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/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"
31 #include <sys/types.h>
32 #if !defined(_MSC_VER) && !defined(__MINGW32__)
36 // Simplistic definitinos of these macros for use in getOpenFile.
38 #define S_ISREG(x) (1)
41 #define S_ISBLK(x) (0)
47 //===----------------------------------------------------------------------===//
48 // MemoryBuffer implementation itself.
49 //===----------------------------------------------------------------------===//
51 MemoryBuffer::~MemoryBuffer() { }
53 /// init - Initialize this MemoryBuffer as a reference to externally allocated
54 /// memory, memory that we know is already null terminated.
55 void MemoryBuffer::init(const char *BufStart, const char *BufEnd,
56 bool RequiresNullTerminator) {
57 assert((!RequiresNullTerminator || BufEnd[0] == 0) &&
58 "Buffer is not null terminated!");
59 BufferStart = BufStart;
63 //===----------------------------------------------------------------------===//
64 // MemoryBufferMem implementation.
65 //===----------------------------------------------------------------------===//
67 /// CopyStringRef - Copies contents of a StringRef into a block of memory and
68 /// null-terminates it.
69 static void CopyStringRef(char *Memory, StringRef Data) {
70 memcpy(Memory, Data.data(), Data.size());
71 Memory[Data.size()] = 0; // Null terminate string.
75 struct NamedBufferAlloc {
77 NamedBufferAlloc(StringRef Name) : Name(Name) {}
81 void *operator new(size_t N, const NamedBufferAlloc &Alloc) {
82 char *Mem = static_cast<char *>(operator new(N + Alloc.Name.size() + 1));
83 CopyStringRef(Mem + N, Alloc.Name);
88 /// MemoryBufferMem - Named MemoryBuffer pointing to a block of memory.
89 class MemoryBufferMem : public MemoryBuffer {
91 MemoryBufferMem(StringRef InputData, bool RequiresNullTerminator) {
92 init(InputData.begin(), InputData.end(), RequiresNullTerminator);
95 virtual const char *getBufferIdentifier() const LLVM_OVERRIDE {
96 // The name is stored after the class itself.
97 return reinterpret_cast<const char*>(this + 1);
100 virtual BufferKind getBufferKind() const LLVM_OVERRIDE {
101 return MemoryBuffer_Malloc;
106 /// getMemBuffer - Open the specified memory range as a MemoryBuffer. Note
107 /// that InputData must be a null terminated if RequiresNullTerminator is true!
108 MemoryBuffer *MemoryBuffer::getMemBuffer(StringRef InputData,
109 StringRef BufferName,
110 bool RequiresNullTerminator) {
111 return new (NamedBufferAlloc(BufferName))
112 MemoryBufferMem(InputData, RequiresNullTerminator);
115 /// getMemBufferCopy - Open the specified memory range as a MemoryBuffer,
116 /// copying the contents and taking ownership of it. This has no requirements
118 MemoryBuffer *MemoryBuffer::getMemBufferCopy(StringRef InputData,
119 StringRef BufferName) {
120 MemoryBuffer *Buf = getNewUninitMemBuffer(InputData.size(), BufferName);
122 memcpy(const_cast<char*>(Buf->getBufferStart()), InputData.data(),
127 /// getNewUninitMemBuffer - Allocate a new MemoryBuffer of the specified size
128 /// that is not initialized. Note that the caller should initialize the
129 /// memory allocated by this method. The memory is owned by the MemoryBuffer
131 MemoryBuffer *MemoryBuffer::getNewUninitMemBuffer(size_t Size,
132 StringRef BufferName) {
133 // Allocate space for the MemoryBuffer, the data and the name. It is important
134 // that MemoryBuffer and data are aligned so PointerIntPair works with them.
135 size_t AlignedStringLen =
136 RoundUpToAlignment(sizeof(MemoryBufferMem) + BufferName.size() + 1,
137 sizeof(void*)); // TODO: Is sizeof(void*) enough?
138 size_t RealLen = AlignedStringLen + Size + 1;
139 char *Mem = static_cast<char*>(operator new(RealLen, std::nothrow));
142 // The name is stored after the class itself.
143 CopyStringRef(Mem + sizeof(MemoryBufferMem), BufferName);
145 // The buffer begins after the name and must be aligned.
