__sync_fetch_and_or_16:
1066     BuiltinIndex = 2;
1067     break;
1068 
1069   case Builtin::BI__sync_fetch_and_and:
1070   case Builtin::BI__sync_fetch_and_and_1:
1071   case Builtin::BI__sync_fetch_and_and_2:
1072   case Builtin::BI__sync_fetch_and_and_4:
1073   case Builtin::BI__sync_fetch_and_and_8:
1074   case Builtin::BI__sync_fetch_and_and_16:
1075     BuiltinIndex = 3;
1076     break;
1077 
1078   case Builtin::BI__sync_fetch_and_xor:
1079   case Builtin::BI__sync_fetch_and_xor_1:
1080   case Builtin::BI__sync_fetch_and_xor_2:
1081   case Builtin::BI__sync_fetch_and_xor_4:
1082   case Builtin::BI__sync_fetch_and_xor_8:
1083   case Builtin::BI__sync_fetch_and_xor_16:
1084     BuiltinIndex = 4;
1085     break;
1086 
1087   case Builtin::BI__sync_add_and_fetch:
1088   case Builtin::BI__sync_add_and_fetch_1:
1089   case Builtin::BI__sync_add_and_fetch_2:
1090   case Builtin::BI__sync_add_and_fetch_4:
1091   case Builtin::BI__sync_add_and_fetch_8:
1092   case Builtin::BI__sync_add_and_fetch_16:
1093     BuiltinIndex = 5;
1094     break;
1095 
1096   case Builtin::BI__sync_sub_and_fetch:
1097   case Builtin::BI__sync_sub_and_fetch_1:
1098   case Builtin::BI__sync_sub_and_fetch_2:
1099   case Builtin::BI__sync_sub_and_fetch_4:
1100   case Builtin::BI__sync_sub_and_fetch_8:
1101   case Builtin::BI__sync_sub_and_fetch_16:
1102     BuiltinIndex = 6;
1103     break;
1104 
1105   case Builtin::BI__sync_and_and_fetch:
1106   case Builtin::BI__sync_and_and_fetch_1:
1107   case Builtin::BI__sync_and_and_fetch_2:
1108   case Builtin::BI__sync_and_and_fetch_4:
1109   case Builtin::BI__sync_and_and_fetch_8:
1110   case Builtin::BI__sync_and_and_fetch_16:
1111     BuiltinIndex = 7;
1112     break;
1113 
1114   case Builtin::BI__sync_or_and_fetch:
1115   case Builtin::BI__sync_or_and_fetch_1:
1116   case Builtin::BI__sync_or_and_fetch_2:
1117   case Builtin::BI__sync_or_and_fetch_4:
1118   case Builtin::BI__sync_or_and_fetch_8:
1119   case Builtin::BI__sync_or_and_fetch_16:
1120     BuiltinIndex = 8;
1121     break;
1122 
1123   case Builtin::BI__sync_xor_and_fetch:
1124   case Builtin::BI__sync_xor_and_fetch_1:
1125   case Builtin::BI__sync_xor_and_fetch_2:
1126   case Builtin::BI__sync_xor_and_fetch_4:
1127   case Builtin::BI__sync_xor_and_fetch_8:
1128   case Builtin::BI__sync_xor_and_fetch_16:
1129     BuiltinIndex = 9;
1130     break;
1131 
1132   case Builtin::BI__sync_val_compare_and_swap:
1133   case Builtin::BI__sync_val_compare_and_swap_1:
1134   case Builtin::BI__sync_val_compare_and_swap_2:
1135   case Builtin::BI__sync_val_compare_and_swap_4:
1136   case Builtin::BI__sync_val_compare_and_swap_8:
1137   case Builtin::BI__sync_val_compare_and_swap_16:
1138     BuiltinIndex = 10;
1139     NumFixed = 2;
1140     break;
1141 
1142   case Builtin::BI__sync_bool_compare_and_swap:
1143   case Builtin::BI__sync_bool_compare_and_swap_1:
1144   case Builtin::BI__sync_bool_compare_and_swap_2:
1145   case Builtin::BI__sync_bool_compare_and_swap_4:
1146   case Builtin::BI__sync_bool_compare_and_swap_8:
1147   case Builtin::BI__sync_bool_compare_and_swap_16:
1148     BuiltinIndex = 11;
1149     NumFixed = 2;
1150     ResultType = Context.BoolTy;
1151     break;
1152 
1153   case Builtin::BI__sync_lock_test_and_set:
1154   case Builtin::BI__sync_lock_test_and_set_1:
1155   case Builtin::BI__sync_lock_test_and_set_2:
1156   case Builtin::BI__sync_lock_test_and_set_4:
1157   case Builtin::BI__sync_lock_test_and_set_8:
1158   case Builtin::BI__sync_lock_test_and_set_16:
1159     BuiltinIndex = 12;
1160     break;
1161 
1162   case Builtin::BI__sync_lock_release:
1163   case Builtin::BI__sync_lock_release_1:
1164   case Builtin::BI__sync_lock_release_2:
1165   case Builtin::BI__sync_lock_release_4:
1166   case Builtin::BI__sync_lock_release_8:
1167   case Builtin::BI__sync_lock_release_16:
1168     BuiltinIndex = 13;
1169     NumFixed = 0;
1170     ResultType = Context.VoidTy;
1171     break;
1172 
1173   case Builtin::BI__sync_swap:
1174   case Builtin::BI__sync_swap_1:
1175   case Builtin::BI__sync_swap_2:
1176   case Builtin::BI__sync_swap_4:
1177   case Builtin::BI__sync_swap_8:
1178   case Builtin::BI__sync_swap_16:
1179     BuiltinIndex = 14;
1180     break;
1181   }
1182 
1183   // Now that we know how many fixed arguments we expect, first check that we
1184   // have at least that many.
1185   if (TheCall->getNumArgs() < 1+NumFixed) {
1186     Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args_at_least)
1187       << 0 << 1+NumFixed << TheCall->getNumArgs()
1188       << TheCall->getCallee()->getSourceRange();
1189     return ExprError();
1190   }
1191 
1192   // Get the decl for the concrete builtin from this, we can tell what the
1193   // concrete integer type we should convert to is.
1194   unsigned NewBuiltinID = BuiltinIndices[BuiltinIndex][SizeIndex];
1195   const char *NewBuiltinName = Context.BuiltinInfo.GetName(NewBuiltinID);
1196   IdentifierInfo *NewBuiltinII = PP.getIdentifierInfo(NewBuiltinName);
1197   FunctionDecl *NewBuiltinDecl =
1198     cast<FunctionDecl>(LazilyCreateBuiltin(NewBuiltinII, NewBuiltinID,
1199                                            TUScope, false, DRE->getLocStart()));
1200 
1201   // The first argument --- the pointer --- has a fixed type; we
1202   // deduce the types of the rest of the arguments accordingly.  Walk
1203   // the remaining arguments, converting them to the deduced value type.
1204   for (unsigned i = 0; i != NumFixed; ++i) {
1205     ExprResult Arg = TheCall->getArg(i+1);
1206 
1207     // GCC does an implicit conversion to the pointer or integer ValType.  This
1208     // can fail in some cases (1i -> int**), check for this error case now.
1209     // Initialize the argument.
1210     InitializedEntity Entity = InitializedEntity::InitializeParameter(Context,
1211                                                    ValType, /*consume*/ false);
1212     Arg = PerformCopyInitialization(Entity, SourceLocation(), Arg);
1213     if (Arg.isInvalid())
1214       return ExprError();
1215 
1216     // Okay, we have something that *can* be converted to the right type.  Check
1217     // to see if there is a potentially weird extension going on here.  This can
1218     // happen when you do an atomic operation on something like an char* and
1219     // pass in 42.  The 42 gets converted to char.  This is even more strange
1220     // for things like 45.123 -> char, etc.
1221     // FIXME: Do this check.
1222     TheCall->setArg(i+1, Arg.take());
1223   }
1224 
1225   ASTContext& Context = this->getASTContext();
1226 
1227   // Create a new DeclRefExpr to refer to the new decl.
1228   DeclRefExpr* NewDRE = DeclRefExpr::Create(
1229       Context,
1230       DRE->getQualifierLoc(),
1231       SourceLocation(),
1232       NewBuiltinDecl,
1233       /*enclosing*/ false,
1234       DRE->getLocation(),
1235       Context.BuiltinFnTy,
1236       DRE->getValueKind());
1237 
1238   // Set the callee in the CallExpr.
1239   // FIXME: This loses syntactic information.
1240   QualType CalleePtrTy = Context.getPointerType(NewBuiltinDecl->getType());
1241   ExprResult PromotedCall = ImpCastExprToType(NewDRE, CalleePtrTy,
1242                                               CK_BuiltinFnToFnPtr);
1243   TheCall->setCallee(PromotedCall.take());
1244 
1245   // Change the result type of the call to match the original value type. This
1246   // is arbitrary, but the codegen for these builtins ins design to handle it
1247   // gracefully.
1248   TheCall->setType(ResultType);
1249 
1250   return TheCallResult;
1251 }
1252 
1253 /// CheckObjCString - Checks that the argument to the builtin
1254 /// CFString constructor is correct
1255 /// Note: It might also make sense to do the UTF-16 conversion here (would
1256 /// simplify the backend).
CheckObjCString(Expr * Arg)1257 bool Sema::CheckObjCString(Expr *Arg) {
1258   Arg = Arg->IgnoreParenCasts();
1259   StringLiteral *Literal = dyn_cast<StringLiteral>(Arg);
1260 
1261   if (!Literal || !Literal->isAscii()) {
1262     Diag(Arg->getLocStart(), diag::err_cfstring_literal_not_string_constant)
1263       << Arg->getSourceRange();
1264     return true;
1265   }
1266 
1267   if (Literal->containsNonAsciiOrNull()) {
1268     StringRef String = Literal->getString();
1269     unsigned NumBytes = String.size();
1270     SmallVector<UTF16, 128> ToBuf(NumBytes);
1271     const UTF8 *FromPtr = (const UTF8 *)String.data();
1272     UTF16 *ToPtr = &ToBuf[0];
1273 
1274     ConversionResult Result = ConvertUTF8toUTF16(&FromPtr, FromPtr + NumBytes,
1275                                                  &ToPtr, ToPtr + NumBytes,
1276                                                  strictConversion);
1277     // Check for conversion failure.
1278     if (Result != conversionOK)
1279       Diag(Arg->getLocStart(),
1280            diag::warn_cfstring_truncated) << Arg->getSourceRange();
1281   }
1282   return false;
1283 }
1284 
1285 /// SemaBuiltinVAStart - Check the arguments to __builtin_va_start for validity.
1286 /// Emit an error and return true on failure, return false on success.
SemaBuiltinVAStart(CallExpr * TheCall)1287 bool Sema::SemaBuiltinVAStart(CallExpr *TheCall) {
1288   Expr *Fn = TheCall->getCallee();
1289   if (TheCall->getNumArgs() > 2) {
1290     Diag(TheCall->getArg(2)->getLocStart(),
1291          diag::err_typecheck_call_too_many_args)
1292       << 0 /*function call*/ << 2 << TheCall->getNumArgs()
1293       << Fn->getSourceRange()
1294       << SourceRange(TheCall->getArg(2)->getLocStart(),
1295                      (*(TheCall->arg_end()-1))->getLocEnd());
1296     return true;
1297   }
1298 
1299   if (TheCall->getNumArgs() < 2) {
1300     return Diag(TheCall->getLocEnd(),
1301       diag::err_typecheck_call_too_few_args_at_least)
1302       << 0 /*function call*/ << 2 << TheCall->getNumArgs();
1303   }
1304 
1305   // Type-check the first argument normally.
1306   if (checkBuiltinArgument(*this, TheCall, 0))
1307     return true;
1308 
1309   // Determine whether the current function is variadic or not.
1310   BlockScopeInfo *CurBlock = getCurBlock();
1311   bool isVariadic;
1312   if (CurBlock)
1313     isVariadic = CurBlock->TheDecl->isVariadic();
1314   else if (FunctionDecl *FD = getCurFunctionDecl())
1315     isVariadic = FD->isVariadic();
1316   else
1317     isVariadic = getCurMethodDecl()->isVariadic();
1318 
1319   if (!isVariadic) {
1320     Diag(Fn->getLocStart(), diag::err_va_start_used_in_non_variadic_function);
1321     return true;
1322   }
1323 
1324   // Verify that the second argument to the builtin is the last argument of the
1325   // current function or method.
1326   bool SecondArgIsLastNamedArgument = false;
1327   const Expr *Arg = TheCall->getArg(1)->IgnoreParenCasts();
1328 
1329   if (const DeclRefExpr *DR = dyn_cast<DeclRefExpr>(Arg)) {
1330     if (const ParmVarDecl *PV = dyn_cast<ParmVarDecl>(DR->getDecl())) {
1331       // FIXME: This isn't correct for methods (results in bogus warning).
1332       // Get the last formal in the current function.
1333       const ParmVarDecl *LastArg;
1334       if (CurBlock)
1335         LastArg = *(CurBlock->TheDecl->param_end()-1);
1336       else if (FunctionDecl *FD = getCurFunctionDecl())
1337         LastArg = *(FD->param_end()-1);
1338       else
1339         LastArg = *(getCurMethodDecl()->param_end()-1);
1340       SecondArgIsLastNamedArgument = PV == LastArg;
1341     }
1342   }
1343 
1344   if (!SecondArgIsLastNamedArgument)
1345     Diag(TheCall->getArg(1)->getLocStart(),
1346          diag::warn_second_parameter_of_va_start_not_last_named_argument);
1347   return false;
1348 }
1349 
1350 /// SemaBuiltinUnorderedCompare - Handle functions like __builtin_isgreater and
1351 /// friends.  This is declared to take (...), so we have to check everything.
SemaBuiltinUnorderedCompare(CallExpr * TheCall)1352 bool Sema::SemaBuiltinUnorderedCompare(CallExpr *TheCall) {
1353   if (TheCall->getNumArgs() < 2)
1354     return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args)
1355       << 0 << 2 << TheCall->getNumArgs()/*function call*/;
1356   if (TheCall->getNumArgs() > 2)
1357     return Diag(TheCall->getArg(2)->getLocStart(),
1358                 diag::err_typecheck_call_too_many_args)
1359       << 0 /*function call*/ << 2 << TheCall->getNumArgs()
1360       << SourceRange(TheCall->getArg(2)->getLocStart(),
1361                      (*(TheCall->arg_end()-1))->getLocEnd());
1362 
1363   ExprResult OrigArg0 = TheCall->getArg(0);
1364   ExprResult OrigArg1 = TheCall->getArg(1);
1365 
1366   // Do standard promotions between the two arguments, returning their common
1367   // type.
1368   QualType Res = UsualArithmeticConversions(OrigArg0, OrigArg1, false);
1369   if (OrigArg0.isInvalid() || OrigArg1.isInvalid())
1370     return true;
1371 
1372   // Make sure any conversions are pushed back into the call; this is
1373   // type safe since unordered compare builtins are declared as "_Bool
1374   // foo(...)".
1375   TheCall->setArg(0, OrigArg0.get());
1376   TheCall->setArg(1, OrigArg1.get());
1377 
1378   if (OrigArg0.get()->isTypeDependent() || OrigArg1.get()->isTypeDependent())
1379     return false;
1380 
1381   // If the common type isn't a real floating type, then the arguments were
1382   // invalid for this operation.
1383   if (Res.isNull() || !Res->isRealFloatingType())
1384     return Diag(OrigArg0.get()->getLocStart(),
1385                 diag::err_typecheck_call_invalid_ordered_compare)
1386       << OrigArg0.get()->getType() << OrigArg1.get()->getType()
1387       << SourceRange(OrigArg0.get()->getLocStart(), OrigArg1.get()->getLocEnd());
1388 
1389   return false;
1390 }
1391 
1392 /// SemaBuiltinSemaBuiltinFPClassification - Handle functions like
1393 /// __builtin_isnan and friends.  This is declared to take (...), so we have
1394 /// to check everything. We expect the last argument to be a floating point
1395 /// value.
SemaBuiltinFPClassification(CallExpr * TheCall,unsigned NumArgs)1396 bool Sema::SemaBuiltinFPClassification(CallExpr *TheCall, unsigned NumArgs) {
1397   if (TheCall->getNumArgs() < NumArgs)
1398     return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args)
1399       << 0 << NumArgs << TheCall->getNumArgs()/*function call*/;
1400   if (TheCall->getNumArgs() > NumArgs)
1401     return Diag(TheCall->getArg(NumArgs)->getLocStart(),
1402                 diag::err_typecheck_call_too_many_args)
1403       << 0 /*function call*/ << NumArgs << TheCall->getNumArgs()
1404       << SourceRange(TheCall->getArg(NumArgs)->getLocStart(),
1405                      (*(TheCall->arg_end()-1))->getLocEnd());
1406 
1407   Expr *OrigArg = TheCall->getArg(NumArgs-1);
1408 
1409   if (OrigArg->isTypeDependent())
1410     return false;
1411 
1412   // This operation requires a non-_Complex floating-point number.
1413   if (!OrigArg->getType()->isRealFloatingType())
1414     return Diag(OrigArg->getLocStart(),
1415                 diag::err_typecheck_call_invalid_unary_fp)
1416       << OrigArg->getType() << OrigArg->getSourceRange();
1417 
1418   // If this is an implicit conversion from float -> double, remove it.
1419   if (ImplicitCastExpr *Cast = dyn_cast<ImplicitCastExpr>(OrigArg)) {
1420     Expr *CastArg = Cast->getSubExpr();
1421     if (CastArg->getType()->isSpecificBuiltinType(BuiltinType::Float)) {
1422       assert(Cast->getType()->isSpecificBuiltinType(BuiltinType::Double) &&
1423              "promotion from float to double is the only expected cast here");
1424       Cast->setSubExpr(0);
1425       TheCall->setArg(NumArgs-1, CastArg);
1426     }
1427   }
1428 
1429   return false;
1430 }
1431 
1432 /// SemaBuiltinShuffleVector - Handle __builtin_shufflevector.
1433 // This is declared to take (...), so we have to check everything.
SemaBuiltinShuffleVector(CallExpr * TheCall)1434 ExprResult Sema::SemaBuiltinShuffleVector(CallExpr *TheCall) {
1435   if (TheCall->getNumArgs() < 2)
1436     return ExprError(Diag(TheCall->getLocEnd(),
1437                           diag::err_typecheck_call_too_few_args_at_least)
1438       << 0 /*function call*/ << 2 << TheCall->getNumArgs()
1439       << TheCall->getSourceRange());
1440 
1441   // Determine which of the following types of shufflevector we're checking:
1442   // 1) unary, vector mask: (lhs, mask)
1443   // 2) binary, vector mask: (lhs, rhs, mask)
1444   // 3) binary, scalar mask: (lhs, rhs, index, ..., index)
1445   QualType resType = TheCall->getArg(0)->getType();
1446   unsigned numElements = 0;
1447 
1448   if (!TheCall->getArg(0)->isTypeDependent() &&
1449       !TheCall->getArg(1)->isTypeDependent()) {
1450     QualType LHSType = TheCall->getArg(0)->getType();
1451     QualType RHSType = TheCall->getArg(1)->getType();
1452 
1453     if (!LHSType->isVectorType() || !RHSType->isVectorType()) {
1454       Diag(TheCall->getLocStart(), diag::err_shufflevector_non_vector)
1455         << SourceRange(TheCall->getArg(0)->getLocStart(),
1456                        TheCall->getArg(1)->getLocEnd());
1457       return ExprError();
1458     }
1459 
1460     numElements = LHSType->getAs<VectorType>()->getNumElements();
1461     unsigned numResElements = TheCall->getNumArgs() - 2;
1462 
1463     // Check to see if we have a call with 2 vector arguments, the unary shuffle
1464     // with mask.  If so, verify that RHS is an integer vector type with the
1465     // same number of elts as lhs.
1466     if (TheCall->getNumArgs() == 2) {
1467       if (!RHSType->hasIntegerRepresentation() ||
1468           RHSType->getAs<VectorType>()->getNumElements() != numElements)
1469         Diag(TheCall->getLocStart(), diag::err_shufflevector_incompatible_vector)
1470           << SourceRange(TheCall->getArg(1)->getLocStart(),
1471                          TheCall->getArg(1)->getLocEnd());
1472       numResElements = numElements;
1473     }
1474     else if (!Context.hasSameUnqualifiedType(LHSType, RHSType)) {
1475       Diag(TheCall->getLocStart(), diag::err_shufflevector_incompatible_vector)
1476         << SourceRange(TheCall->getArg(0)->getLocStart(),
1477                        TheCall->getArg(1)->getLocEnd());
1478       return ExprError();
1479     } else if (numElements != numResElements) {
1480       QualType eltType = LHSType->getAs<VectorType>()->getElementType();
1481       resType = Context.getVectorType(eltType, numResElements,
1482                                       VectorType::GenericVector);
1483     }
1484   }
1485 
1486   for (unsigned i = 2; i < TheCall->getNumArgs(); i++) {
1487     if (TheCall->getArg(i)->isTypeDependent() ||
1488         TheCall->getArg(i)->isValueDependent())
1489       continue;
1490 
1491     llvm::APSInt Result(32);
1492     if (!TheCall->getArg(i)->isIntegerConstantExpr(Result, Context))
1493       return ExprError(Diag(TheCall->getLocStart(),
1494                   diag::err_shufflevector_nonconstant_argument)
1495                 << TheCall->getArg(i)->getSourceRange());
1496 
1497     if (Result.getActiveBits() > 64 || Result.getZExtValue() >= numElements*2)
1498       return ExprError(Diag(TheCall->getLocStart(),
1499                   diag::err_shufflevector_argument_too_large)
1500                << TheCall->getArg(i)->getSourceRange());
1501   }
1502 
1503   SmallVector<Expr*, 32> exprs;
1504 
1505   for (unsigned i = 0, e = TheCall->getNumArgs(); i != e; i++) {
1506     exprs.push_back(TheCall->getArg(i));
1507     TheCall->setArg(i, 0);
1508   }
1509 
1510   return Owned(new (Context) ShuffleVectorExpr(Context, exprs, resType,
1511                                             TheCall->getCallee()->getLocStart(),
1512                                             TheCall->getRParenLoc()));
1513 }
1514 
1515 /// SemaBuiltinPrefetch - Handle __builtin_prefetch.
1516 // This is declared to take (const void*, ...) and can take two
1517 // optional constant int args.
SemaBuiltinPrefetch(CallExpr * TheCall)1518 bool Sema::SemaBuiltinPrefetch(CallExpr *TheCall) {
1519   unsigned NumArgs = TheCall->getNumArgs();
1520 
1521   if (NumArgs > 3)
1522     return Diag(TheCall->getLocEnd(),
1523              diag::err_typecheck_call_too_many_args_at_most)
1524              << 0 /*function call*/ << 3 << NumArgs
1525              << TheCall->getSourceRange();
1526 
1527   // Argument 0 is checked for us and the remaining arguments must be
1528   // constant integers.
1529   for (unsigned i = 1; i != NumArgs; ++i) {
1530     Expr *Arg = TheCall->getArg(i);
1531 
1532     // We can't check the value of a dependent argument.
1533     if (Arg->isTypeDependent() || Arg->isValueDependent())
1534       continue;
1535 
1536     llvm::APSInt Result;
1537     if (SemaBuiltinConstantArg(TheCall, i, Result))
1538       return true;
1539 
1540     // FIXME: gcc issues a warning and rewrites these to 0. These
1541     // seems especially odd for the third argument since the default
1542     // is 3.
1543     if (i == 1) {
1544       if (Result.getLimitedValue() > 1)
1545         return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range)
1546              << "0" << "1" << Arg->getSourceRange();
1547     } else {
1548       if (Result.getLimitedValue() > 3)
1549         return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range)
1550             << "0" << "3" << Arg->getSourceRange();
1551     }
1552   }
1553 
1554   return false;
1555 }
1556 
1557 /// SemaBuiltinConstantArg - Handle a check if argument ArgNum of CallExpr
1558 /// TheCall is a constant expression.
SemaBuiltinConstantArg(CallExpr * TheCall,int ArgNum,llvm::APSInt & Result)1559 bool Sema::SemaBuiltinConstantArg(CallExpr *TheCall, int ArgNum,
1560                                   llvm::APSInt &Result) {
1561   Expr *Arg = TheCall->getArg(ArgNum);
1562   DeclRefExpr *DRE =cast<DeclRefExpr>(TheCall->getCallee()->IgnoreParenCasts());
1563   FunctionDecl *FDecl = cast<FunctionDecl>(DRE->getDecl());
1564 
1565   if (Arg->isTypeDependent() || Arg->isValueDependent()) return false;
1566 
1567   if (!Arg->isIntegerConstantExpr(Result, Context))
1568     return Diag(TheCall->getLocStart(), diag::err_constant_integer_arg_type)
1569                 << FDecl->getDeclName() <<  Arg->getSourceRange();
1570 
1571   return false;
1572 }
1573 
1574 /// SemaBuiltinObjectSize - Handle __builtin_object_size(void *ptr,
1575 /// int type). This simply type checks that type is one of the defined
1576 /// constants (0-3).
1577 // For compatibility check 0-3, llvm only handles 0 and 2.
SemaBuiltinObjectSize(CallExpr * TheCall)1578 bool Sema::SemaBuiltinObjectSize(CallExpr *TheCall) {
1579   llvm::APSInt Result;
1580 
1581   // We can't check the value of a dependent argument.
1582   if (TheCall->getArg(1)->isTypeDependent() ||
1583       TheCall->getArg(1)->isValueDependent())
1584     return false;
1585 
1586   // Check constant-ness first.
1587   if (SemaBuiltinConstantArg(TheCall, 1, Result))
1588     return true;
1589 
1590   Expr *Arg = TheCall->getArg(1);
1591   if (Result.getSExtValue() < 0 || Result.getSExtValue() > 3) {
1592     return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range)
1593              << "0" << "3" << SourceRange(Arg->getLocStart(), Arg->getLocEnd());
1594   }
1595 
1596   return false;
1597 }
1598 
1599 /// SemaBuiltinLongjmp - Handle __builtin_longjmp(void *env[5], int val).
1600 /// This checks that val is a constant 1.
SemaBuiltinLongjmp(CallExpr * TheCall)1601 bool Sema::SemaBuiltinLongjmp(CallExpr *TheCall) {
1602   Expr *Arg = TheCall->getArg(1);
1603   llvm::APSInt Result;
1604 
1605   // TODO: This is less than ideal. Overload this to take a value.
1606   if (SemaBuiltinConstantArg(TheCall, 1, Result))
1607     return true;
1608 
1609   if (Result != 1)
1610     return Diag(TheCall->getLocStart(), diag::err_builtin_longjmp_invalid_val)
1611              << SourceRange(Arg->getLocStart(), Arg->getLocEnd());
1612 
1613   return false;
1614 }
1615 
1616 // Determine if an expression is a string literal or constant string.
1617 // If this function returns false on the arguments to a function expecting a
1618 // format string, we will usually need to emit a warning.
1619 // True string literals are then checked by CheckFormatString.
1620 Sema::StringLiteralCheckType
checkFormatStringExpr(const Expr * E,Expr ** Args,unsigned NumArgs,bool HasVAListArg,unsigned format_idx,unsigned firstDataArg,FormatStringType Type,VariadicCallType CallType,bool inFunctionCall)1621 Sema::checkFormatStringExpr(const Expr *E, Expr **Args,
1622                             unsigned NumArgs, bool HasVAListArg,
1623                             unsigned format_idx, unsigned firstDataArg,
1624                             FormatStringType Type, VariadicCallType CallType,
1625                             bool inFunctionCall) {
1626  tryAgain:
1627   if (E->isTypeDependent() || E->isValueDependent())
1628     return SLCT_NotALiteral;
1629 
1630   E = E->IgnoreParenCasts();
1631 
1632   if (E->isNullPointerConstant(Context, Expr::NPC_ValueDependentIsNotNull))
1633     // Technically -Wformat-nonliteral does not warn about this case.
1634     // The behavior of printf and friends in this case is implementation
1635     // dependent.  Ideally if the format string cannot be null then
1636     // it should have a 'nonnull' attribute in the function prototype.
1637     return SLCT_CheckedLiteral;
1638 
1639   switch (E->getStmtClass()) {
1640   case Stmt::BinaryConditionalOperatorClass:
1641   case Stmt::ConditionalOperatorClass: {
1642     // The expression is a literal if both sub-expressions were, and it was
1643     // completely checked only if both sub-expressions were checked.
1644     const AbstractConditionalOperator *C =
1645         cast<AbstractConditionalOperator>(E);
1646     StringLiteralCheckType Left =
1647         checkFormatStringExpr(C->getTrueExpr(), Args, NumArgs,
1648                               HasVAListArg, format_idx, firstDataArg,
1649                               Type, CallType, inFunctionCall);
1650     if (Left == SLCT_NotALiteral)
1651       return SLCT_NotALiteral;
1652     StringLiteralCheckType Right =
1653         checkFormatStringExpr(C->getFalseExpr(), Args, NumArgs,
1654                               HasVAListArg, format_idx, firstDataArg,
1655                               Type, CallType, inFunctionCall);
1656     return Left < Right ? Left : Right;
1657   }
1658 
1659   case Stmt::ImplicitCastExprClass: {
1660     E = cast<ImplicitCastExpr>(E)->getSubExpr();
1661     goto tryAgain;
1662   }
1663 
1664   case Stmt::OpaqueValueExprClass:
1665     if (const Expr *src = cast<OpaqueValueExpr>(E)->getSourceExpr()) {
1666       E = src;
1667       goto tryAgain;
1668     }
1669     return SLCT_NotALiteral;
1670 
1671   case Stmt::PredefinedExprClass:
1672     // While __func__, etc., are technically not string literals, they
1673     // cannot contain format specifiers and thus are not a security
1674     // liability.
1675     return SLCT_UncheckedLiteral;
1676 
1677   case Stmt::DeclRefExprClass: {
1678     const DeclRefExpr *DR = cast<DeclRefExpr>(E);
1679 
1680     // As an exception, do not flag errors for variables binding to
1681     // const string literals.
