1 //===- tools/dsymutil/MachODebugMapParser.cpp - Parse STABS debug maps ----===//
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
8 //===----------------------------------------------------------------------===//
10 #include "BinaryHolder.h"
13 #include "llvm/ADT/Optional.h"
14 #include "llvm/Object/MachO.h"
15 #include "llvm/Support/Path.h"
16 #include "llvm/Support/raw_ostream.h"
20 using namespace llvm::dsymutil;
21 using namespace llvm::object;
23 class MachODebugMapParser {
25 MachODebugMapParser(StringRef BinaryPath, ArrayRef<std::string> Archs,
26 StringRef PathPrefix = "", bool Verbose = false)
27 : BinaryPath(BinaryPath), Archs(Archs.begin(), Archs.end()),
28 PathPrefix(PathPrefix), MainBinaryHolder(Verbose),
29 CurrentObjectHolder(Verbose), CurrentDebugMapObject(nullptr) {}
31 /// \brief Parses and returns the DebugMaps of the input binary.
32 /// The binary contains multiple maps in case it is a universal
34 /// \returns an error in case the provided BinaryPath doesn't exist
35 /// or isn't of a supported type.
36 ErrorOr<std::vector<std::unique_ptr<DebugMap>>> parse();
38 /// Walk the symbol table and dump it.
42 std::string BinaryPath;
43 SmallVector<StringRef, 1> Archs;
44 std::string PathPrefix;
46 /// Owns the MemoryBuffer for the main binary.
47 BinaryHolder MainBinaryHolder;
48 /// Map of the binary symbol addresses.
49 StringMap<uint64_t> MainBinarySymbolAddresses;
50 StringRef MainBinaryStrings;
51 /// The constructed DebugMap.
52 std::unique_ptr<DebugMap> Result;
54 /// Owns the MemoryBuffer for the currently handled object file.
55 BinaryHolder CurrentObjectHolder;
56 /// Map of the currently processed object file symbol addresses.
57 StringMap<Optional<uint64_t>> CurrentObjectAddresses;
58 /// Element of the debug map corresponfing to the current object file.
59 DebugMapObject *CurrentDebugMapObject;
61 /// Holds function info while function scope processing.
62 const char *CurrentFunctionName;
63 uint64_t CurrentFunctionAddress;
65 std::unique_ptr<DebugMap> parseOneBinary(const MachOObjectFile &MainBinary,
66 StringRef BinaryPath);
68 void switchToNewDebugMapObject(StringRef Filename, sys::TimeValue Timestamp);
69 void resetParserState();
70 uint64_t getMainBinarySymbolAddress(StringRef Name);
71 void loadMainBinarySymbols(const MachOObjectFile &MainBinary);
72 void loadCurrentObjectFileSymbols(const object::MachOObjectFile &Obj);
73 void handleStabSymbolTableEntry(uint32_t StringIndex, uint8_t Type,
74 uint8_t SectionIndex, uint16_t Flags,
77 template <typename STEType> void handleStabDebugMapEntry(const STEType &STE) {
78 handleStabSymbolTableEntry(STE.n_strx, STE.n_type, STE.n_sect, STE.n_desc,
82 /// Dump the symbol table output header.
83 void dumpSymTabHeader(raw_ostream &OS, StringRef Arch);
85 /// Dump the contents of nlist entries.
86 void dumpSymTabEntry(raw_ostream &OS, uint64_t Index, uint32_t StringIndex,
87 uint8_t Type, uint8_t SectionIndex, uint16_t Flags,
90 template <typename STEType>
91 void dumpSymTabEntry(raw_ostream &OS, uint64_t Index, const STEType &STE) {
92 dumpSymTabEntry(OS, Index, STE.n_strx, STE.n_type, STE.n_sect, STE.n_desc,
95 void dumpOneBinaryStab(const MachOObjectFile &MainBinary,
96 StringRef BinaryPath);
99 static void Warning(const Twine &Msg) { errs() << "warning: " + Msg + "\n"; }
100 } // anonymous namespace
102 /// Reset the parser state coresponding to the current object
103 /// file. This is to be called after an object file is finished
105 void MachODebugMapParser::resetParserState() {
106 CurrentObjectAddresses.clear();
107 CurrentDebugMapObject = nullptr;
110 /// Create a new DebugMapObject. This function resets the state of the
111 /// parser that was referring to the last object file and sets
112 /// everything up to add symbols to the new one.
