MachODump.cpp revision dce4a407a24b04eebc6a376f8e62b41aaa7b071f
1//===-- MachODump.cpp - Object file dumping utility for llvm --------------===// 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// This file implements the MachO-specific dumper for llvm-objdump. 11// 12//===----------------------------------------------------------------------===// 13 14#include "llvm-objdump.h" 15#include "llvm/ADT/STLExtras.h" 16#include "llvm/ADT/StringExtras.h" 17#include "llvm/ADT/Triple.h" 18#include "llvm/DebugInfo/DIContext.h" 19#include "llvm/MC/MCAsmInfo.h" 20#include "llvm/MC/MCContext.h" 21#include "llvm/MC/MCDisassembler.h" 22#include "llvm/MC/MCInst.h" 23#include "llvm/MC/MCInstPrinter.h" 24#include "llvm/MC/MCInstrAnalysis.h" 25#include "llvm/MC/MCInstrDesc.h" 26#include "llvm/MC/MCInstrInfo.h" 27#include "llvm/MC/MCRegisterInfo.h" 28#include "llvm/MC/MCSubtargetInfo.h" 29#include "llvm/Object/MachO.h" 30#include "llvm/Support/Casting.h" 31#include "llvm/Support/CommandLine.h" 32#include "llvm/Support/Debug.h" 33#include "llvm/Support/Format.h" 34#include "llvm/Support/GraphWriter.h" 35#include "llvm/Support/MachO.h" 36#include "llvm/Support/MemoryBuffer.h" 37#include "llvm/Support/TargetRegistry.h" 38#include "llvm/Support/TargetSelect.h" 39#include "llvm/Support/raw_ostream.h" 40#include "llvm/Support/system_error.h" 41#include <algorithm> 42#include <cstring> 43using namespace llvm; 44using namespace object; 45 46static cl::opt<bool> 47 UseDbg("g", cl::desc("Print line information from debug info if available")); 48 49static cl::opt<std::string> 50 DSYMFile("dsym", cl::desc("Use .dSYM file for debug info")); 51 52static const Target *GetTarget(const MachOObjectFile *MachOObj) { 53 // Figure out the target triple. 54 if (TripleName.empty()) { 55 llvm::Triple TT("unknown-unknown-unknown"); 56 TT.setArch(Triple::ArchType(MachOObj->getArch())); 57 TripleName = TT.str(); 58 } 59 60 // Get the target specific parser. 61 std::string Error; 62 const Target *TheTarget = TargetRegistry::lookupTarget(TripleName, Error); 63 if (TheTarget) 64 return TheTarget; 65 66 errs() << "llvm-objdump: error: unable to get target for '" << TripleName 67 << "', see --version and --triple.\n"; 68 return nullptr; 69} 70 71struct SymbolSorter { 72 bool operator()(const SymbolRef &A, const SymbolRef &B) { 73 SymbolRef::Type AType, BType; 74 A.getType(AType); 75 B.getType(BType); 76 77 uint64_t AAddr, BAddr; 78 if (AType != SymbolRef::ST_Function) 79 AAddr = 0; 80 else 81 A.getAddress(AAddr); 82 if (BType != SymbolRef::ST_Function) 83 BAddr = 0; 84 else 85 B.getAddress(BAddr); 86 return AAddr < BAddr; 87 } 88}; 89 90// Types for the storted data in code table that is built before disassembly 91// and the predicate function to sort them. 92typedef std::pair<uint64_t, DiceRef> DiceTableEntry; 93typedef std::vector<DiceTableEntry> DiceTable; 94typedef DiceTable::iterator dice_table_iterator; 95 96static bool 97compareDiceTableEntries(const DiceTableEntry i, 98 const DiceTableEntry j) { 99 return i.first == j.first; 100} 101 102static void DumpDataInCode(const char *bytes, uint64_t Size, 103 unsigned short Kind) { 104 uint64_t Value; 105 106 switch (Kind) { 107 case MachO::DICE_KIND_DATA: 108 switch (Size) { 109 case 4: 110 Value = bytes[3] << 24 | 111 bytes[2] << 16 | 112 bytes[1] << 8 | 113 bytes[0]; 114 outs() << "\t.long " << Value; 115 break; 116 case 2: 117 Value = bytes[1] << 8 | 118 bytes[0]; 119 outs() << "\t.short " << Value; 120 break; 121 case 1: 122 Value = bytes[0]; 123 outs() << "\t.byte " << Value; 124 break; 125 } 126 outs() << "\t@ KIND_DATA\n"; 127 break; 128 case MachO::DICE_KIND_JUMP_TABLE8: 129 Value = bytes[0]; 130 outs() << "\t.byte " << Value << "\t@ KIND_JUMP_TABLE8"; 131 break; 132 case MachO::DICE_KIND_JUMP_TABLE16: 133 Value = bytes[1] << 8 | 134 bytes[0]; 135 outs() << "\t.short " << Value << "\t@ KIND_JUMP_TABLE16"; 136 break; 137 case MachO::DICE_KIND_JUMP_TABLE32: 138 Value = bytes[3] << 24 | 139 bytes[2] << 16 | 140 bytes[1] << 8 | 141 bytes[0]; 142 outs() << "\t.long " << Value << "\t@ KIND_JUMP_TABLE32"; 143 break; 144 default: 145 outs() << "\t@ data in code kind = " << Kind << "\n"; 146 break; 147 } 148} 149 150static void getSectionsAndSymbols(const MachO::mach_header Header, 151 MachOObjectFile *MachOObj, 152 std::vector<SectionRef> &Sections, 153 std::vector<SymbolRef> &Symbols, 154 SmallVectorImpl<uint64_t> &FoundFns, 155 uint64_t &BaseSegmentAddress) { 156 for (const SymbolRef &Symbol : MachOObj->symbols()) 157 Symbols.push_back(Symbol); 158 159 for (const SectionRef &Section : MachOObj->sections()) { 160 StringRef SectName; 161 Section.getName(SectName); 162 Sections.push_back(Section); 163 } 164 165 MachOObjectFile::LoadCommandInfo Command = 166 MachOObj->getFirstLoadCommandInfo(); 167 bool BaseSegmentAddressSet = false; 168 for (unsigned i = 0; ; ++i) { 169 if (Command.C.cmd == MachO::LC_FUNCTION_STARTS) { 170 // We found a function starts segment, parse the addresses for later 171 // consumption. 172 MachO::linkedit_data_command LLC = 173 MachOObj->getLinkeditDataLoadCommand(Command); 174 175 MachOObj->ReadULEB128s(LLC.dataoff, FoundFns); 176 } 177 else if (Command.C.cmd == MachO::LC_SEGMENT) { 178 MachO::segment_command SLC = 179 MachOObj->getSegmentLoadCommand(Command); 180 StringRef SegName = SLC.segname; 181 if(!BaseSegmentAddressSet && SegName != "__PAGEZERO") { 182 BaseSegmentAddressSet = true; 183 BaseSegmentAddress = SLC.vmaddr; 184 } 185 } 186 187 if (i == Header.ncmds - 1) 188 break; 189 else 190 Command = MachOObj->getNextLoadCommandInfo(Command); 191 } 192} 193 194static void DisassembleInputMachO2(StringRef Filename, 195 MachOObjectFile *MachOOF); 196 197void llvm::DisassembleInputMachO(StringRef Filename) { 198 std::unique_ptr<MemoryBuffer> Buff; 199 200 if (error_code ec = MemoryBuffer::getFileOrSTDIN(Filename, Buff)) { 201 errs() << "llvm-objdump: " << Filename << ": " << ec.message() << "\n"; 202 return; 203 } 204 205 std::unique_ptr<MachOObjectFile> MachOOF(static_cast<MachOObjectFile *>( 206 ObjectFile::createMachOObjectFile(Buff.release()).get())); 207 208 DisassembleInputMachO2(Filename, MachOOF.get()); 209} 210 211static void DisassembleInputMachO2(StringRef Filename, 212 MachOObjectFile *MachOOF) { 213 const Target *TheTarget = GetTarget(MachOOF); 214 if (!TheTarget) { 215 // GetTarget prints out stuff. 216 return; 217 } 218 std::unique_ptr<const MCInstrInfo> InstrInfo(TheTarget->createMCInstrInfo()); 219 std::unique_ptr<MCInstrAnalysis> InstrAnalysis( 220 TheTarget->createMCInstrAnalysis(InstrInfo.get())); 221 222 // Set up disassembler. 223 std::unique_ptr<const MCRegisterInfo> MRI( 224 TheTarget->createMCRegInfo(TripleName)); 225 std::unique_ptr<const MCAsmInfo> AsmInfo( 226 TheTarget->createMCAsmInfo(*MRI, TripleName)); 227 std::unique_ptr<const MCSubtargetInfo> STI( 228 TheTarget->createMCSubtargetInfo(TripleName, "", "")); 229 MCContext Ctx(AsmInfo.get(), MRI.get(), nullptr); 230 std::unique_ptr<const MCDisassembler> DisAsm( 231 TheTarget->createMCDisassembler(*STI, Ctx)); 232 int AsmPrinterVariant = AsmInfo->getAssemblerDialect(); 233 std::unique_ptr<MCInstPrinter> IP(TheTarget->createMCInstPrinter( 234 AsmPrinterVariant, *AsmInfo, *InstrInfo, *MRI, *STI)); 235 236 if (!InstrAnalysis || !AsmInfo || !STI || !DisAsm || !IP) { 237 errs() << "error: couldn't initialize disassembler for target " 238 << TripleName << '\n'; 239 return; 240 } 241 242 outs() << '\n' << Filename << ":\n\n"; 243 244 MachO::mach_header Header = MachOOF->getHeader(); 245 246 // FIXME: FoundFns isn't used anymore. Using symbols/LC_FUNCTION_STARTS to 247 // determine function locations will eventually go in MCObjectDisassembler. 