SemaChecking.cpp revision 3d692df4b9c58895f9843b03543ec57447c93679
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 "Sema.h" 16#include "clang/AST/ASTContext.h" 17#include "clang/AST/DeclObjC.h" 18#include "clang/AST/ExprCXX.h" 19#include "clang/AST/ExprObjC.h" 20#include "clang/Lex/LiteralSupport.h" 21#include "clang/Lex/Preprocessor.h" 22using namespace clang; 23 24/// getLocationOfStringLiteralByte - Return a source location that points to the 25/// specified byte of the specified string literal. 26/// 27/// Strings are amazingly complex. They can be formed from multiple tokens and 28/// can have escape sequences in them in addition to the usual trigraph and 29/// escaped newline business. This routine handles this complexity. 30/// 31SourceLocation Sema::getLocationOfStringLiteralByte(const StringLiteral *SL, 32 unsigned ByteNo) const { 33 assert(!SL->isWide() && "This doesn't work for wide strings yet"); 34 35 // Loop over all of the tokens in this string until we find the one that 36 // contains the byte we're looking for. 37 unsigned TokNo = 0; 38 while (1) { 39 assert(TokNo < SL->getNumConcatenated() && "Invalid byte number!"); 40 SourceLocation StrTokLoc = SL->getStrTokenLoc(TokNo); 41 42 // Get the spelling of the string so that we can get the data that makes up 43 // the string literal, not the identifier for the macro it is potentially 44 // expanded through. 45 SourceLocation StrTokSpellingLoc = SourceMgr.getSpellingLoc(StrTokLoc); 46 47 // Re-lex the token to get its length and original spelling. 48 std::pair<FileID, unsigned> LocInfo = 49 SourceMgr.getDecomposedLoc(StrTokSpellingLoc); 50 std::pair<const char *,const char *> Buffer = 51 SourceMgr.getBufferData(LocInfo.first); 52 const char *StrData = Buffer.first+LocInfo.second; 53 54 // Create a langops struct and enable trigraphs. This is sufficient for 55 // relexing tokens. 56 LangOptions LangOpts; 57 LangOpts.Trigraphs = true; 58 59 // Create a lexer starting at the beginning of this token. 60 Lexer TheLexer(StrTokSpellingLoc, LangOpts, Buffer.first, StrData, 61 Buffer.second); 62 Token TheTok; 63 TheLexer.LexFromRawLexer(TheTok); 64 65 // Use the StringLiteralParser to compute the length of the string in bytes. 66 StringLiteralParser SLP(&TheTok, 1, PP); 67 unsigned TokNumBytes = SLP.GetStringLength(); 68 69 // If the byte is in this token, return the location of the byte. 70 if (ByteNo < TokNumBytes || 71 (ByteNo == TokNumBytes && TokNo == SL->getNumConcatenated())) { 72 unsigned Offset = 73 StringLiteralParser::getOffsetOfStringByte(TheTok, ByteNo, PP); 74 75 // Now that we know the offset of the token in the spelling, use the 76 // preprocessor to get the offset in the original source. 77 return PP.AdvanceToTokenCharacter(StrTokLoc, Offset); 78 } 79 80 // Move to the next string token. 81 ++TokNo; 82 ByteNo -= TokNumBytes; 83 } 84} 85 86 87/// CheckFunctionCall - Check a direct function call for various correctness 88/// and safety properties not strictly enforced by the C type system. 89Action::OwningExprResult 90Sema::CheckFunctionCall(FunctionDecl *FDecl, CallExpr *TheCall) { 91 OwningExprResult TheCallResult(Owned(TheCall)); 92 // Get the IdentifierInfo* for the called function. 93 IdentifierInfo *FnInfo = FDecl->getIdentifier(); 94 95 // None of the checks below are needed for functions that don't have 96 // simple names (e.g., C++ conversion functions). 97 if (!FnInfo) 98 return move(TheCallResult); 99 100 switch (FDecl->getBuiltinID(Context)) { 101 case Builtin::BI__builtin___CFStringMakeConstantString: 102 assert(TheCall->getNumArgs() == 1 && 103 "Wrong # arguments to builtin CFStringMakeConstantString"); 104 if (CheckObjCString(TheCall->getArg(0))) 105 return ExprError(); 106 return move(TheCallResult); 107 case Builtin::BI__builtin_stdarg_start: 108 case Builtin::BI__builtin_va_start: 109 if (SemaBuiltinVAStart(TheCall)) 110 return ExprError(); 111 return move(TheCallResult); 112 case Builtin::BI__builtin_isgreater: 113 case Builtin::BI__builtin_isgreaterequal: 114 case Builtin::BI__builtin_isless: 115 case Builtin::BI__builtin_islessequal: 116 case Builtin::BI__builtin_islessgreater: 117 case Builtin::BI__builtin_isunordered: 118 if (SemaBuiltinUnorderedCompare(TheCall)) 119 return ExprError(); 120 return move(TheCallResult); 121 case Builtin::BI__builtin_return_address: 122 case Builtin::BI__builtin_frame_address: 123 if (SemaBuiltinStackAddress(TheCall)) 124 return ExprError(); 125 return move(TheCallResult); 126 case Builtin::BI__builtin_shufflevector: 127 return SemaBuiltinShuffleVector(TheCall); 128 // TheCall will be freed by the smart pointer here, but that's fine, since 129 // SemaBuiltinShuffleVector guts it, but then doesn't release it. 130 case Builtin::BI__builtin_prefetch: 131 if (SemaBuiltinPrefetch(TheCall)) 132 return ExprError(); 133 return move(TheCallResult); 134 case Builtin::BI__builtin_object_size: 135 if (SemaBuiltinObjectSize(TheCall)) 136 return ExprError(); 137 } 138 139 // FIXME: This mechanism should be abstracted to be less fragile and 140 // more efficient. For example, just map function ids to custom 141 // handlers. 