1 //===--- MemoryBuffer.cpp - Memory Buffer implementation ------------------===//
2 //
3 // The LLVM Compiler Infrastructure
4 //
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
7 //
8 //===----------------------------------------------------------------------===//
9 //
10 // This file implements the MemoryBuffer interface.
11 //
12 //===----------------------------------------------------------------------===//
13
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"
24 #include <cassert>
25 #include <cerrno>
26 #include <cstring>
27 #include <new>
28 #include <sys/types.h>
29 #include <system_error>
30 #if !defined(_MSC_VER) && !defined(__MINGW32__)
31 #include <unistd.h>
32 #else
33 #include <io.h>
34 #endif
35 using namespace llvm;
36
37 //===----------------------------------------------------------------------===//
38 // MemoryBuffer implementation itself.
39 //===----------------------------------------------------------------------===//
40
~MemoryBuffer()41 MemoryBuffer::~MemoryBuffer() { }
42
43 /// init - Initialize this MemoryBuffer as a reference to externally allocated
44 /// memory, memory that we know is already null terminated.
init(const char * BufStart,const char * BufEnd,bool RequiresNullTerminator)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;
50 BufferEnd = BufEnd;
51 }
52
53 //===----------------------------------------------------------------------===//
54 // MemoryBufferMem implementation.
55 //===----------------------------------------------------------------------===//
56
57 /// CopyStringRef - Copies contents of a StringRef into a block of memory and
58 /// null-terminates it.
CopyStringRef(char * Memory,StringRef Data)59 static void CopyStringRef(char *Memory, StringRef Data) {
60 if (!Data.empty())
61 memcpy(Memory, Data.data(), Data.size());
62 Memory[Data.size()] = 0; // Null terminate string.
63 }
64
65 namespace {
66 struct NamedBufferAlloc {
67 const Twine &Name;
NamedBufferAlloc__anon5fae06310111::NamedBufferAlloc68 NamedBufferAlloc(const Twine &Name) : Name(Name) {}
69 };
70 }
71
operator new(size_t N,const NamedBufferAlloc & Alloc)72 void *operator new(size_t N, const NamedBufferAlloc &Alloc) {
73 SmallString<256> NameBuf;
74 StringRef NameRef = Alloc.Name.toStringRef(NameBuf);
75
76 char *Mem = static_cast<char *>(operator new(N + NameRef.size() + 1));
77 CopyStringRef(Mem + N, NameRef);
78 return Mem;
79 }
80
81 namespace {
82 /// MemoryBufferMem - Named MemoryBuffer pointing to a block of memory.
83 class MemoryBufferMem : public MemoryBuffer {
84 public:
MemoryBufferMem(StringRef InputData,bool RequiresNullTerminator)85 MemoryBufferMem(StringRef InputData, bool RequiresNullTerminator) {
86 init(InputData.begin(), InputData.end(), RequiresNullTerminator);
87 }
88
getBufferIdentifier() const89 const char *getBufferIdentifier() const override {
90 // The name is stored after the class itself.