146 char *Buf = Mem + AlignedStringLen;
147 Buf[Size] = 0; // Null terminate buffer.
149 return new (Mem) MemoryBufferMem(StringRef(Buf, Size), true);
152 /// getNewMemBuffer - Allocate a new MemoryBuffer of the specified size that
153 /// is completely initialized to zeros. Note that the caller should
154 /// initialize the memory allocated by this method. The memory is owned by
155 /// the MemoryBuffer object.
156 MemoryBuffer *MemoryBuffer::getNewMemBuffer(size_t Size, StringRef BufferName) {
157 MemoryBuffer *SB = getNewUninitMemBuffer(Size, BufferName);
159 memset(const_cast<char*>(SB->getBufferStart()), 0, Size);
164 /// getFileOrSTDIN - Open the specified file as a MemoryBuffer, or open stdin
165 /// if the Filename is "-". If an error occurs, this returns null and fills
166 /// in *ErrStr with a reason. If stdin is empty, this API (unlike getSTDIN)
167 /// returns an empty buffer.
168 error_code MemoryBuffer::getFileOrSTDIN(StringRef Filename,
169 OwningPtr<MemoryBuffer> &result,
172 return getSTDIN(result);
173 return getFile(Filename, result, FileSize);
176 //===----------------------------------------------------------------------===//
177 // MemoryBuffer::getFile implementation.
178 //===----------------------------------------------------------------------===//
181 /// \brief Memorry maps a file descriptor using sys::fs::mapped_file_region.
183 /// This handles converting the offset into a legal offset on the platform.
184 class MemoryBufferMMapFile : public MemoryBuffer {
185 sys::fs::mapped_file_region MFR;
187 static uint64_t getLegalMapOffset(uint64_t Offset) {
188 return Offset & ~(sys::fs::mapped_file_region::alignment() - 1);
191 static uint64_t getLegalMapSize(uint64_t Len, uint64_t Offset) {
192 return Len + (Offset - getLegalMapOffset(Offset));
195 const char *getStart(uint64_t Len, uint64_t Offset) {
196 return MFR.const_data() + (Offset - getLegalMapOffset(Offset));
200 MemoryBufferMMapFile(bool RequiresNullTerminator, int FD, uint64_t Len,
201 uint64_t Offset, error_code EC)
202 : MFR(FD, false, sys::fs::mapped_file_region::readonly,
203 getLegalMapSize(Len, Offset), getLegalMapOffset(Offset), EC) {
205 const char *Start = getStart(Len, Offset);
206 init(Start, Start + Len, RequiresNullTerminator);
210 virtual const char *getBufferIdentifier() const LLVM_OVERRIDE {
211 // The name is stored after the class itself.
212 return reinterpret_cast<const char *>(this + 1);
215 virtual BufferKind getBufferKind() const LLVM_OVERRIDE {
216 return MemoryBuffer_MMap;
221 static error_code getMemoryBufferForStream(int FD,
222 StringRef BufferName,
223 OwningPtr<MemoryBuffer> &result) {
224 const ssize_t ChunkSize = 4096*4;
225 SmallString<ChunkSize> Buffer;
227 // Read into Buffer until we hit EOF.
229 Buffer.reserve(Buffer.size() + ChunkSize);
230 ReadBytes = read(FD, Buffer.end(), ChunkSize);
231 if (ReadBytes == -1) {
232 if (errno == EINTR) continue;
233 return error_code(errno, posix_category());
235 Buffer.set_size(Buffer.size() + ReadBytes);
236 } while (ReadBytes != 0);
238 result.reset(MemoryBuffer::getMemBufferCopy(Buffer, BufferName));
239 return error_code::success();
242 error_code MemoryBuffer::getFile(StringRef Filename,
243 OwningPtr<MemoryBuffer> &result,
245 bool RequiresNullTerminator) {
246 // Ensure the path is null terminated.
247 SmallString<256> PathBuf(Filename.begin(), Filename.end());
248 return MemoryBuffer::getFile(PathBuf.c_str(), result, FileSize,
249 RequiresNullTerminator);
252 error_code MemoryBuffer::getFile(const char *Filename,
253 OwningPtr<MemoryBuffer> &result,
255 bool RequiresNullTerminator) {
256 // FIXME: Review if this check is unnecessary on windows as well.