142 143 // Printf checking. 144 if (const FormatAttr *Format = FDecl->getAttr<FormatAttr>()) { 145 if (Format->getType() == "printf") { 146 bool HasVAListArg = Format->getFirstArg() == 0; 147 if (!HasVAListArg) { 148 if (const FunctionProtoType *Proto 149 = FDecl->getType()->getAsFunctionProtoType()) 150 HasVAListArg = !Proto->isVariadic(); 151 } 152 CheckPrintfArguments(TheCall, HasVAListArg, Format->getFormatIdx() - 1, 153 HasVAListArg ? 0 : Format->getFirstArg() - 1); 154 } 155 } 156 157 return move(TheCallResult); 158} 159 160/// CheckObjCString - Checks that the argument to the builtin 161/// CFString constructor is correct 162bool Sema::CheckObjCString(Expr *Arg) { 163 Arg = Arg->IgnoreParenCasts(); 164 StringLiteral *Literal = dyn_cast<StringLiteral>(Arg); 165 166 if (!Literal || Literal->isWide()) { 167 Diag(Arg->getLocStart(), diag::err_cfstring_literal_not_string_constant) 168 << Arg->getSourceRange(); 169 return true; 170 } 171 172 const char *Data = Literal->getStrData(); 173 unsigned Length = Literal->getByteLength(); 174 175 for (unsigned i = 0; i < Length; ++i) { 176 if (!isascii(Data[i])) { 177 Diag(getLocationOfStringLiteralByte(Literal, i), 178 diag::warn_cfstring_literal_contains_non_ascii_character) 179 << Arg->getSourceRange(); 180 break; 181 } 182 183 if (!Data[i]) { 184 Diag(getLocationOfStringLiteralByte(Literal, i), 185 diag::warn_cfstring_literal_contains_nul_character) 186 << Arg->getSourceRange(); 187 break; 188 } 189 } 190 191 return false; 192} 193 194/// SemaBuiltinVAStart - Check the arguments to __builtin_va_start for validity. 195/// Emit an error and return true on failure, return false on success. 196bool Sema::SemaBuiltinVAStart(CallExpr *TheCall) { 197 Expr *Fn = TheCall->getCallee(); 198 if (TheCall->getNumArgs() > 2) { 199 Diag(TheCall->getArg(2)->getLocStart(), 200 diag::err_typecheck_call_too_many_args) 201 << 0 /*function call*/ << Fn->getSourceRange() 202 << SourceRange(TheCall->getArg(2)->getLocStart(), 203 (*(TheCall->arg_end()-1))->getLocEnd()); 204 return true; 205 } 206 207 if (TheCall->getNumArgs() < 2) { 208 return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args) 209 << 0 /*function call*/; 210 } 211 212 // Determine whether the current function is variadic or not. 213 bool isVariadic; 214 if (getCurFunctionDecl()) { 215 if (FunctionProtoType* FTP = 216 dyn_cast<FunctionProtoType>(getCurFunctionDecl()->getType())) 217 isVariadic = FTP->isVariadic(); 218 else 219 isVariadic = false; 220 } else { 221 isVariadic = getCurMethodDecl()->isVariadic(); 222 } 223 224 if (!isVariadic) { 225 Diag(Fn->getLocStart(), diag::err_va_start_used_in_non_variadic_function); 226 return true; 227 } 228 229 // Verify that the second argument to the builtin is the last argument of the 230 // current function or method. 231 bool SecondArgIsLastNamedArgument = false; 232 const Expr *Arg = TheCall->getArg(1)->IgnoreParenCasts(); 233 234 if (const DeclRefExpr *DR = dyn_cast<DeclRefExpr>(Arg)) { 235 if (const ParmVarDecl *PV = dyn_cast<ParmVarDecl>(DR->getDecl())) { 236 // FIXME: This isn't correct for methods (results in bogus warning). 237 // Get the last formal in the current function. 238 const ParmVarDecl *LastArg; 239 if (FunctionDecl *FD = getCurFunctionDecl()) 240 LastArg = *(FD->param_end()-1); 241 else 242 LastArg = *(getCurMethodDecl()->param_end()-1); 243 SecondArgIsLastNamedArgument = PV == LastArg; 244 } 245 } 246 247 if (!SecondArgIsLastNamedArgument) 248 Diag(TheCall->getArg(1)->getLocStart(), 249 diag::warn_second_parameter_of_va_start_not_last_named_argument); 250 return false; 251} 252 253/// SemaBuiltinUnorderedCompare - Handle functions like __builtin_isgreater and 254/// friends. This is declared to take (...), so we have to check everything. 255bool Sema::SemaBuiltinUnorderedCompare(CallExpr *TheCall) { 256 if (TheCall->getNumArgs() < 2) 257 return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args) 258 << 0 /*function call*/; 259 if (TheCall->getNumArgs() > 2) 260 return Diag(TheCall->getArg(2)->getLocStart(), 261 diag::err_typecheck_call_too_many_args) 262 << 0 /*function call*/ 263 << SourceRange(TheCall->getArg(2)->getLocStart(), 264 (*(TheCall->arg_end()-1))->getLocEnd()); 265 266 Expr *OrigArg0 = TheCall->getArg(0); 267 Expr *OrigArg1 = TheCall->getArg(1); 268 269 // Do standard promotions between the two arguments, returning their common 270 // type. 271 QualType Res = UsualArithmeticConversions(OrigArg0, OrigArg1, false); 272 273 // Make sure any conversions are pushed back into the call; this is 274 // type safe since unordered compare builtins are declared as "_Bool 275 // foo(...)". 276 TheCall->setArg(0, OrigArg0); 277 TheCall->setArg(1, OrigArg1); 278 279 // If the common type isn't a real floating type, then the arguments were 280 // invalid for this operation. 