1//= ScanfFormatString.cpp - Analysis of printf format strings --*- C++ -*-===//
2//
3// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
4// See https://llvm.org/LICENSE.txt for license information.
5// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
6//
7//===----------------------------------------------------------------------===//
8//
9// Handling of format string in scanf and friends. The structure of format
10// strings for fscanf() are described in C99 7.19.6.2.
11//
12//===----------------------------------------------------------------------===//
13
14#include "clang/AST/FormatString.h"
15#include "FormatStringParsing.h"
16#include "clang/Basic/TargetInfo.h"
17
18using clang::analyze_format_string::ArgType;
19using clang::analyze_format_string::FormatStringHandler;
20using clang::analyze_format_string::LengthModifier;
21using clang::analyze_format_string::OptionalAmount;
22using clang::analyze_format_string::ConversionSpecifier;
23using clang::analyze_scanf::ScanfConversionSpecifier;
24using clang::analyze_scanf::ScanfSpecifier;
25using clang::UpdateOnReturn;
26using namespace clang;
27
28typedef clang::analyze_format_string::SpecifierResult<ScanfSpecifier>
29 ScanfSpecifierResult;
30
31static bool ParseScanList(FormatStringHandler &H,
32 ScanfConversionSpecifier &CS,
33 const char *&Beg, const char *E) {
34 const char *I = Beg;
35 const char *start = I - 1;
36 UpdateOnReturn <const char*> UpdateBeg(Beg, I);
37
38 // No more characters?
39 if (I == E) {
40 H.HandleIncompleteScanList(start, end: I);
41 return true;
42 }
43
44 // Special case: ']' is the first character.
45 if (*I == ']') {
46 if (++I == E) {
47 H.HandleIncompleteScanList(start, end: I - 1);
48 return true;
49 }
50 }
51
52 // Special case: "^]" are the first characters.
53 if (I + 1 != E && I[0] == '^' && I[1] == ']') {
54 I += 2;
55 if (I == E) {
56 H.HandleIncompleteScanList(start, end: I - 1);
57 return true;
58 }
59 }
60
61 // Look for a ']' character which denotes the end of the scan list.
62 while (*I != ']') {
63 if (++I == E) {
64 H.HandleIncompleteScanList(start, end: I - 1);
65 return true;
66 }
67 }
68
69 CS.setEndScanList(I);
70 return false;
71}
72
73// FIXME: Much of this is copy-paste from ParsePrintfSpecifier.
74// We can possibly refactor.
75static ScanfSpecifierResult ParseScanfSpecifier(FormatStringHandler &H,
76 const char *&Beg,
77 const char *E,
78 unsigned &argIndex,
79 const LangOptions &LO,
80 const TargetInfo &Target) {
81 using namespace clang::analyze_format_string;
82 using namespace clang::analyze_scanf;
83 const char *I = Beg;
84 const char *Start = nullptr;
85 UpdateOnReturn <const char*> UpdateBeg(Beg, I);
86
87 // Look for a '%' character that indicates the start of a format specifier.
88 for ( ; I != E ; ++I) {
89 char c = *I;
90 if (c == '\0') {
91 // Detect spurious null characters, which are likely errors.
92 H.HandleNullChar(nullCharacter: I);
93 return true;
94 }
95 if (c == '%') {
96 Start = I++; // Record the start of the format specifier.
97 break;
98 }
99 }
100
101 // No format specifier found?
102 if (!Start)
103 return false;
104
105 if (I == E) {
106 // No more characters left?
107 H.HandleIncompleteSpecifier(startSpecifier: Start, specifierLen: E - Start);
108 return true;
109 }
110
111 ScanfSpecifier FS;
112 if (ParseArgPosition(H, CS&: FS, Start, Beg&: I, E))
113 return true;
114
115 if (I == E) {
116 // No more characters left?
