SemaChecking.cpp revision d5f8a4fd4d6dfb0415b93bb7ab721bba5cab1332
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/Preprocessor.h" 21#include "clang/Basic/Diagnostic.h" 22#include "SemaUtil.h" 23using namespace clang; 24 25/// CheckFunctionCall - Check a direct function call for various correctness 26/// and safety properties not strictly enforced by the C type system. 27Action::ExprResult 28Sema::CheckFunctionCall(FunctionDecl *FDecl, CallExpr *TheCallRaw) { 29 llvm::OwningPtr<CallExpr> TheCall(TheCallRaw); 30 // Get the IdentifierInfo* for the called function. 31 IdentifierInfo *FnInfo = FDecl->getIdentifier(); 32 33 switch (FnInfo->getBuiltinID()) { 34 case Builtin::BI__builtin___CFStringMakeConstantString: 35 assert(TheCall->getNumArgs() == 1 && 36 "Wrong # arguments to builtin CFStringMakeConstantString"); 37 if (CheckBuiltinCFStringArgument(TheCall->getArg(0))) 38 return true; 39 return TheCall.take(); 40 case Builtin::BI__builtin_stdarg_start: 41 case Builtin::BI__builtin_va_start: 42 if (SemaBuiltinVAStart(TheCall.get())) 43 return true; 44 return TheCall.take(); 45 case Builtin::BI__builtin_isgreater: 46 case Builtin::BI__builtin_isgreaterequal: 47 case Builtin::BI__builtin_isless: 48 case Builtin::BI__builtin_islessequal: 49 case Builtin::BI__builtin_islessgreater: 50 case Builtin::BI__builtin_isunordered: 51 if (SemaBuiltinUnorderedCompare(TheCall.get())) 52 return true; 53 return TheCall.take(); 54 case Builtin::BI__builtin_return_address: 55 case Builtin::BI__builtin_frame_address: 56 if (SemaBuiltinStackAddress(TheCall.get())) 57 return true; 58 return TheCall.take(); 59 case Builtin::BI__builtin_shufflevector: 60 return SemaBuiltinShuffleVector(TheCall.get()); 61 case Builtin::BI__builtin_prefetch: 62 if (SemaBuiltinPrefetch(TheCall.get())) 63 return true; 64 return TheCall.take(); 65 case Builtin::BI__builtin_object_size: 66 if (SemaBuiltinObjectSize(TheCall.get())) 67 return true; 68 } 69 70 // Search the KnownFunctionIDs for the identifier. 71 unsigned i = 0, e = id_num_known_functions; 72 for (; i != e; ++i) { if (KnownFunctionIDs[i] == FnInfo) break; } 73 if (i == e) return TheCall.take(); 74 75 // Printf checking. 76 if (i <= id_vprintf) { 77 // Retrieve the index of the format string parameter and determine 78 // if the function is passed a va_arg argument. 79 unsigned format_idx = 0; 80 bool HasVAListArg = false; 81 82 switch (i) { 83 default: assert(false && "No format string argument index."); 84 case id_printf: format_idx = 0; break; 85 case id_fprintf: format_idx = 1; break; 86 case id_sprintf: format_idx = 1; break; 87 case id_snprintf: format_idx = 2; break; 88 case id_asprintf: format_idx = 1; break; 89 case id_NSLog: format_idx = 0; break; 90 case id_vsnprintf: format_idx = 2; HasVAListArg = true; break; 91 case id_vasprintf: format_idx = 1; HasVAListArg = true; break; 92 case id_vfprintf: format_idx = 1; HasVAListArg = true; break; 93 case id_vsprintf: format_idx = 1; HasVAListArg = true; break; 94 case id_vprintf: format_idx = 0; HasVAListArg = true; break; 95 } 96 97 CheckPrintfArguments(TheCall.get(), HasVAListArg, format_idx); 98 } 99 100 return TheCall.take(); 101} 102 103/// CheckBuiltinCFStringArgument - Checks that the argument to the builtin 104/// CFString constructor is correct 105bool Sema::CheckBuiltinCFStringArgument(Expr* Arg) { 106 Arg = Arg->IgnoreParenCasts(); 107 108 StringLiteral *Literal = dyn_cast<StringLiteral>(Arg); 109 110 if (!Literal || Literal->isWide()) { 111 Diag(Arg->getLocStart(), 112 diag::err_cfstring_literal_not_string_constant, 113 Arg->getSourceRange()); 114 return true; 115 } 116 117 const char *Data = Literal->getStrData(); 118 unsigned Length = Literal->getByteLength(); 119 120 for (unsigned i = 0; i < Length; ++i) { 121 if (!isascii(Data[i])) { 122 Diag(PP.AdvanceToTokenCharacter(Arg->getLocStart(), i + 1), 123 diag::warn_cfstring_literal_contains_non_ascii_character, 124 Arg->getSourceRange()); 125 break; 126 } 127 128 if (!Data[i]) { 129 Diag(PP.AdvanceToTokenCharacter(Arg->getLocStart(), i + 1), 130 diag::warn_cfstring_literal_contains_nul_character, 131 Arg->getSourceRange()); 132 break; 133 } 134 } 135 136 return false; 137} 138 139/// SemaBuiltinVAStart - Check the arguments to __builtin_va_start for validity. 