Chris Lattner | 59907c4 | 2007-08-10 20:18:51 +0000 | [diff] [blame] | 1 | //===--- SemaChecking.cpp - Extra Semantic Checking -----------------------===// |
| 2 | // |
| 3 | // The LLVM Compiler Infrastructure |
| 4 | // |
| 5 | // This file was developed by Ted Kremenek and is distributed under |
| 6 | // the University of Illinois Open Source 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/Decl.h" |
| 18 | #include "clang/AST/Expr.h" |
| 19 | #include "clang/Lex/Preprocessor.h" |
| 20 | #include "clang/Lex/LiteralSupport.h" |
| 21 | #include "clang/Basic/SourceManager.h" |
| 22 | #include "clang/Basic/Diagnostic.h" |
| 23 | #include "clang/Basic/LangOptions.h" |
| 24 | #include "clang/Basic/TargetInfo.h" |
| 25 | #include "llvm/ADT/SmallString.h" |
| 26 | #include "llvm/ADT/StringExtras.h" |
| 27 | using namespace clang; |
| 28 | |
| 29 | /// CheckFunctionCall - Check a direct function call for various correctness |
| 30 | /// and safety properties not strictly enforced by the C type system. |
Anders Carlsson | 71993dd | 2007-08-17 05:31:46 +0000 | [diff] [blame] | 31 | bool |
Ted Kremenek | 71895b9 | 2007-08-14 17:39:48 +0000 | [diff] [blame] | 32 | Sema::CheckFunctionCall(Expr *Fn, |
| 33 | SourceLocation LParenLoc, SourceLocation RParenLoc, |
| 34 | FunctionDecl *FDecl, |
Chris Lattner | 59907c4 | 2007-08-10 20:18:51 +0000 | [diff] [blame] | 35 | Expr** Args, unsigned NumArgsInCall) { |
| 36 | |
| 37 | // Get the IdentifierInfo* for the called function. |
| 38 | IdentifierInfo *FnInfo = FDecl->getIdentifier(); |
| 39 | |
Anders Carlsson | 71993dd | 2007-08-17 05:31:46 +0000 | [diff] [blame] | 40 | if (FnInfo->getBuiltinID() == |
| 41 | Builtin::BI__builtin___CFStringMakeConstantString) { |
| 42 | assert(NumArgsInCall == 1 && |
| 43 | "Wrong number of arguments to builtin CFStringMakeConstantString"); |
| 44 | return CheckBuiltinCFStringArgument(Args[0]); |
| 45 | } |
| 46 | |
Chris Lattner | 59907c4 | 2007-08-10 20:18:51 +0000 | [diff] [blame] | 47 | // Search the KnownFunctionIDs for the identifier. |
| 48 | unsigned i = 0, e = id_num_known_functions; |
Ted Kremenek | 71895b9 | 2007-08-14 17:39:48 +0000 | [diff] [blame] | 49 | for (; i != e; ++i) { if (KnownFunctionIDs[i] == FnInfo) break; } |
Anders Carlsson | 9cdc4d3 | 2007-08-17 15:44:17 +0000 | [diff] [blame] | 50 | if (i == e) return false; |
Chris Lattner | 59907c4 | 2007-08-10 20:18:51 +0000 | [diff] [blame] | 51 | |
| 52 | // Printf checking. |
| 53 | if (i <= id_vprintf) { |
Ted Kremenek | 71895b9 | 2007-08-14 17:39:48 +0000 | [diff] [blame] | 54 | // Retrieve the index of the format string parameter and determine |
| 55 | // if the function is passed a va_arg argument. |
Chris Lattner | 59907c4 | 2007-08-10 20:18:51 +0000 | [diff] [blame] | 56 | unsigned format_idx = 0; |
Ted Kremenek | 71895b9 | 2007-08-14 17:39:48 +0000 | [diff] [blame] | 57 | bool HasVAListArg = false; |
| 58 | |
Chris Lattner | 59907c4 | 2007-08-10 20:18:51 +0000 | [diff] [blame] | 59 | switch (i) { |
| 60 | default: assert(false && "No format string argument index."); |
| 61 | case id_printf: format_idx = 0; break; |
| 62 | case id_fprintf: format_idx = 1; break; |
| 63 | case id_sprintf: format_idx = 1; break; |
| 64 | case id_snprintf: format_idx = 2; break; |
