lib/BinaryFormat/Magic.cpp

   1 //===- llvm/BinaryFormat/Magic.cpp - File magic identification --*- 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 #include "llvm/BinaryFormat/Magic.h"
  11
  12 #include "llvm/BinaryFormat/COFF.h"
  13 #include "llvm/BinaryFormat/ELF.h"
  14 #include "llvm/BinaryFormat/MachO.h"
  15 #include "llvm/Support/Endian.h"
  16 #include "llvm/Support/FileSystem.h"
  17 #include "llvm/Support/MemoryBuffer.h"
  18
  19 #if !defined(_MSC_VER) && !defined(__MINGW32__)
  20 #include <unistd.h>
  21 #else
  22 #include <io.h>
  23 #endif
  24
  25 using namespace llvm;
  26 using namespace llvm::support::endian;
  27 using namespace llvm::sys::fs;
  28
  29 template <size_t N>
  30 static bool startswith(StringRef Magic, const char (&S)[N]) {
  31   return Magic.startswith(StringRef(S, N - 1));
  32 }
  33
  34 /// Identify the magic in magic.
  35 file_magic llvm::identify_magic(StringRef Magic) {
  36   if (Magic.size() < 4)
  37     return file_magic::unknown;
  38   switch ((unsigned char)Magic[0]) {
  39   case 0x00: {
  40     // COFF bigobj, CL.exe's LTO object file, or short import library file
  41     if (startswith(Magic, "\0\0\xFF\xFF")) {
  42       size_t MinSize =
  43           offsetof(COFF::BigObjHeader, UUID) + sizeof(COFF::BigObjMagic);
  44       if (Magic.size() < MinSize)
  45         return file_magic::coff_import_library;
  46
  47       const char *Start = Magic.data() + offsetof(COFF::BigObjHeader, UUID);
  48       if (memcmp(Start, COFF::BigObjMagic, sizeof(COFF::BigObjMagic)) == 0)
  49         return file_magic::coff_object;
  50       if (memcmp(Start, COFF::ClGlObjMagic, sizeof(COFF::BigObjMagic)) == 0)
  51         return file_magic::coff_cl_gl_object;
  52       return file_magic::coff_import_library;
  53     }
  54     // Windows resource file
  55     if (Magic.size() >= sizeof(COFF::WinResMagic) &&
  56         memcmp(Magic.data(), COFF::WinResMagic, sizeof(COFF::WinResMagic)) == 0)
  57       return file_magic::windows_resource;
  58     // 0x0000 = COFF unknown machine type
  59     if (Magic[1] == 0)
  60       return file_magic::coff_object;
  61     if (startswith(Magic, "\0asm"))
  62       return file_magic::wasm_object;
  63     break;
  64   }
  65   case 0xDE: // 0x0B17C0DE = BC wraper
  66     if (startswith(Magic, "\xDE\xC0\x17\x0B"))
  67       return file_magic::bitcode;
  68     break;
  69   case 'B':
  70     if (startswith(Magic, "BC\xC0\xDE"))
  71       return file_magic::bitcode;
  72     break;
  73   case '!':
  74     if (startswith(Magic, "!<arch>\n") || startswith(Magic, "!<thin>\n"))
  75       return file_magic::archive;
  76     break;
  77
  78   case '\177':
  79     if (startswith(Magic, "\177ELF") && Magic.size() >= 18) {
  80       bool Data2MSB = Magic[5] == 2;
  81       unsigned high = Data2MSB ? 16 : 17;
  82       unsigned low = Data2MSB ? 17 : 16;
  83       if (Magic[high] == 0) {
  84         switch (Magic[low]) {
  85         default:
  86           return file_magic::elf;
  87         case 1:
  88           return file_magic::elf_relocatable;
  89         case 2:
  90           return file_magic::elf_executable;
  91         case 3:
  92           return file_magic::elf_shared_object;
  93         case 4:
  94           return file_magic::elf_core;
  95         }
  96       }
  97       // It's still some type of ELF file.
  98       return file_magic::elf;
  99     }
 100     break;
 101
 102   case 0xCA:
 103     if (startswith(Magic, "\xCA\xFE\xBA\xBE") ||
 104         startswith(Magic, "\xCA\xFE\xBA\xBF")) {
 105       // This is complicated by an overlap with Java class files.
 106       // See the Mach-O section in /usr/share/file/magic for details.