140/// Emit an error and return true on failure, return false on success. 141bool Sema::SemaBuiltinVAStart(CallExpr *TheCall) { 142 Expr *Fn = TheCall->getCallee(); 143 if (TheCall->getNumArgs() > 2) { 144 Diag(TheCall->getArg(2)->getLocStart(), 145 diag::err_typecheck_call_too_many_args, Fn->getSourceRange(), 146 SourceRange(TheCall->getArg(2)->getLocStart(), 147 (*(TheCall->arg_end()-1))->getLocEnd())); 148 return true; 149 } 150 151 // Determine whether the current function is variadic or not. 152 bool isVariadic; 153 if (getCurFunctionDecl()) 154 isVariadic = 155 cast<FunctionTypeProto>(getCurFunctionDecl()->getType())->isVariadic(); 156 else 157 isVariadic = getCurMethodDecl()->isVariadic(); 158 159 if (!isVariadic) { 160 Diag(Fn->getLocStart(), diag::err_va_start_used_in_non_variadic_function); 161 return true; 162 } 163 164 // Verify that the second argument to the builtin is the last argument of the 165 // current function or method. 166 bool SecondArgIsLastNamedArgument = false; 167 const Expr *Arg = TheCall->getArg(1)->IgnoreParenCasts(); 168 169 if (const DeclRefExpr *DR = dyn_cast<DeclRefExpr>(Arg)) { 170 if (const ParmVarDecl *PV = dyn_cast<ParmVarDecl>(DR->getDecl())) { 171 // FIXME: This isn't correct for methods (results in bogus warning). 172 // Get the last formal in the current function. 173 const ParmVarDecl *LastArg; 174 if (getCurFunctionDecl()) 175 LastArg = *(getCurFunctionDecl()->param_end()-1); 176 else 177 LastArg = *(getCurMethodDecl()->param_end()-1); 178 SecondArgIsLastNamedArgument = PV == LastArg; 179 } 180 } 181 182 if (!SecondArgIsLastNamedArgument) 183 Diag(TheCall->getArg(1)->getLocStart(), 184 diag::warn_second_parameter_of_va_start_not_last_named_argument); 185 return false; 186} 187 188/// SemaBuiltinUnorderedCompare - Handle functions like __builtin_isgreater and 189/// friends. This is declared to take (...), so we have to check everything. 190bool Sema::SemaBuiltinUnorderedCompare(CallExpr *TheCall) { 191 if (TheCall->getNumArgs() < 2) 192 return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args); 193 if (TheCall->getNumArgs() > 2) 194 return Diag(TheCall->getArg(2)->getLocStart(), 195 diag::err_typecheck_call_too_many_args, 196 SourceRange(TheCall->getArg(2)->getLocStart(), 197 (*(TheCall->arg_end()-1))->getLocEnd())); 198 199 Expr *OrigArg0 = TheCall->getArg(0); 200 Expr *OrigArg1 = TheCall->getArg(1); 201 202 // Do standard promotions between the two arguments, returning their common 203 // type. 204 QualType Res = UsualArithmeticConversions(OrigArg0, OrigArg1, false); 205 206 // If the common type isn't a real floating type, then the arguments were 207 // invalid for this operation. 208 if (!Res->isRealFloatingType()) 209 return Diag(OrigArg0->getLocStart(), 210 diag::err_typecheck_call_invalid_ordered_compare, 211 OrigArg0->getType().getAsString(), 212 OrigArg1->getType().getAsString(), 213 SourceRange(OrigArg0->getLocStart(), OrigArg1->getLocEnd())); 214 215 return false; 216} 217 218bool Sema::SemaBuiltinStackAddress(CallExpr *TheCall) { 219 // The signature for these builtins is exact; the only thing we need 220 // to check is that the argument is a constant. 221 SourceLocation Loc; 222 if (!TheCall->getArg(0)->isIntegerConstantExpr(Context, &Loc)) 223 return Diag(Loc, diag::err_stack_const_level, TheCall->getSourceRange()); 224 225 return false; 226} 227 228/// SemaBuiltinShuffleVector - Handle __builtin_shufflevector. 229// This is declared to take (...), so we have to check everything. 230Action::ExprResult Sema::SemaBuiltinShuffleVector(CallExpr *TheCall) { 231 if (TheCall->getNumArgs() < 3) 232 return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args, 233 TheCall->getSourceRange()); 234 235 QualType FAType = TheCall->getArg(0)->getType(); 236 QualType SAType = TheCall->getArg(1)->getType(); 237 238 if (!FAType->isVectorType() || !SAType->isVectorType()) { 239 Diag(TheCall->getLocStart(), diag::err_shufflevector_non_vector, 240 SourceRange(TheCall->getArg(0)->getLocStart(), 241 TheCall->getArg(1)->getLocEnd())); 242 return true; 243 } 244 245 if (Context.getCanonicalType(FAType).getUnqualifiedType() != 246 Context.getCanonicalType(SAType).getUnqualifiedType()) { 247 Diag(TheCall->getLocStart(), diag::err_shufflevector_incompatible_vector, 