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 /// getMemBufferCopy - Open the specified memory range as a MemoryBuffer,
107 /// copying the contents and taking ownership of it. This has no requirements
109 MemoryBuffer *MemoryBuffer::getMemBufferCopy(StringRef InputData,
110 StringRef BufferName) {
111 MemoryBuffer *Buf = getNewUninitMemBuffer(InputData.size(), BufferName);
112 if (!Buf) return nullptr;
113 memcpy(const_cast<char*>(Buf->getBufferStart()), InputData.data(),
118 /// getNewUninitMemBuffer - Allocate a new MemoryBuffer of the specified size
119 /// that is not initialized. Note that the caller should initialize the
120 /// memory allocated by this method. The memory is owned by the MemoryBuffer
122 MemoryBuffer *MemoryBuffer::getNewUninitMemBuffer(size_t Size,
123 StringRef BufferName) {
124 // Allocate space for the MemoryBuffer, the data and the name. It is important
125 // that MemoryBuffer and data are aligned so PointerIntPair works with them.
126 // TODO: Is 16-byte alignment enough? We copy small object files with large
127 // alignment expectations into this buffer.
128 size_t AlignedStringLen =
129 RoundUpToAlignment(sizeof(MemoryBufferMem) + BufferName.size() + 1, 16);
130 size_t RealLen = AlignedStringLen + Size + 1;
131 char *Mem = static_cast<char*>(operator new(RealLen, std::nothrow));
132 if (!Mem) return nullptr;
134 // The name is stored after the class itself.
135 CopyStringRef(Mem + sizeof(MemoryBufferMem), BufferName);
137 // The buffer begins after the name and must be aligned.
138 char *Buf = Mem + AlignedStringLen;
139 Buf[Size] = 0; // Null terminate buffer.
141 return new (Mem) MemoryBufferMem(StringRef(Buf, Size), true);
144 /// getNewMemBuffer - Allocate a new zero-initialized MemoryBuffer of the
145 /// specified size. Note that the caller need not initialize the memory
146 /// allocated by this method. The memory is owned by the MemoryBuffer object.
147 MemoryBuffer *MemoryBuffer::getNewMemBuffer(size_t Size, StringRef BufferName) {
148 MemoryBuffer *SB = getNewUninitMemBuffer(Size, BufferName);
149 if (!SB) return nullptr;
150 memset(const_cast<char*>(SB->getBufferStart()), 0, Size);
154 ErrorOr<std::unique_ptr<MemoryBuffer>>
155 MemoryBuffer::getFileOrSTDIN(StringRef Filename, int64_t FileSize) {
158 return getFile(Filename, FileSize);
162 //===----------------------------------------------------------------------===//
163 // MemoryBuffer::getFile implementation.
164 //===----------------------------------------------------------------------===//
167 /// \brief Memory maps a file descriptor using sys::fs::mapped_file_region.
169 /// This handles converting the offset into a legal offset on the platform.
170 class MemoryBufferMMapFile : public MemoryBuffer {
171 sys::fs::mapped_file_region MFR;
173 static uint64_t getLegalMapOffset(uint64_t Offset) {
174 return Offset & ~(sys::fs::mapped_file_region::alignment() - 1);
177 static uint64_t getLegalMapSize(uint64_t Len, uint64_t Offset) {
178 return Len + (Offset - getLegalMapOffset(Offset));
181 const char *getStart(uint64_t Len, uint64_t Offset) {
182 return MFR.const_data() + (Offset - getLegalMapOffset(Offset));
186 MemoryBufferMMapFile(bool RequiresNullTerminator, int FD, uint64_t Len,
187 uint64_t Offset, std::error_code EC)
188 : MFR(FD, false, sys::fs::mapped_file_region::readonly,
189 getLegalMapSize(Len, Offset), getLegalMapOffset(Offset), EC) {
191 const char *Start = getStart(Len, Offset);
192 init(Start, Start + Len, RequiresNullTerminator);
196 const char *getBufferIdentifier() const override {
197 // The name is stored after the class itself.
