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/AST/FormatString.h"
16 #include "clang/AST/OSLog.h"
17 #include "FormatStringParsing.h"
18 #include "clang/Basic/TargetInfo.h"
19 
20 using clang::analyze_format_string::ArgType;
21 using clang::analyze_format_string::FormatStringHandler;
22 using clang::analyze_format_string::LengthModifier;
23 using clang::analyze_format_string::OptionalAmount;
24 using clang::analyze_format_string::ConversionSpecifier;
25 using clang::analyze_printf::PrintfSpecifier;
26 
27 using namespace clang;
28 
29 typedef clang::analyze_format_string::SpecifierResult<PrintfSpecifier>
30         PrintfSpecifierResult;
31 
32 //===----------------------------------------------------------------------===//
33 // Methods for parsing format strings.
34 //===----------------------------------------------------------------------===//
35 
36 using analyze_format_string::ParseNonPositionAmount;
37 
38 static bool ParsePrecision(FormatStringHandler &H, PrintfSpecifier &FS,
39                            const char *Start, const char *&Beg, const char *E,
40                            unsigned *argIndex) {
41   if (argIndex) {
42     FS.setPrecision(ParseNonPositionAmount(Beg, E, *argIndex));
43   } else {
44     const OptionalAmount Amt = ParsePositionAmount(H, Start, Beg, E,
45                                            analyze_format_string::PrecisionPos);
46     if (Amt.isInvalid())
47       return true;
48     FS.setPrecision(Amt);
49   }
50   return false;
51 }
52 
53 static bool ParseObjCFlags(FormatStringHandler &H, PrintfSpecifier &FS,
54                            const char *FlagBeg, const char *E, bool Warn) {
55    StringRef Flag(FlagBeg, E - FlagBeg);
56    // Currently there is only one flag.
57    if (Flag == "tt") {
58      FS.setHasObjCTechnicalTerm(FlagBeg);
59      return false;
60    }
61    // Handle either the case of no flag or an invalid flag.
62    if (Warn) {
63      if (Flag == "")
64        H.HandleEmptyObjCModifierFlag(FlagBeg, E  - FlagBeg);
65      else
66        H.HandleInvalidObjCModifierFlag(FlagBeg, E  - FlagBeg);
67    }
68    return true;
69 }
70 
71 static PrintfSpecifierResult ParsePrintfSpecifier(FormatStringHandler &H,
72                                                   const char *&Beg,
73                                                   const char *E,
74                                                   unsigned &argIndex,
75                                                   const LangOptions &LO,
76                                                   const TargetInfo &Target,
77                                                   bool Warn,
78                                                   bool isFreeBSDKPrintf) {
79 
80   using namespace clang::analyze_format_string;
81   using namespace clang::analyze_printf;
82 
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(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     if (Warn)
108       H.HandleIncompleteSpecifier(Start, E - Start);
109     return true;
110   }
111 
112   PrintfSpecifier FS;
113   if (ParseArgPosition(H, FS, Start, I, E))
114     return true;
115 
116   if (I == E) {
117     // No more characters left?
118     if (Warn)
119       H.HandleIncompleteSpecifier(Start, E - Start);
120     return true;
121   }
122 
123   if (*I == '{') {
124     ++I;
125     unsigned char PrivacyFlags = 0;
126     StringRef MatchedStr;
127 
128     do {
129       StringRef Str(I, E - I);
130       std::string Match = "^[[:space:]]*"
131                           "(private|public|sensitive|mask\\.[^[:space:],}]*)"
132                           "[[:space:]]*(,|})";
133       llvm::Regex R(Match);
134       SmallVector<StringRef, 2> Matches;
135 
136       if (R.match(Str, &Matches)) {
137         MatchedStr = Matches[1];
138         I += Matches[0].size();
139 
140         // Set the privacy flag if the privacy annotation in the
141         // comma-delimited segment is at least as strict as the privacy
142         // annotations in previous comma-delimited segments.
143         if (MatchedStr.startswith("mask")) {
144           StringRef MaskType = MatchedStr.substr(sizeof("mask.") - 1);
145           unsigned Size = MaskType.size();
146           if (Warn && (Size == 0 || Size > 8))
147             H.handleInvalidMaskType(MaskType);
148           FS.setMaskType(MaskType);
149         } else if (MatchedStr.equals("sensitive"))
150           PrivacyFlags = clang::analyze_os_log::OSLogBufferItem::IsSensitive;
151         else if (PrivacyFlags !=
152                  clang::analyze_os_log::OSLogBufferItem::IsSensitive &&
153                  MatchedStr.equals("private"))
154           PrivacyFlags = clang::analyze_os_log::OSLogBufferItem::IsPrivate;
155         else if (PrivacyFlags == 0 && MatchedStr.equals("public"))
156           PrivacyFlags = clang::analyze_os_log::OSLogBufferItem::IsPublic;
157       } else {
158         size_t CommaOrBracePos =
159             Str.find_if([](char c) { return c == ',' || c == '}'; });
160 
161         if (CommaOrBracePos == StringRef::npos) {
162           // Neither a comma nor the closing brace was found.