248 SourceRange(TheCall->getArg(0)->getLocStart(), 249 TheCall->getArg(1)->getLocEnd())); 250 return true; 251 } 252 253 unsigned numElements = FAType->getAsVectorType()->getNumElements(); 254 if (TheCall->getNumArgs() != numElements+2) { 255 if (TheCall->getNumArgs() < numElements+2) 256 return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args, 257 TheCall->getSourceRange()); 258 return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_many_args, 259 TheCall->getSourceRange()); 260 } 261 262 for (unsigned i = 2; i < TheCall->getNumArgs(); i++) { 263 llvm::APSInt Result(32); 264 if (!TheCall->getArg(i)->isIntegerConstantExpr(Result, Context)) 265 return Diag(TheCall->getLocStart(), 266 diag::err_shufflevector_nonconstant_argument, 267 TheCall->getArg(i)->getSourceRange()); 268 269 if (Result.getActiveBits() > 64 || Result.getZExtValue() >= numElements*2) 270 return Diag(TheCall->getLocStart(), 271 diag::err_shufflevector_argument_too_large, 272 TheCall->getArg(i)->getSourceRange()); 273 } 274 275 llvm::SmallVector<Expr*, 32> exprs; 276 277 for (unsigned i = 0, e = TheCall->getNumArgs(); i != e; i++) { 278 exprs.push_back(TheCall->getArg(i)); 279 TheCall->setArg(i, 0); 280 } 281 282 return new ShuffleVectorExpr(exprs.begin(), numElements+2, FAType, 283 TheCall->getCallee()->getLocStart(), 284 TheCall->getRParenLoc()); 285} 286 287/// SemaBuiltinPrefetch - Handle __builtin_prefetch. 288// This is declared to take (const void*, ...) and can take two 289// optional constant int args. 290bool Sema::SemaBuiltinPrefetch(CallExpr *TheCall) { 291 unsigned numArgs = TheCall->getNumArgs(); 292 bool res = false; 293 294 if (numArgs > 3) { 295 res |= Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_many_args, 296 TheCall->getSourceRange()); 297 } 298 299 // Argument 0 is checked for us and the remaining arguments must be 300 // constant integers. 301 for (unsigned i=1; i<numArgs; ++i) { 302 Expr *Arg = TheCall->getArg(i); 303 QualType RWType = Arg->getType(); 304 305 const BuiltinType *BT = RWType->getAsBuiltinType(); 306 llvm::APSInt Result; 307 if (!BT || BT->getKind() != BuiltinType::Int || 308 !Arg->isIntegerConstantExpr(Result, Context)) { 309 if (Diag(TheCall->getLocStart(), diag::err_prefetch_invalid_argument, 310 SourceRange(Arg->getLocStart(), Arg->getLocEnd()))) { 311 res = true; 312 continue; 313 } 314 } 315 316 // FIXME: gcc issues a warning and rewrites these to 0. These 317 // seems especially odd for the third argument since the default 318 // is 3. 319 if (i==1) { 320 if (Result.getSExtValue() < 0 || Result.getSExtValue() > 1) 321 res |= Diag(TheCall->getLocStart(), diag::err_argument_invalid_range, 322 "0", "1", 323 SourceRange(Arg->getLocStart(), Arg->getLocEnd())); 324 } else { 325 if (Result.getSExtValue() < 0 || Result.getSExtValue() > 3) 326 res |= Diag(TheCall->getLocStart(), diag::err_argument_invalid_range, 327 "0", "3", 328 SourceRange(Arg->getLocStart(), Arg->getLocEnd())); 329 } 330 } 331 332 return res; 333} 334 335/// SemaBuiltinObjectSize - Handle __builtin_object_size(void *ptr, 336/// int type). This simply type checks that type is one of the defined 337/// constants (0-3). 338bool Sema::SemaBuiltinObjectSize(CallExpr *TheCall) { 339 Expr *Arg = TheCall->getArg(1); 340 QualType ArgType = Arg->getType(); 341 const BuiltinType *BT = ArgType->getAsBuiltinType(); 342 llvm::APSInt Result(32); 343 if (!BT || BT->getKind() != BuiltinType::Int || 344 !Arg->isIntegerConstantExpr(Result, Context)) { 345 return Diag(TheCall->getLocStart(), diag::err_object_size_invalid_argument, 346 SourceRange(Arg->getLocStart(), Arg->getLocEnd())); 347 } 348 349 if (Result.getSExtValue() < 0 || Result.getSExtValue() > 3) { 350 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range, 351 "0", "3", 352 SourceRange(Arg->getLocStart(), Arg->getLocEnd())); 353 } 354 355 return false; 356} 357 358/// CheckPrintfArguments - Check calls to printf (and similar functions) for 359/// correct use of format strings. 360/// 361/// HasVAListArg - A predicate indicating whether the printf-like 362/// function is passed an explicit va_arg argument (e.g., vprintf) 363/// 364/// format_idx - The index into Args for the format string. 