PrintfFormatString.cpp revision e6ca97f2aeb301e28d20988b27c4297a9b540991 
1//== PrintfFormatString.cpp - Analysis of printf format strings --*- C++ -*-==//
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// Handling of format string in printf and friends.  The structure of format
11// strings for fprintf() are described in C99 7.19.6.1.
12//
13//===----------------------------------------------------------------------===//
14
15#include "clang/Analysis/Analyses/FormatString.h"
16#include "FormatStringParsing.h"
17
18using clang::analyze_format_string::ArgTypeResult;
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_printf::PrintfSpecifier;
24
25using namespace clang;
26
27typedef clang::analyze_format_string::SpecifierResult<PrintfSpecifier>
28        PrintfSpecifierResult;
29
30//===----------------------------------------------------------------------===//
31// Methods for parsing format strings.
32//===----------------------------------------------------------------------===//
33
34using analyze_format_string::ParseNonPositionAmount;
35
36static bool ParsePrecision(FormatStringHandler &H, PrintfSpecifier &FS,
37                           const char *Start, const char *&Beg, const char *E,
38                           unsigned *argIndex) {
39  if (argIndex) {
40    FS.setPrecision(ParseNonPositionAmount(Beg, E, *argIndex));
41  } else {
42    const OptionalAmount Amt = ParsePositionAmount(H, Start, Beg, E,
43                                           analyze_format_string::PrecisionPos);
44    if (Amt.isInvalid())
45      return true;
46    FS.setPrecision(Amt);
47  }
48  return false;
49}
50
51static PrintfSpecifierResult ParsePrintfSpecifier(FormatStringHandler &H,
52                                                  const char *&Beg,
53                                                  const char *E,
54                                                  unsigned &argIndex,
55                                                  const LangOptions &LO) {
56
57  using namespace clang::analyze_format_string;
58  using namespace clang::analyze_printf;
59
60  const char *I = Beg;
61  const char *Start = 0;
62  UpdateOnReturn <const char*> UpdateBeg(Beg, I);
63
64  // Look for a '%' character that indicates the start of a format specifier.
65  for ( ; I != E ; ++I) {
66    char c = *I;
67    if (c == '\0') {
68      // Detect spurious null characters, which are likely errors.
69      H.HandleNullChar(I);
70      return true;
71    }
72    if (c == '%') {
73      Start = I++;  // Record the start of the format specifier.
74      break;
75    }
76  }
77
78  // No format specifier found?
79  if (!Start)
80    return false;
81
82  if (I == E) {
83    // No more characters left?
84    H.HandleIncompleteSpecifier(Start, E - Start);
85    return true;
86  }
87
88  PrintfSpecifier FS;
89  if (ParseArgPosition(H, FS, Start, I, E))
90    return true;
91
92  if (I == E) {
93    // No more characters left?
94    H.HandleIncompleteSpecifier(Start, E - Start);
95    return true;
96  }
97
98  // Look for flags (if any).
99  bool hasMore = true;
100  for ( ; I != E; ++I) {
101    switch (*I) {
102      default: hasMore = false; break;
103      case '\'':
104        // FIXME: POSIX specific.  Always accept?
105        FS.setHasThousandsGrouping(I);
106        break;
107      case '-': FS.setIsLeftJustified(I); break;
108      case '+': FS.setHasPlusPrefix(I); break;
109      case ' ': FS.setHasSpacePrefix(I); break;
110      case '#': FS.setHasAlternativeForm(I); break;
111      case '0': FS.setHasLeadingZeros(I); break;
112    }
113    if (!hasMore)
114      break;
115  }
116
117  if (I == E) {
118    // No more characters left?
119    H.HandleIncompleteSpecifier(Start, E - Start);
120    return true;
121  }
122
123  // Look for the field width (if any).
124  if (ParseFieldWidth(H, FS, Start, I, E,
125                      FS.usesPositionalArg() ? 0 : &argIndex))
126    return true;
127
128  if (I == E) {
129    // No more characters left?