117 H.HandleIncompleteSpecifier(startSpecifier: Start, specifierLen: E - Start);
118 return true;
119 }
120
121 // Look for '*' flag if it is present.
122 if (*I == '*') {
123 FS.setSuppressAssignment(I);
124 if (++I == E) {
125 H.HandleIncompleteSpecifier(startSpecifier: Start, specifierLen: E - Start);
126 return true;
127 }
128 }
129
130 // Look for the field width (if any). Unlike printf, this is either
131 // a fixed integer or isn't present.
132 const OptionalAmount &Amt = clang::analyze_format_string::ParseAmount(Beg&: I, E);
133 if (Amt.getHowSpecified() != OptionalAmount::NotSpecified) {
134 assert(Amt.getHowSpecified() == OptionalAmount::Constant);
135 FS.setFieldWidth(Amt);
136
137 if (I == E) {
138 // No more characters left?
139 H.HandleIncompleteSpecifier(startSpecifier: Start, specifierLen: E - Start);
140 return true;
141 }
142 }
143
144 // Look for the length modifier.
145 if (ParseLengthModifier(FS, Beg&: I, E, LO, /*IsScanf=*/true) && I == E) {
146 // No more characters left?
147 H.HandleIncompleteSpecifier(startSpecifier: Start, specifierLen: E - Start);
148 return true;
149 }
150
151 // Detect spurious null characters, which are likely errors.
152 if (*I == '\0') {
153 H.HandleNullChar(nullCharacter: I);
154 return true;
155 }
156
157 // Finally, look for the conversion specifier.
158 const char *conversionPosition = I++;
159 ScanfConversionSpecifier::Kind k = ScanfConversionSpecifier::InvalidSpecifier;
160 switch (*conversionPosition) {
161 default:
162 break;
163 case '%': k = ConversionSpecifier::PercentArg; break;
164 case 'b': k = ConversionSpecifier::bArg; break;
165 case 'A': k = ConversionSpecifier::AArg; break;
166 case 'E': k = ConversionSpecifier::EArg; break;
167 case 'F': k = ConversionSpecifier::FArg; break;
168 case 'G': k = ConversionSpecifier::GArg; break;
169 case 'X': k = ConversionSpecifier::XArg; break;
170 case 'a': k = ConversionSpecifier::aArg; break;
171 case 'd': k = ConversionSpecifier::dArg; break;
172 case 'e': k = ConversionSpecifier::eArg; break;
173 case 'f': k = ConversionSpecifier::fArg; break;
174 case 'g': k = ConversionSpecifier::gArg; break;
175 case 'i': k = ConversionSpecifier::iArg; break;
176 case 'n': k = ConversionSpecifier::nArg; break;
177 case 'c': k = ConversionSpecifier::cArg; break;
178 case 'C': k = ConversionSpecifier::CArg; break;
179 case 'S': k = ConversionSpecifier::SArg; break;
180 case '[': k = ConversionSpecifier::ScanListArg; break;
181 case 'u': k = ConversionSpecifier::uArg; break;
182 case 'x': k = ConversionSpecifier::xArg; break;
183 case 'o': k = ConversionSpecifier::oArg; break;
184 case 's': k = ConversionSpecifier::sArg; break;
185 case 'p': k = ConversionSpecifier::pArg; break;
186 // Apple extensions
187 // Apple-specific
188 case 'D':
189 if (Target.getTriple().isOSDarwin())
190 k = ConversionSpecifier::DArg;
191 break;
192 case 'O':
193 if (Target.getTriple().isOSDarwin())
194 k = ConversionSpecifier::OArg;
195 break;
196 case 'U':
197 if (Target.getTriple().isOSDarwin())
198 k = ConversionSpecifier::UArg;
199 break;
200 }
201 ScanfConversionSpecifier CS(conversionPosition, k);
202 if (k == ScanfConversionSpecifier::ScanListArg) {
203 if (ParseScanList(H, CS, Beg&: I, E))
204 return true;
205 }
206 FS.setConversionSpecifier(CS);
207 if (CS.consumesDataArgument() && !FS.getSuppressAssignment()
208 && !FS.usesPositionalArg())
209 FS.setArgIndex(argIndex++);
210
211 // FIXME: '%' and '*' doesn't make sense. Issue a warning.
