PrintfFormatString.cpp revision 4684778993c667246039b4664acbce59dc99440c 
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:
270      case LengthModifier::AsQuad:
271        return Ctx.LongLongTy;
272      case LengthModifier::AsIntMax:
273        return ArgTypeResult(Ctx.getIntMaxType(), "intmax_t");
274      case LengthModifier::AsSizeT:
275        // FIXME: How to get the corresponding signed version of size_t?
276        return ArgTypeResult();
277      case LengthModifier::AsPtrDiff:
278        return ArgTypeResult(Ctx.getPointerDiffType(), "ptrdiff_t");
279      case LengthModifier::AsAllocate:
280      case LengthModifier::AsMAllocate:
281        return ArgTypeResult::Invalid();
282    }
283
284  if (CS.isUIntArg())
285    switch (LM.getKind()) {
286      case LengthModifier::AsLongDouble:
287        // GNU extension.
288        return Ctx.UnsignedLongLongTy;
289      case LengthModifier::None: return Ctx.UnsignedIntTy;
290      case LengthModifier::AsChar: return Ctx.UnsignedCharTy;
291      case LengthModifier::AsShort: return Ctx.UnsignedShortTy;
292      case LengthModifier::AsLong: return Ctx.UnsignedLongTy;
293      case LengthModifier::AsLongLong:
294      case LengthModifier::AsQuad:
295        return Ctx.UnsignedLongLongTy;
296      case LengthModifier::AsIntMax:
297        return ArgTypeResult(Ctx.getUIntMaxType(), "uintmax_t");
298      case LengthModifier::AsSizeT:
299        return ArgTypeResult(Ctx.getSizeType(), "size_t");
300      case LengthModifier::AsPtrDiff:
301        // FIXME: How to get the corresponding unsigned
302        // version of ptrdiff_t?
303        return ArgTypeResult();
304      case LengthModifier::AsAllocate:
305      case LengthModifier::AsMAllocate:
306        return ArgTypeResult::Invalid();
307    }
308
309  if (CS.isDoubleArg()) {
310    if (LM.getKind() == LengthModifier::AsLongDouble)
311      return Ctx.LongDoubleTy;
312    return Ctx.DoubleTy;
313  }
314
315  switch (CS.getKind()) {
316    case ConversionSpecifier::sArg:
317      if (LM.getKind() == LengthModifier::AsWideChar) {
318        if (IsObjCLiteral)
319          return Ctx.getPointerType(Ctx.UnsignedShortTy.withConst());
320        return ArgTypeResult(ArgTypeResult::WCStrTy, "wchar_t *");
321      }
322      return ArgTypeResult::CStrTy;
323    case ConversionSpecifier::SArg:
324      if (IsObjCLiteral)
325        return Ctx.getPointerType(Ctx.UnsignedShortTy.withConst());
326      return ArgTypeResult(ArgTypeResult::WCStrTy, "wchar_t *");
327    case ConversionSpecifier::CArg:
328      if (IsObjCLiteral)
329        return Ctx.UnsignedShortTy;
330      return ArgTypeResult(Ctx.WCharTy, "wchar_t");
331    case ConversionSpecifier::pArg:
332      return ArgTypeResult::CPointerTy;
333    case ConversionSpecifier::ObjCObjArg:
334      return ArgTypeResult::ObjCPointerTy;
335    default:
336      break;
337  }
338
339  // FIXME: Handle other cases.
340  return ArgTypeResult();
341}
342
343bool PrintfSpecifier::fixType(QualType QT, const LangOptions &LangOpt,
344                              ASTContext &Ctx, bool IsObjCLiteral) {
345  // Handle Objective-C objects first. Note that while the '%@' specifier will
346  // not warn for structure pointer or void pointer arguments (because that's
347  // how CoreFoundation objects are implemented), we only show a fixit for '%@'
348  // if we know it's an object (block, id, class, or __attribute__((NSObject))).