Ted Kremenek | 71895b9 | 2007-08-14 17:39:48 +0000 | [diff] [blame] | 65 | case id_asprintf: format_idx = 1; HasVAListArg = true; break; |
| 66 | case id_vsnprintf: format_idx = 2; HasVAListArg = true; break; |
| 67 | case id_vasprintf: format_idx = 1; HasVAListArg = true; break; |
| 68 | case id_vfprintf: format_idx = 1; HasVAListArg = true; break; |
| 69 | case id_vsprintf: format_idx = 1; HasVAListArg = true; break; |
| 70 | case id_vprintf: format_idx = 0; HasVAListArg = true; break; |
| 71 | } |
| 72 | |
| 73 | CheckPrintfArguments(Fn, LParenLoc, RParenLoc, HasVAListArg, |
Ted Kremenek | 06de276 | 2007-08-17 16:46:58 +0000 | [diff] [blame^] | 74 | FDecl, format_idx, Args, NumArgsInCall); |
Chris Lattner | 59907c4 | 2007-08-10 20:18:51 +0000 | [diff] [blame] | 75 | } |
Anders Carlsson | 71993dd | 2007-08-17 05:31:46 +0000 | [diff] [blame] | 76 | |
Anders Carlsson | 9cdc4d3 | 2007-08-17 15:44:17 +0000 | [diff] [blame] | 77 | return false; |
Anders Carlsson | 71993dd | 2007-08-17 05:31:46 +0000 | [diff] [blame] | 78 | } |
| 79 | |
| 80 | /// CheckBuiltinCFStringArgument - Checks that the argument to the builtin |
| 81 | /// CFString constructor is correct |
| 82 | bool Sema::CheckBuiltinCFStringArgument(Expr* Arg) |
| 83 | { |
| 84 | while (ParenExpr *PE = dyn_cast<ParenExpr>(Arg)) |
| 85 | Arg = PE->getSubExpr(); |
| 86 | |
| 87 | StringLiteral *Literal = dyn_cast<StringLiteral>(Arg); |
| 88 | |
| 89 | if (!Literal || Literal->isWide()) { |
| 90 | Diag(Arg->getLocStart(), |
| 91 | diag::err_cfstring_literal_not_string_constant, |
| 92 | Arg->getSourceRange()); |
Anders Carlsson | 9cdc4d3 | 2007-08-17 15:44:17 +0000 | [diff] [blame] | 93 | return true; |
Anders Carlsson | 71993dd | 2007-08-17 05:31:46 +0000 | [diff] [blame] | 94 | } |
| 95 | |
| 96 | const char *Data = Literal->getStrData(); |
| 97 | unsigned Length = Literal->getByteLength(); |
| 98 | |
| 99 | for (unsigned i = 0; i < Length; ++i) { |
| 100 | if (!isascii(Data[i])) { |
| 101 | Diag(PP.AdvanceToTokenCharacter(Arg->getLocStart(), i + 1), |
| 102 | diag::warn_cfstring_literal_contains_non_ascii_character, |
| 103 | Arg->getSourceRange()); |
| 104 | break; |
| 105 | } |
| 106 | |
| 107 | if (!Data[i]) { |
| 108 | Diag(PP.AdvanceToTokenCharacter(Arg->getLocStart(), i + 1), |
| 109 | diag::warn_cfstring_literal_contains_nul_character, |
| 110 | Arg->getSourceRange()); |
| 111 | break; |
| 112 | } |
| 113 | } |
| 114 | |
Anders Carlsson | 9cdc4d3 | 2007-08-17 15:44:17 +0000 | [diff] [blame] | 115 | return false; |
Chris Lattner | 59907c4 | 2007-08-10 20:18:51 +0000 | [diff] [blame] | 116 | } |
| 117 | |
| 118 | /// CheckPrintfArguments - Check calls to printf (and similar functions) for |
Ted Kremenek | 71895b9 | 2007-08-14 17:39:48 +0000 | [diff] [blame] | 119 | /// correct use of format strings. |
| 120 | /// |
| 121 | /// HasVAListArg - A predicate indicating whether the printf-like |
| 122 | /// function is passed an explicit va_arg argument (e.g., vprintf) |
| 123 | /// |
| 124 | /// format_idx - The index into Args for the format string. |
| 125 | /// |
| 126 | /// Improper format strings to functions in the printf family can be |
| 127 | /// the source of bizarre bugs and very serious security holes. A |
