1//===-- COFFDump.cpp - COFF-specific dumper ---------------------*- 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/// \file
11/// \brief This file implements the COFF-specific dumper for llvm-objdump.
12/// It outputs the Win64 EH data structures as plain text.
13/// The encoding of the unwind codes is described in MSDN:
14/// http://msdn.microsoft.com/en-us/library/ck9asaa9.aspx
15///
16//===----------------------------------------------------------------------===//
17
18#include "llvm-objdump.h"
19#include "llvm/Object/COFF.h"
20#include "llvm/Object/ObjectFile.h"
21#include "llvm/Support/Format.h"
22#include "llvm/Support/SourceMgr.h"
23#include "llvm/Support/Win64EH.h"
24#include "llvm/Support/raw_ostream.h"
25#include <algorithm>
26#include <cstring>
27#include <system_error>
28
29using namespace llvm;
30using namespace object;
31using namespace llvm::Win64EH;
32
33// Returns the name of the unwind code.
34static StringRef getUnwindCodeTypeName(uint8_t Code) {
35  switch(Code) {
36  default: llvm_unreachable("Invalid unwind code");
37  case UOP_PushNonVol: return "UOP_PushNonVol";
38  case UOP_AllocLarge: return "UOP_AllocLarge";
39  case UOP_AllocSmall: return "UOP_AllocSmall";
40  case UOP_SetFPReg: return "UOP_SetFPReg";
41  case UOP_SaveNonVol: return "UOP_SaveNonVol";
42  case UOP_SaveNonVolBig: return "UOP_SaveNonVolBig";
43  case UOP_SaveXMM128: return "UOP_SaveXMM128";
44  case UOP_SaveXMM128Big: return "UOP_SaveXMM128Big";
45  case UOP_PushMachFrame: return "UOP_PushMachFrame";
46  }
47}
48
49// Returns the name of a referenced register.
50static StringRef getUnwindRegisterName(uint8_t Reg) {
51  switch(Reg) {
52  default: llvm_unreachable("Invalid register");
53  case 0: return "RAX";
54  case 1: return "RCX";
55  case 2: return "RDX";
56  case 3: return "RBX";
57  case 4: return "RSP";
58  case 5: return "RBP";
59  case 6: return "RSI";
60  case 7: return "RDI";
61  case 8: return "R8";
62  case 9: return "R9";
63  case 10: return "R10";
64  case 11: return "R11";
65  case 12: return "R12";
66  case 13: return "R13";
67  case 14: return "R14";
68  case 15: return "R15";
69  }
70}
71
72// Calculates the number of array slots required for the unwind code.
73static unsigned getNumUsedSlots(const UnwindCode &UnwindCode) {
74  switch (UnwindCode.getUnwindOp()) {
75  default: llvm_unreachable("Invalid unwind code");
76  case UOP_PushNonVol:
77  case UOP_AllocSmall:
78  case UOP_SetFPReg:
79  case UOP_PushMachFrame:
80    return 1;
81  case UOP_SaveNonVol:
82  case UOP_SaveXMM128:
83    return 2;
84  case UOP_SaveNonVolBig:
85  case UOP_SaveXMM128Big:
86    return 3;
87  case UOP_AllocLarge:
88    return (UnwindCode.getOpInfo() == 0) ? 2 : 3;
89  }
90}
91
92// Prints one unwind code. Because an unwind code can occupy up to 3 slots in
93// the unwind codes array, this function requires that the correct number of
94// slots is provided.