281 if (!Res->isRealFloatingType()) 282 return Diag(OrigArg0->getLocStart(), 283 diag::err_typecheck_call_invalid_ordered_compare) 284 << OrigArg0->getType() << OrigArg1->getType() 285 << SourceRange(OrigArg0->getLocStart(), OrigArg1->getLocEnd()); 286 287 return false; 288} 289 290bool Sema::SemaBuiltinStackAddress(CallExpr *TheCall) { 291 // The signature for these builtins is exact; the only thing we need 292 // to check is that the argument is a constant. 293 SourceLocation Loc; 294 if (!TheCall->getArg(0)->isIntegerConstantExpr(Context, &Loc)) 295 return Diag(Loc, diag::err_stack_const_level) << TheCall->getSourceRange(); 296 297 return false; 298} 299 300/// SemaBuiltinShuffleVector - Handle __builtin_shufflevector. 301// This is declared to take (...), so we have to check everything. 302Action::OwningExprResult Sema::SemaBuiltinShuffleVector(CallExpr *TheCall) { 303 if (TheCall->getNumArgs() < 3) 304 return ExprError(Diag(TheCall->getLocEnd(), 305 diag::err_typecheck_call_too_few_args) 306 << 0 /*function call*/ << TheCall->getSourceRange()); 307 308 QualType FAType = TheCall->getArg(0)->getType(); 309 QualType SAType = TheCall->getArg(1)->getType(); 310 311 if (!FAType->isVectorType() || !SAType->isVectorType()) { 312 Diag(TheCall->getLocStart(), diag::err_shufflevector_non_vector) 313 << SourceRange(TheCall->getArg(0)->getLocStart(), 314 TheCall->getArg(1)->getLocEnd()); 315 return ExprError(); 316 } 317 318 if (Context.getCanonicalType(FAType).getUnqualifiedType() != 319 Context.getCanonicalType(SAType).getUnqualifiedType()) { 320 Diag(TheCall->getLocStart(), diag::err_shufflevector_incompatible_vector) 321 << SourceRange(TheCall->getArg(0)->getLocStart(), 322 TheCall->getArg(1)->getLocEnd()); 323 return ExprError(); 324 } 325 326 unsigned numElements = FAType->getAsVectorType()->getNumElements(); 327 if (TheCall->getNumArgs() != numElements+2) { 328 if (TheCall->getNumArgs() < numElements+2) 329 return ExprError(Diag(TheCall->getLocEnd(), 330 diag::err_typecheck_call_too_few_args) 331 << 0 /*function call*/ << TheCall->getSourceRange()); 332 return ExprError(Diag(TheCall->getLocEnd(), 333 diag::err_typecheck_call_too_many_args) 334 << 0 /*function call*/ << TheCall->getSourceRange()); 335 } 336 337 for (unsigned i = 2; i < TheCall->getNumArgs(); i++) { 338 llvm::APSInt Result(32); 339 if (!TheCall->getArg(i)->isIntegerConstantExpr(Result, Context)) 340 return ExprError(Diag(TheCall->getLocStart(), 341 diag::err_shufflevector_nonconstant_argument) 342 << TheCall->getArg(i)->getSourceRange()); 343 344 if (Result.getActiveBits() > 64 || Result.getZExtValue() >= numElements*2) 345 return ExprError(Diag(TheCall->getLocStart(), 346 diag::err_shufflevector_argument_too_large) 347 << TheCall->getArg(i)->getSourceRange()); 348 } 349 350 llvm::SmallVector<Expr*, 32> exprs; 351 352 for (unsigned i = 0, e = TheCall->getNumArgs(); i != e; i++) { 353 exprs.push_back(TheCall->getArg(i)); 354 TheCall->setArg(i, 0); 355 } 356 357 return Owned(new (Context) ShuffleVectorExpr(exprs.begin(), numElements+2, 358 FAType, 359 TheCall->getCallee()->getLocStart(), 360 TheCall->getRParenLoc())); 361} 362 363/// SemaBuiltinPrefetch - Handle __builtin_prefetch. 364// This is declared to take (const void*, ...) and can take two 365// optional constant int args. 366bool Sema::SemaBuiltinPrefetch(CallExpr *TheCall) { 367 unsigned NumArgs = TheCall->getNumArgs(); 368 369 if (NumArgs > 3) 370 return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_many_args) 371 << 0 /*function call*/ << TheCall->getSourceRange(); 372 373 // Argument 0 is checked for us and the remaining arguments must be 374 // constant integers. 375 for (unsigned i = 1; i != NumArgs; ++i) { 376 Expr *Arg = TheCall->getArg(i); 377 QualType RWType = Arg->getType(); 378 379 const BuiltinType *BT = RWType->getAsBuiltinType(); 380 llvm::APSInt Result; 381 if (!BT || BT->getKind() != BuiltinType::Int || 382 !Arg->isIntegerConstantExpr(Result, Context)) 383 return Diag(TheCall->getLocStart(), diag::err_prefetch_invalid_argument) 384 << SourceRange(Arg->getLocStart(), Arg->getLocEnd()); 385 386 // FIXME: gcc issues a warning and rewrites these to 0. These 387 // seems especially odd for the third argument since the default 388 // is 3. 389 if (i == 1) { 390 if (Result.getSExtValue() < 0 || Result.getSExtValue() > 1) 391 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 392 << "0" << "1" << SourceRange(Arg->getLocStart(), Arg->getLocEnd()); 393 } else { 394 if (Result.getSExtValue() < 0 || Result.getSExtValue() > 3) 395 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 396 << "0" << "3" << SourceRange(Arg->getLocStart(), Arg->getLocEnd()); 397 } 398 } 399 400 return false; 401} 402 403/// SemaBuiltinObjectSize - Handle __builtin_object_size(void *ptr, 404/// int type). This simply type checks that type is one of the defined 405/// constants (0-3). 