130    H.HandleIncompleteSpecifier(Start, E - Start);
131    return true;
132  }
133
134  // Look for the precision (if any).
135  if (*I == '.') {
136    ++I;
137    if (I == E) {
138      H.HandleIncompleteSpecifier(Start, E - Start);
139      return true;
140    }
141
142    if (ParsePrecision(H, FS, Start, I, E,
143                       FS.usesPositionalArg() ? 0 : &argIndex))
144      return true;
145
146    if (I == E) {
147      // No more characters left?
148      H.HandleIncompleteSpecifier(Start, E - Start);
149      return true;
150    }
151  }
152
153  // Look for the length modifier.
154  if (ParseLengthModifier(FS, I, E, LO) && I == E) {
155    // No more characters left?
156    H.HandleIncompleteSpecifier(Start, E - Start);
157    return true;
158  }
159
160  if (*I == '\0') {
161    // Detect spurious null characters, which are likely errors.
162    H.HandleNullChar(I);
163    return true;
164  }
165
166  // Finally, look for the conversion specifier.
167  const char *conversionPosition = I++;
168  ConversionSpecifier::Kind k = ConversionSpecifier::InvalidSpecifier;
169  switch (*conversionPosition) {
170    default:
171      break;
172    // C99: 7.19.6.1 (section 8).
173    case '%': k = ConversionSpecifier::PercentArg;   break;
174    case 'A': k = ConversionSpecifier::AArg; break;
175    case 'E': k = ConversionSpecifier::EArg; break;
176    case 'F': k = ConversionSpecifier::FArg; break;
177    case 'G': k = ConversionSpecifier::GArg; break;
178    case 'X': k = ConversionSpecifier::XArg; break;
179    case 'a': k = ConversionSpecifier::aArg; break;
180    case 'c': k = ConversionSpecifier::cArg; break;
181    case 'd': k = ConversionSpecifier::dArg; break;
182    case 'e': k = ConversionSpecifier::eArg; break;
183    case 'f': k = ConversionSpecifier::fArg; break;
184    case 'g': k = ConversionSpecifier::gArg; break;
185    case 'i': k = ConversionSpecifier::iArg; break;
186    case 'n': k = ConversionSpecifier::nArg; break;
187    case 'o': k = ConversionSpecifier::oArg; break;
188    case 'p': k = ConversionSpecifier::pArg;   break;
189    case 's': k = ConversionSpecifier::sArg;      break;
190    case 'u': k = ConversionSpecifier::uArg; break;
191    case 'x': k = ConversionSpecifier::xArg; break;
192    // POSIX specific.
193    case 'C': k = ConversionSpecifier::CArg; break;
194    case 'S': k = ConversionSpecifier::SArg; break;
195    // Objective-C.
196    case '@': k = ConversionSpecifier::ObjCObjArg; break;
197    // Glibc specific.
198    case 'm': k = ConversionSpecifier::PrintErrno; break;
199  }
200  PrintfConversionSpecifier CS(conversionPosition, k);
201  FS.setConversionSpecifier(CS);
202  if (CS.consumesDataArgument() && !FS.usesPositionalArg())
203    FS.setArgIndex(argIndex++);
204
205  if (k == ConversionSpecifier::InvalidSpecifier) {
206    // Assume the conversion takes one argument.
207    return !H.HandleInvalidPrintfConversionSpecifier(FS, Start, I - Start);
208  }
209  return PrintfSpecifierResult(Start, FS);
210}
211
212bool clang::analyze_format_string::ParsePrintfString(FormatStringHandler &H,
213                                                     const char *I,
214                                                     const char *E,
215                                                     const LangOptions &LO) {
216
217  unsigned argIndex = 0;
218
219  // Keep looking for a format specifier until we have exhausted the string.
220  while (I != E) {
221    const PrintfSpecifierResult &FSR = ParsePrintfSpecifier(H, I, E, argIndex,
222                                                            LO);
223    // Did a fail-stop error of any kind occur when parsing the specifier?
224    // If so, don't do any more processing.
225    if (FSR.shouldStop())
226      return true;;
227    // Did we exhaust the string or encounter an error that
228    // we can recover from?