95static void printUnwindCode(ArrayRef<UnwindCode> UCs) {
96  assert(UCs.size() >= getNumUsedSlots(UCs[0]));
97  outs() <<  format("      0x%02x: ", unsigned(UCs[0].u.CodeOffset))
98         << getUnwindCodeTypeName(UCs[0].getUnwindOp());
99  switch (UCs[0].getUnwindOp()) {
100  case UOP_PushNonVol:
101    outs() << " " << getUnwindRegisterName(UCs[0].getOpInfo());
102    break;
103  case UOP_AllocLarge:
104    if (UCs[0].getOpInfo() == 0) {
105      outs() << " " << UCs[1].FrameOffset;
106    } else {
107      outs() << " " << UCs[1].FrameOffset
108                       + (static_cast<uint32_t>(UCs[2].FrameOffset) << 16);
109    }
110    break;
111  case UOP_AllocSmall:
112    outs() << " " << ((UCs[0].getOpInfo() + 1) * 8);
113    break;
114  case UOP_SetFPReg:
115    outs() << " ";
116    break;
117  case UOP_SaveNonVol:
118    outs() << " " << getUnwindRegisterName(UCs[0].getOpInfo())
119           << format(" [0x%04x]", 8 * UCs[1].FrameOffset);
120    break;
121  case UOP_SaveNonVolBig:
122    outs() << " " << getUnwindRegisterName(UCs[0].getOpInfo())
123           << format(" [0x%08x]", UCs[1].FrameOffset
124                    + (static_cast<uint32_t>(UCs[2].FrameOffset) << 16));
125    break;
126  case UOP_SaveXMM128:
127    outs() << " XMM" << static_cast<uint32_t>(UCs[0].getOpInfo())
128           << format(" [0x%04x]", 16 * UCs[1].FrameOffset);
129    break;
130  case UOP_SaveXMM128Big:
131    outs() << " XMM" << UCs[0].getOpInfo()
132           << format(" [0x%08x]", UCs[1].FrameOffset
133                           + (static_cast<uint32_t>(UCs[2].FrameOffset) << 16));
134    break;
135  case UOP_PushMachFrame:
136    outs() << " " << (UCs[0].getOpInfo() ? "w/o" : "w")
137           << " error code";
138    break;
139  }
140  outs() << "\n";
141}
142
143static void printAllUnwindCodes(ArrayRef<UnwindCode> UCs) {
144  for (const UnwindCode *I = UCs.begin(), *E = UCs.end(); I < E; ) {
145    unsigned UsedSlots = getNumUsedSlots(*I);
146    if (UsedSlots > UCs.size()) {
147      outs() << "Unwind data corrupted: Encountered unwind op "
148             << getUnwindCodeTypeName((*I).getUnwindOp())
149             << " which requires " << UsedSlots
150             << " slots, but only " << UCs.size()
151             << " remaining in buffer";
152      return ;
153    }
154    printUnwindCode(ArrayRef<UnwindCode>(I, E));
155    I += UsedSlots;
156  }
157}
158
159// Given a symbol sym this functions returns the address and section of it.
160static std::error_code
161resolveSectionAndAddress(const COFFObjectFile *Obj, const SymbolRef &Sym,
162                         const coff_section *&ResolvedSection,
163                         uint64_t &ResolvedAddr) {
164  if (std::error_code EC = Sym.getAddress(ResolvedAddr))
165    return EC;
166  section_iterator iter(Obj->section_begin());
167  if (std::error_code EC = Sym.getSection(iter))
168    return EC;
169  ResolvedSection = Obj->getCOFFSection(*iter);
170  return object_error::success;
171}
172
173// Given a vector of relocations for a section and an offset into this section
174// the function returns the symbol used for the relocation at the offset.
175static std::error_code resolveSymbol(const std::vector<RelocationRef> &Rels,
176                                     uint64_t Offset, SymbolRef &Sym) {
177  for (std::vector<RelocationRef>::const_iterator I = Rels.begin(),
178                                                  E = Rels.end();
179                                                  I != E; ++I) {
180    uint64_t Ofs;
181    if (std::error_code EC = I->getOffset(Ofs))
182      return EC;
183    if (Ofs == Offset) {
184      Sym = *I->getSymbol();
185      return object_error::success;
186    }
187  }
188  return object_error::parse_failed;
189}
190
191// Given a vector of relocations for a section and an offset into this section
192// the function resolves the symbol used for the relocation at the offset and
193// returns the section content and the address inside the content pointed to
194// by the symbol.
