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 <algorithm>
41#include <cstring>
42#include <system_error>
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  ErrorOr<std::unique_ptr<MemoryBuffer>> Buff =
199      MemoryBuffer::getFileOrSTDIN(Filename);
200  if (std::error_code EC = Buff.getError()) {
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.get()).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      ErrorOr<std::unique_ptr<MemoryBuffer>> Buf =
292          MemoryBuffer::getFileOrSTDIN(DSYMFile);
293      if (std::error_code EC = Buf.getError()) {
294        errs() << "llvm-objdump: " << Filename << ": " << EC.message() << '\n';
295        return;
296      }
297      DbgObj = ObjectFile::createMachOObjectFile(Buf.get()).get();
298    }
299
300    // Setup the DIContext
301    diContext.reset(DIContext::getDWARFContext(DbgObj));
302  }
303
304  for (unsigned SectIdx = 0; SectIdx != Sections.size(); SectIdx++) {
305
306    bool SectIsText = false;
307    Sections[SectIdx].isText(SectIsText);
308    if (SectIsText == false)
309      continue;
310
311    StringRef SectName;
312    if (Sections[SectIdx].getName(SectName) ||
313        SectName != "__text")
314      continue; // Skip non-text sections
315
316    DataRefImpl DR = Sections[SectIdx].getRawDataRefImpl();
317
318    StringRef SegmentName = MachOOF->getSectionFinalSegmentName(DR);
319    if (SegmentName != "__TEXT")
320      continue;
321
322    StringRef Bytes;
323    Sections[SectIdx].getContents(Bytes);
324    StringRefMemoryObject memoryObject(Bytes);
325    bool symbolTableWorked = false;
326
327    // Parse relocations.
328    std::vector<std::pair<uint64_t, SymbolRef>> Relocs;
329    for (const RelocationRef &Reloc : Sections[SectIdx].relocations()) {
330      uint64_t RelocOffset, SectionAddress;
331      Reloc.getOffset(RelocOffset);
332      Sections[SectIdx].getAddress(SectionAddress);
333      RelocOffset -= SectionAddress;
334
335      symbol_iterator RelocSym = Reloc.getSymbol();
336
337      Relocs.push_back(std::make_pair(RelocOffset, *RelocSym));
338    }
339    array_pod_sort(Relocs.begin(), Relocs.end());
340
341    // Disassemble symbol by symbol.
342    for (unsigned SymIdx = 0; SymIdx != Symbols.size(); SymIdx++) {
343      StringRef SymName;
344      Symbols[SymIdx].getName(SymName);
345
346      SymbolRef::Type ST;
347      Symbols[SymIdx].getType(ST);
348      if (ST != SymbolRef::ST_Function)
349        continue;
350
351      // Make sure the symbol is defined in this section.
352      bool containsSym = false;
353      Sections[SectIdx].containsSymbol(Symbols[SymIdx], containsSym);
354      if (!containsSym)
355        continue;
356
357      // Start at the address of the symbol relative to the section's address.
358      uint64_t SectionAddress = 0;
359      uint64_t Start = 0;
360      Sections[SectIdx].getAddress(SectionAddress);
361      Symbols[SymIdx].getAddress(Start);
362      Start -= SectionAddress;
363
364      // Stop disassembling either at the beginning of the next symbol or at
365      // the end of the section.
366      bool containsNextSym = false;
367      uint64_t NextSym = 0;
368      uint64_t NextSymIdx = SymIdx+1;
369      while (Symbols.size() > NextSymIdx) {
370        SymbolRef::Type NextSymType;
371        Symbols[NextSymIdx].getType(NextSymType);
372        if (NextSymType == SymbolRef::ST_Function) {
373          Sections[SectIdx].containsSymbol(Symbols[NextSymIdx],
374                                           containsNextSym);
375          Symbols[NextSymIdx].getAddress(NextSym);
376          NextSym -= SectionAddress;
377          break;
378        }
379        ++NextSymIdx;
380      }
381
382      uint64_t SectSize;
383      Sections[SectIdx].getSize(SectSize);
384      uint64_t End = containsNextSym ?  NextSym : SectSize;
385      uint64_t Size;
386
387      symbolTableWorked = true;
388
389      outs() << SymName << ":\n";
390      DILineInfo lastLine;
391      for (uint64_t Index = Start; Index < End; Index += Size) {
392        MCInst Inst;
393
394        uint64_t SectAddress = 0;
395        Sections[SectIdx].getAddress(SectAddress);
396        outs() << format("%8" PRIx64 ":\t", SectAddress + Index);
397
398        // Check the data in code table here to see if this is data not an
399        // instruction to be disassembled.
400        DiceTable Dice;
401        Dice.push_back(std::make_pair(SectAddress + Index, DiceRef()));
402        dice_table_iterator DTI = std::search(Dices.begin(), Dices.end(),
403                                              Dice.begin(), Dice.end(),
404                                              compareDiceTableEntries);
405        if (DTI != Dices.end()){
406          uint16_t Length;
407          DTI->second.getLength(Length);
408          DumpBytes(StringRef(Bytes.data() + Index, Length));
409          uint16_t Kind;
410          DTI->second.getKind(Kind);
411          DumpDataInCode(Bytes.data() + Index, Length, Kind);
412          continue;
413        }
414
415        if (DisAsm->getInstruction(Inst, Size, memoryObject, Index,
416                                   DebugOut, nulls())) {
417          DumpBytes(StringRef(Bytes.data() + Index, Size));
418          IP->printInst(&Inst, outs(), "");
419
420          // Print debug info.
421          if (diContext) {
422            DILineInfo dli =
423              diContext->getLineInfoForAddress(SectAddress + Index);
424            // Print valid line info if it changed.
425            if (dli != lastLine && dli.Line != 0)
426              outs() << "\t## " << dli.FileName << ':' << dli.Line << ':'
427                     << dli.Column;
428            lastLine = dli;
429          }
430          outs() << "\n";
431        } else {
432          errs() << "llvm-objdump: warning: invalid instruction encoding\n";
433          if (Size == 0)
434            Size = 1; // skip illegible bytes
435        }
436      }
437    }
438    if (!symbolTableWorked) {
439      // Reading the symbol table didn't work, disassemble the whole section.
440      uint64_t SectAddress;
441      Sections[SectIdx].getAddress(SectAddress);
442      uint64_t SectSize;
443      Sections[SectIdx].getSize(SectSize);
444      uint64_t InstSize;
445      for (uint64_t Index = 0; Index < SectSize; Index += InstSize) {
446        MCInst Inst;
447
448        if (DisAsm->getInstruction(Inst, InstSize, memoryObject, Index,
449                                   DebugOut, nulls())) {
450          outs() << format("%8" PRIx64 ":\t", SectAddress + Index);
451          DumpBytes(StringRef(Bytes.data() + Index, InstSize));
452          IP->printInst(&Inst, outs(), "");
453          outs() << "\n";
454        } else {
455          errs() << "llvm-objdump: warning: invalid instruction encoding\n";
456          if (InstSize == 0)
457            InstSize = 1; // skip illegible bytes
458        }
459      }
460    }
461  }
462}
463