1682     if (const VarDecl *VD = dyn_cast<VarDecl>(DR->getDecl())) {
1683       bool isConstant = false;
1684       QualType T = DR->getType();
1685 
1686       if (const ArrayType *AT = Context.getAsArrayType(T)) {
1687         isConstant = AT->getElementType().isConstant(Context);
1688       } else if (const PointerType *PT = T->getAs<PointerType>()) {
1689         isConstant = T.isConstant(Context) &&
1690                      PT->getPointeeType().isConstant(Context);
1691       } else if (T->isObjCObjectPointerType()) {
1692         // In ObjC, there is usually no "const ObjectPointer" type,
1693         // so don't check if the pointee type is constant.
1694         isConstant = T.isConstant(Context);
1695       }
1696 
1697       if (isConstant) {
1698         if (const Expr *Init = VD->getAnyInitializer()) {
1699           // Look through initializers like const char c[] = { "foo" }
1700           if (const InitListExpr *InitList = dyn_cast<InitListExpr>(Init)) {
1701             if (InitList->isStringLiteralInit())
1702               Init = InitList->getInit(0)->IgnoreParenImpCasts();
1703           }
1704           return checkFormatStringExpr(Init, Args, NumArgs,
1705                                        HasVAListArg, format_idx,
1706                                        firstDataArg, Type, CallType,
1707                                        /*inFunctionCall*/false);
1708         }
1709       }
1710 
1711       // For vprintf* functions (i.e., HasVAListArg==true), we add a
1712       // special check to see if the format string is a function parameter
1713       // of the function calling the printf function.  If the function
1714       // has an attribute indicating it is a printf-like function, then we
1715       // should suppress warnings concerning non-literals being used in a call
1716       // to a vprintf function.  For example:
1717       //
1718       // void
1719       // logmessage(char const *fmt __attribute__ (format (printf, 1, 2)), ...){
1720       //      va_list ap;
1721       //      va_start(ap, fmt);
1722       //      vprintf(fmt, ap);  // Do NOT emit a warning about "fmt".
1723       //      ...
1724       //
1725       if (HasVAListArg) {
1726         if (const ParmVarDecl *PV = dyn_cast<ParmVarDecl>(VD)) {
1727           if (const NamedDecl *ND = dyn_cast<NamedDecl>(PV->getDeclContext())) {
1728             int PVIndex = PV->getFunctionScopeIndex() + 1;
1729             for (specific_attr_iterator<FormatAttr>
1730                  i = ND->specific_attr_begin<FormatAttr>(),
1731                  e = ND->specific_attr_end<FormatAttr>(); i != e ; ++i) {
1732               FormatAttr *PVFormat = *i;
1733               // adjust for implicit parameter
1734               if (const CXXMethodDecl *MD = dyn_cast<CXXMethodDecl>(ND))
1735                 if (MD->isInstance())
1736                   ++PVIndex;
1737               // We also check if the formats are compatible.
1738               // We can't pass a 'scanf' string to a 'printf' function.
1739               if (PVIndex == PVFormat->getFormatIdx() &&
1740                   Type == GetFormatStringType(PVFormat))
1741                 return SLCT_UncheckedLiteral;
1742             }
1743           }
1744         }
1745       }
1746     }
1747 
1748     return SLCT_NotALiteral;
1749   }
1750 
1751   case Stmt::CallExprClass:
1752   case Stmt::CXXMemberCallExprClass: {
1753     const CallExpr *CE = cast<CallExpr>(E);
1754     if (const NamedDecl *ND = dyn_cast_or_null<NamedDecl>(CE->getCalleeDecl())) {
1755       if (const FormatArgAttr *FA = ND->getAttr<FormatArgAttr>()) {
1756         unsigned ArgIndex = FA->getFormatIdx();
1757         if (const CXXMethodDecl *MD = dyn_cast<CXXMethodDecl>(ND))
1758           if (MD->isInstance())
1759             --ArgIndex;
1760         const Expr *Arg = CE->getArg(ArgIndex - 1);
1761 
1762         return checkFormatStringExpr(Arg, Args, NumArgs,
1763                                      HasVAListArg, format_idx, firstDataArg,
1764                                      Type, CallType, inFunctionCall);
1765       } else if (const FunctionDecl *FD = dyn_cast<FunctionDecl>(ND)) {
1766         unsigned BuiltinID = FD->getBuiltinID();
1767         if (BuiltinID == Builtin::BI__builtin___CFStringMakeConstantString ||
1768             BuiltinID == Builtin::BI__builtin___NSStringMakeConstantString) {
1769           const Expr *Arg = CE->getArg(0);
1770           return checkFormatStringExpr(Arg, Args, NumArgs,
1771                                        HasVAListArg, format_idx,
1772                                        firstDataArg, Type, CallType,
1773                                        inFunctionCall);
1774         }
1775       }
1776     }
1777 
1778     return SLCT_NotALiteral;
1779   }
1780   case Stmt::ObjCStringLiteralClass:
1781   case Stmt::StringLiteralClass: {
1782     const StringLiteral *StrE = NULL;
1783 
1784     if (const ObjCStringLiteral *ObjCFExpr = dyn_cast<ObjCStringLiteral>(E))
1785       StrE = ObjCFExpr->getString();
1786     else
1787       StrE = cast<StringLiteral>(E);
1788 
1789     if (StrE) {
1790       CheckFormatString(StrE, E, Args, NumArgs, HasVAListArg, format_idx,
1791                         firstDataArg, Type, inFunctionCall, CallType);
1792       return SLCT_CheckedLiteral;
1793     }
1794 
1795     return SLCT_NotALiteral;
1796   }
1797 
1798   default:
1799     return SLCT_NotALiteral;
1800   }
1801 }
1802 
1803 void
CheckNonNullArguments(const NonNullAttr * NonNull,const Expr * const * ExprArgs,SourceLocation CallSiteLoc)1804 Sema::CheckNonNullArguments(const NonNullAttr *NonNull,
1805                             const Expr * const *ExprArgs,
1806                             SourceLocation CallSiteLoc) {
1807   for (NonNullAttr::args_iterator i = NonNull->args_begin(),
1808                                   e = NonNull->args_end();
1809        i != e; ++i) {
1810     const Expr *ArgExpr = ExprArgs[*i];
1811     if (ArgExpr->isNullPointerConstant(Context,
1812                                        Expr::NPC_ValueDependentIsNotNull))
1813       Diag(CallSiteLoc, diag::warn_null_arg) << ArgExpr->getSourceRange();
1814   }
1815 }
1816 
GetFormatStringType(const FormatAttr * Format)1817 Sema::FormatStringType Sema::GetFormatStringType(const FormatAttr *Format) {
1818   return llvm::StringSwitch<FormatStringType>(Format->getType())
1819   .Case("scanf", FST_Scanf)
1820   .Cases("printf", "printf0", FST_Printf)
1821   .Cases("NSString", "CFString", FST_NSString)
1822   .Case("strftime", FST_Strftime)
1823   .Case("strfmon", FST_Strfmon)
1824   .Cases("kprintf", "cmn_err", "vcmn_err", "zcmn_err", FST_Kprintf)
1825   .Default(FST_Unknown);
1826 }
1827 
1828 /// CheckFormatArguments - Check calls to printf and scanf (and similar
1829 /// functions) for correct use of format strings.
1830 /// Returns true if a format string has been fully checked.
CheckFormatArguments(const FormatAttr * Format,Expr ** Args,unsigned NumArgs,bool IsCXXMember,VariadicCallType CallType,SourceLocation Loc,SourceRange Range)1831 bool Sema::CheckFormatArguments(const FormatAttr *Format, Expr **Args,
1832                                 unsigned NumArgs, bool IsCXXMember,
1833                                 VariadicCallType CallType,
1834                                 SourceLocation Loc, SourceRange Range) {
1835   FormatStringInfo FSI;
1836   if (getFormatStringInfo(Format, IsCXXMember, &FSI))
1837     return CheckFormatArguments(Args, NumArgs, FSI.HasVAListArg, FSI.FormatIdx,
1838                                 FSI.FirstDataArg, GetFormatStringType(Format),
1839                                 CallType, Loc, Range);
1840   return false;
1841 }
1842 
CheckFormatArguments(Expr ** Args,unsigned NumArgs,bool HasVAListArg,unsigned format_idx,unsigned firstDataArg,FormatStringType Type,VariadicCallType CallType,SourceLocation Loc,SourceRange Range)1843 bool Sema::CheckFormatArguments(Expr **Args, unsigned NumArgs,
1844                                 bool HasVAListArg, unsigned format_idx,
1845                                 unsigned firstDataArg, FormatStringType Type,
1846                                 VariadicCallType CallType,
1847                                 SourceLocation Loc, SourceRange Range) {
1848   // CHECK: printf/scanf-like function is called with no format string.
1849   if (format_idx >= NumArgs) {
1850     Diag(Loc, diag::warn_missing_format_string) << Range;
1851     return false;
1852   }
1853 
1854   const Expr *OrigFormatExpr = Args[format_idx]->IgnoreParenCasts();
1855 
1856   // CHECK: format string is not a string literal.
1857   //
1858   // Dynamically generated format strings are difficult to
1859   // automatically vet at compile time.  Requiring that format strings
1860   // are string literals: (1) permits the checking of format strings by
1861   // the compiler and thereby (2) can practically remove the source of
1862   // many format string exploits.
1863 
1864   // Format string can be either ObjC string (e.g. @"%d") or
1865   // C string (e.g. "%d")
1866   // ObjC string uses the same format specifiers as C string, so we can use
1867   // the same format string checking logic for both ObjC and C strings.
1868   StringLiteralCheckType CT =
1869       checkFormatStringExpr(OrigFormatExpr, Args, NumArgs, HasVAListArg,
1870                             format_idx, firstDataArg, Type, CallType);
1871   if (CT != SLCT_NotALiteral)
1872     // Literal format string found, check done!
1873     return CT == SLCT_CheckedLiteral;
1874 
1875   // Strftime is particular as it always uses a single 'time' argument,
1876   // so it is safe to pass a non-literal string.
1877   if (Type == FST_Strftime)
1878     return false;
1879 
1880   // Do not emit diag when the string param is a macro expansion and the
1881   // format is either NSString or CFString. This is a hack to prevent
1882   // diag when using the NSLocalizedString and CFCopyLocalizedString macros
1883   // which are usually used in place of NS and CF string literals.
1884   if (Type == FST_NSString &&
1885       SourceMgr.isInSystemMacro(Args[format_idx]->getLocStart()))
1886     return false;
1887 
1888   // If there are no arguments specified, warn with -Wformat-security, otherwise
1889   // warn only with -Wformat-nonliteral.
1890   if (NumArgs == format_idx+1)
1891     Diag(Args[format_idx]->getLocStart(),
1892          diag::warn_format_nonliteral_noargs)
1893       << OrigFormatExpr->getSourceRange();
1894   else
1895     Diag(Args[format_idx]->getLocStart(),
1896          diag::warn_format_nonliteral)
1897            << OrigFormatExpr->getSourceRange();
1898   return false;
1899 }
1900 
1901 namespace {
1902 class CheckFormatHandler : public analyze_format_string::FormatStringHandler {
1903 protected:
1904   Sema &S;
1905   const StringLiteral *FExpr;
1906   const Expr *OrigFormatExpr;
1907   const unsigned FirstDataArg;
1908   const unsigned NumDataArgs;
1909   const char *Beg; // Start of format string.
1910   const bool HasVAListArg;
1911   const Expr * const *Args;
1912   const unsigned NumArgs;
1913   unsigned FormatIdx;
1914   llvm::BitVector CoveredArgs;
1915   bool usesPositionalArgs;
1916   bool atFirstArg;
1917   bool inFunctionCall;
1918   Sema::VariadicCallType CallType;
1919 public:
CheckFormatHandler(Sema & s,const StringLiteral * fexpr,const Expr * origFormatExpr,unsigned firstDataArg,unsigned numDataArgs,const char * beg,bool hasVAListArg,Expr ** args,unsigned numArgs,unsigned formatIdx,bool inFunctionCall,Sema::VariadicCallType callType)1920   CheckFormatHandler(Sema &s, const StringLiteral *fexpr,
1921                      const Expr *origFormatExpr, unsigned firstDataArg,
1922                      unsigned numDataArgs, const char *beg, bool hasVAListArg,
1923                      Expr **args, unsigned numArgs,
1924                      unsigned formatIdx, bool inFunctionCall,
1925                      Sema::VariadicCallType callType)
1926     : S(s), FExpr(fexpr), OrigFormatExpr(origFormatExpr),
1927       FirstDataArg(firstDataArg), NumDataArgs(numDataArgs),
1928       Beg(beg), HasVAListArg(hasVAListArg),
1929       Args(args), NumArgs(numArgs), FormatIdx(formatIdx),
1930       usesPositionalArgs(false), atFirstArg(true),
1931       inFunctionCall(inFunctionCall), CallType(callType) {
1932         CoveredArgs.resize(numDataArgs);
1933         CoveredArgs.reset();
1934       }
1935 
1936   void DoneProcessing();
1937 
1938   void HandleIncompleteSpecifier(const char *startSpecifier,
1939                                  unsigned specifierLen);
1940 
1941   void HandleInvalidLengthModifier(
1942       const analyze_format_string::FormatSpecifier &FS,
1943       const analyze_format_string::ConversionSpecifier &CS,
1944       const char *startSpecifier, unsigned specifierLen, unsigned DiagID);
1945 
1946   void HandleNonStandardLengthModifier(
1947       const analyze_format_string::FormatSpecifier &FS,
1948       const char *startSpecifier, unsigned specifierLen);
1949 
1950   void HandleNonStandardConversionSpecifier(
1951       const analyze_format_string::ConversionSpecifier &CS,
1952       const char *startSpecifier, unsigned specifierLen);
1953 
1954   virtual void HandlePosition(const char *startPos, unsigned posLen);
1955 
1956   virtual void HandleInvalidPosition(const char *startSpecifier,
1957                                      unsigned specifierLen,
1958                                      analyze_format_string::PositionContext p);
1959 
1960   virtual void HandleZeroPosition(const char *startPos, unsigned posLen);
1961 
1962   void HandleNullChar(const char *nullCharacter);
1963 
1964   template <typename Range>
1965   static void EmitFormatDiagnostic(Sema &S, bool inFunctionCall,
1966                                    const Expr *ArgumentExpr,
1967                                    PartialDiagnostic PDiag,
1968                                    SourceLocation StringLoc,
1969                                    bool IsStringLocation, Range StringRange,
1970                             ArrayRef<FixItHint> Fixit = ArrayRef<FixItHint>());
1971 
1972 protected:
1973   bool HandleInvalidConversionSpecifier(unsigned argIndex, SourceLocation Loc,
1974                                         const char *startSpec,
1975                                         unsigned specifierLen,
1976                                         const char *csStart, unsigned csLen);
1977 
1978   void HandlePositionalNonpositionalArgs(SourceLocation Loc,
1979                                          const char *startSpec,
1980                                          unsigned specifierLen);
1981 
1982   SourceRange getFormatStringRange();
1983   CharSourceRange getSpecifierRange(const char *startSpecifier,
1984                                     unsigned specifierLen);
1985   SourceLocation getLocationOfByte(const char *x);
1986 
1987   const Expr *getDataArg(unsigned i) const;
1988 
1989   bool CheckNumArgs(const analyze_format_string::FormatSpecifier &FS,
1990                     const analyze_format_string::ConversionSpecifier &CS,
1991                     const char *startSpecifier, unsigned specifierLen,
1992                     unsigned argIndex);
1993 
1994   template <typename Range>
1995   void EmitFormatDiagnostic(PartialDiagnostic PDiag, SourceLocation StringLoc,
1996                             bool IsStringLocation, Range StringRange,
1997                             ArrayRef<FixItHint> Fixit = ArrayRef<FixItHint>());
1998 
1999   void CheckPositionalAndNonpositionalArgs(
2000       const analyze_format_string::FormatSpecifier *FS);
2001 };
2002 }
2003 
getFormatStringRange()2004 SourceRange CheckFormatHandler::getFormatStringRange() {
2005   return OrigFormatExpr->getSourceRange();
2006 }
2007 
2008 CharSourceRange CheckFormatHandler::
getSpecifierRange(const char * startSpecifier,unsigned specifierLen)2009 getSpecifierRange(const char *startSpecifier, unsigned specifierLen) {
2010   SourceLocation Start = getLocationOfByte(startSpecifier);
2011   SourceLocation End   = getLocationOfByte(startSpecifier + specifierLen - 1);
2012 
2013   // Advance the end SourceLocation by one due to half-open ranges.
2014   End = End.getLocWithOffset(1);
2015 
2016   return CharSourceRange::getCharRange(Start, End);
2017 }
2018 
getLocationOfByte(const char * x)2019 SourceLocation CheckFormatHandler::getLocationOfByte(const char *x) {
2020   return S.getLocationOfStringLiteralByte(FExpr, x - Beg);
2021 }
2022 
HandleIncompleteSpecifier(const char * startSpecifier,unsigned specifierLen)2023 void CheckFormatHandler::HandleIncompleteSpecifier(const char *startSpecifier,
2024                                                    unsigned specifierLen){
2025   EmitFormatDiagnostic(S.PDiag(diag::warn_printf_incomplete_specifier),
2026                        getLocationOfByte(startSpecifier),
2027                        /*IsStringLocation*/true,
2028                        getSpecifierRange(startSpecifier, specifierLen));
2029 }
2030 
HandleInvalidLengthModifier(const analyze_format_string::FormatSpecifier & FS,const analyze_format_string::ConversionSpecifier & CS,const char * startSpecifier,unsigned specifierLen,unsigned DiagID)2031 void CheckFormatHandler::HandleInvalidLengthModifier(
2032     const analyze_format_string::FormatSpecifier &FS,
2033     const analyze_format_string::ConversionSpecifier &CS,
2034     const char *startSpecifier, unsigned specifierLen, unsigned DiagID) {
2035   using namespace analyze_format_string;
2036 
2037   const LengthModifier &LM = FS.getLengthModifier();
2038   CharSourceRange LMRange = getSpecifierRange(LM.getStart(), LM.getLength());
2039 
2040   // See if we know how to fix this length modifier.
2041   llvm::Optional<LengthModifier> FixedLM = FS.getCorrectedLengthModifier();
2042   if (FixedLM) {
2043     EmitFormatDiagnostic(S.PDiag(DiagID) << LM.toString() << CS.toString(),
2044                          getLocationOfByte(LM.getStart()),
2045                          /*IsStringLocation*/true,
2046                          getSpecifierRange(startSpecifier, specifierLen));
2047 
2048     S.Diag(getLocationOfByte(LM.getStart()), diag::note_format_fix_specifier)
2049       << FixedLM->toString()
2050       << FixItHint::CreateReplacement(LMRange, FixedLM->toString());
2051 
2052   } else {
2053     FixItHint Hint;
2054     if (DiagID == diag::warn_format_nonsensical_length)
2055       Hint = FixItHint::CreateRemoval(LMRange);
2056 
2057     EmitFormatDiagnostic(S.PDiag(DiagID) << LM.toString() << CS.toString(),
2058                          getLocationOfByte(LM.getStart()),
2059                          /*IsStringLocation*/true,
2060                          getSpecifierRange(startSpecifier, specifierLen),
2061                          Hint);
2062   }
2063 }
2064 
HandleNonStandardLengthModifier(const analyze_format_string::FormatSpecifier & FS,const char * startSpecifier,unsigned specifierLen)2065 void CheckFormatHandler::HandleNonStandardLengthModifier(
2066     const analyze_format_string::FormatSpecifier &FS,
2067     const char *startSpecifier, unsigned specifierLen) {
2068   using namespace analyze_format_string;
2069 
2070   const LengthModifier &LM = FS.getLengthModifier();
2071   CharSourceRange LMRange = getSpecifierRange(LM.getStart(), LM.getLength());
2072 
2073   // See if we know how to fix this length modifier.
2074   llvm::Optional<LengthModifier> FixedLM = FS.getCorrectedLengthModifier();
2075   if (FixedLM) {
2076     EmitFormatDiagnostic(S.PDiag(diag::warn_format_non_standard)
2077                            << LM.toString() << 0,
2078                          getLocationOfByte(LM.getStart()),
2079                          /*IsStringLocation*/true,
2080                          getSpecifierRange(startSpecifier, specifierLen));
2081 
2082     S.Diag(getLocationOfByte(LM.getStart()), diag::note_format_fix_specifier)
2083       << FixedLM->toString()
2084       << FixItHint::CreateReplacement(LMRange, FixedLM->toString());
2085 
2086   } else {
2087     EmitFormatDiagnostic(S.PDiag(diag::warn_format_non_standard)
2088                            << LM.toString() << 0,
2089                          getLocationOfByte(LM.getStart()),
2090                          /*IsStringLocation*/true,
2091                          getSpecifierRange(startSpecifier, specifierLen));
2092   }
2093 }
2094 
HandleNonStandardConversionSpecifier(const analyze_format_string::ConversionSpecifier & CS,const char * startSpecifier,unsigned specifierLen)2095 void CheckFormatHandler::HandleNonStandardConversionSpecifier(
2096     const analyze_format_string::ConversionSpecifier &CS,
2097     const char *startSpecifier, unsigned specifierLen) {
2098   EmitFormatDiagnostic(S.PDiag(diag::warn_format_non_standard) << CS.toString()
2099                        << 1,
2100                        getLocationOfByte(CS.getStart()),
2101                        /*IsStringLocation*/true,
2102                        getSpecifierRange(startSpecifier, specifierLen));
2103 }
2104 
HandlePosition(const char * startPos,unsigned posLen)2105 void CheckFormatHandler::HandlePosition(const char *startPos,
2106                                         unsigned posLen) {
2107   EmitFormatDiagnostic(S.PDiag(diag::warn_format_non_standard_positional_arg),
2108                                getLocationOfByte(startPos),
2109                                /*IsStringLocation*/true,
2110                                getSpecifierRange(startPos, posLen));
2111 }
2112 
2113 void
HandleInvalidPosition(const char * startPos,unsigned posLen,analyze_format_string::PositionContext p)2114 CheckFormatHandler::HandleInvalidPosition(const char *startPos, unsigned posLen,
2115                                      analyze_format_string::PositionContext p) {
2116   EmitFormatDiagnostic(S.PDiag(diag::warn_format_invalid_positional_specifier)
2117                          << (unsigned) p,
2118                        getLocationOfByte(startPos), /*IsStringLocation*/true,
2119                        getSpecifierRange(startPos, posLen));
2120 }
2121 
HandleZeroPosition(const char * startPos,unsigned posLen)2122 void CheckFormatHandler::HandleZeroPosition(const char *startPos,
2123                                             unsigned posLen) {
2124   EmitFormatDiagnostic(S.PDiag(diag::warn_format_zero_positional_specifier),
2125                                getLocationOfByte(startPos),
2126                                /*IsStringLocation*/true,
2127                                getSpecifierRange(startPos, posLen));
2128 }
2129 
HandleNullChar(const char * nullCharacter)2130 void CheckFormatHandler::HandleNullChar(const char *nullCharacter) {
2131   if (!isa<ObjCStringLiteral>(OrigFormatExpr)) {
2132     // The presence of a null character is likely an error.
2133     EmitFormatDiagnostic(
2134       S.PDiag(diag::warn_printf_format_string_contains_null_char),
2135       getLocationOfByte(nullCharacter), /*IsStringLocation*/true,
2136       getFormatStringRange());
2137   }
2138 }
2139 
2140 // Note that this may return NULL if there was an error parsing or building
2141 // one of the argument expressions.
getDataArg(unsigned i) const2142 const Expr *CheckFormatHandler::getDataArg(unsigned i) const {
2143   return Args[FirstDataArg + i];
2144 }
2145 
DoneProcessing()2146 void CheckFormatHandler::DoneProcessing() {
2147     // Does the number of data arguments exceed the number of
2148     // format conversions in the format string?
2149   if (!HasVAListArg) {
2150       // Find any arguments that weren't covered.
2151     CoveredArgs.flip();
2152     signed notCoveredArg = CoveredArgs.find_first();
2153     if (notCoveredArg >= 0) {
2154       assert((unsigned)notCoveredArg < NumDataArgs);
2155       if (const Expr *E = getDataArg((unsigned) notCoveredArg)) {
2156         SourceLocation Loc = E->getLocStart();
2157         if (!S.getSourceManager().isInSystemMacro(Loc)) {
2158           EmitFormatDiagnostic(S.PDiag(diag::warn_printf_data_arg_not_used),
2159                                Loc, /*IsStringLocation*/false,
2160                                getFormatStringRange());
2161         }
2162       }
2163     }
2164   }
2165 }
2166 
2167 bool
HandleInvalidConversionSpecifier(unsigned argIndex,SourceLocation Loc,const char * startSpec,unsigned specifierLen,const char * csStart,unsigned csLen)2168 CheckFormatHandler::HandleInvalidConversionSpecifier(unsigned argIndex,
2169                                                      SourceLocation Loc,
2170                                                      const char *startSpec,
2171                                                      unsigned specifierLen,
2172                                                      const char *csStart,
2173                                                      unsigned csLen) {
2174 
2175   bool keepGoing = true;
2176   if (argIndex < NumDataArgs) {
2177     // Consider the argument coverered, even though the specifier doesn't
2178     // make sense.
2179     CoveredArgs.set(argIndex);
2180   }
2181   else {
2182     // If argIndex exceeds the number of data arguments we
2183     // don't issue a warning because that is just a cascade of warnings (and
2184     // they may have intended '%%' anyway). We don't want to continue processing
2185     // the format string after this point, however, as we will like just get
2186     // gibberish when trying to match arguments.
2187     keepGoing = false;
2188   }
2189 
2190   EmitFormatDiagnostic(S.PDiag(diag::warn_format_invalid_conversion)
2191                          << StringRef(csStart, csLen),
2192                        Loc, /*IsStringLocation*/true,
2193                        getSpecifierRange(startSpec, specifierLen));
2194 
2195   return keepGoing;
2196 }
2197 
2198 void
HandlePositionalNonpositionalArgs(SourceLocation Loc,const char * startSpec,unsigned specifierLen)2199 CheckFormatHandler::HandlePositionalNonpositionalArgs(SourceLocation Loc,
2200                                                       const char *startSpec,
2201                                                       unsigned specifierLen) {
2202   EmitFormatDiagnostic(
2203     S.PDiag(diag::warn_format_mix_positional_nonpositional_args),
2204     Loc, /*isStringLoc*/true, getSpecifierRange(startSpec, specifierLen));
2205 }
2206 
2207 bool
CheckNumArgs(const analyze_format_string::FormatSpecifier & FS,const analyze_format_string::ConversionSpecifier & CS,const char * startSpecifier,unsigned specifierLen,unsigned argIndex)2208 CheckFormatHandler::CheckNumArgs(
2209   const analyze_format_string::FormatSpecifier &FS,
2210   const analyze_format_string::ConversionSpecifier &CS,
2211   const char *startSpecifier, unsigned specifierLen, unsigned argIndex) {
2212 
2213   if (argIndex >= NumDataArgs) {
2214     PartialDiagnostic PDiag = FS.usesPositionalArg()
2215       ? (S.PDiag(diag::warn_printf_positional_arg_exceeds_data_args)
2216            << (argIndex+1) << NumDataArgs)
2217       : S.PDiag(diag::warn_printf_insufficient_data_args);
2218     EmitFormatDiagnostic(
2219       PDiag, getLocationOfByte(CS.getStart()), /*IsStringLocation*/true,
2220       getSpecifierRange(startSpecifier, specifierLen));
2221     return false;
2222   }
2223   return true;
2224 }
2225 
2226 template<typename Range>
EmitFormatDiagnostic(PartialDiagnostic PDiag,SourceLocation Loc,bool IsStringLocation,Range StringRange,ArrayRef<FixItHint> FixIt)2227 void CheckFormatHandler::EmitFormatDiagnostic(PartialDiagnostic PDiag,
2228                                               SourceLocation Loc,
2229                                               bool IsStringLocation,
2230                                               Range StringRange,
2231                                               ArrayRef<FixItHint> FixIt) {
2232   EmitFormatDiagnostic(S, inFunctionCall, Args[FormatIdx], PDiag,
2233                        Loc, IsStringLocation, StringRange, FixIt);
2234 }
2235 
2236 /// \brief If the format string is not within the funcion call, emit a note
2237 /// so that the function call and string are in diagnostic messages.
2238 ///
2239 /// \param InFunctionCall if true, the format string is within the function
2240 /// call and only one diagnostic message will be produced.  Otherwise, an
2241 /// extra note will be emitted pointing to location of the format string.
2242 ///
2243 /// \param ArgumentExpr the expression that is passed as the format string
2244 /// argument in the function call.  Used for getting locations when two
2245 /// diagnostics are emitted.
2246 ///
2247 /// \param PDiag the callee should already have provided any strings for the
2248 /// diagnostic message.  This function only adds locations and fixits
2249 /// to diagnostics.
2250 ///
2251 /// \param Loc primary location for diagnostic.  If two diagnostics are
2252 /// required, one will be at Loc and a new SourceLocation will be created for
2253 /// the other one.
2254 ///
2255 /// \param IsStringLocation if true, Loc points to the format string should be
2256 /// used for the note.  Otherwise, Loc points to the argument list and will
2257 /// be used with PDiag.
2258 ///
2259 /// \param StringRange some or all of the string to highlight.  This is
2260 /// templated so it can accept either a CharSourceRange or a SourceRange.
2261 ///
2262 /// \param FixIt optional fix it hint for the format string.
