lib/Support/TarWriter.cpp

   1 //===-- TarWriter.cpp - Tar archive file creator --------------------------===//
   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 // TarWriter class provides a feature to create a tar archive file.
  11 //
  12 // I put emphasis on simplicity over comprehensiveness when implementing this
  13 // class because we don't need a full-fledged archive file generator in LLVM
  14 // at the moment.
  15 //
  16 // The filename field in the Unix V7 tar header is 100 bytes. Longer filenames
  17 // are stored using the PAX extension. The PAX header is standardized in
  18 // POSIX.1-2001.
  19 //
  20 // The struct definition of UstarHeader is copied from
  21 // https://www.freebsd.org/cgi/man.cgi?query=tar&sektion=5
  22 //
  23 //===----------------------------------------------------------------------===//
  24
  25 #include "llvm/Support/TarWriter.h"
  26 #include "llvm/ADT/StringRef.h"
  27 #include "llvm/Support/FileSystem.h"
  28 #include "llvm/Support/MathExtras.h"
  29 #include "llvm/Support/Path.h"
  30
  31 using namespace llvm;
  32
  33 // Each file in an archive must be aligned to this block size.
  34 static const int BlockSize = 512;
  35
  36 struct UstarHeader {
  37   char Name[100];
  38   char Mode[8];
  39   char Uid[8];
  40   char Gid[8];
  41   char Size[12];
  42   char Mtime[12];
  43   char Checksum[8];
  44   char TypeFlag;
  45   char Linkname[100];
  46   char Magic[6];
  47   char Version[2];
  48   char Uname[32];
  49   char Gname[32];
  50   char DevMajor[8];
  51   char DevMinor[8];
  52   char Prefix[155];
  53   char Pad[12];
  54 };
  55 static_assert(sizeof(UstarHeader) == BlockSize, "invalid Ustar header");
  56
  57 static UstarHeader makeUstarHeader() {
  58   UstarHeader Hdr = {};
  59   memcpy(Hdr.Magic, "ustar", 5); // Ustar magic
  60   memcpy(Hdr.Version, "00", 2);  // Ustar version
  61   return Hdr;
  62 }
  63
  64 // A PAX attribute is in the form of "<length> <key>=<value>\n"
  65 // where <length> is the length of the entire string including
  66 // the length field itself. An example string is this.
  67 //
  68 //   25 ctime=1084839148.1212\n
  69 //
  70 // This function create such string.
  71 static std::string formatPax(StringRef Key, StringRef Val) {
  72   int Len = Key.size() + Val.size() + 3; // +3 for " ", "=" and "\n"
  73
  74   // We need to compute total size twice because appending
  75   // a length field could change total size by one.
  76   int Total = Len + Twine(Len).str().size();
  77   Total = Len + Twine(Total).str().size();
  78   return (Twine(Total) + " " + Key + "=" + Val + "\n").str();
  79 }
  80
  81 // Headers in tar files must be aligned to 512 byte boundaries.
  82 // This function forwards the current file position to the next boundary.
  83 static void pad(raw_fd_ostream &OS) {
  84   uint64_t Pos = OS.tell();
  85   OS.seek(alignTo(Pos, BlockSize));
  86 }
  87
  88 // Computes a checksum for a tar header.
  89 static void computeChecksum(UstarHeader &Hdr) {
  90   // Before computing a checksum, checksum field must be
  91   // filled with space characters.
  92   memset(Hdr.Checksum, ' ', sizeof(Hdr.Checksum));
  93
  94   // Compute a checksum and set it to the checksum field.
  95   unsigned Chksum = 0;
  96   for (size_t I = 0; I < sizeof(Hdr); ++I)
  97     Chksum += reinterpret_cast<uint8_t *>(&Hdr)[I];
  98   snprintf(Hdr.Checksum, sizeof(Hdr.Checksum), "%06o", Chksum);
  99 }
 100
 101 // Create a tar header and write it to a given output stream.
 102 static void writePaxHeader(raw_fd_ostream &OS, StringRef Path) {
 103   // A PAX header consists of a 512-byte header followed
 104   // by key-value strings. First, create key-value strings.
 105   std::string PaxAttr = formatPax("path", Path);
 106
 107   // Create a 512-byte header.
