PrintfFormatString.cpp revision be6126a2a784e1446460b8d15c2b26f880c871fc 
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,
245                                          bool IsObjCLiteral) const {
246  const PrintfConversionSpecifier &CS = getConversionSpecifier();
247
248  if (!CS.consumesDataArgument())
249    return ArgTypeResult::Invalid();
250
251  if (CS.getKind() == ConversionSpecifier::cArg)
252    switch (LM.getKind()) {
253      case LengthModifier::None: return Ctx.IntTy;
254      case LengthModifier::AsLong:
255        return ArgTypeResult(ArgTypeResult::WIntTy, "wint_t");
256      default:
257        return ArgTypeResult::Invalid();
258    }
259
260  if (CS.isIntArg())
261    switch (LM.getKind()) {
262      case LengthModifier::AsLongDouble:
263        // GNU extension.
264        return Ctx.LongLongTy;
265      case LengthModifier::None: return Ctx.IntTy;
266      case LengthModifier::AsChar: return ArgTypeResult::AnyCharTy;
267      case LengthModifier::AsShort: return Ctx.ShortTy;
268      case LengthModifier::AsLong: return Ctx.LongTy;
269      case LengthModifier::AsLongLong: return Ctx.LongLongTy;
270      case LengthModifier::AsIntMax:
271        return ArgTypeResult(Ctx.getIntMaxType(), "intmax_t");
272      case LengthModifier::AsSizeT:
273        // FIXME: How to get the corresponding signed version of size_t?
274        return ArgTypeResult();
275      case LengthModifier::AsPtrDiff:
276        return ArgTypeResult(Ctx.getPointerDiffType(), "ptrdiff_t");
277      case LengthModifier::AsAllocate:
278      case LengthModifier::AsMAllocate:
279        return ArgTypeResult::Invalid();
280    }
281
282  if (CS.isUIntArg())
283    switch (LM.getKind()) {
284      case LengthModifier::AsLongDouble:
285        // GNU extension.
286        return Ctx.UnsignedLongLongTy;
287      case LengthModifier::None: return Ctx.UnsignedIntTy;
288      case LengthModifier::AsChar: return Ctx.UnsignedCharTy;
289      case LengthModifier::AsShort: return Ctx.UnsignedShortTy;
290      case LengthModifier::AsLong: return Ctx.UnsignedLongTy;
291      case LengthModifier::AsLongLong: return Ctx.UnsignedLongLongTy;
292      case LengthModifier::AsIntMax:
293        return ArgTypeResult(Ctx.getUIntMaxType(), "uintmax_t");
294      case LengthModifier::AsSizeT:
295        return ArgTypeResult(Ctx.getSizeType(), "size_t");
296      case LengthModifier::AsPtrDiff:
297        // FIXME: How to get the corresponding unsigned
298        // version of ptrdiff_t?
299        return ArgTypeResult();
300      case LengthModifier::AsAllocate:
301      case LengthModifier::AsMAllocate:
302        return ArgTypeResult::Invalid();
303    }
304
305  if (CS.isDoubleArg()) {
306    if (LM.getKind() == LengthModifier::AsLongDouble)
307      return Ctx.LongDoubleTy;
308    return Ctx.DoubleTy;
309  }
310
311  switch (CS.getKind()) {
312    case ConversionSpecifier::sArg:
313      if (LM.getKind() == LengthModifier::AsWideChar) {
314        if (IsObjCLiteral)
315          return Ctx.getPointerType(Ctx.UnsignedShortTy.withConst());
316        return ArgTypeResult(ArgTypeResult::WCStrTy, "wchar_t *");
317      }
318      return ArgTypeResult::CStrTy;
319    case ConversionSpecifier::SArg:
320      if (IsObjCLiteral)
321        return Ctx.getPointerType(Ctx.UnsignedShortTy.withConst());
322      return ArgTypeResult(ArgTypeResult::WCStrTy, "wchar_t *");
323    case ConversionSpecifier::CArg:
324      if (IsObjCLiteral)
325        return Ctx.UnsignedShortTy;
326      return ArgTypeResult(Ctx.WCharTy, "wchar_t");
327    case ConversionSpecifier::pArg:
328      return ArgTypeResult::CPointerTy;
329    case ConversionSpecifier::ObjCObjArg:
330      return ArgTypeResult::ObjCPointerTy;
331    default:
332      break;
333  }
334
335  // FIXME: Handle other cases.
