xref: /llvm-project/llvm/lib/Support/MemoryBuffer.cpp (revision 890e8c8f7e9e448a870224068d6b4181d96a293d)
1 //===--- MemoryBuffer.cpp - Memory Buffer implementation ------------------===//
2 //
3 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
4 // See https://llvm.org/LICENSE.txt for license information.
5 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
6 //
7 //===----------------------------------------------------------------------===//
8 //
9 //  This file implements the MemoryBuffer interface.
10 //
11 //===----------------------------------------------------------------------===//
12 
13 #include "llvm/Support/MemoryBuffer.h"
14 #include "llvm/ADT/SmallString.h"
15 #include "llvm/Config/config.h"
16 #include "llvm/Support/AutoConvert.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 "llvm/Support/SmallVectorMemoryBuffer.h"
25 #include <cassert>
26 #include <cerrno>
27 #include <cstring>
28 #include <new>
29 #include <sys/types.h>
30 #include <system_error>
31 #if !defined(_MSC_VER) && !defined(__MINGW32__)
32 #include <unistd.h>
33 #else
34 #include <io.h>
35 #endif
36 using namespace llvm;
37 
38 //===----------------------------------------------------------------------===//
39 // MemoryBuffer implementation itself.
40 //===----------------------------------------------------------------------===//
41 
42 MemoryBuffer::~MemoryBuffer() { }
43 
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;
51   BufferEnd = BufEnd;
52 }
53 
54 //===----------------------------------------------------------------------===//
55 // MemoryBufferMem implementation.
56 //===----------------------------------------------------------------------===//
57 
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   if (!Data.empty())
62     memcpy(Memory, Data.data(), Data.size());
63   Memory[Data.size()] = 0; // Null terminate string.
64 }
65 
66 namespace {
67 struct NamedBufferAlloc {
68   const Twine &Name;
69   NamedBufferAlloc(const Twine &Name) : Name(Name) {}
70 };
71 } // namespace
72 
73 void *operator new(size_t N, const NamedBufferAlloc &Alloc) {
74   SmallString<256> NameBuf;
75   StringRef NameRef = Alloc.Name.toStringRef(NameBuf);
76 
77   char *Mem = static_cast<char *>(operator new(N + NameRef.size() + 1));
78   CopyStringRef(Mem + N, NameRef);
79   return Mem;
80 }
81 
82 namespace {
83 /// MemoryBufferMem - Named MemoryBuffer pointing to a block of memory.
84 template<typename MB>
85 class MemoryBufferMem : public MB {
86 public:
87   MemoryBufferMem(StringRef InputData, bool RequiresNullTerminator) {
88     MemoryBuffer::init(InputData.begin(), InputData.end(),
89                        RequiresNullTerminator);
90   }
91 
92   /// Disable sized deallocation for MemoryBufferMem, because it has
93   /// tail-allocated data.
94   void operator delete(void *p) { ::operator delete(p); }
95 
96   StringRef getBufferIdentifier() const override {
97     // The name is stored after the class itself.