349  if (QT->isObjCRetainableType()) {
350    if (!IsObjCLiteral)
351      return false;
352
353    CS.setKind(ConversionSpecifier::ObjCObjArg);
354
355    // Disable irrelevant flags
356    HasThousandsGrouping = false;
357    HasPlusPrefix = false;
358    HasSpacePrefix = false;
359    HasAlternativeForm = false;
360    HasLeadingZeroes = false;
361    Precision.setHowSpecified(OptionalAmount::NotSpecified);
362    LM.setKind(LengthModifier::None);
363
364    return true;
365  }
366
367  // Handle strings next (char *, wchar_t *)
368  if (QT->isPointerType() && (QT->getPointeeType()->isAnyCharacterType())) {
369    CS.setKind(ConversionSpecifier::sArg);
370
371    // Disable irrelevant flags
372    HasAlternativeForm = 0;
373    HasLeadingZeroes = 0;
374
375    // Set the long length modifier for wide characters
376    if (QT->getPointeeType()->isWideCharType())
377      LM.setKind(LengthModifier::AsWideChar);
378    else
379      LM.setKind(LengthModifier::None);
380
381    return true;
382  }
383
384  // If it's an enum, get its underlying type.
385  if (const EnumType *ETy = QT->getAs<EnumType>())
386    QT = ETy->getDecl()->getIntegerType();
387
388  // We can only work with builtin types.
389  const BuiltinType *BT = QT->getAs<BuiltinType>();
390  if (!BT)
391    return false;
392
393  // Set length modifier
394  switch (BT->getKind()) {
395  case BuiltinType::Bool:
396  case BuiltinType::WChar_U:
397  case BuiltinType::WChar_S:
398  case BuiltinType::Char16:
399  case BuiltinType::Char32:
400  case BuiltinType::UInt128:
401  case BuiltinType::Int128:
402  case BuiltinType::Half:
403    // Various types which are non-trivial to correct.
404    return false;
405
406#define SIGNED_TYPE(Id, SingletonId)
407#define UNSIGNED_TYPE(Id, SingletonId)
408#define FLOATING_TYPE(Id, SingletonId)
409#define BUILTIN_TYPE(Id, SingletonId) \
410  case BuiltinType::Id:
411#include "clang/AST/BuiltinTypes.def"
412    // Misc other stuff which doesn't make sense here.
413    return false;
414
415  case BuiltinType::UInt:
416  case BuiltinType::Int:
417  case BuiltinType::Float:
418  case BuiltinType::Double:
419    LM.setKind(LengthModifier::None);
420    break;
421
422  case BuiltinType::Char_U:
423  case BuiltinType::UChar:
424  case BuiltinType::Char_S:
425  case BuiltinType::SChar:
426    LM.setKind(LengthModifier::AsChar);
427    break;
428
429  case BuiltinType::Short:
430  case BuiltinType::UShort:
431    LM.setKind(LengthModifier::AsShort);
432    break;
433
434  case BuiltinType::Long:
435  case BuiltinType::ULong:
436    LM.setKind(LengthModifier::AsLong);
437    break;
438
439  case BuiltinType::LongLong:
440  case BuiltinType::ULongLong:
441    LM.setKind(LengthModifier::AsLongLong);
442    break;
443
444  case BuiltinType::LongDouble:
445    LM.setKind(LengthModifier::AsLongDouble);
446    break;
447  }
448
449  // Handle size_t, ptrdiff_t, etc. that have dedicated length modifiers in C99.
450  if (isa<TypedefType>(QT) && (LangOpt.C99 || LangOpt.CPlusPlus0x))
451    namedTypeToLengthModifier(QT, LM);
452
453  // If fixing the length modifier was enough, we are done.