163           if (Warn)
164             H.HandleIncompleteSpecifier(Start, E - Start);
165           return true;
166         }
167 
168         I += CommaOrBracePos + 1;
169       }
170       // Continue until the closing brace is found.
171     } while (*(I - 1) == ',');
172 
173     // Set the privacy flag.
174     switch (PrivacyFlags) {
175     case 0:
176       break;
177     case clang::analyze_os_log::OSLogBufferItem::IsPrivate:
178       FS.setIsPrivate(MatchedStr.data());
179       break;
180     case clang::analyze_os_log::OSLogBufferItem::IsPublic:
181       FS.setIsPublic(MatchedStr.data());
182       break;
183     case clang::analyze_os_log::OSLogBufferItem::IsSensitive:
184       FS.setIsSensitive(MatchedStr.data());
185       break;
186     default:
187       llvm_unreachable("Unexpected privacy flag value");
188     }
189   }
190 
191   // Look for flags (if any).
192   bool hasMore = true;
193   for ( ; I != E; ++I) {
194     switch (*I) {
195       default: hasMore = false; break;
196       case '\'':
197         // FIXME: POSIX specific.  Always accept?
198         FS.setHasThousandsGrouping(I);
199         break;
200       case '-': FS.setIsLeftJustified(I); break;
201       case '+': FS.setHasPlusPrefix(I); break;
202       case ' ': FS.setHasSpacePrefix(I); break;
203       case '#': FS.setHasAlternativeForm(I); break;
204       case '0': FS.setHasLeadingZeros(I); break;
205     }
206     if (!hasMore)
207       break;
208   }
209 
210   if (I == E) {
211     // No more characters left?
212     if (Warn)
213       H.HandleIncompleteSpecifier(Start, E - Start);
214     return true;
215   }
216 
217   // Look for the field width (if any).
218   if (ParseFieldWidth(H, FS, Start, I, E,
219                       FS.usesPositionalArg() ? nullptr : &argIndex))
220     return true;
221 
222   if (I == E) {
223     // No more characters left?
224     if (Warn)
225       H.HandleIncompleteSpecifier(Start, E - Start);
226     return true;
227   }
228 
229   // Look for the precision (if any).
230   if (*I == '.') {
231     ++I;
232     if (I == E) {
233       if (Warn)
234         H.HandleIncompleteSpecifier(Start, E - Start);
235       return true;
236     }
237 
238     if (ParsePrecision(H, FS, Start, I, E,
239                        FS.usesPositionalArg() ? nullptr : &argIndex))
240       return true;
241 
242     if (I == E) {
243       // No more characters left?
244       if (Warn)
245         H.HandleIncompleteSpecifier(Start, E - Start);
246       return true;
247     }
248   }
249 
250   // Look for the length modifier.
251   if (ParseLengthModifier(FS, I, E, LO) && I == E) {
252     // No more characters left?
253     if (Warn)
254       H.HandleIncompleteSpecifier(Start, E - Start);
255     return true;
256   }
257 
258   // Look for the Objective-C modifier flags, if any.
259   // We parse these here, even if they don't apply to
260   // the conversion specifier, and then emit an error
261   // later if the conversion specifier isn't '@'.  This
262   // enables better recovery, and we don't know if
263   // these flags are applicable until later.
264   const char *ObjCModifierFlagsStart = nullptr,
265              *ObjCModifierFlagsEnd = nullptr;
266   if (*I == '[') {
267     ObjCModifierFlagsStart = I;
268     ++I;
269     auto flagStart = I;
270     for (;; ++I) {
271       ObjCModifierFlagsEnd = I;
272       if (I == E) {
273         if (Warn)
274           H.HandleIncompleteSpecifier(Start, E - Start);
275         return true;
276       }
277       // Did we find the closing ']'?
278       if (*I == ']') {
279         if (ParseObjCFlags(H, FS, flagStart, I, Warn))
280           return true;
281         ++I;
282         break;
283       }
284       // There are no separators defined yet for multiple
285       // Objective-C modifier flags.  When those are
286       // defined, this is the place to check.
287     }
288   }
289 
290   if (*I == '\0') {
291     // Detect spurious null characters, which are likely errors.
292     H.HandleNullChar(I);
293     return true;
294   }
295 
296   // Finally, look for the conversion specifier.
297   const char *conversionPosition = I++;
298   ConversionSpecifier::Kind k = ConversionSpecifier::InvalidSpecifier;
299   switch (*conversionPosition) {
300     default:
301       break;
302     // C99: 7.19.6.1 (section 8).