| 128 | /// good source of information is available in the following paper |
| 129 | /// (which includes additional references): |
Chris Lattner | 59907c4 | 2007-08-10 20:18:51 +0000 | [diff] [blame] | 130 | /// |
| 131 | /// FormatGuard: Automatic Protection From printf Format String |
| 132 | /// Vulnerabilities, Proceedings of the 10th USENIX Security Symposium, 2001. |
Ted Kremenek | 71895b9 | 2007-08-14 17:39:48 +0000 | [diff] [blame] | 133 | /// |
| 134 | /// Functionality implemented: |
| 135 | /// |
| 136 | /// We can statically check the following properties for string |
| 137 | /// literal format strings for non v.*printf functions (where the |
| 138 | /// arguments are passed directly): |
| 139 | // |
| 140 | /// (1) Are the number of format conversions equal to the number of |
| 141 | /// data arguments? |
| 142 | /// |
| 143 | /// (2) Does each format conversion correctly match the type of the |
| 144 | /// corresponding data argument? (TODO) |
| 145 | /// |
| 146 | /// Moreover, for all printf functions we can: |
| 147 | /// |
| 148 | /// (3) Check for a missing format string (when not caught by type checking). |
| 149 | /// |
| 150 | /// (4) Check for no-operation flags; e.g. using "#" with format |
| 151 | /// conversion 'c' (TODO) |
| 152 | /// |
| 153 | /// (5) Check the use of '%n', a major source of security holes. |
| 154 | /// |
| 155 | /// (6) Check for malformed format conversions that don't specify anything. |
| 156 | /// |
| 157 | /// (7) Check for empty format strings. e.g: printf(""); |
| 158 | /// |
| 159 | /// (8) Check that the format string is a wide literal. |
| 160 | /// |
| 161 | /// All of these checks can be done by parsing the format string. |
| 162 | /// |
| 163 | /// For now, we ONLY do (1), (3), (5), (6), (7), and (8). |
Chris Lattner | 59907c4 | 2007-08-10 20:18:51 +0000 | [diff] [blame] | 164 | void |
Ted Kremenek | 71895b9 | 2007-08-14 17:39:48 +0000 | [diff] [blame] | 165 | Sema::CheckPrintfArguments(Expr *Fn, |
| 166 | SourceLocation LParenLoc, SourceLocation RParenLoc, |
| 167 | bool HasVAListArg, FunctionDecl *FDecl, |
Ted Kremenek | 8207710 | 2007-08-10 21:21:05 +0000 | [diff] [blame] | 168 | unsigned format_idx, Expr** Args, |
| 169 | unsigned NumArgsInCall) { |
Ted Kremenek | 71895b9 | 2007-08-14 17:39:48 +0000 | [diff] [blame] | 170 | // CHECK: printf-like function is called with no format string. |
| 171 | if (format_idx >= NumArgsInCall) { |
| 172 | Diag(RParenLoc, diag::warn_printf_missing_format_string, |
| 173 | Fn->getSourceRange()); |
| 174 | return; |
| 175 | } |
| 176 | |
Chris Lattner | 59907c4 | 2007-08-10 20:18:51 +0000 | [diff] [blame] | 177 | // CHECK: format string is not a string literal. |
| 178 | // |
Ted Kremenek | 71895b9 | 2007-08-14 17:39:48 +0000 | [diff] [blame] | 179 | // Dynamically generated format strings are difficult to |
| 180 | // automatically vet at compile time. Requiring that format strings |
| 181 | // are string literals: (1) permits the checking of format strings by |
| 182 | // the compiler and thereby (2) can practically remove the source of |
| 183 | // many format string exploits. |
Chris Lattner | 59907c4 | 2007-08-10 20:18:51 +0000 | [diff] [blame] | 184 | StringLiteral *FExpr = dyn_cast<StringLiteral>(Args[format_idx]); |