98     return StringRef(reinterpret_cast<const char *>(this + 1));
99   }
100 
101   MemoryBuffer::BufferKind getBufferKind() const override {
102     return MemoryBuffer::MemoryBuffer_Malloc;
103   }
104 };
105 } // namespace
106 
107 template <typename MB>
108 static ErrorOr<std::unique_ptr<MB>>
109 getFileAux(const Twine &Filename, uint64_t MapSize, uint64_t Offset,
110            bool IsText, bool RequiresNullTerminator, bool IsVolatile);
111 
112 std::unique_ptr<MemoryBuffer>
113 MemoryBuffer::getMemBuffer(StringRef InputData, StringRef BufferName,
114                            bool RequiresNullTerminator) {
115   auto *Ret = new (NamedBufferAlloc(BufferName))
116       MemoryBufferMem<MemoryBuffer>(InputData, RequiresNullTerminator);
117   return std::unique_ptr<MemoryBuffer>(Ret);
118 }
119 
120 std::unique_ptr<MemoryBuffer>
121 MemoryBuffer::getMemBuffer(MemoryBufferRef Ref, bool RequiresNullTerminator) {
122   return std::unique_ptr<MemoryBuffer>(getMemBuffer(
123       Ref.getBuffer(), Ref.getBufferIdentifier(), RequiresNullTerminator));
124 }
125 
126 static ErrorOr<std::unique_ptr<WritableMemoryBuffer>>
127 getMemBufferCopyImpl(StringRef InputData, const Twine &BufferName) {
128   auto Buf = WritableMemoryBuffer::getNewUninitMemBuffer(InputData.size(), BufferName);
129   if (!Buf)
130     return make_error_code(errc::not_enough_memory);
131   memcpy(Buf->getBufferStart(), InputData.data(), InputData.size());
132   return std::move(Buf);
133 }
134 
135 std::unique_ptr<MemoryBuffer>
136 MemoryBuffer::getMemBufferCopy(StringRef InputData, const Twine &BufferName) {
137   auto Buf = getMemBufferCopyImpl(InputData, BufferName);
138   if (Buf)
139     return std::move(*Buf);
140   return nullptr;
141 }
142 
143 ErrorOr<std::unique_ptr<MemoryBuffer>>
144 MemoryBuffer::getFileOrSTDIN(const Twine &Filename, bool IsText,
145                              bool RequiresNullTerminator) {
146   SmallString<256> NameBuf;
147   StringRef NameRef = Filename.toStringRef(NameBuf);
148 
149   if (NameRef == "-")
150     return getSTDIN();
151   return getFile(Filename, IsText, RequiresNullTerminator,
152                  /*IsVolatile=*/false);
153 }
154 
155 ErrorOr<std::unique_ptr<MemoryBuffer>>
156 MemoryBuffer::getFileSlice(const Twine &FilePath, uint64_t MapSize,
157                            uint64_t Offset, bool IsVolatile) {
158   return getFileAux<MemoryBuffer>(FilePath, MapSize, Offset, /*IsText=*/false,
159                                   /*RequiresNullTerminator=*/false, IsVolatile);
160 }
161 
162 //===----------------------------------------------------------------------===//
163 // MemoryBuffer::getFile implementation.
164 //===----------------------------------------------------------------------===//
165 
166 namespace {
167 
168 template <typename MB>
169 constexpr sys::fs::mapped_file_region::mapmode Mapmode =
170     sys::fs::mapped_file_region::readonly;
171 template <>
172 constexpr sys::fs::mapped_file_region::mapmode Mapmode<MemoryBuffer> =
173     sys::fs::mapped_file_region::readonly;
174 template <>
175 constexpr sys::fs::mapped_file_region::mapmode Mapmode<WritableMemoryBuffer> =
176     sys::fs::mapped_file_region::priv;
177 template <>
178 constexpr sys::fs::mapped_file_region::mapmode
179     Mapmode<WriteThroughMemoryBuffer> = sys::fs::mapped_file_region::readwrite;
180 
181 /// Memory maps a file descriptor using sys::fs::mapped_file_region.
182 ///
183 /// This handles converting the offset into a legal offset on the platform.
184 template<typename MB>
185 class MemoryBufferMMapFile : public MB {
186   sys::fs::mapped_file_region MFR;
187 
188   static uint64_t getLegalMapOffset(uint64_t Offset) {
189     return Offset & ~(sys::fs::mapped_file_region::alignment() - 1);
190   }
191 
192   static uint64_t getLegalMapSize(uint64_t Len, uint64_t Offset) {
193     return Len + (Offset - getLegalMapOffset(Offset));
194   }
195 
196   const char *getStart(uint64_t Len, uint64_t Offset) {
197     return MFR.const_data() + (Offset - getLegalMapOffset(Offset));
198   }
199 
200 public:
201   MemoryBufferMMapFile(bool RequiresNullTerminator, sys::fs::file_t FD, uint64_t Len,
202                        uint64_t Offset, std::error_code &EC)
203       : MFR(FD, Mapmode<MB>, getLegalMapSize(Len, Offset),
204             getLegalMapOffset(Offset), EC) {
205     if (!EC) {
206       const char *Start = getStart(Len, Offset);
207       MemoryBuffer::init(Start, Start + Len, RequiresNullTerminator);
208     }
209   }
210 
211   /// Disable sized deallocation for MemoryBufferMMapFile, because it has
212   /// tail-allocated data.