454  const analyze_printf::ArgTypeResult &ATR = getArgType(Ctx, IsObjCLiteral);
455  if (hasValidLengthModifier() && ATR.isValid() && ATR.matchesType(Ctx, QT))
456    return true;
457
458  // Set conversion specifier and disable any flags which do not apply to it.
459  // Let typedefs to char fall through to int, as %c is silly for uint8_t.
460  if (isa<TypedefType>(QT) && QT->isAnyCharacterType()) {
461    CS.setKind(ConversionSpecifier::cArg);
462    LM.setKind(LengthModifier::None);
463    Precision.setHowSpecified(OptionalAmount::NotSpecified);
464    HasAlternativeForm = 0;
465    HasLeadingZeroes = 0;
466    HasPlusPrefix = 0;
467  }
468  // Test for Floating type first as LongDouble can pass isUnsignedIntegerType
469  else if (QT->isRealFloatingType()) {
470    CS.setKind(ConversionSpecifier::fArg);
471  }
472  else if (QT->isSignedIntegerType()) {
473    CS.setKind(ConversionSpecifier::dArg);
474    HasAlternativeForm = 0;
475  }
476  else if (QT->isUnsignedIntegerType()) {
477    CS.setKind(ConversionSpecifier::uArg);
478    HasAlternativeForm = 0;
479    HasPlusPrefix = 0;
480  } else {
481    llvm_unreachable("Unexpected type");
482  }
483
484  return true;
485}
486
487void PrintfSpecifier::toString(raw_ostream &os) const {
488  // Whilst some features have no defined order, we are using the order
489  // appearing in the C99 standard (ISO/IEC 9899:1999 (E) 7.19.6.1)
490  os << "%";
491
492  // Positional args
493  if (usesPositionalArg()) {
494    os << getPositionalArgIndex() << "$";
495  }
496
497  // Conversion flags
498  if (IsLeftJustified)    os << "-";
499  if (HasPlusPrefix)      os << "+";
500  if (HasSpacePrefix)     os << " ";
501  if (HasAlternativeForm) os << "#";
502  if (HasLeadingZeroes)   os << "0";
503
504  // Minimum field width
505  FieldWidth.toString(os);
506  // Precision
507  Precision.toString(os);
508  // Length modifier
509  os << LM.toString();
510  // Conversion specifier
511  os << CS.toString();
512}
513
514bool PrintfSpecifier::hasValidPlusPrefix() const {
515  if (!HasPlusPrefix)
516    return true;
517
518  // The plus prefix only makes sense for signed conversions
519  switch (CS.getKind()) {
520  case ConversionSpecifier::dArg:
521  case ConversionSpecifier::iArg:
522  case ConversionSpecifier::fArg:
523  case ConversionSpecifier::FArg:
524  case ConversionSpecifier::eArg:
525  case ConversionSpecifier::EArg:
526  case ConversionSpecifier::gArg:
527  case ConversionSpecifier::GArg:
528  case ConversionSpecifier::aArg:
529  case ConversionSpecifier::AArg:
530    return true;
531
532  default:
533    return false;
534  }
535}
536
537bool PrintfSpecifier::hasValidAlternativeForm() const {
538  if (!HasAlternativeForm)
539    return true;
540
541  // Alternate form flag only valid with the oxXaAeEfFgG conversions
542  switch (CS.getKind()) {
543  case ConversionSpecifier::oArg:
544  case ConversionSpecifier::xArg:
545  case ConversionSpecifier::XArg:
546  case ConversionSpecifier::aArg:
547  case ConversionSpecifier::AArg:
548  case ConversionSpecifier::eArg:
549  case ConversionSpecifier::EArg:
550  case ConversionSpecifier::fArg:
551  case ConversionSpecifier::FArg:
552  case ConversionSpecifier::gArg:
553  case ConversionSpecifier::GArg:
554    return true;
555
556  default:
557    return false;
558  }
559}
560
561bool PrintfSpecifier::hasValidLeadingZeros() const {
562  if (!HasLeadingZeroes)