2263 template<typename Range>
EmitFormatDiagnostic(Sema & S,bool InFunctionCall,const Expr * ArgumentExpr,PartialDiagnostic PDiag,SourceLocation Loc,bool IsStringLocation,Range StringRange,ArrayRef<FixItHint> FixIt)2264 void CheckFormatHandler::EmitFormatDiagnostic(Sema &S, bool InFunctionCall,
2265                                               const Expr *ArgumentExpr,
2266                                               PartialDiagnostic PDiag,
2267                                               SourceLocation Loc,
2268                                               bool IsStringLocation,
2269                                               Range StringRange,
2270                                               ArrayRef<FixItHint> FixIt) {
2271   if (InFunctionCall) {
2272     const Sema::SemaDiagnosticBuilder &D = S.Diag(Loc, PDiag);
2273     D << StringRange;
2274     for (ArrayRef<FixItHint>::iterator I = FixIt.begin(), E = FixIt.end();
2275          I != E; ++I) {
2276       D << *I;
2277     }
2278   } else {
2279     S.Diag(IsStringLocation ? ArgumentExpr->getExprLoc() : Loc, PDiag)
2280       << ArgumentExpr->getSourceRange();
2281 
2282     const Sema::SemaDiagnosticBuilder &Note =
2283       S.Diag(IsStringLocation ? Loc : StringRange.getBegin(),
2284              diag::note_format_string_defined);
2285 
2286     Note << StringRange;
2287     for (ArrayRef<FixItHint>::iterator I = FixIt.begin(), E = FixIt.end();
2288          I != E; ++I) {
2289       Note << *I;
2290     }
2291   }
2292 }
2293 
2294 //===--- CHECK: Printf format string checking ------------------------------===//
2295 
2296 namespace {
2297 class CheckPrintfHandler : public CheckFormatHandler {
2298   bool ObjCContext;
2299 public:
CheckPrintfHandler(Sema & s,const StringLiteral * fexpr,const Expr * origFormatExpr,unsigned firstDataArg,unsigned numDataArgs,bool isObjC,const char * beg,bool hasVAListArg,Expr ** Args,unsigned NumArgs,unsigned formatIdx,bool inFunctionCall,Sema::VariadicCallType CallType)2300   CheckPrintfHandler(Sema &s, const StringLiteral *fexpr,
2301                      const Expr *origFormatExpr, unsigned firstDataArg,
2302                      unsigned numDataArgs, bool isObjC,
2303                      const char *beg, bool hasVAListArg,
2304                      Expr **Args, unsigned NumArgs,
2305                      unsigned formatIdx, bool inFunctionCall,
2306                      Sema::VariadicCallType CallType)
2307   : CheckFormatHandler(s, fexpr, origFormatExpr, firstDataArg,
2308                        numDataArgs, beg, hasVAListArg, Args, NumArgs,
2309                        formatIdx, inFunctionCall, CallType), ObjCContext(isObjC)
2310   {}
2311 
2312 
2313   bool HandleInvalidPrintfConversionSpecifier(
2314                                       const analyze_printf::PrintfSpecifier &FS,
2315                                       const char *startSpecifier,
2316                                       unsigned specifierLen);
2317 
2318   bool HandlePrintfSpecifier(const analyze_printf::PrintfSpecifier &FS,
2319                              const char *startSpecifier,
2320                              unsigned specifierLen);
2321   bool checkFormatExpr(const analyze_printf::PrintfSpecifier &FS,
2322                        const char *StartSpecifier,
2323                        unsigned SpecifierLen,
2324                        const Expr *E);
2325 
2326   bool HandleAmount(const analyze_format_string::OptionalAmount &Amt, unsigned k,
2327                     const char *startSpecifier, unsigned specifierLen);
2328   void HandleInvalidAmount(const analyze_printf::PrintfSpecifier &FS,
2329                            const analyze_printf::OptionalAmount &Amt,
2330                            unsigned type,
2331                            const char *startSpecifier, unsigned specifierLen);
2332   void HandleFlag(const analyze_printf::PrintfSpecifier &FS,
2333                   const analyze_printf::OptionalFlag &flag,
2334                   const char *startSpecifier, unsigned specifierLen);
2335   void HandleIgnoredFlag(const analyze_printf::PrintfSpecifier &FS,
2336                          const analyze_printf::OptionalFlag &ignoredFlag,
2337                          const analyze_printf::OptionalFlag &flag,
2338                          const char *startSpecifier, unsigned specifierLen);
2339   bool checkForCStrMembers(const analyze_printf::ArgType &AT,
2340                            const Expr *E, const CharSourceRange &CSR);
2341 
2342 };
2343 }
2344 
HandleInvalidPrintfConversionSpecifier(const analyze_printf::PrintfSpecifier & FS,const char * startSpecifier,unsigned specifierLen)2345 bool CheckPrintfHandler::HandleInvalidPrintfConversionSpecifier(
2346                                       const analyze_printf::PrintfSpecifier &FS,
2347                                       const char *startSpecifier,
2348                                       unsigned specifierLen) {
2349   const analyze_printf::PrintfConversionSpecifier &CS =
2350     FS.getConversionSpecifier();
2351 
2352   return HandleInvalidConversionSpecifier(FS.getArgIndex(),
2353                                           getLocationOfByte(CS.getStart()),
2354                                           startSpecifier, specifierLen,
2355                                           CS.getStart(), CS.getLength());
2356 }
2357 
HandleAmount(const analyze_format_string::OptionalAmount & Amt,unsigned k,const char * startSpecifier,unsigned specifierLen)2358 bool CheckPrintfHandler::HandleAmount(
2359                                const analyze_format_string::OptionalAmount &Amt,
2360                                unsigned k, const char *startSpecifier,
2361                                unsigned specifierLen) {
2362 
2363   if (Amt.hasDataArgument()) {
2364     if (!HasVAListArg) {
2365       unsigned argIndex = Amt.getArgIndex();
2366       if (argIndex >= NumDataArgs) {
2367         EmitFormatDiagnostic(S.PDiag(diag::warn_printf_asterisk_missing_arg)
2368                                << k,
2369                              getLocationOfByte(Amt.getStart()),
2370                              /*IsStringLocation*/true,
2371                              getSpecifierRange(startSpecifier, specifierLen));
2372         // Don't do any more checking.  We will just emit
2373         // spurious errors.
2374         return false;
2375       }
2376 
2377       // Type check the data argument.  It should be an 'int'.
2378       // Although not in conformance with C99, we also allow the argument to be
2379       // an 'unsigned int' as that is a reasonably safe case.  GCC also
2380       // doesn't emit a warning for that case.
2381       CoveredArgs.set(argIndex);
2382       const Expr *Arg = getDataArg(argIndex);
2383       if (!Arg)
2384         return false;
2385 
2386       QualType T = Arg->getType();
2387 
2388       const analyze_printf::ArgType &AT = Amt.getArgType(S.Context);
2389       assert(AT.isValid());
2390 
2391       if (!AT.matchesType(S.Context, T)) {
2392         EmitFormatDiagnostic(S.PDiag(diag::warn_printf_asterisk_wrong_type)
2393                                << k << AT.getRepresentativeTypeName(S.Context)
2394                                << T << Arg->getSourceRange(),
2395                              getLocationOfByte(Amt.getStart()),
2396                              /*IsStringLocation*/true,
2397                              getSpecifierRange(startSpecifier, specifierLen));
2398         // Don't do any more checking.  We will just emit
2399         // spurious errors.
2400         return false;
2401       }
2402     }
2403   }
2404   return true;
2405 }
2406 
HandleInvalidAmount(const analyze_printf::PrintfSpecifier & FS,const analyze_printf::OptionalAmount & Amt,unsigned type,const char * startSpecifier,unsigned specifierLen)2407 void CheckPrintfHandler::HandleInvalidAmount(
2408                                       const analyze_printf::PrintfSpecifier &FS,
2409                                       const analyze_printf::OptionalAmount &Amt,
2410                                       unsigned type,
2411                                       const char *startSpecifier,
2412                                       unsigned specifierLen) {
2413   const analyze_printf::PrintfConversionSpecifier &CS =
2414     FS.getConversionSpecifier();
2415 
2416   FixItHint fixit =
2417     Amt.getHowSpecified() == analyze_printf::OptionalAmount::Constant
2418       ? FixItHint::CreateRemoval(getSpecifierRange(Amt.getStart(),
2419                                  Amt.getConstantLength()))
2420       : FixItHint();
2421 
2422   EmitFormatDiagnostic(S.PDiag(diag::warn_printf_nonsensical_optional_amount)
2423                          << type << CS.toString(),
2424                        getLocationOfByte(Amt.getStart()),
2425                        /*IsStringLocation*/true,
2426                        getSpecifierRange(startSpecifier, specifierLen),
2427                        fixit);
2428 }
2429 
HandleFlag(const analyze_printf::PrintfSpecifier & FS,const analyze_printf::OptionalFlag & flag,const char * startSpecifier,unsigned specifierLen)2430 void CheckPrintfHandler::HandleFlag(const analyze_printf::PrintfSpecifier &FS,
2431                                     const analyze_printf::OptionalFlag &flag,
2432                                     const char *startSpecifier,
2433                                     unsigned specifierLen) {
2434   // Warn about pointless flag with a fixit removal.
2435   const analyze_printf::PrintfConversionSpecifier &CS =
2436     FS.getConversionSpecifier();
2437   EmitFormatDiagnostic(S.PDiag(diag::warn_printf_nonsensical_flag)
2438                          << flag.toString() << CS.toString(),
2439                        getLocationOfByte(flag.getPosition()),
2440                        /*IsStringLocation*/true,
2441                        getSpecifierRange(startSpecifier, specifierLen),
2442                        FixItHint::CreateRemoval(
2443                          getSpecifierRange(flag.getPosition(), 1)));
2444 }
2445 
HandleIgnoredFlag(const analyze_printf::PrintfSpecifier & FS,const analyze_printf::OptionalFlag & ignoredFlag,const analyze_printf::OptionalFlag & flag,const char * startSpecifier,unsigned specifierLen)2446 void CheckPrintfHandler::HandleIgnoredFlag(
2447                                 const analyze_printf::PrintfSpecifier &FS,
2448                                 const analyze_printf::OptionalFlag &ignoredFlag,
2449                                 const analyze_printf::OptionalFlag &flag,
2450                                 const char *startSpecifier,
2451                                 unsigned specifierLen) {
2452   // Warn about ignored flag with a fixit removal.
2453   EmitFormatDiagnostic(S.PDiag(diag::warn_printf_ignored_flag)
2454                          << ignoredFlag.toString() << flag.toString(),
2455                        getLocationOfByte(ignoredFlag.getPosition()),
2456                        /*IsStringLocation*/true,
2457                        getSpecifierRange(startSpecifier, specifierLen),
2458                        FixItHint::CreateRemoval(
2459                          getSpecifierRange(ignoredFlag.getPosition(), 1)));
2460 }
2461 
2462 // Determines if the specified is a C++ class or struct containing
2463 // a member with the specified name and kind (e.g. a CXXMethodDecl named
2464 // "c_str()").
2465 template<typename MemberKind>
2466 static llvm::SmallPtrSet<MemberKind*, 1>
CXXRecordMembersNamed(StringRef Name,Sema & S,QualType Ty)2467 CXXRecordMembersNamed(StringRef Name, Sema &S, QualType Ty) {
2468   const RecordType *RT = Ty->getAs<RecordType>();
2469   llvm::SmallPtrSet<MemberKind*, 1> Results;
2470 
2471   if (!RT)
2472     return Results;
2473   const CXXRecordDecl *RD = dyn_cast<CXXRecordDecl>(RT->getDecl());
2474   if (!RD)
2475     return Results;
2476 
2477   LookupResult R(S, &S.PP.getIdentifierTable().get(Name), SourceLocation(),
2478                  Sema::LookupMemberName);
2479 
2480   // We just need to include all members of the right kind turned up by the
2481   // filter, at this point.
2482   if (S.LookupQualifiedName(R, RT->getDecl()))
2483     for (LookupResult::iterator I = R.begin(), E = R.end(); I != E; ++I) {
2484       NamedDecl *decl = (*I)->getUnderlyingDecl();
2485       if (MemberKind *FK = dyn_cast<MemberKind>(decl))
2486         Results.insert(FK);
2487     }
2488   return Results;
2489 }
2490 
2491 // Check if a (w)string was passed when a (w)char* was needed, and offer a
2492 // better diagnostic if so. AT is assumed to be valid.
2493 // Returns true when a c_str() conversion method is found.
checkForCStrMembers(const analyze_printf::ArgType & AT,const Expr * E,const CharSourceRange & CSR)2494 bool CheckPrintfHandler::checkForCStrMembers(
2495     const analyze_printf::ArgType &AT, const Expr *E,
2496     const CharSourceRange &CSR) {
2497   typedef llvm::SmallPtrSet<CXXMethodDecl*, 1> MethodSet;
2498 
2499   MethodSet Results =
2500       CXXRecordMembersNamed<CXXMethodDecl>("c_str", S, E->getType());
2501 
2502   for (MethodSet::iterator MI = Results.begin(), ME = Results.end();
2503        MI != ME; ++MI) {
2504     const CXXMethodDecl *Method = *MI;
2505     if (Method->getNumParams() == 0 &&
2506           AT.matchesType(S.Context, Method->getResultType())) {
2507       // FIXME: Suggest parens if the expression needs them.
2508       SourceLocation EndLoc =
2509           S.getPreprocessor().getLocForEndOfToken(E->getLocEnd());
2510       S.Diag(E->getLocStart(), diag::note_printf_c_str)
2511           << "c_str()"
2512           << FixItHint::CreateInsertion(EndLoc, ".c_str()");
2513       return true;
2514     }
2515   }
2516 
2517   return false;
2518 }
2519 
2520 bool
HandlePrintfSpecifier(const analyze_printf::PrintfSpecifier & FS,const char * startSpecifier,unsigned specifierLen)2521 CheckPrintfHandler::HandlePrintfSpecifier(const analyze_printf::PrintfSpecifier
2522                                             &FS,
2523                                           const char *startSpecifier,
2524                                           unsigned specifierLen) {
2525 
2526   using namespace analyze_format_string;
2527   using namespace analyze_printf;
2528   const PrintfConversionSpecifier &CS = FS.getConversionSpecifier();
2529 
2530   if (FS.consumesDataArgument()) {
2531     if (atFirstArg) {
2532         atFirstArg = false;
2533         usesPositionalArgs = FS.usesPositionalArg();
2534     }
2535     else if (usesPositionalArgs != FS.usesPositionalArg()) {
2536       HandlePositionalNonpositionalArgs(getLocationOfByte(CS.getStart()),
2537                                         startSpecifier, specifierLen);
2538       return false;
2539     }
2540   }
2541 
2542   // First check if the field width, precision, and conversion specifier
2543   // have matching data arguments.
2544   if (!HandleAmount(FS.getFieldWidth(), /* field width */ 0,
2545                     startSpecifier, specifierLen)) {
2546     return false;
2547   }
2548 
2549   if (!HandleAmount(FS.getPrecision(), /* precision */ 1,
2550                     startSpecifier, specifierLen)) {
2551     return false;
2552   }
2553 
2554   if (!CS.consumesDataArgument()) {
2555     // FIXME: Technically specifying a precision or field width here
2556     // makes no sense.  Worth issuing a warning at some point.
2557     return true;
2558   }
2559 
2560   // Consume the argument.
2561   unsigned argIndex = FS.getArgIndex();
2562   if (argIndex < NumDataArgs) {
2563     // The check to see if the argIndex is valid will come later.
2564     // We set the bit here because we may exit early from this
2565     // function if we encounter some other error.
2566     CoveredArgs.set(argIndex);
2567   }
2568 
2569   // Check for using an Objective-C specific conversion specifier
2570   // in a non-ObjC literal.
2571   if (!ObjCContext && CS.isObjCArg()) {
2572     return HandleInvalidPrintfConversionSpecifier(FS, startSpecifier,
2573                                                   specifierLen);
2574   }
2575 
2576   // Check for invalid use of field width
2577   if (!FS.hasValidFieldWidth()) {
2578     HandleInvalidAmount(FS, FS.getFieldWidth(), /* field width */ 0,
2579         startSpecifier, specifierLen);
2580   }
2581 
2582   // Check for invalid use of precision
2583   if (!FS.hasValidPrecision()) {
2584     HandleInvalidAmount(FS, FS.getPrecision(), /* precision */ 1,
2585         startSpecifier, specifierLen);
2586   }
2587 
2588   // Check each flag does not conflict with any other component.
2589   if (!FS.hasValidThousandsGroupingPrefix())
2590     HandleFlag(FS, FS.hasThousandsGrouping(), startSpecifier, specifierLen);
2591   if (!FS.hasValidLeadingZeros())
2592     HandleFlag(FS, FS.hasLeadingZeros(), startSpecifier, specifierLen);
2593   if (!FS.hasValidPlusPrefix())
2594     HandleFlag(FS, FS.hasPlusPrefix(), startSpecifier, specifierLen);
2595   if (!FS.hasValidSpacePrefix())
2596     HandleFlag(FS, FS.hasSpacePrefix(), startSpecifier, specifierLen);
2597   if (!FS.hasValidAlternativeForm())
2598     HandleFlag(FS, FS.hasAlternativeForm(), startSpecifier, specifierLen);
2599   if (!FS.hasValidLeftJustified())
2600     HandleFlag(FS, FS.isLeftJustified(), startSpecifier, specifierLen);
2601 
2602   // Check that flags are not ignored by another flag
2603   if (FS.hasSpacePrefix() && FS.hasPlusPrefix()) // ' ' ignored by '+'
2604     HandleIgnoredFlag(FS, FS.hasSpacePrefix(), FS.hasPlusPrefix(),
2605         startSpecifier, specifierLen);
2606   if (FS.hasLeadingZeros() && FS.isLeftJustified()) // '0' ignored by '-'
2607     HandleIgnoredFlag(FS, FS.hasLeadingZeros(), FS.isLeftJustified(),
2608             startSpecifier, specifierLen);
2609 
2610   // Check the length modifier is valid with the given conversion specifier.
2611   if (!FS.hasValidLengthModifier(S.getASTContext().getTargetInfo()))
2612     HandleInvalidLengthModifier(FS, CS, startSpecifier, specifierLen,
2613                                 diag::warn_format_nonsensical_length);
2614   else if (!FS.hasStandardLengthModifier())
2615     HandleNonStandardLengthModifier(FS, startSpecifier, specifierLen);
2616   else if (!FS.hasStandardLengthConversionCombination())
2617     HandleInvalidLengthModifier(FS, CS, startSpecifier, specifierLen,
2618                                 diag::warn_format_non_standard_conversion_spec);
2619 
2620   if (!FS.hasStandardConversionSpecifier(S.getLangOpts()))
2621     HandleNonStandardConversionSpecifier(CS, startSpecifier, specifierLen);
2622 
2623   // The remaining checks depend on the data arguments.
2624   if (HasVAListArg)
2625     return true;
2626 
2627   if (!CheckNumArgs(FS, CS, startSpecifier, specifierLen, argIndex))
2628     return false;
2629 
2630   const Expr *Arg = getDataArg(argIndex);
2631   if (!Arg)
2632     return true;
2633 
2634   return checkFormatExpr(FS, startSpecifier, specifierLen, Arg);
2635 }
2636 
requiresParensToAddCast(const Expr * E)2637 static bool requiresParensToAddCast(const Expr *E) {
2638   // FIXME: We should have a general way to reason about operator
2639   // precedence and whether parens are actually needed here.
2640   // Take care of a few common cases where they aren't.
2641   const Expr *Inside = E->IgnoreImpCasts();
2642   if (const PseudoObjectExpr *POE = dyn_cast<PseudoObjectExpr>(Inside))
2643     Inside = POE->getSyntacticForm()->IgnoreImpCasts();
2644 
2645   switch (Inside->getStmtClass()) {
2646   case Stmt::ArraySubscriptExprClass:
2647   case Stmt::CallExprClass:
2648   case Stmt::DeclRefExprClass:
2649   case Stmt::MemberExprClass:
2650   case Stmt::ObjCIvarRefExprClass:
2651   case Stmt::ObjCMessageExprClass:
2652   case Stmt::ObjCPropertyRefExprClass:
2653   case Stmt::ParenExprClass:
2654   case Stmt::UnaryOperatorClass:
2655     return false;
2656   default:
2657     return true;
2658   }
2659 }
2660 
2661 bool
checkFormatExpr(const analyze_printf::PrintfSpecifier & FS,const char * StartSpecifier,unsigned SpecifierLen,const Expr * E)2662 CheckPrintfHandler::checkFormatExpr(const analyze_printf::PrintfSpecifier &FS,
2663                                     const char *StartSpecifier,
2664                                     unsigned SpecifierLen,
2665                                     const Expr *E) {
2666   using namespace analyze_format_string;
2667   using namespace analyze_printf;
2668   // Now type check the data expression that matches the
2669   // format specifier.
2670   const analyze_printf::ArgType &AT = FS.getArgType(S.Context,
2671                                                     ObjCContext);
2672   if (!AT.isValid())
2673     return true;
2674 
2675   QualType IntendedTy = E->getType();
2676   if (AT.matchesType(S.Context, IntendedTy))
2677     return true;
2678 
2679   // Look through argument promotions for our error message's reported type.
2680   // This includes the integral and floating promotions, but excludes array
2681   // and function pointer decay; seeing that an argument intended to be a
2682   // string has type 'char [6]' is probably more confusing than 'char *'.
2683   if (const ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(E)) {
2684     if (ICE->getCastKind() == CK_IntegralCast ||
2685         ICE->getCastKind() == CK_FloatingCast) {
2686       E = ICE->getSubExpr();
2687       IntendedTy = E->getType();
2688 
2689       // Check if we didn't match because of an implicit cast from a 'char'
2690       // or 'short' to an 'int'.  This is done because printf is a varargs
2691       // function.
2692       if (ICE->getType() == S.Context.IntTy ||
2693           ICE->getType() == S.Context.UnsignedIntTy) {
2694         // All further checking is done on the subexpression.
2695         if (AT.matchesType(S.Context, IntendedTy))
2696           return true;
2697       }
2698     }
2699   }
2700 
2701   if (S.Context.getTargetInfo().getTriple().isOSDarwin()) {
2702     // Special-case some of Darwin's platform-independence types.
2703     if (const TypedefType *UserTy = IntendedTy->getAs<TypedefType>()) {
2704       StringRef Name = UserTy->getDecl()->getName();
2705       IntendedTy = llvm::StringSwitch<QualType>(Name)
2706         .Case("NSInteger", S.Context.LongTy)
2707         .Case("NSUInteger", S.Context.UnsignedLongTy)
2708         .Case("SInt32", S.Context.IntTy)
2709         .Case("UInt32", S.Context.UnsignedIntTy)
2710         .Default(IntendedTy);
2711     }
2712   }
2713 
2714   // We may be able to offer a FixItHint if it is a supported type.
2715   PrintfSpecifier fixedFS = FS;
2716   bool success = fixedFS.fixType(IntendedTy, S.getLangOpts(),
2717                                  S.Context, ObjCContext);
2718 
2719   if (success) {
2720     // Get the fix string from the fixed format specifier
2721     SmallString<16> buf;
2722     llvm::raw_svector_ostream os(buf);
2723     fixedFS.toString(os);
2724 
2725     CharSourceRange SpecRange = getSpecifierRange(StartSpecifier, SpecifierLen);
2726 
2727     if (IntendedTy != E->getType()) {
2728       // The canonical type for formatting this value is different from the
2729       // actual type of the expression. (This occurs, for example, with Darwin's
2730       // NSInteger on 32-bit platforms, where it is typedef'd as 'int', but
2731       // should be printed as 'long' for 64-bit compatibility.)
2732       // Rather than emitting a normal format/argument mismatch, we want to
2733       // add a cast to the recommended type (and correct the format string
2734       // if necessary).
2735       SmallString<16> CastBuf;
2736       llvm::raw_svector_ostream CastFix(CastBuf);
2737       CastFix << "(";
2738       IntendedTy.print(CastFix, S.Context.getPrintingPolicy());
2739       CastFix << ")";
2740 
2741       SmallVector<FixItHint,4> Hints;
2742       if (!AT.matchesType(S.Context, IntendedTy))
2743         Hints.push_back(FixItHint::CreateReplacement(SpecRange, os.str()));
2744 
2745       if (const CStyleCastExpr *CCast = dyn_cast<CStyleCastExpr>(E)) {
2746         // If there's already a cast present, just replace it.
2747         SourceRange CastRange(CCast->getLParenLoc(), CCast->getRParenLoc());
2748         Hints.push_back(FixItHint::CreateReplacement(CastRange, CastFix.str()));
2749 
2750       } else if (!requiresParensToAddCast(E)) {
2751         // If the expression has high enough precedence,
2752         // just write the C-style cast.
2753         Hints.push_back(FixItHint::CreateInsertion(E->getLocStart(),
2754                                                    CastFix.str()));
2755       } else {
2756         // Otherwise, add parens around the expression as well as the cast.
2757         CastFix << "(";
2758         Hints.push_back(FixItHint::CreateInsertion(E->getLocStart(),
2759                                                    CastFix.str()));
2760 
2761         SourceLocation After = S.PP.getLocForEndOfToken(E->getLocEnd());
2762         Hints.push_back(FixItHint::CreateInsertion(After, ")"));
2763       }
2764 
2765       // We extract the name from the typedef because we don't want to show
2766       // the underlying type in the diagnostic.
2767       const TypedefType *UserTy = cast<TypedefType>(E->getType());
2768       StringRef Name = UserTy->getDecl()->getName();
2769 
2770       // Finally, emit the diagnostic.
2771       EmitFormatDiagnostic(S.PDiag(diag::warn_format_argument_needs_cast)
2772                              << Name << IntendedTy
2773                              << E->getSourceRange(),
2774                            E->getLocStart(), /*IsStringLocation=*/false,
2775                            SpecRange, Hints);
2776     } else {
2777       EmitFormatDiagnostic(
2778         S.PDiag(diag::warn_printf_conversion_argument_type_mismatch)
2779           << AT.getRepresentativeTypeName(S.Context) << IntendedTy
2780           << E->getSourceRange(),
2781         E->getLocStart(),
2782         /*IsStringLocation*/false,
2783         SpecRange,
2784         FixItHint::CreateReplacement(SpecRange, os.str()));
2785     }
2786   } else {
2787     const CharSourceRange &CSR = getSpecifierRange(StartSpecifier,
2788                                                    SpecifierLen);
2789     // Since the warning for passing non-POD types to variadic functions
2790     // was deferred until now, we emit a warning for non-POD
2791     // arguments here.
2792     if (S.isValidVarArgType(E->getType()) == Sema::VAK_Invalid) {
2793       unsigned DiagKind;
2794       if (E->getType()->isObjCObjectType())
2795         DiagKind = diag::err_cannot_pass_objc_interface_to_vararg_format;
2796       else
2797         DiagKind = diag::warn_non_pod_vararg_with_format_string;
2798 
2799       EmitFormatDiagnostic(
2800         S.PDiag(DiagKind)
2801           << S.getLangOpts().CPlusPlus0x
2802           << E->getType()
2803           << CallType
2804           << AT.getRepresentativeTypeName(S.Context)
2805           << CSR
2806           << E->getSourceRange(),
2807         E->getLocStart(), /*IsStringLocation*/false, CSR);
2808 
2809       checkForCStrMembers(AT, E, CSR);
2810     } else
2811       EmitFormatDiagnostic(
2812         S.PDiag(diag::warn_printf_conversion_argument_type_mismatch)
2813           << AT.getRepresentativeTypeName(S.Context) << E->getType()
2814           << CSR
2815           << E->getSourceRange(),
2816         E->getLocStart(), /*IsStringLocation*/false, CSR);
2817   }
2818 
2819   return true;
2820 }
2821 
2822 //===--- CHECK: Scanf format string checking ------------------------------===//
2823 
2824 namespace {
2825 class CheckScanfHandler : public CheckFormatHandler {
2826 public:
CheckScanfHandler(Sema & s,const StringLiteral * fexpr,const Expr * origFormatExpr,unsigned firstDataArg,unsigned numDataArgs,const char * beg,bool hasVAListArg,Expr ** Args,unsigned NumArgs,unsigned formatIdx,bool inFunctionCall,Sema::VariadicCallType CallType)2827   CheckScanfHandler(Sema &s, const StringLiteral *fexpr,
2828                     const Expr *origFormatExpr, unsigned firstDataArg,
2829                     unsigned numDataArgs, const char *beg, bool hasVAListArg,
2830                     Expr **Args, unsigned NumArgs,
2831                     unsigned formatIdx, bool inFunctionCall,
2832                     Sema::VariadicCallType CallType)
2833   : CheckFormatHandler(s, fexpr, origFormatExpr, firstDataArg,
2834                        numDataArgs, beg, hasVAListArg,
2835                        Args, NumArgs, formatIdx, inFunctionCall, CallType)
2836   {}
2837 
2838   bool HandleScanfSpecifier(const analyze_scanf::ScanfSpecifier &FS,
2839                             const char *startSpecifier,
2840                             unsigned specifierLen);
2841 
2842   bool HandleInvalidScanfConversionSpecifier(
2843           const analyze_scanf::ScanfSpecifier &FS,
2844           const char *startSpecifier,
2845           unsigned specifierLen);
2846 
2847   void HandleIncompleteScanList(const char *start, const char *end);
2848 };
2849 }
2850 
HandleIncompleteScanList(const char * start,const char * end)2851 void CheckScanfHandler::HandleIncompleteScanList(const char *start,
2852                                                  const char *end) {
2853   EmitFormatDiagnostic(S.PDiag(diag::warn_scanf_scanlist_incomplete),
2854                        getLocationOfByte(end), /*IsStringLocation*/true,
2855                        getSpecifierRange(start, end - start));
2856 }
2857 
HandleInvalidScanfConversionSpecifier(const analyze_scanf::ScanfSpecifier & FS,const char * startSpecifier,unsigned specifierLen)2858 bool CheckScanfHandler::HandleInvalidScanfConversionSpecifier(
2859                                         const analyze_scanf::ScanfSpecifier &FS,
2860                                         const char *startSpecifier,
2861                                         unsigned specifierLen) {
2862 
2863   const analyze_scanf::ScanfConversionSpecifier &CS =
2864     FS.getConversionSpecifier();
2865 
2866   return HandleInvalidConversionSpecifier(FS.getArgIndex(),
2867                                           getLocationOfByte(CS.getStart()),
2868                                           startSpecifier, specifierLen,
2869                                           CS.getStart(), CS.getLength());
2870 }
2871 
HandleScanfSpecifier(const analyze_scanf::ScanfSpecifier & FS,const char * startSpecifier,unsigned specifierLen)2872 bool CheckScanfHandler::HandleScanfSpecifier(
2873                                        const analyze_scanf::ScanfSpecifier &FS,
2874                                        const char *startSpecifier,
2875                                        unsigned specifierLen) {
2876 
2877   using namespace analyze_scanf;
2878   using namespace analyze_format_string;
2879 
2880   const ScanfConversionSpecifier &CS = FS.getConversionSpecifier();
2881 
2882   // Handle case where '%' and '*' don't consume an argument.  These shouldn't
2883   // be used to decide if we are using positional arguments consistently.