195static std::error_code
196getSectionContents(const COFFObjectFile *Obj,
197                   const std::vector<RelocationRef> &Rels, uint64_t Offset,
198                   ArrayRef<uint8_t> &Contents, uint64_t &Addr) {
199  SymbolRef Sym;
200  if (std::error_code EC = resolveSymbol(Rels, Offset, Sym))
201    return EC;
202  const coff_section *Section;
203  if (std::error_code EC = resolveSectionAndAddress(Obj, Sym, Section, Addr))
204    return EC;
205  if (std::error_code EC = Obj->getSectionContents(Section, Contents))
206    return EC;
207  return object_error::success;
208}
209
210// Given a vector of relocations for a section and an offset into this section
211// the function returns the name of the symbol used for the relocation at the
212// offset.
213static std::error_code resolveSymbolName(const std::vector<RelocationRef> &Rels,
214                                         uint64_t Offset, StringRef &Name) {
215  SymbolRef Sym;
216  if (std::error_code EC = resolveSymbol(Rels, Offset, Sym))
217    return EC;
218  if (std::error_code EC = Sym.getName(Name))
219    return EC;
220  return object_error::success;
221}
222
223static void printCOFFSymbolAddress(llvm::raw_ostream &Out,
224                                   const std::vector<RelocationRef> &Rels,
225                                   uint64_t Offset, uint32_t Disp) {
226  StringRef Sym;
227  if (!resolveSymbolName(Rels, Offset, Sym)) {
228    Out << Sym;
229    if (Disp > 0)
230      Out << format(" + 0x%04x", Disp);
231  } else {
232    Out << format("0x%04x", Disp);
233  }
234}
235
236static void
237printSEHTable(const COFFObjectFile *Obj, uint32_t TableVA, int Count) {
238  if (Count == 0)
239    return;
240
241  const pe32_header *PE32Header;
242  if (error(Obj->getPE32Header(PE32Header)))
243    return;
244  uint32_t ImageBase = PE32Header->ImageBase;
245  uintptr_t IntPtr = 0;
246  if (error(Obj->getVaPtr(TableVA, IntPtr)))
247    return;
248  const support::ulittle32_t *P = (const support::ulittle32_t *)IntPtr;
249  outs() << "SEH Table:";
250  for (int I = 0; I < Count; ++I)
251    outs() << format(" 0x%x", P[I] + ImageBase);
252  outs() << "\n\n";
253}
254
255static void printLoadConfiguration(const COFFObjectFile *Obj) {
256  // Skip if it's not executable.
257  const pe32_header *PE32Header;
258  if (error(Obj->getPE32Header(PE32Header)))
259    return;
260  if (!PE32Header)
261    return;
262
263  const coff_file_header *Header;
264  if (error(Obj->getCOFFHeader(Header)))
265    return;
266  // Currently only x86 is supported
267  if (Header->Machine != COFF::IMAGE_FILE_MACHINE_I386)
268    return;
269
270  const data_directory *DataDir;
271  if (error(Obj->getDataDirectory(COFF::LOAD_CONFIG_TABLE, DataDir)))
272    return;
273  uintptr_t IntPtr = 0;
274  if (DataDir->RelativeVirtualAddress == 0)
275    return;
276  if (error(Obj->getRvaPtr(DataDir->RelativeVirtualAddress, IntPtr)))
277    return;
278
279  auto *LoadConf = reinterpret_cast<const coff_load_configuration32 *>(IntPtr);
280  outs() << "Load configuration:"
281         << "\n  Timestamp: " << LoadConf->TimeDateStamp
282         << "\n  Major Version: " << LoadConf->MajorVersion
283         << "\n  Minor Version: " << LoadConf->MinorVersion
284         << "\n  GlobalFlags Clear: " << LoadConf->GlobalFlagsClear
285         << "\n  GlobalFlags Set: " << LoadConf->GlobalFlagsSet
286         << "\n  Critical Section Default Timeout: " << LoadConf->CriticalSectionDefaultTimeout
287         << "\n  Decommit Free Block Threshold: " << LoadConf->DeCommitFreeBlockThreshold
288         << "\n  Decommit Total Free Threshold: " << LoadConf->DeCommitTotalFreeThreshold
