1 //===--- MemoryBuffer.cpp - Memory Buffer implementation ------------------===//
3 // The LLVM Compiler Infrastructure
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
8 //===----------------------------------------------------------------------===//
10 // This file implements the MemoryBuffer interface.
12 //===----------------------------------------------------------------------===//
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"
30 #include <sys/types.h>
32 #if !defined(_MSC_VER) && !defined(__MINGW32__)
40 //===----------------------------------------------------------------------===//
41 // MemoryBuffer implementation itself.
42 //===----------------------------------------------------------------------===//
44 MemoryBuffer::~MemoryBuffer() { }
46 /// init - Initialize this MemoryBuffer as a reference to externally allocated
47 /// memory, memory that we know is already null terminated.
48 void 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;
56 //===----------------------------------------------------------------------===//
57 // MemoryBufferMem implementation.
58 //===----------------------------------------------------------------------===//
60 /// CopyStringRef - Copies contents of a StringRef into a block of memory and
61 /// null-terminates it.
62 static void CopyStringRef(char *Memory, StringRef Data) {
63 memcpy(Memory, Data.data(), Data.size());
64 Memory[Data.size()] = 0; // Null terminate string.
67 /// GetNamedBuffer - Allocates a new MemoryBuffer with Name copied after it.
69 static 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);
77 /// MemoryBufferMem - Named MemoryBuffer pointing to a block of memory.
78 class MemoryBufferMem : public MemoryBuffer {
80 MemoryBufferMem(StringRef InputData, bool RequiresNullTerminator) {
81 init(InputData.begin(), InputData.end(), RequiresNullTerminator);
84 virtual const char *getBufferIdentifier() const {
85 // The name is stored after the class itself.
86 return reinterpret_cast<const char*>(this + 1);
89 virtual BufferKind getBufferKind() const {
90 return MemoryBuffer_Malloc;
95 /// getMemBuffer - Open the specified memory range as a MemoryBuffer. Note
96 /// that InputData must be a null terminated if RequiresNullTerminator is true!
97 MemoryBuffer *MemoryBuffer::getMemBuffer(StringRef InputData,
99 bool RequiresNullTerminator) {
100 return GetNamedBuffer<MemoryBufferMem>(InputData, BufferName,
101 RequiresNullTerminator);
104 /// getMemBufferCopy - Open the specified memory range as a MemoryBuffer,
105 /// copying the contents and taking ownership of it. This has no requirements
107 MemoryBuffer *MemoryBuffer::getMemBufferCopy(StringRef InputData,
108 StringRef BufferName) {
109 MemoryBuffer *Buf = getNewUninitMemBuffer(InputData.size(), BufferName);
111 memcpy(const_cast<char*>(Buf->getBufferStart()), InputData.data(),
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
120 MemoryBuffer *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));
131 // The name is stored after the class itself.
132 CopyStringRef(Mem + sizeof(MemoryBufferMem), BufferName);
134 // The buffer begins after the name and must be aligned.
135 char *Buf = Mem + AlignedStringLen;
136 Buf[Size] = 0; // Null terminate buffer.
138 return new (Mem) MemoryBufferMem(StringRef(Buf, Size), true);
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.
145 MemoryBuffer *MemoryBuffer::getNewMemBuffer(size_t Size, StringRef BufferName) {
146 MemoryBuffer *SB = getNewUninitMemBuffer(Size, BufferName);
148 memset(const_cast<char*>(SB->getBufferStart()), 0, Size);
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.
157 error_code MemoryBuffer::getFileOrSTDIN(StringRef Filename,
158 OwningPtr<MemoryBuffer> &result,
161 return getSTDIN(result);
162 return getFile(Filename, result, FileSize);
165 error_code MemoryBuffer::getFileOrSTDIN(const char *Filename,
166 OwningPtr<MemoryBuffer> &result,
168 if (strcmp(Filename, "-") == 0)
169 return getSTDIN(result);
170 return getFile(Filename, result, FileSize);
173 //===----------------------------------------------------------------------===//
174 // MemoryBuffer::getFile implementation.
175 //===----------------------------------------------------------------------===//
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.