212 // FIXME: 'ConsumedSoFar' and '*' doesn't make sense.
213
214 if (k == ScanfConversionSpecifier::InvalidSpecifier) {
215 unsigned Len = I - Beg;
216 if (ParseUTF8InvalidSpecifier(SpecifierBegin: Beg, FmtStrEnd: E, Len)) {
217 CS.setEndScanList(Beg + Len);
218 FS.setConversionSpecifier(CS);
219 }
220 // Assume the conversion takes one argument.
221 return !H.HandleInvalidScanfConversionSpecifier(FS, startSpecifier: Beg, specifierLen: Len);
222 }
223 return ScanfSpecifierResult(Start, FS);
224}
225
226ArgType ScanfSpecifier::getArgType(ASTContext &Ctx) const {
227 const ScanfConversionSpecifier &CS = getConversionSpecifier();
228
229 if (!CS.consumesDataArgument())
230 return ArgType::Invalid();
231
232 switch(CS.getKind()) {
233 // Signed int.
234 case ConversionSpecifier::dArg:
235 case ConversionSpecifier::DArg:
236 case ConversionSpecifier::iArg:
237 switch (LM.getKind()) {
238 case LengthModifier::None:
239 return ArgType::PtrTo(A: Ctx.IntTy);
240 case LengthModifier::AsChar:
241 return ArgType::PtrTo(A: ArgType::AnyCharTy);
242 case LengthModifier::AsShort:
243 return ArgType::PtrTo(A: Ctx.ShortTy);
244 case LengthModifier::AsLong:
245 return ArgType::PtrTo(A: Ctx.LongTy);
246 case LengthModifier::AsLongLong:
247 case LengthModifier::AsQuad:
248 return ArgType::PtrTo(A: Ctx.LongLongTy);
249 case LengthModifier::AsInt64:
250 return ArgType::PtrTo(A: ArgType(Ctx.LongLongTy, "__int64"));
251 case LengthModifier::AsIntMax:
252 return ArgType::PtrTo(A: ArgType(Ctx.getIntMaxType(), "intmax_t"));
253 case LengthModifier::AsSizeT:
254 return ArgType::PtrTo(A: ArgType(Ctx.getSignedSizeType(), "ssize_t"));
255 case LengthModifier::AsPtrDiff:
256 return ArgType::PtrTo(A: ArgType(Ctx.getPointerDiffType(), "ptrdiff_t"));
257 case LengthModifier::AsLongDouble:
258 // GNU extension.
259 return ArgType::PtrTo(A: Ctx.LongLongTy);
260 case LengthModifier::AsAllocate:
261 case LengthModifier::AsMAllocate:
262 case LengthModifier::AsInt32:
263 case LengthModifier::AsInt3264:
264 case LengthModifier::AsWide:
265 case LengthModifier::AsShortLong:
266 return ArgType::Invalid();
267 }
268 llvm_unreachable("Unsupported LengthModifier Type");
269
270 // Unsigned int.