| 185 | |
Ted Kremenek | 71895b9 | 2007-08-14 17:39:48 +0000 | [diff] [blame] | 186 | if (FExpr == NULL) { |
| 187 | Diag(Args[format_idx]->getLocStart(), |
| 188 | diag::warn_printf_not_string_constant, Fn->getSourceRange()); |
| 189 | return; |
| 190 | } |
| 191 | |
| 192 | // CHECK: is the format string a wide literal? |
| 193 | if (FExpr->isWide()) { |
| 194 | Diag(Args[format_idx]->getLocStart(), |
| 195 | diag::warn_printf_format_string_is_wide_literal, |
| 196 | Fn->getSourceRange()); |
| 197 | return; |
| 198 | } |
| 199 | |
| 200 | // Str - The format string. NOTE: this is NOT null-terminated! |
| 201 | const char * const Str = FExpr->getStrData(); |
| 202 | |
| 203 | // CHECK: empty format string? |
| 204 | const unsigned StrLen = FExpr->getByteLength(); |
| 205 | |
| 206 | if (StrLen == 0) { |
| 207 | Diag(Args[format_idx]->getLocStart(), |
| 208 | diag::warn_printf_empty_format_string, Fn->getSourceRange()); |
| 209 | return; |
| 210 | } |
| 211 | |
| 212 | // We process the format string using a binary state machine. The |
| 213 | // current state is stored in CurrentState. |
| 214 | enum { |
| 215 | state_OrdChr, |
| 216 | state_Conversion |
| 217 | } CurrentState = state_OrdChr; |
| 218 | |
| 219 | // numConversions - The number of conversions seen so far. This is |
| 220 | // incremented as we traverse the format string. |
| 221 | unsigned numConversions = 0; |
| 222 | |
| 223 | // numDataArgs - The number of data arguments after the format |
| 224 | // string. This can only be determined for non vprintf-like |
| 225 | // functions. For those functions, this value is 1 (the sole |
| 226 | // va_arg argument). |
| 227 | unsigned numDataArgs = NumArgsInCall-(format_idx+1); |
| 228 | |
| 229 | // Inspect the format string. |
| 230 | unsigned StrIdx = 0; |
| 231 | |
| 232 | // LastConversionIdx - Index within the format string where we last saw |
| 233 | // a '%' character that starts a new format conversion. |
| 234 | unsigned LastConversionIdx = 0; |
| 235 | |
| 236 | for ( ; StrIdx < StrLen ; ++StrIdx ) { |
| 237 | |
| 238 | // Is the number of detected conversion conversions greater than |
| 239 | // the number of matching data arguments? If so, stop. |
| 240 | if (!HasVAListArg && numConversions > numDataArgs) break; |
| 241 | |
| 242 | // Handle "\0" |
| 243 | if(Str[StrIdx] == '\0' ) { |
| 244 | // The string returned by getStrData() is not null-terminated, |
| 245 | // so the presence of a null character is likely an error. |
| 246 | |
| 247 | SourceLocation Loc = |
| 248 | PP.AdvanceToTokenCharacter(Args[format_idx]->getLocStart(),StrIdx+1); |
| 249 | |
| 250 | Diag(Loc, diag::warn_printf_format_string_contains_null_char, |
| 251 | Fn->getSourceRange()); |
| 252 | |
| 253 | return; |
| 254 | } |
| 255 | |
| 256 | // Ordinary characters (not processing a format conversion). |
| 257 | if (CurrentState == state_OrdChr) { |
| 258 | if (Str[StrIdx] == '%') { |
| 259 | CurrentState = state_Conversion; |
| 260 | LastConversionIdx = StrIdx; |
| 261 | } |
| 262 | continue; |
| 263 | } |
| 264 | |
| 265 | // Seen '%'. Now processing a format conversion. |
| 266 | switch (Str[StrIdx]) { |
| 267 | // Characters which can terminate a format conversion |