213   void operator delete(void *p) { ::operator delete(p); }
214 
215   StringRef getBufferIdentifier() const override {
216     // The name is stored after the class itself.
217     return StringRef(reinterpret_cast<const char *>(this + 1));
218   }
219 
220   MemoryBuffer::BufferKind getBufferKind() const override {
221     return MemoryBuffer::MemoryBuffer_MMap;
222   }
223 
224   void dontNeedIfMmap() override { MFR.dontNeed(); }
225 };
226 } // namespace
227 
228 static ErrorOr<std::unique_ptr<WritableMemoryBuffer>>
229 getMemoryBufferForStream(sys::fs::file_t FD, const Twine &BufferName) {
230   const ssize_t ChunkSize = 4096*4;
231   SmallString<ChunkSize> Buffer;
232 
233   // Read into Buffer until we hit EOF.
234   size_t Size = Buffer.size();
235   for (;;) {
236     Buffer.resize_for_overwrite(Size + ChunkSize);
237     Expected<size_t> ReadBytes = sys::fs::readNativeFile(
238         FD, makeMutableArrayRef(Buffer.begin() + Size, ChunkSize));
239     if (!ReadBytes)
240       return errorToErrorCode(ReadBytes.takeError());
241     if (*ReadBytes == 0)
242       break;
243     Size += *ReadBytes;
244   }
245   Buffer.truncate(Size);
246 
247   return getMemBufferCopyImpl(Buffer, BufferName);
248 }
249 
250 ErrorOr<std::unique_ptr<MemoryBuffer>>
251 MemoryBuffer::getFile(const Twine &Filename, bool IsText,
252                       bool RequiresNullTerminator, bool IsVolatile) {
253   return getFileAux<MemoryBuffer>(Filename, /*MapSize=*/-1, /*Offset=*/0,
254                                   IsText, RequiresNullTerminator, IsVolatile);
255 }
256 
257 template <typename MB>
258 static ErrorOr<std::unique_ptr<MB>>
259 getOpenFileImpl(sys::fs::file_t FD, const Twine &Filename, uint64_t FileSize,
260                 uint64_t MapSize, int64_t Offset, bool RequiresNullTerminator,
261                 bool IsVolatile);
262 
263 template <typename MB>
264 static ErrorOr<std::unique_ptr<MB>>
265 getFileAux(const Twine &Filename, uint64_t MapSize, uint64_t Offset,
266            bool IsText, bool RequiresNullTerminator, bool IsVolatile) {
267   Expected<sys::fs::file_t> FDOrErr = sys::fs::openNativeFileForRead(
268       Filename, IsText ? sys::fs::OF_TextWithCRLF : sys::fs::OF_None);
269   if (!FDOrErr)
270     return errorToErrorCode(FDOrErr.takeError());
271   sys::fs::file_t FD = *FDOrErr;
272   auto Ret = getOpenFileImpl<MB>(FD, Filename, /*FileSize=*/-1, MapSize, Offset,
273                                  RequiresNullTerminator, IsVolatile);
274   sys::fs::closeFile(FD);
275   return Ret;
276 }
277 
278 ErrorOr<std::unique_ptr<WritableMemoryBuffer>>
279 WritableMemoryBuffer::getFile(const Twine &Filename, bool IsVolatile) {
280   return getFileAux<WritableMemoryBuffer>(
281       Filename, /*MapSize=*/-1, /*Offset=*/0, /*IsText=*/false,
282       /*RequiresNullTerminator=*/false, IsVolatile);
283 }
284 
285 ErrorOr<std::unique_ptr<WritableMemoryBuffer>>
286 WritableMemoryBuffer::getFileSlice(const Twine &Filename, uint64_t MapSize,
287                                    uint64_t Offset, bool IsVolatile) {
288   return getFileAux<WritableMemoryBuffer>(
289       Filename, MapSize, Offset, /*IsText=*/false,
290       /*RequiresNullTerminator=*/false, IsVolatile);
291 }
292 
293 std::unique_ptr<WritableMemoryBuffer>
294 WritableMemoryBuffer::getNewUninitMemBuffer(size_t Size, const Twine &BufferName) {
295   using MemBuffer = MemoryBufferMem<WritableMemoryBuffer>;
296   // Allocate space for the MemoryBuffer, the data and the name. It is important
297   // that MemoryBuffer and data are aligned so PointerIntPair works with them.
