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 | 71993dd | 2007-08-17 05:31:46 +0000 | [diff] [blame^] | 50 | if (i == e) return true; |
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, |
| 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 | |
| 77 | return true; |
| 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()); |
| 93 | return false; |
| 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 | |
| 115 | return true; |
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, |
| 350 | std::string(Str+LastConversionIdx, Str+StrIdx), |
| 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 | } |