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"
28 #include <sys/types.h>
29 #include <system_error>
30 #if !defined(_MSC_VER) && !defined(__MINGW32__)
37 //===----------------------------------------------------------------------===//
38 // MemoryBuffer implementation itself.
39 //===----------------------------------------------------------------------===//
41 MemoryBuffer::~MemoryBuffer() { }
43 /// init - Initialize this MemoryBuffer as a reference to externally allocated
44 /// memory, memory that we know is already null terminated.
45 void MemoryBuffer::init(const char *BufStart, const char *BufEnd,
46 bool RequiresNullTerminator) {
47 assert((!RequiresNullTerminator || BufEnd[0] == 0) &&
48 "Buffer is not null terminated!");
49 BufferStart = BufStart;
53 //===----------------------------------------------------------------------===//
54 // MemoryBufferMem implementation.
55 //===----------------------------------------------------------------------===//
57 /// CopyStringRef - Copies contents of a StringRef into a block of memory and
58 /// null-terminates it.
59 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(const Twine &Name) : Name(Name) {}
72 void *operator new(size_t N, const NamedBufferAlloc &Alloc) {
73 SmallString<256> NameBuf;
74 StringRef NameRef = Alloc.Name.toStringRef(NameBuf);
76 char *Mem = static_cast<char *>(operator new(N + NameRef.size() + 1));
77 CopyStringRef(Mem + N, NameRef);
82 /// MemoryBufferMem - Named MemoryBuffer pointing to a block of memory.
84 class MemoryBufferMem : public MB {
86 MemoryBufferMem(StringRef InputData, bool RequiresNullTerminator) {
87 MemoryBuffer::init(InputData.begin(), InputData.end(),
88 RequiresNullTerminator);
91 /// Disable sized deallocation for MemoryBufferMem, because it has
92 /// tail-allocated data.
93 void operator delete(void *p) { ::operator delete(p); }
95 StringRef getBufferIdentifier() const override {
96 // The name is stored after the class itself.
97 return StringRef(reinterpret_cast<const char *>(this + 1));
100 MemoryBuffer::BufferKind getBufferKind() const override {
101 return MemoryBuffer::MemoryBuffer_Malloc;
106 template <typename MB>
107 static ErrorOr<std::unique_ptr<MB>>
108 getFileAux(const Twine &Filename, int64_t FileSize, uint64_t MapSize,
109 uint64_t Offset, bool RequiresNullTerminator, bool IsVolatile);
111 std::unique_ptr<MemoryBuffer>
112 MemoryBuffer::getMemBuffer(StringRef InputData, StringRef BufferName,
113 bool RequiresNullTerminator) {
114 auto *Ret = new (NamedBufferAlloc(BufferName))
115 MemoryBufferMem<MemoryBuffer>(InputData, RequiresNullTerminator);
116 return std::unique_ptr<MemoryBuffer>(Ret);
119 std::unique_ptr<MemoryBuffer>
120 MemoryBuffer::getMemBuffer(MemoryBufferRef Ref, bool RequiresNullTerminator) {
121 return std::unique_ptr<MemoryBuffer>(getMemBuffer(
122 Ref.getBuffer(), Ref.getBufferIdentifier(), RequiresNullTerminator));
125 static ErrorOr<std::unique_ptr<WritableMemoryBuffer>>
126 getMemBufferCopyImpl(StringRef InputData, const Twine &BufferName) {
127 auto Buf = WritableMemoryBuffer::getNewUninitMemBuffer(InputData.size(), BufferName);
129 return make_error_code(errc::not_enough_memory);
130 memcpy(Buf->getBufferStart(), InputData.data(), InputData.size());
131 return std::move(Buf);
134 std::unique_ptr<MemoryBuffer>
135 MemoryBuffer::getMemBufferCopy(StringRef InputData, const Twine &BufferName) {
136 auto Buf = getMemBufferCopyImpl(InputData, BufferName);
138 return std::move(*Buf);
142 std::unique_ptr<MemoryBuffer>
143 MemoryBuffer::getNewMemBuffer(size_t Size, StringRef BufferName) {
144 return WritableMemoryBuffer::getNewMemBuffer(Size, BufferName);
147 ErrorOr<std::unique_ptr<MemoryBuffer>>
148 MemoryBuffer::getFileOrSTDIN(const Twine &Filename, int64_t FileSize,
149 bool RequiresNullTerminator) {
150 SmallString<256> NameBuf;
151 StringRef NameRef = Filename.toStringRef(NameBuf);
155 return getFile(Filename, FileSize, RequiresNullTerminator);
158 ErrorOr<std::unique_ptr<MemoryBuffer>>
159 MemoryBuffer::getFileSlice(const Twine &FilePath, uint64_t MapSize,
160 uint64_t Offset, bool IsVolatile) {
161 return getFileAux<MemoryBuffer>(FilePath, -1, MapSize, Offset, false,
165 //===----------------------------------------------------------------------===//
166 // MemoryBuffer::getFile implementation.