303     case '%': k = ConversionSpecifier::PercentArg;   break;
304     case 'A': k = ConversionSpecifier::AArg; break;
305     case 'E': k = ConversionSpecifier::EArg; break;
306     case 'F': k = ConversionSpecifier::FArg; break;
307     case 'G': k = ConversionSpecifier::GArg; break;
308     case 'X': k = ConversionSpecifier::XArg; break;
309     case 'a': k = ConversionSpecifier::aArg; break;
310     case 'c': k = ConversionSpecifier::cArg; break;
311     case 'd': k = ConversionSpecifier::dArg; break;
312     case 'e': k = ConversionSpecifier::eArg; break;
313     case 'f': k = ConversionSpecifier::fArg; break;
314     case 'g': k = ConversionSpecifier::gArg; break;
315     case 'i': k = ConversionSpecifier::iArg; break;
316     case 'n': k = ConversionSpecifier::nArg; break;
317     case 'o': k = ConversionSpecifier::oArg; break;
318     case 'p': k = ConversionSpecifier::pArg; break;
319     case 's': k = ConversionSpecifier::sArg; break;
320     case 'u': k = ConversionSpecifier::uArg; break;
321     case 'x': k = ConversionSpecifier::xArg; break;
322     // POSIX specific.
323     case 'C': k = ConversionSpecifier::CArg; break;
324     case 'S': k = ConversionSpecifier::SArg; break;
325     // Apple extension for os_log
326     case 'P':
327       k = ConversionSpecifier::PArg;
328       break;
329     // Objective-C.
330     case '@': k = ConversionSpecifier::ObjCObjArg; break;
331     // Glibc specific.
332     case 'm': k = ConversionSpecifier::PrintErrno; break;
333     // FreeBSD kernel specific.
334     case 'b':
335       if (isFreeBSDKPrintf)
336         k = ConversionSpecifier::FreeBSDbArg; // int followed by char *
337       break;
338     case 'r':
339       if (isFreeBSDKPrintf)
340         k = ConversionSpecifier::FreeBSDrArg; // int
341       break;
342     case 'y':
343       if (isFreeBSDKPrintf)
344         k = ConversionSpecifier::FreeBSDyArg; // int
345       break;
346     // Apple-specific.
347     case 'D':
348       if (isFreeBSDKPrintf)
349         k = ConversionSpecifier::FreeBSDDArg; // void * followed by char *
350       else if (Target.getTriple().isOSDarwin())
351         k = ConversionSpecifier::DArg;
352       break;
353     case 'O':
354       if (Target.getTriple().isOSDarwin())
355         k = ConversionSpecifier::OArg;
356       break;
357     case 'U':
358       if (Target.getTriple().isOSDarwin())
359         k = ConversionSpecifier::UArg;
360       break;
361     // MS specific.
362     case 'Z':
363       if (Target.getTriple().isOSMSVCRT())
364         k = ConversionSpecifier::ZArg;
365   }
366 
367   // Check to see if we used the Objective-C modifier flags with
368   // a conversion specifier other than '@'.
369   if (k != ConversionSpecifier::ObjCObjArg &&
370       k != ConversionSpecifier::InvalidSpecifier &&
371       ObjCModifierFlagsStart) {
372     H.HandleObjCFlagsWithNonObjCConversion(ObjCModifierFlagsStart,
373                                            ObjCModifierFlagsEnd + 1,
374                                            conversionPosition);
375     return true;
376   }
377 
378   PrintfConversionSpecifier CS(conversionPosition, k);
379   FS.setConversionSpecifier(CS);
380   if (CS.consumesDataArgument() && !FS.usesPositionalArg())
381     FS.setArgIndex(argIndex++);
382   // FreeBSD kernel specific.
383   if (k == ConversionSpecifier::FreeBSDbArg ||
384       k == ConversionSpecifier::FreeBSDDArg)
385     argIndex++;
386 
387   if (k == ConversionSpecifier::InvalidSpecifier) {
388     unsigned Len = I - Start;
389     if (ParseUTF8InvalidSpecifier(Start, E, Len)) {
390       CS.setEndScanList(Start + Len);
391       FS.setConversionSpecifier(CS);
392     }
393     // Assume the conversion takes one argument.
394     return !H.HandleInvalidPrintfConversionSpecifier(FS, Start, Len);
395   }
396   return PrintfSpecifierResult(Start, FS);
397 }
398 
399 bool clang::analyze_format_string::ParsePrintfString(FormatStringHandler &H,
400                                                      const char *I,
401                                                      const char *E,
402                                                      const LangOptions &LO,
403                                                      const TargetInfo &Target,
404                                                      bool isFreeBSDKPrintf) {
405 
406   unsigned argIndex = 0;
407 
408   // Keep looking for a format specifier until we have exhausted the string.
409   while (I != E) {
410     const PrintfSpecifierResult &FSR = ParsePrintfSpecifier(H, I, E, argIndex,
411                                                             LO, Target, true,
412                                                             isFreeBSDKPrintf);
413     // Did a fail-stop error of any kind occur when parsing the specifier?
414     // If so, don't do any more processing.
415     if (FSR.shouldStop())
416       return true;
417     // Did we exhaust the string or encounter an error that
418     // we can recover from?
419     if (!FSR.hasValue())
420       continue;
421     // We have a format specifier.  Pass it to the callback.