| 268 | // (e.g. "%d"). Characters that specify length modifiers or |
| 269 | // other flags are handled by the default case below. |
| 270 | // |
| 271 | // TODO: additional checks will go into the following cases. |
| 272 | case 'i': |
| 273 | case 'd': |
| 274 | case 'o': |
| 275 | case 'u': |
| 276 | case 'x': |
| 277 | case 'X': |
| 278 | case 'D': |
| 279 | case 'O': |
| 280 | case 'U': |
| 281 | case 'e': |
| 282 | case 'E': |
| 283 | case 'f': |
| 284 | case 'F': |
| 285 | case 'g': |
| 286 | case 'G': |
| 287 | case 'a': |
| 288 | case 'A': |
| 289 | case 'c': |
| 290 | case 'C': |
| 291 | case 'S': |
| 292 | case 's': |
| 293 | case 'P': |
| 294 | ++numConversions; |
| 295 | CurrentState = state_OrdChr; |
| 296 | break; |
| 297 | |
| 298 | // CHECK: Are we using "%n"? Issue a warning. |
| 299 | case 'n': { |
| 300 | ++numConversions; |
| 301 | CurrentState = state_OrdChr; |
| 302 | SourceLocation Loc = |
| 303 | PP.AdvanceToTokenCharacter(Args[format_idx]->getLocStart(), |
| 304 | LastConversionIdx+1); |
| 305 | |
| 306 | Diag(Loc, diag::warn_printf_write_back, Fn->getSourceRange()); |
| 307 | break; |
| 308 | } |
| 309 | |
| 310 | // Handle "%%" |
| 311 | case '%': |
| 312 | // Sanity check: Was the first "%" character the previous one? |
| 313 | // If not, we will assume that we have a malformed format |
| 314 | // conversion, and that the current "%" character is the start |
| 315 | // of a new conversion. |
| 316 | if (StrIdx - LastConversionIdx == 1) |
| 317 | CurrentState = state_OrdChr; |
| 318 | else { |
| 319 | // Issue a warning: invalid format conversion. |
| 320 | SourceLocation Loc = |
| 321 | PP.AdvanceToTokenCharacter(Args[format_idx]->getLocStart(), |
| 322 | LastConversionIdx+1); |
| 323 | |
| 324 | Diag(Loc, diag::warn_printf_invalid_conversion, |
| 325 | std::string(Str+LastConversionIdx, Str+StrIdx), |
| 326 | Fn->getSourceRange()); |
| 327 | |
| 328 | // This conversion is broken. Advance to the next format |
| 329 | // conversion. |
| 330 | LastConversionIdx = StrIdx; |
| 331 | ++numConversions; |
| 332 | } |
| 333 | |
| 334 | break; |
| 335 | |
| 336 | default: |
| 337 | // This case catches all other characters: flags, widths, etc. |
| 338 | // We should eventually process those as well. |
| 339 | break; |
| 340 | } |
| 341 | } |
| 342 | |
| 343 | if (CurrentState == state_Conversion) { |
| 344 | // Issue a warning: invalid format conversion. |
| 345 | SourceLocation Loc = |
| 346 | PP.AdvanceToTokenCharacter(Args[format_idx]->getLocStart(), |
| 347 | LastConversionIdx+1); |
| 348 | |
| 349 | Diag(Loc, diag::warn_printf_invalid_conversion, |
Ted Kremenek | 06de276 | 2007-08-17 16:46:58 +0000 | [diff] [blame^] | 350 | std::string(Str+LastConversionIdx, Str+StrIdx), |
Ted Kremenek | 71895b9 | 2007-08-14 17:39:48 +0000 | [diff] [blame] | 351 | Fn->getSourceRange()); |
| 352 | return; |
| 353 | } |
| 354 | |
| 355 | if (!HasVAListArg) { |
| 356 | // CHECK: Does the number of format conversions exceed the number |
| 357 | // of data arguments? |
| 358 | if (numConversions > numDataArgs) { |
| 359 | SourceLocation Loc = |
| 360 | PP.AdvanceToTokenCharacter(Args[format_idx]->getLocStart(), |
| 361 | LastConversionIdx); |
| 362 | |
| 363 | Diag(Loc, diag::warn_printf_insufficient_data_args, |