2884   if (FS.consumesDataArgument()) {
2885     if (atFirstArg) {
2886       atFirstArg = false;
2887       usesPositionalArgs = FS.usesPositionalArg();
2888     }
2889     else if (usesPositionalArgs != FS.usesPositionalArg()) {
2890       HandlePositionalNonpositionalArgs(getLocationOfByte(CS.getStart()),
2891                                         startSpecifier, specifierLen);
2892       return false;
2893     }
2894   }
2895 
2896   // Check if the field with is non-zero.
2897   const OptionalAmount &Amt = FS.getFieldWidth();
2898   if (Amt.getHowSpecified() == OptionalAmount::Constant) {
2899     if (Amt.getConstantAmount() == 0) {
2900       const CharSourceRange &R = getSpecifierRange(Amt.getStart(),
2901                                                    Amt.getConstantLength());
2902       EmitFormatDiagnostic(S.PDiag(diag::warn_scanf_nonzero_width),
2903                            getLocationOfByte(Amt.getStart()),
2904                            /*IsStringLocation*/true, R,
2905                            FixItHint::CreateRemoval(R));
2906     }
2907   }
2908 
2909   if (!FS.consumesDataArgument()) {
2910     // FIXME: Technically specifying a precision or field width here
2911     // makes no sense.  Worth issuing a warning at some point.
2912     return true;
2913   }
2914 
2915   // Consume the argument.
2916   unsigned argIndex = FS.getArgIndex();
2917   if (argIndex < NumDataArgs) {
2918       // The check to see if the argIndex is valid will come later.
2919       // We set the bit here because we may exit early from this
2920       // function if we encounter some other error.
2921     CoveredArgs.set(argIndex);
2922   }
2923 
2924   // Check the length modifier is valid with the given conversion specifier.
2925   if (!FS.hasValidLengthModifier(S.getASTContext().getTargetInfo()))
2926     HandleInvalidLengthModifier(FS, CS, startSpecifier, specifierLen,
2927                                 diag::warn_format_nonsensical_length);
2928   else if (!FS.hasStandardLengthModifier())
2929     HandleNonStandardLengthModifier(FS, startSpecifier, specifierLen);
2930   else if (!FS.hasStandardLengthConversionCombination())
2931     HandleInvalidLengthModifier(FS, CS, startSpecifier, specifierLen,
2932                                 diag::warn_format_non_standard_conversion_spec);
2933 
2934   if (!FS.hasStandardConversionSpecifier(S.getLangOpts()))
2935     HandleNonStandardConversionSpecifier(CS, startSpecifier, specifierLen);
2936 
2937   // The remaining checks depend on the data arguments.
2938   if (HasVAListArg)
2939     return true;
2940 
2941   if (!CheckNumArgs(FS, CS, startSpecifier, specifierLen, argIndex))
2942     return false;
2943 
2944   // Check that the argument type matches the format specifier.
2945   const Expr *Ex = getDataArg(argIndex);
2946   if (!Ex)
2947     return true;
2948 
2949   const analyze_format_string::ArgType &AT = FS.getArgType(S.Context);
2950   if (AT.isValid() && !AT.matchesType(S.Context, Ex->getType())) {
2951     ScanfSpecifier fixedFS = FS;
2952     bool success = fixedFS.fixType(Ex->getType(), S.getLangOpts(),
2953                                    S.Context);
2954 
2955     if (success) {
2956       // Get the fix string from the fixed format specifier.
2957       SmallString<128> buf;
2958       llvm::raw_svector_ostream os(buf);
2959       fixedFS.toString(os);
2960 
2961       EmitFormatDiagnostic(
2962         S.PDiag(diag::warn_printf_conversion_argument_type_mismatch)
2963           << AT.getRepresentativeTypeName(S.Context) << Ex->getType()
2964           << Ex->getSourceRange(),
2965         Ex->getLocStart(),
2966         /*IsStringLocation*/false,
2967         getSpecifierRange(startSpecifier, specifierLen),
2968         FixItHint::CreateReplacement(
2969           getSpecifierRange(startSpecifier, specifierLen),
2970           os.str()));
2971     } else {
2972       EmitFormatDiagnostic(
2973         S.PDiag(diag::warn_printf_conversion_argument_type_mismatch)
2974           << AT.getRepresentativeTypeName(S.Context) << Ex->getType()
2975           << Ex->getSourceRange(),
2976         Ex->getLocStart(),
2977         /*IsStringLocation*/false,
2978         getSpecifierRange(startSpecifier, specifierLen));
2979     }
2980   }
2981 
2982   return true;
2983 }
2984 
CheckFormatString(const StringLiteral * FExpr,const Expr * OrigFormatExpr,Expr ** Args,unsigned NumArgs,bool HasVAListArg,unsigned format_idx,unsigned firstDataArg,FormatStringType Type,bool inFunctionCall,VariadicCallType CallType)2985 void Sema::CheckFormatString(const StringLiteral *FExpr,
2986                              const Expr *OrigFormatExpr,
2987                              Expr **Args, unsigned NumArgs,
2988                              bool HasVAListArg, unsigned format_idx,
2989                              unsigned firstDataArg, FormatStringType Type,
2990                              bool inFunctionCall, VariadicCallType CallType) {
2991 
2992   // CHECK: is the format string a wide literal?
2993   if (!FExpr->isAscii() && !FExpr->isUTF8()) {
2994     CheckFormatHandler::EmitFormatDiagnostic(
2995       *this, inFunctionCall, Args[format_idx],
2996       PDiag(diag::warn_format_string_is_wide_literal), FExpr->getLocStart(),
2997       /*IsStringLocation*/true, OrigFormatExpr->getSourceRange());
2998     return;
2999   }
3000 
3001   // Str - The format string.  NOTE: this is NOT null-terminated!
3002   StringRef StrRef = FExpr->getString();
3003   const char *Str = StrRef.data();
3004   unsigned StrLen = StrRef.size();
3005   const unsigned numDataArgs = NumArgs - firstDataArg;
3006 
3007   // CHECK: empty format string?
3008   if (StrLen == 0 && numDataArgs > 0) {
3009     CheckFormatHandler::EmitFormatDiagnostic(
3010       *this, inFunctionCall, Args[format_idx],
3011       PDiag(diag::warn_empty_format_string), FExpr->getLocStart(),
3012       /*IsStringLocation*/true, OrigFormatExpr->getSourceRange());
3013     return;
3014   }
3015 
3016   if (Type == FST_Printf || Type == FST_NSString) {
3017     CheckPrintfHandler H(*this, FExpr, OrigFormatExpr, firstDataArg,
3018                          numDataArgs, (Type == FST_NSString),
3019                          Str, HasVAListArg, Args, NumArgs, format_idx,
3020                          inFunctionCall, CallType);
3021 
3022     if (!analyze_format_string::ParsePrintfString(H, Str, Str + StrLen,
3023                                                   getLangOpts()))
3024       H.DoneProcessing();
3025   } else if (Type == FST_Scanf) {
3026     CheckScanfHandler H(*this, FExpr, OrigFormatExpr, firstDataArg, numDataArgs,
3027                         Str, HasVAListArg, Args, NumArgs, format_idx,
3028                         inFunctionCall, CallType);
3029 
3030     if (!analyze_format_string::ParseScanfString(H, Str, Str + StrLen,
3031                                                  getLangOpts()))
3032       H.DoneProcessing();
3033   } // TODO: handle other formats
3034 }
3035 
3036 //===--- CHECK: Standard memory functions ---------------------------------===//
3037 
3038 /// \brief Determine whether the given type is a dynamic class type (e.g.,
3039 /// whether it has a vtable).
isDynamicClassType(QualType T)3040 static bool isDynamicClassType(QualType T) {
3041   if (CXXRecordDecl *Record = T->getAsCXXRecordDecl())
3042     if (CXXRecordDecl *Definition = Record->getDefinition())
3043       if (Definition->isDynamicClass())
3044         return true;
3045 
3046   return false;
3047 }
3048 
3049 /// \brief If E is a sizeof expression, returns its argument expression,
3050 /// otherwise returns NULL.
getSizeOfExprArg(const Expr * E)3051 static const Expr *getSizeOfExprArg(const Expr* E) {
3052   if (const UnaryExprOrTypeTraitExpr *SizeOf =
3053       dyn_cast<UnaryExprOrTypeTraitExpr>(E))
3054     if (SizeOf->getKind() == clang::UETT_SizeOf && !SizeOf->isArgumentType())
3055       return SizeOf->getArgumentExpr()->IgnoreParenImpCasts();
3056 
3057   return 0;
3058 }
3059 
3060 /// \brief If E is a sizeof expression, returns its argument type.
getSizeOfArgType(const Expr * E)3061 static QualType getSizeOfArgType(const Expr* E) {
3062   if (const UnaryExprOrTypeTraitExpr *SizeOf =
3063       dyn_cast<UnaryExprOrTypeTraitExpr>(E))
3064     if (SizeOf->getKind() == clang::UETT_SizeOf)
3065       return SizeOf->getTypeOfArgument();
3066 
3067   return QualType();
3068 }
3069 
3070 /// \brief Check for dangerous or invalid arguments to memset().
3071 ///
3072 /// This issues warnings on known problematic, dangerous or unspecified
3073 /// arguments to the standard 'memset', 'memcpy', 'memmove', and 'memcmp'
3074 /// function calls.
3075 ///
3076 /// \param Call The call expression to diagnose.
CheckMemaccessArguments(const CallExpr * Call,unsigned BId,IdentifierInfo * FnName)3077 void Sema::CheckMemaccessArguments(const CallExpr *Call,
3078                                    unsigned BId,
3079                                    IdentifierInfo *FnName) {
3080   assert(BId != 0);
3081 
3082   // It is possible to have a non-standard definition of memset.  Validate
3083   // we have enough arguments, and if not, abort further checking.
3084   unsigned ExpectedNumArgs = (BId == Builtin::BIstrndup ? 2 : 3);
3085   if (Call->getNumArgs() < ExpectedNumArgs)
3086     return;
3087 
3088   unsigned LastArg = (BId == Builtin::BImemset ||
3089                       BId == Builtin::BIstrndup ? 1 : 2);
3090   unsigned LenArg = (BId == Builtin::BIstrndup ? 1 : 2);
3091   const Expr *LenExpr = Call->getArg(LenArg)->IgnoreParenImpCasts();
3092 
3093   // We have special checking when the length is a sizeof expression.
3094   QualType SizeOfArgTy = getSizeOfArgType(LenExpr);
3095   const Expr *SizeOfArg = getSizeOfExprArg(LenExpr);
3096   llvm::FoldingSetNodeID SizeOfArgID;
3097 
3098   for (unsigned ArgIdx = 0; ArgIdx != LastArg; ++ArgIdx) {
3099     const Expr *Dest = Call->getArg(ArgIdx)->IgnoreParenImpCasts();
3100     SourceRange ArgRange = Call->getArg(ArgIdx)->getSourceRange();
3101 
3102     QualType DestTy = Dest->getType();
3103     if (const PointerType *DestPtrTy = DestTy->getAs<PointerType>()) {
3104       QualType PointeeTy = DestPtrTy->getPointeeType();
3105 
3106       // Never warn about void type pointers. This can be used to suppress
3107       // false positives.
3108       if (PointeeTy->isVoidType())
3109         continue;
3110 
3111       // Catch "memset(p, 0, sizeof(p))" -- needs to be sizeof(*p). Do this by
3112       // actually comparing the expressions for equality. Because computing the
3113       // expression IDs can be expensive, we only do this if the diagnostic is
3114       // enabled.
3115       if (SizeOfArg &&
3116           Diags.getDiagnosticLevel(diag::warn_sizeof_pointer_expr_memaccess,
3117                                    SizeOfArg->getExprLoc())) {
3118         // We only compute IDs for expressions if the warning is enabled, and
3119         // cache the sizeof arg's ID.
3120         if (SizeOfArgID == llvm::FoldingSetNodeID())
3121           SizeOfArg->Profile(SizeOfArgID, Context, true);
3122         llvm::FoldingSetNodeID DestID;
3123         Dest->Profile(DestID, Context, true);
3124         if (DestID == SizeOfArgID) {
3125           // TODO: For strncpy() and friends, this could suggest sizeof(dst)
3126           //       over sizeof(src) as well.
3127           unsigned ActionIdx = 0; // Default is to suggest dereferencing.
3128           StringRef ReadableName = FnName->getName();
3129 
3130           if (const UnaryOperator *UnaryOp = dyn_cast<UnaryOperator>(Dest))
3131             if (UnaryOp->getOpcode() == UO_AddrOf)
3132               ActionIdx = 1; // If its an address-of operator, just remove it.
3133           if (Context.getTypeSize(PointeeTy) == Context.getCharWidth())
3134             ActionIdx = 2; // If the pointee's size is sizeof(char),
3135                            // suggest an explicit length.
3136 
3137           // If the function is defined as a builtin macro, do not show macro
3138           // expansion.
3139           SourceLocation SL = SizeOfArg->getExprLoc();
3140           SourceRange DSR = Dest->getSourceRange();
3141           SourceRange SSR = SizeOfArg->getSourceRange();
3142           SourceManager &SM  = PP.getSourceManager();
3143 
3144           if (SM.isMacroArgExpansion(SL)) {
3145             ReadableName = Lexer::getImmediateMacroName(SL, SM, LangOpts);
3146             SL = SM.getSpellingLoc(SL);
3147             DSR = SourceRange(SM.getSpellingLoc(DSR.getBegin()),
3148                              SM.getSpellingLoc(DSR.getEnd()));
3149             SSR = SourceRange(SM.getSpellingLoc(SSR.getBegin()),
3150                              SM.getSpellingLoc(SSR.getEnd()));
3151           }
3152 
3153           DiagRuntimeBehavior(SL, SizeOfArg,
3154                               PDiag(diag::warn_sizeof_pointer_expr_memaccess)
3155                                 << ReadableName
3156                                 << PointeeTy
3157                                 << DestTy
3158                                 << DSR
3159                                 << SSR);
3160           DiagRuntimeBehavior(SL, SizeOfArg,
3161                          PDiag(diag::warn_sizeof_pointer_expr_memaccess_note)
3162                                 << ActionIdx
3163                                 << SSR);
3164 
3165           break;
3166         }
3167       }
3168 
3169       // Also check for cases where the sizeof argument is the exact same
3170       // type as the memory argument, and where it points to a user-defined
3171       // record type.
3172       if (SizeOfArgTy != QualType()) {
3173         if (PointeeTy->isRecordType() &&
3174             Context.typesAreCompatible(SizeOfArgTy, DestTy)) {
3175           DiagRuntimeBehavior(LenExpr->getExprLoc(), Dest,
3176                               PDiag(diag::warn_sizeof_pointer_type_memaccess)
3177                                 << FnName << SizeOfArgTy << ArgIdx
3178                                 << PointeeTy << Dest->getSourceRange()
3179                                 << LenExpr->getSourceRange());
3180           break;
3181         }
3182       }
3183 
3184       // Always complain about dynamic classes.
3185       if (isDynamicClassType(PointeeTy)) {
3186 
3187         unsigned OperationType = 0;
3188         // "overwritten" if we're warning about the destination for any call
3189         // but memcmp; otherwise a verb appropriate to the call.
3190         if (ArgIdx != 0 || BId == Builtin::BImemcmp) {
3191           if (BId == Builtin::BImemcpy)
3192             OperationType = 1;
3193           else if(BId == Builtin::BImemmove)
3194             OperationType = 2;
3195           else if (BId == Builtin::BImemcmp)
3196             OperationType = 3;
3197         }
3198 
3199         DiagRuntimeBehavior(
3200           Dest->getExprLoc(), Dest,
3201           PDiag(diag::warn_dyn_class_memaccess)
3202             << (BId == Builtin::BImemcmp ? ArgIdx + 2 : ArgIdx)
3203             << FnName << PointeeTy
3204             << OperationType
3205             << Call->getCallee()->getSourceRange());
3206       } else if (PointeeTy.hasNonTrivialObjCLifetime() &&
3207                BId != Builtin::BImemset)
3208         DiagRuntimeBehavior(
3209           Dest->getExprLoc(), Dest,
3210           PDiag(diag::warn_arc_object_memaccess)
3211             << ArgIdx << FnName << PointeeTy
3212             << Call->getCallee()->getSourceRange());
3213       else
3214         continue;
3215 
3216       DiagRuntimeBehavior(
3217         Dest->getExprLoc(), Dest,
3218         PDiag(diag::note_bad_memaccess_silence)
3219           << FixItHint::CreateInsertion(ArgRange.getBegin(), "(void*)"));
3220       break;
3221     }
3222   }
3223 }
3224 
3225 // A little helper routine: ignore addition and subtraction of integer literals.
3226 // This intentionally does not ignore all integer constant expressions because
3227 // we don't want to remove sizeof().
ignoreLiteralAdditions(const Expr * Ex,ASTContext & Ctx)3228 static const Expr *ignoreLiteralAdditions(const Expr *Ex, ASTContext &Ctx) {
3229   Ex = Ex->IgnoreParenCasts();
3230 
3231   for (;;) {
3232     const BinaryOperator * BO = dyn_cast<BinaryOperator>(Ex);
3233     if (!BO || !BO->isAdditiveOp())
3234       break;
3235 
3236     const Expr *RHS = BO->getRHS()->IgnoreParenCasts();
3237     const Expr *LHS = BO->getLHS()->IgnoreParenCasts();
3238 
3239     if (isa<IntegerLiteral>(RHS))
3240       Ex = LHS;
3241     else if (isa<IntegerLiteral>(LHS))
3242       Ex = RHS;
3243     else
3244       break;
3245   }
3246 
3247   return Ex;
3248 }
3249 
isConstantSizeArrayWithMoreThanOneElement(QualType Ty,ASTContext & Context)3250 static bool isConstantSizeArrayWithMoreThanOneElement(QualType Ty,
3251                                                       ASTContext &Context) {
3252   // Only handle constant-sized or VLAs, but not flexible members.
3253   if (const ConstantArrayType *CAT = Context.getAsConstantArrayType(Ty)) {
3254     // Only issue the FIXIT for arrays of size > 1.
3255     if (CAT->getSize().getSExtValue() <= 1)
3256       return false;
3257   } else if (!Ty->isVariableArrayType()) {
3258     return false;
3259   }
3260   return true;
3261 }
3262 
3263 // Warn if the user has made the 'size' argument to strlcpy or strlcat
3264 // be the size of the source, instead of the destination.
CheckStrlcpycatArguments(const CallExpr * Call,IdentifierInfo * FnName)3265 void Sema::CheckStrlcpycatArguments(const CallExpr *Call,
3266                                     IdentifierInfo *FnName) {
3267 
3268   // Don't crash if the user has the wrong number of arguments
3269   if (Call->getNumArgs() != 3)
3270     return;
3271 
3272   const Expr *SrcArg = ignoreLiteralAdditions(Call->getArg(1), Context);
3273   const Expr *SizeArg = ignoreLiteralAdditions(Call->getArg(2), Context);
3274   const Expr *CompareWithSrc = NULL;
3275 
3276   // Look for 'strlcpy(dst, x, sizeof(x))'
3277   if (const Expr *Ex = getSizeOfExprArg(SizeArg))
3278     CompareWithSrc = Ex;
3279   else {
3280     // Look for 'strlcpy(dst, x, strlen(x))'
3281     if (const CallExpr *SizeCall = dyn_cast<CallExpr>(SizeArg)) {
3282       if (SizeCall->isBuiltinCall() == Builtin::BIstrlen
3283           && SizeCall->getNumArgs() == 1)
3284         CompareWithSrc = ignoreLiteralAdditions(SizeCall->getArg(0), Context);
3285     }
3286   }
3287 
3288   if (!CompareWithSrc)
3289     return;
3290 
3291   // Determine if the argument to sizeof/strlen is equal to the source
3292   // argument.  In principle there's all kinds of things you could do
3293   // here, for instance creating an == expression and evaluating it with
3294   // EvaluateAsBooleanCondition, but this uses a more direct technique:
3295   const DeclRefExpr *SrcArgDRE = dyn_cast<DeclRefExpr>(SrcArg);
3296   if (!SrcArgDRE)
3297     return;
3298 
3299   const DeclRefExpr *CompareWithSrcDRE = dyn_cast<DeclRefExpr>(CompareWithSrc);
3300   if (!CompareWithSrcDRE ||
3301       SrcArgDRE->getDecl() != CompareWithSrcDRE->getDecl())
3302     return;
3303 
3304   const Expr *OriginalSizeArg = Call->getArg(2);
3305   Diag(CompareWithSrcDRE->getLocStart(), diag::warn_strlcpycat_wrong_size)
3306     << OriginalSizeArg->getSourceRange() << FnName;
3307 
3308   // Output a FIXIT hint if the destination is an array (rather than a
3309   // pointer to an array).  This could be enhanced to handle some
3310   // pointers if we know the actual size, like if DstArg is 'array+2'
3311   // we could say 'sizeof(array)-2'.
3312   const Expr *DstArg = Call->getArg(0)->IgnoreParenImpCasts();
3313   if (!isConstantSizeArrayWithMoreThanOneElement(DstArg->getType(), Context))
3314     return;
3315 
3316   SmallString<128> sizeString;
3317   llvm::raw_svector_ostream OS(sizeString);
3318   OS << "sizeof(";
3319   DstArg->printPretty(OS, 0, getPrintingPolicy());
3320   OS << ")";
3321 
3322   Diag(OriginalSizeArg->getLocStart(), diag::note_strlcpycat_wrong_size)
3323     << FixItHint::CreateReplacement(OriginalSizeArg->getSourceRange(),
3324                                     OS.str());
3325 }
3326 
3327 /// Check if two expressions refer to the same declaration.
referToTheSameDecl(const Expr * E1,const Expr * E2)3328 static bool referToTheSameDecl(const Expr *E1, const Expr *E2) {
3329   if (const DeclRefExpr *D1 = dyn_cast_or_null<DeclRefExpr>(E1))
3330     if (const DeclRefExpr *D2 = dyn_cast_or_null<DeclRefExpr>(E2))
3331       return D1->getDecl() == D2->getDecl();
3332   return false;
3333 }
3334 
getStrlenExprArg(const Expr * E)3335 static const Expr *getStrlenExprArg(const Expr *E) {
3336   if (const CallExpr *CE = dyn_cast<CallExpr>(E)) {
3337     const FunctionDecl *FD = CE->getDirectCallee();
3338     if (!FD || FD->getMemoryFunctionKind() != Builtin::BIstrlen)
3339       return 0;
3340     return CE->getArg(0)->IgnoreParenCasts();
3341   }
3342   return 0;
3343 }
3344 
3345 // Warn on anti-patterns as the 'size' argument to strncat.
3346 // The correct size argument should look like following:
3347 //   strncat(dst, src, sizeof(dst) - strlen(dest) - 1);
CheckStrncatArguments(const CallExpr * CE,IdentifierInfo * FnName)3348 void Sema::CheckStrncatArguments(const CallExpr *CE,
3349                                  IdentifierInfo *FnName) {
3350   // Don't crash if the user has the wrong number of arguments.
3351   if (CE->getNumArgs() < 3)
3352     return;
3353   const Expr *DstArg = CE->getArg(0)->IgnoreParenCasts();
3354   const Expr *SrcArg = CE->getArg(1)->IgnoreParenCasts();
3355   const Expr *LenArg = CE->getArg(2)->IgnoreParenCasts();
3356 
3357   // Identify common expressions, which are wrongly used as the size argument
3358   // to strncat and may lead to buffer overflows.
3359   unsigned PatternType = 0;
3360   if (const Expr *SizeOfArg = getSizeOfExprArg(LenArg)) {
3361     // - sizeof(dst)
3362     if (referToTheSameDecl(SizeOfArg, DstArg))
3363       PatternType = 1;
3364     // - sizeof(src)
3365     else if (referToTheSameDecl(SizeOfArg, SrcArg))
3366       PatternType = 2;
3367   } else if (const BinaryOperator *BE = dyn_cast<BinaryOperator>(LenArg)) {
3368     if (BE->getOpcode() == BO_Sub) {
3369       const Expr *L = BE->getLHS()->IgnoreParenCasts();
3370       const Expr *R = BE->getRHS()->IgnoreParenCasts();
3371       // - sizeof(dst) - strlen(dst)
3372       if (referToTheSameDecl(DstArg, getSizeOfExprArg(L)) &&
3373           referToTheSameDecl(DstArg, getStrlenExprArg(R)))
3374         PatternType = 1;
3375       // - sizeof(src) - (anything)
3376       else if (referToTheSameDecl(SrcArg, getSizeOfExprArg(L)))
3377         PatternType = 2;
3378     }
3379   }
3380 
3381   if (PatternType == 0)
3382     return;
3383 
3384   // Generate the diagnostic.
3385   SourceLocation SL = LenArg->getLocStart();
3386   SourceRange SR = LenArg->getSourceRange();
3387   SourceManager &SM  = PP.getSourceManager();
3388 
3389   // If the function is defined as a builtin macro, do not show macro expansion.
3390   if (SM.isMacroArgExpansion(SL)) {
3391     SL = SM.getSpellingLoc(SL);
3392     SR = SourceRange(SM.getSpellingLoc(SR.getBegin()),
3393                      SM.getSpellingLoc(SR.getEnd()));
3394   }
3395 
3396   // Check if the destination is an array (rather than a pointer to an array).
3397   QualType DstTy = DstArg->getType();
3398   bool isKnownSizeArray = isConstantSizeArrayWithMoreThanOneElement(DstTy,
3399                                                                     Context);
3400   if (!isKnownSizeArray) {
3401     if (PatternType == 1)
3402       Diag(SL, diag::warn_strncat_wrong_size) << SR;
3403     else
3404       Diag(SL, diag::warn_strncat_src_size) << SR;
3405     return;
3406   }
3407 
3408   if (PatternType == 1)
3409     Diag(SL, diag::warn_strncat_large_size) << SR;
3410   else
3411     Diag(SL, diag::warn_strncat_src_size) << SR;
3412 
3413   SmallString<128> sizeString;
3414   llvm::raw_svector_ostream OS(sizeString);
3415   OS << "sizeof(";
3416   DstArg->printPretty(OS, 0, getPrintingPolicy());
3417   OS << ") - ";
3418   OS << "strlen(";
3419   DstArg->printPretty(OS, 0, getPrintingPolicy());
3420   OS << ") - 1";
3421 
3422   Diag(SL, diag::note_strncat_wrong_size)
3423     << FixItHint::CreateReplacement(SR, OS.str());
3424 }
3425 
3426 //===--- CHECK: Return Address of Stack Variable --------------------------===//
3427 
3428 static Expr *EvalVal(Expr *E, SmallVectorImpl<DeclRefExpr *> &refVars,
3429                      Decl *ParentDecl);
3430 static Expr *EvalAddr(Expr* E, SmallVectorImpl<DeclRefExpr *> &refVars,
3431                       Decl *ParentDecl);
3432 
3433 /// CheckReturnStackAddr - Check if a return statement returns the address
3434 ///   of a stack variable.
3435 void
CheckReturnStackAddr(Expr * RetValExp,QualType lhsType,SourceLocation ReturnLoc)3436 Sema::CheckReturnStackAddr(Expr *RetValExp, QualType lhsType,
3437                            SourceLocation ReturnLoc) {
3438 
3439   Expr *stackE = 0;
3440   SmallVector<DeclRefExpr *, 8> refVars;
3441 
3442   // Perform checking for returned stack addresses, local blocks,
3443   // label addresses or references to temporaries.
3444   if (lhsType->isPointerType() ||
3445       (!getLangOpts().ObjCAutoRefCount && lhsType->isBlockPointerType())) {
3446     stackE = EvalAddr(RetValExp, refVars, /*ParentDecl=*/0);
3447   } else if (lhsType->isReferenceType()) {
3448     stackE = EvalVal(RetValExp, refVars, /*ParentDecl=*/0);
3449   }
3450 
3451   if (stackE == 0)
3452     return; // Nothing suspicious was found.
3453 
3454   SourceLocation diagLoc;
3455   SourceRange diagRange;
3456   if (refVars.empty()) {
3457     diagLoc = stackE->getLocStart();
3458     diagRange = stackE->getSourceRange();
3459   } else {
3460     // We followed through a reference variable. 'stackE' contains the
3461     // problematic expression but we will warn at the return statement pointing
3462     // at the reference variable. We will later display the "trail" of
3463     // reference variables using notes.
3464     diagLoc = refVars[0]->getLocStart();
3465     diagRange = refVars[0]->getSourceRange();
3466   }
3467 
3468   if (DeclRefExpr *DR = dyn_cast<DeclRefExpr>(stackE)) { //address of local var.
3469     Diag(diagLoc, lhsType->isReferenceType() ? diag::warn_ret_stack_ref
3470                                              : diag::warn_ret_stack_addr)
3471      << DR->getDecl()->getDeclName() << diagRange;
3472   } else if (isa<BlockExpr>(stackE)) { // local block.
3473     Diag(diagLoc, diag::err_ret_local_block) << diagRange;
3474   } else if (isa<AddrLabelExpr>(stackE)) { // address of label.
3475     Diag(diagLoc, diag::warn_ret_addr_label) << diagRange;
3476   } else { // local temporary.