271 case ConversionSpecifier::bArg:
272 case ConversionSpecifier::oArg:
273 case ConversionSpecifier::OArg:
274 case ConversionSpecifier::uArg:
275 case ConversionSpecifier::UArg:
276 case ConversionSpecifier::xArg:
277 case ConversionSpecifier::XArg:
278 switch (LM.getKind()) {
279 case LengthModifier::None:
280 return ArgType::PtrTo(A: Ctx.UnsignedIntTy);
281 case LengthModifier::AsChar:
282 return ArgType::PtrTo(A: Ctx.UnsignedCharTy);
283 case LengthModifier::AsShort:
284 return ArgType::PtrTo(A: Ctx.UnsignedShortTy);
285 case LengthModifier::AsLong:
286 return ArgType::PtrTo(A: Ctx.UnsignedLongTy);
287 case LengthModifier::AsLongLong:
288 case LengthModifier::AsQuad:
289 return ArgType::PtrTo(A: Ctx.UnsignedLongLongTy);
290 case LengthModifier::AsInt64:
291 return ArgType::PtrTo(A: ArgType(Ctx.UnsignedLongLongTy, "unsigned __int64"));
292 case LengthModifier::AsIntMax:
293 return ArgType::PtrTo(A: ArgType(Ctx.getUIntMaxType(), "uintmax_t"));
294 case LengthModifier::AsSizeT:
295 return ArgType::PtrTo(A: ArgType(Ctx.getSizeType(), "size_t"));
296 case LengthModifier::AsPtrDiff:
297 return ArgType::PtrTo(
298 A: ArgType(Ctx.getUnsignedPointerDiffType(), "unsigned ptrdiff_t"));
299 case LengthModifier::AsLongDouble:
300 // GNU extension.
301 return ArgType::PtrTo(A: Ctx.UnsignedLongLongTy);
302 case LengthModifier::AsAllocate:
303 case LengthModifier::AsMAllocate:
304 case LengthModifier::AsInt32:
305 case LengthModifier::AsInt3264:
306 case LengthModifier::AsWide:
307 case LengthModifier::AsShortLong:
308 return ArgType::Invalid();
309 }
310 llvm_unreachable("Unsupported LengthModifier Type");
311
312 // Float.
313 case ConversionSpecifier::aArg:
314 case ConversionSpecifier::AArg:
315 case ConversionSpecifier::eArg:
316 case ConversionSpecifier::EArg:
317 case ConversionSpecifier::fArg:
318 case ConversionSpecifier::FArg:
319 case ConversionSpecifier::gArg:
320 case ConversionSpecifier::GArg:
321 switch (LM.getKind()) {
322 case LengthModifier::None:
323 return ArgType::PtrTo(A: Ctx.FloatTy);
324 case LengthModifier::AsLong:
325 return ArgType::PtrTo(A: Ctx.DoubleTy);
326 case LengthModifier::AsLongDouble:
327 return ArgType::PtrTo(A: Ctx.LongDoubleTy);
328 default:
329 return ArgType::Invalid();
330 }
331
332 // Char, string and scanlist.
333 case ConversionSpecifier::cArg:
334 case ConversionSpecifier::sArg:
335 case ConversionSpecifier::ScanListArg:
336 switch (LM.getKind()) {
337 case LengthModifier::None:
338 return ArgType::PtrTo(A: ArgType::AnyCharTy);
339 case LengthModifier::AsLong:
340 case LengthModifier::AsWide:
341 return ArgType::PtrTo(A: ArgType(Ctx.getWideCharType(), "wchar_t"));
342 case LengthModifier::AsAllocate:
343 case LengthModifier::AsMAllocate:
344 return ArgType::PtrTo(A: ArgType::CStrTy);
345 case LengthModifier::AsShort:
346 if (Ctx.getTargetInfo().getTriple().isOSMSVCRT())
347 return ArgType::PtrTo(A: ArgType::AnyCharTy);
348 [[fallthrough]];
349 default:
350 return ArgType::Invalid();
351 }
352 case ConversionSpecifier::CArg:
353 case ConversionSpecifier::SArg:
354 // FIXME: Mac OS X specific?