198 return reinterpret_cast<const char *>(this + 1);
201 BufferKind getBufferKind() const override {
202 return MemoryBuffer_MMap;
207 static ErrorOr<std::unique_ptr<MemoryBuffer>>
208 getMemoryBufferForStream(int FD, StringRef BufferName) {
209 const ssize_t ChunkSize = 4096*4;
210 SmallString<ChunkSize> Buffer;
212 // Read into Buffer until we hit EOF.
214 Buffer.reserve(Buffer.size() + ChunkSize);
215 ReadBytes = read(FD, Buffer.end(), ChunkSize);
216 if (ReadBytes == -1) {
217 if (errno == EINTR) continue;
218 return std::error_code(errno, std::generic_category());
220 Buffer.set_size(Buffer.size() + ReadBytes);
221 } while (ReadBytes != 0);
223 std::unique_ptr<MemoryBuffer> Ret(
224 MemoryBuffer::getMemBufferCopy(Buffer, BufferName));
225 return std::move(Ret);
228 static ErrorOr<std::unique_ptr<MemoryBuffer>>
229 getFileAux(const char *Filename, int64_t FileSize, bool RequiresNullTerminator,
230 bool IsVolatileSize);
232 ErrorOr<std::unique_ptr<MemoryBuffer>>
233 MemoryBuffer::getFile(Twine Filename, int64_t FileSize,
234 bool RequiresNullTerminator, bool IsVolatileSize) {
235 // Ensure the path is null terminated.
236 SmallString<256> PathBuf;
237 StringRef NullTerminatedName = Filename.toNullTerminatedStringRef(PathBuf);
238 return getFileAux(NullTerminatedName.data(), FileSize, RequiresNullTerminator,
242 static ErrorOr<std::unique_ptr<MemoryBuffer>>
243 getOpenFileImpl(int FD, const char *Filename, uint64_t FileSize,
244 uint64_t MapSize, int64_t Offset, bool RequiresNullTerminator,
245 bool IsVolatileSize);
247 static ErrorOr<std::unique_ptr<MemoryBuffer>>
248 getFileAux(const char *Filename, int64_t FileSize, bool RequiresNullTerminator,
249 bool IsVolatileSize) {
251 std::error_code EC = sys::fs::openFileForRead(Filename, FD);
255 ErrorOr<std::unique_ptr<MemoryBuffer>> Ret =
256 getOpenFileImpl(FD, Filename, FileSize, FileSize, 0,
257 RequiresNullTerminator, IsVolatileSize);
262 static bool shouldUseMmap(int FD,
266 bool RequiresNullTerminator,
268 bool IsVolatileSize) {
269 // mmap may leave the buffer without null terminator if the file size changed
270 // by the time the last page is mapped in, so avoid it if the file size is
275 // We don't use mmap for small files because this can severely fragment our
277 if (MapSize < 4 * 4096 || MapSize < (unsigned)PageSize)
280 if (!RequiresNullTerminator)
284 // If we don't know the file size, use fstat to find out. fstat on an open
285 // file descriptor is cheaper than stat on a random path.
286 // FIXME: this chunk of code is duplicated, but it avoids a fstat when
287 // RequiresNullTerminator = false and MapSize != -1.
288 if (FileSize == size_t(-1)) {
289 sys::fs::file_status Status;
290 if (sys::fs::status(FD, Status))
292 FileSize = Status.getSize();
295 // If we need a null terminator and the end of the map is inside the file,
296 // we cannot use mmap.
297 size_t End = Offset + MapSize;
298 assert(End <= FileSize);
302 // Don't try to map files that are exactly a multiple of the system page size
303 // if we need a null terminator.
304 if ((FileSize & (PageSize -1)) == 0)
307 #if defined(__CYGWIN__)
308 // Don't try to map files that are exactly a multiple of the physical page size
309 // if we need a null terminator.
310 // FIXME: We should reorganize again getPageSize() on Win32.