3477     Diag(diagLoc, lhsType->isReferenceType() ? diag::warn_ret_local_temp_ref
3478                                              : diag::warn_ret_local_temp_addr)
3479      << diagRange;
3480   }
3481 
3482   // Display the "trail" of reference variables that we followed until we
3483   // found the problematic expression using notes.
3484   for (unsigned i = 0, e = refVars.size(); i != e; ++i) {
3485     VarDecl *VD = cast<VarDecl>(refVars[i]->getDecl());
3486     // If this var binds to another reference var, show the range of the next
3487     // var, otherwise the var binds to the problematic expression, in which case
3488     // show the range of the expression.
3489     SourceRange range = (i < e-1) ? refVars[i+1]->getSourceRange()
3490                                   : stackE->getSourceRange();
3491     Diag(VD->getLocation(), diag::note_ref_var_local_bind)
3492       << VD->getDeclName() << range;
3493   }
3494 }
3495 
3496 /// EvalAddr - EvalAddr and EvalVal are mutually recursive functions that
3497 ///  check if the expression in a return statement evaluates to an address
3498 ///  to a location on the stack, a local block, an address of a label, or a
3499 ///  reference to local temporary. The recursion is used to traverse the
3500 ///  AST of the return expression, with recursion backtracking when we
3501 ///  encounter a subexpression that (1) clearly does not lead to one of the
3502 ///  above problematic expressions (2) is something we cannot determine leads to
3503 ///  a problematic expression based on such local checking.
3504 ///
3505 ///  Both EvalAddr and EvalVal follow through reference variables to evaluate
3506 ///  the expression that they point to. Such variables are added to the
3507 ///  'refVars' vector so that we know what the reference variable "trail" was.
3508 ///
3509 ///  EvalAddr processes expressions that are pointers that are used as
3510 ///  references (and not L-values).  EvalVal handles all other values.
3511 ///  At the base case of the recursion is a check for the above problematic
3512 ///  expressions.
3513 ///
3514 ///  This implementation handles:
3515 ///
3516 ///   * pointer-to-pointer casts
3517 ///   * implicit conversions from array references to pointers
3518 ///   * taking the address of fields
3519 ///   * arbitrary interplay between "&" and "*" operators
3520 ///   * pointer arithmetic from an address of a stack variable
3521 ///   * taking the address of an array element where the array is on the stack
EvalAddr(Expr * E,SmallVectorImpl<DeclRefExpr * > & refVars,Decl * ParentDecl)3522 static Expr *EvalAddr(Expr *E, SmallVectorImpl<DeclRefExpr *> &refVars,
3523                       Decl *ParentDecl) {
3524   if (E->isTypeDependent())
3525       return NULL;
3526 
3527   // We should only be called for evaluating pointer expressions.
3528   assert((E->getType()->isAnyPointerType() ||
3529           E->getType()->isBlockPointerType() ||
3530           E->getType()->isObjCQualifiedIdType()) &&
3531          "EvalAddr only works on pointers");
3532 
3533   E = E->IgnoreParens();
3534 
3535   // Our "symbolic interpreter" is just a dispatch off the currently
3536   // viewed AST node.  We then recursively traverse the AST by calling
3537   // EvalAddr and EvalVal appropriately.
3538   switch (E->getStmtClass()) {
3539   case Stmt::DeclRefExprClass: {
3540     DeclRefExpr *DR = cast<DeclRefExpr>(E);
3541 
3542     if (VarDecl *V = dyn_cast<VarDecl>(DR->getDecl()))
3543       // If this is a reference variable, follow through to the expression that
3544       // it points to.
3545       if (V->hasLocalStorage() &&
3546           V->getType()->isReferenceType() && V->hasInit()) {
3547         // Add the reference variable to the "trail".
3548         refVars.push_back(DR);
3549         return EvalAddr(V->getInit(), refVars, ParentDecl);
3550       }
3551 
3552     return NULL;
3553   }
3554 
3555   case Stmt::UnaryOperatorClass: {
3556     // The only unary operator that make sense to handle here
3557     // is AddrOf.  All others don't make sense as pointers.
3558     UnaryOperator *U = cast<UnaryOperator>(E);
3559 
3560     if (U->getOpcode() == UO_AddrOf)
3561       return EvalVal(U->getSubExpr(), refVars, ParentDecl);
3562     else
3563       return NULL;
3564   }
3565 
3566   case Stmt::BinaryOperatorClass: {
3567     // Handle pointer arithmetic.  All other binary operators are not valid
3568     // in this context.
3569     BinaryOperator *B = cast<BinaryOperator>(E);
3570     BinaryOperatorKind op = B->getOpcode();
3571 
3572     if (op != BO_Add && op != BO_Sub)
3573       return NULL;
3574 
3575     Expr *Base = B->getLHS();
3576 
3577     // Determine which argument is the real pointer base.  It could be
3578     // the RHS argument instead of the LHS.
3579     if (!Base->getType()->isPointerType()) Base = B->getRHS();
3580 
3581     assert (Base->getType()->isPointerType());
3582     return EvalAddr(Base, refVars, ParentDecl);
3583   }
3584 
3585   // For conditional operators we need to see if either the LHS or RHS are
3586   // valid DeclRefExpr*s.  If one of them is valid, we return it.
3587   case Stmt::ConditionalOperatorClass: {
3588     ConditionalOperator *C = cast<ConditionalOperator>(E);
3589 
3590     // Handle the GNU extension for missing LHS.
3591     if (Expr *lhsExpr = C->getLHS()) {
3592     // In C++, we can have a throw-expression, which has 'void' type.
3593       if (!lhsExpr->getType()->isVoidType())
3594         if (Expr* LHS = EvalAddr(lhsExpr, refVars, ParentDecl))
3595           return LHS;
3596     }
3597 
3598     // In C++, we can have a throw-expression, which has 'void' type.
3599     if (C->getRHS()->getType()->isVoidType())
3600       return NULL;
3601 
3602     return EvalAddr(C->getRHS(), refVars, ParentDecl);
3603   }
3604 
3605   case Stmt::BlockExprClass:
3606     if (cast<BlockExpr>(E)->getBlockDecl()->hasCaptures())
3607       return E; // local block.
3608     return NULL;
3609 
3610   case Stmt::AddrLabelExprClass:
3611     return E; // address of label.
3612 
3613   case Stmt::ExprWithCleanupsClass:
3614     return EvalAddr(cast<ExprWithCleanups>(E)->getSubExpr(), refVars,
3615                     ParentDecl);
3616 
3617   // For casts, we need to handle conversions from arrays to
3618   // pointer values, and pointer-to-pointer conversions.
3619   case Stmt::ImplicitCastExprClass:
3620   case Stmt::CStyleCastExprClass:
3621   case Stmt::CXXFunctionalCastExprClass:
3622   case Stmt::ObjCBridgedCastExprClass:
3623   case Stmt::CXXStaticCastExprClass:
3624   case Stmt::CXXDynamicCastExprClass:
3625   case Stmt::CXXConstCastExprClass:
3626   case Stmt::CXXReinterpretCastExprClass: {
3627     Expr* SubExpr = cast<CastExpr>(E)->getSubExpr();
3628     switch (cast<CastExpr>(E)->getCastKind()) {
3629     case CK_BitCast:
3630     case CK_LValueToRValue:
3631     case CK_NoOp:
3632     case CK_BaseToDerived:
3633     case CK_DerivedToBase:
3634     case CK_UncheckedDerivedToBase:
3635     case CK_Dynamic:
3636     case CK_CPointerToObjCPointerCast:
3637     case CK_BlockPointerToObjCPointerCast:
3638     case CK_AnyPointerToBlockPointerCast:
3639       return EvalAddr(SubExpr, refVars, ParentDecl);
3640 
3641     case CK_ArrayToPointerDecay:
3642       return EvalVal(SubExpr, refVars, ParentDecl);
3643 
3644     default:
3645       return 0;
3646     }
3647   }
3648 
3649   case Stmt::MaterializeTemporaryExprClass:
3650     if (Expr *Result = EvalAddr(
3651                          cast<MaterializeTemporaryExpr>(E)->GetTemporaryExpr(),
3652                                 refVars, ParentDecl))
3653       return Result;
3654 
3655     return E;
3656 
3657   // Everything else: we simply don't reason about them.
3658   default:
3659     return NULL;
3660   }
3661 }
3662 
3663 
3664 ///  EvalVal - This function is complements EvalAddr in the mutual recursion.
3665 ///   See the comments for EvalAddr for more details.
EvalVal(Expr * E,SmallVectorImpl<DeclRefExpr * > & refVars,Decl * ParentDecl)3666 static Expr *EvalVal(Expr *E, SmallVectorImpl<DeclRefExpr *> &refVars,
3667                      Decl *ParentDecl) {
3668 do {
3669   // We should only be called for evaluating non-pointer expressions, or
3670   // expressions with a pointer type that are not used as references but instead
3671   // are l-values (e.g., DeclRefExpr with a pointer type).
3672 
3673   // Our "symbolic interpreter" is just a dispatch off the currently
3674   // viewed AST node.  We then recursively traverse the AST by calling
3675   // EvalAddr and EvalVal appropriately.
3676 
3677   E = E->IgnoreParens();
3678   switch (E->getStmtClass()) {
3679   case Stmt::ImplicitCastExprClass: {
3680     ImplicitCastExpr *IE = cast<ImplicitCastExpr>(E);
3681     if (IE->getValueKind() == VK_LValue) {
3682       E = IE->getSubExpr();
3683       continue;
3684     }
3685     return NULL;
3686   }
3687 
3688   case Stmt::ExprWithCleanupsClass:
3689     return EvalVal(cast<ExprWithCleanups>(E)->getSubExpr(), refVars,ParentDecl);
3690 
3691   case Stmt::DeclRefExprClass: {
3692     // When we hit a DeclRefExpr we are looking at code that refers to a
3693     // variable's name. If it's not a reference variable we check if it has
3694     // local storage within the function, and if so, return the expression.
3695     DeclRefExpr *DR = cast<DeclRefExpr>(E);
3696 
3697     if (VarDecl *V = dyn_cast<VarDecl>(DR->getDecl())) {
3698       // Check if it refers to itself, e.g. "int& i = i;".
3699       if (V == ParentDecl)
3700         return DR;
3701 
3702       if (V->hasLocalStorage()) {
3703         if (!V->getType()->isReferenceType())
3704           return DR;
3705 
3706         // Reference variable, follow through to the expression that
3707         // it points to.
3708         if (V->hasInit()) {
3709           // Add the reference variable to the "trail".
3710           refVars.push_back(DR);
3711           return EvalVal(V->getInit(), refVars, V);
3712         }
3713       }
3714     }
3715 
3716     return NULL;
3717   }
3718 
3719   case Stmt::UnaryOperatorClass: {
3720     // The only unary operator that make sense to handle here
3721     // is Deref.  All others don't resolve to a "name."  This includes
3722     // handling all sorts of rvalues passed to a unary operator.
3723     UnaryOperator *U = cast<UnaryOperator>(E);
3724 
3725     if (U->getOpcode() == UO_Deref)
3726       return EvalAddr(U->getSubExpr(), refVars, ParentDecl);
3727 
3728     return NULL;
3729   }
3730 
3731   case Stmt::ArraySubscriptExprClass: {
3732     // Array subscripts are potential references to data on the stack.  We
3733     // retrieve the DeclRefExpr* for the array variable if it indeed
3734     // has local storage.
3735     return EvalAddr(cast<ArraySubscriptExpr>(E)->getBase(), refVars,ParentDecl);
3736   }
3737 
3738   case Stmt::ConditionalOperatorClass: {
3739     // For conditional operators we need to see if either the LHS or RHS are
3740     // non-NULL Expr's.  If one is non-NULL, we return it.
3741     ConditionalOperator *C = cast<ConditionalOperator>(E);
3742 
3743     // Handle the GNU extension for missing LHS.
3744     if (Expr *lhsExpr = C->getLHS())
3745       if (Expr *LHS = EvalVal(lhsExpr, refVars, ParentDecl))
3746         return LHS;
3747 
3748     return EvalVal(C->getRHS(), refVars, ParentDecl);
3749   }
3750 
3751   // Accesses to members are potential references to data on the stack.
3752   case Stmt::MemberExprClass: {
3753     MemberExpr *M = cast<MemberExpr>(E);
3754 
3755     // Check for indirect access.  We only want direct field accesses.
3756     if (M->isArrow())
3757       return NULL;
3758 
3759     // Check whether the member type is itself a reference, in which case
3760     // we're not going to refer to the member, but to what the member refers to.
3761     if (M->getMemberDecl()->getType()->isReferenceType())
3762       return NULL;
3763 
3764     return EvalVal(M->getBase(), refVars, ParentDecl);
3765   }
3766 
3767   case Stmt::MaterializeTemporaryExprClass:
3768     if (Expr *Result = EvalVal(
3769                           cast<MaterializeTemporaryExpr>(E)->GetTemporaryExpr(),
3770                                refVars, ParentDecl))
3771       return Result;
3772 
3773     return E;
3774 
3775   default:
3776     // Check that we don't return or take the address of a reference to a
3777     // temporary. This is only useful in C++.
3778     if (!E->isTypeDependent() && E->isRValue())
3779       return E;
3780 
3781     // Everything else: we simply don't reason about them.
3782     return NULL;
3783   }
3784 } while (true);
3785 }
3786 
3787 //===--- CHECK: Floating-Point comparisons (-Wfloat-equal) ---------------===//
3788 
3789 /// Check for comparisons of floating point operands using != and ==.
3790 /// Issue a warning if these are no self-comparisons, as they are not likely
3791 /// to do what the programmer intended.
CheckFloatComparison(SourceLocation Loc,Expr * LHS,Expr * RHS)3792 void Sema::CheckFloatComparison(SourceLocation Loc, Expr* LHS, Expr *RHS) {
3793   Expr* LeftExprSansParen = LHS->IgnoreParenImpCasts();
3794   Expr* RightExprSansParen = RHS->IgnoreParenImpCasts();
3795 
3796   // Special case: check for x == x (which is OK).
3797   // Do not emit warnings for such cases.
3798   if (DeclRefExpr* DRL = dyn_cast<DeclRefExpr>(LeftExprSansParen))
3799     if (DeclRefExpr* DRR = dyn_cast<DeclRefExpr>(RightExprSansParen))
3800       if (DRL->getDecl() == DRR->getDecl())
3801         return;
3802 
3803 
3804   // Special case: check for comparisons against literals that can be exactly
3805   //  represented by APFloat.  In such cases, do not emit a warning.  This
3806   //  is a heuristic: often comparison against such literals are used to
3807   //  detect if a value in a variable has not changed.  This clearly can
3808   //  lead to false negatives.
3809   if (FloatingLiteral* FLL = dyn_cast<FloatingLiteral>(LeftExprSansParen)) {
3810     if (FLL->isExact())
3811       return;
3812   } else
3813     if (FloatingLiteral* FLR = dyn_cast<FloatingLiteral>(RightExprSansParen))
3814       if (FLR->isExact())
3815         return;
3816 
3817   // Check for comparisons with builtin types.
3818   if (CallExpr* CL = dyn_cast<CallExpr>(LeftExprSansParen))
3819     if (CL->isBuiltinCall())
3820       return;
3821 
3822   if (CallExpr* CR = dyn_cast<CallExpr>(RightExprSansParen))
3823     if (CR->isBuiltinCall())
3824       return;
3825 
3826   // Emit the diagnostic.
3827   Diag(Loc, diag::warn_floatingpoint_eq)
3828     << LHS->getSourceRange() << RHS->getSourceRange();
3829 }
3830 
3831 //===--- CHECK: Integer mixed-sign comparisons (-Wsign-compare) --------===//
3832 //===--- CHECK: Lossy implicit conversions (-Wconversion) --------------===//
3833 
3834 namespace {
3835 
3836 /// Structure recording the 'active' range of an integer-valued
3837 /// expression.
3838 struct IntRange {
3839   /// The number of bits active in the int.
3840   unsigned Width;
3841 
3842   /// True if the int is known not to have negative values.
3843   bool NonNegative;
3844 
IntRange__anon32971bd90511::IntRange3845   IntRange(unsigned Width, bool NonNegative)
3846     : Width(Width), NonNegative(NonNegative)
3847   {}
3848 
3849   /// Returns the range of the bool type.
forBoolType__anon32971bd90511::IntRange3850   static IntRange forBoolType() {
3851     return IntRange(1, true);
3852   }
3853 
3854   /// Returns the range of an opaque value of the given integral type.
forValueOfType__anon32971bd90511::IntRange3855   static IntRange forValueOfType(ASTContext &C, QualType T) {
3856     return forValueOfCanonicalType(C,
3857                           T->getCanonicalTypeInternal().getTypePtr());
3858   }
3859 
3860   /// Returns the range of an opaque value of a canonical integral type.
forValueOfCanonicalType__anon32971bd90511::IntRange3861   static IntRange forValueOfCanonicalType(ASTContext &C, const Type *T) {
3862     assert(T->isCanonicalUnqualified());
3863 
3864     if (const VectorType *VT = dyn_cast<VectorType>(T))
3865       T = VT->getElementType().getTypePtr();
3866     if (const ComplexType *CT = dyn_cast<ComplexType>(T))
3867       T = CT->getElementType().getTypePtr();
3868 
3869     // For enum types, use the known bit width of the enumerators.
3870     if (const EnumType *ET = dyn_cast<EnumType>(T)) {
3871       EnumDecl *Enum = ET->getDecl();
3872       if (!Enum->isCompleteDefinition())
3873         return IntRange(C.getIntWidth(QualType(T, 0)), false);
3874 
3875       unsigned NumPositive = Enum->getNumPositiveBits();
3876       unsigned NumNegative = Enum->getNumNegativeBits();
3877 
3878       return IntRange(std::max(NumPositive, NumNegative), NumNegative == 0);
3879     }
3880 
3881     const BuiltinType *BT = cast<BuiltinType>(T);
3882     assert(BT->isInteger());
3883 
3884     return IntRange(C.getIntWidth(QualType(T, 0)), BT->isUnsignedInteger());
3885   }
3886 
3887   /// Returns the "target" range of a canonical integral type, i.e.
3888   /// the range of values expressible in the type.
3889   ///
3890   /// This matches forValueOfCanonicalType except that enums have the
3891   /// full range of their type, not the range of their enumerators.
forTargetOfCanonicalType__anon32971bd90511::IntRange3892   static IntRange forTargetOfCanonicalType(ASTContext &C, const Type *T) {
3893     assert(T->isCanonicalUnqualified());
3894 
3895     if (const VectorType *VT = dyn_cast<VectorType>(T))
3896       T = VT->getElementType().getTypePtr();
3897     if (const ComplexType *CT = dyn_cast<ComplexType>(T))
3898       T = CT->getElementType().getTypePtr();
3899     if (const EnumType *ET = dyn_cast<EnumType>(T))
3900       T = C.getCanonicalType(ET->getDecl()->getIntegerType()).getTypePtr();
3901 
3902     const BuiltinType *BT = cast<BuiltinType>(T);
3903     assert(BT->isInteger());
3904 
3905     return IntRange(C.getIntWidth(QualType(T, 0)), BT->isUnsignedInteger());
3906   }
3907 
3908   /// Returns the supremum of two ranges: i.e. their conservative merge.
join__anon32971bd90511::IntRange3909   static IntRange join(IntRange L, IntRange R) {
3910     return IntRange(std::max(L.Width, R.Width),
3911                     L.NonNegative && R.NonNegative);
3912   }
3913 
3914   /// Returns the infinum of two ranges: i.e. their aggressive merge.
meet__anon32971bd90511::IntRange3915   static IntRange meet(IntRange L, IntRange R) {
3916     return IntRange(std::min(L.Width, R.Width),
3917                     L.NonNegative || R.NonNegative);
3918   }
3919 };
3920 
GetValueRange(ASTContext & C,llvm::APSInt & value,unsigned MaxWidth)3921 static IntRange GetValueRange(ASTContext &C, llvm::APSInt &value,
3922                               unsigned MaxWidth) {
3923   if (value.isSigned() && value.isNegative())
3924     return IntRange(value.getMinSignedBits(), false);
3925 
3926   if (value.getBitWidth() > MaxWidth)
3927     value = value.trunc(MaxWidth);
3928 
3929   // isNonNegative() just checks the sign bit without considering
3930   // signedness.
3931   return IntRange(value.getActiveBits(), true);
3932 }
3933 
GetValueRange(ASTContext & C,APValue & result,QualType Ty,unsigned MaxWidth)3934 static IntRange GetValueRange(ASTContext &C, APValue &result, QualType Ty,
3935                               unsigned MaxWidth) {
3936   if (result.isInt())
3937     return GetValueRange(C, result.getInt(), MaxWidth);
3938 
3939   if (result.isVector()) {
3940     IntRange R = GetValueRange(C, result.getVectorElt(0), Ty, MaxWidth);
3941     for (unsigned i = 1, e = result.getVectorLength(); i != e; ++i) {
3942       IntRange El = GetValueRange(C, result.getVectorElt(i), Ty, MaxWidth);
3943       R = IntRange::join(R, El);
3944     }
3945     return R;
3946   }
3947 
3948   if (result.isComplexInt()) {
3949     IntRange R = GetValueRange(C, result.getComplexIntReal(), MaxWidth);
3950     IntRange I = GetValueRange(C, result.getComplexIntImag(), MaxWidth);
3951     return IntRange::join(R, I);
3952   }
3953 
3954   // This can happen with lossless casts to intptr_t of "based" lvalues.
3955   // Assume it might use arbitrary bits.
3956   // FIXME: The only reason we need to pass the type in here is to get
3957   // the sign right on this one case.  It would be nice if APValue
3958   // preserved this.
3959   assert(result.isLValue() || result.isAddrLabelDiff());
3960   return IntRange(MaxWidth, Ty->isUnsignedIntegerOrEnumerationType());
3961 }
3962 
3963 /// Pseudo-evaluate the given integer expression, estimating the
3964 /// range of values it might take.
3965 ///
3966 /// \param MaxWidth - the width to which the value will be truncated
GetExprRange(ASTContext & C,Expr * E,unsigned MaxWidth)3967 static IntRange GetExprRange(ASTContext &C, Expr *E, unsigned MaxWidth) {
3968   E = E->IgnoreParens();
3969 
3970   // Try a full evaluation first.
3971   Expr::EvalResult result;
3972   if (E->EvaluateAsRValue(result, C))
3973     return GetValueRange(C, result.Val, E->getType(), MaxWidth);
3974 
3975   // I think we only want to look through implicit casts here; if the
3976   // user has an explicit widening cast, we should treat the value as
3977   // being of the new, wider type.
3978   if (ImplicitCastExpr *CE = dyn_cast<ImplicitCastExpr>(E)) {
3979     if (CE->getCastKind() == CK_NoOp || CE->getCastKind() == CK_LValueToRValue)
3980       return GetExprRange(C, CE->getSubExpr(), MaxWidth);
3981 
3982     IntRange OutputTypeRange = IntRange::forValueOfType(C, CE->getType());
3983 
3984     bool isIntegerCast = (CE->getCastKind() == CK_IntegralCast);
3985 
3986     // Assume that non-integer casts can span the full range of the type.
3987     if (!isIntegerCast)
3988       return OutputTypeRange;
3989 
3990     IntRange SubRange
3991       = GetExprRange(C, CE->getSubExpr(),
3992                      std::min(MaxWidth, OutputTypeRange.Width));
3993 
3994     // Bail out if the subexpr's range is as wide as the cast type.
3995     if (SubRange.Width >= OutputTypeRange.Width)
3996       return OutputTypeRange;
3997 
3998     // Otherwise, we take the smaller width, and we're non-negative if
3999     // either the output type or the subexpr is.
4000     return IntRange(SubRange.Width,
4001                     SubRange.NonNegative || OutputTypeRange.NonNegative);
4002   }
4003 
4004   if (ConditionalOperator *CO = dyn_cast<ConditionalOperator>(E)) {
4005     // If we can fold the condition, just take that operand.
4006     bool CondResult;
4007     if (CO->getCond()->EvaluateAsBooleanCondition(CondResult, C))
4008       return GetExprRange(C, CondResult ? CO->getTrueExpr()
4009                                         : CO->getFalseExpr(),
4010                           MaxWidth);
4011 
4012     // Otherwise, conservatively merge.
4013     IntRange L = GetExprRange(C, CO->getTrueExpr(), MaxWidth);
4014     IntRange R = GetExprRange(C, CO->getFalseExpr(), MaxWidth);
4015     return IntRange::join(L, R);
4016   }
4017 
4018   if (BinaryOperator *BO = dyn_cast<BinaryOperator>(E)) {
4019     switch (BO->getOpcode()) {
4020 
4021     // Boolean-valued operations are single-bit and positive.
4022     case BO_LAnd:
4023     case BO_LOr:
4024     case BO_LT:
4025     case BO_GT:
4026     case BO_LE:
4027     case BO_GE:
4028     case BO_EQ:
4029     case BO_NE:
4030       return IntRange::forBoolType();
4031 
4032     // The type of the assignments is the type of the LHS, so the RHS
4033     // is not necessarily the same type.
4034     case BO_MulAssign:
4035     case BO_DivAssign:
4036     case BO_RemAssign:
4037     case BO_AddAssign:
4038     case BO_SubAssign:
4039     case BO_XorAssign:
4040     case BO_OrAssign:
4041       // TODO: bitfields?
4042       return IntRange::forValueOfType(C, E->getType());
4043 
4044     // Simple assignments just pass through the RHS, which will have
4045     // been coerced to the LHS type.
4046     case BO_Assign:
4047       // TODO: bitfields?
4048       return GetExprRange(C, BO->getRHS(), MaxWidth);
4049 
4050     // Operations with opaque sources are black-listed.
4051     case BO_PtrMemD:
4052     case BO_PtrMemI:
4053       return IntRange::forValueOfType(C, E->getType());
4054 
4055     // Bitwise-and uses the *infinum* of the two source ranges.
4056     case BO_And:
4057     case BO_AndAssign:
4058       return IntRange::meet(GetExprRange(C, BO->getLHS(), MaxWidth),
4059                             GetExprRange(C, BO->getRHS(), MaxWidth));
4060 
4061     // Left shift gets black-listed based on a judgement call.
4062     case BO_Shl:
4063       // ...except that we want to treat '1 << (blah)' as logically
4064       // positive.  It's an important idiom.
4065       if (IntegerLiteral *I
4066             = dyn_cast<IntegerLiteral>(BO->getLHS()->IgnoreParenCasts())) {
4067         if (I->getValue() == 1) {
4068           IntRange R = IntRange::forValueOfType(C, E->getType());
4069           return IntRange(R.Width, /*NonNegative*/ true);
4070         }
4071       }
4072       // fallthrough
4073 
4074     case BO_ShlAssign:
4075       return IntRange::forValueOfType(C, E->getType());
4076 
4077     // Right shift by a constant can narrow its left argument.
4078     case BO_Shr:
4079     case BO_ShrAssign: {
4080       IntRange L = GetExprRange(C, BO->getLHS(), MaxWidth);
4081 
4082       // If the shift amount is a positive constant, drop the width by
4083       // that much.
4084       llvm::APSInt shift;
4085       if (BO->getRHS()->isIntegerConstantExpr(shift, C) &&
4086           shift.isNonNegative()) {
4087         unsigned zext = shift.getZExtValue();
4088         if (zext >= L.Width)
4089           L.Width = (L.NonNegative ? 0 : 1);
4090         else
4091           L.Width -= zext;
4092       }
4093 
4094       return L;
4095     }
4096 
4097     // Comma acts as its right operand.
4098     case BO_Comma:
4099       return GetExprRange(C, BO->getRHS(), MaxWidth);
4100 
4101     // Black-list pointer subtractions.
4102     case BO_Sub:
4103       if (BO->getLHS()->getType()->isPointerType())
4104         return IntRange::forValueOfType(C, E->getType());
4105       break;
4106 
4107     // The width of a division result is mostly determined by the size
4108     // of the LHS.
4109     case BO_Div: {
4110       // Don't 'pre-truncate' the operands.
4111       unsigned opWidth = C.getIntWidth(E->getType());
4112       IntRange L = GetExprRange(C, BO->getLHS(), opWidth);
4113 
4114       // If the divisor is constant, use that.
4115       llvm::APSInt divisor;
4116       if (BO->getRHS()->isIntegerConstantExpr(divisor, C)) {
4117         unsigned log2 = divisor.logBase2(); // floor(log_2(divisor))
4118         if (log2 >= L.Width)
4119           L.Width = (L.NonNegative ? 0 : 1);
4120         else
4121           L.Width = std::min(L.Width - log2, MaxWidth);
4122         return L;
4123       }
4124 
4125       // Otherwise, just use the LHS's width.
4126       IntRange R = GetExprRange(C, BO->getRHS(), opWidth);
4127       return IntRange(L.Width, L.NonNegative && R.NonNegative);
4128     }
4129 
4130     // The result of a remainder can't be larger than the result of
4131     // either side.
4132     case BO_Rem: {
4133       // Don't 'pre-truncate' the operands.
4134       unsigned opWidth = C.getIntWidth(E->getType());
4135       IntRange L = GetExprRange(C, BO->getLHS(), opWidth);
4136       IntRange R = GetExprRange(C, BO->getRHS(), opWidth);
4137 
4138       IntRange meet = IntRange::meet(L, R);
4139       meet.Width = std::min(meet.Width, MaxWidth);
4140       return meet;
4141     }
4142 
4143     // The default behavior is okay for these.
4144     case BO_Mul:
4145     case BO_Add:
4146     case BO_Xor:
4147     case BO_Or:
4148       break;
4149     }
4150 
4151     // The default case is to treat the operation as if it were closed
4152     // on the narrowest type that encompasses both operands.
4153     IntRange L = GetExprRange(C, BO->getLHS(), MaxWidth);
4154     IntRange R = GetExprRange(C, BO->getRHS(), MaxWidth);
4155     return IntRange::join(L, R);
4156   }
4157 
4158   if (UnaryOperator *UO = dyn_cast<UnaryOperator>(E)) {
4159     switch (UO->getOpcode()) {
4160     // Boolean-valued operations are white-listed.