365/// 366/// Improper format strings to functions in the printf family can be 367/// the source of bizarre bugs and very serious security holes. A 368/// good source of information is available in the following paper 369/// (which includes additional references): 370/// 371/// FormatGuard: Automatic Protection From printf Format String 372/// Vulnerabilities, Proceedings of the 10th USENIX Security Symposium, 2001. 373/// 374/// Functionality implemented: 375/// 376/// We can statically check the following properties for string 377/// literal format strings for non v.*printf functions (where the 378/// arguments are passed directly): 379// 380/// (1) Are the number of format conversions equal to the number of 381/// data arguments? 382/// 383/// (2) Does each format conversion correctly match the type of the 384/// corresponding data argument? (TODO) 385/// 386/// Moreover, for all printf functions we can: 387/// 388/// (3) Check for a missing format string (when not caught by type checking). 389/// 390/// (4) Check for no-operation flags; e.g. using "#" with format 391/// conversion 'c' (TODO) 392/// 393/// (5) Check the use of '%n', a major source of security holes. 394/// 395/// (6) Check for malformed format conversions that don't specify anything. 396/// 397/// (7) Check for empty format strings. e.g: printf(""); 398/// 399/// (8) Check that the format string is a wide literal. 400/// 401/// (9) Also check the arguments of functions with the __format__ attribute. 402/// (TODO). 403/// 404/// All of these checks can be done by parsing the format string. 405/// 406/// For now, we ONLY do (1), (3), (5), (6), (7), and (8). 407void 408Sema::CheckPrintfArguments(CallExpr *TheCall, bool HasVAListArg, 409 unsigned format_idx) { 410 Expr *Fn = TheCall->getCallee(); 411 412 // CHECK: printf-like function is called with no format string. 413 if (format_idx >= TheCall->getNumArgs()) { 414 Diag(TheCall->getRParenLoc(), diag::warn_printf_missing_format_string, 415 Fn->getSourceRange()); 416 return; 417 } 418 419 Expr *OrigFormatExpr = TheCall->getArg(format_idx)->IgnoreParenCasts(); 420 421 // CHECK: format string is not a string literal. 422 // 423 // Dynamically generated format strings are difficult to 424 // automatically vet at compile time. Requiring that format strings 425 // are string literals: (1) permits the checking of format strings by 426 // the compiler and thereby (2) can practically remove the source of 427 // many format string exploits. 428 429 // Format string can be either ObjC string (e.g. @"%d") or 430 // C string (e.g. "%d") 431 // ObjC string uses the same format specifiers as C string, so we can use 432 // the same format string checking logic for both ObjC and C strings. 433 ObjCStringLiteral *ObjCFExpr = dyn_cast<ObjCStringLiteral>(OrigFormatExpr); 434 StringLiteral *FExpr = NULL; 435 436 if(ObjCFExpr != NULL) 437 FExpr = ObjCFExpr->getString(); 438 else 439 FExpr = dyn_cast<StringLiteral>(OrigFormatExpr); 440 441 if (FExpr == NULL) { 442 // For vprintf* functions (i.e., HasVAListArg==true), we add a 443 // special check to see if the format string is a function parameter 444 // of the function calling the printf function. If the function 445 // has an attribute indicating it is a printf-like function, then we 446 // should suppress warnings concerning non-literals being used in a call 447 // to a vprintf function. For example: 448 // 449 // void 450 // logmessage(char const *fmt __attribute__ (format (printf, 1, 2)), ...) { 451 // va_list ap; 452 // va_start(ap, fmt); 453 // vprintf(fmt, ap); // Do NOT emit a warning about "fmt". 454 // ... 455 // 456 // 457 // FIXME: We don't have full attribute support yet, so just check to see 458 // if the argument is a DeclRefExpr that references a parameter. We'll 459 // add proper support for checking the attribute later. 460 if (HasVAListArg) 461 if (DeclRefExpr* DR = dyn_cast<DeclRefExpr>(OrigFormatExpr)) 462 if (isa<ParmVarDecl>(DR->getDecl())) 463 return; 464 465 Diag(TheCall->getArg(format_idx)->getLocStart(), 466 diag::warn_printf_not_string_constant, 467 OrigFormatExpr->getSourceRange()); 468 return; 469 } 470 471 // CHECK: is the format string a wide literal? 472 if (FExpr->isWide()) { 473 Diag(FExpr->getLocStart(), 474 diag::warn_printf_format_string_is_wide_literal, 475 OrigFormatExpr->getSourceRange()); 476 return; 477 } 478 479 // Str - The format string. NOTE: this is NOT null-terminated! 