229    if (!FSR.hasValue())
230      continue;
231    // We have a format specifier.  Pass it to the callback.
232    if (!H.HandlePrintfSpecifier(FSR.getValue(), FSR.getStart(),
233                                 I - FSR.getStart()))
234      return true;
235  }
236  assert(I == E && "Format string not exhausted");
237  return false;
238}
239
240//===----------------------------------------------------------------------===//
241// Methods on PrintfSpecifier.
242//===----------------------------------------------------------------------===//
243
244ArgTypeResult PrintfSpecifier::getArgType(ASTContext &Ctx) const {
245  const PrintfConversionSpecifier &CS = getConversionSpecifier();
246
247  if (!CS.consumesDataArgument())
248    return ArgTypeResult::Invalid();
249
250  if (CS.getKind() == ConversionSpecifier::cArg)
251    switch (LM.getKind()) {
252      case LengthModifier::None: return Ctx.IntTy;
253      case LengthModifier::AsLong:
254        return ArgTypeResult(ArgTypeResult::WIntTy, "wint_t");
255      default:
256        return ArgTypeResult::Invalid();
257    }
258
259  if (CS.isIntArg())
260    switch (LM.getKind()) {
261      case LengthModifier::AsLongDouble:
262        // GNU extension.
263        return Ctx.LongLongTy;
264      case LengthModifier::None: return Ctx.IntTy;
265      case LengthModifier::AsChar: return ArgTypeResult::AnyCharTy;
266      case LengthModifier::AsShort: return Ctx.ShortTy;
267      case LengthModifier::AsLong: return Ctx.LongTy;
268      case LengthModifier::AsLongLong: return Ctx.LongLongTy;
269      case LengthModifier::AsIntMax:
270        return ArgTypeResult(Ctx.getIntMaxType(), "intmax_t");
271      case LengthModifier::AsSizeT:
272        // FIXME: How to get the corresponding signed version of size_t?
273        return ArgTypeResult();
274      case LengthModifier::AsPtrDiff:
275        return ArgTypeResult(Ctx.getPointerDiffType(), "ptrdiff_t");
276      case LengthModifier::AsAllocate:
277      case LengthModifier::AsMAllocate:
278        return ArgTypeResult::Invalid();
279    }
280
281  if (CS.isUIntArg())
282    switch (LM.getKind()) {
283      case LengthModifier::AsLongDouble:
284        // GNU extension.
285        return Ctx.UnsignedLongLongTy;
286      case LengthModifier::None: return Ctx.UnsignedIntTy;
287      case LengthModifier::AsChar: return Ctx.UnsignedCharTy;
288      case LengthModifier::AsShort: return Ctx.UnsignedShortTy;
289      case LengthModifier::AsLong: return Ctx.UnsignedLongTy;
290      case LengthModifier::AsLongLong: return Ctx.UnsignedLongLongTy;
291      case LengthModifier::AsIntMax:
292        return ArgTypeResult(Ctx.getUIntMaxType(), "uintmax_t");
293      case LengthModifier::AsSizeT:
294        return ArgTypeResult(Ctx.getSizeType(), "size_t");
295      case LengthModifier::AsPtrDiff:
296        // FIXME: How to get the corresponding unsigned
297        // version of ptrdiff_t?
298        return ArgTypeResult();
299      case LengthModifier::AsAllocate:
300      case LengthModifier::AsMAllocate:
301        return ArgTypeResult::Invalid();
302    }
303
304  if (CS.isDoubleArg()) {
305    if (LM.getKind() == LengthModifier::AsLongDouble)
306      return Ctx.LongDoubleTy;
307    return Ctx.DoubleTy;
308  }
309
310  switch (CS.getKind()) {
311    case ConversionSpecifier::sArg:
312      if (LM.getKind() == LengthModifier::AsWideChar)
313        return ArgTypeResult(ArgTypeResult::WCStrTy, "wchar_t *");
314      return ArgTypeResult::CStrTy;
315    case ConversionSpecifier::SArg:
316      // FIXME: This appears to be Mac OS X specific.
