Home | History | Annotate | Download | only in Support
      1 //===--- MemoryBuffer.cpp - Memory Buffer implementation ------------------===//
      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 MemoryBuffer interface.
     11 //
     12 //===----------------------------------------------------------------------===//
     13 
     14 #include "llvm/Support/MemoryBuffer.h"
     15 #include "llvm/ADT/OwningPtr.h"
     16 #include "llvm/ADT/SmallString.h"
     17 #include "llvm/Config/config.h"
     18 #include "llvm/Support/Errno.h"
     19 #include "llvm/Support/FileSystem.h"
     20 #include "llvm/Support/MathExtras.h"
     21 #include "llvm/Support/Path.h"
     22 #include "llvm/Support/Process.h"
     23 #include "llvm/Support/Program.h"
     24 #include "llvm/Support/system_error.h"
     25 #include <cassert>
     26 #include <cerrno>
     27 #include <cstdio>
     28 #include <cstring>
     29 #include <new>
     30 #include <sys/stat.h>
     31 #include <sys/types.h>
     32 #if !defined(_MSC_VER) && !defined(__MINGW32__)
     33 #include <unistd.h>
     34 #else
     35 #include <io.h>
     36 // Simplistic definitinos of these macros to allow files to be read with
     37 // MapInFilePages.
     38 #ifndef S_ISREG
     39 #define S_ISREG(x) (1)
     40 #endif
     41 #ifndef S_ISBLK
     42 #define S_ISBLK(x) (0)
     43 #endif
     44 #endif
     45 #include <fcntl.h>
     46 using namespace llvm;
     47 
     48 //===----------------------------------------------------------------------===//
     49 // MemoryBuffer implementation itself.
     50 //===----------------------------------------------------------------------===//
     51 
     52 MemoryBuffer::~MemoryBuffer() { }
     53 
     54 /// init - Initialize this MemoryBuffer as a reference to externally allocated
     55 /// memory, memory that we know is already null terminated.
     56 void MemoryBuffer::init(const char *BufStart, const char *BufEnd,
     57                         bool RequiresNullTerminator) {
     58   assert((!RequiresNullTerminator || BufEnd[0] == 0) &&
     59          "Buffer is not null terminated!");
     60   BufferStart = BufStart;
     61   BufferEnd = BufEnd;
     62 }
     63 
     64 //===----------------------------------------------------------------------===//
     65 // MemoryBufferMem implementation.
     66 //===----------------------------------------------------------------------===//
     67 
     68 /// CopyStringRef - Copies contents of a StringRef into a block of memory and
     69 /// null-terminates it.
     70 static void CopyStringRef(char *Memory, StringRef Data) {
     71   memcpy(Memory, Data.data(), Data.size());
     72   Memory[Data.size()] = 0; // Null terminate string.
     73 }
     74 
     75 struct NamedBufferAlloc {
     76   StringRef Name;
     77   NamedBufferAlloc(StringRef Name) : Name(Name) {}
     78 };
     79 
     80 void *operator new(size_t N, const NamedBufferAlloc &Alloc) {
     81   char *Mem = static_cast<char *>(operator new(N + Alloc.Name.size() + 1));
     82   CopyStringRef(Mem + N, Alloc.Name);
     83   return Mem;
     84 }
     85 
     86 namespace {
     87 /// MemoryBufferMem - Named MemoryBuffer pointing to a block of memory.
     88 class MemoryBufferMem : public MemoryBuffer {
     89 public:
     90   MemoryBufferMem(StringRef InputData, bool RequiresNullTerminator) {
     91     init(InputData.begin(), InputData.end(), RequiresNullTerminator);
     92   }
     93 
     94   virtual const char *getBufferIdentifier() const LLVM_OVERRIDE {
     95      // The name is stored after the class itself.
     96     return reinterpret_cast<const char*>(this + 1);
     97   }
     98 
     99   virtual BufferKind getBufferKind() const LLVM_OVERRIDE {
    100     return MemoryBuffer_Malloc;
    101   }
    102 };
    103 }
    104 
    105 /// getMemBuffer - Open the specified memory range as a MemoryBuffer.  Note
    106 /// that InputData must be a null terminated if RequiresNullTerminator is true!
    107 MemoryBuffer *MemoryBuffer::getMemBuffer(StringRef InputData,
    108                                          StringRef BufferName,
    109                                          bool RequiresNullTerminator) {
    110   return new (NamedBufferAlloc(BufferName))
    111       MemoryBufferMem(InputData, RequiresNullTerminator);
    112 }
    113 
    114 /// getMemBufferCopy - Open the specified memory range as a MemoryBuffer,
    115 /// copying the contents and taking ownership of it.  This has no requirements
    116 /// on EndPtr[0].