406bool Sema::SemaBuiltinObjectSize(CallExpr *TheCall) { 407 Expr *Arg = TheCall->getArg(1); 408 QualType ArgType = Arg->getType(); 409 const BuiltinType *BT = ArgType->getAsBuiltinType(); 410 llvm::APSInt Result(32); 411 if (!BT || BT->getKind() != BuiltinType::Int || 412 !Arg->isIntegerConstantExpr(Result, Context)) { 413 return Diag(TheCall->getLocStart(), diag::err_object_size_invalid_argument) 414 << SourceRange(Arg->getLocStart(), Arg->getLocEnd()); 415 } 416 417 if (Result.getSExtValue() < 0 || Result.getSExtValue() > 3) { 418 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 419 << "0" << "3" << SourceRange(Arg->getLocStart(), Arg->getLocEnd()); 420 } 421 422 return false; 423} 424 425// Handle i > 1 ? "x" : "y", recursivelly 426bool Sema::SemaCheckStringLiteral(Expr *E, CallExpr *TheCall, bool HasVAListArg, 427 unsigned format_idx, unsigned firstDataArg) { 428 429 switch (E->getStmtClass()) { 430 case Stmt::ConditionalOperatorClass: { 431 ConditionalOperator *C = cast<ConditionalOperator>(E); 432 return SemaCheckStringLiteral(C->getLHS(), TheCall, 433 HasVAListArg, format_idx, firstDataArg) 434 && SemaCheckStringLiteral(C->getRHS(), TheCall, 435 HasVAListArg, format_idx, firstDataArg); 436 } 437 438 case Stmt::ImplicitCastExprClass: { 439 ImplicitCastExpr *Expr = dyn_cast<ImplicitCastExpr>(E); 440 return SemaCheckStringLiteral(Expr->getSubExpr(), TheCall, HasVAListArg, 441 format_idx, firstDataArg); 442 } 443 444 case Stmt::ParenExprClass: { 445 ParenExpr *Expr = dyn_cast<ParenExpr>(E); 446 return SemaCheckStringLiteral(Expr->getSubExpr(), TheCall, HasVAListArg, 447 format_idx, firstDataArg); 448 } 449 450 default: { 451 ObjCStringLiteral *ObjCFExpr = dyn_cast<ObjCStringLiteral>(E); 452 StringLiteral *StrE = NULL; 453 454 if (ObjCFExpr) 455 StrE = ObjCFExpr->getString(); 456 else 457 StrE = dyn_cast<StringLiteral>(E); 458 459 if (StrE) { 460 CheckPrintfString(StrE, E, TheCall, HasVAListArg, format_idx, 461 firstDataArg); 462 return true; 463 } 464 465 return false; 466 } 467 } 468} 469 470 471/// CheckPrintfArguments - Check calls to printf (and similar functions) for 472/// correct use of format strings. 473/// 474/// HasVAListArg - A predicate indicating whether the printf-like 475/// function is passed an explicit va_arg argument (e.g., vprintf) 476/// 477/// format_idx - The index into Args for the format string. 478/// 479/// Improper format strings to functions in the printf family can be 480/// the source of bizarre bugs and very serious security holes. A 481/// good source of information is available in the following paper 482/// (which includes additional references): 483/// 484/// FormatGuard: Automatic Protection From printf Format String 485/// Vulnerabilities, Proceedings of the 10th USENIX Security Symposium, 2001. 486/// 487/// Functionality implemented: 488/// 489/// We can statically check the following properties for string 490/// literal format strings for non v.*printf functions (where the 491/// arguments are passed directly): 492// 493/// (1) Are the number of format conversions equal to the number of 494/// data arguments? 495/// 496/// (2) Does each format conversion correctly match the type of the 497/// corresponding data argument? (TODO) 498/// 499/// Moreover, for all printf functions we can: 500/// 501/// (3) Check for a missing format string (when not caught by type checking). 502/// 503/// (4) Check for no-operation flags; e.g. using "#" with format 504/// conversion 'c' (TODO) 505/// 506/// (5) Check the use of '%n', a major source of security holes. 507/// 508/// (6) Check for malformed format conversions that don't specify anything. 509/// 510/// (7) Check for empty format strings. e.g: printf(""); 511/// 512/// (8) Check that the format string is a wide literal. 513/// 514/// (9) Also check the arguments of functions with the __format__ attribute. 515/// (TODO). 516/// 517/// All of these checks can be done by parsing the format string. 518/// 519/// For now, we ONLY do (1), (3), (5), (6), (7), and (8). 520void 521Sema::CheckPrintfArguments(CallExpr *TheCall, bool HasVAListArg, 522 unsigned format_idx, unsigned firstDataArg) { 523 Expr *Fn = TheCall->getCallee(); 524 525 // CHECK: printf-like function is called with no format string. 526 if (format_idx >= TheCall->getNumArgs()) { 527 Diag(TheCall->getRParenLoc(), diag::warn_printf_missing_format_string) 528 << Fn->getSourceRange(); 529 return; 530 } 531 532 Expr *OrigFormatExpr = TheCall->getArg(format_idx)->IgnoreParenCasts(); 533 534 // CHECK: format string is not a string literal. 535 // 536 // Dynamically generated format strings are difficult to 537 // automatically vet at compile time. Requiring that format strings 538 // are string literals: (1) permits the checking of format strings by 539 // the compiler and thereby (2) can practically remove the source of 540 // many format string exploits. 