167 //===----------------------------------------------------------------------===//
170 /// \brief Memory maps a file descriptor using sys::fs::mapped_file_region.
172 /// This handles converting the offset into a legal offset on the platform.
173 template<typename MB>
174 class MemoryBufferMMapFile : public MB {
175 sys::fs::mapped_file_region MFR;
177 static uint64_t getLegalMapOffset(uint64_t Offset) {
178 return Offset & ~(sys::fs::mapped_file_region::alignment() - 1);
181 static uint64_t getLegalMapSize(uint64_t Len, uint64_t Offset) {
182 return Len + (Offset - getLegalMapOffset(Offset));
185 const char *getStart(uint64_t Len, uint64_t Offset) {
186 return MFR.const_data() + (Offset - getLegalMapOffset(Offset));
190 MemoryBufferMMapFile(bool RequiresNullTerminator, int FD, uint64_t Len,
191 uint64_t Offset, std::error_code &EC)
193 MB::Writable ? sys::fs::mapped_file_region::priv
194 : sys::fs::mapped_file_region::readonly,
195 getLegalMapSize(Len, Offset), getLegalMapOffset(Offset), EC) {
197 const char *Start = getStart(Len, Offset);
198 MemoryBuffer::init(Start, Start + Len, RequiresNullTerminator);
202 /// Disable sized deallocation for MemoryBufferMMapFile, because it has
203 /// tail-allocated data.
204 void operator delete(void *p) { ::operator delete(p); }
206 StringRef getBufferIdentifier() const override {
207 // The name is stored after the class itself.
208 return StringRef(reinterpret_cast<const char *>(this + 1));
211 MemoryBuffer::BufferKind getBufferKind() const override {
212 return MemoryBuffer::MemoryBuffer_MMap;
217 static ErrorOr<std::unique_ptr<WritableMemoryBuffer>>
218 getMemoryBufferForStream(int FD, const Twine &BufferName) {
219 const ssize_t ChunkSize = 4096*4;
220 SmallString<ChunkSize> Buffer;
222 // Read into Buffer until we hit EOF.
224 Buffer.reserve(Buffer.size() + ChunkSize);
225 ReadBytes = sys::RetryAfterSignal(-1, read, FD, Buffer.end(), ChunkSize);
227 return std::error_code(errno, std::generic_category());
228 Buffer.set_size(Buffer.size() + ReadBytes);
229 } while (ReadBytes != 0);
231 return getMemBufferCopyImpl(Buffer, BufferName);
235 ErrorOr<std::unique_ptr<MemoryBuffer>>
236 MemoryBuffer::getFile(const Twine &Filename, int64_t FileSize,
237 bool RequiresNullTerminator, bool IsVolatile) {
238 return getFileAux<MemoryBuffer>(Filename, FileSize, FileSize, 0,
239 RequiresNullTerminator, IsVolatile);
242 template <typename MB>
243 static ErrorOr<std::unique_ptr<MB>>
244 getOpenFileImpl(int FD, const Twine &Filename, uint64_t FileSize,
245 uint64_t MapSize, int64_t Offset, bool RequiresNullTerminator,
248 template <typename MB>
249 static ErrorOr<std::unique_ptr<MB>>
250 getFileAux(const Twine &Filename, int64_t FileSize, uint64_t MapSize,
251 uint64_t Offset, bool RequiresNullTerminator, bool IsVolatile) {
253 std::error_code EC = sys::fs::openFileForRead(Filename, FD);
258 auto Ret = getOpenFileImpl<MB>(FD, Filename, FileSize, MapSize, Offset,
259 RequiresNullTerminator, IsVolatile);
264 ErrorOr<std::unique_ptr<WritableMemoryBuffer>>
265 WritableMemoryBuffer::getFile(const Twine &Filename, int64_t FileSize,