480 const char * const Str = FExpr->getStrData(); 481 482 // CHECK: empty format string? 483 const unsigned StrLen = FExpr->getByteLength(); 484 485 if (StrLen == 0) { 486 Diag(FExpr->getLocStart(), diag::warn_printf_empty_format_string, 487 OrigFormatExpr->getSourceRange()); 488 return; 489 } 490 491 // We process the format string using a binary state machine. The 492 // current state is stored in CurrentState. 493 enum { 494 state_OrdChr, 495 state_Conversion 496 } CurrentState = state_OrdChr; 497 498 // numConversions - The number of conversions seen so far. This is 499 // incremented as we traverse the format string. 500 unsigned numConversions = 0; 501 502 // numDataArgs - The number of data arguments after the format 503 // string. This can only be determined for non vprintf-like 504 // functions. For those functions, this value is 1 (the sole 505 // va_arg argument). 506 unsigned numDataArgs = TheCall->getNumArgs()-(format_idx+1); 507 508 // Inspect the format string. 509 unsigned StrIdx = 0; 510 511 // LastConversionIdx - Index within the format string where we last saw 512 // a '%' character that starts a new format conversion. 513 unsigned LastConversionIdx = 0; 514 515 for (; StrIdx < StrLen; ++StrIdx) { 516 517 // Is the number of detected conversion conversions greater than 518 // the number of matching data arguments? If so, stop. 519 if (!HasVAListArg && numConversions > numDataArgs) break; 520 521 // Handle "\0" 522 if (Str[StrIdx] == '\0') { 523 // The string returned by getStrData() is not null-terminated, 524 // so the presence of a null character is likely an error. 525 Diag(PP.AdvanceToTokenCharacter(FExpr->getLocStart(), StrIdx+1), 526 diag::warn_printf_format_string_contains_null_char, 527 OrigFormatExpr->getSourceRange()); 528 return; 529 } 530 531 // Ordinary characters (not processing a format conversion). 532 if (CurrentState == state_OrdChr) { 533 if (Str[StrIdx] == '%') { 534 CurrentState = state_Conversion; 535 LastConversionIdx = StrIdx; 536 } 537 continue; 538 } 539 540 // Seen '%'. Now processing a format conversion. 541 switch (Str[StrIdx]) { 542 // Handle dynamic precision or width specifier. 543 case '*': { 544 ++numConversions; 545 546 if (!HasVAListArg && numConversions > numDataArgs) { 547 SourceLocation Loc = FExpr->getLocStart(); 548 Loc = PP.AdvanceToTokenCharacter(Loc, StrIdx+1); 549 550 if (Str[StrIdx-1] == '.') 551 Diag(Loc, diag::warn_printf_asterisk_precision_missing_arg, 552 OrigFormatExpr->getSourceRange()); 553 else 554 Diag(Loc, diag::warn_printf_asterisk_width_missing_arg, 555 OrigFormatExpr->getSourceRange()); 556 557 // Don't do any more checking. We'll just emit spurious errors. 558 return; 559 } 560 561 // Perform type checking on width/precision specifier. 562 Expr *E = TheCall->getArg(format_idx+numConversions); 563 if (const BuiltinType *BT = E->getType()->getAsBuiltinType()) 564 if (BT->getKind() == BuiltinType::Int) 565 break; 566 567 SourceLocation Loc = 568 PP.AdvanceToTokenCharacter(FExpr->getLocStart(), StrIdx+1); 569 570 if (Str[StrIdx-1] == '.') 571 Diag(Loc, diag::warn_printf_asterisk_precision_wrong_type, 572 E->getType().getAsString(), E->getSourceRange()); 573 else 574 Diag(Loc, diag::warn_printf_asterisk_width_wrong_type, 575 E->getType().getAsString(), E->getSourceRange()); 576 577 break; 578 } 579 580 // Characters which can terminate a format conversion 581 // (e.g. "%d"). Characters that specify length modifiers or 582 // other flags are handled by the default case below. 583 // 584 // FIXME: additional checks will go into the following cases. 585 case 'i': 586 case 'd': 587 case 'o': 588 case 'u': 589 case 'x': 590 case 'X': 591 case 'D': 592 case 'O': 593 case 'U': 594 case 'e': 595 case 'E': 596 case 'f': 597 case 'F': 598 case 'g': 599 case 'G': 600 case 'a': 601 case 'A': 602 case 'c': 603 case 'C': 604 case 'S': 605 case 's': 606 case 'p': 607 ++numConversions; 608 CurrentState = state_OrdChr; 609 break; 610 611 // CHECK: Are we using "%n"? Issue a warning. 