SemaChecking.cpp revision 082d936a5b8323ac2c04558d8bca277a647831a3
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(const Expr *E, const CallExpr *TheCall, 427 bool HasVAListArg, 428 unsigned format_idx, unsigned firstDataArg) { 429 430 switch (E->getStmtClass()) { 431 case Stmt::ConditionalOperatorClass: { 432 const ConditionalOperator *C = cast<ConditionalOperator>(E); 433 return SemaCheckStringLiteral(C->getLHS(), TheCall, 434 HasVAListArg, format_idx, firstDataArg) 435 && SemaCheckStringLiteral(C->getRHS(), TheCall, 436 HasVAListArg, format_idx, firstDataArg); 437 } 438 439 case Stmt::ImplicitCastExprClass: { 440 const ImplicitCastExpr *Expr = cast<ImplicitCastExpr>(E); 441 return SemaCheckStringLiteral(Expr->getSubExpr(), TheCall, HasVAListArg, 442 format_idx, firstDataArg); 443 } 444 445 case Stmt::ParenExprClass: { 446 const ParenExpr *Expr = cast<ParenExpr>(E); 447 return SemaCheckStringLiteral(Expr->getSubExpr(), TheCall, HasVAListArg, 448 format_idx, firstDataArg); 449 } 450 451 case Stmt::DeclRefExprClass: { 452 const DeclRefExpr *DR = cast<DeclRefExpr>(E); 453 454 // As an exception, do not flag errors for variables binding to 455 // const string literals. 456 if (const VarDecl *VD = dyn_cast<VarDecl>(DR->getDecl())) { 457 bool isConstant = false; 458 QualType T = DR->getType(); 459 460 if (const ArrayType *AT = Context.getAsArrayType(T)) { 461 isConstant = AT->getElementType().isConstant(Context); 462 } 463 else if (const PointerType *PT = T->getAsPointerType()) { 464 isConstant = T.isConstant(Context) && 465 PT->getPointeeType().isConstant(Context); 466 } 467 468 if (isConstant) { 469 const VarDecl *Def = 0; 470 if (const Expr *Init = VD->getDefinition(Def)) 471 return SemaCheckStringLiteral(Init, TheCall, 472 HasVAListArg, format_idx, firstDataArg); 473 } 474 } 475 476 return false; 477 } 478 479 case Stmt::ObjCStringLiteralClass: 480 case Stmt::StringLiteralClass: { 481 const StringLiteral *StrE = NULL; 482 483 if (const ObjCStringLiteral *ObjCFExpr = dyn_cast<ObjCStringLiteral>(E)) 484 StrE = ObjCFExpr->getString(); 485 else 486 StrE = cast<StringLiteral>(E); 487 488 if (StrE) { 489 CheckPrintfString(StrE, E, TheCall, HasVAListArg, format_idx, 490 firstDataArg); 491 return true; 492 } 493 494 return false; 495 } 496 497 default: 498 return false; 499 } 500} 501 502 503/// CheckPrintfArguments - Check calls to printf (and similar functions) for 504/// correct use of format strings. 505/// 506/// HasVAListArg - A predicate indicating whether the printf-like 507/// function is passed an explicit va_arg argument (e.g., vprintf) 508/// 509/// format_idx - The index into Args for the format string. 510/// 511/// Improper format strings to functions in the printf family can be 512/// the source of bizarre bugs and very serious security holes. A 513/// good source of information is available in the following paper 514/// (which includes additional references): 515/// 516/// FormatGuard: Automatic Protection From printf Format String 517/// Vulnerabilities, Proceedings of the 10th USENIX Security Symposium, 2001. 518/// 519/// Functionality implemented: 520/// 521/// We can statically check the following properties for string 522/// literal format strings for non v.*printf functions (where the 523/// arguments are passed directly): 524// 525/// (1) Are the number of format conversions equal to the number of 526/// data arguments? 527/// 528/// (2) Does each format conversion correctly match the type of the 529/// corresponding data argument? (TODO) 530/// 531/// Moreover, for all printf functions we can: 532/// 533/// (3) Check for a missing format string (when not caught by type checking). 