298   // TODO: Is 16-byte alignment enough?  We copy small object files with large
299   // alignment expectations into this buffer.
300   SmallString<256> NameBuf;
301   StringRef NameRef = BufferName.toStringRef(NameBuf);
302   size_t AlignedStringLen = alignTo(sizeof(MemBuffer) + NameRef.size() + 1, 16);
303   size_t RealLen = AlignedStringLen + Size + 1;
304   char *Mem = static_cast<char*>(operator new(RealLen, std::nothrow));
305   if (!Mem)
306     return nullptr;
307 
308   // The name is stored after the class itself.
309   CopyStringRef(Mem + sizeof(MemBuffer), NameRef);
310 
311   // The buffer begins after the name and must be aligned.
312   char *Buf = Mem + AlignedStringLen;
313   Buf[Size] = 0; // Null terminate buffer.
314 
315   auto *Ret = new (Mem) MemBuffer(StringRef(Buf, Size), true);
316   return std::unique_ptr<WritableMemoryBuffer>(Ret);
317 }
318 
319 std::unique_ptr<WritableMemoryBuffer>
320 WritableMemoryBuffer::getNewMemBuffer(size_t Size, const Twine &BufferName) {
321   auto SB = WritableMemoryBuffer::getNewUninitMemBuffer(Size, BufferName);
322   if (!SB)
323     return nullptr;
324   memset(SB->getBufferStart(), 0, Size);
325   return SB;
326 }
327 
328 static bool shouldUseMmap(sys::fs::file_t FD,
329                           size_t FileSize,
330                           size_t MapSize,
331                           off_t Offset,
332                           bool RequiresNullTerminator,
333                           int PageSize,
334                           bool IsVolatile) {
335   // mmap may leave the buffer without null terminator if the file size changed
336   // by the time the last page is mapped in, so avoid it if the file size is
337   // likely to change.
338   if (IsVolatile && RequiresNullTerminator)
339     return false;
340 
341   // We don't use mmap for small files because this can severely fragment our
342   // address space.
343   if (MapSize < 4 * 4096 || MapSize < (unsigned)PageSize)
344     return false;
345 
346   if (!RequiresNullTerminator)
347     return true;
348 
349   // If we don't know the file size, use fstat to find out.  fstat on an open
350   // file descriptor is cheaper than stat on a random path.
351   // FIXME: this chunk of code is duplicated, but it avoids a fstat when
352   // RequiresNullTerminator = false and MapSize != -1.
353   if (FileSize == size_t(-1)) {
354     sys::fs::file_status Status;
355     if (sys::fs::status(FD, Status))
356       return false;
357     FileSize = Status.getSize();
358   }
359 
360   // If we need a null terminator and the end of the map is inside the file,
361   // we cannot use mmap.
362   size_t End = Offset + MapSize;
363   assert(End <= FileSize);
364   if (End != FileSize)
365     return false;
366 
367   // Don't try to map files that are exactly a multiple of the system page size
368   // if we need a null terminator.
369   if ((FileSize & (PageSize -1)) == 0)
370     return false;
371 
372 #if defined(__CYGWIN__)
373   // Don't try to map files that are exactly a multiple of the physical page size
374   // if we need a null terminator.
