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/SmallString.h"
16 #include "llvm/Config/config.h"
17 #include "llvm/Support/Errc.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"
29 #include <sys/types.h>
30 #include <system_error>
31 #if !defined(_MSC_VER) && !defined(__MINGW32__)
38 //===----------------------------------------------------------------------===//
39 // MemoryBuffer implementation itself.
40 //===----------------------------------------------------------------------===//
42 MemoryBuffer::~MemoryBuffer() { }
44 /// init - Initialize this MemoryBuffer as a reference to externally allocated
45 /// memory, memory that we know is already null terminated.
46 void MemoryBuffer::init(const char *BufStart, const char *BufEnd,
47 bool RequiresNullTerminator) {
48 assert((!RequiresNullTerminator || BufEnd[0] == 0) &&
49 "Buffer is not null terminated!");
50 BufferStart = BufStart;
54 //===----------------------------------------------------------------------===//
55 // MemoryBufferMem implementation.
56 //===----------------------------------------------------------------------===//
58 /// CopyStringRef - Copies contents of a StringRef into a block of memory and
59 /// null-terminates it.
60 static void CopyStringRef(char *Memory, StringRef Data) {
61 memcpy(Memory, Data.data(), Data.size());
62 Memory[Data.size()] = 0; // Null terminate string.
66 struct NamedBufferAlloc {
68 NamedBufferAlloc(StringRef Name) : Name(Name) {}
72 void *operator new(size_t N, const NamedBufferAlloc &Alloc) {
73 char *Mem = static_cast<char *>(operator new(N + Alloc.Name.size() + 1));
74 CopyStringRef(Mem + N, Alloc.Name);
79 /// MemoryBufferMem - Named MemoryBuffer pointing to a block of memory.
80 class MemoryBufferMem : public MemoryBuffer {
82 MemoryBufferMem(StringRef InputData, bool RequiresNullTerminator) {
83 init(InputData.begin(), InputData.end(), RequiresNullTerminator);
86 const char *getBufferIdentifier() const override {
87 // The name is stored after the class itself.
88 return reinterpret_cast<const char*>(this + 1);
91 BufferKind getBufferKind() const override {
92 return MemoryBuffer_Malloc;
97 /// getMemBuffer - Open the specified memory range as a MemoryBuffer. Note
98 /// that InputData must be a null terminated if RequiresNullTerminator is true!
99 MemoryBuffer *MemoryBuffer::getMemBuffer(StringRef InputData,
100 StringRef BufferName,
101 bool RequiresNullTerminator) {
102 return new (NamedBufferAlloc(BufferName))
103 MemoryBufferMem(InputData, RequiresNullTerminator);
106 std::unique_ptr<MemoryBuffer>
107 MemoryBuffer::getMemBuffer(MemoryBufferRef Ref, bool RequiresNullTerminator) {
108 return std::unique_ptr<MemoryBuffer>(getMemBuffer(
109 Ref.getBuffer(), Ref.getBufferIdentifier(), RequiresNullTerminator));
112 /// getMemBufferCopy - Open the specified memory range as a MemoryBuffer,
113 /// copying the contents and taking ownership of it. This has no requirements
115 MemoryBuffer *MemoryBuffer::getMemBufferCopy(StringRef InputData,
116 StringRef BufferName) {
117 MemoryBuffer *Buf = getNewUninitMemBuffer(InputData.size(), BufferName);
118 if (!Buf) return nullptr;
119 memcpy(const_cast<char*>(Buf->getBufferStart()), InputData.data(),
124 /// getNewUninitMemBuffer - Allocate a new MemoryBuffer of the specified size
125 /// that is not initialized. Note that the caller should initialize the
126 /// memory allocated by this method. The memory is owned by the MemoryBuffer
128 MemoryBuffer *MemoryBuffer::getNewUninitMemBuffer(size_t Size,
129 StringRef BufferName) {
130 // Allocate space for the MemoryBuffer, the data and the name. It is important
131 // that MemoryBuffer and data are aligned so PointerIntPair works with them.
132 // TODO: Is 16-byte alignment enough? We copy small object files with large
133 // alignment expectations into this buffer.
134 size_t AlignedStringLen =
135 RoundUpToAlignment(sizeof(MemoryBufferMem) + BufferName.size() + 1, 16);
136 size_t RealLen = AlignedStringLen + Size + 1;
137 char *Mem = static_cast<char*>(operator new(RealLen, std::nothrow));
138 if (!Mem) return nullptr;
140 // The name is stored after the class itself.
141 CopyStringRef(Mem + sizeof(MemoryBufferMem), BufferName);
143 // The buffer begins after the name and must be aligned.