267 return getFileAux<WritableMemoryBuffer>(Filename, FileSize, FileSize, 0,
268 /*RequiresNullTerminator*/ false,
272 ErrorOr<std::unique_ptr<WritableMemoryBuffer>>
273 WritableMemoryBuffer::getFileSlice(const Twine &Filename, uint64_t MapSize,
274 uint64_t Offset, bool IsVolatile) {
275 return getFileAux<WritableMemoryBuffer>(Filename, -1, MapSize, Offset, false,
279 std::unique_ptr<WritableMemoryBuffer>
280 WritableMemoryBuffer::getNewUninitMemBuffer(size_t Size, const Twine &BufferName) {
281 using MemBuffer = MemoryBufferMem<WritableMemoryBuffer>;
282 // Allocate space for the MemoryBuffer, the data and the name. It is important
283 // that MemoryBuffer and data are aligned so PointerIntPair works with them.
284 // TODO: Is 16-byte alignment enough? We copy small object files with large
285 // alignment expectations into this buffer.
286 SmallString<256> NameBuf;
287 StringRef NameRef = BufferName.toStringRef(NameBuf);
288 size_t AlignedStringLen = alignTo(sizeof(MemBuffer) + NameRef.size() + 1, 16);
289 size_t RealLen = AlignedStringLen + Size + 1;
290 char *Mem = static_cast<char*>(operator new(RealLen, std::nothrow));
294 // The name is stored after the class itself.
295 CopyStringRef(Mem + sizeof(MemBuffer), NameRef);
297 // The buffer begins after the name and must be aligned.
298 char *Buf = Mem + AlignedStringLen;
299 Buf[Size] = 0; // Null terminate buffer.
301 auto *Ret = new (Mem) MemBuffer(StringRef(Buf, Size), true);
302 return std::unique_ptr<WritableMemoryBuffer>(Ret);
305 std::unique_ptr<WritableMemoryBuffer>
306 WritableMemoryBuffer::getNewMemBuffer(size_t Size, const Twine &BufferName) {
307 auto SB = WritableMemoryBuffer::getNewUninitMemBuffer(Size, BufferName);
310 memset(SB->getBufferStart(), 0, Size);
314 static bool shouldUseMmap(int FD,
318 bool RequiresNullTerminator,
321 // mmap may leave the buffer without null terminator if the file size changed
322 // by the time the last page is mapped in, so avoid it if the file size is
327 // We don't use mmap for small files because this can severely fragment our
329 if (MapSize < 4 * 4096 || MapSize < (unsigned)PageSize)
332 if (!RequiresNullTerminator)
335 // If we don't know the file size, use fstat to find out. fstat on an open
336 // file descriptor is cheaper than stat on a random path.
337 // FIXME: this chunk of code is duplicated, but it avoids a fstat when
338 // RequiresNullTerminator = false and MapSize != -1.
339 if (FileSize == size_t(-1)) {
340 sys::fs::file_status Status;
341 if (sys::fs::status(FD, Status))
343 FileSize = Status.getSize();
346 // If we need a null terminator and the end of the map is inside the file,
347 // we cannot use mmap.
348 size_t End = Offset + MapSize;
349 assert(End <= FileSize);
353 // Don't try to map files that are exactly a multiple of the system page size
354 // if we need a null terminator.
355 if ((FileSize & (PageSize -1)) == 0)
358 #if defined(__CYGWIN__)
359 // Don't try to map files that are exactly a multiple of the physical page size
360 // if we need a null terminator.
361 // FIXME: We should reorganize again getPageSize() on Win32.