| 364 | Fn->getSourceRange()); |
| 365 | } |
| 366 | // CHECK: Does the number of data arguments exceed the number of |
| 367 | // format conversions in the format string? |
| 368 | else if (numConversions < numDataArgs) |
| 369 | Diag(Args[format_idx+numConversions+1]->getLocStart(), |
| 370 | diag::warn_printf_too_many_data_args, Fn->getSourceRange()); |
| 371 | } |
| 372 | } |
Ted Kremenek | 06de276 | 2007-08-17 16:46:58 +0000 | [diff] [blame^] | 373 | |
| 374 | //===--- CHECK: Return Address of Stack Variable --------------------------===// |
| 375 | |
| 376 | static DeclRefExpr* EvalVal(Expr *E); |
| 377 | static DeclRefExpr* EvalAddr(Expr* E); |
| 378 | |
| 379 | /// CheckReturnStackAddr - Check if a return statement returns the address |
| 380 | /// of a stack variable. |
| 381 | void |
| 382 | Sema::CheckReturnStackAddr(Expr *RetValExp, QualType lhsType, |
| 383 | SourceLocation ReturnLoc) { |
| 384 | |
| 385 | // Perform checking for returned stack addresses. |
| 386 | if (lhsType->isPointerType()) { |
| 387 | if (DeclRefExpr *DR = EvalAddr(RetValExp)) |
| 388 | Diag(DR->getLocStart(), diag::warn_ret_stack_addr, |
| 389 | DR->getDecl()->getIdentifier()->getName(), |
| 390 | RetValExp->getSourceRange()); |
| 391 | } |
| 392 | // Perform checking for stack values returned by reference. |
| 393 | else if (lhsType->isReferenceType()) { |
| 394 | if (DeclRefExpr *DR = EvalVal(RetValExp)) |
| 395 | Diag(DR->getLocStart(), diag::warn_ret_stack_ref, |
| 396 | DR->getDecl()->getIdentifier()->getName(), |
| 397 | RetValExp->getSourceRange()); |
| 398 | } |
| 399 | } |
| 400 | |
| 401 | /// EvalAddr - EvalAddr and EvalVal are mutually recursive functions that |
| 402 | /// check if the expression in a return statement evaluates to an address |
| 403 | /// to a location on the stack. The recursion is used to traverse the |
| 404 | /// AST of the return expression, with recursion backtracking when we |
| 405 | /// encounter a subexpression that (1) clearly does not lead to the address |
| 406 | /// of a stack variable or (2) is something we cannot determine leads to |
| 407 | /// the address of a stack variable based on such local checking. |
| 408 | /// |
| 409 | /// EvalAddr processes expressions that are pointers, and EvalVal handles |
| 410 | /// expressions that are rvalues or variable references. |
| 411 | /// At the base case of the recursion is a check for a DeclRefExpr* in |
| 412 | /// the refers to a stack variable. |
| 413 | /// |
| 414 | /// This implementation handles: |
| 415 | /// |
| 416 | /// * pointer-to-pointer casts |
| 417 | /// * implicit conversions from array references to pointers |
| 418 | /// * taking the address of fields |
| 419 | /// * arbitrary interplay between "&" and "*" operators |
| 420 | /// * pointer arithmetic from an address of a stack variable |
| 421 | /// * taking the address of an array element where the array is on the stack |
| 422 | static DeclRefExpr* EvalAddr(Expr *E) { |
| 423 | |
| 424 | // We should only be called for evaluating pointer expressions. |
| 425 | assert (E->getType()->isPointerType() && "EvalAddr only works on pointers"); |
| 426 | |
| 427 | // Our "symbolic interpreter" is just a dispatch off the currently |
| 428 | // viewed AST node. We then recursively traverse the AST by calling |