144 char *Buf = Mem + AlignedStringLen;
145 Buf[Size] = 0; // Null terminate buffer.
147 return new (Mem) MemoryBufferMem(StringRef(Buf, Size), true);
150 /// getNewMemBuffer - Allocate a new zero-initialized MemoryBuffer of the
151 /// specified size. Note that the caller need not initialize the memory
152 /// allocated by this method. The memory is owned by the MemoryBuffer object.
153 MemoryBuffer *MemoryBuffer::getNewMemBuffer(size_t Size, StringRef BufferName) {
154 MemoryBuffer *SB = getNewUninitMemBuffer(Size, BufferName);
155 if (!SB) return nullptr;
156 memset(const_cast<char*>(SB->getBufferStart()), 0, Size);
160 ErrorOr<std::unique_ptr<MemoryBuffer>>
161 MemoryBuffer::getFileOrSTDIN(StringRef Filename, int64_t FileSize) {
164 return getFile(Filename, FileSize);
168 //===----------------------------------------------------------------------===//
169 // MemoryBuffer::getFile implementation.
170 //===----------------------------------------------------------------------===//
173 /// \brief Memory maps a file descriptor using sys::fs::mapped_file_region.
175 /// This handles converting the offset into a legal offset on the platform.
176 class MemoryBufferMMapFile : public MemoryBuffer {
177 sys::fs::mapped_file_region MFR;
179 static uint64_t getLegalMapOffset(uint64_t Offset) {
180 return Offset & ~(sys::fs::mapped_file_region::alignment() - 1);
183 static uint64_t getLegalMapSize(uint64_t Len, uint64_t Offset) {
184 return Len + (Offset - getLegalMapOffset(Offset));
187 const char *getStart(uint64_t Len, uint64_t Offset) {
188 return MFR.const_data() + (Offset - getLegalMapOffset(Offset));
192 MemoryBufferMMapFile(bool RequiresNullTerminator, int FD, uint64_t Len,
193 uint64_t Offset, std::error_code EC)
194 : MFR(FD, false, sys::fs::mapped_file_region::readonly,
195 getLegalMapSize(Len, Offset), getLegalMapOffset(Offset), EC) {
197 const char *Start = getStart(Len, Offset);
198 init(Start, Start + Len, RequiresNullTerminator);
202 const char *getBufferIdentifier() const override {
203 // The name is stored after the class itself.
204 return reinterpret_cast<const char *>(this + 1);
207 BufferKind getBufferKind() const override {
208 return MemoryBuffer_MMap;
213 static ErrorOr<std::unique_ptr<MemoryBuffer>>
214 getMemoryBufferForStream(int FD, StringRef BufferName) {
215 const ssize_t ChunkSize = 4096*4;
216 SmallString<ChunkSize> Buffer;
218 // Read into Buffer until we hit EOF.
220 Buffer.reserve(Buffer.size() + ChunkSize);
221 ReadBytes = read(FD, Buffer.end(), ChunkSize);
222 if (ReadBytes == -1) {
223 if (errno == EINTR) continue;
224 return std::error_code(errno, std::generic_category());
226 Buffer.set_size(Buffer.size() + ReadBytes);
227 } while (ReadBytes != 0);
229 std::unique_ptr<MemoryBuffer> Ret(
230 MemoryBuffer::getMemBufferCopy(Buffer, BufferName));
231 return std::move(Ret);
234 static ErrorOr<std::unique_ptr<MemoryBuffer>>
235 getFileAux(const char *Filename, int64_t FileSize, bool RequiresNullTerminator,
236 bool IsVolatileSize);
238 ErrorOr<std::unique_ptr<MemoryBuffer>>
239 MemoryBuffer::getFile(Twine Filename, int64_t FileSize,
240 bool RequiresNullTerminator, bool IsVolatileSize) {
241 // Ensure the path is null terminated.
242 SmallString<256> PathBuf;
243 StringRef NullTerminatedName = Filename.toNullTerminatedStringRef(PathBuf);
244 return getFileAux(NullTerminatedName.data(), FileSize, RequiresNullTerminator,
248 static ErrorOr<std::unique_ptr<MemoryBuffer>>
249 getOpenFileImpl(int FD, const char *Filename, uint64_t FileSize,
250 uint64_t MapSize, int64_t Offset, bool RequiresNullTerminator,
251 bool IsVolatileSize);
253 static ErrorOr<std::unique_ptr<MemoryBuffer>>
254 getFileAux(const char *Filename, int64_t FileSize, bool RequiresNullTerminator,
255 bool IsVolatileSize) {
257 std::error_code EC = sys::fs::openFileForRead(Filename, FD);
261 ErrorOr<std::unique_ptr<MemoryBuffer>> Ret =
262 getOpenFileImpl(FD, Filename, FileSize, FileSize, 0,
263 RequiresNullTerminator, IsVolatileSize);
268 static bool shouldUseMmap(int FD,
272 bool RequiresNullTerminator,
274 bool IsVolatileSize) {
275 // mmap may leave the buffer without null terminator if the file size changed
276 // by the time the last page is mapped in, so avoid it if the file size is
281 // We don't use mmap for small files because this can severely fragment our
283 if (MapSize < 4 * 4096 || MapSize < (unsigned)PageSize)
286 if (!RequiresNullTerminator)
290 // If we don't know the file size, use fstat to find out. fstat on an open
291 // file descriptor is cheaper than stat on a random path.