 108   UstarHeader Hdr = makeUstarHeader();
 109   snprintf(Hdr.Size, sizeof(Hdr.Size), "%011zo", PaxAttr.size());
 110   Hdr.TypeFlag = 'x'; // PAX magic
 111   computeChecksum(Hdr);
 112
 113   // Write them down.
 114   OS << StringRef(reinterpret_cast<char *>(&Hdr), sizeof(Hdr));
 115   OS << PaxAttr;
 116   pad(OS);
 117 }
 118
 119 // Path fits in a Ustar header if
 120 //
 121 // - Path is less than 100 characters long, or
 122 // - Path is in the form of "<prefix>/<name>" where <prefix> is less
 123 //   than or equal to 155 characters long and <name> is less than 100
 124 //   characters long. Both <prefix> and <name> can contain extra '/'.
 125 //
 126 // If Path fits in a Ustar header, updates Prefix and Name and returns true.
 127 // Otherwise, returns false.
 128 static bool splitUstar(StringRef Path, StringRef &Prefix, StringRef &Name) {
 129   if (Path.size() < sizeof(UstarHeader::Name)) {
 130     Prefix = "";
 131     Name = Path;
 132     return true;
 133   }
 134
 135   size_t Sep = Path.rfind('/', sizeof(UstarHeader::Prefix) + 1);
 136   if (Sep == StringRef::npos)
 137     return false;
 138   if (Path.size() - Sep - 1 >= sizeof(UstarHeader::Name))
 139     return false;
 140
 141   Prefix = Path.substr(0, Sep);
 142   Name = Path.substr(Sep + 1);
 143   return true;
 144 }
 145
 146 // The PAX header is an extended format, so a PAX header needs
 147 // to be followed by a "real" header.
 148 static void writeUstarHeader(raw_fd_ostream &OS, StringRef Prefix,
 149                              StringRef Name, size_t Size) {
 150   UstarHeader Hdr = makeUstarHeader();
 151   memcpy(Hdr.Name, Name.data(), Name.size());
 152   memcpy(Hdr.Mode, "0000664", 8);
 153   snprintf(Hdr.Size, sizeof(Hdr.Size), "%011zo", Size);
 154   memcpy(Hdr.Prefix, Prefix.data(), Prefix.size());
 155   computeChecksum(Hdr);
 156   OS << StringRef(reinterpret_cast<char *>(&Hdr), sizeof(Hdr));
 157 }
 158
 159 // Creates a TarWriter instance and returns it.
 160 Expected<std::unique_ptr<TarWriter>> TarWriter::create(StringRef OutputPath,
 161                                                        StringRef BaseDir) {
 162   using namespace sys::fs;
 163   int FD;
 164   if (std::error_code EC =
 165           openFileForWrite(OutputPath, FD, CD_CreateAlways, OF_None))
 166     return make_error<StringError>("cannot open " + OutputPath, EC);
 167   return std::unique_ptr<TarWriter>(new TarWriter(FD, BaseDir));
 168 }
 169
 170 TarWriter::TarWriter(int FD, StringRef BaseDir)
 171     : OS(FD, /*shouldClose=*/true, /*unbuffered=*/false), BaseDir(BaseDir) {}
 172
 173 // Append a given file to an archive.
 174 void TarWriter::append(StringRef Path, StringRef Data) {
 175   // Write Path and Data.
 176   std::string Fullpath = BaseDir + "/" + sys::path::convert_to_slash(Path);
 177
 178   // We do not want to include the same file more than once.
 179   if (!Files.insert(Fullpath).second)
 180     return;
 181
 182   StringRef Prefix;
 183   StringRef Name;
 184   if (splitUstar(Fullpath, Prefix, Name)) {
 185     writeUstarHeader(OS, Prefix, Name, Data.size());
 186   } else {
 187     writePaxHeader(OS, Fullpath);
 188     writeUstarHeader(OS, "", "", Data.size());
 189   }
 190
 191   OS << Data;
 192   pad(OS);
 193
 194   // POSIX requires tar archives end with two null blocks.
 195   // Here, we write the terminator and then seek back, so that
 196   // the file being output is terminated correctly at any moment.
 197   uint64_t Pos = OS.tell();
 198   OS << std::string(BlockSize * 2, '\0');
 199   OS.seek(Pos);
 200   OS.flush();
 201 }