317      return ArgTypeResult(ArgTypeResult::WCStrTy, "wchar_t *");
318    case ConversionSpecifier::CArg:
319      return ArgTypeResult(Ctx.WCharTy, "wchar_t");
320    case ConversionSpecifier::pArg:
321      return ArgTypeResult::CPointerTy;
322    case ConversionSpecifier::ObjCObjArg:
323      return ArgTypeResult::ObjCPointerTy;
324    default:
325      break;
326  }
327
328  // FIXME: Handle other cases.
329  return ArgTypeResult();
330}
331
332bool PrintfSpecifier::fixType(QualType QT, const LangOptions &LangOpt) {
333  // Handle strings first (char *, wchar_t *)
334  if (QT->isPointerType() && (QT->getPointeeType()->isAnyCharacterType())) {
335    CS.setKind(ConversionSpecifier::sArg);
336
337    // Disable irrelevant flags
338    HasAlternativeForm = 0;
339    HasLeadingZeroes = 0;
340
341    // Set the long length modifier for wide characters
342    if (QT->getPointeeType()->isWideCharType())
343      LM.setKind(LengthModifier::AsWideChar);
344    else
345      LM.setKind(LengthModifier::None);
346
347    return true;
348  }
349
350  // We can only work with builtin types.
351  const BuiltinType *BT = QT->getAs<BuiltinType>();
352  if (!BT)
353    return false;
354
355  // Set length modifier
356  switch (BT->getKind()) {
357  case BuiltinType::Bool:
358  case BuiltinType::WChar_U:
359  case BuiltinType::WChar_S:
360  case BuiltinType::Char16:
361  case BuiltinType::Char32:
362  case BuiltinType::UInt128:
363  case BuiltinType::Int128:
364  case BuiltinType::Half:
365    // Various types which are non-trivial to correct.
366    return false;
367
368#define SIGNED_TYPE(Id, SingletonId)
369#define UNSIGNED_TYPE(Id, SingletonId)
370#define FLOATING_TYPE(Id, SingletonId)
371#define BUILTIN_TYPE(Id, SingletonId) \
372  case BuiltinType::Id:
373#include "clang/AST/BuiltinTypes.def"
374    // Misc other stuff which doesn't make sense here.
375    return false;
376
377  case BuiltinType::UInt:
378  case BuiltinType::Int:
379  case BuiltinType::Float:
380  case BuiltinType::Double:
381    LM.setKind(LengthModifier::None);
382    break;
383
384  case BuiltinType::Char_U:
385  case BuiltinType::UChar:
386  case BuiltinType::Char_S:
387  case BuiltinType::SChar:
388    LM.setKind(LengthModifier::AsChar);
389    break;
390
391  case BuiltinType::Short:
392  case BuiltinType::UShort:
393    LM.setKind(LengthModifier::AsShort);
394    break;
395
396  case BuiltinType::Long:
397  case BuiltinType::ULong:
398    LM.setKind(LengthModifier::AsLong);
399    break;
400
401  case BuiltinType::LongLong:
402  case BuiltinType::ULongLong:
403    LM.setKind(LengthModifier::AsLongLong);
404    break;
405
406  case BuiltinType::LongDouble:
407    LM.setKind(LengthModifier::AsLongDouble);
408    break;
409  }
410
411  // Handle size_t, ptrdiff_t, etc. that have dedicated length modifiers in C99.
412  if (isa<TypedefType>(QT) && (LangOpt.C99 || LangOpt.CPlusPlus0x)) {
413    const IdentifierInfo *Identifier = QT.getBaseTypeIdentifier();
414    if (Identifier->getName() == "size_t") {
415      LM.setKind(LengthModifier::AsSizeT);
416    } else if (Identifier->getName() == "ssize_t") {
417      // Not C99, but common in Unix.