258 // First check that the "file" is not a directory
260 error_code err = sys::fs::is_directory(Filename, is_dir);
264 return make_error_code(errc::is_a_directory);
267 int OpenFlags = O_RDONLY;
269 OpenFlags |= O_BINARY; // Open input file in binary mode on win32.
271 int FD = ::open(Filename, OpenFlags);
273 return error_code(errno, posix_category());
275 error_code ret = getOpenFile(FD, Filename, result, FileSize, FileSize,
276 0, RequiresNullTerminator);
281 static bool shouldUseMmap(int FD,
285 bool RequiresNullTerminator,
287 // We don't use mmap for small files because this can severely fragment our
289 if (MapSize < 4096*4)
292 if (!RequiresNullTerminator)
296 // If we don't know the file size, use fstat to find out. fstat on an open
297 // file descriptor is cheaper than stat on a random path.
298 // FIXME: this chunk of code is duplicated, but it avoids a fstat when
299 // RequiresNullTerminator = false and MapSize != -1.
300 if (FileSize == size_t(-1)) {
301 struct stat FileInfo;
302 // TODO: This should use fstat64 when available.
303 if (fstat(FD, &FileInfo) == -1) {
304 return error_code(errno, posix_category());
306 FileSize = FileInfo.st_size;
309 // If we need a null terminator and the end of the map is inside the file,
310 // we cannot use mmap.
311 size_t End = Offset + MapSize;
312 assert(End <= FileSize);
316 // Don't try to map files that are exactly a multiple of the system page size
317 // if we need a null terminator.
318 if ((FileSize & (PageSize -1)) == 0)
324 error_code MemoryBuffer::getOpenFile(int FD, const char *Filename,
325 OwningPtr<MemoryBuffer> &result,
326 uint64_t FileSize, uint64_t MapSize,
328 bool RequiresNullTerminator) {
329 static int PageSize = sys::process::get_self()->page_size();
331 // Default is to map the full file.
332 if (MapSize == uint64_t(-1)) {
333 // If we don't know the file size, use fstat to find out. fstat on an open
334 // file descriptor is cheaper than stat on a random path.
335 if (FileSize == uint64_t(-1)) {
336 struct stat FileInfo;
337 // TODO: This should use fstat64 when available.
338 if (fstat(FD, &FileInfo) == -1) {
339 return error_code(errno, posix_category());
342 // If this not a file or a block device (e.g. it's a named pipe
343 // or character device), we can't trust the size. Create the memory
344 // buffer by copying off the stream.
345 if (!S_ISREG(FileInfo.st_mode) && !S_ISBLK(FileInfo.st_mode)) {
346 return getMemoryBufferForStream(FD, Filename, result);
349 FileSize = FileInfo.st_size;
354 if (shouldUseMmap(FD, FileSize, MapSize, Offset, RequiresNullTerminator,
357 result.reset(new (NamedBufferAlloc(Filename)) MemoryBufferMMapFile(
358 RequiresNullTerminator, FD, MapSize, Offset, EC));
360 return error_code::success();
363 MemoryBuffer *Buf = MemoryBuffer::getNewUninitMemBuffer(MapSize, Filename);
365 // Failed to create a buffer. The only way it can fail is if
366 // new(std::nothrow) returns 0.
367 return make_error_code(errc::not_enough_memory);
370 OwningPtr<MemoryBuffer> SB(Buf);
371 char *BufPtr = const_cast<char*>(SB->getBufferStart());
373 size_t BytesLeft = MapSize;
375 if (lseek(FD, Offset, SEEK_SET) == -1)
376 return error_code(errno, posix_category());
381 ssize_t NumRead = ::pread(FD, BufPtr, BytesLeft, MapSize-BytesLeft+Offset);
383 ssize_t NumRead = ::read(FD, BufPtr, BytesLeft);
388 // Error while reading.
389 return error_code(errno, posix_category());
392 assert(0 && "We got inaccurate FileSize value or fstat reported an "
393 "invalid file size.");
394 *BufPtr = '\0'; // null-terminate at the actual size.
397 BytesLeft -= NumRead;
402 return error_code::success();
405 //===----------------------------------------------------------------------===//
406 // MemoryBuffer::getSTDIN implementation.
407 //===----------------------------------------------------------------------===//
409 error_code MemoryBuffer::getSTDIN(OwningPtr<MemoryBuffer> &result) {
410 // Read in all of the data from stdin, we cannot mmap stdin.
412 // FIXME: That isn't necessarily true, we should try to mmap stdin and
413 // fallback if it fails.
414 sys::ChangeStdinToBinary();
416 return getMemoryBufferForStream(0, "<stdin>", result);