311 if ((FileSize & (4096 - 1)) == 0)
318 static ErrorOr<std::unique_ptr<MemoryBuffer>>
319 getOpenFileImpl(int FD, const char *Filename, uint64_t FileSize,
320 uint64_t MapSize, int64_t Offset, bool RequiresNullTerminator,
321 bool IsVolatileSize) {
322 static int PageSize = sys::process::get_self()->page_size();
324 // Default is to map the full file.
325 if (MapSize == uint64_t(-1)) {
326 // If we don't know the file size, use fstat to find out. fstat on an open
327 // file descriptor is cheaper than stat on a random path.
328 if (FileSize == uint64_t(-1)) {
329 sys::fs::file_status Status;
330 std::error_code EC = sys::fs::status(FD, Status);
334 // If this not a file or a block device (e.g. it's a named pipe
335 // or character device), we can't trust the size. Create the memory
336 // buffer by copying off the stream.
337 sys::fs::file_type Type = Status.type();
338 if (Type != sys::fs::file_type::regular_file &&
339 Type != sys::fs::file_type::block_file)
340 return getMemoryBufferForStream(FD, Filename);
342 FileSize = Status.getSize();
347 if (shouldUseMmap(FD, FileSize, MapSize, Offset, RequiresNullTerminator,
348 PageSize, IsVolatileSize)) {
350 std::unique_ptr<MemoryBuffer> Result(
351 new (NamedBufferAlloc(Filename))
352 MemoryBufferMMapFile(RequiresNullTerminator, FD, MapSize, Offset, EC));
354 return std::move(Result);
357 MemoryBuffer *Buf = MemoryBuffer::getNewUninitMemBuffer(MapSize, Filename);
359 // Failed to create a buffer. The only way it can fail is if
360 // new(std::nothrow) returns 0.
361 return make_error_code(errc::not_enough_memory);
364 std::unique_ptr<MemoryBuffer> SB(Buf);
365 char *BufPtr = const_cast<char*>(SB->getBufferStart());
367 size_t BytesLeft = MapSize;
369 if (lseek(FD, Offset, SEEK_SET) == -1)
370 return std::error_code(errno, std::generic_category());
375 ssize_t NumRead = ::pread(FD, BufPtr, BytesLeft, MapSize-BytesLeft+Offset);
377 ssize_t NumRead = ::read(FD, BufPtr, BytesLeft);
382 // Error while reading.
383 return std::error_code(errno, std::generic_category());
386 memset(BufPtr, 0, BytesLeft); // zero-initialize rest of the buffer.
389 BytesLeft -= NumRead;
393 return std::move(SB);
396 ErrorOr<std::unique_ptr<MemoryBuffer>>
397 MemoryBuffer::getOpenFile(int FD, const char *Filename, uint64_t FileSize,
398 bool RequiresNullTerminator, bool IsVolatileSize) {
399 return getOpenFileImpl(FD, Filename, FileSize, FileSize, 0,
400 RequiresNullTerminator, IsVolatileSize);
403 ErrorOr<std::unique_ptr<MemoryBuffer>>
404 MemoryBuffer::getOpenFileSlice(int FD, const char *Filename, uint64_t MapSize,
405 int64_t Offset, bool IsVolatileSize) {
406 return getOpenFileImpl(FD, Filename, -1, MapSize, Offset, false,
410 ErrorOr<std::unique_ptr<MemoryBuffer>> MemoryBuffer::getSTDIN() {
411 // Read in all of the data from stdin, we cannot mmap stdin.
413 // FIXME: That isn't necessarily true, we should try to mmap stdin and
414 // fallback if it fails.
415 sys::ChangeStdinToBinary();
417 return getMemoryBufferForStream(0, "<stdin>");
420 MemoryBufferRef MemoryBuffer::getMemBufferRef() const {
421 StringRef Data = getBuffer();
422 StringRef Identifier = getBufferIdentifier();
423 return MemoryBufferRef(Data, Identifier);