534/// 535/// (4) Check for no-operation flags; e.g. using "#" with format 536/// conversion 'c' (TODO) 537/// 538/// (5) Check the use of '%n', a major source of security holes. 539/// 540/// (6) Check for malformed format conversions that don't specify anything. 541/// 542/// (7) Check for empty format strings. e.g: printf(""); 543/// 544/// (8) Check that the format string is a wide literal. 545/// 546/// (9) Also check the arguments of functions with the __format__ attribute. 547/// (TODO). 548/// 549/// All of these checks can be done by parsing the format string. 550/// 551/// For now, we ONLY do (1), (3), (5), (6), (7), and (8). 552void 553Sema::CheckPrintfArguments(const CallExpr *TheCall, bool HasVAListArg, 554 unsigned format_idx, unsigned firstDataArg) { 555 const Expr *Fn = TheCall->getCallee(); 556 557 // CHECK: printf-like function is called with no format string. 558 if (format_idx >= TheCall->getNumArgs()) { 559 Diag(TheCall->getRParenLoc(), diag::warn_printf_missing_format_string) 560 << Fn->getSourceRange(); 561 return; 562 } 563 564 const Expr *OrigFormatExpr = TheCall->getArg(format_idx)->IgnoreParenCasts(); 565 566 // CHECK: format string is not a string literal. 567 // 568 // Dynamically generated format strings are difficult to 569 // automatically vet at compile time. Requiring that format strings 570 // are string literals: (1) permits the checking of format strings by 571 // the compiler and thereby (2) can practically remove the source of 572 // many format string exploits. 573 574 // Format string can be either ObjC string (e.g. @"%d") or 575 // C string (e.g. "%d") 576 // ObjC string uses the same format specifiers as C string, so we can use 577 // the same format string checking logic for both ObjC and C strings. 578 bool isFExpr = SemaCheckStringLiteral(OrigFormatExpr, TheCall, 579 HasVAListArg, format_idx, 580 firstDataArg); 581 582 if (!isFExpr) { 583 // For vprintf* functions (i.e., HasVAListArg==true), we add a 584 // special check to see if the format string is a function parameter 585 // of the function calling the printf function. If the function 586 // has an attribute indicating it is a printf-like function, then we 587 // should suppress warnings concerning non-literals being used in a call 588 // to a vprintf function. For example: 589 // 590 // void 591 // logmessage(char const *fmt __attribute__ (format (printf, 1, 2)), ...) { 592 // va_list ap; 593 // va_start(ap, fmt); 594 // vprintf(fmt, ap); // Do NOT emit a warning about "fmt". 595 // ... 596 // 597 // 598 // FIXME: We don't have full attribute support yet, so just check to see 599 // if the argument is a DeclRefExpr that references a parameter. We'll 600 // add proper support for checking the attribute later. 601 if (HasVAListArg) 602 if (const DeclRefExpr* DR = dyn_cast<DeclRefExpr>(OrigFormatExpr)) 603 if (isa<ParmVarDecl>(DR->getDecl())) 604 return; 605 606 Diag(TheCall->getArg(format_idx)->getLocStart(), 607 diag::warn_printf_not_string_constant) 608 << OrigFormatExpr->getSourceRange(); 609 return; 610 } 611} 612 613void Sema::CheckPrintfString(const StringLiteral *FExpr, 614 const Expr *OrigFormatExpr, 615 const CallExpr *TheCall, bool HasVAListArg, 616 unsigned format_idx, unsigned firstDataArg) { 617 618 const ObjCStringLiteral *ObjCFExpr = 619 dyn_cast<ObjCStringLiteral>(OrigFormatExpr); 620 621 // CHECK: is the format string a wide literal? 