292 // FIXME: this chunk of code is duplicated, but it avoids a fstat when
293 // RequiresNullTerminator = false and MapSize != -1.
294 if (FileSize == size_t(-1)) {
295 sys::fs::file_status Status;
296 if (sys::fs::status(FD, Status))
298 FileSize = Status.getSize();
301 // If we need a null terminator and the end of the map is inside the file,
302 // we cannot use mmap.
303 size_t End = Offset + MapSize;
304 assert(End <= FileSize);
308 // Don't try to map files that are exactly a multiple of the system page size
309 // if we need a null terminator.
310 if ((FileSize & (PageSize -1)) == 0)
313 #if defined(__CYGWIN__)
314 // Don't try to map files that are exactly a multiple of the physical page size
315 // if we need a null terminator.
316 // FIXME: We should reorganize again getPageSize() on Win32.
317 if ((FileSize & (4096 - 1)) == 0)
324 static ErrorOr<std::unique_ptr<MemoryBuffer>>
325 getOpenFileImpl(int FD, const char *Filename, uint64_t FileSize,
326 uint64_t MapSize, int64_t Offset, bool RequiresNullTerminator,
327 bool IsVolatileSize) {
328 static int PageSize = sys::process::get_self()->page_size();
330 // Default is to map the full file.
331 if (MapSize == uint64_t(-1)) {
332 // If we don't know the file size, use fstat to find out. fstat on an open
333 // file descriptor is cheaper than stat on a random path.
334 if (FileSize == uint64_t(-1)) {
335 sys::fs::file_status Status;
336 std::error_code EC = sys::fs::status(FD, Status);
340 // If this not a file or a block device (e.g. it's a named pipe
341 // or character device), we can't trust the size. Create the memory
342 // buffer by copying off the stream.
343 sys::fs::file_type Type = Status.type();
344 if (Type != sys::fs::file_type::regular_file &&
345 Type != sys::fs::file_type::block_file)
346 return getMemoryBufferForStream(FD, Filename);
348 FileSize = Status.getSize();
353 if (shouldUseMmap(FD, FileSize, MapSize, Offset, RequiresNullTerminator,
354 PageSize, IsVolatileSize)) {
356 std::unique_ptr<MemoryBuffer> Result(
357 new (NamedBufferAlloc(Filename))
358 MemoryBufferMMapFile(RequiresNullTerminator, FD, MapSize, Offset, EC));
360 return std::move(Result);
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 std::unique_ptr<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 std::error_code(errno, std::generic_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 std::error_code(errno, std::generic_category());
392 memset(BufPtr, 0, BytesLeft); // zero-initialize rest of the buffer.
395 BytesLeft -= NumRead;
399 return std::move(SB);
402 ErrorOr<std::unique_ptr<MemoryBuffer>>
403 MemoryBuffer::getOpenFile(int FD, const char *Filename, uint64_t FileSize,
404 bool RequiresNullTerminator, bool IsVolatileSize) {
405 return getOpenFileImpl(FD, Filename, FileSize, FileSize, 0,
406 RequiresNullTerminator, IsVolatileSize);
409 ErrorOr<std::unique_ptr<MemoryBuffer>>
410 MemoryBuffer::getOpenFileSlice(int FD, const char *Filename, uint64_t MapSize,
411 int64_t Offset, bool IsVolatileSize) {
412 return getOpenFileImpl(FD, Filename, -1, MapSize, Offset, false,
416 ErrorOr<std::unique_ptr<MemoryBuffer>> MemoryBuffer::getSTDIN() {
417 // Read in all of the data from stdin, we cannot mmap stdin.
419 // FIXME: That isn't necessarily true, we should try to mmap stdin and
420 // fallback if it fails.
421 sys::ChangeStdinToBinary();
423 return getMemoryBufferForStream(0, "<stdin>");
426 MemoryBufferRef MemoryBuffer::getMemBufferRef() const {
427 StringRef Data = getBuffer();
428 StringRef Identifier = getBufferIdentifier();
429 return MemoryBufferRef(Data, Identifier);