91 return reinterpret_cast<const char*>(this + 1);
92 }
93
getBufferKind() const94 BufferKind getBufferKind() const override {
95 return MemoryBuffer_Malloc;
96 }
97 };
98 }
99
100 static ErrorOr<std::unique_ptr<MemoryBuffer>>
101 getFileAux(const Twine &Filename, int64_t FileSize, uint64_t MapSize,
102 uint64_t Offset, bool RequiresNullTerminator, bool IsVolatileSize);
103
104 std::unique_ptr<MemoryBuffer>
getMemBuffer(StringRef InputData,StringRef BufferName,bool RequiresNullTerminator)105 MemoryBuffer::getMemBuffer(StringRef InputData, StringRef BufferName,
106 bool RequiresNullTerminator) {
107 auto *Ret = new (NamedBufferAlloc(BufferName))
108 MemoryBufferMem(InputData, RequiresNullTerminator);
109 return std::unique_ptr<MemoryBuffer>(Ret);
110 }
111
112 std::unique_ptr<MemoryBuffer>
getMemBuffer(MemoryBufferRef Ref,bool RequiresNullTerminator)113 MemoryBuffer::getMemBuffer(MemoryBufferRef Ref, bool RequiresNullTerminator) {
114 return std::unique_ptr<MemoryBuffer>(getMemBuffer(
115 Ref.getBuffer(), Ref.getBufferIdentifier(), RequiresNullTerminator));
116 }
117
118 std::unique_ptr<MemoryBuffer>
getMemBufferCopy(StringRef InputData,const Twine & BufferName)119 MemoryBuffer::getMemBufferCopy(StringRef InputData, const Twine &BufferName) {
120 std::unique_ptr<MemoryBuffer> Buf =
121 getNewUninitMemBuffer(InputData.size(), BufferName);
122 if (!Buf)
123 return nullptr;
124 memcpy(const_cast<char*>(Buf->getBufferStart()), InputData.data(),
125 InputData.size());
126 return Buf;
127 }
128
129 std::unique_ptr<MemoryBuffer>
getNewUninitMemBuffer(size_t Size,const Twine & BufferName)130 MemoryBuffer::getNewUninitMemBuffer(size_t Size, const Twine &BufferName) {
131 // Allocate space for the MemoryBuffer, the data and the name. It is important
132 // that MemoryBuffer and data are aligned so PointerIntPair works with them.
133 // TODO: Is 16-byte alignment enough? We copy small object files with large
134 // alignment expectations into this buffer.
135 SmallString<256> NameBuf;
136 StringRef NameRef = BufferName.toStringRef(NameBuf);
137 size_t AlignedStringLen =
138 RoundUpToAlignment(sizeof(MemoryBufferMem) + NameRef.size() + 1, 16);
139 size_t RealLen = AlignedStringLen + Size + 1;
140 char *Mem = static_cast<char*>(operator new(RealLen, std::nothrow));
141 if (!Mem)
142 return nullptr;
143
144 // The name is stored after the class itself.
145 CopyStringRef(Mem + sizeof(MemoryBufferMem), NameRef);
146
147 // The buffer begins after the name and must be aligned.
148 char *Buf = Mem + AlignedStringLen;
149 Buf[Size] = 0; // Null terminate buffer.
150
151 auto *Ret = new (Mem) MemoryBufferMem(StringRef(Buf, Size), true);
152 return std::unique_ptr<MemoryBuffer>(Ret);
153 }
154
155 std::unique_ptr<MemoryBuffer>
getNewMemBuffer(size_t Size,StringRef BufferName)156 MemoryBuffer::getNewMemBuffer(size_t Size, StringRef BufferName) {
157 std::unique_ptr<MemoryBuffer> SB = getNewUninitMemBuffer(Size, BufferName);
158 if (!SB)
159 return nullptr;
160 memset(const_cast<char*>(SB->getBufferStart()), 0, Size);
161 return SB;
162 }
163
164 ErrorOr<std::unique_ptr<MemoryBuffer>>
getFileOrSTDIN(const Twine & Filename,int64_t FileSize,bool RequiresNullTerminator)165 MemoryBuffer::getFileOrSTDIN(const Twine &Filename, int64_t FileSize,
166 bool RequiresNullTerminator) {
167 SmallString<256> NameBuf;
168 StringRef NameRef = Filename.toStringRef(NameBuf);
169
170 if (NameRef == "-")
171 return getSTDIN();
172 return getFile(Filename, FileSize, RequiresNullTerminator);
173 }
174
175 ErrorOr<std::unique_ptr<MemoryBuffer>>
getFileSlice(const Twine & FilePath,uint64_t MapSize,uint64_t Offset)176 MemoryBuffer::getFileSlice(const Twine &FilePath, uint64_t MapSize,
177 uint64_t Offset) {
178 return getFileAux(FilePath, -1, MapSize, Offset, false, false);
179 }
180
181
182 //===----------------------------------------------------------------------===//
183 // MemoryBuffer::getFile implementation.