622 if (FExpr->isWide()) { 623 Diag(FExpr->getLocStart(), 624 diag::warn_printf_format_string_is_wide_literal) 625 << OrigFormatExpr->getSourceRange(); 626 return; 627 } 628 629 // Str - The format string. NOTE: this is NOT null-terminated! 630 const char * const Str = FExpr->getStrData(); 631 632 // CHECK: empty format string? 633 const unsigned StrLen = FExpr->getByteLength(); 634 635 if (StrLen == 0) { 636 Diag(FExpr->getLocStart(), diag::warn_printf_empty_format_string) 637 << OrigFormatExpr->getSourceRange(); 638 return; 639 } 640 641 // We process the format string using a binary state machine. The 642 // current state is stored in CurrentState. 643 enum { 644 state_OrdChr, 645 state_Conversion 646 } CurrentState = state_OrdChr; 647 648 // numConversions - The number of conversions seen so far. This is 649 // incremented as we traverse the format string. 650 unsigned numConversions = 0; 651 652 // numDataArgs - The number of data arguments after the format 653 // string. This can only be determined for non vprintf-like 654 // functions. For those functions, this value is 1 (the sole 655 // va_arg argument). 656 unsigned numDataArgs = TheCall->getNumArgs()-firstDataArg; 657 658 // Inspect the format string. 659 unsigned StrIdx = 0; 660 661 // LastConversionIdx - Index within the format string where we last saw 662 // a '%' character that starts a new format conversion. 663 unsigned LastConversionIdx = 0; 664 665 for (; StrIdx < StrLen; ++StrIdx) { 666 667 // Is the number of detected conversion conversions greater than 668 // the number of matching data arguments? If so, stop. 669 if (!HasVAListArg && numConversions > numDataArgs) break; 670 671 // Handle "\0" 672 if (Str[StrIdx] == '\0') { 673 // The string returned by getStrData() is not null-terminated, 674 // so the presence of a null character is likely an error. 675 Diag(getLocationOfStringLiteralByte(FExpr, StrIdx), 676 diag::warn_printf_format_string_contains_null_char) 677 << OrigFormatExpr->getSourceRange(); 678 return; 679 } 680 681 // Ordinary characters (not processing a format conversion). 682 if (CurrentState == state_OrdChr) { 683 if (Str[StrIdx] == '%') { 684 CurrentState = state_Conversion; 685 LastConversionIdx = StrIdx; 686 } 687 continue; 688 } 689 690 // Seen '%'. Now processing a format conversion. 691 switch (Str[StrIdx]) { 692 // Handle dynamic precision or width specifier. 693 case '*': { 694 ++numConversions; 695 696 if (!HasVAListArg && numConversions > numDataArgs) { 697 SourceLocation Loc = getLocationOfStringLiteralByte(FExpr, StrIdx); 698 699 if (Str[StrIdx-1] == '.') 700 Diag(Loc, diag::warn_printf_asterisk_precision_missing_arg) 701 << OrigFormatExpr->getSourceRange(); 702 else 703 Diag(Loc, diag::warn_printf_asterisk_width_missing_arg) 704 << OrigFormatExpr->getSourceRange(); 705 706 // Don't do any more checking. We'll just emit spurious errors. 707 return; 708 } 709 710 // Perform type checking on width/precision specifier. 711 const Expr *E = TheCall->getArg(format_idx+numConversions); 712 if (const BuiltinType *BT = E->getType()->getAsBuiltinType()) 713 if (BT->getKind() == BuiltinType::Int) 714 break; 715 716 SourceLocation Loc = getLocationOfStringLiteralByte(FExpr, StrIdx); 717 718 if (Str[StrIdx-1] == '.') 719 Diag(Loc, diag::warn_printf_asterisk_precision_wrong_type) 720 << E->getType() << E->getSourceRange(); 721 else 722 Diag(Loc, diag::warn_printf_asterisk_width_wrong_type) 723 << E->getType() << E->getSourceRange(); 724 725 break; 726 } 727 728 // Characters which can terminate a format conversion 729 // (e.g. "%d"). Characters that specify length modifiers or 730 // other flags are handled by the default case below. 