541 542 // Format string can be either ObjC string (e.g. @"%d") or 543 // C string (e.g. "%d") 544 // ObjC string uses the same format specifiers as C string, so we can use 545 // the same format string checking logic for both ObjC and C strings. 546 bool isFExpr = SemaCheckStringLiteral(OrigFormatExpr, TheCall, 547 HasVAListArg, format_idx, 548 firstDataArg); 549 550 if (!isFExpr) { 551 // For vprintf* functions (i.e., HasVAListArg==true), we add a 552 // special check to see if the format string is a function parameter 553 // of the function calling the printf function. If the function 554 // has an attribute indicating it is a printf-like function, then we 555 // should suppress warnings concerning non-literals being used in a call 556 // to a vprintf function. For example: 557 // 558 // void 559 // logmessage(char const *fmt __attribute__ (format (printf, 1, 2)), ...) { 560 // va_list ap; 561 // va_start(ap, fmt); 562 // vprintf(fmt, ap); // Do NOT emit a warning about "fmt". 563 // ... 564 // 565 // 566 // FIXME: We don't have full attribute support yet, so just check to see 567 // if the argument is a DeclRefExpr that references a parameter. We'll 568 // add proper support for checking the attribute later. 569 if (HasVAListArg) 570 if (DeclRefExpr* DR = dyn_cast<DeclRefExpr>(OrigFormatExpr)) 571 if (isa<ParmVarDecl>(DR->getDecl())) 572 return; 573 574 Diag(TheCall->getArg(format_idx)->getLocStart(), 575 diag::warn_printf_not_string_constant) 576 << OrigFormatExpr->getSourceRange(); 577 return; 578 } 579} 580 581void Sema::CheckPrintfString(StringLiteral *FExpr, Expr *OrigFormatExpr, 582 CallExpr *TheCall, bool HasVAListArg, unsigned format_idx, 583 unsigned firstDataArg) { 584 585 ObjCStringLiteral *ObjCFExpr = dyn_cast<ObjCStringLiteral>(OrigFormatExpr); 586 // CHECK: is the format string a wide literal? 587 if (FExpr->isWide()) { 588 Diag(FExpr->getLocStart(), 589 diag::warn_printf_format_string_is_wide_literal) 590 << OrigFormatExpr->getSourceRange(); 591 return; 592 } 593 594 // Str - The format string. NOTE: this is NOT null-terminated! 595 const char * const Str = FExpr->getStrData(); 596 597 // CHECK: empty format string? 598 const unsigned StrLen = FExpr->getByteLength(); 599 600 if (StrLen == 0) { 601 Diag(FExpr->getLocStart(), diag::warn_printf_empty_format_string) 602 << OrigFormatExpr->getSourceRange(); 603 return; 604 } 605 606 // We process the format string using a binary state machine. The 607 // current state is stored in CurrentState. 608 enum { 609 state_OrdChr, 610 state_Conversion 611 } CurrentState = state_OrdChr; 612 613 // numConversions - The number of conversions seen so far. This is 614 // incremented as we traverse the format string. 615 unsigned numConversions = 0; 616 617 // numDataArgs - The number of data arguments after the format 618 // string. This can only be determined for non vprintf-like 619 // functions. For those functions, this value is 1 (the sole 620 // va_arg argument). 621 unsigned numDataArgs = TheCall->getNumArgs()-firstDataArg; 622 623 // Inspect the format string. 624 unsigned StrIdx = 0; 625 626 // LastConversionIdx - Index within the format string where we last saw 627 // a '%' character that starts a new format conversion. 628 unsigned LastConversionIdx = 0; 629 630 for (; StrIdx < StrLen; ++StrIdx) { 631 632 // Is the number of detected conversion conversions greater than 633 // the number of matching data arguments? If so, stop. 634 if (!HasVAListArg && numConversions > numDataArgs) break; 635 636 // Handle "\0" 637 if (Str[StrIdx] == '\0') { 638 // The string returned by getStrData() is not null-terminated, 639 // so the presence of a null character is likely an error. 640 Diag(getLocationOfStringLiteralByte(FExpr, StrIdx), 641 diag::warn_printf_format_string_contains_null_char) 642 << OrigFormatExpr->getSourceRange(); 643 return; 644 } 645 646 // Ordinary characters (not processing a format conversion). 647 if (CurrentState == state_OrdChr) { 648 if (Str[StrIdx] == '%') { 649 CurrentState = state_Conversion; 650 LastConversionIdx = StrIdx; 651 } 652 continue; 653 } 654 655 // Seen '%'. Now processing a format conversion. 656 switch (Str[StrIdx]) { 657 // Handle dynamic precision or width specifier. 658 case '*': { 659 ++numConversions; 660 661 if (!HasVAListArg && numConversions > numDataArgs) { 662 SourceLocation Loc = getLocationOfStringLiteralByte(FExpr, StrIdx); 663 664 if (Str[StrIdx-1] == '.') 665 Diag(Loc, diag::warn_printf_asterisk_precision_missing_arg) 666 << OrigFormatExpr->getSourceRange(); 667 else 668 Diag(Loc, diag::warn_printf_asterisk_width_missing_arg) 669 << OrigFormatExpr->getSourceRange(); 670 671 // Don't do any more checking. We'll just emit spurious errors. 672 return; 673 } 674 675 // Perform type checking on width/precision specifier. 676 Expr *E = TheCall->getArg(format_idx+numConversions); 677 if (const BuiltinType *BT = E->getType()->getAsBuiltinType()) 678 if (BT->getKind() == BuiltinType::Int) 679 break; 680 681 SourceLocation Loc = getLocationOfStringLiteralByte(FExpr, StrIdx); 682 683 if (Str[StrIdx-1] == '.') 