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 std::unique_ptr<MemoryBuffer>
98 MemoryBuffer::getMemBuffer(StringRef InputData, StringRef BufferName,
99 bool RequiresNullTerminator) {
100 auto *Ret = new (NamedBufferAlloc(BufferName))
101 MemoryBufferMem(InputData, RequiresNullTerminator);
102 return std::unique_ptr<MemoryBuffer>(Ret);
105 std::unique_ptr<MemoryBuffer>
106 MemoryBuffer::getMemBuffer(MemoryBufferRef Ref, bool RequiresNullTerminator) {
107 return std::unique_ptr<MemoryBuffer>(getMemBuffer(
108 Ref.getBuffer(), Ref.getBufferIdentifier(), RequiresNullTerminator));
111 std::unique_ptr<MemoryBuffer>
112 MemoryBuffer::getMemBufferCopy(StringRef InputData, StringRef BufferName) {
113 std::unique_ptr<MemoryBuffer> Buf =
114 getNewUninitMemBuffer(InputData.size(), BufferName);
117 memcpy(const_cast<char*>(Buf->getBufferStart()), InputData.data(),
122 std::unique_ptr<MemoryBuffer>
123 MemoryBuffer::getNewUninitMemBuffer(size_t Size, 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));
135 // The name is stored after the class itself.
136 CopyStringRef(Mem + sizeof(MemoryBufferMem), BufferName);
138 // The buffer begins after the name and must be aligned.
139 char *Buf = Mem + AlignedStringLen;
140 Buf[Size] = 0; // Null terminate buffer.
142 auto *Ret = new (Mem) MemoryBufferMem(StringRef(Buf, Size), true);
143 return std::unique_ptr<MemoryBuffer>(Ret);
146 std::unique_ptr<MemoryBuffer>
147 MemoryBuffer::getNewMemBuffer(size_t Size, StringRef BufferName) {
148 std::unique_ptr<MemoryBuffer> SB = getNewUninitMemBuffer(Size, BufferName);
151 memset(const_cast<char*>(SB->getBufferStart()), 0, Size);
155 ErrorOr<std::unique_ptr<MemoryBuffer>>
156 MemoryBuffer::getFileOrSTDIN(StringRef Filename, int64_t FileSize) {
159 return getFile(Filename, FileSize);
163 //===----------------------------------------------------------------------===//
164 // MemoryBuffer::getFile implementation.
165 //===----------------------------------------------------------------------===//
168 /// \brief Memory maps a file descriptor using sys::fs::mapped_file_region.
170 /// This handles converting the offset into a legal offset on the platform.
171 class MemoryBufferMMapFile : public MemoryBuffer {
172 sys::fs::mapped_file_region MFR;
174 static uint64_t getLegalMapOffset(uint64_t Offset) {
175 return Offset & ~(sys::fs::mapped_file_region::alignment() - 1);
178 static uint64_t getLegalMapSize(uint64_t Len, uint64_t Offset) {
179 return Len + (Offset - getLegalMapOffset(Offset));
182 const char *getStart(uint64_t Len, uint64_t Offset) {
183 return MFR.const_data() + (Offset - getLegalMapOffset(Offset));
187 MemoryBufferMMapFile(bool RequiresNullTerminator, int FD, uint64_t Len,
188 uint64_t Offset, std::error_code EC)
189 : MFR(FD, false, sys::fs::mapped_file_region::readonly,
190 getLegalMapSize(Len, Offset), getLegalMapOffset(Offset), EC) {
192 const char *Start = getStart(Len, Offset);
193 init(Start, Start + Len, RequiresNullTerminator);
197 const char *getBufferIdentifier() const override {
198 // The name is stored after the class itself.
199 return reinterpret_cast<const char *>(this + 1);
202 BufferKind getBufferKind() const override {
203 return MemoryBuffer_MMap;
208 static ErrorOr<std::unique_ptr<MemoryBuffer>>
209 getMemoryBufferForStream(int FD, StringRef BufferName) {
210 const ssize_t ChunkSize = 4096*4;
211 SmallString<ChunkSize> Buffer;
213 // Read into Buffer until we hit EOF.
215 Buffer.reserve(Buffer.size() + ChunkSize);
216 ReadBytes = read(FD, Buffer.end(), ChunkSize);
217 if (ReadBytes == -1) {
218 if (errno == EINTR) continue;
219 return std::error_code(errno, std::generic_category());
221 Buffer.set_size(Buffer.size() + ReadBytes);
222 } while (ReadBytes != 0);
224 return MemoryBuffer::getMemBufferCopy(Buffer, BufferName);
227 static ErrorOr<std::unique_ptr<MemoryBuffer>>
228 getFileAux(const char *Filename, int64_t FileSize, bool RequiresNullTerminator,
229 bool IsVolatileSize);
231 ErrorOr<std::unique_ptr<MemoryBuffer>>
232 MemoryBuffer::getFile(Twine Filename, int64_t FileSize,
233 bool RequiresNullTerminator, bool IsVolatileSize) {
234 // Ensure the path is null terminated.