362 if ((FileSize & (4096 - 1)) == 0)
369 template <typename MB>
370 static ErrorOr<std::unique_ptr<MB>>
371 getOpenFileImpl(int FD, const Twine &Filename, uint64_t FileSize,
372 uint64_t MapSize, int64_t Offset, bool RequiresNullTerminator,
374 static int PageSize = sys::Process::getPageSize();
376 // Default is to map the full file.
377 if (MapSize == uint64_t(-1)) {
378 // If we don't know the file size, use fstat to find out. fstat on an open
379 // file descriptor is cheaper than stat on a random path.
380 if (FileSize == uint64_t(-1)) {
381 sys::fs::file_status Status;
382 std::error_code EC = sys::fs::status(FD, Status);
386 // If this not a file or a block device (e.g. it's a named pipe
387 // or character device), we can't trust the size. Create the memory
388 // buffer by copying off the stream.
389 sys::fs::file_type Type = Status.type();
390 if (Type != sys::fs::file_type::regular_file &&
391 Type != sys::fs::file_type::block_file)
392 return getMemoryBufferForStream(FD, Filename);
394 FileSize = Status.getSize();
399 if (shouldUseMmap(FD, FileSize, MapSize, Offset, RequiresNullTerminator,
400 PageSize, IsVolatile)) {
402 std::unique_ptr<MB> Result(
403 new (NamedBufferAlloc(Filename)) MemoryBufferMMapFile<MB>(
404 RequiresNullTerminator, FD, MapSize, Offset, EC));
406 return std::move(Result);
409 auto Buf = WritableMemoryBuffer::getNewUninitMemBuffer(MapSize, Filename);
411 // Failed to create a buffer. The only way it can fail is if
412 // new(std::nothrow) returns 0.
413 return make_error_code(errc::not_enough_memory);
416 char *BufPtr = Buf.get()->getBufferStart();
418 size_t BytesLeft = MapSize;
420 if (lseek(FD, Offset, SEEK_SET) == -1)
421 return std::error_code(errno, std::generic_category());
426 ssize_t NumRead = sys::RetryAfterSignal(-1, ::pread, FD, BufPtr, BytesLeft,
427 MapSize - BytesLeft + Offset);
429 ssize_t NumRead = sys::RetryAfterSignal(-1, ::read, FD, BufPtr, BytesLeft);
432 // Error while reading.
433 return std::error_code(errno, std::generic_category());
436 memset(BufPtr, 0, BytesLeft); // zero-initialize rest of the buffer.
439 BytesLeft -= NumRead;
443 return std::move(Buf);
446 ErrorOr<std::unique_ptr<MemoryBuffer>>
447 MemoryBuffer::getOpenFile(int FD, const Twine &Filename, uint64_t FileSize,
448 bool RequiresNullTerminator, bool IsVolatile) {
449 return getOpenFileImpl<MemoryBuffer>(FD, Filename, FileSize, FileSize, 0,
450 RequiresNullTerminator, IsVolatile);
453 ErrorOr<std::unique_ptr<MemoryBuffer>>
454 MemoryBuffer::getOpenFileSlice(int FD, const Twine &Filename, uint64_t MapSize,
455 int64_t Offset, bool IsVolatile) {
456 assert(MapSize != uint64_t(-1));
457 return getOpenFileImpl<MemoryBuffer>(FD, Filename, -1, MapSize, Offset, false,
461 ErrorOr<std::unique_ptr<MemoryBuffer>> MemoryBuffer::getSTDIN() {
462 // Read in all of the data from stdin, we cannot mmap stdin.
464 // FIXME: That isn't necessarily true, we should try to mmap stdin and
465 // fallback if it fails.
466 sys::ChangeStdinToBinary();
468 return getMemoryBufferForStream(0, "<stdin>");
471 ErrorOr<std::unique_ptr<MemoryBuffer>>
472 MemoryBuffer::getFileAsStream(const Twine &Filename) {
474 std::error_code EC = sys::fs::openFileForRead(Filename, FD);
477 ErrorOr<std::unique_ptr<MemoryBuffer>> Ret =
478 getMemoryBufferForStream(FD, Filename);
483 MemoryBufferRef MemoryBuffer::getMemBufferRef() const {
484 StringRef Data = getBuffer();
485 StringRef Identifier = getBufferIdentifier();
486 return MemoryBufferRef(Data, Identifier);