355 switch (LM.getKind()) {
356 case LengthModifier::None:
357 case LengthModifier::AsWide:
358 return ArgType::PtrTo(A: ArgType(Ctx.getWideCharType(), "wchar_t"));
359 case LengthModifier::AsAllocate:
360 case LengthModifier::AsMAllocate:
361 return ArgType::PtrTo(A: ArgType(ArgType::WCStrTy, "wchar_t *"));
362 case LengthModifier::AsShort:
363 if (Ctx.getTargetInfo().getTriple().isOSMSVCRT())
364 return ArgType::PtrTo(A: ArgType::AnyCharTy);
365 [[fallthrough]];
366 default:
367 return ArgType::Invalid();
368 }
369
370 // Pointer.
371 case ConversionSpecifier::pArg:
372 return ArgType::PtrTo(A: ArgType::CPointerTy);
373
374 // Write-back.
375 case ConversionSpecifier::nArg:
376 switch (LM.getKind()) {
377 case LengthModifier::None:
378 return ArgType::PtrTo(A: Ctx.IntTy);
379 case LengthModifier::AsChar:
380 return ArgType::PtrTo(A: Ctx.SignedCharTy);
381 case LengthModifier::AsShort:
382 return ArgType::PtrTo(A: Ctx.ShortTy);
383 case LengthModifier::AsLong:
384 return ArgType::PtrTo(A: Ctx.LongTy);
385 case LengthModifier::AsLongLong:
386 case LengthModifier::AsQuad:
387 return ArgType::PtrTo(A: Ctx.LongLongTy);
388 case LengthModifier::AsInt64:
389 return ArgType::PtrTo(A: ArgType(Ctx.LongLongTy, "__int64"));
390 case LengthModifier::AsIntMax:
391 return ArgType::PtrTo(A: ArgType(Ctx.getIntMaxType(), "intmax_t"));
392 case LengthModifier::AsSizeT:
393 return ArgType::PtrTo(A: ArgType(Ctx.getSignedSizeType(), "ssize_t"));
394 case LengthModifier::AsPtrDiff:
395 return ArgType::PtrTo(A: ArgType(Ctx.getPointerDiffType(), "ptrdiff_t"));
396 case LengthModifier::AsLongDouble:
397 return ArgType(); // FIXME: Is this a known extension?
398 case LengthModifier::AsAllocate:
399 case LengthModifier::AsMAllocate:
400 case LengthModifier::AsInt32:
401 case LengthModifier::AsInt3264:
402 case LengthModifier::AsWide:
403 case LengthModifier::AsShortLong:
404 return ArgType::Invalid();
405 }
406
407 default:
408 break;
409 }
410
411 return ArgType();
412}
413
414bool ScanfSpecifier::fixType(QualType QT, QualType RawQT,
415 const LangOptions &LangOpt,
416 ASTContext &Ctx) {
417
418 // %n is different from other conversion specifiers; don't try to fix it.
419 if (CS.getKind() == ConversionSpecifier::nArg)
420 return false;
421
422 if (!QT->isPointerType())
423 return false;
424
425 QualType PT = QT->getPointeeType();
426
427 // If it's an enum, get its underlying type.
428 if (const EnumType *ETy = PT->getAs<EnumType>()) {
429 // Don't try to fix incomplete enums.
430 if (!ETy->getDecl()->isComplete())
431 return false;
432 PT = ETy->getDecl()->getIntegerType();
433 }
434
435 const BuiltinType *BT = PT->getAs<BuiltinType>();
436 if (!BT)
437 return false;
438
439 // Pointer to a character.
440 if (PT->isAnyCharacterType()) {
441 CS.setKind(ConversionSpecifier::sArg);
442 if (PT->isWideCharType())
443 LM.setKind(LengthModifier::AsWideChar);
444 else
445 LM.setKind(LengthModifier::None);
446
447 // If we know the target array length, we can use it as a field width.
448 if (const ConstantArrayType *CAT = Ctx.getAsConstantArrayType(T: RawQT)) {
449 if (CAT->getSizeModifier() == ArraySizeModifier::Normal)
450 FieldWidth = OptionalAmount(OptionalAmount::Constant,
451 CAT->getZExtSize() - 1, "", 0, false);
452 }
453 return true;
454 }
455
456 // Figure out the length modifier.