422     if (!H.HandlePrintfSpecifier(FSR.getValue(), FSR.getStart(),
423                                  I - FSR.getStart()))
424       return true;
425   }
426   assert(I == E && "Format string not exhausted");
427   return false;
428 }
429 
430 bool clang::analyze_format_string::ParseFormatStringHasSArg(const char *I,
431                                                             const char *E,
432                                                             const LangOptions &LO,
433                                                             const TargetInfo &Target) {
434 
435   unsigned argIndex = 0;
436 
437   // Keep looking for a %s format specifier until we have exhausted the string.
438   FormatStringHandler H;
439   while (I != E) {
440     const PrintfSpecifierResult &FSR = ParsePrintfSpecifier(H, I, E, argIndex,
441                                                             LO, Target, false,
442                                                             false);
443     // Did a fail-stop error of any kind occur when parsing the specifier?
444     // If so, don't do any more processing.
445     if (FSR.shouldStop())
446       return false;
447     // Did we exhaust the string or encounter an error that
448     // we can recover from?
449     if (!FSR.hasValue())
450       continue;
451     const analyze_printf::PrintfSpecifier &FS = FSR.getValue();
452     // Return true if this a %s format specifier.
453     if (FS.getConversionSpecifier().getKind() == ConversionSpecifier::Kind::sArg)
454       return true;
455   }
456   return false;
457 }
458 
459 //===----------------------------------------------------------------------===//
460 // Methods on PrintfSpecifier.
461 //===----------------------------------------------------------------------===//
462 
463 ArgType PrintfSpecifier::getArgType(ASTContext &Ctx,
464                                     bool IsObjCLiteral) const {
465   const PrintfConversionSpecifier &CS = getConversionSpecifier();
466 
467   if (!CS.consumesDataArgument())
468     return ArgType::Invalid();
469 
470   if (CS.getKind() == ConversionSpecifier::cArg)
471     switch (LM.getKind()) {
472       case LengthModifier::None:
473         return Ctx.IntTy;
474       case LengthModifier::AsLong:
475       case LengthModifier::AsWide:
476         return ArgType(ArgType::WIntTy, "wint_t");
477       case LengthModifier::AsShort:
478         if (Ctx.getTargetInfo().getTriple().isOSMSVCRT())
479           return Ctx.IntTy;
480         LLVM_FALLTHROUGH;
481       default:
482         return ArgType::Invalid();
483     }
484 
485   if (CS.isIntArg())
486     switch (LM.getKind()) {
487       case LengthModifier::AsLongDouble:
488         // GNU extension.
489         return Ctx.LongLongTy;
490       case LengthModifier::None:
491         return Ctx.IntTy;
492       case LengthModifier::AsInt32:
493         return ArgType(Ctx.IntTy, "__int32");
494       case LengthModifier::AsChar: return ArgType::AnyCharTy;
495       case LengthModifier::AsShort: return Ctx.ShortTy;
496       case LengthModifier::AsLong: return Ctx.LongTy;
497       case LengthModifier::AsLongLong:
498       case LengthModifier::AsQuad:
499         return Ctx.LongLongTy;
500       case LengthModifier::AsInt64:
501         return ArgType(Ctx.LongLongTy, "__int64");
502       case LengthModifier::AsIntMax:
503         return ArgType(Ctx.getIntMaxType(), "intmax_t");
504       case LengthModifier::AsSizeT:
505         return ArgType::makeSizeT(ArgType(Ctx.getSignedSizeType(), "ssize_t"));
506       case LengthModifier::AsInt3264:
507         return Ctx.getTargetInfo().getTriple().isArch64Bit()
508                    ? ArgType(Ctx.LongLongTy, "__int64")
509                    : ArgType(Ctx.IntTy, "__int32");
510       case LengthModifier::AsPtrDiff:
511         return ArgType::makePtrdiffT(
512             ArgType(Ctx.getPointerDiffType(), "ptrdiff_t"));
513       case LengthModifier::AsAllocate:
514       case LengthModifier::AsMAllocate:
515       case LengthModifier::AsWide:
516         return ArgType::Invalid();
517     }
518 
519   if (CS.isUIntArg())
520     switch (LM.getKind()) {
521       case LengthModifier::AsLongDouble:
522         // GNU extension.