184 //===----------------------------------------------------------------------===//
185
186 namespace {
187 /// \brief Memory maps a file descriptor using sys::fs::mapped_file_region.
188 ///
189 /// This handles converting the offset into a legal offset on the platform.
190 class MemoryBufferMMapFile : public MemoryBuffer {
191 sys::fs::mapped_file_region MFR;
192
getLegalMapOffset(uint64_t Offset)193 static uint64_t getLegalMapOffset(uint64_t Offset) {
194 return Offset & ~(sys::fs::mapped_file_region::alignment() - 1);
195 }
196
getLegalMapSize(uint64_t Len,uint64_t Offset)197 static uint64_t getLegalMapSize(uint64_t Len, uint64_t Offset) {
198 return Len + (Offset - getLegalMapOffset(Offset));
199 }
200
getStart(uint64_t Len,uint64_t Offset)201 const char *getStart(uint64_t Len, uint64_t Offset) {
202 return MFR.const_data() + (Offset - getLegalMapOffset(Offset));
203 }
204
205 public:
MemoryBufferMMapFile(bool RequiresNullTerminator,int FD,uint64_t Len,uint64_t Offset,std::error_code & EC)206 MemoryBufferMMapFile(bool RequiresNullTerminator, int FD, uint64_t Len,
207 uint64_t Offset, std::error_code &EC)
208 : MFR(FD, sys::fs::mapped_file_region::readonly,
209 getLegalMapSize(Len, Offset), getLegalMapOffset(Offset), EC) {
210 if (!EC) {
211 const char *Start = getStart(Len, Offset);
212 init(Start, Start + Len, RequiresNullTerminator);
213 }
214 }
215
getBufferIdentifier() const216 const char *getBufferIdentifier() const override {
217 // The name is stored after the class itself.
218 return reinterpret_cast<const char *>(this + 1);
219 }
220
getBufferKind() const221 BufferKind getBufferKind() const override {
222 return MemoryBuffer_MMap;
223 }
224 };
225 }
226
227 static ErrorOr<std::unique_ptr<MemoryBuffer>>
getMemoryBufferForStream(int FD,const Twine & BufferName)228 getMemoryBufferForStream(int FD, const Twine &BufferName) {
229 const ssize_t ChunkSize = 4096*4;
230 SmallString<ChunkSize> Buffer;
231 ssize_t ReadBytes;
232 // Read into Buffer until we hit EOF.
233 do {
234 Buffer.reserve(Buffer.size() + ChunkSize);
235 ReadBytes = read(FD, Buffer.end(), ChunkSize);
236 if (ReadBytes == -1) {
237 if (errno == EINTR) continue;
238 return std::error_code(errno, std::generic_category());
239 }
240 Buffer.set_size(Buffer.size() + ReadBytes);
241 } while (ReadBytes != 0);
242
243 return MemoryBuffer::getMemBufferCopy(Buffer, BufferName);
244 }
245
246
247 ErrorOr<std::unique_ptr<MemoryBuffer>>
getFile(const Twine & Filename,int64_t FileSize,bool RequiresNullTerminator,bool IsVolatileSize)248 MemoryBuffer::getFile(const Twine &Filename, int64_t FileSize,
249 bool RequiresNullTerminator, bool IsVolatileSize) {
250 return getFileAux(Filename, FileSize, FileSize, 0,
251 RequiresNullTerminator, IsVolatileSize);
252 }
253
254 static ErrorOr<std::unique_ptr<MemoryBuffer>>
255 getOpenFileImpl(int FD, const Twine &Filename, uint64_t FileSize,
256 uint64_t MapSize, int64_t Offset, bool RequiresNullTerminator,
257 bool IsVolatileSize);
258
259 static ErrorOr<std::unique_ptr<MemoryBuffer>>