336  return ArgTypeResult();
337}
338
339bool PrintfSpecifier::fixType(QualType QT, const LangOptions &LangOpt,
340                              ASTContext &Ctx, bool IsObjCLiteral) {
341  // Handle strings first (char *, wchar_t *)
342  if (QT->isPointerType() && (QT->getPointeeType()->isAnyCharacterType())) {
343    CS.setKind(ConversionSpecifier::sArg);
344
345    // Disable irrelevant flags
346    HasAlternativeForm = 0;
347    HasLeadingZeroes = 0;
348
349    // Set the long length modifier for wide characters
350    if (QT->getPointeeType()->isWideCharType())
351      LM.setKind(LengthModifier::AsWideChar);
352    else
353      LM.setKind(LengthModifier::None);
354
355    return true;
356  }
357
358  // We can only work with builtin types.
359  const BuiltinType *BT = QT->getAs<BuiltinType>();
360  if (!BT)
361    return false;
362
363  // Set length modifier
364  switch (BT->getKind()) {
365  case BuiltinType::Bool:
366  case BuiltinType::WChar_U:
367  case BuiltinType::WChar_S:
368  case BuiltinType::Char16:
369  case BuiltinType::Char32:
370  case BuiltinType::UInt128:
371  case BuiltinType::Int128:
372  case BuiltinType::Half:
373    // Various types which are non-trivial to correct.
374    return false;
375
376#define SIGNED_TYPE(Id, SingletonId)
377#define UNSIGNED_TYPE(Id, SingletonId)
378#define FLOATING_TYPE(Id, SingletonId)
379#define BUILTIN_TYPE(Id, SingletonId) \
380  case BuiltinType::Id:
381#include "clang/AST/BuiltinTypes.def"
382    // Misc other stuff which doesn't make sense here.
383    return false;
384
385  case BuiltinType::UInt:
386  case BuiltinType::Int:
387  case BuiltinType::Float:
388  case BuiltinType::Double:
389    LM.setKind(LengthModifier::None);
390    break;
391
392  case BuiltinType::Char_U:
393  case BuiltinType::UChar:
394  case BuiltinType::Char_S:
395  case BuiltinType::SChar:
396    LM.setKind(LengthModifier::AsChar);
397    break;
398
399  case BuiltinType::Short:
400  case BuiltinType::UShort:
401    LM.setKind(LengthModifier::AsShort);
402    break;
403
404  case BuiltinType::Long:
405  case BuiltinType::ULong:
406    LM.setKind(LengthModifier::AsLong);
407    break;
408
409  case BuiltinType::LongLong:
410  case BuiltinType::ULongLong:
411    LM.setKind(LengthModifier::AsLongLong);
412    break;
413
414  case BuiltinType::LongDouble:
415    LM.setKind(LengthModifier::AsLongDouble);
416    break;
417  }
418
419  // Handle size_t, ptrdiff_t, etc. that have dedicated length modifiers in C99.
420  if (isa<TypedefType>(QT) && (LangOpt.C99 || LangOpt.CPlusPlus0x)) {
421    const IdentifierInfo *Identifier = QT.getBaseTypeIdentifier();
422    if (Identifier->getName() == "size_t") {
423      LM.setKind(LengthModifier::AsSizeT);
424    } else if (Identifier->getName() == "ssize_t") {
425      // Not C99, but common in Unix.
426      LM.setKind(LengthModifier::AsSizeT);
427    } else if (Identifier->getName() == "intmax_t") {
428      LM.setKind(LengthModifier::AsIntMax);
429    } else if (Identifier->getName() == "uintmax_t") {
430      LM.setKind(LengthModifier::AsIntMax);
431    } else if (Identifier->getName() == "ptrdiff_t") {
432      LM.setKind(LengthModifier::AsPtrDiff);
433    }
434  }
435
436  // If fixing the length modifier was enough, we are done.