523         return Ctx.UnsignedLongLongTy;
524       case LengthModifier::None:
525         return Ctx.UnsignedIntTy;
526       case LengthModifier::AsInt32:
527         return ArgType(Ctx.UnsignedIntTy, "unsigned __int32");
528       case LengthModifier::AsChar: return Ctx.UnsignedCharTy;
529       case LengthModifier::AsShort: return Ctx.UnsignedShortTy;
530       case LengthModifier::AsLong: return Ctx.UnsignedLongTy;
531       case LengthModifier::AsLongLong:
532       case LengthModifier::AsQuad:
533         return Ctx.UnsignedLongLongTy;
534       case LengthModifier::AsInt64:
535         return ArgType(Ctx.UnsignedLongLongTy, "unsigned __int64");
536       case LengthModifier::AsIntMax:
537         return ArgType(Ctx.getUIntMaxType(), "uintmax_t");
538       case LengthModifier::AsSizeT:
539         return ArgType::makeSizeT(ArgType(Ctx.getSizeType(), "size_t"));
540       case LengthModifier::AsInt3264:
541         return Ctx.getTargetInfo().getTriple().isArch64Bit()
542                    ? ArgType(Ctx.UnsignedLongLongTy, "unsigned __int64")
543                    : ArgType(Ctx.UnsignedIntTy, "unsigned __int32");
544       case LengthModifier::AsPtrDiff:
545         return ArgType::makePtrdiffT(
546             ArgType(Ctx.getUnsignedPointerDiffType(), "unsigned ptrdiff_t"));
547       case LengthModifier::AsAllocate:
548       case LengthModifier::AsMAllocate:
549       case LengthModifier::AsWide:
550         return ArgType::Invalid();
551     }
552 
553   if (CS.isDoubleArg()) {
554     if (LM.getKind() == LengthModifier::AsLongDouble)
555       return Ctx.LongDoubleTy;
556     return Ctx.DoubleTy;
557   }
558 
559   if (CS.getKind() == ConversionSpecifier::nArg) {
560     switch (LM.getKind()) {
561       case LengthModifier::None:
562         return ArgType::PtrTo(Ctx.IntTy);
563       case LengthModifier::AsChar:
564         return ArgType::PtrTo(Ctx.SignedCharTy);
565       case LengthModifier::AsShort:
566         return ArgType::PtrTo(Ctx.ShortTy);
567       case LengthModifier::AsLong:
568         return ArgType::PtrTo(Ctx.LongTy);
569       case LengthModifier::AsLongLong:
570       case LengthModifier::AsQuad:
571         return ArgType::PtrTo(Ctx.LongLongTy);
572       case LengthModifier::AsIntMax:
573         return ArgType::PtrTo(ArgType(Ctx.getIntMaxType(), "intmax_t"));
574       case LengthModifier::AsSizeT:
575         return ArgType::PtrTo(ArgType(Ctx.getSignedSizeType(), "ssize_t"));
576       case LengthModifier::AsPtrDiff:
577         return ArgType::PtrTo(ArgType(Ctx.getPointerDiffType(), "ptrdiff_t"));
578       case LengthModifier::AsLongDouble:
579         return ArgType(); // FIXME: Is this a known extension?
580       case LengthModifier::AsAllocate:
581       case LengthModifier::AsMAllocate:
582       case LengthModifier::AsInt32:
583       case LengthModifier::AsInt3264:
584       case LengthModifier::AsInt64:
585       case LengthModifier::AsWide:
586         return ArgType::Invalid();
587     }
588   }
589 
590   switch (CS.getKind()) {
591     case ConversionSpecifier::sArg:
592       if (LM.getKind() == LengthModifier::AsWideChar) {
593         if (IsObjCLiteral)
594           return ArgType(Ctx.getPointerType(Ctx.UnsignedShortTy.withConst()),
595                          "const unichar *");
596         return ArgType(ArgType::WCStrTy, "wchar_t *");
597       }
598       if (LM.getKind() == LengthModifier::AsWide)
599         return ArgType(ArgType::WCStrTy, "wchar_t *");
600       return ArgType::CStrTy;
601     case ConversionSpecifier::SArg:
602       if (IsObjCLiteral)
603         return ArgType(Ctx.getPointerType(Ctx.UnsignedShortTy.withConst()),
604                        "const unichar *");
605       if (Ctx.getTargetInfo().getTriple().isOSMSVCRT() &&
606           LM.getKind() == LengthModifier::AsShort)
607         return ArgType::CStrTy;
608       return ArgType(ArgType::WCStrTy, "wchar_t *");
609     case ConversionSpecifier::CArg:
610       if (IsObjCLiteral)
611         return ArgType(Ctx.UnsignedShortTy, "unichar");
612       if (Ctx.getTargetInfo().getTriple().isOSMSVCRT() &&
613           LM.getKind() == LengthModifier::AsShort)
614         return Ctx.IntTy;
615       return ArgType(Ctx.WideCharTy, "wchar_t");
616     case ConversionSpecifier::pArg:
617     case ConversionSpecifier::PArg:
618       return ArgType::CPointerTy;
619     case ConversionSpecifier::ObjCObjArg:
620       return ArgType::ObjCPointerTy;
621     default:
622       break;
623   }
624 
625   // FIXME: Handle other cases.
626   return ArgType();
627 }
628 
629 bool PrintfSpecifier::fixType(QualType QT, const LangOptions &LangOpt,
630                               ASTContext &Ctx, bool IsObjCLiteral) {
631   // %n is different from other conversion specifiers; don't try to fix it.
632   if (CS.getKind() == ConversionSpecifier::nArg)
633     return false;
634 
635   // Handle Objective-C objects first. Note that while the '%@' specifier will
636   // not warn for structure pointer or void pointer arguments (because that's
637   // how CoreFoundation objects are implemented), we only show a fixit for '%@'
638   // if we know it's an object (block, id, class, or __attribute__((NSObject))).