getFileAux(const Twine & Filename,int64_t FileSize,uint64_t MapSize,uint64_t Offset,bool RequiresNullTerminator,bool IsVolatileSize)260 getFileAux(const Twine &Filename, int64_t FileSize, uint64_t MapSize,
261 uint64_t Offset, bool RequiresNullTerminator, bool IsVolatileSize) {
262 int FD;
263 std::error_code EC = sys::fs::openFileForRead(Filename, FD);
264 if (EC)
265 return EC;
266
267 ErrorOr<std::unique_ptr<MemoryBuffer>> Ret =
268 getOpenFileImpl(FD, Filename, FileSize, MapSize, Offset,
269 RequiresNullTerminator, IsVolatileSize);
270 close(FD);
271 return Ret;
272 }
273
shouldUseMmap(int FD,size_t FileSize,size_t MapSize,off_t Offset,bool RequiresNullTerminator,int PageSize,bool IsVolatileSize)274 static bool shouldUseMmap(int FD,
275 size_t FileSize,
276 size_t MapSize,
277 off_t Offset,
278 bool RequiresNullTerminator,
279 int PageSize,
280 bool IsVolatileSize) {
281 // mmap may leave the buffer without null terminator if the file size changed
282 // by the time the last page is mapped in, so avoid it if the file size is
283 // likely to change.
284 if (IsVolatileSize)
285 return false;
286
287 // We don't use mmap for small files because this can severely fragment our
288 // address space.
289 if (MapSize < 4 * 4096 || MapSize < (unsigned)PageSize)
290 return false;
291
292 if (!RequiresNullTerminator)
293 return true;
294
295
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 sys::fs::file_status Status;
302 if (sys::fs::status(FD, Status))
303 return false;
304 FileSize = Status.getSize();
305 }
306
307 // If we need a null terminator and the end of the map is inside the file,
308 // we cannot use mmap.
309 size_t End = Offset + MapSize;
310 assert(End <= FileSize);
311 if (End != FileSize)
312 return false;
313
314 // Don't try to map files that are exactly a multiple of the system page size
315 // if we need a null terminator.
316 if ((FileSize & (PageSize -1)) == 0)
317 return false;
318
319 #if defined(__CYGWIN__)
320 // Don't try to map files that are exactly a multiple of the physical page size
321 // if we need a null terminator.
322 // FIXME: We should reorganize again getPageSize() on Win32.
323 if ((FileSize & (4096 - 1)) == 0)
324 return false;
325 #endif
326
327 return true;
328 }
329
330 static ErrorOr<std::unique_ptr<MemoryBuffer>>
getOpenFileImpl(int FD,const Twine & Filename,uint64_t FileSize,uint64_t MapSize,int64_t Offset,bool RequiresNullTerminator,bool IsVolatileSize)331 getOpenFileImpl(int FD, const Twine &Filename, uint64_t FileSize,
332 uint64_t MapSize, int64_t Offset, bool RequiresNullTerminator,
333 bool IsVolatileSize) {
334 static int PageSize = sys::Process::getPageSize();
335
336 // Default is to map the full file.
337 if (MapSize == uint64_t(-1)) {
338 // If we don't know the file size, use fstat to find out. fstat on an open
339 // file descriptor is cheaper than stat on a random path.
340 if (FileSize == uint64_t(-1)) {
341 sys::fs::file_status Status;
342 std::error_code EC = sys::fs::status(FD, Status);
343 if (EC)
344 return EC;
345
346 // If this not a file or a block device (e.g. it's a named pipe
347 // or character device), we can't trust the size. Create the memory
348 // buffer by copying off the stream.