437  const analyze_printf::ArgTypeResult &ATR = getArgType(Ctx, IsObjCLiteral);
438  if (hasValidLengthModifier() && ATR.isValid() && ATR.matchesType(Ctx, QT))
439    return true;
440
441  // Set conversion specifier and disable any flags which do not apply to it.
442  // Let typedefs to char fall through to int, as %c is silly for uint8_t.
443  if (isa<TypedefType>(QT) && QT->isAnyCharacterType()) {
444    CS.setKind(ConversionSpecifier::cArg);
445    LM.setKind(LengthModifier::None);
446    Precision.setHowSpecified(OptionalAmount::NotSpecified);
447    HasAlternativeForm = 0;
448    HasLeadingZeroes = 0;
449    HasPlusPrefix = 0;
450  }
451  // Test for Floating type first as LongDouble can pass isUnsignedIntegerType
452  else if (QT->isRealFloatingType()) {
453    CS.setKind(ConversionSpecifier::fArg);
454  }
455  else if (QT->isSignedIntegerType()) {
456    CS.setKind(ConversionSpecifier::dArg);
457    HasAlternativeForm = 0;
458  }
459  else if (QT->isUnsignedIntegerType()) {
460    CS.setKind(ConversionSpecifier::uArg);
461    HasAlternativeForm = 0;
462    HasPlusPrefix = 0;
463  } else {
464    llvm_unreachable("Unexpected type");
465  }
466
467  return true;
468}
469
470void PrintfSpecifier::toString(raw_ostream &os) const {
471  // Whilst some features have no defined order, we are using the order
472  // appearing in the C99 standard (ISO/IEC 9899:1999 (E) 7.19.6.1)
473  os << "%";
474
475  // Positional args
476  if (usesPositionalArg()) {
477    os << getPositionalArgIndex() << "$";
478  }
479
480  // Conversion flags
481  if (IsLeftJustified)    os << "-";
482  if (HasPlusPrefix)      os << "+";
483  if (HasSpacePrefix)     os << " ";
484  if (HasAlternativeForm) os << "#";
485  if (HasLeadingZeroes)   os << "0";
486
487  // Minimum field width
488  FieldWidth.toString(os);
489  // Precision
490  Precision.toString(os);
491  // Length modifier
492  os << LM.toString();
493  // Conversion specifier
494  os << CS.toString();
495}
496
497bool PrintfSpecifier::hasValidPlusPrefix() const {
498  if (!HasPlusPrefix)
499    return true;
500
501  // The plus prefix only makes sense for signed conversions
502  switch (CS.getKind()) {
503  case ConversionSpecifier::dArg:
504  case ConversionSpecifier::iArg:
505  case ConversionSpecifier::fArg:
506  case ConversionSpecifier::FArg:
507  case ConversionSpecifier::eArg:
508  case ConversionSpecifier::EArg:
509  case ConversionSpecifier::gArg:
510  case ConversionSpecifier::GArg:
511  case ConversionSpecifier::aArg:
512  case ConversionSpecifier::AArg:
513    return true;
514
515  default:
516    return false;
517  }
518}
519
520bool PrintfSpecifier::hasValidAlternativeForm() const {
521  if (!HasAlternativeForm)
522    return true;
523
524  // Alternate form flag only valid with the oxXaAeEfFgG conversions
525  switch (CS.getKind()) {
526  case ConversionSpecifier::oArg:
527  case ConversionSpecifier::xArg:
528  case ConversionSpecifier::XArg:
529  case ConversionSpecifier::aArg:
530  case ConversionSpecifier::AArg:
531  case ConversionSpecifier::eArg:
532  case ConversionSpecifier::EArg:
533  case ConversionSpecifier::fArg:
534  case ConversionSpecifier::FArg:
535  case ConversionSpecifier::gArg:
536  case ConversionSpecifier::GArg:
537    return true;
538
539  default:
540    return false;
541  }
542}
543
544bool PrintfSpecifier::hasValidLeadingZeros() const {
545  if (!HasLeadingZeroes)