639   if (QT->isObjCRetainableType()) {
640     if (!IsObjCLiteral)
641       return false;
642 
643     CS.setKind(ConversionSpecifier::ObjCObjArg);
644 
645     // Disable irrelevant flags
646     HasThousandsGrouping = false;
647     HasPlusPrefix = false;
648     HasSpacePrefix = false;
649     HasAlternativeForm = false;
650     HasLeadingZeroes = false;
651     Precision.setHowSpecified(OptionalAmount::NotSpecified);
652     LM.setKind(LengthModifier::None);
653 
654     return true;
655   }
656 
657   // Handle strings next (char *, wchar_t *)
658   if (QT->isPointerType() && (QT->getPointeeType()->isAnyCharacterType())) {
659     CS.setKind(ConversionSpecifier::sArg);
660 
661     // Disable irrelevant flags
662     HasAlternativeForm = 0;
663     HasLeadingZeroes = 0;
664 
665     // Set the long length modifier for wide characters
666     if (QT->getPointeeType()->isWideCharType())
667       LM.setKind(LengthModifier::AsWideChar);
668     else
669       LM.setKind(LengthModifier::None);
670 
671     return true;
672   }
673 
674   // If it's an enum, get its underlying type.
675   if (const EnumType *ETy = QT->getAs<EnumType>())
676     QT = ETy->getDecl()->getIntegerType();
677 
678   // We can only work with builtin types.
679   const BuiltinType *BT = QT->getAs<BuiltinType>();
680   if (!BT)
681     return false;
682 
683   // Set length modifier
684   switch (BT->getKind()) {
685   case BuiltinType::Bool:
686   case BuiltinType::WChar_U:
687   case BuiltinType::WChar_S:
688   case BuiltinType::Char8: // FIXME: Treat like 'char'?
689   case BuiltinType::Char16:
690   case BuiltinType::Char32:
691   case BuiltinType::UInt128:
692   case BuiltinType::Int128:
693   case BuiltinType::Half:
694   case BuiltinType::Float16:
695   case BuiltinType::Float128:
696   case BuiltinType::ShortAccum:
697   case BuiltinType::Accum:
698   case BuiltinType::LongAccum:
699   case BuiltinType::UShortAccum:
700   case BuiltinType::UAccum:
701   case BuiltinType::ULongAccum:
702   case BuiltinType::ShortFract:
703   case BuiltinType::Fract:
704   case BuiltinType::LongFract:
705   case BuiltinType::UShortFract:
706   case BuiltinType::UFract:
707   case BuiltinType::ULongFract:
708   case BuiltinType::SatShortAccum:
709   case BuiltinType::SatAccum:
710   case BuiltinType::SatLongAccum:
711   case BuiltinType::SatUShortAccum:
712   case BuiltinType::SatUAccum:
713   case BuiltinType::SatULongAccum:
714   case BuiltinType::SatShortFract:
715   case BuiltinType::SatFract:
716   case BuiltinType::SatLongFract:
717   case BuiltinType::SatUShortFract:
718   case BuiltinType::SatUFract:
719   case BuiltinType::SatULongFract:
720     // Various types which are non-trivial to correct.
721     return false;
722 
723 #define IMAGE_TYPE(ImgType, Id, SingletonId, Access, Suffix) \
724   case BuiltinType::Id:
725 #include "clang/Basic/OpenCLImageTypes.def"
726 #define SIGNED_TYPE(Id, SingletonId)
727 #define UNSIGNED_TYPE(Id, SingletonId)
728 #define FLOATING_TYPE(Id, SingletonId)
729 #define BUILTIN_TYPE(Id, SingletonId) \
730   case BuiltinType::Id:
731 #include "clang/AST/BuiltinTypes.def"
732     // Misc other stuff which doesn't make sense here.
733     return false;
734 
735   case BuiltinType::UInt:
736   case BuiltinType::Int:
737   case BuiltinType::Float:
738   case BuiltinType::Double:
739     LM.setKind(LengthModifier::None);
740     break;
741 
742   case BuiltinType::Char_U:
743   case BuiltinType::UChar:
744   case BuiltinType::Char_S:
745   case BuiltinType::SChar:
746     LM.setKind(LengthModifier::AsChar);
747     break;
748 
749   case BuiltinType::Short:
750   case BuiltinType::UShort:
751     LM.setKind(LengthModifier::AsShort);
752     break;
753 
754   case BuiltinType::Long:
755   case BuiltinType::ULong:
756     LM.setKind(LengthModifier::AsLong);
757     break;
758 
759   case BuiltinType::LongLong:
760   case BuiltinType::ULongLong:
761     LM.setKind(LengthModifier::AsLongLong);
762     break;
763 
764   case BuiltinType::LongDouble:
765     LM.setKind(LengthModifier::AsLongDouble);
766     break;
767   }
768 
769   // Handle size_t, ptrdiff_t, etc. that have dedicated length modifiers in C99.
770   if (isa<TypedefType>(QT) && (LangOpt.C99 || LangOpt.CPlusPlus11))
771     namedTypeToLengthModifier(QT, LM);
772 
773   // If fixing the length modifier was enough, we might be done.