612 case 'n': { 613 ++numConversions; 614 CurrentState = state_OrdChr; 615 SourceLocation Loc = PP.AdvanceToTokenCharacter(FExpr->getLocStart(), 616 LastConversionIdx+1); 617 618 Diag(Loc, diag::warn_printf_write_back, OrigFormatExpr->getSourceRange()); 619 break; 620 } 621 622 // Handle "%@" 623 case '@': 624 // %@ is allowed in ObjC format strings only. 625 if(ObjCFExpr != NULL) 626 CurrentState = state_OrdChr; 627 else { 628 // Issue a warning: invalid format conversion. 629 SourceLocation Loc = PP.AdvanceToTokenCharacter(FExpr->getLocStart(), 630 LastConversionIdx+1); 631 632 Diag(Loc, diag::warn_printf_invalid_conversion, 633 std::string(Str+LastConversionIdx, 634 Str+std::min(LastConversionIdx+2, StrLen)), 635 OrigFormatExpr->getSourceRange()); 636 } 637 ++numConversions; 638 break; 639 640 // Handle "%%" 641 case '%': 642 // Sanity check: Was the first "%" character the previous one? 643 // If not, we will assume that we have a malformed format 644 // conversion, and that the current "%" character is the start 645 // of a new conversion. 646 if (StrIdx - LastConversionIdx == 1) 647 CurrentState = state_OrdChr; 648 else { 649 // Issue a warning: invalid format conversion. 650 SourceLocation Loc = PP.AdvanceToTokenCharacter(FExpr->getLocStart(), 651 LastConversionIdx+1); 652 653 Diag(Loc, diag::warn_printf_invalid_conversion, 654 std::string(Str+LastConversionIdx, Str+StrIdx), 655 OrigFormatExpr->getSourceRange()); 656 657 // This conversion is broken. Advance to the next format 658 // conversion. 659 LastConversionIdx = StrIdx; 660 ++numConversions; 661 } 662 break; 663 664 default: 665 // This case catches all other characters: flags, widths, etc. 666 // We should eventually process those as well. 667 break; 668 } 669 } 670 671 if (CurrentState == state_Conversion) { 672 // Issue a warning: invalid format conversion. 673 SourceLocation Loc = PP.AdvanceToTokenCharacter(FExpr->getLocStart(), 674 LastConversionIdx+1); 675 676 Diag(Loc, diag::warn_printf_invalid_conversion, 677 std::string(Str+LastConversionIdx, 678 Str+std::min(LastConversionIdx+2, StrLen)), 679 OrigFormatExpr->getSourceRange()); 680 return; 681 } 682 683 if (!HasVAListArg) { 684 // CHECK: Does the number of format conversions exceed the number 685 // of data arguments? 686 if (numConversions > numDataArgs) { 687 SourceLocation Loc = PP.AdvanceToTokenCharacter(FExpr->getLocStart(), 688 LastConversionIdx); 689 690 Diag(Loc, diag::warn_printf_insufficient_data_args, 691 OrigFormatExpr->getSourceRange()); 692 } 693 // CHECK: Does the number of data arguments exceed the number of 694 // format conversions in the format string? 695 else if (numConversions < numDataArgs) 696 Diag(TheCall->getArg(format_idx+numConversions+1)->getLocStart(), 697 diag::warn_printf_too_many_data_args, 698 OrigFormatExpr->getSourceRange()); 699 } 700} 701 702//===--- CHECK: Return Address of Stack Variable --------------------------===// 703 704static DeclRefExpr* EvalVal(Expr *E); 705static DeclRefExpr* EvalAddr(Expr* E); 706 707/// CheckReturnStackAddr - Check if a return statement returns the address 708/// of a stack variable. 709void 710Sema::CheckReturnStackAddr(Expr *RetValExp, QualType lhsType, 711 SourceLocation ReturnLoc) { 712 713 // Perform checking for returned stack addresses. 714 if (lhsType->isPointerType()) { 715 if (DeclRefExpr *DR = EvalAddr(RetValExp)) 716 Diag(DR->getLocStart(), diag::warn_ret_stack_addr, 717 DR->getDecl()->getIdentifier()->getName(), 718 RetValExp->getSourceRange()); 719 } 720 // Perform checking for stack values returned by reference. 721 else if (lhsType->isReferenceType()) { 722 // Check for an implicit cast to a reference. 723 if (ImplicitCastExpr *I = dyn_cast<ImplicitCastExpr>(RetValExp)) 724 if (DeclRefExpr *DR = EvalVal(I->getSubExpr())) 725 Diag(DR->getLocStart(), diag::warn_ret_stack_ref, 726 DR->getDecl()->getIdentifier()->getName(), 727 RetValExp->getSourceRange()); 728 } 729} 730 731/// EvalAddr - EvalAddr and EvalVal are mutually recursive functions that 732/// check if the expression in a return statement evaluates to an address 733/// to a location on the stack. The recursion is used to traverse the 734/// AST of the return expression, with recursion backtracking when we 735/// encounter a subexpression that (1) clearly does not lead to the address 736/// of a stack variable or (2) is something we cannot determine leads to 737/// the address of a stack variable based on such local checking. 