4161     case UO_LNot:
4162       return IntRange::forBoolType();
4163 
4164     // Operations with opaque sources are black-listed.
4165     case UO_Deref:
4166     case UO_AddrOf: // should be impossible
4167       return IntRange::forValueOfType(C, E->getType());
4168 
4169     default:
4170       return GetExprRange(C, UO->getSubExpr(), MaxWidth);
4171     }
4172   }
4173 
4174   if (dyn_cast<OffsetOfExpr>(E)) {
4175     IntRange::forValueOfType(C, E->getType());
4176   }
4177 
4178   if (FieldDecl *BitField = E->getBitField())
4179     return IntRange(BitField->getBitWidthValue(C),
4180                     BitField->getType()->isUnsignedIntegerOrEnumerationType());
4181 
4182   return IntRange::forValueOfType(C, E->getType());
4183 }
4184 
GetExprRange(ASTContext & C,Expr * E)4185 static IntRange GetExprRange(ASTContext &C, Expr *E) {
4186   return GetExprRange(C, E, C.getIntWidth(E->getType()));
4187 }
4188 
4189 /// Checks whether the given value, which currently has the given
4190 /// source semantics, has the same value when coerced through the
4191 /// target semantics.
IsSameFloatAfterCast(const llvm::APFloat & value,const llvm::fltSemantics & Src,const llvm::fltSemantics & Tgt)4192 static bool IsSameFloatAfterCast(const llvm::APFloat &value,
4193                                  const llvm::fltSemantics &Src,
4194                                  const llvm::fltSemantics &Tgt) {
4195   llvm::APFloat truncated = value;
4196 
4197   bool ignored;
4198   truncated.convert(Src, llvm::APFloat::rmNearestTiesToEven, &ignored);
4199   truncated.convert(Tgt, llvm::APFloat::rmNearestTiesToEven, &ignored);
4200 
4201   return truncated.bitwiseIsEqual(value);
4202 }
4203 
4204 /// Checks whether the given value, which currently has the given
4205 /// source semantics, has the same value when coerced through the
4206 /// target semantics.
4207 ///
4208 /// The value might be a vector of floats (or a complex number).
IsSameFloatAfterCast(const APValue & value,const llvm::fltSemantics & Src,const llvm::fltSemantics & Tgt)4209 static bool IsSameFloatAfterCast(const APValue &value,
4210                                  const llvm::fltSemantics &Src,
4211                                  const llvm::fltSemantics &Tgt) {
4212   if (value.isFloat())
4213     return IsSameFloatAfterCast(value.getFloat(), Src, Tgt);
4214 
4215   if (value.isVector()) {
4216     for (unsigned i = 0, e = value.getVectorLength(); i != e; ++i)
4217       if (!IsSameFloatAfterCast(value.getVectorElt(i), Src, Tgt))
4218         return false;
4219     return true;
4220   }
4221 
4222   assert(value.isComplexFloat());
4223   return (IsSameFloatAfterCast(value.getComplexFloatReal(), Src, Tgt) &&
4224           IsSameFloatAfterCast(value.getComplexFloatImag(), Src, Tgt));
4225 }
4226 
4227 static void AnalyzeImplicitConversions(Sema &S, Expr *E, SourceLocation CC);
4228 
IsZero(Sema & S,Expr * E)4229 static bool IsZero(Sema &S, Expr *E) {
4230   // Suppress cases where we are comparing against an enum constant.
4231   if (const DeclRefExpr *DR =
4232       dyn_cast<DeclRefExpr>(E->IgnoreParenImpCasts()))
4233     if (isa<EnumConstantDecl>(DR->getDecl()))
4234       return false;
4235 
4236   // Suppress cases where the '0' value is expanded from a macro.
4237   if (E->getLocStart().isMacroID())
4238     return false;
4239 
4240   llvm::APSInt Value;
4241   return E->isIntegerConstantExpr(Value, S.Context) && Value == 0;
4242 }
4243 
HasEnumType(Expr * E)4244 static bool HasEnumType(Expr *E) {
4245   // Strip off implicit integral promotions.
4246   while (ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(E)) {
4247     if (ICE->getCastKind() != CK_IntegralCast &&
4248         ICE->getCastKind() != CK_NoOp)
4249       break;
4250     E = ICE->getSubExpr();
4251   }
4252 
4253   return E->getType()->isEnumeralType();
4254 }
4255 
CheckTrivialUnsignedComparison(Sema & S,BinaryOperator * E)4256 static void CheckTrivialUnsignedComparison(Sema &S, BinaryOperator *E) {
4257   BinaryOperatorKind op = E->getOpcode();
4258   if (E->isValueDependent())
4259     return;
4260 
4261   if (op == BO_LT && IsZero(S, E->getRHS())) {
4262     S.Diag(E->getOperatorLoc(), diag::warn_lunsigned_always_true_comparison)
4263       << "< 0" << "false" << HasEnumType(E->getLHS())
4264       << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange();
4265   } else if (op == BO_GE && IsZero(S, E->getRHS())) {
4266     S.Diag(E->getOperatorLoc(), diag::warn_lunsigned_always_true_comparison)
4267       << ">= 0" << "true" << HasEnumType(E->getLHS())
4268       << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange();
4269   } else if (op == BO_GT && IsZero(S, E->getLHS())) {
4270     S.Diag(E->getOperatorLoc(), diag::warn_runsigned_always_true_comparison)
4271       << "0 >" << "false" << HasEnumType(E->getRHS())
4272       << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange();
4273   } else if (op == BO_LE && IsZero(S, E->getLHS())) {
4274     S.Diag(E->getOperatorLoc(), diag::warn_runsigned_always_true_comparison)
4275       << "0 <=" << "true" << HasEnumType(E->getRHS())
4276       << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange();
4277   }
4278 }
4279 
4280 /// Analyze the operands of the given comparison.  Implements the
4281 /// fallback case from AnalyzeComparison.
AnalyzeImpConvsInComparison(Sema & S,BinaryOperator * E)4282 static void AnalyzeImpConvsInComparison(Sema &S, BinaryOperator *E) {
4283   AnalyzeImplicitConversions(S, E->getLHS(), E->getOperatorLoc());
4284   AnalyzeImplicitConversions(S, E->getRHS(), E->getOperatorLoc());
4285 }
4286 
4287 /// \brief Implements -Wsign-compare.
4288 ///
4289 /// \param E the binary operator to check for warnings
AnalyzeComparison(Sema & S,BinaryOperator * E)4290 static void AnalyzeComparison(Sema &S, BinaryOperator *E) {
4291   // The type the comparison is being performed in.
4292   QualType T = E->getLHS()->getType();
4293   assert(S.Context.hasSameUnqualifiedType(T, E->getRHS()->getType())
4294          && "comparison with mismatched types");
4295 
4296   // We don't do anything special if this isn't an unsigned integral
4297   // comparison:  we're only interested in integral comparisons, and
4298   // signed comparisons only happen in cases we don't care to warn about.
4299   //
4300   // We also don't care about value-dependent expressions or expressions
4301   // whose result is a constant.
4302   if (!T->hasUnsignedIntegerRepresentation()
4303       || E->isValueDependent() || E->isIntegerConstantExpr(S.Context))
4304     return AnalyzeImpConvsInComparison(S, E);
4305 
4306   Expr *LHS = E->getLHS()->IgnoreParenImpCasts();
4307   Expr *RHS = E->getRHS()->IgnoreParenImpCasts();
4308 
4309   // Check to see if one of the (unmodified) operands is of different
4310   // signedness.
4311   Expr *signedOperand, *unsignedOperand;
4312   if (LHS->getType()->hasSignedIntegerRepresentation()) {
4313     assert(!RHS->getType()->hasSignedIntegerRepresentation() &&
4314            "unsigned comparison between two signed integer expressions?");
4315     signedOperand = LHS;
4316     unsignedOperand = RHS;
4317   } else if (RHS->getType()->hasSignedIntegerRepresentation()) {
4318     signedOperand = RHS;
4319     unsignedOperand = LHS;
4320   } else {
4321     CheckTrivialUnsignedComparison(S, E);
4322     return AnalyzeImpConvsInComparison(S, E);
4323   }
4324 
4325   // Otherwise, calculate the effective range of the signed operand.
4326   IntRange signedRange = GetExprRange(S.Context, signedOperand);
4327 
4328   // Go ahead and analyze implicit conversions in the operands.  Note
4329   // that we skip the implicit conversions on both sides.
4330   AnalyzeImplicitConversions(S, LHS, E->getOperatorLoc());
4331   AnalyzeImplicitConversions(S, RHS, E->getOperatorLoc());
4332 
4333   // If the signed range is non-negative, -Wsign-compare won't fire,
4334   // but we should still check for comparisons which are always true
4335   // or false.
4336   if (signedRange.NonNegative)
4337     return CheckTrivialUnsignedComparison(S, E);
4338 
4339   // For (in)equality comparisons, if the unsigned operand is a
4340   // constant which cannot collide with a overflowed signed operand,
4341   // then reinterpreting the signed operand as unsigned will not
4342   // change the result of the comparison.
4343   if (E->isEqualityOp()) {
4344     unsigned comparisonWidth = S.Context.getIntWidth(T);
4345     IntRange unsignedRange = GetExprRange(S.Context, unsignedOperand);
4346 
4347     // We should never be unable to prove that the unsigned operand is
4348     // non-negative.
4349     assert(unsignedRange.NonNegative && "unsigned range includes negative?");
4350 
4351     if (unsignedRange.Width < comparisonWidth)
4352       return;
4353   }
4354 
4355   S.DiagRuntimeBehavior(E->getOperatorLoc(), E,
4356     S.PDiag(diag::warn_mixed_sign_comparison)
4357       << LHS->getType() << RHS->getType()
4358       << LHS->getSourceRange() << RHS->getSourceRange());
4359 }
4360 
4361 /// Analyzes an attempt to assign the given value to a bitfield.
4362 ///
4363 /// Returns true if there was something fishy about the attempt.
AnalyzeBitFieldAssignment(Sema & S,FieldDecl * Bitfield,Expr * Init,SourceLocation InitLoc)4364 static bool AnalyzeBitFieldAssignment(Sema &S, FieldDecl *Bitfield, Expr *Init,
4365                                       SourceLocation InitLoc) {
4366   assert(Bitfield->isBitField());
4367   if (Bitfield->isInvalidDecl())
4368     return false;
4369 
4370   // White-list bool bitfields.
4371   if (Bitfield->getType()->isBooleanType())
4372     return false;
4373 
4374   // Ignore value- or type-dependent expressions.
4375   if (Bitfield->getBitWidth()->isValueDependent() ||
4376       Bitfield->getBitWidth()->isTypeDependent() ||
4377       Init->isValueDependent() ||
4378       Init->isTypeDependent())
4379     return false;
4380 
4381   Expr *OriginalInit = Init->IgnoreParenImpCasts();
4382 
4383   llvm::APSInt Value;
4384   if (!OriginalInit->EvaluateAsInt(Value, S.Context, Expr::SE_AllowSideEffects))
4385     return false;
4386 
4387   unsigned OriginalWidth = Value.getBitWidth();
4388   unsigned FieldWidth = Bitfield->getBitWidthValue(S.Context);
4389 
4390   if (OriginalWidth <= FieldWidth)
4391     return false;
4392 
4393   // Compute the value which the bitfield will contain.
4394   llvm::APSInt TruncatedValue = Value.trunc(FieldWidth);
4395   TruncatedValue.setIsSigned(Bitfield->getType()->isSignedIntegerType());
4396 
4397   // Check whether the stored value is equal to the original value.
4398   TruncatedValue = TruncatedValue.extend(OriginalWidth);
4399   if (llvm::APSInt::isSameValue(Value, TruncatedValue))
4400     return false;
4401 
4402   // Special-case bitfields of width 1: booleans are naturally 0/1, and
4403   // therefore don't strictly fit into a signed bitfield of width 1.
4404   if (FieldWidth == 1 && Value == 1)
4405     return false;
4406 
4407   std::string PrettyValue = Value.toString(10);
4408   std::string PrettyTrunc = TruncatedValue.toString(10);
4409 
4410   S.Diag(InitLoc, diag::warn_impcast_bitfield_precision_constant)
4411     << PrettyValue << PrettyTrunc << OriginalInit->getType()
4412     << Init->getSourceRange();
4413 
4414   return true;
4415 }
4416 
4417 /// Analyze the given simple or compound assignment for warning-worthy
4418 /// operations.
AnalyzeAssignment(Sema & S,BinaryOperator * E)4419 static void AnalyzeAssignment(Sema &S, BinaryOperator *E) {
4420   // Just recurse on the LHS.
4421   AnalyzeImplicitConversions(S, E->getLHS(), E->getOperatorLoc());
4422 
4423   // We want to recurse on the RHS as normal unless we're assigning to
4424   // a bitfield.
4425   if (FieldDecl *Bitfield = E->getLHS()->getBitField()) {
4426     if (AnalyzeBitFieldAssignment(S, Bitfield, E->getRHS(),
4427                                   E->getOperatorLoc())) {
4428       // Recurse, ignoring any implicit conversions on the RHS.
4429       return AnalyzeImplicitConversions(S, E->getRHS()->IgnoreParenImpCasts(),
4430                                         E->getOperatorLoc());
4431     }
4432   }
4433 
4434   AnalyzeImplicitConversions(S, E->getRHS(), E->getOperatorLoc());
4435 }
4436 
4437 /// 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)4438 static void DiagnoseImpCast(Sema &S, Expr *E, QualType SourceType, QualType T,
4439                             SourceLocation CContext, unsigned diag,
4440                             bool pruneControlFlow = false) {
4441   if (pruneControlFlow) {
4442     S.DiagRuntimeBehavior(E->getExprLoc(), E,
4443                           S.PDiag(diag)
4444                             << SourceType << T << E->getSourceRange()
4445                             << SourceRange(CContext));
4446     return;
4447   }
4448   S.Diag(E->getExprLoc(), diag)
4449     << SourceType << T << E->getSourceRange() << SourceRange(CContext);
4450 }
4451 
4452 /// Diagnose an implicit cast;  purely a helper for CheckImplicitConversion.
DiagnoseImpCast(Sema & S,Expr * E,QualType T,SourceLocation CContext,unsigned diag,bool pruneControlFlow=false)4453 static void DiagnoseImpCast(Sema &S, Expr *E, QualType T,
4454                             SourceLocation CContext, unsigned diag,
4455                             bool pruneControlFlow = false) {
4456   DiagnoseImpCast(S, E, E->getType(), T, CContext, diag, pruneControlFlow);
4457 }
4458 
4459 /// Diagnose an implicit cast from a literal expression. Does not warn when the
4460 /// cast wouldn't lose information.
DiagnoseFloatingLiteralImpCast(Sema & S,FloatingLiteral * FL,QualType T,SourceLocation CContext)4461 void DiagnoseFloatingLiteralImpCast(Sema &S, FloatingLiteral *FL, QualType T,
4462                                     SourceLocation CContext) {
4463   // Try to convert the literal exactly to an integer. If we can, don't warn.
4464   bool isExact = false;
4465   const llvm::APFloat &Value = FL->getValue();
4466   llvm::APSInt IntegerValue(S.Context.getIntWidth(T),
4467                             T->hasUnsignedIntegerRepresentation());
4468   if (Value.convertToInteger(IntegerValue,
4469                              llvm::APFloat::rmTowardZero, &isExact)
4470       == llvm::APFloat::opOK && isExact)
4471     return;
4472 
4473   SmallString<16> PrettySourceValue;
4474   Value.toString(PrettySourceValue);
4475   SmallString<16> PrettyTargetValue;
4476   if (T->isSpecificBuiltinType(BuiltinType::Bool))
4477     PrettyTargetValue = IntegerValue == 0 ? "false" : "true";
4478   else
4479     IntegerValue.toString(PrettyTargetValue);
4480 
4481   S.Diag(FL->getExprLoc(), diag::warn_impcast_literal_float_to_integer)
4482     << FL->getType() << T.getUnqualifiedType() << PrettySourceValue
4483     << PrettyTargetValue << FL->getSourceRange() << SourceRange(CContext);
4484 }
4485 
PrettyPrintInRange(const llvm::APSInt & Value,IntRange Range)4486 std::string PrettyPrintInRange(const llvm::APSInt &Value, IntRange Range) {
4487   if (!Range.Width) return "0";
4488 
4489   llvm::APSInt ValueInRange = Value;
4490   ValueInRange.setIsSigned(!Range.NonNegative);
4491   ValueInRange = ValueInRange.trunc(Range.Width);
4492   return ValueInRange.toString(10);
4493 }
4494 
IsImplicitBoolFloatConversion(Sema & S,Expr * Ex,bool ToBool)4495 static bool IsImplicitBoolFloatConversion(Sema &S, Expr *Ex, bool ToBool) {
4496   if (!isa<ImplicitCastExpr>(Ex))
4497     return false;
4498 
4499   Expr *InnerE = Ex->IgnoreParenImpCasts();
4500   const Type *Target = S.Context.getCanonicalType(Ex->getType()).getTypePtr();
4501   const Type *Source =
4502     S.Context.getCanonicalType(InnerE->getType()).getTypePtr();
4503   if (Target->isDependentType())
4504     return false;
4505 
4506   const BuiltinType *FloatCandidateBT =
4507     dyn_cast<BuiltinType>(ToBool ? Source : Target);
4508   const Type *BoolCandidateType = ToBool ? Target : Source;
4509 
4510   return (BoolCandidateType->isSpecificBuiltinType(BuiltinType::Bool) &&
4511           FloatCandidateBT && (FloatCandidateBT->isFloatingPoint()));
4512 }
4513 
CheckImplicitArgumentConversions(Sema & S,CallExpr * TheCall,SourceLocation CC)4514 void CheckImplicitArgumentConversions(Sema &S, CallExpr *TheCall,
4515                                       SourceLocation CC) {
4516   unsigned NumArgs = TheCall->getNumArgs();
4517   for (unsigned i = 0; i < NumArgs; ++i) {
4518     Expr *CurrA = TheCall->getArg(i);
4519     if (!IsImplicitBoolFloatConversion(S, CurrA, true))
4520       continue;
4521 
4522     bool IsSwapped = ((i > 0) &&
4523         IsImplicitBoolFloatConversion(S, TheCall->getArg(i - 1), false));
4524     IsSwapped |= ((i < (NumArgs - 1)) &&
4525         IsImplicitBoolFloatConversion(S, TheCall->getArg(i + 1), false));
4526     if (IsSwapped) {
4527       // Warn on this floating-point to bool conversion.
4528       DiagnoseImpCast(S, CurrA->IgnoreParenImpCasts(),
4529                       CurrA->getType(), CC,
4530                       diag::warn_impcast_floating_point_to_bool);
4531     }
4532   }
4533 }
4534 
CheckImplicitConversion(Sema & S,Expr * E,QualType T,SourceLocation CC,bool * ICContext=0)4535 void CheckImplicitConversion(Sema &S, Expr *E, QualType T,
4536                              SourceLocation CC, bool *ICContext = 0) {
4537   if (E->isTypeDependent() || E->isValueDependent()) return;
4538 
4539   const Type *Source = S.Context.getCanonicalType(E->getType()).getTypePtr();
4540   const Type *Target = S.Context.getCanonicalType(T).getTypePtr();
4541   if (Source == Target) return;
4542   if (Target->isDependentType()) return;
4543 
4544   // If the conversion context location is invalid don't complain. We also
4545   // don't want to emit a warning if the issue occurs from the expansion of
4546   // a system macro. The problem is that 'getSpellingLoc()' is slow, so we
4547   // delay this check as long as possible. Once we detect we are in that
4548   // scenario, we just return.
4549   if (CC.isInvalid())
4550     return;
4551 
4552   // Diagnose implicit casts to bool.
4553   if (Target->isSpecificBuiltinType(BuiltinType::Bool)) {
4554     if (isa<StringLiteral>(E))
4555       // Warn on string literal to bool.  Checks for string literals in logical
4556       // expressions, for instances, assert(0 && "error here"), is prevented
4557       // by a check in AnalyzeImplicitConversions().
4558       return DiagnoseImpCast(S, E, T, CC,
4559                              diag::warn_impcast_string_literal_to_bool);
4560     if (Source->isFunctionType()) {
4561       // Warn on function to bool. Checks free functions and static member
4562       // functions. Weakly imported functions are excluded from the check,
4563       // since it's common to test their value to check whether the linker
4564       // found a definition for them.
4565       ValueDecl *D = 0;
4566       if (DeclRefExpr* R = dyn_cast<DeclRefExpr>(E)) {
4567         D = R->getDecl();
4568       } else if (MemberExpr *M = dyn_cast<MemberExpr>(E)) {
4569         D = M->getMemberDecl();
4570       }
4571 
4572       if (D && !D->isWeak()) {
4573         if (FunctionDecl* F = dyn_cast<FunctionDecl>(D)) {
4574           S.Diag(E->getExprLoc(), diag::warn_impcast_function_to_bool)
4575             << F << E->getSourceRange() << SourceRange(CC);
4576           S.Diag(E->getExprLoc(), diag::note_function_to_bool_silence)
4577             << FixItHint::CreateInsertion(E->getExprLoc(), "&");
4578           QualType ReturnType;
4579           UnresolvedSet<4> NonTemplateOverloads;
4580           S.isExprCallable(*E, ReturnType, NonTemplateOverloads);
4581           if (!ReturnType.isNull()
4582               && ReturnType->isSpecificBuiltinType(BuiltinType::Bool))
4583             S.Diag(E->getExprLoc(), diag::note_function_to_bool_call)
4584               << FixItHint::CreateInsertion(
4585                  S.getPreprocessor().getLocForEndOfToken(E->getLocEnd()), "()");
4586           return;
4587         }
4588       }
4589     }
4590   }
4591 
4592   // Strip vector types.
4593   if (isa<VectorType>(Source)) {
4594     if (!isa<VectorType>(Target)) {
4595       if (S.SourceMgr.isInSystemMacro(CC))
4596         return;
4597       return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_vector_scalar);
4598     }
4599 
4600     // If the vector cast is cast between two vectors of the same size, it is
4601     // a bitcast, not a conversion.
4602     if (S.Context.getTypeSize(Source) == S.Context.getTypeSize(Target))
4603       return;
4604 
4605     Source = cast<VectorType>(Source)->getElementType().getTypePtr();
4606     Target = cast<VectorType>(Target)->getElementType().getTypePtr();
4607   }
4608 
4609   // Strip complex types.
4610   if (isa<ComplexType>(Source)) {
4611     if (!isa<ComplexType>(Target)) {
4612       if (S.SourceMgr.isInSystemMacro(CC))
4613         return;
4614 
4615       return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_complex_scalar);
4616     }
4617 
4618     Source = cast<ComplexType>(Source)->getElementType().getTypePtr();
4619     Target = cast<ComplexType>(Target)->getElementType().getTypePtr();
4620   }
4621 
4622   const BuiltinType *SourceBT = dyn_cast<BuiltinType>(Source);
4623   const BuiltinType *TargetBT = dyn_cast<BuiltinType>(Target);
4624 
4625   // If the source is floating point...
4626   if (SourceBT && SourceBT->isFloatingPoint()) {
4627     // ...and the target is floating point...
4628     if (TargetBT && TargetBT->isFloatingPoint()) {
4629       // ...then warn if we're dropping FP rank.
4630 
4631       // Builtin FP kinds are ordered by increasing FP rank.
4632       if (SourceBT->getKind() > TargetBT->getKind()) {
4633         // Don't warn about float constants that are precisely
4634         // representable in the target type.
4635         Expr::EvalResult result;
4636         if (E->EvaluateAsRValue(result, S.Context)) {
4637           // Value might be a float, a float vector, or a float complex.
4638           if (IsSameFloatAfterCast(result.Val,
4639                    S.Context.getFloatTypeSemantics(QualType(TargetBT, 0)),
4640                    S.Context.getFloatTypeSemantics(QualType(SourceBT, 0))))
4641             return;
4642         }
4643 
4644         if (S.SourceMgr.isInSystemMacro(CC))
4645           return;
4646 
4647         DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_float_precision);
4648       }
4649       return;
4650     }
4651 
4652     // If the target is integral, always warn.
4653     if (TargetBT && TargetBT->isInteger()) {
4654       if (S.SourceMgr.isInSystemMacro(CC))
4655         return;
4656 
4657       Expr *InnerE = E->IgnoreParenImpCasts();
4658       // We also want to warn on, e.g., "int i = -1.234"
4659       if (UnaryOperator *UOp = dyn_cast<UnaryOperator>(InnerE))
4660         if (UOp->getOpcode() == UO_Minus || UOp->getOpcode() == UO_Plus)
4661           InnerE = UOp->getSubExpr()->IgnoreParenImpCasts();
4662 
4663       if (FloatingLiteral *FL = dyn_cast<FloatingLiteral>(InnerE)) {
4664         DiagnoseFloatingLiteralImpCast(S, FL, T, CC);
4665       } else {
4666         DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_float_integer);
4667       }
4668     }
4669 
4670     // If the target is bool, warn if expr is a function or method call.
4671     if (Target->isSpecificBuiltinType(BuiltinType::Bool) &&
4672         isa<CallExpr>(E)) {
4673       // Check last argument of function call to see if it is an
4674       // implicit cast from a type matching the type the result
4675       // is being cast to.
4676       CallExpr *CEx = cast<CallExpr>(E);
4677       unsigned NumArgs = CEx->getNumArgs();
4678       if (NumArgs > 0) {
4679         Expr *LastA = CEx->getArg(NumArgs - 1);
4680         Expr *InnerE = LastA->IgnoreParenImpCasts();
4681         const Type *InnerType =
4682           S.Context.getCanonicalType(InnerE->getType()).getTypePtr();
4683         if (isa<ImplicitCastExpr>(LastA) && (InnerType == Target)) {
4684           // Warn on this floating-point to bool conversion
4685           DiagnoseImpCast(S, E, T, CC,
4686                           diag::warn_impcast_floating_point_to_bool);
4687         }
4688       }
4689     }
4690     return;
4691   }
4692 
4693   if ((E->isNullPointerConstant(S.Context, Expr::NPC_ValueDependentIsNotNull)
4694            == Expr::NPCK_GNUNull) && !Target->isAnyPointerType()
4695       && !Target->isBlockPointerType() && !Target->isMemberPointerType()) {
4696     SourceLocation Loc = E->getSourceRange().getBegin();
4697     if (Loc.isMacroID())
4698       Loc = S.SourceMgr.getImmediateExpansionRange(Loc).first;
4699     if (!Loc.isMacroID() || CC.isMacroID())
4700       S.Diag(Loc, diag::warn_impcast_null_pointer_to_integer)
4701           << T << clang::SourceRange(CC)
4702           << FixItHint::CreateReplacement(Loc, S.getFixItZeroLiteralForType(T));
4703   }
4704 
4705   if (!Source->isIntegerType() || !Target->isIntegerType())
4706     return;
4707 
4708   // TODO: remove this early return once the false positives for constant->bool
4709   // in templates, macros, etc, are reduced or removed.
4710   if (Target->isSpecificBuiltinType(BuiltinType::Bool))
4711     return;
4712 
4713   IntRange SourceRange = GetExprRange(S.Context, E);
4714   IntRange TargetRange = IntRange::forTargetOfCanonicalType(S.Context, Target);
4715 
4716   if (SourceRange.Width > TargetRange.Width) {
4717     // If the source is a constant, use a default-on diagnostic.
4718     // TODO: this should happen for bitfield stores, too.
4719     llvm::APSInt Value(32);
4720     if (E->isIntegerConstantExpr(Value, S.Context)) {
4721       if (S.SourceMgr.isInSystemMacro(CC))
4722         return;
4723 
4724       std::string PrettySourceValue = Value.toString(10);
4725       std::string PrettyTargetValue = PrettyPrintInRange(Value, TargetRange);
4726 
4727       S.DiagRuntimeBehavior(E->getExprLoc(), E,
4728         S.PDiag(diag::warn_impcast_integer_precision_constant)
4729             << PrettySourceValue << PrettyTargetValue
4730             << E->getType() << T << E->getSourceRange()
4731             << clang::SourceRange(CC));
4732       return;
4733     }
4734 
4735     // People want to build with -Wshorten-64-to-32 and not -Wconversion.
4736     if (S.SourceMgr.isInSystemMacro(CC))
4737       return;
4738 
4739     if (TargetRange.Width == 32 && S.Context.getIntWidth(E->getType()) == 64)
4740       return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_integer_64_32,
4741                              /* pruneControlFlow */ true);
4742     return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_integer_precision);
4743   }
4744 
4745   if ((TargetRange.NonNegative && !SourceRange.NonNegative) ||
4746       (!TargetRange.NonNegative && SourceRange.NonNegative &&
4747        SourceRange.Width == TargetRange.Width)) {
4748 
4749     if (S.SourceMgr.isInSystemMacro(CC))
4750       return;
4751 
4752     unsigned DiagID = diag::warn_impcast_integer_sign;
4753 
4754     // Traditionally, gcc has warned about this under -Wsign-compare.
4755     // We also want to warn about it in -Wconversion.
4756     // So if -Wconversion is off, use a completely identical diagnostic
4757     // in the sign-compare group.
4758     // The conditional-checking code will
4759     if (ICContext) {
4760       DiagID = diag::warn_impcast_integer_sign_conditional;
4761       *ICContext = true;
4762     }
4763 
4764     return DiagnoseImpCast(S, E, T, CC, DiagID);
4765   }
4766 
4767   // Diagnose conversions between different enumeration types.
4768   // In C, we pretend that the type of an EnumConstantDecl is its enumeration
4769   // type, to give us better diagnostics.