375   // FIXME: We should reorganize again getPageSize() on Win32.
376   if ((FileSize & (4096 - 1)) == 0)
377     return false;
378 #endif
379 
380   return true;
381 }
382 
383 static ErrorOr<std::unique_ptr<WriteThroughMemoryBuffer>>
384 getReadWriteFile(const Twine &Filename, uint64_t FileSize, uint64_t MapSize,
385                  uint64_t Offset) {
386   Expected<sys::fs::file_t> FDOrErr = sys::fs::openNativeFileForReadWrite(
387       Filename, sys::fs::CD_OpenExisting, sys::fs::OF_None);
388   if (!FDOrErr)
389     return errorToErrorCode(FDOrErr.takeError());
390   sys::fs::file_t FD = *FDOrErr;
391 
392   // Default is to map the full file.
393   if (MapSize == uint64_t(-1)) {
394     // If we don't know the file size, use fstat to find out.  fstat on an open
395     // file descriptor is cheaper than stat on a random path.
396     if (FileSize == uint64_t(-1)) {
397       sys::fs::file_status Status;
398       std::error_code EC = sys::fs::status(FD, Status);
399       if (EC)
400         return EC;
401 
402       // If this not a file or a block device (e.g. it's a named pipe
403       // or character device), we can't mmap it, so error out.
404       sys::fs::file_type Type = Status.type();
405       if (Type != sys::fs::file_type::regular_file &&
406           Type != sys::fs::file_type::block_file)
407         return make_error_code(errc::invalid_argument);
408 
409       FileSize = Status.getSize();
410     }
411     MapSize = FileSize;
412   }
413 
414   std::error_code EC;
415   std::unique_ptr<WriteThroughMemoryBuffer> Result(
416       new (NamedBufferAlloc(Filename))
417           MemoryBufferMMapFile<WriteThroughMemoryBuffer>(false, FD, MapSize,
418                                                          Offset, EC));
419   if (EC)
420     return EC;
421   return std::move(Result);
422 }
423 
424 ErrorOr<std::unique_ptr<WriteThroughMemoryBuffer>>
425 WriteThroughMemoryBuffer::getFile(const Twine &Filename, int64_t FileSize) {
426   return getReadWriteFile(Filename, FileSize, FileSize, 0);
427 }
428 
429 /// Map a subrange of the specified file as a WritableMemoryBuffer.
430 ErrorOr<std::unique_ptr<WriteThroughMemoryBuffer>>
431 WriteThroughMemoryBuffer::getFileSlice(const Twine &Filename, uint64_t MapSize,
432                                        uint64_t Offset) {
433   return getReadWriteFile(Filename, -1, MapSize, Offset);
434 }
435 
436 template <typename MB>
437 static ErrorOr<std::unique_ptr<MB>>
438 getOpenFileImpl(sys::fs::file_t FD, const Twine &Filename, uint64_t FileSize,
439                 uint64_t MapSize, int64_t Offset, bool RequiresNullTerminator,
440                 bool IsVolatile) {
441   static int PageSize = sys::Process::getPageSizeEstimate();
442 
443   // Default is to map the full file.
444   if (MapSize == uint64_t(-1)) {
445     // If we don't know the file size, use fstat to find out.  fstat on an open
446     // file descriptor is cheaper than stat on a random path.
447     if (FileSize == uint64_t(-1)) {
448       sys::fs::file_status Status;
449       std::error_code EC = sys::fs::status(FD, Status);
450       if (EC)
451         return EC;
452 
453       // If this not a file or a block device (e.g. it's a named pipe
454       // or character device), we can't trust the size. Create the memory
455       // buffer by copying off the stream.
