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