774   if (hasValidLengthModifier(Ctx.getTargetInfo())) {
775     // If we're going to offer a fix anyway, make sure the sign matches.
776     switch (CS.getKind()) {
777     case ConversionSpecifier::uArg:
778     case ConversionSpecifier::UArg:
779       if (QT->isSignedIntegerType())
780         CS.setKind(clang::analyze_format_string::ConversionSpecifier::dArg);
781       break;
782     case ConversionSpecifier::dArg:
783     case ConversionSpecifier::DArg:
784     case ConversionSpecifier::iArg:
785       if (QT->isUnsignedIntegerType() && !HasPlusPrefix)
786         CS.setKind(clang::analyze_format_string::ConversionSpecifier::uArg);
787       break;
788     default:
789       // Other specifiers do not have signed/unsigned variants.
790       break;
791     }
792 
793     const analyze_printf::ArgType &ATR = getArgType(Ctx, IsObjCLiteral);
794     if (ATR.isValid() && ATR.matchesType(Ctx, QT))
795       return true;
796   }
797 
798   // Set conversion specifier and disable any flags which do not apply to it.
799   // Let typedefs to char fall through to int, as %c is silly for uint8_t.
800   if (!isa<TypedefType>(QT) && QT->isCharType()) {
801     CS.setKind(ConversionSpecifier::cArg);
802     LM.setKind(LengthModifier::None);
803     Precision.setHowSpecified(OptionalAmount::NotSpecified);
804     HasAlternativeForm = 0;
805     HasLeadingZeroes = 0;
806     HasPlusPrefix = 0;
807   }
808   // Test for Floating type first as LongDouble can pass isUnsignedIntegerType
809   else if (QT->isRealFloatingType()) {
810     CS.setKind(ConversionSpecifier::fArg);
811   }
812   else if (QT->isSignedIntegerType()) {
813     CS.setKind(ConversionSpecifier::dArg);
814     HasAlternativeForm = 0;
815   }
816   else if (QT->isUnsignedIntegerType()) {
817     CS.setKind(ConversionSpecifier::uArg);
818     HasAlternativeForm = 0;
819     HasPlusPrefix = 0;
820   } else {
821     llvm_unreachable("Unexpected type");
822   }
823 
824   return true;
825 }
826 
827 void PrintfSpecifier::toString(raw_ostream &os) const {
828   // Whilst some features have no defined order, we are using the order
829   // appearing in the C99 standard (ISO/IEC 9899:1999 (E) 7.19.6.1)
830   os << "%";
831 
832   // Positional args
833   if (usesPositionalArg()) {
834     os << getPositionalArgIndex() << "$";
835   }
836 
837   // Conversion flags
838   if (IsLeftJustified)    os << "-";
839   if (HasPlusPrefix)      os << "+";
840   if (HasSpacePrefix)     os << " ";
841   if (HasAlternativeForm) os << "#";
842   if (HasLeadingZeroes)   os << "0";
843 
844   // Minimum field width
845   FieldWidth.toString(os);
846   // Precision
847   Precision.toString(os);
848   // Length modifier
849   os << LM.toString();
850   // Conversion specifier
851   os << CS.toString();
852 }
853 
854 bool PrintfSpecifier::hasValidPlusPrefix() const {
855   if (!HasPlusPrefix)
856     return true;
857 
858   // The plus prefix only makes sense for signed conversions
859   switch (CS.getKind()) {
860   case ConversionSpecifier::dArg:
861   case ConversionSpecifier::DArg:
862   case ConversionSpecifier::iArg:
863   case ConversionSpecifier::fArg:
864   case ConversionSpecifier::FArg:
865   case ConversionSpecifier::eArg:
866   case ConversionSpecifier::EArg:
867   case ConversionSpecifier::gArg:
868   case ConversionSpecifier::GArg:
869   case ConversionSpecifier::aArg:
870   case ConversionSpecifier::AArg:
871   case ConversionSpecifier::FreeBSDrArg:
872   case ConversionSpecifier::FreeBSDyArg:
873     return true;
874 
875   default:
876     return false;
877   }
878 }
879 
880 bool PrintfSpecifier::hasValidAlternativeForm() const {
881   if (!HasAlternativeForm)
882     return true;
883 
884   // Alternate form flag only valid with the oxXaAeEfFgG conversions
885   switch (CS.getKind()) {
886   case ConversionSpecifier::oArg:
887   case ConversionSpecifier::OArg:
888   case ConversionSpecifier::xArg:
889   case ConversionSpecifier::XArg:
890   case ConversionSpecifier::aArg:
891   case ConversionSpecifier::AArg:
892   case ConversionSpecifier::eArg:
893   case ConversionSpecifier::EArg:
894   case ConversionSpecifier::fArg:
895   case ConversionSpecifier::FArg:
896   case ConversionSpecifier::gArg:
897   case ConversionSpecifier::GArg:
898   case ConversionSpecifier::FreeBSDrArg:
899   case ConversionSpecifier::FreeBSDyArg:
900     return true;