418      LM.setKind(LengthModifier::AsSizeT);
419    } else if (Identifier->getName() == "intmax_t") {
420      LM.setKind(LengthModifier::AsIntMax);
421    } else if (Identifier->getName() == "uintmax_t") {
422      LM.setKind(LengthModifier::AsIntMax);
423    } else if (Identifier->getName() == "ptrdiff_t") {
424      LM.setKind(LengthModifier::AsPtrDiff);
425    }
426  }
427
428  // Set conversion specifier and disable any flags which do not apply to it.
429  // Let typedefs to char fall through to int, as %c is silly for uint8_t.
430  if (isa<TypedefType>(QT) && QT->isAnyCharacterType()) {
431    CS.setKind(ConversionSpecifier::cArg);
432    LM.setKind(LengthModifier::None);
433    Precision.setHowSpecified(OptionalAmount::NotSpecified);
434    HasAlternativeForm = 0;
435    HasLeadingZeroes = 0;
436    HasPlusPrefix = 0;
437  }
438  // Test for Floating type first as LongDouble can pass isUnsignedIntegerType
439  else if (QT->isRealFloatingType()) {
440    CS.setKind(ConversionSpecifier::fArg);
441  }
442  else if (QT->isSignedIntegerType()) {
443    CS.setKind(ConversionSpecifier::dArg);
444    HasAlternativeForm = 0;
445  }
446  else if (QT->isUnsignedIntegerType()) {
447    // Preserve the original formatting, e.g. 'X', 'o'.
448    if (!cast<PrintfConversionSpecifier>(CS).isUIntArg())
449      CS.setKind(ConversionSpecifier::uArg);
450    HasAlternativeForm = 0;
451    HasPlusPrefix = 0;
452  } else {
453    llvm_unreachable("Unexpected type");
454  }
455
456  return true;
457}
458
459void PrintfSpecifier::toString(raw_ostream &os) const {
460  // Whilst some features have no defined order, we are using the order
461  // appearing in the C99 standard (ISO/IEC 9899:1999 (E) 7.19.6.1)
462  os << "%";
463
464  // Positional args
465  if (usesPositionalArg()) {
466    os << getPositionalArgIndex() << "$";
467  }
468
469  // Conversion flags
470  if (IsLeftJustified)    os << "-";
471  if (HasPlusPrefix)      os << "+";
472  if (HasSpacePrefix)     os << " ";
473  if (HasAlternativeForm) os << "#";
474  if (HasLeadingZeroes)   os << "0";
475
476  // Minimum field width
477  FieldWidth.toString(os);
478  // Precision
479  Precision.toString(os);
480  // Length modifier
481  os << LM.toString();
482  // Conversion specifier
483  os << CS.toString();
484}
485
486bool PrintfSpecifier::hasValidPlusPrefix() const {
487  if (!HasPlusPrefix)
488    return true;
489
490  // The plus prefix only makes sense for signed conversions
491  switch (CS.getKind()) {
492  case ConversionSpecifier::dArg:
493  case ConversionSpecifier::iArg:
494  case ConversionSpecifier::fArg:
495  case ConversionSpecifier::FArg:
496  case ConversionSpecifier::eArg:
497  case ConversionSpecifier::EArg:
498  case ConversionSpecifier::gArg:
499  case ConversionSpecifier::GArg:
500  case ConversionSpecifier::aArg:
501  case ConversionSpecifier::AArg:
502    return true;
503
504  default:
505    return false;
506  }
507}
508
509bool PrintfSpecifier::hasValidAlternativeForm() const {
510  if (!HasAlternativeForm)
511    return true;
512
513  // Alternate form flag only valid with the oxXaAeEfFgG conversions
514  switch (CS.getKind()) {
515  case ConversionSpecifier::oArg:
516  case ConversionSpecifier::xArg:
517  case ConversionSpecifier::XArg:
518  case ConversionSpecifier::aArg:
519  case ConversionSpecifier::AArg:
520  case ConversionSpecifier::eArg:
521  case ConversionSpecifier::EArg:
522  case ConversionSpecifier::fArg:
523  case ConversionSpecifier::FArg:
524  case ConversionSpecifier::gArg:
525  case ConversionSpecifier::GArg:
526    return true;
527
528  default:
529    return false;
530  }
531}
532
533bool PrintfSpecifier::hasValidLeadingZeros() const {
534  if (!HasLeadingZeroes)
535    return true;
536
537  // Leading zeroes flag only valid with the diouxXaAeEfFgG conversions
538  switch (CS.getKind()) {
539  case ConversionSpecifier::dArg:
540  case ConversionSpecifier::iArg:
541  case ConversionSpecifier::oArg:
542  case ConversionSpecifier::uArg:
543  case ConversionSpecifier::xArg:
544  case ConversionSpecifier::XArg:
545  case ConversionSpecifier::aArg:
546  case ConversionSpecifier::AArg:
547  case ConversionSpecifier::eArg:
548  case ConversionSpecifier::EArg:
549  case ConversionSpecifier::fArg:
550  case ConversionSpecifier::FArg:
551  case ConversionSpecifier::gArg:
552  case ConversionSpecifier::GArg:
553    return true;
554
555  default:
556    return false;
557  }
558}
559
560bool PrintfSpecifier::hasValidSpacePrefix() const {
561  if (!HasSpacePrefix)
562    return true;
563
564  // The space prefix only makes sense for signed conversions
565  switch (CS.getKind()) {
566  case ConversionSpecifier::dArg:
567  case ConversionSpecifier::iArg:
568  case ConversionSpecifier::fArg:
569  case ConversionSpecifier::FArg:
570  case ConversionSpecifier::eArg:
571  case ConversionSpecifier::EArg:
572  case ConversionSpecifier::gArg:
573  case ConversionSpecifier::GArg:
574  case ConversionSpecifier::aArg:
575  case ConversionSpecifier::AArg:
576    return true;
577
578  default:
579    return false;
580  }
581}
582
583bool PrintfSpecifier::hasValidLeftJustified() const {
584  if (!IsLeftJustified)
585    return true;
586
587  // The left justified flag is valid for all conversions except n
588  switch (CS.getKind()) {
589  case ConversionSpecifier::nArg:
590    return false;
591
592  default:
593    return true;
594  }
595}
596
597bool PrintfSpecifier::hasValidThousandsGroupingPrefix() const {
598  if (!HasThousandsGrouping)
599    return true;
600
601  switch (CS.getKind()) {
602    case ConversionSpecifier::dArg:
603    case ConversionSpecifier::iArg:
604    case ConversionSpecifier::uArg:
605    case ConversionSpecifier::fArg:
606    case ConversionSpecifier::FArg:
607    case ConversionSpecifier::gArg:
608    case ConversionSpecifier::GArg:
609      return true;
610    default:
611      return false;
612  }
613}
614
615bool PrintfSpecifier::hasValidPrecision() const {
616  if (Precision.getHowSpecified() == OptionalAmount::NotSpecified)
617    return true;
618
619  // Precision is only valid with the diouxXaAeEfFgGs conversions
620  switch (CS.getKind()) {
621  case ConversionSpecifier::dArg:
622  case ConversionSpecifier::iArg:
623  case ConversionSpecifier::oArg:
624  case ConversionSpecifier::uArg:
625  case ConversionSpecifier::xArg:
626  case ConversionSpecifier::XArg:
627  case ConversionSpecifier::aArg:
628  case ConversionSpecifier::AArg:
629  case ConversionSpecifier::eArg:
630  case ConversionSpecifier::EArg:
631  case ConversionSpecifier::fArg:
632  case ConversionSpecifier::FArg:
633  case ConversionSpecifier::gArg:
634  case ConversionSpecifier::GArg:
635  case ConversionSpecifier::sArg:
636    return true;
637
638  default:
639    return false;
640  }
641}
642bool PrintfSpecifier::hasValidFieldWidth() const {
643  if (FieldWidth.getHowSpecified() == OptionalAmount::NotSpecified)
644      return true;
645
646  // The field width is valid for all conversions except n
647  switch (CS.getKind()) {
648  case ConversionSpecifier::nArg:
649    return false;
650
651  default:
652    return true;
653  }
654}
655