    117 MemoryBuffer *MemoryBuffer::getMemBufferCopy(StringRef InputData,
    118                                              StringRef BufferName) {
    119   MemoryBuffer *Buf = getNewUninitMemBuffer(InputData.size(), BufferName);
    120   if (!Buf) return 0;
    121   memcpy(const_cast<char*>(Buf->getBufferStart()), InputData.data(),
    122          InputData.size());
    123   return Buf;
    124 }
    125 
    126 /// getNewUninitMemBuffer - Allocate a new MemoryBuffer of the specified size
    127 /// that is not initialized.  Note that the caller should initialize the
    128 /// memory allocated by this method.  The memory is owned by the MemoryBuffer
    129 /// object.
    130 MemoryBuffer *MemoryBuffer::getNewUninitMemBuffer(size_t Size,
    131                                                   StringRef BufferName) {
    132   // Allocate space for the MemoryBuffer, the data and the name. It is important
    133   // that MemoryBuffer and data are aligned so PointerIntPair works with them.
    134   size_t AlignedStringLen =
    135     RoundUpToAlignment(sizeof(MemoryBufferMem) + BufferName.size() + 1,
    136                        sizeof(void*)); // TODO: Is sizeof(void*) enough?
    137   size_t RealLen = AlignedStringLen + Size + 1;
    138   char *Mem = static_cast<char*>(operator new(RealLen, std::nothrow));
    139   if (!Mem) return 0;
    140 
    141   // The name is stored after the class itself.
    142   CopyStringRef(Mem + sizeof(MemoryBufferMem), BufferName);
    143 
    144   // The buffer begins after the name and must be aligned.
    145   char *Buf = Mem + AlignedStringLen;
    146   Buf[Size] = 0; // Null terminate buffer.
    147 
    148   return new (Mem) MemoryBufferMem(StringRef(Buf, Size), true);
    149 }
    150 
    151 /// getNewMemBuffer - Allocate a new MemoryBuffer of the specified size that
    152 /// is completely initialized to zeros.  Note that the caller should
    153 /// initialize the memory allocated by this method.  The memory is owned by
    154 /// the MemoryBuffer object.
    155 MemoryBuffer *MemoryBuffer::getNewMemBuffer(size_t Size, StringRef BufferName) {
    156   MemoryBuffer *SB = getNewUninitMemBuffer(Size, BufferName);
    157   if (!SB) return 0;
    158   memset(const_cast<char*>(SB->getBufferStart()), 0, Size);
    159   return SB;
    160 }
    161 
    162 
    163 /// getFileOrSTDIN - Open the specified file as a MemoryBuffer, or open stdin
    164 /// if the Filename is "-".  If an error occurs, this returns null and fills
    165 /// in *ErrStr with a reason.  If stdin is empty, this API (unlike getSTDIN)
    166 /// returns an empty buffer.
    167 error_code MemoryBuffer::getFileOrSTDIN(StringRef Filename,
    168                                         OwningPtr<MemoryBuffer> &result,
    169                                         int64_t FileSize) {
    170   if (Filename == "-")
    171     return getSTDIN(result);
    172   return getFile(Filename, result, FileSize);
    173 }
    174 
    175 error_code MemoryBuffer::getFileOrSTDIN(const char *Filename,
    176                                         OwningPtr<MemoryBuffer> &result,
    177                                         int64_t FileSize) {
    178   if (strcmp(Filename, "-") == 0)
    179     return getSTDIN(result);
    180   return getFile(Filename, result, FileSize);
    181 }
    182 
    183 //===----------------------------------------------------------------------===//
    184 // MemoryBuffer::getFile implementation.
    185 //===----------------------------------------------------------------------===//
    186 
    187 namespace {
    188 /// \brief Memorry maps a file descriptor using sys::fs::mapped_file_region.
    189 ///
    190 /// This handles converting the offset into a legal offset on the platform.