901 
902   default:
903     return false;
904   }
905 }
906 
907 bool PrintfSpecifier::hasValidLeadingZeros() const {
908   if (!HasLeadingZeroes)
909     return true;
910 
911   // Leading zeroes flag only valid with the diouxXaAeEfFgG conversions
912   switch (CS.getKind()) {
913   case ConversionSpecifier::dArg:
914   case ConversionSpecifier::DArg:
915   case ConversionSpecifier::iArg:
916   case ConversionSpecifier::oArg:
917   case ConversionSpecifier::OArg:
918   case ConversionSpecifier::uArg:
919   case ConversionSpecifier::UArg:
920   case ConversionSpecifier::xArg:
921   case ConversionSpecifier::XArg:
922   case ConversionSpecifier::aArg:
923   case ConversionSpecifier::AArg:
924   case ConversionSpecifier::eArg:
925   case ConversionSpecifier::EArg:
926   case ConversionSpecifier::fArg:
927   case ConversionSpecifier::FArg:
928   case ConversionSpecifier::gArg:
929   case ConversionSpecifier::GArg:
930   case ConversionSpecifier::FreeBSDrArg:
931   case ConversionSpecifier::FreeBSDyArg:
932     return true;
933 
934   default:
935     return false;
936   }
937 }
938 
939 bool PrintfSpecifier::hasValidSpacePrefix() const {
940   if (!HasSpacePrefix)
941     return true;
942 
943   // The space prefix only makes sense for signed conversions
944   switch (CS.getKind()) {
945   case ConversionSpecifier::dArg:
946   case ConversionSpecifier::DArg:
947   case ConversionSpecifier::iArg:
948   case ConversionSpecifier::fArg:
949   case ConversionSpecifier::FArg:
950   case ConversionSpecifier::eArg:
951   case ConversionSpecifier::EArg:
952   case ConversionSpecifier::gArg:
953   case ConversionSpecifier::GArg:
954   case ConversionSpecifier::aArg:
955   case ConversionSpecifier::AArg:
956   case ConversionSpecifier::FreeBSDrArg:
957   case ConversionSpecifier::FreeBSDyArg:
958     return true;
959 
960   default:
961     return false;
962   }
963 }
964 
965 bool PrintfSpecifier::hasValidLeftJustified() const {
966   if (!IsLeftJustified)
967     return true;
968 
969   // The left justified flag is valid for all conversions except n
970   switch (CS.getKind()) {
971   case ConversionSpecifier::nArg:
972     return false;
973 
974   default:
975     return true;
976   }
977 }
978 
979 bool PrintfSpecifier::hasValidThousandsGroupingPrefix() const {
980   if (!HasThousandsGrouping)
981     return true;
982 
983   switch (CS.getKind()) {
984     case ConversionSpecifier::dArg:
985     case ConversionSpecifier::DArg:
986     case ConversionSpecifier::iArg:
987     case ConversionSpecifier::uArg:
988     case ConversionSpecifier::UArg:
989     case ConversionSpecifier::fArg:
990     case ConversionSpecifier::FArg:
991     case ConversionSpecifier::gArg:
992     case ConversionSpecifier::GArg:
993       return true;
994     default:
995       return false;
996   }
997 }
998 
999 bool PrintfSpecifier::hasValidPrecision() const {
1000   if (Precision.getHowSpecified() == OptionalAmount::NotSpecified)
1001     return true;
1002 
1003   // Precision is only valid with the diouxXaAeEfFgGsP conversions
1004   switch (CS.getKind()) {
1005   case ConversionSpecifier::dArg:
1006   case ConversionSpecifier::DArg:
1007   case ConversionSpecifier::iArg:
1008   case ConversionSpecifier::oArg:
1009   case ConversionSpecifier::OArg:
1010   case ConversionSpecifier::uArg:
1011   case ConversionSpecifier::UArg:
1012   case ConversionSpecifier::xArg:
1013   case ConversionSpecifier::XArg:
1014   case ConversionSpecifier::aArg:
1015   case ConversionSpecifier::AArg:
1016   case ConversionSpecifier::eArg:
1017   case ConversionSpecifier::EArg:
1018   case ConversionSpecifier::fArg:
1019   case ConversionSpecifier::FArg:
1020   case ConversionSpecifier::gArg:
1021   case ConversionSpecifier::GArg:
1022   case ConversionSpecifier::sArg:
1023   case ConversionSpecifier::FreeBSDrArg:
1024   case ConversionSpecifier::FreeBSDyArg:
1025   case ConversionSpecifier::PArg:
1026     return true;
1027 
1028   default:
1029     return false;
1030   }
1031 }
1032 bool PrintfSpecifier::hasValidFieldWidth() const {
1033   if (FieldWidth.getHowSpecified() == OptionalAmount::NotSpecified)
1034       return true;
1035 
1036   // The field width is valid for all conversions except n
1037   switch (CS.getKind()) {
1038   case ConversionSpecifier::nArg:
1039     return false;
1040 
1041   default:
1042     return true;
1043   }
1044 }
1045