349 sys::fs::file_type Type = Status.type();
350 if (Type != sys::fs::file_type::regular_file &&
351 Type != sys::fs::file_type::block_file)
352 return getMemoryBufferForStream(FD, Filename);
353
354 FileSize = Status.getSize();
355 }
356 MapSize = FileSize;
357 }
358
359 if (shouldUseMmap(FD, FileSize, MapSize, Offset, RequiresNullTerminator,
360 PageSize, IsVolatileSize)) {
361 std::error_code EC;
362 std::unique_ptr<MemoryBuffer> Result(
363 new (NamedBufferAlloc(Filename))
364 MemoryBufferMMapFile(RequiresNullTerminator, FD, MapSize, Offset, EC));
365 if (!EC)
366 return std::move(Result);
367 }
368
369 std::unique_ptr<MemoryBuffer> Buf =
370 MemoryBuffer::getNewUninitMemBuffer(MapSize, Filename);
371 if (!Buf) {
372 // Failed to create a buffer. The only way it can fail is if
373 // new(std::nothrow) returns 0.
374 return make_error_code(errc::not_enough_memory);
375 }
376
377 char *BufPtr = const_cast<char *>(Buf->getBufferStart());
378
379 size_t BytesLeft = MapSize;
380 #ifndef HAVE_PREAD
381 if (lseek(FD, Offset, SEEK_SET) == -1)
382 return std::error_code(errno, std::generic_category());
383 #endif
384
385 while (BytesLeft) {
386 #ifdef HAVE_PREAD
387 ssize_t NumRead = ::pread(FD, BufPtr, BytesLeft, MapSize-BytesLeft+Offset);
388 #else
389 ssize_t NumRead = ::read(FD, BufPtr, BytesLeft);
390 #endif
391 if (NumRead == -1) {
392 if (errno == EINTR)
393 continue;
394 // Error while reading.
395 return std::error_code(errno, std::generic_category());
396 }
397 if (NumRead == 0) {
398 memset(BufPtr, 0, BytesLeft); // zero-initialize rest of the buffer.
399 break;
400 }
401 BytesLeft -= NumRead;
402 BufPtr += NumRead;
403 }
404
405 return std::move(Buf);
406 }
407
408 ErrorOr<std::unique_ptr<MemoryBuffer>>
getOpenFile(int FD,const Twine & Filename,uint64_t FileSize,bool RequiresNullTerminator,bool IsVolatileSize)409 MemoryBuffer::getOpenFile(int FD, const Twine &Filename, uint64_t FileSize,
410 bool RequiresNullTerminator, bool IsVolatileSize) {
411 return getOpenFileImpl(FD, Filename, FileSize, FileSize, 0,
412 RequiresNullTerminator, IsVolatileSize);
413 }
414
415 ErrorOr<std::unique_ptr<MemoryBuffer>>
getOpenFileSlice(int FD,const Twine & Filename,uint64_t MapSize,int64_t Offset)416 MemoryBuffer::getOpenFileSlice(int FD, const Twine &Filename, uint64_t MapSize,
417 int64_t Offset) {
418 assert(MapSize != uint64_t(-1));
419 return getOpenFileImpl(FD, Filename, -1, MapSize, Offset, false,
420 /*IsVolatileSize*/ false);
421 }
422
getSTDIN()423 ErrorOr<std::unique_ptr<MemoryBuffer>> MemoryBuffer::getSTDIN() {
424 // Read in all of the data from stdin, we cannot mmap stdin.
425 //
426 // FIXME: That isn't necessarily true, we should try to mmap stdin and
427 // fallback if it fails.
428 sys::ChangeStdinToBinary();
429
430 return getMemoryBufferForStream(0, "<stdin>");
431 }
432
getMemBufferRef() const433 MemoryBufferRef MemoryBuffer::getMemBufferRef() const {
434 StringRef Data = getBuffer();
435 StringRef Identifier = getBufferIdentifier();
436 return MemoryBufferRef(Data, Identifier);
437 }
438