113 void MachODebugMapParser::switchToNewDebugMapObject(StringRef Filename,
114 sys::TimeValue Timestamp) {
117 SmallString<80> Path(PathPrefix);
118 sys::path::append(Path, Filename);
121 CurrentObjectHolder.GetFilesAs<MachOObjectFile>(Path, Timestamp);
122 if (auto Error = MachOOrError.getError()) {
123 Warning(Twine("cannot open debug object \"") + Path.str() + "\": " +
124 Error.message() + "\n");
129 CurrentObjectHolder.GetAs<MachOObjectFile>(Result->getTriple());
130 if (auto Err = ErrOrAchObj.getError()) {
131 return Warning(Twine("cannot open debug object \"") + Path.str() + "\": " +
132 Err.message() + "\n");
135 CurrentDebugMapObject = &Result->addDebugMapObject(Path, Timestamp);
136 loadCurrentObjectFileSymbols(*ErrOrAchObj);
139 static std::string getArchName(const object::MachOObjectFile &Obj) {
141 Triple T = Obj.getArch(nullptr, &ThumbTriple);
142 return T.getArchName();
145 std::unique_ptr<DebugMap>
146 MachODebugMapParser::parseOneBinary(const MachOObjectFile &MainBinary,
147 StringRef BinaryPath) {
148 loadMainBinarySymbols(MainBinary);
150 make_unique<DebugMap>(BinaryHolder::getTriple(MainBinary), BinaryPath);
151 MainBinaryStrings = MainBinary.getStringTableData();
152 for (const SymbolRef &Symbol : MainBinary.symbols()) {
153 const DataRefImpl &DRI = Symbol.getRawDataRefImpl();
154 if (MainBinary.is64Bit())
155 handleStabDebugMapEntry(MainBinary.getSymbol64TableEntry(DRI));
157 handleStabDebugMapEntry(MainBinary.getSymbolTableEntry(DRI));
161 return std::move(Result);
164 // Table that maps Darwin's Mach-O stab constants to strings to allow printing.
165 // llvm-nm has very similar code, the strings used here are however slightly
166 // different and part of the interface of dsymutil (some project's build-systems
167 // parse the ouptut of dsymutil -s), thus they shouldn't be changed.
168 struct DarwinStabName {
173 static const struct DarwinStabName DarwinStabNames[] = {
174 {MachO::N_GSYM, "N_GSYM"}, {MachO::N_FNAME, "N_FNAME"},
175 {MachO::N_FUN, "N_FUN"}, {MachO::N_STSYM, "N_STSYM"},
176 {MachO::N_LCSYM, "N_LCSYM"}, {MachO::N_BNSYM, "N_BNSYM"},
177 {MachO::N_PC, "N_PC"}, {MachO::N_AST, "N_AST"},
178 {MachO::N_OPT, "N_OPT"}, {MachO::N_RSYM, "N_RSYM"},
179 {MachO::N_SLINE, "N_SLINE"}, {MachO::N_ENSYM, "N_ENSYM"},
180 {MachO::N_SSYM, "N_SSYM"}, {MachO::N_SO, "N_SO"},
181 {MachO::N_OSO, "N_OSO"}, {MachO::N_LSYM, "N_LSYM"},
182 {MachO::N_BINCL, "N_BINCL"}, {MachO::N_SOL, "N_SOL"},
183 {MachO::N_PARAMS, "N_PARAM"}, {MachO::N_VERSION, "N_VERS"},
184 {MachO::N_OLEVEL, "N_OLEV"}, {MachO::N_PSYM, "N_PSYM"},
185 {MachO::N_EINCL, "N_EINCL"}, {MachO::N_ENTRY, "N_ENTRY"},
186 {MachO::N_LBRAC, "N_LBRAC"}, {MachO::N_EXCL, "N_EXCL"},
187 {MachO::N_RBRAC, "N_RBRAC"}, {MachO::N_BCOMM, "N_BCOMM"},
188 {MachO::N_ECOMM, "N_ECOMM"}, {MachO::N_ECOML, "N_ECOML"},
189 {MachO::N_LENG, "N_LENG"}, {0, nullptr}};
191 static const char *getDarwinStabString(uint8_t NType) {
192 for (unsigned i = 0; DarwinStabNames[i].Name; i++) {
193 if (DarwinStabNames[i].NType == NType)
194 return DarwinStabNames[i].Name;
199 void MachODebugMapParser::dumpSymTabHeader(raw_ostream &OS, StringRef Arch) {
200 OS << "-----------------------------------"
201 "-----------------------------------\n";
202 OS << "Symbol table for: '" << BinaryPath << "' (" << Arch.data() << ")\n";
203 OS << "-----------------------------------"
204 "-----------------------------------\n";
205 OS << "Index n_strx n_type n_sect n_desc n_value\n";
206 OS << "======== -------- ------------------ ------ ------ ----------------\n";