731 // 732 // FIXME: additional checks will go into the following cases. 733 case 'i': 734 case 'd': 735 case 'o': 736 case 'u': 737 case 'x': 738 case 'X': 739 case 'D': 740 case 'O': 741 case 'U': 742 case 'e': 743 case 'E': 744 case 'f': 745 case 'F': 746 case 'g': 747 case 'G': 748 case 'a': 749 case 'A': 750 case 'c': 751 case 'C': 752 case 'S': 753 case 's': 754 case 'p': 755 ++numConversions; 756 CurrentState = state_OrdChr; 757 break; 758 759 // CHECK: Are we using "%n"? Issue a warning. 760 case 'n': { 761 ++numConversions; 762 CurrentState = state_OrdChr; 763 SourceLocation Loc = getLocationOfStringLiteralByte(FExpr, 764 LastConversionIdx); 765 766 Diag(Loc, diag::warn_printf_write_back)<<OrigFormatExpr->getSourceRange(); 767 break; 768 } 769 770 // Handle "%@" 771 case '@': 772 // %@ is allowed in ObjC format strings only. 773 if(ObjCFExpr != NULL) 774 CurrentState = state_OrdChr; 775 else { 776 // Issue a warning: invalid format conversion. 777 SourceLocation Loc = 778 getLocationOfStringLiteralByte(FExpr, LastConversionIdx); 779 780 Diag(Loc, diag::warn_printf_invalid_conversion) 781 << std::string(Str+LastConversionIdx, 782 Str+std::min(LastConversionIdx+2, StrLen)) 783 << OrigFormatExpr->getSourceRange(); 784 } 785 ++numConversions; 786 break; 787 788 // Handle "%%" 789 case '%': 790 // Sanity check: Was the first "%" character the previous one? 791 // If not, we will assume that we have a malformed format 792 // conversion, and that the current "%" character is the start 793 // of a new conversion. 794 if (StrIdx - LastConversionIdx == 1) 795 CurrentState = state_OrdChr; 796 else { 797 // Issue a warning: invalid format conversion. 798 SourceLocation Loc = 799 getLocationOfStringLiteralByte(FExpr, LastConversionIdx); 800 801 Diag(Loc, diag::warn_printf_invalid_conversion) 802 << std::string(Str+LastConversionIdx, Str+StrIdx) 803 << OrigFormatExpr->getSourceRange(); 804 805 // This conversion is broken. Advance to the next format 806 // conversion. 807 LastConversionIdx = StrIdx; 808 ++numConversions; 809 } 810 break; 811 812 default: 813 // This case catches all other characters: flags, widths, etc. 814 // We should eventually process those as well. 815 break; 816 } 817 } 818 819 if (CurrentState == state_Conversion) { 820 // Issue a warning: invalid format conversion. 821 SourceLocation Loc = 822 getLocationOfStringLiteralByte(FExpr, LastConversionIdx); 823 824 Diag(Loc, diag::warn_printf_invalid_conversion) 825 << std::string(Str+LastConversionIdx, 826 Str+std::min(LastConversionIdx+2, StrLen)) 827 << OrigFormatExpr->getSourceRange(); 828 return; 829 } 830 831 if (!HasVAListArg) { 832 // CHECK: Does the number of format conversions exceed the number 833 // of data arguments? 834 if (numConversions > numDataArgs) { 835 SourceLocation Loc = 836 getLocationOfStringLiteralByte(FExpr, LastConversionIdx); 837 838 Diag(Loc, diag::warn_printf_insufficient_data_args) 839 << OrigFormatExpr->getSourceRange(); 840 } 841 // CHECK: Does the number of data arguments exceed the number of 842 // format conversions in the format string? 843 else if (numConversions < numDataArgs) 844 Diag(TheCall->getArg(format_idx+numConversions+1)->getLocStart(), 845 diag::warn_printf_too_many_data_args) 846 << OrigFormatExpr->getSourceRange(); 847 } 848} 849 850//===--- CHECK: Return Address of Stack Variable --------------------------===// 851 852static DeclRefExpr* EvalVal(Expr *E); 853static DeclRefExpr* EvalAddr(Expr* E); 854 855/// CheckReturnStackAddr - Check if a return statement returns the address 856/// of a stack variable. 