    191 class MemoryBufferMMapFile : public MemoryBuffer {
    192   sys::fs::mapped_file_region MFR;
    193 
    194   static uint64_t getLegalMapOffset(uint64_t Offset) {
    195     return Offset & ~(sys::fs::mapped_file_region::alignment() - 1);
    196   }
    197 
    198   static uint64_t getLegalMapSize(uint64_t Len, uint64_t Offset) {
    199     return Len + (Offset - getLegalMapOffset(Offset));
    200   }
    201 
    202   const char *getStart(uint64_t Len, uint64_t Offset) {
    203     return MFR.const_data() + (Offset - getLegalMapOffset(Offset));
    204   }
    205 
    206 public:
    207   MemoryBufferMMapFile(bool RequiresNullTerminator, int FD, uint64_t Len,
    208                        uint64_t Offset, error_code EC)
    209       : MFR(FD, false, sys::fs::mapped_file_region::readonly,
    210             getLegalMapSize(Len, Offset), getLegalMapOffset(Offset), EC) {
    211     if (!EC) {
    212       const char *Start = getStart(Len, Offset);
    213       init(Start, Start + Len, RequiresNullTerminator);
    214     }
    215   }
    216 
    217   virtual const char *getBufferIdentifier() const LLVM_OVERRIDE {
    218     // The name is stored after the class itself.
    219     return reinterpret_cast<const char *>(this + 1);
    220   }
    221 
    222   virtual BufferKind getBufferKind() const LLVM_OVERRIDE {
    223     return MemoryBuffer_MMap;
    224   }
    225 };
    226 }
    227 
    228 static error_code getMemoryBufferForStream(int FD,
    229                                            StringRef BufferName,
    230                                            OwningPtr<MemoryBuffer> &result) {
    231   const ssize_t ChunkSize = 4096*4;
    232   SmallString<ChunkSize> Buffer;
    233   ssize_t ReadBytes;
    234   // Read into Buffer until we hit EOF.
    235   do {
    236     Buffer.reserve(Buffer.size() + ChunkSize);
    237     ReadBytes = read(FD, Buffer.end(), ChunkSize);
    238     if (ReadBytes == -1) {
    239       if (errno == EINTR) continue;
    240       return error_code(errno, posix_category());
    241     }
    242     Buffer.set_size(Buffer.size() + ReadBytes);
    243   } while (ReadBytes != 0);
    244 
    245   result.reset(MemoryBuffer::getMemBufferCopy(Buffer, BufferName));
    246   return error_code::success();
    247 }
    248 
    249 error_code MemoryBuffer::getFile(StringRef Filename,
    250                                  OwningPtr<MemoryBuffer> &result,
    251                                  int64_t FileSize,
    252                                  bool RequiresNullTerminator) {
    253   // Ensure the path is null terminated.
    254   SmallString<256> PathBuf(Filename.begin(), Filename.end());
    255   return MemoryBuffer::getFile(PathBuf.c_str(), result, FileSize,
    256                                RequiresNullTerminator);
    257 }
    258 
    259 error_code MemoryBuffer::getFile(const char *Filename,
    260                                  OwningPtr<MemoryBuffer> &result,
    261                                  int64_t FileSize,
    262                                  bool RequiresNullTerminator) {
    263   // FIXME: Review if this check is unnecessary on windows as well.
    264 #ifdef LLVM_ON_WIN32
    265   // First check that the "file" is not a directory
    266   bool is_dir = false;
    267   error_code err = sys::fs::is_directory(Filename, is_dir);
    268   if (err)
    269     return err;
    270   if (is_dir)
    271     return make_error_code(errc::is_a_directory);
    272 #endif
    273 
    274   int OpenFlags = O_RDONLY;
    275 #ifdef O_BINARY
    276   OpenFlags |= O_BINARY;  // Open input file in binary mode on win32.
    277 #endif
    278   int FD = ::open(Filename, OpenFlags);
    279   if (FD == -1)
    280     return error_code(errno, posix_category());
    281 
    282   error_code ret = getOpenFile(FD, Filename, result, FileSize, FileSize,
    283                                0, RequiresNullTerminator);
    284   close(FD);
    285   return ret;
    286 }
    287 
    288 static bool shouldUseMmap(int FD,
    289                           size_t FileSize,
    290                           size_t MapSize,
    291                           off_t Offset,
    292                           bool RequiresNullTerminator,
    293                           int PageSize) {
    294   // We don't use mmap for small files because this can severely fragment our
    295   // address space.
    296   if (MapSize < 4096*4)
    297     return false;
    298 
    299   if (!RequiresNullTerminator)
    300     return true;
    301 
    302 
    303   // If we don't know the file size, use fstat to find out.  fstat on an open
    304   // file descriptor is cheaper than stat on a random path.
    305   // FIXME: this chunk of code is duplicated, but it avoids a fstat when
    306   // RequiresNullTerminator = false and MapSize != -1.
    307   if (FileSize == size_t(-1)) {
    308     struct stat FileInfo;
    309     // TODO: This should use fstat64 when available.