546    return true;
547
548  // Leading zeroes flag only valid with the diouxXaAeEfFgG conversions
549  switch (CS.getKind()) {
550  case ConversionSpecifier::dArg:
551  case ConversionSpecifier::iArg:
552  case ConversionSpecifier::oArg:
553  case ConversionSpecifier::uArg:
554  case ConversionSpecifier::xArg:
555  case ConversionSpecifier::XArg:
556  case ConversionSpecifier::aArg:
557  case ConversionSpecifier::AArg:
558  case ConversionSpecifier::eArg:
559  case ConversionSpecifier::EArg:
560  case ConversionSpecifier::fArg:
561  case ConversionSpecifier::FArg:
562  case ConversionSpecifier::gArg:
563  case ConversionSpecifier::GArg:
564    return true;
565
566  default:
567    return false;
568  }
569}
570
571bool PrintfSpecifier::hasValidSpacePrefix() const {
572  if (!HasSpacePrefix)
573    return true;
574
575  // The space prefix only makes sense for signed conversions
576  switch (CS.getKind()) {
577  case ConversionSpecifier::dArg:
578  case ConversionSpecifier::iArg:
579  case ConversionSpecifier::fArg:
580  case ConversionSpecifier::FArg:
581  case ConversionSpecifier::eArg:
582  case ConversionSpecifier::EArg:
583  case ConversionSpecifier::gArg:
584  case ConversionSpecifier::GArg:
585  case ConversionSpecifier::aArg:
586  case ConversionSpecifier::AArg:
587    return true;
588
589  default:
590    return false;
591  }
592}
593
594bool PrintfSpecifier::hasValidLeftJustified() const {
595  if (!IsLeftJustified)
596    return true;
597
598  // The left justified flag is valid for all conversions except n
599  switch (CS.getKind()) {
600  case ConversionSpecifier::nArg:
601    return false;
602
603  default:
604    return true;
605  }
606}
607
608bool PrintfSpecifier::hasValidThousandsGroupingPrefix() const {
609  if (!HasThousandsGrouping)
610    return true;
611
612  switch (CS.getKind()) {
613    case ConversionSpecifier::dArg:
614    case ConversionSpecifier::iArg:
615    case ConversionSpecifier::uArg:
616    case ConversionSpecifier::fArg:
617    case ConversionSpecifier::FArg:
618    case ConversionSpecifier::gArg:
619    case ConversionSpecifier::GArg:
620      return true;
621    default:
622      return false;
623  }
624}
625
626bool PrintfSpecifier::hasValidPrecision() const {
627  if (Precision.getHowSpecified() == OptionalAmount::NotSpecified)
628    return true;
629
630  // Precision is only valid with the diouxXaAeEfFgGs conversions
631  switch (CS.getKind()) {
632  case ConversionSpecifier::dArg:
633  case ConversionSpecifier::iArg:
634  case ConversionSpecifier::oArg:
635  case ConversionSpecifier::uArg:
636  case ConversionSpecifier::xArg:
637  case ConversionSpecifier::XArg:
638  case ConversionSpecifier::aArg:
639  case ConversionSpecifier::AArg:
640  case ConversionSpecifier::eArg:
641  case ConversionSpecifier::EArg:
642  case ConversionSpecifier::fArg:
643  case ConversionSpecifier::FArg:
644  case ConversionSpecifier::gArg:
645  case ConversionSpecifier::GArg:
646  case ConversionSpecifier::sArg:
647    return true;
648
649  default:
650    return false;
651  }
652}
653bool PrintfSpecifier::hasValidFieldWidth() const {
654  if (FieldWidth.getHowSpecified() == OptionalAmount::NotSpecified)
655      return true;
656
657  // The field width is valid for all conversions except n
658  switch (CS.getKind()) {
659  case ConversionSpecifier::nArg:
660    return false;
661
662  default:
663    return true;
664  }
665}
666