738/// 739/// EvalAddr processes expressions that are pointers that are used as 740/// references (and not L-values). EvalVal handles all other values. 741/// At the base case of the recursion is a check for a DeclRefExpr* in 742/// the refers to a stack variable. 743/// 744/// This implementation handles: 745/// 746/// * pointer-to-pointer casts 747/// * implicit conversions from array references to pointers 748/// * taking the address of fields 749/// * arbitrary interplay between "&" and "*" operators 750/// * pointer arithmetic from an address of a stack variable 751/// * taking the address of an array element where the array is on the stack 752static DeclRefExpr* EvalAddr(Expr *E) { 753 // We should only be called for evaluating pointer expressions. 754 assert((E->getType()->isPointerType() || 755 E->getType()->isObjCQualifiedIdType()) && 756 "EvalAddr only works on pointers"); 757 758 // Our "symbolic interpreter" is just a dispatch off the currently 759 // viewed AST node. We then recursively traverse the AST by calling 760 // EvalAddr and EvalVal appropriately. 761 switch (E->getStmtClass()) { 762 case Stmt::ParenExprClass: 763 // Ignore parentheses. 764 return EvalAddr(cast<ParenExpr>(E)->getSubExpr()); 765 766 case Stmt::UnaryOperatorClass: { 767 // The only unary operator that make sense to handle here 768 // is AddrOf. All others don't make sense as pointers. 769 UnaryOperator *U = cast<UnaryOperator>(E); 770 771 if (U->getOpcode() == UnaryOperator::AddrOf) 772 return EvalVal(U->getSubExpr()); 773 else 774 return NULL; 775 } 776 777 case Stmt::BinaryOperatorClass: { 778 // Handle pointer arithmetic. All other binary operators are not valid 779 // in this context. 780 BinaryOperator *B = cast<BinaryOperator>(E); 781 BinaryOperator::Opcode op = B->getOpcode(); 782 783 if (op != BinaryOperator::Add && op != BinaryOperator::Sub) 784 return NULL; 785 786 Expr *Base = B->getLHS(); 787 788 // Determine which argument is the real pointer base. It could be 789 // the RHS argument instead of the LHS. 790 if (!Base->getType()->isPointerType()) Base = B->getRHS(); 791 792 assert (Base->getType()->isPointerType()); 793 return EvalAddr(Base); 794 } 795 796 // For conditional operators we need to see if either the LHS or RHS are 797 // valid DeclRefExpr*s. If one of them is valid, we return it. 798 case Stmt::ConditionalOperatorClass: { 799 ConditionalOperator *C = cast<ConditionalOperator>(E); 800 801 // Handle the GNU extension for missing LHS. 802 if (Expr *lhsExpr = C->getLHS()) 803 if (DeclRefExpr* LHS = EvalAddr(lhsExpr)) 804 return LHS; 805 806 return EvalAddr(C->getRHS()); 807 } 808 809 // For casts, we need to handle conversions from arrays to 810 // pointer values, and pointer-to-pointer conversions. 811 case Stmt::ExplicitCastExprClass: 812 case Stmt::ImplicitCastExprClass: { 813 814 Expr* SubExpr = cast<CastExpr>(E)->getSubExpr(); 815 QualType T = SubExpr->getType(); 816 817 if (T->isPointerType() || T->isObjCQualifiedIdType()) 818 return EvalAddr(SubExpr); 819 else if (T->isArrayType()) 820 return EvalVal(SubExpr); 821 else 822 return 0; 823 } 824 825 // C++ casts. For dynamic casts, static casts, and const casts, we 826 // are always converting from a pointer-to-pointer, so we just blow 827 // through the cast. In the case the dynamic cast doesn't fail 828 // (and return NULL), we take the conservative route and report cases 829 // where we return the address of a stack variable. For Reinterpre 830 case Stmt::CXXCastExprClass: { 831 CXXCastExpr *C = cast<CXXCastExpr>(E); 832 833 if (C->getOpcode() == CXXCastExpr::ReinterpretCast) { 834 Expr *S = C->getSubExpr(); 835 if (S->getType()->isPointerType()) 836 return EvalAddr(S); 837 else 838 return NULL; 839 } 840 else 841 return EvalAddr(C->getSubExpr()); 842 } 843 844 // Everything else: we simply don't reason about them. 845 default: 846 return NULL; 847 } 848} 849 850 851/// EvalVal - This function is complements EvalAddr in the mutual recursion. 852/// See the comments for EvalAddr for more details. 853static DeclRefExpr* EvalVal(Expr *E) { 854 855 // We should only be called for evaluating non-pointer expressions, or 856 // expressions with a pointer type that are not used as references but instead 857 // are l-values (e.g., DeclRefExpr with a pointer type). 