| 429 | // EvalAddr and EvalVal appropriately. |
| 430 | switch (E->getStmtClass()) { |
| 431 | |
| 432 | case Stmt::ParenExprClass: |
| 433 | // Ignore parentheses. |
| 434 | return EvalAddr(cast<ParenExpr>(E)->getSubExpr()); |
| 435 | |
| 436 | case Stmt::UnaryOperatorClass: { |
| 437 | // The only unary operator that make sense to handle here |
| 438 | // is AddrOf. All others don't make sense as pointers. |
| 439 | UnaryOperator *U = cast<UnaryOperator>(E); |
| 440 | |
| 441 | if (U->getOpcode() == UnaryOperator::AddrOf) |
| 442 | return EvalVal(U->getSubExpr()); |
| 443 | else |
| 444 | return NULL; |
| 445 | } |
| 446 | |
| 447 | case Stmt::BinaryOperatorClass: { |
| 448 | // Handle pointer arithmetic. All other binary operators are not valid |
| 449 | // in this context. |
| 450 | BinaryOperator *B = cast<BinaryOperator>(E); |
| 451 | BinaryOperator::Opcode op = B->getOpcode(); |
| 452 | |
| 453 | if (op != BinaryOperator::Add && op != BinaryOperator::Sub) |
| 454 | return NULL; |
| 455 | |
| 456 | Expr *Base = B->getLHS(); |
| 457 | |
| 458 | // Determine which argument is the real pointer base. It could be |
| 459 | // the RHS argument instead of the LHS. |
| 460 | if (!Base->getType()->isPointerType()) Base = B->getRHS(); |
| 461 | |
| 462 | assert (Base->getType()->isPointerType()); |
| 463 | return EvalAddr(Base); |
| 464 | } |
| 465 | |
| 466 | // For conditional operators we need to see if either the LHS or RHS are |
| 467 | // valid DeclRefExpr*s. If one of them is valid, we return it. |
| 468 | case Stmt::ConditionalOperatorClass: { |
| 469 | ConditionalOperator *C = cast<ConditionalOperator>(E); |
| 470 | |
| 471 | if (DeclRefExpr* LHS = EvalAddr(C->getLHS())) |
| 472 | return LHS; |
| 473 | else |
| 474 | return EvalAddr(C->getRHS()); |
| 475 | } |
| 476 | |
| 477 | // For implicit casts, we need to handle conversions from arrays to |
| 478 | // pointer values, and implicit pointer-to-pointer conversions. |
| 479 | case Stmt::ImplicitCastExprClass: { |
| 480 | ImplicitCastExpr *IE = cast<ImplicitCastExpr>(E); |
| 481 | Expr* SubExpr = IE->getSubExpr(); |
| 482 | |
| 483 | if (SubExpr->getType()->isPointerType()) |
| 484 | return EvalAddr(SubExpr); |
| 485 | else |
| 486 | return EvalVal(SubExpr); |
| 487 | } |
| 488 | |
| 489 | // For casts, we handle pointer-to-pointer conversions (which |
| 490 | // is essentially a no-op from our mini-interpreter's standpoint). |
| 491 | // For other casts we abort. |
| 492 | case Stmt::CastExprClass: { |
| 493 | CastExpr *C = cast<CastExpr>(E); |
| 494 | Expr *SubExpr = C->getSubExpr(); |
| 495 | |
| 496 | if (SubExpr->getType()->isPointerType()) |
| 497 | return EvalAddr(SubExpr); |
| 498 | else |
| 499 | return NULL; |
| 500 | } |
| 501 | |
| 502 | // TODO: C++ casts. |
| 503 | case Stmt::CXXCastExprClass: |
| 504 | return NULL; |
| 505 | |
| 506 | // Everything else: we simply don't reason about them. |
| 507 | default: |
| 508 | return NULL; |
| 509 | } |
| 510 | } |
| 511 | |
| 512 | |
| 513 | /// EvalVal - This function is complements EvalAddr in the mutual recursion. |
| 514 | /// See the comments for EvalAddr for more details. |
| 515 | static DeclRefExpr* EvalVal(Expr *E) { |
| 516 | |