684 Diag(Loc, diag::warn_printf_asterisk_precision_wrong_type) 685 << E->getType() << E->getSourceRange(); 686 else 687 Diag(Loc, diag::warn_printf_asterisk_width_wrong_type) 688 << E->getType() << E->getSourceRange(); 689 690 break; 691 } 692 693 // Characters which can terminate a format conversion 694 // (e.g. "%d"). Characters that specify length modifiers or 695 // other flags are handled by the default case below. 696 // 697 // FIXME: additional checks will go into the following cases. 698 case 'i': 699 case 'd': 700 case 'o': 701 case 'u': 702 case 'x': 703 case 'X': 704 case 'D': 705 case 'O': 706 case 'U': 707 case 'e': 708 case 'E': 709 case 'f': 710 case 'F': 711 case 'g': 712 case 'G': 713 case 'a': 714 case 'A': 715 case 'c': 716 case 'C': 717 case 'S': 718 case 's': 719 case 'p': 720 ++numConversions; 721 CurrentState = state_OrdChr; 722 break; 723 724 // CHECK: Are we using "%n"? Issue a warning. 725 case 'n': { 726 ++numConversions; 727 CurrentState = state_OrdChr; 728 SourceLocation Loc = getLocationOfStringLiteralByte(FExpr, 729 LastConversionIdx); 730 731 Diag(Loc, diag::warn_printf_write_back)<<OrigFormatExpr->getSourceRange(); 732 break; 733 } 734 735 // Handle "%@" 736 case '@': 737 // %@ is allowed in ObjC format strings only. 738 if(ObjCFExpr != NULL) 739 CurrentState = state_OrdChr; 740 else { 741 // Issue a warning: invalid format conversion. 742 SourceLocation Loc = 743 getLocationOfStringLiteralByte(FExpr, LastConversionIdx); 744 745 Diag(Loc, diag::warn_printf_invalid_conversion) 746 << std::string(Str+LastConversionIdx, 747 Str+std::min(LastConversionIdx+2, StrLen)) 748 << OrigFormatExpr->getSourceRange(); 749 } 750 ++numConversions; 751 break; 752 753 // Handle "%%" 754 case '%': 755 // Sanity check: Was the first "%" character the previous one? 756 // If not, we will assume that we have a malformed format 757 // conversion, and that the current "%" character is the start 758 // of a new conversion. 759 if (StrIdx - LastConversionIdx == 1) 760 CurrentState = state_OrdChr; 761 else { 762 // Issue a warning: invalid format conversion. 763 SourceLocation Loc = 764 getLocationOfStringLiteralByte(FExpr, LastConversionIdx); 765 766 Diag(Loc, diag::warn_printf_invalid_conversion) 767 << std::string(Str+LastConversionIdx, Str+StrIdx) 768 << OrigFormatExpr->getSourceRange(); 769 770 // This conversion is broken. Advance to the next format 771 // conversion. 772 LastConversionIdx = StrIdx; 773 ++numConversions; 774 } 775 break; 776 777 default: 778 // This case catches all other characters: flags, widths, etc. 779 // We should eventually process those as well. 780 break; 781 } 782 } 783 784 if (CurrentState == state_Conversion) { 785 // Issue a warning: invalid format conversion. 786 SourceLocation Loc = 787 getLocationOfStringLiteralByte(FExpr, LastConversionIdx); 788 789 Diag(Loc, diag::warn_printf_invalid_conversion) 790 << std::string(Str+LastConversionIdx, 791 Str+std::min(LastConversionIdx+2, StrLen)) 792 << OrigFormatExpr->getSourceRange(); 793 return; 794 } 795 796 if (!HasVAListArg) { 797 // CHECK: Does the number of format conversions exceed the number 798 // of data arguments? 799 if (numConversions > numDataArgs) { 800 SourceLocation Loc = 801 getLocationOfStringLiteralByte(FExpr, LastConversionIdx); 802 803 Diag(Loc, diag::warn_printf_insufficient_data_args) 804 << OrigFormatExpr->getSourceRange(); 805 } 806 // CHECK: Does the number of data arguments exceed the number of 807 // format conversions in the format string? 808 else if (numConversions < numDataArgs) 809 Diag(TheCall->getArg(format_idx+numConversions+1)->getLocStart(), 810 diag::warn_printf_too_many_data_args) 811 << OrigFormatExpr->getSourceRange(); 812 } 813} 814 815//===--- CHECK: Return Address of Stack Variable --------------------------===// 816 817static DeclRefExpr* EvalVal(Expr *E); 818static DeclRefExpr* EvalAddr(Expr* E); 819 820/// CheckReturnStackAddr - Check if a return statement returns the address 821/// of a stack variable. 822void 823Sema::CheckReturnStackAddr(Expr *RetValExp, QualType lhsType, 824 SourceLocation ReturnLoc) { 825 826 // Perform checking for returned stack addresses. 827 if (lhsType->isPointerType() || lhsType->isBlockPointerType()) { 828 if (DeclRefExpr *DR = EvalAddr(RetValExp)) 829 Diag(DR->getLocStart(), diag::warn_ret_stack_addr) 830 << DR->getDecl()->getDeclName() << RetValExp->getSourceRange(); 831 832 // Skip over implicit cast expressions when checking for block expressions. 833 if (ImplicitCastExpr *IcExpr = 834 dyn_cast_or_null<ImplicitCastExpr>(RetValExp)) 835 RetValExp = IcExpr->getSubExpr(); 836 837 if (BlockExpr *C = dyn_cast_or_null<BlockExpr>(RetValExp)) 838 Diag(C->getLocStart(), diag::err_ret_local_block) 839 << C->getSourceRange(); 840 } 841 // Perform checking for stack values returned by reference. 