858 859 // Our "symbolic interpreter" is just a dispatch off the currently 860 // viewed AST node. We then recursively traverse the AST by calling 861 // EvalAddr and EvalVal appropriately. 862 switch (E->getStmtClass()) { 863 case Stmt::DeclRefExprClass: { 864 // DeclRefExpr: the base case. When we hit a DeclRefExpr we are looking 865 // at code that refers to a variable's name. We check if it has local 866 // storage within the function, and if so, return the expression. 867 DeclRefExpr *DR = cast<DeclRefExpr>(E); 868 869 if (VarDecl *V = dyn_cast<VarDecl>(DR->getDecl())) 870 if(V->hasLocalStorage()) return DR; 871 872 return NULL; 873 } 874 875 case Stmt::ParenExprClass: 876 // Ignore parentheses. 877 return EvalVal(cast<ParenExpr>(E)->getSubExpr()); 878 879 case Stmt::UnaryOperatorClass: { 880 // The only unary operator that make sense to handle here 881 // is Deref. All others don't resolve to a "name." This includes 882 // handling all sorts of rvalues passed to a unary operator. 883 UnaryOperator *U = cast<UnaryOperator>(E); 884 885 if (U->getOpcode() == UnaryOperator::Deref) 886 return EvalAddr(U->getSubExpr()); 887 888 return NULL; 889 } 890 891 case Stmt::ArraySubscriptExprClass: { 892 // Array subscripts are potential references to data on the stack. We 893 // retrieve the DeclRefExpr* for the array variable if it indeed 894 // has local storage. 895 return EvalAddr(cast<ArraySubscriptExpr>(E)->getBase()); 896 } 897 898 case Stmt::ConditionalOperatorClass: { 899 // For conditional operators we need to see if either the LHS or RHS are 900 // non-NULL DeclRefExpr's. If one is non-NULL, we return it. 901 ConditionalOperator *C = cast<ConditionalOperator>(E); 902 903 // Handle the GNU extension for missing LHS. 904 if (Expr *lhsExpr = C->getLHS()) 905 if (DeclRefExpr *LHS = EvalVal(lhsExpr)) 906 return LHS; 907 908 return EvalVal(C->getRHS()); 909 } 910 911 // Accesses to members are potential references to data on the stack. 912 case Stmt::MemberExprClass: { 913 MemberExpr *M = cast<MemberExpr>(E); 914 915 // Check for indirect access. We only want direct field accesses. 916 if (!M->isArrow()) 917 return EvalVal(M->getBase()); 918 else 919 return NULL; 920 } 921 922 // Everything else: we simply don't reason about them. 923 default: 924 return NULL; 925 } 926} 927 928//===--- CHECK: Floating-Point comparisons (-Wfloat-equal) ---------------===// 929 930/// Check for comparisons of floating point operands using != and ==. 931/// Issue a warning if these are no self-comparisons, as they are not likely 932/// to do what the programmer intended. 933void Sema::CheckFloatComparison(SourceLocation loc, Expr* lex, Expr *rex) { 934 bool EmitWarning = true; 935 936 Expr* LeftExprSansParen = lex->IgnoreParens(); 937 Expr* RightExprSansParen = rex->IgnoreParens(); 938 939 // Special case: check for x == x (which is OK). 940 // Do not emit warnings for such cases. 941 if (DeclRefExpr* DRL = dyn_cast<DeclRefExpr>(LeftExprSansParen)) 942 if (DeclRefExpr* DRR = dyn_cast<DeclRefExpr>(RightExprSansParen)) 943 if (DRL->getDecl() == DRR->getDecl()) 944 EmitWarning = false; 945 946 947 // Special case: check for comparisons against literals that can be exactly 948 // represented by APFloat. In such cases, do not emit a warning. This 949 // is a heuristic: often comparison against such literals are used to 950 // detect if a value in a variable has not changed. This clearly can 951 // lead to false negatives. 952 if (EmitWarning) { 953 if (FloatingLiteral* FLL = dyn_cast<FloatingLiteral>(LeftExprSansParen)) { 954 if (FLL->isExact()) 955 EmitWarning = false; 956 } 957 else 958 if (FloatingLiteral* FLR = dyn_cast<FloatingLiteral>(RightExprSansParen)){ 959 if (FLR->isExact()) 960 EmitWarning = false; 961 } 962 } 963 964 // Check for comparisons with builtin types. 965 if (EmitWarning) 966 if (CallExpr* CL = dyn_cast<CallExpr>(LeftExprSansParen)) 967 if (isCallBuiltin(CL)) 968 EmitWarning = false; 969 970 if (EmitWarning) 971 if (CallExpr* CR = dyn_cast<CallExpr>(RightExprSansParen)) 972 if (isCallBuiltin(CR)) 973 EmitWarning = false; 974 975 // Emit the diagnostic. 976 if (EmitWarning) 977 Diag(loc, diag::warn_floatingpoint_eq, 978 lex->getSourceRange(),rex->getSourceRange()); 979} 980