| 517 | // We should only be called for evaluating non-pointer expressions. |
| 518 | assert (!E->getType()->isPointerType() && "EvalVal doesn't work on pointers"); |
| 519 | |
| 520 | // Our "symbolic interpreter" is just a dispatch off the currently |
| 521 | // viewed AST node. We then recursively traverse the AST by calling |
| 522 | // EvalAddr and EvalVal appropriately. |
| 523 | switch (E->getStmtClass()) { |
| 524 | |
| 525 | case Stmt::DeclRefExprClass: { |
| 526 | // DeclRefExpr: the base case. When we hit a DeclRefExpr we are looking |
| 527 | // at code that refers to a variable's name. We check if it has local |
| 528 | // storage within the function, and if so, return the expression. |
| 529 | DeclRefExpr *DR = cast<DeclRefExpr>(E); |
| 530 | |
| 531 | if (VarDecl *V = dyn_cast<VarDecl>(DR->getDecl())) |
| 532 | if(V->hasLocalStorage()) return DR; |
| 533 | |
| 534 | return NULL; |
| 535 | } |
| 536 | |
| 537 | case Stmt::ParenExprClass: |
| 538 | // Ignore parentheses. |
| 539 | return EvalVal(cast<ParenExpr>(E)->getSubExpr()); |
| 540 | |
| 541 | case Stmt::UnaryOperatorClass: { |
| 542 | // The only unary operator that make sense to handle here |
| 543 | // is Deref. All others don't resolve to a "name." This includes |
| 544 | // handling all sorts of rvalues passed to a unary operator. |
| 545 | UnaryOperator *U = cast<UnaryOperator>(E); |
| 546 | |
| 547 | if (U->getOpcode() == UnaryOperator::Deref) |
| 548 | return EvalAddr(U->getSubExpr()); |
| 549 | |
| 550 | return NULL; |
| 551 | } |
| 552 | |
| 553 | case Stmt::ArraySubscriptExprClass: { |
| 554 | // Array subscripts are potential references to data on the stack. We |
| 555 | // retrieve the DeclRefExpr* for the array variable if it indeed |
| 556 | // has local storage. |
| 557 | ArraySubscriptExpr *A = cast<ArraySubscriptExpr>(E); |
| 558 | |
| 559 | // The array access could be written A[4] or 4[A] (both are equivalent). |
| 560 | // In the second case, the "base" is the offset and the "Idx" is |
| 561 | // the base. We test for this case by seeing if the Base expression |
| 562 | // has a pointer type. |
| 563 | Expr* Base = A->getBase(); |
| 564 | |
| 565 | if (Base->getType()->isPointerType()) |
| 566 | return EvalAddr(Base); |
| 567 | else |
| 568 | return EvalAddr(A->getIdx()); |
| 569 | } |
| 570 | |
| 571 | case Stmt::ConditionalOperatorClass: { |
| 572 | // For conditional operators we need to see if either the LHS or RHS are |
| 573 | // non-NULL DeclRefExpr's. If one is non-NULL, we return it. |
| 574 | ConditionalOperator *C = cast<ConditionalOperator>(E); |
| 575 | |
| 576 | if (DeclRefExpr *LHS = EvalVal(C->getLHS())) |
| 577 | return LHS; |
| 578 | else |
| 579 | return EvalVal(C->getRHS()); |
| 580 | } |
| 581 | |
| 582 | // Accesses to members are potential references to data on the stack. |
| 583 | case Stmt::MemberExprClass: { |
| 584 | MemberExpr *M = cast<MemberExpr>(E); |
| 585 | |
| 586 | // Check for indirect access. We only want direct field accesses. |
| 587 | if (!M->isArrow()) |
| 588 | return EvalVal(M->getBase()); |
| 589 | else |
| 590 | return NULL; |
| 591 | } |
| 592 | |
| 593 | // Everything else: we simply don't reason about them. |
| 594 | default: |
| 595 | return NULL; |
| 596 | } |
| 597 | } |