457 switch (BT->getKind()) {
458 // no modifier
459 case BuiltinType::UInt:
460 case BuiltinType::Int:
461 case BuiltinType::Float:
462 LM.setKind(LengthModifier::None);
463 break;
464
465 // hh
466 case BuiltinType::Char_U:
467 case BuiltinType::UChar:
468 case BuiltinType::Char_S:
469 case BuiltinType::SChar:
470 LM.setKind(LengthModifier::AsChar);
471 break;
472
473 // h
474 case BuiltinType::Short:
475 case BuiltinType::UShort:
476 LM.setKind(LengthModifier::AsShort);
477 break;
478
479 // l
480 case BuiltinType::Long:
481 case BuiltinType::ULong:
482 case BuiltinType::Double:
483 LM.setKind(LengthModifier::AsLong);
484 break;
485
486 // ll
487 case BuiltinType::LongLong:
488 case BuiltinType::ULongLong:
489 LM.setKind(LengthModifier::AsLongLong);
490 break;
491
492 // L
493 case BuiltinType::LongDouble:
494 LM.setKind(LengthModifier::AsLongDouble);
495 break;
496
497 // Don't know.
498 default:
499 return false;
500 }
501
502 // Handle size_t, ptrdiff_t, etc. that have dedicated length modifiers in C99.
503 if (LangOpt.C99 || LangOpt.CPlusPlus11)
504 namedTypeToLengthModifier(QT: PT, LM);
505
506 // If fixing the length modifier was enough, we are done.
507 if (hasValidLengthModifier(Target: Ctx.getTargetInfo(), LO: LangOpt)) {
508 const analyze_scanf::ArgType &AT = getArgType(Ctx);
509 if (AT.isValid() && AT.matchesType(C&: Ctx, argTy: QT))
510 return true;
511 }
512
513 // Figure out the conversion specifier.
514 if (PT->isRealFloatingType())
515 CS.setKind(ConversionSpecifier::fArg);
516 else if (PT->isSignedIntegerType())
517 CS.setKind(ConversionSpecifier::dArg);
518 else if (PT->isUnsignedIntegerType())
519 CS.setKind(ConversionSpecifier::uArg);
520 else
521 llvm_unreachable("Unexpected type");
522
523 return true;
524}
525
526void ScanfSpecifier::toString(raw_ostream &os) const {
527 os << "%";
528
529 if (usesPositionalArg())
530 os << getPositionalArgIndex() << "$";
531 if (SuppressAssignment)
532 os << "*";
533
534 FieldWidth.toString(os);
535 os << LM.toString();
536 os << CS.toString();
537}
538
539bool clang::analyze_format_string::ParseScanfString(FormatStringHandler &H,
540 const char *I,
541 const char *E,
542 const LangOptions &LO,
543 const TargetInfo &Target) {
544
545 unsigned argIndex = 0;
546
547 // Keep looking for a format specifier until we have exhausted the string.
548 while (I != E) {
549 const ScanfSpecifierResult &FSR = ParseScanfSpecifier(H, Beg&: I, E, argIndex,
550 LO, Target);
551 // Did a fail-stop error of any kind occur when parsing the specifier?
552 // If so, don't do any more processing.
553 if (FSR.shouldStop())
554 return true;
555 // Did we exhaust the string or encounter an error that
556 // we can recover from?
557 if (!FSR.hasValue())
558 continue;
559 // We have a format specifier. Pass it to the callback.
560 if (!H.HandleScanfSpecifier(FS: FSR.getValue(), startSpecifier: FSR.getStart(),
561 specifierLen: I - FSR.getStart())) {
562 return true;
563 }
564 }
565 assert(I == E && "Format string not exhausted");
566 return false;
567}
568