209 void MachODebugMapParser::dumpSymTabEntry(raw_ostream &OS, uint64_t Index,
210 uint32_t StringIndex, uint8_t Type,
211 uint8_t SectionIndex, uint16_t Flags,
214 OS << '[' << format_decimal(Index, 6) << "] "
216 << format_hex_no_prefix(StringIndex, 8) << ' '
218 << format_hex_no_prefix(Type, 2) << " (";
220 if (Type & MachO::N_STAB)
221 OS << left_justify(getDarwinStabString(Type), 13);
223 if (Type & MachO::N_PEXT)
227 switch (Type & MachO::N_TYPE) {
228 case MachO::N_UNDF: // 0x0 undefined, n_sect == NO_SECT
231 case MachO::N_ABS: // 0x2 absolute, n_sect == NO_SECT
234 case MachO::N_SECT: // 0xe defined in section number n_sect
237 case MachO::N_PBUD: // 0xc prebound undefined (defined in a dylib)
240 case MachO::N_INDR: // 0xa indirect
244 OS << format_hex_no_prefix(Type, 2) << " ";
247 if (Type & MachO::N_EXT)
255 << format_hex_no_prefix(SectionIndex, 2) << " "
257 << format_hex_no_prefix(Flags, 4) << " "
259 << format_hex_no_prefix(Value, 16);
261 const char *Name = &MainBinaryStrings.data()[StringIndex];
263 OS << " '" << Name << "'";
268 void MachODebugMapParser::dumpOneBinaryStab(const MachOObjectFile &MainBinary,
269 StringRef BinaryPath) {
270 loadMainBinarySymbols(MainBinary);
271 MainBinaryStrings = MainBinary.getStringTableData();
272 raw_ostream &OS(llvm::outs());
274 dumpSymTabHeader(OS, getArchName(MainBinary));
276 for (const SymbolRef &Symbol : MainBinary.symbols()) {
277 const DataRefImpl &DRI = Symbol.getRawDataRefImpl();
278 if (MainBinary.is64Bit())
279 dumpSymTabEntry(OS, Idx, MainBinary.getSymbol64TableEntry(DRI));
281 dumpSymTabEntry(OS, Idx, MainBinary.getSymbolTableEntry(DRI));
289 static bool shouldLinkArch(SmallVectorImpl<StringRef> &Archs, StringRef Arch) {
291 std::find(Archs.begin(), Archs.end(), "all") != Archs.end() ||
292 std::find(Archs.begin(), Archs.end(), "*") != Archs.end())
295 if (Arch.startswith("arm") && Arch != "arm64" &&
296 std::find(Archs.begin(), Archs.end(), "arm") != Archs.end())
299 return std::find(Archs.begin(), Archs.end(), Arch) != Archs.end();
302 bool MachODebugMapParser::dumpStab() {
303 auto MainBinOrError =
304 MainBinaryHolder.GetFilesAs<MachOObjectFile>(BinaryPath);
305 if (auto Error = MainBinOrError.getError()) {
306 llvm::errs() << "Cannot get '" << BinaryPath
307 << "' as MachO file: " << Error.message() << "\n";
312 for (const auto *Binary : *MainBinOrError)
313 if (shouldLinkArch(Archs, Binary->getArch(nullptr, &T).getArchName()))
314 dumpOneBinaryStab(*Binary, BinaryPath);
319 /// This main parsing routine tries to open the main binary and if
320 /// successful iterates over the STAB entries. The real parsing is
321 /// done in handleStabSymbolTableEntry.
322 ErrorOr<std::vector<std::unique_ptr<DebugMap>>> MachODebugMapParser::parse() {
323 auto MainBinOrError =
324 MainBinaryHolder.GetFilesAs<MachOObjectFile>(BinaryPath);
325 if (auto Error = MainBinOrError.getError())
328 std::vector<std::unique_ptr<DebugMap>> Results;
330 for (const auto *Binary : *MainBinOrError)
331 if (shouldLinkArch(Archs, Binary->getArch(nullptr, &T).getArchName()))
332 Results.push_back(parseOneBinary(*Binary, BinaryPath));
334 return std::move(Results);
337 /// Interpret the STAB entries to fill the DebugMap.
338 void MachODebugMapParser::handleStabSymbolTableEntry(uint32_t StringIndex,
340 uint8_t SectionIndex,
343 if (!(Type & MachO::N_STAB))
346 const char *Name = &MainBinaryStrings.data()[StringIndex];
348 // An N_OSO entry represents the start of a new object file description.