857void 858Sema::CheckReturnStackAddr(Expr *RetValExp, QualType lhsType, 859 SourceLocation ReturnLoc) { 860 861 // Perform checking for returned stack addresses. 862 if (lhsType->isPointerType() || lhsType->isBlockPointerType()) { 863 if (DeclRefExpr *DR = EvalAddr(RetValExp)) 864 Diag(DR->getLocStart(), diag::warn_ret_stack_addr) 865 << DR->getDecl()->getDeclName() << RetValExp->getSourceRange(); 866 867 // Skip over implicit cast expressions when checking for block expressions. 868 if (ImplicitCastExpr *IcExpr = 869 dyn_cast_or_null<ImplicitCastExpr>(RetValExp)) 870 RetValExp = IcExpr->getSubExpr(); 871 872 if (BlockExpr *C = dyn_cast_or_null<BlockExpr>(RetValExp)) 873 Diag(C->getLocStart(), diag::err_ret_local_block) 874 << C->getSourceRange(); 875 } 876 // Perform checking for stack values returned by reference. 877 else if (lhsType->isReferenceType()) { 878 // Check for a reference to the stack 879 if (DeclRefExpr *DR = EvalVal(RetValExp)) 880 Diag(DR->getLocStart(), diag::warn_ret_stack_ref) 881 << DR->getDecl()->getDeclName() << RetValExp->getSourceRange(); 882 } 883} 884 885/// EvalAddr - EvalAddr and EvalVal are mutually recursive functions that 886/// check if the expression in a return statement evaluates to an address 887/// to a location on the stack. The recursion is used to traverse the 888/// AST of the return expression, with recursion backtracking when we 889/// encounter a subexpression that (1) clearly does not lead to the address 890/// of a stack variable or (2) is something we cannot determine leads to 891/// the address of a stack variable based on such local checking. 892/// 893/// EvalAddr processes expressions that are pointers that are used as 894/// references (and not L-values). EvalVal handles all other values. 895/// At the base case of the recursion is a check for a DeclRefExpr* in 896/// the refers to a stack variable. 897/// 898/// This implementation handles: 899/// 900/// * pointer-to-pointer casts 901/// * implicit conversions from array references to pointers 902/// * taking the address of fields 903/// * arbitrary interplay between "&" and "*" operators 904/// * pointer arithmetic from an address of a stack variable 905/// * taking the address of an array element where the array is on the stack 906static DeclRefExpr* EvalAddr(Expr *E) { 907 // We should only be called for evaluating pointer expressions. 908 assert((E->getType()->isPointerType() || 909 E->getType()->isBlockPointerType() || 910 E->getType()->isObjCQualifiedIdType()) && 911 "EvalAddr only works on pointers"); 912 913 // Our "symbolic interpreter" is just a dispatch off the currently 914 // viewed AST node. We then recursively traverse the AST by calling 915 // EvalAddr and EvalVal appropriately. 916 switch (E->getStmtClass()) { 917 case Stmt::ParenExprClass: 918 // Ignore parentheses. 919 return EvalAddr(cast<ParenExpr>(E)->getSubExpr()); 920 921 case Stmt::UnaryOperatorClass: { 922 // The only unary operator that make sense to handle here 923 // is AddrOf. All others don't make sense as pointers. 924 UnaryOperator *U = cast<UnaryOperator>(E); 925 926 if (U->getOpcode() == UnaryOperator::AddrOf) 927 return EvalVal(U->getSubExpr()); 928 else 929 return NULL; 930 } 931 932 case Stmt::BinaryOperatorClass: { 933 // Handle pointer arithmetic. All other binary operators are not valid 934 // in this context. 