4770   QualType SourceType = E->getType();
4771   if (!S.getLangOpts().CPlusPlus) {
4772     if (DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(E))
4773       if (EnumConstantDecl *ECD = dyn_cast<EnumConstantDecl>(DRE->getDecl())) {
4774         EnumDecl *Enum = cast<EnumDecl>(ECD->getDeclContext());
4775         SourceType = S.Context.getTypeDeclType(Enum);
4776         Source = S.Context.getCanonicalType(SourceType).getTypePtr();
4777       }
4778   }
4779 
4780   if (const EnumType *SourceEnum = Source->getAs<EnumType>())
4781     if (const EnumType *TargetEnum = Target->getAs<EnumType>())
4782       if ((SourceEnum->getDecl()->getIdentifier() ||
4783            SourceEnum->getDecl()->getTypedefNameForAnonDecl()) &&
4784           (TargetEnum->getDecl()->getIdentifier() ||
4785            TargetEnum->getDecl()->getTypedefNameForAnonDecl()) &&
4786           SourceEnum != TargetEnum) {
4787         if (S.SourceMgr.isInSystemMacro(CC))
4788           return;
4789 
4790         return DiagnoseImpCast(S, E, SourceType, T, CC,
4791                                diag::warn_impcast_different_enum_types);
4792       }
4793 
4794   return;
4795 }
4796 
4797 void CheckConditionalOperator(Sema &S, ConditionalOperator *E,
4798                               SourceLocation CC, QualType T);
4799 
CheckConditionalOperand(Sema & S,Expr * E,QualType T,SourceLocation CC,bool & ICContext)4800 void CheckConditionalOperand(Sema &S, Expr *E, QualType T,
4801                              SourceLocation CC, bool &ICContext) {
4802   E = E->IgnoreParenImpCasts();
4803 
4804   if (isa<ConditionalOperator>(E))
4805     return CheckConditionalOperator(S, cast<ConditionalOperator>(E), CC, T);
4806 
4807   AnalyzeImplicitConversions(S, E, CC);
4808   if (E->getType() != T)
4809     return CheckImplicitConversion(S, E, T, CC, &ICContext);
4810   return;
4811 }
4812 
CheckConditionalOperator(Sema & S,ConditionalOperator * E,SourceLocation CC,QualType T)4813 void CheckConditionalOperator(Sema &S, ConditionalOperator *E,
4814                               SourceLocation CC, QualType T) {
4815   AnalyzeImplicitConversions(S, E->getCond(), CC);
4816 
4817   bool Suspicious = false;
4818   CheckConditionalOperand(S, E->getTrueExpr(), T, CC, Suspicious);
4819   CheckConditionalOperand(S, E->getFalseExpr(), T, CC, Suspicious);
4820 
4821   // If -Wconversion would have warned about either of the candidates
4822   // for a signedness conversion to the context type...
4823   if (!Suspicious) return;
4824 
4825   // ...but it's currently ignored...
4826   if (S.Diags.getDiagnosticLevel(diag::warn_impcast_integer_sign_conditional,
4827                                  CC))
4828     return;
4829 
4830   // ...then check whether it would have warned about either of the
4831   // candidates for a signedness conversion to the condition type.
4832   if (E->getType() == T) return;
4833 
4834   Suspicious = false;
4835   CheckImplicitConversion(S, E->getTrueExpr()->IgnoreParenImpCasts(),
4836                           E->getType(), CC, &Suspicious);
4837   if (!Suspicious)
4838     CheckImplicitConversion(S, E->getFalseExpr()->IgnoreParenImpCasts(),
4839                             E->getType(), CC, &Suspicious);
4840 }
4841 
4842 /// AnalyzeImplicitConversions - Find and report any interesting
4843 /// implicit conversions in the given expression.  There are a couple
4844 /// of competing diagnostics here, -Wconversion and -Wsign-compare.
AnalyzeImplicitConversions(Sema & S,Expr * OrigE,SourceLocation CC)4845 void AnalyzeImplicitConversions(Sema &S, Expr *OrigE, SourceLocation CC) {
4846   QualType T = OrigE->getType();
4847   Expr *E = OrigE->IgnoreParenImpCasts();
4848 
4849   if (E->isTypeDependent() || E->isValueDependent())
4850     return;
4851 
4852   // For conditional operators, we analyze the arguments as if they
4853   // were being fed directly into the output.
4854   if (isa<ConditionalOperator>(E)) {
4855     ConditionalOperator *CO = cast<ConditionalOperator>(E);
4856     CheckConditionalOperator(S, CO, CC, T);
4857     return;
4858   }
4859 
4860   // Check implicit argument conversions for function calls.
4861   if (CallExpr *Call = dyn_cast<CallExpr>(E))
4862     CheckImplicitArgumentConversions(S, Call, CC);
4863 
4864   // Go ahead and check any implicit conversions we might have skipped.
4865   // The non-canonical typecheck is just an optimization;
4866   // CheckImplicitConversion will filter out dead implicit conversions.
4867   if (E->getType() != T)
4868     CheckImplicitConversion(S, E, T, CC);
4869 
4870   // Now continue drilling into this expression.
4871 
4872   // Skip past explicit casts.
4873   if (isa<ExplicitCastExpr>(E)) {
4874     E = cast<ExplicitCastExpr>(E)->getSubExpr()->IgnoreParenImpCasts();
4875     return AnalyzeImplicitConversions(S, E, CC);
4876   }
4877 
4878   if (BinaryOperator *BO = dyn_cast<BinaryOperator>(E)) {
4879     // Do a somewhat different check with comparison operators.
4880     if (BO->isComparisonOp())
4881       return AnalyzeComparison(S, BO);
4882 
4883     // And with simple assignments.
4884     if (BO->getOpcode() == BO_Assign)
4885       return AnalyzeAssignment(S, BO);
4886   }
4887 
4888   // These break the otherwise-useful invariant below.  Fortunately,
4889   // we don't really need to recurse into them, because any internal
4890   // expressions should have been analyzed already when they were
4891   // built into statements.
4892   if (isa<StmtExpr>(E)) return;
4893 
4894   // Don't descend into unevaluated contexts.
4895   if (isa<UnaryExprOrTypeTraitExpr>(E)) return;
4896 
4897   // Now just recurse over the expression's children.
4898   CC = E->getExprLoc();
4899   BinaryOperator *BO = dyn_cast<BinaryOperator>(E);
4900   bool IsLogicalOperator = BO && BO->isLogicalOp();
4901   for (Stmt::child_range I = E->children(); I; ++I) {
4902     Expr *ChildExpr = dyn_cast_or_null<Expr>(*I);
4903     if (!ChildExpr)
4904       continue;
4905 
4906     if (IsLogicalOperator &&
4907         isa<StringLiteral>(ChildExpr->IgnoreParenImpCasts()))
4908       // Ignore checking string literals that are in logical operators.
4909       continue;
4910     AnalyzeImplicitConversions(S, ChildExpr, CC);
4911   }
4912 }
4913 
4914 } // end anonymous namespace
4915 
4916 /// Diagnoses "dangerous" implicit conversions within the given
4917 /// expression (which is a full expression).  Implements -Wconversion
4918 /// and -Wsign-compare.
4919 ///
4920 /// \param CC the "context" location of the implicit conversion, i.e.
4921 ///   the most location of the syntactic entity requiring the implicit
4922 ///   conversion
CheckImplicitConversions(Expr * E,SourceLocation CC)4923 void Sema::CheckImplicitConversions(Expr *E, SourceLocation CC) {
4924   // Don't diagnose in unevaluated contexts.
4925   if (isUnevaluatedContext())
4926     return;
4927 
4928   // Don't diagnose for value- or type-dependent expressions.
4929   if (E->isTypeDependent() || E->isValueDependent())
4930     return;
4931 
4932   // Check for array bounds violations in cases where the check isn't triggered
4933   // elsewhere for other Expr types (like BinaryOperators), e.g. when an
4934   // ArraySubscriptExpr is on the RHS of a variable initialization.
4935   CheckArrayAccess(E);
4936 
4937   // This is not the right CC for (e.g.) a variable initialization.
4938   AnalyzeImplicitConversions(*this, E, CC);
4939 }
4940 
CheckBitFieldInitialization(SourceLocation InitLoc,FieldDecl * BitField,Expr * Init)4941 void Sema::CheckBitFieldInitialization(SourceLocation InitLoc,
4942                                        FieldDecl *BitField,
4943                                        Expr *Init) {
4944   (void) AnalyzeBitFieldAssignment(*this, BitField, Init, InitLoc);
4945 }
4946 
4947 /// CheckParmsForFunctionDef - Check that the parameters of the given
4948 /// function are appropriate for the definition of a function. This
4949 /// takes care of any checks that cannot be performed on the
4950 /// declaration itself, e.g., that the types of each of the function
4951 /// parameters are complete.
CheckParmsForFunctionDef(ParmVarDecl ** P,ParmVarDecl ** PEnd,bool CheckParameterNames)4952 bool Sema::CheckParmsForFunctionDef(ParmVarDecl **P, ParmVarDecl **PEnd,
4953                                     bool CheckParameterNames) {
4954   bool HasInvalidParm = false;
4955   for (; P != PEnd; ++P) {
4956     ParmVarDecl *Param = *P;
4957 
4958     // C99 6.7.5.3p4: the parameters in a parameter type list in a
4959     // function declarator that is part of a function definition of
4960     // that function shall not have incomplete type.
4961     //
4962     // This is also C++ [dcl.fct]p6.
4963     if (!Param->isInvalidDecl() &&
4964         RequireCompleteType(Param->getLocation(), Param->getType(),
4965                             diag::err_typecheck_decl_incomplete_type)) {
4966       Param->setInvalidDecl();
4967       HasInvalidParm = true;
4968     }
4969 
4970     // C99 6.9.1p5: If the declarator includes a parameter type list, the
4971     // declaration of each parameter shall include an identifier.
4972     if (CheckParameterNames &&
4973         Param->getIdentifier() == 0 &&
4974         !Param->isImplicit() &&
4975         !getLangOpts().CPlusPlus)
4976       Diag(Param->getLocation(), diag::err_parameter_name_omitted);
4977 
4978     // C99 6.7.5.3p12:
4979     //   If the function declarator is not part of a definition of that
4980     //   function, parameters may have incomplete type and may use the [*]
4981     //   notation in their sequences of declarator specifiers to specify
4982     //   variable length array types.
4983     QualType PType = Param->getOriginalType();
4984     if (const ArrayType *AT = Context.getAsArrayType(PType)) {
4985       if (AT->getSizeModifier() == ArrayType::Star) {
4986         // FIXME: This diagnosic should point the '[*]' if source-location
4987         // information is added for it.
4988         Diag(Param->getLocation(), diag::err_array_star_in_function_definition);
4989       }
4990     }
4991   }
4992 
4993   return HasInvalidParm;
4994 }
4995 
4996 /// CheckCastAlign - Implements -Wcast-align, which warns when a
4997 /// pointer cast increases the alignment requirements.
CheckCastAlign(Expr * Op,QualType T,SourceRange TRange)4998 void Sema::CheckCastAlign(Expr *Op, QualType T, SourceRange TRange) {
4999   // This is actually a lot of work to potentially be doing on every
5000   // cast; don't do it if we're ignoring -Wcast_align (as is the default).
5001   if (getDiagnostics().getDiagnosticLevel(diag::warn_cast_align,
5002                                           TRange.getBegin())
5003         == DiagnosticsEngine::Ignored)
5004     return;
5005 
5006   // Ignore dependent types.
5007   if (T->isDependentType() || Op->getType()->isDependentType())
5008     return;
5009 
5010   // Require that the destination be a pointer type.
5011   const PointerType *DestPtr = T->getAs<PointerType>();
5012   if (!DestPtr) return;
5013 
5014   // If the destination has alignment 1, we're done.
5015   QualType DestPointee = DestPtr->getPointeeType();
5016   if (DestPointee->isIncompleteType()) return;
5017   CharUnits DestAlign = Context.getTypeAlignInChars(DestPointee);
5018   if (DestAlign.isOne()) return;
5019 
5020   // Require that the source be a pointer type.
5021   const PointerType *SrcPtr = Op->getType()->getAs<PointerType>();
5022   if (!SrcPtr) return;
5023   QualType SrcPointee = SrcPtr->getPointeeType();
5024 
5025   // Whitelist casts from cv void*.  We already implicitly
5026   // whitelisted casts to cv void*, since they have alignment 1.
5027   // Also whitelist casts involving incomplete types, which implicitly
5028   // includes 'void'.
5029   if (SrcPointee->isIncompleteType()) return;
5030 
5031   CharUnits SrcAlign = Context.getTypeAlignInChars(SrcPointee);
5032   if (SrcAlign >= DestAlign) return;
5033 
5034   Diag(TRange.getBegin(), diag::warn_cast_align)
5035     << Op->getType() << T
5036     << static_cast<unsigned>(SrcAlign.getQuantity())
5037     << static_cast<unsigned>(DestAlign.getQuantity())
5038     << TRange << Op->getSourceRange();
5039 }
5040 
getElementType(const Expr * BaseExpr)5041 static const Type* getElementType(const Expr *BaseExpr) {
5042   const Type* EltType = BaseExpr->getType().getTypePtr();
5043   if (EltType->isAnyPointerType())
5044     return EltType->getPointeeType().getTypePtr();
5045   else if (EltType->isArrayType())
5046     return EltType->getBaseElementTypeUnsafe();
5047   return EltType;
5048 }
5049 
5050 /// \brief Check whether this array fits the idiom of a size-one tail padded
5051 /// array member of a struct.
5052 ///
5053 /// We avoid emitting out-of-bounds access warnings for such arrays as they are
5054 /// commonly used to emulate flexible arrays in C89 code.
IsTailPaddedMemberArray(Sema & S,llvm::APInt Size,const NamedDecl * ND)5055 static bool IsTailPaddedMemberArray(Sema &S, llvm::APInt Size,
5056                                     const NamedDecl *ND) {
5057   if (Size != 1 || !ND) return false;
5058 
5059   const FieldDecl *FD = dyn_cast<FieldDecl>(ND);
5060   if (!FD) return false;
5061 
5062   // Don't consider sizes resulting from macro expansions or template argument
5063   // substitution to form C89 tail-padded arrays.
5064 
5065   TypeSourceInfo *TInfo = FD->getTypeSourceInfo();
5066   while (TInfo) {
5067     TypeLoc TL = TInfo->getTypeLoc();
5068     // Look through typedefs.
5069     const TypedefTypeLoc *TTL = dyn_cast<TypedefTypeLoc>(&TL);
5070     if (TTL) {
5071       const TypedefNameDecl *TDL = TTL->getTypedefNameDecl();
5072       TInfo = TDL->getTypeSourceInfo();
5073       continue;
5074     }
5075     ConstantArrayTypeLoc CTL = cast<ConstantArrayTypeLoc>(TL);
5076     const Expr *SizeExpr = dyn_cast<IntegerLiteral>(CTL.getSizeExpr());
5077     if (!SizeExpr || SizeExpr->getExprLoc().isMacroID())
5078       return false;
5079     break;
5080   }
5081 
5082   const RecordDecl *RD = dyn_cast<RecordDecl>(FD->getDeclContext());
5083   if (!RD) return false;
5084   if (RD->isUnion()) return false;
5085   if (const CXXRecordDecl *CRD = dyn_cast<CXXRecordDecl>(RD)) {
5086     if (!CRD->isStandardLayout()) return false;
5087   }
5088 
5089   // See if this is the last field decl in the record.
5090   const Decl *D = FD;
5091   while ((D = D->getNextDeclInContext()))
5092     if (isa<FieldDecl>(D))
5093       return false;
5094   return true;
5095 }
5096 
CheckArrayAccess(const Expr * BaseExpr,const Expr * IndexExpr,const ArraySubscriptExpr * ASE,bool AllowOnePastEnd,bool IndexNegated)5097 void Sema::CheckArrayAccess(const Expr *BaseExpr, const Expr *IndexExpr,
5098                             const ArraySubscriptExpr *ASE,
5099                             bool AllowOnePastEnd, bool IndexNegated) {
5100   IndexExpr = IndexExpr->IgnoreParenImpCasts();
5101   if (IndexExpr->isValueDependent())
5102     return;
5103 
5104   const Type *EffectiveType = getElementType(BaseExpr);
5105   BaseExpr = BaseExpr->IgnoreParenCasts();
5106   const ConstantArrayType *ArrayTy =
5107     Context.getAsConstantArrayType(BaseExpr->getType());
5108   if (!ArrayTy)
5109     return;
5110 
5111   llvm::APSInt index;
5112   if (!IndexExpr->EvaluateAsInt(index, Context))
5113     return;
5114   if (IndexNegated)
5115     index = -index;
5116 
5117   const NamedDecl *ND = NULL;
5118   if (const DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(BaseExpr))
5119     ND = dyn_cast<NamedDecl>(DRE->getDecl());
5120   if (const MemberExpr *ME = dyn_cast<MemberExpr>(BaseExpr))
5121     ND = dyn_cast<NamedDecl>(ME->getMemberDecl());
5122 
5123   if (index.isUnsigned() || !index.isNegative()) {
5124     llvm::APInt size = ArrayTy->getSize();
5125     if (!size.isStrictlyPositive())
5126       return;
5127 
5128     const Type* BaseType = getElementType(BaseExpr);
5129     if (BaseType != EffectiveType) {
5130       // Make sure we're comparing apples to apples when comparing index to size
5131       uint64_t ptrarith_typesize = Context.getTypeSize(EffectiveType);
5132       uint64_t array_typesize = Context.getTypeSize(BaseType);
5133       // Handle ptrarith_typesize being zero, such as when casting to void*
5134       if (!ptrarith_typesize) ptrarith_typesize = 1;
5135       if (ptrarith_typesize != array_typesize) {
5136         // There's a cast to a different size type involved
5137         uint64_t ratio = array_typesize / ptrarith_typesize;
5138         // TODO: Be smarter about handling cases where array_typesize is not a
5139         // multiple of ptrarith_typesize
5140         if (ptrarith_typesize * ratio == array_typesize)
5141           size *= llvm::APInt(size.getBitWidth(), ratio);
5142       }
5143     }
5144 
5145     if (size.getBitWidth() > index.getBitWidth())
5146       index = index.zext(size.getBitWidth());
5147     else if (size.getBitWidth() < index.getBitWidth())
5148       size = size.zext(index.getBitWidth());
5149 
5150     // For array subscripting the index must be less than size, but for pointer
5151     // arithmetic also allow the index (offset) to be equal to size since
5152     // computing the next address after the end of the array is legal and
5153     // commonly done e.g. in C++ iterators and range-based for loops.
5154     if (AllowOnePastEnd ? index.ule(size) : index.ult(size))
5155       return;
5156 
5157     // Also don't warn for arrays of size 1 which are members of some
5158     // structure. These are often used to approximate flexible arrays in C89
5159     // code.
5160     if (IsTailPaddedMemberArray(*this, size, ND))
5161       return;
5162 
5163     // Suppress the warning if the subscript expression (as identified by the
5164     // ']' location) and the index expression are both from macro expansions
5165     // within a system header.
5166     if (ASE) {
5167       SourceLocation RBracketLoc = SourceMgr.getSpellingLoc(
5168           ASE->getRBracketLoc());
5169       if (SourceMgr.isInSystemHeader(RBracketLoc)) {
5170         SourceLocation IndexLoc = SourceMgr.getSpellingLoc(
5171             IndexExpr->getLocStart());
5172         if (SourceMgr.isFromSameFile(RBracketLoc, IndexLoc))
5173           return;
5174       }
5175     }
5176 
5177     unsigned DiagID = diag::warn_ptr_arith_exceeds_bounds;
5178     if (ASE)
5179       DiagID = diag::warn_array_index_exceeds_bounds;
5180 
5181     DiagRuntimeBehavior(BaseExpr->getLocStart(), BaseExpr,
5182                         PDiag(DiagID) << index.toString(10, true)
5183                           << size.toString(10, true)
5184                           << (unsigned)size.getLimitedValue(~0U)
5185                           << IndexExpr->getSourceRange());
5186   } else {
5187     unsigned DiagID = diag::warn_array_index_precedes_bounds;
5188     if (!ASE) {
5189       DiagID = diag::warn_ptr_arith_precedes_bounds;
5190       if (index.isNegative()) index = -index;
5191     }
5192 
5193     DiagRuntimeBehavior(BaseExpr->getLocStart(), BaseExpr,
5194                         PDiag(DiagID) << index.toString(10, true)
5195                           << IndexExpr->getSourceRange());
5196   }
5197 
5198   if (!ND) {
5199     // Try harder to find a NamedDecl to point at in the note.
5200     while (const ArraySubscriptExpr *ASE =
5201            dyn_cast<ArraySubscriptExpr>(BaseExpr))
5202       BaseExpr = ASE->getBase()->IgnoreParenCasts();
5203     if (const DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(BaseExpr))
5204       ND = dyn_cast<NamedDecl>(DRE->getDecl());
5205     if (const MemberExpr *ME = dyn_cast<MemberExpr>(BaseExpr))
5206       ND = dyn_cast<NamedDecl>(ME->getMemberDecl());
5207   }
5208 
5209   if (ND)
5210     DiagRuntimeBehavior(ND->getLocStart(), BaseExpr,
5211                         PDiag(diag::note_array_index_out_of_bounds)
5212                           << ND->getDeclName());
5213 }
5214 
CheckArrayAccess(const Expr * expr)5215 void Sema::CheckArrayAccess(const Expr *expr) {
5216   int AllowOnePastEnd = 0;
5217   while (expr) {
5218     expr = expr->IgnoreParenImpCasts();
5219     switch (expr->getStmtClass()) {
5220       case Stmt::ArraySubscriptExprClass: {
5221         const ArraySubscriptExpr *ASE = cast<ArraySubscriptExpr>(expr);
5222         CheckArrayAccess(ASE->getBase(), ASE->getIdx(), ASE,
5223                          AllowOnePastEnd > 0);
5224         return;
5225       }
5226       case Stmt::UnaryOperatorClass: {
5227         // Only unwrap the * and & unary operators
5228         const UnaryOperator *UO = cast<UnaryOperator>(expr);
5229         expr = UO->getSubExpr();
5230         switch (UO->getOpcode()) {
5231           case UO_AddrOf:
5232             AllowOnePastEnd++;
5233             break;
5234           case UO_Deref:
5235             AllowOnePastEnd--;
5236             break;
5237           default:
5238             return;
5239         }
5240         break;
5241       }
5242       case Stmt::ConditionalOperatorClass: {
5243         const ConditionalOperator *cond = cast<ConditionalOperator>(expr);
5244         if (const Expr *lhs = cond->getLHS())
5245           CheckArrayAccess(lhs);
5246         if (const Expr *rhs = cond->getRHS())
5247           CheckArrayAccess(rhs);
5248         return;
5249       }
5250       default:
5251         return;
5252     }
5253   }
5254 }
5255 
5256 //===--- CHECK: Objective-C retain cycles ----------------------------------//
5257 
5258 namespace {
5259   struct RetainCycleOwner {
RetainCycleOwner__anon32971bd90611::RetainCycleOwner5260     RetainCycleOwner() : Variable(0), Indirect(false) {}
5261     VarDecl *Variable;
5262     SourceRange Range;
5263     SourceLocation Loc;
5264     bool Indirect;
5265 
setLocsFrom__anon32971bd90611::RetainCycleOwner5266     void setLocsFrom(Expr *e) {
5267       Loc = e->getExprLoc();
5268       Range = e->getSourceRange();
5269     }
5270   };
5271 }
5272 
5273 /// Consider whether capturing the given variable can possibly lead to
5274 /// a retain cycle.
considerVariable(VarDecl * var,Expr * ref,RetainCycleOwner & owner)5275 static bool considerVariable(VarDecl *var, Expr *ref, RetainCycleOwner &owner) {
5276   // In ARC, it's captured strongly iff the variable has __strong
5277   // lifetime.  In MRR, it's captured strongly if the variable is
5278   // __block and has an appropriate type.
5279   if (var->getType().getObjCLifetime() != Qualifiers::OCL_Strong)
5280     return false;
5281 
5282   owner.Variable = var;
5283   owner.setLocsFrom(ref);
5284   return true;
5285 }
5286 
findRetainCycleOwner(Sema & S,Expr * e,RetainCycleOwner & owner)5287 static bool findRetainCycleOwner(Sema &S, Expr *e, RetainCycleOwner &owner) {
5288   while (true) {
5289     e = e->IgnoreParens();
5290     if (CastExpr *cast = dyn_cast<CastExpr>(e)) {
5291       switch (cast->getCastKind()) {
5292       case CK_BitCast:
5293       case CK_LValueBitCast:
5294       case CK_LValueToRValue:
5295       case CK_ARCReclaimReturnedObject:
5296         e = cast->getSubExpr();
5297         continue;
5298 
5299       default:
5300         return false;
5301       }
5302     }
5303 
5304     if (ObjCIvarRefExpr *ref = dyn_cast<ObjCIvarRefExpr>(e)) {
5305       ObjCIvarDecl *ivar = ref->getDecl();
5306       if (ivar->getType().getObjCLifetime() != Qualifiers::OCL_Strong)
5307         return false;
5308 
5309       // Try to find a retain cycle in the base.
5310       if (!findRetainCycleOwner(S, ref->getBase(), owner))
5311         return false;
5312 
5313       if (ref->isFreeIvar()) owner.setLocsFrom(ref);
5314       owner.Indirect = true;
5315       return true;
5316     }
5317 
5318     if (DeclRefExpr *ref = dyn_cast<DeclRefExpr>(e)) {
5319       VarDecl *var = dyn_cast<VarDecl>(ref->getDecl());
5320       if (!var) return false;
5321       return considerVariable(var, ref, owner);
5322     }
5323 
5324     if (MemberExpr *member = dyn_cast<MemberExpr>(e)) {
5325       if (member->isArrow()) return false;
5326 
5327       // Don't count this as an indirect ownership.
5328       e = member->getBase();
5329       continue;
5330     }
5331 
5332     if (PseudoObjectExpr *pseudo = dyn_cast<PseudoObjectExpr>(e)) {
5333       // Only pay attention to pseudo-objects on property references.
5334       ObjCPropertyRefExpr *pre
5335         = dyn_cast<ObjCPropertyRefExpr>(pseudo->getSyntacticForm()
5336                                               ->IgnoreParens());
5337       if (!pre) return false;
5338       if (pre->isImplicitProperty()) return false;
5339       ObjCPropertyDecl *property = pre->getExplicitProperty();
5340       if (!property->isRetaining() &&
5341           !(property->getPropertyIvarDecl() &&
5342             property->getPropertyIvarDecl()->getType()
5343               .getObjCLifetime() == Qualifiers::OCL_Strong))
5344           return false;
5345 
5346       owner.Indirect = true;
5347       if (pre->isSuperReceiver()) {
5348         owner.Variable = S.getCurMethodDecl()->getSelfDecl();
5349         if (!owner.Variable)
5350           return false;
5351         owner.Loc = pre->getLocation();
5352         owner.Range = pre->getSourceRange();
5353         return true;
5354       }
5355       e = const_cast<Expr*>(cast<OpaqueValueExpr>(pre->getBase())
5356                               ->getSourceExpr());
5357       continue;
5358     }
5359 
5360     // Array ivars?
5361 
5362     return false;
5363   }
5364 }
5365 
5366 namespace {
5367   struct FindCaptureVisitor : EvaluatedExprVisitor<FindCaptureVisitor> {
FindCaptureVisitor__anon32971bd90711::FindCaptureVisitor5368     FindCaptureVisitor(ASTContext &Context, VarDecl *variable)
5369       : EvaluatedExprVisitor<FindCaptureVisitor>(Context),
5370         Variable(variable), Capturer(0) {}
5371 
5372     VarDecl *Variable;
5373     Expr *Capturer;
5374 
VisitDeclRefExpr__anon32971bd90711::FindCaptureVisitor5375     void VisitDeclRefExpr(DeclRefExpr *ref) {
5376       if (ref->getDecl() == Variable && !Capturer)
5377         Capturer = ref;
5378     }
5379 
VisitObjCIvarRefExpr__anon32971bd90711::FindCaptureVisitor5380     void VisitObjCIvarRefExpr(ObjCIvarRefExpr *ref) {
5381       if (Capturer) return;
5382       Visit(ref->getBase());
5383       if (Capturer && ref->isFreeIvar())
5384         Capturer = ref;
5385     }
5386 
VisitBlockExpr__anon32971bd90711::FindCaptureVisitor5387     void VisitBlockExpr(BlockExpr *block) {
5388       // Look inside nested blocks
5389       if (block->getBlockDecl()->capturesVariable(Variable))
5390         Visit(block->getBlockDecl()->getBody());
5391     }
5392 
VisitOpaqueValueExpr__anon32971bd90711::FindCaptureVisitor5393     void VisitOpaqueValueExpr(OpaqueValueExpr *OVE) {
5394       if (Capturer) return;
5395       if (OVE->getSourceExpr())
5396         Visit(OVE->getSourceExpr());
5397     }
5398   };
5399 }
5400 
5401 /// Check whether the given argument is a block which captures a
5402 /// variable.
findCapturingExpr(Sema & S,Expr * e,RetainCycleOwner & owner)5403 static Expr *findCapturingExpr(Sema &S, Expr *e, RetainCycleOwner &owner) {
5404   assert(owner.Variable && owner.Loc.isValid());
5405 
5406   e = e->IgnoreParenCasts();
5407   BlockExpr *block = dyn_cast<BlockExpr>(e);
5408   if (!block || !block->getBlockDecl()->capturesVariable(owner.Variable))
5409     return 0;
5410 
5411   FindCaptureVisitor visitor(S.Context, owner.Variable);
5412   visitor.Visit(block->getBlockDecl()->getBody());
5413   return visitor.Capturer;
5414 }
5415 
diagnoseRetainCycle(Sema & S,Expr * capturer,RetainCycleOwner & owner)5416 static void diagnoseRetainCycle(Sema &S, Expr *capturer,
5417                                 RetainCycleOwner &owner) {
5418   assert(capturer);
5419   assert(owner.Variable && owner.Loc.isValid());
5420 
5421   S.Diag(capturer->getExprLoc(), diag::warn_arc_retain_cycle)
5422     << owner.Variable << capturer->getSourceRange();
5423   S.Diag(owner.Loc, diag::note_arc_retain_cycle_owner)
5424     << owner.Indirect << owner.Range;
5425 }
5426 
5427 /// Check for a keyword selector that starts with the word 'add' or
5428 /// 'set'.
isSetterLikeSelector(Selector sel)5429 static bool isSetterLikeSelector(Selector sel) {
5430   if (sel.isUnarySelector()) return false;
5431 
5432   StringRef str = sel.getNameForSlot(0);
5433   while (!str.empty() && str.front() == '_') str = str.substr(1);
5434   if (str.startswith("set"))
5435     str = str.substr(3);
5436   else if (str.startswith("add")) {
5437     // Specially whitelist 'addOperationWithBlock:'.