    310     if (fstat(FD, &FileInfo) == -1) {
    311       return error_code(errno, posix_category());
    312     }
    313     FileSize = FileInfo.st_size;
    314   }
    315 
    316   // If we need a null terminator and the end of the map is inside the file,
    317   // we cannot use mmap.
    318   size_t End = Offset + MapSize;
    319   assert(End <= FileSize);
    320   if (End != FileSize)
    321     return false;
    322 
    323   // Don't try to map files that are exactly a multiple of the system page size
    324   // if we need a null terminator.
    325   if ((FileSize & (PageSize -1)) == 0)
    326     return false;
    327 
    328   return true;
    329 }
    330 
    331 error_code MemoryBuffer::getOpenFile(int FD, const char *Filename,
    332                                      OwningPtr<MemoryBuffer> &result,
    333                                      uint64_t FileSize, uint64_t MapSize,
    334                                      int64_t Offset,
    335                                      bool RequiresNullTerminator) {
    336   static int PageSize = sys::process::get_self()->page_size();
    337 
    338   // Default is to map the full file.
    339   if (MapSize == uint64_t(-1)) {
    340     // If we don't know the file size, use fstat to find out.  fstat on an open
    341     // file descriptor is cheaper than stat on a random path.
    342     if (FileSize == uint64_t(-1)) {
    343       struct stat FileInfo;
    344       // TODO: This should use fstat64 when available.
    345       if (fstat(FD, &FileInfo) == -1) {
    346         return error_code(errno, posix_category());
    347       }
    348 
    349       // If this not a file or a block device (e.g. it's a named pipe
    350       // or character device), we can't trust the size. Create the memory
    351       // buffer by copying off the stream.
    352       if (!S_ISREG(FileInfo.st_mode) && !S_ISBLK(FileInfo.st_mode)) {
    353         return getMemoryBufferForStream(FD, Filename, result);
    354       }
    355 
    356       FileSize = FileInfo.st_size;
    357     }
    358     MapSize = FileSize;
    359   }
    360 
    361   if (shouldUseMmap(FD, FileSize, MapSize, Offset, RequiresNullTerminator,
    362                     PageSize)) {
    363     error_code EC;
    364     result.reset(new (NamedBufferAlloc(Filename)) MemoryBufferMMapFile(
    365         RequiresNullTerminator, FD, MapSize, Offset, EC));
    366     if (!EC)
    367       return error_code::success();
    368   }
    369 
    370   MemoryBuffer *Buf = MemoryBuffer::getNewUninitMemBuffer(MapSize, Filename);
    371   if (!Buf) {
    372     // Failed to create a buffer. The only way it can fail is if
    373     // new(std::nothrow) returns 0.
    374     return make_error_code(errc::not_enough_memory);
    375   }
    376 
    377   OwningPtr<MemoryBuffer> SB(Buf);
    378   char *BufPtr = const_cast<char*>(SB->getBufferStart());
    379 
    380   size_t BytesLeft = MapSize;
    381 #ifndef HAVE_PREAD
    382   if (lseek(FD, Offset, SEEK_SET) == -1)
    383     return error_code(errno, posix_category());
    384 #endif
    385 
    386   while (BytesLeft) {
    387 #ifdef HAVE_PREAD
    388     ssize_t NumRead = ::pread(FD, BufPtr, BytesLeft, MapSize-BytesLeft+Offset);
    389 #else
    390     ssize_t NumRead = ::read(FD, BufPtr, BytesLeft);
    391 #endif
    392     if (NumRead == -1) {
    393       if (errno == EINTR)
    394         continue;
    395       // Error while reading.
    396       return error_code(errno, posix_category());
    397     }
    398     if (NumRead == 0) {
    399       assert(0 && "We got inaccurate FileSize value or fstat reported an "
    400                    "invalid file size.");
    401       *BufPtr = '\0'; // null-terminate at the actual size.
    402       break;
    403     }
    404     BytesLeft -= NumRead;
    405     BufPtr += NumRead;
    406   }
    407 
    408   result.swap(SB);
    409   return error_code::success();
    410 }
    411 
    412 //===----------------------------------------------------------------------===//
    413 // MemoryBuffer::getSTDIN implementation.
    414 //===----------------------------------------------------------------------===//
    415 
    416 error_code MemoryBuffer::getSTDIN(OwningPtr<MemoryBuffer> &result) {
    417   // Read in all of the data from stdin, we cannot mmap stdin.
    418   //
    419   // FIXME: That isn't necessarily true, we should try to mmap stdin and
    420   // fallback if it fails.
    421   sys::Program::ChangeStdinToBinary();
    422 
    423   return getMemoryBufferForStream(0, "<stdin>", result);
    424 }
    425