842 else if (lhsType->isReferenceType()) { 843 // Check for a reference to the stack 844 if (DeclRefExpr *DR = EvalVal(RetValExp)) 845 Diag(DR->getLocStart(), diag::warn_ret_stack_ref) 846 << DR->getDecl()->getDeclName() << RetValExp->getSourceRange(); 847 } 848} 849 850/// EvalAddr - EvalAddr and EvalVal are mutually recursive functions that 851/// check if the expression in a return statement evaluates to an address 852/// to a location on the stack. The recursion is used to traverse the 853/// AST of the return expression, with recursion backtracking when we 854/// encounter a subexpression that (1) clearly does not lead to the address 855/// of a stack variable or (2) is something we cannot determine leads to 856/// the address of a stack variable based on such local checking. 857/// 858/// EvalAddr processes expressions that are pointers that are used as 859/// references (and not L-values). EvalVal handles all other values. 860/// At the base case of the recursion is a check for a DeclRefExpr* in 861/// the refers to a stack variable. 862/// 863/// This implementation handles: 864/// 865/// * pointer-to-pointer casts 866/// * implicit conversions from array references to pointers 867/// * taking the address of fields 868/// * arbitrary interplay between "&" and "*" operators 869/// * pointer arithmetic from an address of a stack variable 870/// * taking the address of an array element where the array is on the stack 871static DeclRefExpr* EvalAddr(Expr *E) { 872 // We should only be called for evaluating pointer expressions. 873 assert((E->getType()->isPointerType() || 874 E->getType()->isBlockPointerType() || 875 E->getType()->isObjCQualifiedIdType()) && 876 "EvalAddr only works on pointers"); 877 878 // Our "symbolic interpreter" is just a dispatch off the currently 879 // viewed AST node. We then recursively traverse the AST by calling 880 // EvalAddr and EvalVal appropriately. 881 switch (E->getStmtClass()) { 882 case Stmt::ParenExprClass: 883 // Ignore parentheses. 884 return EvalAddr(cast<ParenExpr>(E)->getSubExpr()); 885 886 case Stmt::UnaryOperatorClass: { 887 // The only unary operator that make sense to handle here 888 // is AddrOf. All others don't make sense as pointers. 889 UnaryOperator *U = cast<UnaryOperator>(E); 890 891 if (U->getOpcode() == UnaryOperator::AddrOf) 892 return EvalVal(U->getSubExpr()); 893 else 894 return NULL; 895 } 896 897 case Stmt::BinaryOperatorClass: { 898 // Handle pointer arithmetic. All other binary operators are not valid 899 // in this context. 900 BinaryOperator *B = cast<BinaryOperator>(E); 901 BinaryOperator::Opcode op = B->getOpcode(); 902 903 if (op != BinaryOperator::Add && op != BinaryOperator::Sub) 904 return NULL; 905 906 Expr *Base = B->getLHS(); 907 908 // Determine which argument is the real pointer base. It could be 909 // the RHS argument instead of the LHS. 910 if (!Base->getType()->isPointerType()) Base = B->getRHS(); 911 912 assert (Base->getType()->isPointerType()); 913 return EvalAddr(Base); 914 } 915 916 // For conditional operators we need to see if either the LHS or RHS are 917 // valid DeclRefExpr*s. If one of them is valid, we return it. 918 case Stmt::ConditionalOperatorClass: { 919 ConditionalOperator *C = cast<ConditionalOperator>(E); 920 921 // Handle the GNU extension for missing LHS. 922 if (Expr *lhsExpr = C->getLHS()) 923 if (DeclRefExpr* LHS = EvalAddr(lhsExpr)) 924 return LHS; 925 926 return EvalAddr(C->getRHS()); 927 } 928 929 // For casts, we need to handle conversions from arrays to 930 // pointer values, and pointer-to-pointer conversions. 931 case Stmt::ImplicitCastExprClass: 932 case Stmt::CStyleCastExprClass: 933 case Stmt::CXXFunctionalCastExprClass: { 934 Expr* SubExpr = cast<CastExpr>(E)->getSubExpr(); 935 QualType T = SubExpr->getType(); 936 937 if (SubExpr->getType()->isPointerType() || 938 SubExpr->getType()->isBlockPointerType() || 939 SubExpr->getType()->isObjCQualifiedIdType()) 940 return EvalAddr(SubExpr); 941 else if (T->isArrayType()) 942 return EvalVal(SubExpr); 943 else 944 return 0; 945 } 946 947 // C++ casts. For dynamic casts, static casts, and const casts, we 948 // are always converting from a pointer-to-pointer, so we just blow 949 // through the cast. In the case the dynamic cast doesn't fail (and 950 // return NULL), we take the conservative route and report cases 951 // where we return the address of a stack variable. For Reinterpre 952 // FIXME: The comment about is wrong; we're not always converting 953 // from pointer to pointer. I'm guessing that this code should also 954 // handle references to objects. 955 case Stmt::CXXStaticCastExprClass: 956 case Stmt::CXXDynamicCastExprClass: 957 case Stmt::CXXConstCastExprClass: 958 case Stmt::CXXReinterpretCastExprClass: { 959 Expr *S = cast<CXXNamedCastExpr>(E)->getSubExpr(); 960 if (S->getType()->isPointerType() || S->getType()->isBlockPointerType()) 961 return EvalAddr(S); 962 else 963 return NULL; 964 } 965 966 // Everything else: we simply don't reason about them. 967 default: 968 return NULL; 969 } 970} 971 972 973/// EvalVal - This function is complements EvalAddr in the mutual recursion. 