181 class MemoryBufferMMapFile : public MemoryBufferMem {
183 MemoryBufferMMapFile(StringRef Buffer, bool RequiresNullTerminator)
184 : MemoryBufferMem(Buffer, RequiresNullTerminator) { }
186 ~MemoryBufferMMapFile() {
187 static int PageSize = sys::Process::GetPageSize();
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);
194 sys::Path::UnMapFilePages(reinterpret_cast<const char*>(RealStart),
198 virtual BufferKind getBufferKind() const {
199 return MemoryBuffer_MMap;
204 error_code MemoryBuffer::getFile(StringRef Filename,
205 OwningPtr<MemoryBuffer> &result,
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);
214 error_code MemoryBuffer::getFile(const char *Filename,
215 OwningPtr<MemoryBuffer> &result,
217 bool RequiresNullTerminator) {
218 // First check that the "file" is not a directory
220 error_code err = sys::fs::is_directory(Filename, is_dir);
224 return make_error_code(errc::is_a_directory);
226 int OpenFlags = O_RDONLY;
228 OpenFlags |= O_BINARY; // Open input file in binary mode on win32.
230 int FD = ::open(Filename, OpenFlags);
232 return error_code(errno, posix_category());
234 error_code ret = getOpenFile(FD, Filename, result, FileSize, FileSize,
235 0, RequiresNullTerminator);
240 static bool shouldUseMmap(int FD,
244 bool RequiresNullTerminator,
246 // We don't use mmap for small files because this can severely fragment our
248 if (MapSize < 4096*4)
251 if (!RequiresNullTerminator)
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());
265 FileSize = FileInfo.st_size;
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);
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)
283 error_code MemoryBuffer::getOpenFile(int FD, const char *Filename,
284 OwningPtr<MemoryBuffer> &result,
285 uint64_t FileSize, uint64_t MapSize,
287 bool RequiresNullTerminator) {
288 static int PageSize = sys::Process::GetPageSize();
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());
300 FileSize = FileInfo.st_size;
305 if (shouldUseMmap(FD, FileSize, MapSize, Offset, RequiresNullTerminator,
307 off_t RealMapOffset = Offset & ~(PageSize - 1);
308 off_t Delta = Offset - RealMapOffset;
309 size_t RealMapSize = MapSize + Delta;
311 if (const char *Pages = sys::Path::MapInFilePages(FD,
314 result.reset(GetNamedBuffer<MemoryBufferMMapFile>(
315 StringRef(Pages + Delta, MapSize), Filename, RequiresNullTerminator));
317 if (RequiresNullTerminator && result->getBufferEnd()[0] != '\0') {
318 // There could be a racing issue that resulted in the file being larger
319 // than the FileSize passed by the caller. We already have an assertion
320 // for this in MemoryBuffer::init() but have a runtime guarantee that
321 // the buffer will be null-terminated here, so do a copy that adds a
323 result.reset(MemoryBuffer::getMemBufferCopy(result->getBuffer(),
326 return error_code::success();
330 MemoryBuffer *Buf = MemoryBuffer::getNewUninitMemBuffer(MapSize, Filename);
332 // Failed to create a buffer. The only way it can fail is if
333 // new(std::nothrow) returns 0.
334 return make_error_code(errc::not_enough_memory);
337 OwningPtr<MemoryBuffer> SB(Buf);
338 char *BufPtr = const_cast<char*>(SB->getBufferStart());
340 size_t BytesLeft = MapSize;
342 if (lseek(FD, Offset, SEEK_SET) == -1)
343 return error_code(errno, posix_category());
348 ssize_t NumRead = ::pread(FD, BufPtr, BytesLeft, MapSize-BytesLeft+Offset);
350 ssize_t NumRead = ::read(FD, BufPtr, BytesLeft);
355 // Error while reading.
356 return error_code(errno, posix_category());
359 assert(0 && "We got inaccurate FileSize value or fstat reported an "
360 "invalid file size.");
361 *BufPtr = '\0'; // null-terminate at the actual size.
364 BytesLeft -= NumRead;
369 return error_code::success();
372 //===----------------------------------------------------------------------===//
373 // MemoryBuffer::getSTDIN implementation.
374 //===----------------------------------------------------------------------===//
376 error_code MemoryBuffer::getSTDIN(OwningPtr<MemoryBuffer> &result) {
377 // Read in all of the data from stdin, we cannot mmap stdin.
379 // FIXME: That isn't necessarily true, we should try to mmap stdin and
380 // fallback if it fails.
381 sys::Program::ChangeStdinToBinary();
383 const ssize_t ChunkSize = 4096*4;
384 SmallString<ChunkSize> Buffer;
386 // Read into Buffer until we hit EOF.
388 Buffer.reserve(Buffer.size() + ChunkSize);
389 ReadBytes = read(0, Buffer.end(), ChunkSize);
390 if (ReadBytes == -1) {
391 if (errno == EINTR) continue;
392 return error_code(errno, posix_category());
394 Buffer.set_size(Buffer.size() + ReadBytes);
395 } while (ReadBytes != 0);
397 result.reset(getMemBufferCopy(Buffer, "<stdin>"));
398 return error_code::success();