289         << "\n  Lock Prefix Table: " << LoadConf->LockPrefixTable
290         << "\n  Maximum Allocation Size: " << LoadConf->MaximumAllocationSize
291         << "\n  Virtual Memory Threshold: " << LoadConf->VirtualMemoryThreshold
292         << "\n  Process Affinity Mask: " << LoadConf->ProcessAffinityMask
293         << "\n  Process Heap Flags: " << LoadConf->ProcessHeapFlags
294         << "\n  CSD Version: " << LoadConf->CSDVersion
295         << "\n  Security Cookie: " << LoadConf->SecurityCookie
296         << "\n  SEH Table: " << LoadConf->SEHandlerTable
297         << "\n  SEH Count: " << LoadConf->SEHandlerCount
298         << "\n\n";
299  printSEHTable(Obj, LoadConf->SEHandlerTable, LoadConf->SEHandlerCount);
300  outs() << "\n";
301}
302
303// Prints import tables. The import table is a table containing the list of
304// DLL name and symbol names which will be linked by the loader.
305static void printImportTables(const COFFObjectFile *Obj) {
306  import_directory_iterator I = Obj->import_directory_begin();
307  import_directory_iterator E = Obj->import_directory_end();
308  if (I == E)
309    return;
310  outs() << "The Import Tables:\n";
311  for (; I != E; I = ++I) {
312    const import_directory_table_entry *Dir;
313    StringRef Name;
314    if (I->getImportTableEntry(Dir)) return;
315    if (I->getName(Name)) return;
316
317    outs() << format("  lookup %08x time %08x fwd %08x name %08x addr %08x\n\n",
318                     static_cast<uint32_t>(Dir->ImportLookupTableRVA),
319                     static_cast<uint32_t>(Dir->TimeDateStamp),
320                     static_cast<uint32_t>(Dir->ForwarderChain),
321                     static_cast<uint32_t>(Dir->NameRVA),
322                     static_cast<uint32_t>(Dir->ImportAddressTableRVA));
323    outs() << "    DLL Name: " << Name << "\n";
324    outs() << "    Hint/Ord  Name\n";
325    const import_lookup_table_entry32 *entry;
326    if (I->getImportLookupEntry(entry))
327      return;
328    for (; entry->data; ++entry) {
329      if (entry->isOrdinal()) {
330        outs() << format("      % 6d\n", entry->getOrdinal());
331        continue;
332      }
333      uint16_t Hint;
334      StringRef Name;
335      if (Obj->getHintName(entry->getHintNameRVA(), Hint, Name))
336        return;
337      outs() << format("      % 6d  ", Hint) << Name << "\n";
338    }
339    outs() << "\n";
340  }
341}
342
343// Prints export tables. The export table is a table containing the list of
344// exported symbol from the DLL.
345static void printExportTable(const COFFObjectFile *Obj) {
346  outs() << "Export Table:\n";
347  export_directory_iterator I = Obj->export_directory_begin();
348  export_directory_iterator E = Obj->export_directory_end();
349  if (I == E)
350    return;
351  StringRef DllName;
352  uint32_t OrdinalBase;
353  if (I->getDllName(DllName))
354    return;
355  if (I->getOrdinalBase(OrdinalBase))
356    return;
357  outs() << " DLL name: " << DllName << "\n";
358  outs() << " Ordinal base: " << OrdinalBase << "\n";
359  outs() << " Ordinal      RVA  Name\n";
360  for (; I != E; I = ++I) {
361    uint32_t Ordinal;
362    if (I->getOrdinal(Ordinal))
363      return;
364    uint32_t RVA;
365    if (I->getExportRVA(RVA))
366      return;
367    outs() << format("    % 4d %# 8x", Ordinal, RVA);
368
369    StringRef Name;
370    if (I->getSymbolName(Name))
371      continue;
372    if (!Name.empty())
373      outs() << "  " << Name;
374    outs() << "\n";
375  }
376}
377
378// Given the COFF object file, this function returns the relocations for .pdata
379// and the pointer to "runtime function" structs.