974/// See the comments for EvalAddr for more details. 975static DeclRefExpr* EvalVal(Expr *E) { 976 977 // We should only be called for evaluating non-pointer expressions, or 978 // expressions with a pointer type that are not used as references but instead 979 // are l-values (e.g., DeclRefExpr with a pointer type). 980 981 // Our "symbolic interpreter" is just a dispatch off the currently 982 // viewed AST node. We then recursively traverse the AST by calling 983 // EvalAddr and EvalVal appropriately. 984 switch (E->getStmtClass()) { 985 case Stmt::DeclRefExprClass: 986 case Stmt::QualifiedDeclRefExprClass: { 987 // DeclRefExpr: the base case. When we hit a DeclRefExpr we are looking 988 // at code that refers to a variable's name. We check if it has local 989 // storage within the function, and if so, return the expression. 990 DeclRefExpr *DR = cast<DeclRefExpr>(E); 991 992 if (VarDecl *V = dyn_cast<VarDecl>(DR->getDecl())) 993 if(V->hasLocalStorage() && !V->getType()->isReferenceType()) return DR; 994 995 return NULL; 996 } 997 998 case Stmt::ParenExprClass: 999 // Ignore parentheses. 1000 return EvalVal(cast<ParenExpr>(E)->getSubExpr()); 1001 1002 case Stmt::UnaryOperatorClass: { 1003 // The only unary operator that make sense to handle here 1004 // is Deref. All others don't resolve to a "name." This includes 1005 // handling all sorts of rvalues passed to a unary operator. 1006 UnaryOperator *U = cast<UnaryOperator>(E); 1007 1008 if (U->getOpcode() == UnaryOperator::Deref) 1009 return EvalAddr(U->getSubExpr()); 1010 1011 return NULL; 1012 } 1013 1014 case Stmt::ArraySubscriptExprClass: { 1015 // Array subscripts are potential references to data on the stack. We 1016 // retrieve the DeclRefExpr* for the array variable if it indeed 1017 // has local storage. 1018 return EvalAddr(cast<ArraySubscriptExpr>(E)->getBase()); 1019 } 1020 1021 case Stmt::ConditionalOperatorClass: { 1022 // For conditional operators we need to see if either the LHS or RHS are 1023 // non-NULL DeclRefExpr's. If one is non-NULL, we return it. 1024 ConditionalOperator *C = cast<ConditionalOperator>(E); 1025 1026 // Handle the GNU extension for missing LHS. 1027 if (Expr *lhsExpr = C->getLHS()) 1028 if (DeclRefExpr *LHS = EvalVal(lhsExpr)) 1029 return LHS; 1030 1031 return EvalVal(C->getRHS()); 1032 } 1033 1034 // Accesses to members are potential references to data on the stack. 1035 case Stmt::MemberExprClass: { 1036 MemberExpr *M = cast<MemberExpr>(E); 1037 1038 // Check for indirect access. We only want direct field accesses. 1039 if (!M->isArrow()) 1040 return EvalVal(M->getBase()); 1041 else 1042 return NULL; 1043 } 1044 1045 // Everything else: we simply don't reason about them. 1046 default: 1047 return NULL; 1048 } 1049} 1050 1051//===--- CHECK: Floating-Point comparisons (-Wfloat-equal) ---------------===// 1052 1053/// Check for comparisons of floating point operands using != and ==. 1054/// Issue a warning if these are no self-comparisons, as they are not likely 1055/// to do what the programmer intended. 1056void Sema::CheckFloatComparison(SourceLocation loc, Expr* lex, Expr *rex) { 1057 bool EmitWarning = true; 1058 1059 Expr* LeftExprSansParen = lex->IgnoreParens(); 1060 Expr* RightExprSansParen = rex->IgnoreParens(); 1061 1062 // Special case: check for x == x (which is OK). 1063 // Do not emit warnings for such cases. 1064 if (DeclRefExpr* DRL = dyn_cast<DeclRefExpr>(LeftExprSansParen)) 1065 if (DeclRefExpr* DRR = dyn_cast<DeclRefExpr>(RightExprSansParen)) 1066 if (DRL->getDecl() == DRR->getDecl()) 1067 EmitWarning = false; 1068 1069 1070 // Special case: check for comparisons against literals that can be exactly 1071 // represented by APFloat. In such cases, do not emit a warning. This 1072 // is a heuristic: often comparison against such literals are used to 1073 // detect if a value in a variable has not changed. This clearly can 1074 // lead to false negatives. 1075 if (EmitWarning) { 1076 if (FloatingLiteral* FLL = dyn_cast<FloatingLiteral>(LeftExprSansParen)) { 1077 if (FLL->isExact()) 1078 EmitWarning = false; 1079 } 1080 else 1081 if (FloatingLiteral* FLR = dyn_cast<FloatingLiteral>(RightExprSansParen)){ 1082 if (FLR->isExact()) 1083 EmitWarning = false; 1084 } 1085 } 1086 1087 // Check for comparisons with builtin types. 1088 if (EmitWarning) 1089 if (CallExpr* CL = dyn_cast<CallExpr>(LeftExprSansParen)) 1090 if (CL->isBuiltinCall(Context)) 1091 EmitWarning = false; 1092 1093 if (EmitWarning) 1094 if (CallExpr* CR = dyn_cast<CallExpr>(RightExprSansParen)) 1095 if (CR->isBuiltinCall(Context)) 1096 EmitWarning = false; 1097 1098 // Emit the diagnostic. 1099 if (EmitWarning) 1100 Diag(loc, diag::warn_floatingpoint_eq) 1101 << lex->getSourceRange() << rex->getSourceRange(); 1102} 1103