935 BinaryOperator *B = cast<BinaryOperator>(E); 936 BinaryOperator::Opcode op = B->getOpcode(); 937 938 if (op != BinaryOperator::Add && op != BinaryOperator::Sub) 939 return NULL; 940 941 Expr *Base = B->getLHS(); 942 943 // Determine which argument is the real pointer base. It could be 944 // the RHS argument instead of the LHS. 945 if (!Base->getType()->isPointerType()) Base = B->getRHS(); 946 947 assert (Base->getType()->isPointerType()); 948 return EvalAddr(Base); 949 } 950 951 // For conditional operators we need to see if either the LHS or RHS are 952 // valid DeclRefExpr*s. If one of them is valid, we return it. 953 case Stmt::ConditionalOperatorClass: { 954 ConditionalOperator *C = cast<ConditionalOperator>(E); 955 956 // Handle the GNU extension for missing LHS. 957 if (Expr *lhsExpr = C->getLHS()) 958 if (DeclRefExpr* LHS = EvalAddr(lhsExpr)) 959 return LHS; 960 961 return EvalAddr(C->getRHS()); 962 } 963 964 // For casts, we need to handle conversions from arrays to 965 // pointer values, and pointer-to-pointer conversions. 966 case Stmt::ImplicitCastExprClass: 967 case Stmt::CStyleCastExprClass: 968 case Stmt::CXXFunctionalCastExprClass: { 969 Expr* SubExpr = cast<CastExpr>(E)->getSubExpr(); 970 QualType T = SubExpr->getType(); 971 972 if (SubExpr->getType()->isPointerType() || 973 SubExpr->getType()->isBlockPointerType() || 974 SubExpr->getType()->isObjCQualifiedIdType()) 975 return EvalAddr(SubExpr); 976 else if (T->isArrayType()) 977 return EvalVal(SubExpr); 978 else 979 return 0; 980 } 981 982 // C++ casts. For dynamic casts, static casts, and const casts, we 983 // are always converting from a pointer-to-pointer, so we just blow 984 // through the cast. In the case the dynamic cast doesn't fail (and 985 // return NULL), we take the conservative route and report cases 986 // where we return the address of a stack variable. For Reinterpre 987 // FIXME: The comment about is wrong; we're not always converting 988 // from pointer to pointer. I'm guessing that this code should also 989 // handle references to objects. 990 case Stmt::CXXStaticCastExprClass: 991 case Stmt::CXXDynamicCastExprClass: 992 case Stmt::CXXConstCastExprClass: 993 case Stmt::CXXReinterpretCastExprClass: { 994 Expr *S = cast<CXXNamedCastExpr>(E)->getSubExpr(); 995 if (S->getType()->isPointerType() || S->getType()->isBlockPointerType()) 996 return EvalAddr(S); 997 else 998 return NULL; 999 } 1000 1001 // Everything else: we simply don't reason about them. 1002 default: 1003 return NULL; 1004 } 1005} 1006 1007 1008/// EvalVal - This function is complements EvalAddr in the mutual recursion. 1009/// See the comments for EvalAddr for more details. 1010static DeclRefExpr* EvalVal(Expr *E) { 1011 1012 // We should only be called for evaluating non-pointer expressions, or 1013 // expressions with a pointer type that are not used as references but instead 1014 // are l-values (e.g., DeclRefExpr with a pointer type). 1015 1016 // Our "symbolic interpreter" is just a dispatch off the currently 1017 // viewed AST node. We then recursively traverse the AST by calling 1018 // EvalAddr and EvalVal appropriately. 1019 switch (E->getStmtClass()) { 1020 case Stmt::DeclRefExprClass: 1021 case Stmt::QualifiedDeclRefExprClass: { 1022 // DeclRefExpr: the base case. When we hit a DeclRefExpr we are looking 1023 // at code that refers to a variable's name. We check if it has local 1024 // storage within the function, and if so, return the expression. 1025 DeclRefExpr *DR = cast<DeclRefExpr>(E); 1026 1027 if (VarDecl *V = dyn_cast<VarDecl>(DR->getDecl())) 1028 if(V->hasLocalStorage() && !V->getType()->isReferenceType()) return DR; 1029 1030 return NULL; 1031 } 1032 1033 case Stmt::ParenExprClass: 1034 // Ignore parentheses. 