380static bool getPDataSection(const COFFObjectFile *Obj,
381                            std::vector<RelocationRef> &Rels,
382                            const RuntimeFunction *&RFStart, int &NumRFs) {
383  for (const SectionRef &Section : Obj->sections()) {
384    StringRef Name;
385    if (error(Section.getName(Name)))
386      continue;
387    if (Name != ".pdata")
388      continue;
389
390    const coff_section *Pdata = Obj->getCOFFSection(Section);
391    for (const RelocationRef &Reloc : Section.relocations())
392      Rels.push_back(Reloc);
393
394    // Sort relocations by address.
395    std::sort(Rels.begin(), Rels.end(), RelocAddressLess);
396
397    ArrayRef<uint8_t> Contents;
398    if (error(Obj->getSectionContents(Pdata, Contents)))
399      continue;
400    if (Contents.empty())
401      continue;
402
403    RFStart = reinterpret_cast<const RuntimeFunction *>(Contents.data());
404    NumRFs = Contents.size() / sizeof(RuntimeFunction);
405    return true;
406  }
407  return false;
408}
409
410static void printWin64EHUnwindInfo(const Win64EH::UnwindInfo *UI) {
411  // The casts to int are required in order to output the value as number.
412  // Without the casts the value would be interpreted as char data (which
413  // results in garbage output).
414  outs() << "    Version: " << static_cast<int>(UI->getVersion()) << "\n";
415  outs() << "    Flags: " << static_cast<int>(UI->getFlags());
416  if (UI->getFlags()) {
417    if (UI->getFlags() & UNW_ExceptionHandler)
418      outs() << " UNW_ExceptionHandler";
419    if (UI->getFlags() & UNW_TerminateHandler)
420      outs() << " UNW_TerminateHandler";
421    if (UI->getFlags() & UNW_ChainInfo)
422      outs() << " UNW_ChainInfo";
423  }
424  outs() << "\n";
425  outs() << "    Size of prolog: " << static_cast<int>(UI->PrologSize) << "\n";
426  outs() << "    Number of Codes: " << static_cast<int>(UI->NumCodes) << "\n";
427  // Maybe this should move to output of UOP_SetFPReg?
428  if (UI->getFrameRegister()) {
429    outs() << "    Frame register: "
430           << getUnwindRegisterName(UI->getFrameRegister()) << "\n";
431    outs() << "    Frame offset: " << 16 * UI->getFrameOffset() << "\n";
432  } else {
433    outs() << "    No frame pointer used\n";
434  }
435  if (UI->getFlags() & (UNW_ExceptionHandler | UNW_TerminateHandler)) {
436    // FIXME: Output exception handler data
437  } else if (UI->getFlags() & UNW_ChainInfo) {
438    // FIXME: Output chained unwind info
439  }
440
441  if (UI->NumCodes)
442    outs() << "    Unwind Codes:\n";
443
444  printAllUnwindCodes(ArrayRef<UnwindCode>(&UI->UnwindCodes[0], UI->NumCodes));
445
446  outs() << "\n";
447  outs().flush();
448}
449
450/// Prints out the given RuntimeFunction struct for x64, assuming that Obj is
451/// pointing to an executable file.