563    return true;
564
565  // Leading zeroes flag only valid with the diouxXaAeEfFgG conversions
566  switch (CS.getKind()) {
567  case ConversionSpecifier::dArg:
568  case ConversionSpecifier::iArg:
569  case ConversionSpecifier::oArg:
570  case ConversionSpecifier::uArg:
571  case ConversionSpecifier::xArg:
572  case ConversionSpecifier::XArg:
573  case ConversionSpecifier::aArg:
574  case ConversionSpecifier::AArg:
575  case ConversionSpecifier::eArg:
576  case ConversionSpecifier::EArg:
577  case ConversionSpecifier::fArg:
578  case ConversionSpecifier::FArg:
579  case ConversionSpecifier::gArg:
580  case ConversionSpecifier::GArg:
581    return true;
582
583  default:
584    return false;
585  }
586}
587
588bool PrintfSpecifier::hasValidSpacePrefix() const {
589  if (!HasSpacePrefix)
590    return true;
591
592  // The space prefix only makes sense for signed conversions
593  switch (CS.getKind()) {
594  case ConversionSpecifier::dArg:
595  case ConversionSpecifier::iArg:
596  case ConversionSpecifier::fArg:
597  case ConversionSpecifier::FArg:
598  case ConversionSpecifier::eArg:
599  case ConversionSpecifier::EArg:
600  case ConversionSpecifier::gArg:
601  case ConversionSpecifier::GArg:
602  case ConversionSpecifier::aArg:
603  case ConversionSpecifier::AArg:
604    return true;
605
606  default:
607    return false;
608  }
609}
610
611bool PrintfSpecifier::hasValidLeftJustified() const {
612  if (!IsLeftJustified)
613    return true;
614
615  // The left justified flag is valid for all conversions except n
616  switch (CS.getKind()) {
617  case ConversionSpecifier::nArg:
618    return false;
619
620  default:
621    return true;
622  }
623}
624
625bool PrintfSpecifier::hasValidThousandsGroupingPrefix() const {
626  if (!HasThousandsGrouping)
627    return true;
628
629  switch (CS.getKind()) {
630    case ConversionSpecifier::dArg:
631    case ConversionSpecifier::iArg:
632    case ConversionSpecifier::uArg:
633    case ConversionSpecifier::fArg:
634    case ConversionSpecifier::FArg:
635    case ConversionSpecifier::gArg:
636    case ConversionSpecifier::GArg:
637      return true;
638    default:
639      return false;
640  }
641}
642
643bool PrintfSpecifier::hasValidPrecision() const {
644  if (Precision.getHowSpecified() == OptionalAmount::NotSpecified)
645    return true;
646
647  // Precision is only valid with the diouxXaAeEfFgGs conversions
648  switch (CS.getKind()) {
649  case ConversionSpecifier::dArg:
650  case ConversionSpecifier::iArg:
651  case ConversionSpecifier::oArg:
652  case ConversionSpecifier::uArg:
653  case ConversionSpecifier::xArg:
654  case ConversionSpecifier::XArg:
655  case ConversionSpecifier::aArg:
656  case ConversionSpecifier::AArg:
657  case ConversionSpecifier::eArg:
658  case ConversionSpecifier::EArg:
659  case ConversionSpecifier::fArg:
660  case ConversionSpecifier::FArg:
661  case ConversionSpecifier::gArg:
662  case ConversionSpecifier::GArg:
663  case ConversionSpecifier::sArg:
664    return true;
665
666  default:
667    return false;
668  }
669}
670bool PrintfSpecifier::hasValidFieldWidth() const {
671  if (FieldWidth.getHowSpecified() == OptionalAmount::NotSpecified)
672      return true;
673
674  // The field width is valid for all conversions except n
675  switch (CS.getKind()) {
676  case ConversionSpecifier::nArg:
677    return false;
678
679  default:
680    return true;
681  }
682}
683