248 // FIXME: Using the -cfg command line option, this code used to be able to 249 // annotate relocations with the referenced symbol's name, and if this was 250 // inside a __[cf]string section, the data it points to. This is now replaced 251 // by the upcoming MCSymbolizer, which needs the appropriate setup done above. 252 std::vector<SectionRef> Sections; 253 std::vector<SymbolRef> Symbols; 254 SmallVector<uint64_t, 8> FoundFns; 255 uint64_t BaseSegmentAddress; 256 257 getSectionsAndSymbols(Header, MachOOF, Sections, Symbols, FoundFns, 258 BaseSegmentAddress); 259 260 // Sort the symbols by address, just in case they didn't come in that way. 261 std::sort(Symbols.begin(), Symbols.end(), SymbolSorter()); 262 263 // Build a data in code table that is sorted on by the address of each entry. 264 uint64_t BaseAddress = 0; 265 if (Header.filetype == MachO::MH_OBJECT) 266 Sections[0].getAddress(BaseAddress); 267 else 268 BaseAddress = BaseSegmentAddress; 269 DiceTable Dices; 270 for (dice_iterator DI = MachOOF->begin_dices(), DE = MachOOF->end_dices(); 271 DI != DE; ++DI) { 272 uint32_t Offset; 273 DI->getOffset(Offset); 274 Dices.push_back(std::make_pair(BaseAddress + Offset, *DI)); 275 } 276 array_pod_sort(Dices.begin(), Dices.end()); 277 278#ifndef NDEBUG 279 raw_ostream &DebugOut = DebugFlag ? dbgs() : nulls(); 280#else 281 raw_ostream &DebugOut = nulls(); 282#endif 283 284 std::unique_ptr<DIContext> diContext; 285 ObjectFile *DbgObj = MachOOF; 286 // Try to find debug info and set up the DIContext for it. 287 if (UseDbg) { 288 // A separate DSym file path was specified, parse it as a macho file, 289 // get the sections and supply it to the section name parsing machinery. 290 if (!DSYMFile.empty()) { 291 std::unique_ptr<MemoryBuffer> Buf; 292 if (error_code ec = MemoryBuffer::getFileOrSTDIN(DSYMFile, Buf)) { 293 errs() << "llvm-objdump: " << Filename << ": " << ec.message() << '\n'; 294 return; 295 } 296 DbgObj = ObjectFile::createMachOObjectFile(Buf.release()).get(); 297 } 298 299 // Setup the DIContext 300 diContext.reset(DIContext::getDWARFContext(DbgObj)); 301 } 302 303 for (unsigned SectIdx = 0; SectIdx != Sections.size(); SectIdx++) { 304 305 bool SectIsText = false; 306 Sections[SectIdx].isText(SectIsText); 307 if (SectIsText == false) 308 continue; 309 310 StringRef SectName; 311 if (Sections[SectIdx].getName(SectName) || 312 SectName != "__text") 313 continue; // Skip non-text sections 314 315 DataRefImpl DR = Sections[SectIdx].getRawDataRefImpl(); 316 317 StringRef SegmentName = MachOOF->getSectionFinalSegmentName(DR); 318 if (SegmentName != "__TEXT") 319 continue; 320 321 StringRef Bytes; 322 Sections[SectIdx].getContents(Bytes); 323 StringRefMemoryObject memoryObject(Bytes); 324 bool symbolTableWorked = false; 325 326 // Parse relocations. 327 std::vector<std::pair<uint64_t, SymbolRef>> Relocs; 328 for (const RelocationRef &Reloc : Sections[SectIdx].relocations()) { 329 uint64_t RelocOffset, SectionAddress; 330 Reloc.getOffset(RelocOffset); 331 Sections[SectIdx].getAddress(SectionAddress); 332 RelocOffset -= SectionAddress; 333 334 symbol_iterator RelocSym = Reloc.getSymbol(); 335 336 Relocs.push_back(std::make_pair(RelocOffset, *RelocSym)); 337 } 338 array_pod_sort(Relocs.begin(), Relocs.end()); 339 340 // Disassemble symbol by symbol. 