349 if (Type == MachO::N_OSO) {
350 sys::TimeValue Timestamp;
351 Timestamp.fromEpochTime(Value);
352 return switchToNewDebugMapObject(Name, Timestamp);
355 // If the last N_OSO object file wasn't found,
356 // CurrentDebugMapObject will be null. Do not update anything
357 // until we find the next valid N_OSO entry.
358 if (!CurrentDebugMapObject)
364 // This is a global variable. We need to query the main binary
365 // symbol table to find its address as it might not be in the
366 // debug map (for common symbols).
367 Value = getMainBinarySymbolAddress(Name);
370 // Functions are scopes in STABS. They have an end marker that
371 // contains the function size.
372 if (Name[0] == '\0') {
374 Value = CurrentFunctionAddress;
375 Name = CurrentFunctionName;
378 CurrentFunctionName = Name;
379 CurrentFunctionAddress = Value;
388 auto ObjectSymIt = CurrentObjectAddresses.find(Name);
389 if (ObjectSymIt == CurrentObjectAddresses.end())
390 return Warning("could not find object file symbol for symbol " +
392 if (!CurrentDebugMapObject->addSymbol(Name, ObjectSymIt->getValue(), Value,
394 return Warning(Twine("failed to insert symbol '") + Name +
395 "' in the debug map.");
398 /// Load the current object file symbols into CurrentObjectAddresses.
399 void MachODebugMapParser::loadCurrentObjectFileSymbols(
400 const object::MachOObjectFile &Obj) {
401 CurrentObjectAddresses.clear();
403 for (auto Sym : Obj.symbols()) {
404 uint64_t Addr = Sym.getValue();
405 ErrorOr<StringRef> Name = Sym.getName();
408 // The value of some categories of symbols isn't meaningful. For
409 // example common symbols store their size in the value field, not
410 // their address. Absolute symbols have a fixed address that can
411 // conflict with standard symbols. These symbols (especially the
412 // common ones), might still be referenced by relocations. These
413 // relocations will use the symbol itself, and won't need an
414 // object file address. The object file address field is optional
415 // in the DebugMap, leave it unassigned for these symbols.
416 if (Sym.getFlags() & (SymbolRef::SF_Absolute | SymbolRef::SF_Common))
417 CurrentObjectAddresses[*Name] = None;
419 CurrentObjectAddresses[*Name] = Addr;
423 /// Lookup a symbol address in the main binary symbol table. The
424 /// parser only needs to query common symbols, thus not every symbol's
425 /// address is available through this function.
426 uint64_t MachODebugMapParser::getMainBinarySymbolAddress(StringRef Name) {
427 auto Sym = MainBinarySymbolAddresses.find(Name);
428 if (Sym == MainBinarySymbolAddresses.end())
433 /// Load the interesting main binary symbols' addresses into
434 /// MainBinarySymbolAddresses.
435 void MachODebugMapParser::loadMainBinarySymbols(
436 const MachOObjectFile &MainBinary) {
437 section_iterator Section = MainBinary.section_end();
438 MainBinarySymbolAddresses.clear();
439 for (const auto &Sym : MainBinary.symbols()) {
440 SymbolRef::Type Type = Sym.getType();
441 // Skip undefined and STAB entries.
442 if ((Type & SymbolRef::ST_Debug) || (Type & SymbolRef::ST_Unknown))
444 // The only symbols of interest are the global variables. These
445 // are the only ones that need to be queried because the address
446 // of common data won't be described in the debug map. All other
447 // addresses should be fetched for the debug map.
448 if (!(Sym.getFlags() & SymbolRef::SF_Global))
450 ErrorOr<section_iterator> SectionOrErr = Sym.getSection();
453 Section = *SectionOrErr;
454 if (Section == MainBinary.section_end() || Section->isText())
456 uint64_t Addr = Sym.getValue();
457 ErrorOr<StringRef> NameOrErr = Sym.getName();
460 StringRef Name = *NameOrErr;
461 if (Name.size() == 0 || Name[0] == '\0')
463 MainBinarySymbolAddresses[Name] = Addr;
469 llvm::ErrorOr<std::vector<std::unique_ptr<DebugMap>>>
470 parseDebugMap(StringRef InputFile, ArrayRef<std::string> Archs,
471 StringRef PrependPath, bool Verbose, bool InputIsYAML) {
473 MachODebugMapParser Parser(InputFile, Archs, PrependPath, Verbose);
474 return Parser.parse();
476 return DebugMap::parseYAMLDebugMap(InputFile, PrependPath, Verbose);
480 bool dumpStab(StringRef InputFile, ArrayRef<std::string> Archs,
481 StringRef PrependPath) {
482 MachODebugMapParser Parser(InputFile, Archs, PrependPath, false);
483 return Parser.dumpStab();
485 } // namespace dsymutil