452static void printRuntimeFunction(const COFFObjectFile *Obj,
453                                 const RuntimeFunction &RF) {
454  if (!RF.StartAddress)
455    return;
456  outs() << "Function Table:\n"
457         << format("  Start Address: 0x%04x\n",
458                   static_cast<uint32_t>(RF.StartAddress))
459         << format("  End Address: 0x%04x\n",
460                   static_cast<uint32_t>(RF.EndAddress))
461         << format("  Unwind Info Address: 0x%04x\n",
462                   static_cast<uint32_t>(RF.UnwindInfoOffset));
463  uintptr_t addr;
464  if (Obj->getRvaPtr(RF.UnwindInfoOffset, addr))
465    return;
466  printWin64EHUnwindInfo(reinterpret_cast<const Win64EH::UnwindInfo *>(addr));
467}
468
469/// Prints out the given RuntimeFunction struct for x64, assuming that Obj is
470/// pointing to an object file. Unlike executable, fields in RuntimeFunction
471/// struct are filled with zeros, but instead there are relocations pointing to
472/// them so that the linker will fill targets' RVAs to the fields at link
473/// time. This function interprets the relocations to find the data to be used
474/// in the resulting executable.
475static void printRuntimeFunctionRels(const COFFObjectFile *Obj,
476                                     const RuntimeFunction &RF,
477                                     uint64_t SectionOffset,
478                                     const std::vector<RelocationRef> &Rels) {
479  outs() << "Function Table:\n";
480  outs() << "  Start Address: ";
481  printCOFFSymbolAddress(outs(), Rels,
482                         SectionOffset +
483                             /*offsetof(RuntimeFunction, StartAddress)*/ 0,
484                         RF.StartAddress);
485  outs() << "\n";
486
487  outs() << "  End Address: ";
488  printCOFFSymbolAddress(outs(), Rels,
489                         SectionOffset +
490                             /*offsetof(RuntimeFunction, EndAddress)*/ 4,
491                         RF.EndAddress);
492  outs() << "\n";
493
494  outs() << "  Unwind Info Address: ";
495  printCOFFSymbolAddress(outs(), Rels,
496                         SectionOffset +
497                             /*offsetof(RuntimeFunction, UnwindInfoOffset)*/ 8,
498                         RF.UnwindInfoOffset);
499  outs() << "\n";
500
501  ArrayRef<uint8_t> XContents;
502  uint64_t UnwindInfoOffset = 0;
503  if (error(getSectionContents(
504          Obj, Rels, SectionOffset +
505                         /*offsetof(RuntimeFunction, UnwindInfoOffset)*/ 8,
506          XContents, UnwindInfoOffset)))
507    return;
508  if (XContents.empty())
509    return;
510
511  UnwindInfoOffset += RF.UnwindInfoOffset;
512  if (UnwindInfoOffset > XContents.size())
513    return;
514
515  auto *UI = reinterpret_cast<const Win64EH::UnwindInfo *>(XContents.data() +
516                                                           UnwindInfoOffset);
517  printWin64EHUnwindInfo(UI);
518}
519
520void llvm::printCOFFUnwindInfo(const COFFObjectFile *Obj) {
521  const coff_file_header *Header;
522  if (error(Obj->getCOFFHeader(Header)))
523    return;
524
525  if (Header->Machine != COFF::IMAGE_FILE_MACHINE_AMD64) {
526    errs() << "Unsupported image machine type "
527              "(currently only AMD64 is supported).\n";
528    return;
529  }
530
531  std::vector<RelocationRef> Rels;
532  const RuntimeFunction *RFStart;
533  int NumRFs;
534  if (!getPDataSection(Obj, Rels, RFStart, NumRFs))
535    return;
536  ArrayRef<RuntimeFunction> RFs(RFStart, NumRFs);
537
538  bool IsExecutable = Rels.empty();
539  if (IsExecutable) {
540    for (const RuntimeFunction &RF : RFs)
541      printRuntimeFunction(Obj, RF);
542    return;
543  }
544
545  for (const RuntimeFunction &RF : RFs) {
546    uint64_t SectionOffset =
547        std::distance(RFs.begin(), &RF) * sizeof(RuntimeFunction);
548    printRuntimeFunctionRels(Obj, RF, SectionOffset, Rels);
549  }
550}
551
552void llvm::printCOFFFileHeader(const object::ObjectFile *Obj) {
553  const COFFObjectFile *file = dyn_cast<const COFFObjectFile>(Obj);
554  printLoadConfiguration(file);
555  printImportTables(file);
556  printExportTable(file);
557}
558