5438     if (sel.getNumArgs() == 1 && str.startswith("addOperationWithBlock"))
5439       return false;
5440     str = str.substr(3);
5441   }
5442   else
5443     return false;
5444 
5445   if (str.empty()) return true;
5446   return !islower(str.front());
5447 }
5448 
5449 /// Check a message send to see if it's likely to cause a retain cycle.
checkRetainCycles(ObjCMessageExpr * msg)5450 void Sema::checkRetainCycles(ObjCMessageExpr *msg) {
5451   // Only check instance methods whose selector looks like a setter.
5452   if (!msg->isInstanceMessage() || !isSetterLikeSelector(msg->getSelector()))
5453     return;
5454 
5455   // Try to find a variable that the receiver is strongly owned by.
5456   RetainCycleOwner owner;
5457   if (msg->getReceiverKind() == ObjCMessageExpr::Instance) {
5458     if (!findRetainCycleOwner(*this, msg->getInstanceReceiver(), owner))
5459       return;
5460   } else {
5461     assert(msg->getReceiverKind() == ObjCMessageExpr::SuperInstance);
5462     owner.Variable = getCurMethodDecl()->getSelfDecl();
5463     owner.Loc = msg->getSuperLoc();
5464     owner.Range = msg->getSuperLoc();
5465   }
5466 
5467   // Check whether the receiver is captured by any of the arguments.
5468   for (unsigned i = 0, e = msg->getNumArgs(); i != e; ++i)
5469     if (Expr *capturer = findCapturingExpr(*this, msg->getArg(i), owner))
5470       return diagnoseRetainCycle(*this, capturer, owner);
5471 }
5472 
5473 /// Check a property assign to see if it's likely to cause a retain cycle.
checkRetainCycles(Expr * receiver,Expr * argument)5474 void Sema::checkRetainCycles(Expr *receiver, Expr *argument) {
5475   RetainCycleOwner owner;
5476   if (!findRetainCycleOwner(*this, receiver, owner))
5477     return;
5478 
5479   if (Expr *capturer = findCapturingExpr(*this, argument, owner))
5480     diagnoseRetainCycle(*this, capturer, owner);
5481 }
5482 
checkUnsafeAssigns(SourceLocation Loc,QualType LHS,Expr * RHS)5483 bool Sema::checkUnsafeAssigns(SourceLocation Loc,
5484                               QualType LHS, Expr *RHS) {
5485   Qualifiers::ObjCLifetime LT = LHS.getObjCLifetime();
5486   if (LT != Qualifiers::OCL_Weak && LT != Qualifiers::OCL_ExplicitNone)
5487     return false;
5488   // strip off any implicit cast added to get to the one arc-specific
5489   while (ImplicitCastExpr *cast = dyn_cast<ImplicitCastExpr>(RHS)) {
5490     if (cast->getCastKind() == CK_ARCConsumeObject) {
5491       Diag(Loc, diag::warn_arc_retained_assign)
5492         << (LT == Qualifiers::OCL_ExplicitNone) << 1
5493         << RHS->getSourceRange();
5494       return true;
5495     }
5496     RHS = cast->getSubExpr();
5497   }
5498   return false;
5499 }
5500 
checkUnsafeExprAssigns(SourceLocation Loc,Expr * LHS,Expr * RHS)5501 void Sema::checkUnsafeExprAssigns(SourceLocation Loc,
5502                               Expr *LHS, Expr *RHS) {
5503   QualType LHSType;
5504   // PropertyRef on LHS type need be directly obtained from
5505   // its declaration as it has a PsuedoType.
5506   ObjCPropertyRefExpr *PRE
5507     = dyn_cast<ObjCPropertyRefExpr>(LHS->IgnoreParens());
5508   if (PRE && !PRE->isImplicitProperty()) {
5509     const ObjCPropertyDecl *PD = PRE->getExplicitProperty();
5510     if (PD)
5511       LHSType = PD->getType();
5512   }
5513 
5514   if (LHSType.isNull())
5515     LHSType = LHS->getType();
5516   if (checkUnsafeAssigns(Loc, LHSType, RHS))
5517     return;
5518   Qualifiers::ObjCLifetime LT = LHSType.getObjCLifetime();
5519   // FIXME. Check for other life times.
5520   if (LT != Qualifiers::OCL_None)
5521     return;
5522 
5523   if (PRE) {
5524     if (PRE->isImplicitProperty())
5525       return;
5526     const ObjCPropertyDecl *PD = PRE->getExplicitProperty();
5527     if (!PD)
5528       return;
5529 
5530     unsigned Attributes = PD->getPropertyAttributes();
5531     if (Attributes & ObjCPropertyDecl::OBJC_PR_assign) {
5532       // when 'assign' attribute was not explicitly specified
5533       // by user, ignore it and rely on property type itself
5534       // for lifetime info.
5535       unsigned AsWrittenAttr = PD->getPropertyAttributesAsWritten();
5536       if (!(AsWrittenAttr & ObjCPropertyDecl::OBJC_PR_assign) &&
5537           LHSType->isObjCRetainableType())
5538         return;
5539 
5540       while (ImplicitCastExpr *cast = dyn_cast<ImplicitCastExpr>(RHS)) {
5541         if (cast->getCastKind() == CK_ARCConsumeObject) {
5542           Diag(Loc, diag::warn_arc_retained_property_assign)
5543           << RHS->getSourceRange();
5544           return;
5545         }
5546         RHS = cast->getSubExpr();
5547       }
5548     }
5549     else if (Attributes & ObjCPropertyDecl::OBJC_PR_weak) {
5550       while (ImplicitCastExpr *cast = dyn_cast<ImplicitCastExpr>(RHS)) {
5551         if (cast->getCastKind() == CK_ARCConsumeObject) {
5552           Diag(Loc, diag::warn_arc_retained_assign)
5553           << 0 << 0<< RHS->getSourceRange();
5554           return;
5555         }
5556         RHS = cast->getSubExpr();
5557       }
5558     }
5559   }
5560 }
5561 
5562 //===--- CHECK: Empty statement body (-Wempty-body) ---------------------===//
5563 
5564 namespace {
ShouldDiagnoseEmptyStmtBody(const SourceManager & SourceMgr,SourceLocation StmtLoc,const NullStmt * Body)5565 bool ShouldDiagnoseEmptyStmtBody(const SourceManager &SourceMgr,
5566                                  SourceLocation StmtLoc,
5567                                  const NullStmt *Body) {
5568   // Do not warn if the body is a macro that expands to nothing, e.g:
5569   //
5570   // #define CALL(x)
5571   // if (condition)
5572   //   CALL(0);
5573   //
5574   if (Body->hasLeadingEmptyMacro())
5575     return false;
5576 
5577   // Get line numbers of statement and body.
5578   bool StmtLineInvalid;
5579   unsigned StmtLine = SourceMgr.getSpellingLineNumber(StmtLoc,
5580                                                       &StmtLineInvalid);
5581   if (StmtLineInvalid)
5582     return false;
5583 
5584   bool BodyLineInvalid;
5585   unsigned BodyLine = SourceMgr.getSpellingLineNumber(Body->getSemiLoc(),
5586                                                       &BodyLineInvalid);
5587   if (BodyLineInvalid)
5588     return false;
5589 
5590   // Warn if null statement and body are on the same line.
5591   if (StmtLine != BodyLine)
5592     return false;
5593 
5594   return true;
5595 }
5596 } // Unnamed namespace
5597 
DiagnoseEmptyStmtBody(SourceLocation StmtLoc,const Stmt * Body,unsigned DiagID)5598 void Sema::DiagnoseEmptyStmtBody(SourceLocation StmtLoc,
5599                                  const Stmt *Body,
5600                                  unsigned DiagID) {
5601   // Since this is a syntactic check, don't emit diagnostic for template
5602   // instantiations, this just adds noise.
5603   if (CurrentInstantiationScope)
5604     return;
5605 
5606   // The body should be a null statement.
5607   const NullStmt *NBody = dyn_cast<NullStmt>(Body);
5608   if (!NBody)
5609     return;
5610 
5611   // Do the usual checks.
5612   if (!ShouldDiagnoseEmptyStmtBody(SourceMgr, StmtLoc, NBody))
5613     return;
5614 
5615   Diag(NBody->getSemiLoc(), DiagID);
5616   Diag(NBody->getSemiLoc(), diag::note_empty_body_on_separate_line);
5617 }
5618 
DiagnoseEmptyLoopBody(const Stmt * S,const Stmt * PossibleBody)5619 void Sema::DiagnoseEmptyLoopBody(const Stmt *S,
5620                                  const Stmt *PossibleBody) {
5621   assert(!CurrentInstantiationScope); // Ensured by caller
5622 
5623   SourceLocation StmtLoc;
5624   const Stmt *Body;
5625   unsigned DiagID;
5626   if (const ForStmt *FS = dyn_cast<ForStmt>(S)) {
5627     StmtLoc = FS->getRParenLoc();
5628     Body = FS->getBody();
5629     DiagID = diag::warn_empty_for_body;
5630   } else if (const WhileStmt *WS = dyn_cast<WhileStmt>(S)) {
5631     StmtLoc = WS->getCond()->getSourceRange().getEnd();
5632     Body = WS->getBody();
5633     DiagID = diag::warn_empty_while_body;
5634   } else
5635     return; // Neither `for' nor `while'.
5636 
5637   // The body should be a null statement.
5638   const NullStmt *NBody = dyn_cast<NullStmt>(Body);
5639   if (!NBody)
5640     return;
5641 
5642   // Skip expensive checks if diagnostic is disabled.
5643   if (Diags.getDiagnosticLevel(DiagID, NBody->getSemiLoc()) ==
5644           DiagnosticsEngine::Ignored)
5645     return;
5646 
5647   // Do the usual checks.
5648   if (!ShouldDiagnoseEmptyStmtBody(SourceMgr, StmtLoc, NBody))
5649     return;
5650 
5651   // `for(...);' and `while(...);' are popular idioms, so in order to keep
5652   // noise level low, emit diagnostics only if for/while is followed by a
5653   // CompoundStmt, e.g.:
5654   //    for (int i = 0; i < n; i++);
5655   //    {
5656   //      a(i);
5657   //    }
5658   // or if for/while is followed by a statement with more indentation
5659   // than for/while itself:
5660   //    for (int i = 0; i < n; i++);
5661   //      a(i);
5662   bool ProbableTypo = isa<CompoundStmt>(PossibleBody);
5663   if (!ProbableTypo) {
5664     bool BodyColInvalid;
5665     unsigned BodyCol = SourceMgr.getPresumedColumnNumber(
5666                              PossibleBody->getLocStart(),
5667                              &BodyColInvalid);
5668     if (BodyColInvalid)
5669       return;
5670 
5671     bool StmtColInvalid;
5672     unsigned StmtCol = SourceMgr.getPresumedColumnNumber(
5673                              S->getLocStart(),
5674                              &StmtColInvalid);
5675     if (StmtColInvalid)
5676       return;
5677 
5678     if (BodyCol > StmtCol)
5679       ProbableTypo = true;
5680   }
5681 
5682   if (ProbableTypo) {
5683     Diag(NBody->getSemiLoc(), DiagID);
5684     Diag(NBody->getSemiLoc(), diag::note_empty_body_on_separate_line);
5685   }
5686 }
5687 
5688 //===--- Layout compatibility ----------------------------------------------//
5689 
5690 namespace {
5691 
5692 bool isLayoutCompatible(ASTContext &C, QualType T1, QualType T2);
5693 
5694 /// \brief Check if two enumeration types are layout-compatible.
isLayoutCompatible(ASTContext & C,EnumDecl * ED1,EnumDecl * ED2)5695 bool isLayoutCompatible(ASTContext &C, EnumDecl *ED1, EnumDecl *ED2) {
5696   // C++11 [dcl.enum] p8:
5697   // Two enumeration types are layout-compatible if they have the same
5698   // underlying type.
5699   return ED1->isComplete() && ED2->isComplete() &&
5700          C.hasSameType(ED1->getIntegerType(), ED2->getIntegerType());
5701 }
5702 
5703 /// \brief Check if two fields are layout-compatible.
isLayoutCompatible(ASTContext & C,FieldDecl * Field1,FieldDecl * Field2)5704 bool isLayoutCompatible(ASTContext &C, FieldDecl *Field1, FieldDecl *Field2) {
5705   if (!isLayoutCompatible(C, Field1->getType(), Field2->getType()))
5706     return false;
5707 
5708   if (Field1->isBitField() != Field2->isBitField())
5709     return false;
5710 
5711   if (Field1->isBitField()) {
5712     // Make sure that the bit-fields are the same length.
5713     unsigned Bits1 = Field1->getBitWidthValue(C);
5714     unsigned Bits2 = Field2->getBitWidthValue(C);
5715 
5716     if (Bits1 != Bits2)
5717       return false;
5718   }
5719 
5720   return true;
5721 }
5722 
5723 /// \brief Check if two standard-layout structs are layout-compatible.
5724 /// (C++11 [class.mem] p17)
isLayoutCompatibleStruct(ASTContext & C,RecordDecl * RD1,RecordDecl * RD2)5725 bool isLayoutCompatibleStruct(ASTContext &C,
5726                               RecordDecl *RD1,
5727                               RecordDecl *RD2) {
5728   // If both records are C++ classes, check that base classes match.
5729   if (const CXXRecordDecl *D1CXX = dyn_cast<CXXRecordDecl>(RD1)) {
5730     // If one of records is a CXXRecordDecl we are in C++ mode,
5731     // thus the other one is a CXXRecordDecl, too.
5732     const CXXRecordDecl *D2CXX = cast<CXXRecordDecl>(RD2);
5733     // Check number of base classes.
5734     if (D1CXX->getNumBases() != D2CXX->getNumBases())
5735       return false;
5736 
5737     // Check the base classes.
5738     for (CXXRecordDecl::base_class_const_iterator
5739                Base1 = D1CXX->bases_begin(),
5740            BaseEnd1 = D1CXX->bases_end(),
5741               Base2 = D2CXX->bases_begin();
5742          Base1 != BaseEnd1;
5743          ++Base1, ++Base2) {
5744       if (!isLayoutCompatible(C, Base1->getType(), Base2->getType()))
5745         return false;
5746     }
5747   } else if (const CXXRecordDecl *D2CXX = dyn_cast<CXXRecordDecl>(RD2)) {
5748     // If only RD2 is a C++ class, it should have zero base classes.
5749     if (D2CXX->getNumBases() > 0)
5750       return false;
5751   }
5752 
5753   // Check the fields.
5754   RecordDecl::field_iterator Field2 = RD2->field_begin(),
5755                              Field2End = RD2->field_end(),
5756                              Field1 = RD1->field_begin(),
5757                              Field1End = RD1->field_end();
5758   for ( ; Field1 != Field1End && Field2 != Field2End; ++Field1, ++Field2) {
5759     if (!isLayoutCompatible(C, *Field1, *Field2))
5760       return false;
5761   }
5762   if (Field1 != Field1End || Field2 != Field2End)
5763     return false;
5764 
5765   return true;
5766 }
5767 
5768 /// \brief Check if two standard-layout unions are layout-compatible.
5769 /// (C++11 [class.mem] p18)
isLayoutCompatibleUnion(ASTContext & C,RecordDecl * RD1,RecordDecl * RD2)5770 bool isLayoutCompatibleUnion(ASTContext &C,
5771                              RecordDecl *RD1,
5772                              RecordDecl *RD2) {
5773   llvm::SmallPtrSet<FieldDecl *, 8> UnmatchedFields;
5774   for (RecordDecl::field_iterator Field2 = RD2->field_begin(),
5775                                   Field2End = RD2->field_end();
5776        Field2 != Field2End; ++Field2) {
5777     UnmatchedFields.insert(*Field2);
5778   }
5779 
5780   for (RecordDecl::field_iterator Field1 = RD1->field_begin(),
5781                                   Field1End = RD1->field_end();
5782        Field1 != Field1End; ++Field1) {
5783     llvm::SmallPtrSet<FieldDecl *, 8>::iterator
5784         I = UnmatchedFields.begin(),
5785         E = UnmatchedFields.end();
5786 
5787     for ( ; I != E; ++I) {
5788       if (isLayoutCompatible(C, *Field1, *I)) {
5789         bool Result = UnmatchedFields.erase(*I);
5790         (void) Result;
5791         assert(Result);
5792         break;
5793       }
5794     }
5795     if (I == E)
5796       return false;
5797   }
5798 
5799   return UnmatchedFields.empty();
5800 }
5801 
isLayoutCompatible(ASTContext & C,RecordDecl * RD1,RecordDecl * RD2)5802 bool isLayoutCompatible(ASTContext &C, RecordDecl *RD1, RecordDecl *RD2) {
5803   if (RD1->isUnion() != RD2->isUnion())
5804     return false;
5805 
5806   if (RD1->isUnion())
5807     return isLayoutCompatibleUnion(C, RD1, RD2);
5808   else
5809     return isLayoutCompatibleStruct(C, RD1, RD2);
5810 }
5811 
5812 /// \brief Check if two types are layout-compatible in C++11 sense.
isLayoutCompatible(ASTContext & C,QualType T1,QualType T2)5813 bool isLayoutCompatible(ASTContext &C, QualType T1, QualType T2) {
5814   if (T1.isNull() || T2.isNull())
5815     return false;
5816 
5817   // C++11 [basic.types] p11:
5818   // If two types T1 and T2 are the same type, then T1 and T2 are
5819   // layout-compatible types.
5820   if (C.hasSameType(T1, T2))
5821     return true;
5822 
5823   T1 = T1.getCanonicalType().getUnqualifiedType();
5824   T2 = T2.getCanonicalType().getUnqualifiedType();
5825 
5826   const Type::TypeClass TC1 = T1->getTypeClass();
5827   const Type::TypeClass TC2 = T2->getTypeClass();
5828 
5829   if (TC1 != TC2)
5830     return false;
5831 
5832   if (TC1 == Type::Enum) {
5833     return isLayoutCompatible(C,
5834                               cast<EnumType>(T1)->getDecl(),
5835                               cast<EnumType>(T2)->getDecl());
5836   } else if (TC1 == Type::Record) {
5837     if (!T1->isStandardLayoutType() || !T2->isStandardLayoutType())
5838       return false;
5839 
5840     return isLayoutCompatible(C,
5841                               cast<RecordType>(T1)->getDecl(),
5842                               cast<RecordType>(T2)->getDecl());
5843   }
5844 
5845   return false;
5846 }
5847 }
5848 
5849 //===--- CHECK: pointer_with_type_tag attribute: datatypes should match ----//
5850 
5851 namespace {
5852 /// \brief Given a type tag expression find the type tag itself.
5853 ///
5854 /// \param TypeExpr Type tag expression, as it appears in user's code.
5855 ///
5856 /// \param VD Declaration of an identifier that appears in a type tag.
5857 ///
5858 /// \param MagicValue Type tag magic value.
FindTypeTagExpr(const Expr * TypeExpr,const ASTContext & Ctx,const ValueDecl ** VD,uint64_t * MagicValue)5859 bool FindTypeTagExpr(const Expr *TypeExpr, const ASTContext &Ctx,
5860                      const ValueDecl **VD, uint64_t *MagicValue) {
5861   while(true) {
5862     if (!TypeExpr)
5863       return false;
5864 
5865     TypeExpr = TypeExpr->IgnoreParenImpCasts()->IgnoreParenCasts();
5866 
5867     switch (TypeExpr->getStmtClass()) {
5868     case Stmt::UnaryOperatorClass: {
5869       const UnaryOperator *UO = cast<UnaryOperator>(TypeExpr);
5870       if (UO->getOpcode() == UO_AddrOf || UO->getOpcode() == UO_Deref) {
5871         TypeExpr = UO->getSubExpr();
5872         continue;
5873       }
5874       return false;
5875     }
5876 
5877     case Stmt::DeclRefExprClass: {
5878       const DeclRefExpr *DRE = cast<DeclRefExpr>(TypeExpr);
5879       *VD = DRE->getDecl();
5880       return true;
5881     }
5882 
5883     case Stmt::IntegerLiteralClass: {
5884       const IntegerLiteral *IL = cast<IntegerLiteral>(TypeExpr);
5885       llvm::APInt MagicValueAPInt = IL->getValue();
5886       if (MagicValueAPInt.getActiveBits() <= 64) {
5887         *MagicValue = MagicValueAPInt.getZExtValue();
5888         return true;
5889       } else
5890         return false;
5891     }
5892 
5893     case Stmt::BinaryConditionalOperatorClass:
5894     case Stmt::ConditionalOperatorClass: {
5895       const AbstractConditionalOperator *ACO =
5896           cast<AbstractConditionalOperator>(TypeExpr);
5897       bool Result;
5898       if (ACO->getCond()->EvaluateAsBooleanCondition(Result, Ctx)) {
5899         if (Result)
5900           TypeExpr = ACO->getTrueExpr();
5901         else
5902           TypeExpr = ACO->getFalseExpr();
5903         continue;
5904       }
5905       return false;
5906     }
5907 
5908     case Stmt::BinaryOperatorClass: {
5909       const BinaryOperator *BO = cast<BinaryOperator>(TypeExpr);
5910       if (BO->getOpcode() == BO_Comma) {
5911         TypeExpr = BO->getRHS();
5912         continue;
5913       }
5914       return false;
5915     }
5916 
5917     default:
5918       return false;
5919     }
5920   }
5921 }
5922 
5923 /// \brief Retrieve the C type corresponding to type tag TypeExpr.
5924 ///
5925 /// \param TypeExpr Expression that specifies a type tag.
5926 ///
5927 /// \param MagicValues Registered magic values.
5928 ///
5929 /// \param FoundWrongKind Set to true if a type tag was found, but of a wrong
5930 ///        kind.
5931 ///
5932 /// \param TypeInfo Information about the corresponding C type.
5933 ///
5934 /// \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)5935 bool GetMatchingCType(
5936         const IdentifierInfo *ArgumentKind,
5937         const Expr *TypeExpr, const ASTContext &Ctx,
5938         const llvm::DenseMap<Sema::TypeTagMagicValue,
5939                              Sema::TypeTagData> *MagicValues,
5940         bool &FoundWrongKind,
5941         Sema::TypeTagData &TypeInfo) {
5942   FoundWrongKind = false;
5943 
5944   // Variable declaration that has type_tag_for_datatype attribute.
5945   const ValueDecl *VD = NULL;
5946 
5947   uint64_t MagicValue;
5948 
5949   if (!FindTypeTagExpr(TypeExpr, Ctx, &VD, &MagicValue))
5950     return false;
5951 
5952   if (VD) {
5953     for (specific_attr_iterator<TypeTagForDatatypeAttr>
5954              I = VD->specific_attr_begin<TypeTagForDatatypeAttr>(),
5955              E = VD->specific_attr_end<TypeTagForDatatypeAttr>();
5956          I != E; ++I) {
5957       if (I->getArgumentKind() != ArgumentKind) {
5958         FoundWrongKind = true;
5959         return false;
5960       }
5961       TypeInfo.Type = I->getMatchingCType();
5962       TypeInfo.LayoutCompatible = I->getLayoutCompatible();
5963       TypeInfo.MustBeNull = I->getMustBeNull();
5964       return true;
5965     }
5966     return false;
5967   }
5968 
5969   if (!MagicValues)
5970     return false;
5971 
5972   llvm::DenseMap<Sema::TypeTagMagicValue,
5973                  Sema::TypeTagData>::const_iterator I =
5974       MagicValues->find(std::make_pair(ArgumentKind, MagicValue));
5975   if (I == MagicValues->end())
5976     return false;
5977 
5978   TypeInfo = I->second;
5979   return true;
5980 }
5981 } // unnamed namespace
5982 
RegisterTypeTagForDatatype(const IdentifierInfo * ArgumentKind,uint64_t MagicValue,QualType Type,bool LayoutCompatible,bool MustBeNull)5983 void Sema::RegisterTypeTagForDatatype(const IdentifierInfo *ArgumentKind,
5984                                       uint64_t MagicValue, QualType Type,
5985                                       bool LayoutCompatible,
5986                                       bool MustBeNull) {
5987   if (!TypeTagForDatatypeMagicValues)
5988     TypeTagForDatatypeMagicValues.reset(
5989         new llvm::DenseMap<TypeTagMagicValue, TypeTagData>);
5990 
5991   TypeTagMagicValue Magic(ArgumentKind, MagicValue);
5992   (*TypeTagForDatatypeMagicValues)[Magic] =
5993       TypeTagData(Type, LayoutCompatible, MustBeNull);
5994 }
5995 
5996 namespace {
IsSameCharType(QualType T1,QualType T2)5997 bool IsSameCharType(QualType T1, QualType T2) {
5998   const BuiltinType *BT1 = T1->getAs<BuiltinType>();
5999   if (!BT1)
6000     return false;
6001 
6002   const BuiltinType *BT2 = T2->getAs<BuiltinType>();
6003   if (!BT2)
6004     return false;
6005 
6006   BuiltinType::Kind T1Kind = BT1->getKind();
6007   BuiltinType::Kind T2Kind = BT2->getKind();
6008 
6009   return (T1Kind == BuiltinType::SChar  && T2Kind == BuiltinType::Char_S) ||
6010          (T1Kind == BuiltinType::UChar  && T2Kind == BuiltinType::Char_U) ||
6011          (T1Kind == BuiltinType::Char_U && T2Kind == BuiltinType::UChar) ||
6012          (T1Kind == BuiltinType::Char_S && T2Kind == BuiltinType::SChar);
6013 }
6014 } // unnamed namespace
6015 
CheckArgumentWithTypeTag(const ArgumentWithTypeTagAttr * Attr,const Expr * const * ExprArgs)6016 void Sema::CheckArgumentWithTypeTag(const ArgumentWithTypeTagAttr *Attr,
6017                                     const Expr * const *ExprArgs) {
6018   const IdentifierInfo *ArgumentKind = Attr->getArgumentKind();
6019   bool IsPointerAttr = Attr->getIsPointer();
6020 
6021   const Expr *TypeTagExpr = ExprArgs[Attr->getTypeTagIdx()];
6022   bool FoundWrongKind;
6023   TypeTagData TypeInfo;
6024   if (!GetMatchingCType(ArgumentKind, TypeTagExpr, Context,
6025                         TypeTagForDatatypeMagicValues.get(),
6026                         FoundWrongKind, TypeInfo)) {
6027     if (FoundWrongKind)
6028       Diag(TypeTagExpr->getExprLoc(),
6029            diag::warn_type_tag_for_datatype_wrong_kind)
6030         << TypeTagExpr->getSourceRange();
6031     return;
6032   }
6033 
6034   const Expr *ArgumentExpr = ExprArgs[Attr->getArgumentIdx()];
6035   if (IsPointerAttr) {
6036     // Skip implicit cast of pointer to `void *' (as a function argument).
6037     if (const ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(ArgumentExpr))
6038       if (ICE->getType()->isVoidPointerType())
6039         ArgumentExpr = ICE->getSubExpr();
6040   }
6041   QualType ArgumentType = ArgumentExpr->getType();
6042 
6043   // Passing a `void*' pointer shouldn't trigger a warning.
6044   if (IsPointerAttr && ArgumentType->isVoidPointerType())
6045     return;
6046 
6047   if (TypeInfo.MustBeNull) {
6048     // Type tag with matching void type requires a null pointer.
6049     if (!ArgumentExpr->isNullPointerConstant(Context,
6050                                              Expr::NPC_ValueDependentIsNotNull)) {
6051       Diag(ArgumentExpr->getExprLoc(),
6052            diag::warn_type_safety_null_pointer_required)
6053           << ArgumentKind->getName()
6054           << ArgumentExpr->getSourceRange()
6055           << TypeTagExpr->getSourceRange();
6056     }
6057     return;
6058   }
6059 
6060   QualType RequiredType = TypeInfo.Type;
6061   if (IsPointerAttr)
6062     RequiredType = Context.getPointerType(RequiredType);
6063 
6064   bool mismatch = false;
6065   if (!TypeInfo.LayoutCompatible) {
6066     mismatch = !Context.hasSameType(ArgumentType, RequiredType);
6067 
6068     // C++11 [basic.fundamental] p1:
6069     // Plain char, signed char, and unsigned char are three distinct types.
6070     //
6071     // But we treat plain `char' as equivalent to `signed char' or `unsigned
6072     // char' depending on the current char signedness mode.
6073     if (mismatch)
6074       if ((IsPointerAttr && IsSameCharType(ArgumentType->getPointeeType(),
6075                                            RequiredType->getPointeeType())) ||
6076           (!IsPointerAttr && IsSameCharType(ArgumentType, RequiredType)))
6077         mismatch = false;
6078   } else
6079     if (IsPointerAttr)
6080       mismatch = !isLayoutCompatible(Context,
6081                                      ArgumentType->getPointeeType(),
6082                                      RequiredType->getPointeeType());
6083     else
6084       mismatch = !isLayoutCompatible(Context, ArgumentType, RequiredType);
6085 
6086   if (mismatch)
6087     Diag(ArgumentExpr->getExprLoc(), diag::warn_type_safety_type_mismatch)
6088         << ArgumentType << ArgumentKind->getName()
6089         << TypeInfo.LayoutCompatible << RequiredType
6090         << ArgumentExpr->getSourceRange()
6091         << TypeTagExpr->getSourceRange();
6092 }
6093 
6094