1035 return EvalVal(cast<ParenExpr>(E)->getSubExpr()); 1036 1037 case Stmt::UnaryOperatorClass: { 1038 // The only unary operator that make sense to handle here 1039 // is Deref. All others don't resolve to a "name." This includes 1040 // handling all sorts of rvalues passed to a unary operator. 1041 UnaryOperator *U = cast<UnaryOperator>(E); 1042 1043 if (U->getOpcode() == UnaryOperator::Deref) 1044 return EvalAddr(U->getSubExpr()); 1045 1046 return NULL; 1047 } 1048 1049 case Stmt::ArraySubscriptExprClass: { 1050 // Array subscripts are potential references to data on the stack. We 1051 // retrieve the DeclRefExpr* for the array variable if it indeed 1052 // has local storage. 1053 return EvalAddr(cast<ArraySubscriptExpr>(E)->getBase()); 1054 } 1055 1056 case Stmt::ConditionalOperatorClass: { 1057 // For conditional operators we need to see if either the LHS or RHS are 1058 // non-NULL DeclRefExpr's. If one is non-NULL, we return it. 1059 ConditionalOperator *C = cast<ConditionalOperator>(E); 1060 1061 // Handle the GNU extension for missing LHS. 1062 if (Expr *lhsExpr = C->getLHS()) 1063 if (DeclRefExpr *LHS = EvalVal(lhsExpr)) 1064 return LHS; 1065 1066 return EvalVal(C->getRHS()); 1067 } 1068 1069 // Accesses to members are potential references to data on the stack. 1070 case Stmt::MemberExprClass: { 1071 MemberExpr *M = cast<MemberExpr>(E); 1072 1073 // Check for indirect access. We only want direct field accesses. 1074 if (!M->isArrow()) 1075 return EvalVal(M->getBase()); 1076 else 1077 return NULL; 1078 } 1079 1080 // Everything else: we simply don't reason about them. 1081 default: 1082 return NULL; 1083 } 1084} 1085 1086//===--- CHECK: Floating-Point comparisons (-Wfloat-equal) ---------------===// 1087 1088/// Check for comparisons of floating point operands using != and ==. 1089/// Issue a warning if these are no self-comparisons, as they are not likely 1090/// to do what the programmer intended. 1091void Sema::CheckFloatComparison(SourceLocation loc, Expr* lex, Expr *rex) { 1092 bool EmitWarning = true; 1093 1094 Expr* LeftExprSansParen = lex->IgnoreParens(); 1095 Expr* RightExprSansParen = rex->IgnoreParens(); 1096 1097 // Special case: check for x == x (which is OK). 1098 // Do not emit warnings for such cases. 1099 if (DeclRefExpr* DRL = dyn_cast<DeclRefExpr>(LeftExprSansParen)) 1100 if (DeclRefExpr* DRR = dyn_cast<DeclRefExpr>(RightExprSansParen)) 1101 if (DRL->getDecl() == DRR->getDecl()) 1102 EmitWarning = false; 1103 1104 1105 // Special case: check for comparisons against literals that can be exactly 1106 // represented by APFloat. In such cases, do not emit a warning. This 1107 // is a heuristic: often comparison against such literals are used to 1108 // detect if a value in a variable has not changed. This clearly can 1109 // lead to false negatives. 1110 if (EmitWarning) { 1111 if (FloatingLiteral* FLL = dyn_cast<FloatingLiteral>(LeftExprSansParen)) { 1112 if (FLL->isExact()) 1113 EmitWarning = false; 1114 } 1115 else 1116 if (FloatingLiteral* FLR = dyn_cast<FloatingLiteral>(RightExprSansParen)){ 1117 if (FLR->isExact()) 1118 EmitWarning = false; 1119 } 1120 } 1121 1122 // Check for comparisons with builtin types. 1123 if (EmitWarning) 1124 if (CallExpr* CL = dyn_cast<CallExpr>(LeftExprSansParen)) 1125 if (CL->isBuiltinCall(Context)) 1126 EmitWarning = false; 1127 1128 if (EmitWarning) 1129 if (CallExpr* CR = dyn_cast<CallExpr>(RightExprSansParen)) 1130 if (CR->isBuiltinCall(Context)) 1131 EmitWarning = false; 1132 1133 // Emit the diagnostic. 1134 if (EmitWarning) 1135 Diag(loc, diag::warn_floatingpoint_eq) 1136 << lex->getSourceRange() << rex->getSourceRange(); 1137} 1138