341 for (unsigned SymIdx = 0; SymIdx != Symbols.size(); SymIdx++) { 342 StringRef SymName; 343 Symbols[SymIdx].getName(SymName); 344 345 SymbolRef::Type ST; 346 Symbols[SymIdx].getType(ST); 347 if (ST != SymbolRef::ST_Function) 348 continue; 349 350 // Make sure the symbol is defined in this section. 351 bool containsSym = false; 352 Sections[SectIdx].containsSymbol(Symbols[SymIdx], containsSym); 353 if (!containsSym) 354 continue; 355 356 // Start at the address of the symbol relative to the section's address. 357 uint64_t SectionAddress = 0; 358 uint64_t Start = 0; 359 Sections[SectIdx].getAddress(SectionAddress); 360 Symbols[SymIdx].getAddress(Start); 361 Start -= SectionAddress; 362 363 // Stop disassembling either at the beginning of the next symbol or at 364 // the end of the section. 365 bool containsNextSym = false; 366 uint64_t NextSym = 0; 367 uint64_t NextSymIdx = SymIdx+1; 368 while (Symbols.size() > NextSymIdx) { 369 SymbolRef::Type NextSymType; 370 Symbols[NextSymIdx].getType(NextSymType); 371 if (NextSymType == SymbolRef::ST_Function) { 372 Sections[SectIdx].containsSymbol(Symbols[NextSymIdx], 373 containsNextSym); 374 Symbols[NextSymIdx].getAddress(NextSym); 375 NextSym -= SectionAddress; 376 break; 377 } 378 ++NextSymIdx; 379 } 380 381 uint64_t SectSize; 382 Sections[SectIdx].getSize(SectSize); 383 uint64_t End = containsNextSym ? NextSym : SectSize; 384 uint64_t Size; 385 386 symbolTableWorked = true; 387 388 outs() << SymName << ":\n"; 389 DILineInfo lastLine; 390 for (uint64_t Index = Start; Index < End; Index += Size) { 391 MCInst Inst; 392 393 uint64_t SectAddress = 0; 394 Sections[SectIdx].getAddress(SectAddress); 395 outs() << format("%8" PRIx64 ":\t", SectAddress + Index); 396 397 // Check the data in code table here to see if this is data not an 398 // instruction to be disassembled. 399 DiceTable Dice; 400 Dice.push_back(std::make_pair(SectAddress + Index, DiceRef())); 401 dice_table_iterator DTI = std::search(Dices.begin(), Dices.end(), 402 Dice.begin(), Dice.end(), 403 compareDiceTableEntries); 404 if (DTI != Dices.end()){ 405 uint16_t Length; 406 DTI->second.getLength(Length); 407 DumpBytes(StringRef(Bytes.data() + Index, Length)); 408 uint16_t Kind; 409 DTI->second.getKind(Kind); 410 DumpDataInCode(Bytes.data() + Index, Length, Kind); 411 continue; 412 } 413 414 if (DisAsm->getInstruction(Inst, Size, memoryObject, Index, 415 DebugOut, nulls())) { 416 DumpBytes(StringRef(Bytes.data() + Index, Size)); 417 IP->printInst(&Inst, outs(), ""); 418 419 // Print debug info. 420 if (diContext) { 421 DILineInfo dli = 422 diContext->getLineInfoForAddress(SectAddress + Index); 423 // Print valid line info if it changed. 424 if (dli != lastLine && dli.Line != 0) 425 outs() << "\t## " << dli.FileName << ':' << dli.Line << ':' 426 << dli.Column; 427 lastLine = dli; 428 } 429 outs() << "\n"; 430 } else { 431 errs() << "llvm-objdump: warning: invalid instruction encoding\n"; 432 if (Size == 0) 433 Size = 1; // skip illegible bytes 434 } 435 } 436 } 437 if (!symbolTableWorked) { 438 // Reading the symbol table didn't work, disassemble the whole section. 439 uint64_t SectAddress; 440 Sections[SectIdx].getAddress(SectAddress); 441 uint64_t SectSize; 442 Sections[SectIdx].getSize(SectSize); 443 uint64_t InstSize; 444 for (uint64_t Index = 0; Index < SectSize; Index += InstSize) { 445 MCInst Inst; 446 447 if (DisAsm->getInstruction(Inst, InstSize, memoryObject, Index, 448 DebugOut, nulls())) { 449 outs() << format("%8" PRIx64 ":\t", SectAddress + Index); 450 DumpBytes(StringRef(Bytes.data() + Index, InstSize)); 451 IP->printInst(&Inst, outs(), ""); 452 outs() << "\n"; 453 } else { 454 errs() << "llvm-objdump: warning: invalid instruction encoding\n"; 455 if (InstSize == 0) 456 InstSize = 1; // skip illegible bytes 457 } 458 } 459 } 460 } 461} 462