235 SmallString<256> PathBuf;
236 StringRef NullTerminatedName = Filename.toNullTerminatedStringRef(PathBuf);
237 return getFileAux(NullTerminatedName.data(), FileSize, RequiresNullTerminator,
241 static ErrorOr<std::unique_ptr<MemoryBuffer>>
242 getOpenFileImpl(int FD, const char *Filename, uint64_t FileSize,
243 uint64_t MapSize, int64_t Offset, bool RequiresNullTerminator,
244 bool IsVolatileSize);
246 static ErrorOr<std::unique_ptr<MemoryBuffer>>
247 getFileAux(const char *Filename, int64_t FileSize, bool RequiresNullTerminator,
248 bool IsVolatileSize) {
250 std::error_code EC = sys::fs::openFileForRead(Filename, FD);
254 ErrorOr<std::unique_ptr<MemoryBuffer>> Ret =
255 getOpenFileImpl(FD, Filename, FileSize, FileSize, 0,
256 RequiresNullTerminator, IsVolatileSize);
261 static bool shouldUseMmap(int FD,
265 bool RequiresNullTerminator,
267 bool IsVolatileSize) {
268 // mmap may leave the buffer without null terminator if the file size changed
269 // by the time the last page is mapped in, so avoid it if the file size is
274 // We don't use mmap for small files because this can severely fragment our
276 if (MapSize < 4 * 4096 || MapSize < (unsigned)PageSize)
279 if (!RequiresNullTerminator)
283 // If we don't know the file size, use fstat to find out. fstat on an open
284 // file descriptor is cheaper than stat on a random path.
285 // FIXME: this chunk of code is duplicated, but it avoids a fstat when
286 // RequiresNullTerminator = false and MapSize != -1.
287 if (FileSize == size_t(-1)) {
288 sys::fs::file_status Status;
289 if (sys::fs::status(FD, Status))
291 FileSize = Status.getSize();
294 // If we need a null terminator and the end of the map is inside the file,
295 // we cannot use mmap.
296 size_t End = Offset + MapSize;
297 assert(End <= FileSize);
301 // Don't try to map files that are exactly a multiple of the system page size
302 // if we need a null terminator.
303 if ((FileSize & (PageSize -1)) == 0)
306 #if defined(__CYGWIN__)
307 // Don't try to map files that are exactly a multiple of the physical page size
308 // if we need a null terminator.
309 // FIXME: We should reorganize again getPageSize() on Win32.
310 if ((FileSize & (4096 - 1)) == 0)
317 static ErrorOr<std::unique_ptr<MemoryBuffer>>
318 getOpenFileImpl(int FD, const char *Filename, uint64_t FileSize,
319 uint64_t MapSize, int64_t Offset, bool RequiresNullTerminator,
320 bool IsVolatileSize) {
321 static int PageSize = sys::process::get_self()->page_size();
323 // Default is to map the full file.
324 if (MapSize == uint64_t(-1)) {
325 // If we don't know the file size, use fstat to find out. fstat on an open
326 // file descriptor is cheaper than stat on a random path.
327 if (FileSize == uint64_t(-1)) {
328 sys::fs::file_status Status;
329 std::error_code EC = sys::fs::status(FD, Status);
333 // If this not a file or a block device (e.g. it's a named pipe
334 // or character device), we can't trust the size. Create the memory
335 // buffer by copying off the stream.
336 sys::fs::file_type Type = Status.type();
337 if (Type != sys::fs::file_type::regular_file &&
338 Type != sys::fs::file_type::block_file)
339 return getMemoryBufferForStream(FD, Filename);
341 FileSize = Status.getSize();
346 if (shouldUseMmap(FD, FileSize, MapSize, Offset, RequiresNullTerminator,
347 PageSize, IsVolatileSize)) {
349 std::unique_ptr<MemoryBuffer> Result(
350 new (NamedBufferAlloc(Filename))
351 MemoryBufferMMapFile(RequiresNullTerminator, FD, MapSize, Offset, EC));
353 return std::move(Result);
356 std::unique_ptr<MemoryBuffer> Buf =
357 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 char *BufPtr = const_cast<char *>(Buf->getBufferStart());
366 size_t BytesLeft = MapSize;
368 if (lseek(FD, Offset, SEEK_SET) == -1)
369 return std::error_code(errno, std::generic_category());
374 ssize_t NumRead = ::pread(FD, BufPtr, BytesLeft, MapSize-BytesLeft+Offset);
376 ssize_t NumRead = ::read(FD, BufPtr, BytesLeft);
381 // Error while reading.
382 return std::error_code(errno, std::generic_category());
385 memset(BufPtr, 0, BytesLeft); // zero-initialize rest of the buffer.
388 BytesLeft -= NumRead;
392 return std::move(Buf);
395 ErrorOr<std::unique_ptr<MemoryBuffer>>
396 MemoryBuffer::getOpenFile(int FD, const char *Filename, uint64_t FileSize,
397 bool RequiresNullTerminator, bool IsVolatileSize) {
398 return getOpenFileImpl(FD, Filename, FileSize, FileSize, 0,
399 RequiresNullTerminator, IsVolatileSize);
402 ErrorOr<std::unique_ptr<MemoryBuffer>>
403 MemoryBuffer::getOpenFileSlice(int FD, const char *Filename, uint64_t MapSize,
404 int64_t Offset, bool IsVolatileSize) {
405 return getOpenFileImpl(FD, Filename, -1, MapSize, Offset, false,
409 ErrorOr<std::unique_ptr<MemoryBuffer>> MemoryBuffer::getSTDIN() {
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>");
419 MemoryBufferRef MemoryBuffer::getMemBufferRef() const {
420 StringRef Data = getBuffer();
421 StringRef Identifier = getBufferIdentifier();
422 return MemoryBufferRef(Data, Identifier);