 107       if (Magic.size() >= 8 && Magic[7] < 43)
 108         return file_magic::macho_universal_binary;
 109     }
 110     break;
 111
 112   // The two magic numbers for mach-o are:
 113   // 0xfeedface - 32-bit mach-o
 114   // 0xfeedfacf - 64-bit mach-o
 115   case 0xFE:
 116   case 0xCE:
 117   case 0xCF: {
 118     uint16_t type = 0;
 119     if (startswith(Magic, "\xFE\xED\xFA\xCE") ||
 120         startswith(Magic, "\xFE\xED\xFA\xCF")) {
 121       /* Native endian */
 122       size_t MinSize;
 123       if (Magic[3] == char(0xCE))
 124         MinSize = sizeof(MachO::mach_header);
 125       else
 126         MinSize = sizeof(MachO::mach_header_64);
 127       if (Magic.size() >= MinSize)
 128         type = Magic[12] << 24 | Magic[13] << 12 | Magic[14] << 8 | Magic[15];
 129     } else if (startswith(Magic, "\xCE\xFA\xED\xFE") ||
 130                startswith(Magic, "\xCF\xFA\xED\xFE")) {
 131       /* Reverse endian */
 132       size_t MinSize;
 133       if (Magic[0] == char(0xCE))
 134         MinSize = sizeof(MachO::mach_header);
 135       else
 136         MinSize = sizeof(MachO::mach_header_64);
 137       if (Magic.size() >= MinSize)
 138         type = Magic[15] << 24 | Magic[14] << 12 | Magic[13] << 8 | Magic[12];
 139     }
 140     switch (type) {
 141     default:
 142       break;
 143     case 1:
 144       return file_magic::macho_object;
 145     case 2:
 146       return file_magic::macho_executable;
 147     case 3:
 148       return file_magic::macho_fixed_virtual_memory_shared_lib;
 149     case 4:
 150       return file_magic::macho_core;
 151     case 5:
 152       return file_magic::macho_preload_executable;
 153     case 6:
 154       return file_magic::macho_dynamically_linked_shared_lib;
 155     case 7:
 156       return file_magic::macho_dynamic_linker;
 157     case 8:
 158       return file_magic::macho_bundle;
 159     case 9:
 160       return file_magic::macho_dynamically_linked_shared_lib_stub;
 161     case 10:
 162       return file_magic::macho_dsym_companion;
 163     case 11:
 164       return file_magic::macho_kext_bundle;
 165     }
 166     break;
 167   }
 168   case 0xF0: // PowerPC Windows
 169   case 0x83: // Alpha 32-bit
 170   case 0x84: // Alpha 64-bit
 171   case 0x66: // MPS R4000 Windows
 172   case 0x50: // mc68K
 173   case 0x4c: // 80386 Windows
 174   case 0xc4: // ARMNT Windows
 175     if (Magic[1] == 0x01)
 176       return file_magic::coff_object;
 177     LLVM_FALLTHROUGH;
 178
 179   case 0x90: // PA-RISC Windows
 180   case 0x68: // mc68K Windows
 181     if (Magic[1] == 0x02)
 182       return file_magic::coff_object;
 183     break;
 184
 185   case 'M': // Possible MS-DOS stub on Windows PE file or MSF/PDB file.
 186     if (startswith(Magic, "MZ") && Magic.size() >= 0x3c + 4) {
 187       uint32_t off = read32le(Magic.data() + 0x3c);
 188       // PE/COFF file, either EXE or DLL.
 189       if (Magic.substr(off).startswith(
 190               StringRef(COFF::PEMagic, sizeof(COFF::PEMagic))))
 191         return file_magic::pecoff_executable;
 192     }
 193     if (Magic.startswith("Microsoft C/C++ MSF 7.00\r\n"))
 194       return file_magic::pdb;
 195     break;
 196
 197   case 0x64: // x86-64 or ARM64 Windows.
 198     if (Magic[1] == char(0x86) || Magic[1] == char(0xaa))
 199       return file_magic::coff_object;
 200     break;
 201
 202   default:
 203     break;
 204   }
 205   return file_magic::unknown;
 206 }
 207
 208 std::error_code llvm::identify_magic(const Twine &Path, file_magic &Result) {
 209   auto FileOrError = MemoryBuffer::getFile(Path);
 210   if (!FileOrError)
 211     return FileOrError.getError();
 212
 213   std::unique_ptr<MemoryBuffer> FileBuffer = std::move(*FileOrError);
 214   Result = identify_magic(FileBuffer->getBuffer());
 215
 216   return std::error_code();
 217 }