456       sys::fs::file_type Type = Status.type();
457       if (Type != sys::fs::file_type::regular_file &&
458           Type != sys::fs::file_type::block_file)
459         return getMemoryBufferForStream(FD, Filename);
460 
461       FileSize = Status.getSize();
462     }
463     MapSize = FileSize;
464   }
465 
466   if (shouldUseMmap(FD, FileSize, MapSize, Offset, RequiresNullTerminator,
467                     PageSize, IsVolatile)) {
468     std::error_code EC;
469     std::unique_ptr<MB> Result(
470         new (NamedBufferAlloc(Filename)) MemoryBufferMMapFile<MB>(
471             RequiresNullTerminator, FD, MapSize, Offset, EC));
472     if (!EC)
473       return std::move(Result);
474   }
475 
476 #ifdef __MVS__
477   // Set codepage auto-conversion for z/OS.
478   if (auto EC = llvm::enableAutoConversion(FD))
479     return EC;
480 #endif
481 
482   auto Buf = WritableMemoryBuffer::getNewUninitMemBuffer(MapSize, Filename);
483   if (!Buf) {
484     // Failed to create a buffer. The only way it can fail is if
485     // new(std::nothrow) returns 0.
486     return make_error_code(errc::not_enough_memory);
487   }
488 
489   // Read until EOF, zero-initialize the rest.
490   MutableArrayRef<char> ToRead = Buf->getBuffer();
491   while (!ToRead.empty()) {
492     Expected<size_t> ReadBytes =
493         sys::fs::readNativeFileSlice(FD, ToRead, Offset);
494     if (!ReadBytes)
495       return errorToErrorCode(ReadBytes.takeError());
496     if (*ReadBytes == 0) {
497       std::memset(ToRead.data(), 0, ToRead.size());
498       break;
499     }
500     ToRead = ToRead.drop_front(*ReadBytes);
501     Offset += *ReadBytes;
502   }
503 
504   return std::move(Buf);
505 }
506 
507 ErrorOr<std::unique_ptr<MemoryBuffer>>
508 MemoryBuffer::getOpenFile(sys::fs::file_t FD, const Twine &Filename, uint64_t FileSize,
509                           bool RequiresNullTerminator, bool IsVolatile) {
510   return getOpenFileImpl<MemoryBuffer>(FD, Filename, FileSize, FileSize, 0,
511                          RequiresNullTerminator, IsVolatile);
512 }
513 
514 ErrorOr<std::unique_ptr<MemoryBuffer>>
515 MemoryBuffer::getOpenFileSlice(sys::fs::file_t FD, const Twine &Filename, uint64_t MapSize,
516                                int64_t Offset, bool IsVolatile) {
517   assert(MapSize != uint64_t(-1));
518   return getOpenFileImpl<MemoryBuffer>(FD, Filename, -1, MapSize, Offset, false,
519                                        IsVolatile);
520 }
521 
522 ErrorOr<std::unique_ptr<MemoryBuffer>> MemoryBuffer::getSTDIN() {
523   // Read in all of the data from stdin, we cannot mmap stdin.
524   //
525   // FIXME: That isn't necessarily true, we should try to mmap stdin and
526   // fallback if it fails.
527   sys::ChangeStdinMode(sys::fs::OF_Text);
528 
529   return getMemoryBufferForStream(sys::fs::getStdinHandle(), "<stdin>");
530 }
531 
532 ErrorOr<std::unique_ptr<MemoryBuffer>>
533 MemoryBuffer::getFileAsStream(const Twine &Filename) {
534   Expected<sys::fs::file_t> FDOrErr =
535       sys::fs::openNativeFileForRead(Filename, sys::fs::OF_None);
536   if (!FDOrErr)
537     return errorToErrorCode(FDOrErr.takeError());
538   sys::fs::file_t FD = *FDOrErr;
539   ErrorOr<std::unique_ptr<MemoryBuffer>> Ret =
540       getMemoryBufferForStream(FD, Filename);
541   sys::fs::closeFile(FD);
542   return Ret;
543 }
544 
545 MemoryBufferRef MemoryBuffer::getMemBufferRef() const {
546   StringRef Data = getBuffer();
547   StringRef Identifier = getBufferIdentifier();
548   return MemoryBufferRef(Data, Identifier);
549 }
550 
551 SmallVectorMemoryBuffer::~SmallVectorMemoryBuffer() {}
552