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/OwningPtr.h" 16 #include "llvm/ADT/SmallString.h" 17 #include "llvm/Config/config.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/system_error.h" 25 #include <cassert> 26 #include <cerrno> 27 #include <cstdio> 28 #include <cstring> 29 #include <new> 30 #include <sys/stat.h> 31 #include <sys/types.h> 32 #if !defined(_MSC_VER) && !defined(__MINGW32__) 33 #include <unistd.h> 34 #else 35 #include <io.h> 36 // Simplistic definitinos of these macros for use in getOpenFile. 37 #ifndef S_ISREG 38 #define S_ISREG(x) (1) 39 #endif 40 #ifndef S_ISBLK 41 #define S_ISBLK(x) (0) 42 #endif 43 #endif 44 using namespace llvm; 45 46 //===----------------------------------------------------------------------===// 47 // MemoryBuffer implementation itself. 48 //===----------------------------------------------------------------------===// 49 50 MemoryBuffer::~MemoryBuffer() { } 51 52 /// init - Initialize this MemoryBuffer as a reference to externally allocated 53 /// memory, memory that we know is already null terminated. 54 void MemoryBuffer::init(const char *BufStart, const char *BufEnd, 55 bool RequiresNullTerminator) { 56 assert((!RequiresNullTerminator || BufEnd[0] == 0) && 57 "Buffer is not null terminated!"); 58 BufferStart = BufStart; 59 BufferEnd = BufEnd; 60 } 61 62 //===----------------------------------------------------------------------===// 63 // MemoryBufferMem implementation. 64 //===----------------------------------------------------------------------===// 65 66 /// CopyStringRef - Copies contents of a StringRef into a block of memory and 67 /// null-terminates it. 68 static void CopyStringRef(char *Memory, StringRef Data) { 69 memcpy(Memory, Data.data(), Data.size()); 70 Memory[Data.size()] = 0; // Null terminate string. 71 } 72 73 namespace { 74 struct NamedBufferAlloc { 75 StringRef Name; 76 NamedBufferAlloc(StringRef Name) : Name(Name) {} 77 }; 78 } 79 80 void *operator new(size_t N, const NamedBufferAlloc &Alloc) { 81 char *Mem = static_cast<char *>(operator new(N + Alloc.Name.size() + 1)); 82 CopyStringRef(Mem + N, Alloc.Name); 83 return Mem; 84 } 85 86 namespace { 87 /// MemoryBufferMem - Named MemoryBuffer pointing to a block of memory. 88 class MemoryBufferMem : public MemoryBuffer { 89 public: 90 MemoryBufferMem(StringRef InputData, bool RequiresNullTerminator) { 91 init(InputData.begin(), InputData.end(), RequiresNullTerminator); 92 } 93 94 virtual const char *getBufferIdentifier() const LLVM_OVERRIDE { 95 // The name is stored after the class itself. 96 return reinterpret_cast<const char*>(this + 1); 97 } 98 99 virtual BufferKind getBufferKind() const LLVM_OVERRIDE { 100 return MemoryBuffer_Malloc; 101 } 102 }; 103 } 104 105 /// getMemBuffer - Open the specified memory range as a MemoryBuffer. Note 106 /// that InputData must be a null terminated if RequiresNullTerminator is true! 107 MemoryBuffer *MemoryBuffer::getMemBuffer(StringRef InputData, 108 StringRef BufferName, 109 bool RequiresNullTerminator) { 110 return new (NamedBufferAlloc(BufferName)) 111 MemoryBufferMem(InputData, RequiresNullTerminator); 112 } 113 114 /// getMemBufferCopy - Open the specified memory range as a MemoryBuffer, 115 /// copying the contents and taking ownership of it. This has no requirements 116 /// on EndPtr[0]. 117 MemoryBuffer *MemoryBuffer::getMemBufferCopy(StringRef InputData, 118 StringRef BufferName) { 119 MemoryBuffer *Buf = getNewUninitMemBuffer(InputData.size(), BufferName); 120 if (!Buf) return 0; 121 memcpy(const_cast<char*>(Buf->getBufferStart()), InputData.data(), 122 InputData.size()); 123 return Buf; 124 } 125 126 /// getNewUninitMemBuffer - Allocate a new MemoryBuffer of the specified size 127 /// that is not initialized. Note that the caller should initialize the 128 /// memory allocated by this method. The memory is owned by the MemoryBuffer 129 /// object. 130 MemoryBuffer *MemoryBuffer::getNewUninitMemBuffer(size_t Size, 131 StringRef BufferName) { 132 // Allocate space for the MemoryBuffer, the data and the name. It is important 133 // that MemoryBuffer and data are aligned so PointerIntPair works with them. 134 size_t AlignedStringLen = 135 RoundUpToAlignment(sizeof(MemoryBufferMem) + BufferName.size() + 1, 136 sizeof(void*)); // TODO: Is sizeof(void*) enough? 137 size_t RealLen = AlignedStringLen + Size + 1; 138 char *Mem = static_cast<char*>(operator new(RealLen, std::nothrow)); 139 if (!Mem) return 0; 140 141 // The name is stored after the class itself. 142 CopyStringRef(Mem + sizeof(MemoryBufferMem), BufferName); 143 144 // The buffer begins after the name and must be aligned. 145 char *Buf = Mem + AlignedStringLen; 146 Buf[Size] = 0; // Null terminate buffer. 147 148 return new (Mem) MemoryBufferMem(StringRef(Buf, Size), true); 149 } 150 151 /// getNewMemBuffer - Allocate a new MemoryBuffer of the specified size that 152 /// is completely initialized to zeros. Note that the caller should 153 /// initialize the memory allocated by this method. The memory is owned by 154 /// the MemoryBuffer object. 155 MemoryBuffer *MemoryBuffer::getNewMemBuffer(size_t Size, StringRef BufferName) { 156 MemoryBuffer *SB = getNewUninitMemBuffer(Size, BufferName); 157 if (!SB) return 0; 158 memset(const_cast<char*>(SB->getBufferStart()), 0, Size); 159 return SB; 160 } 161 162 163 /// getFileOrSTDIN - Open the specified file as a MemoryBuffer, or open stdin 164 /// if the Filename is "-". If an error occurs, this returns null and fills 165 /// in *ErrStr with a reason. If stdin is empty, this API (unlike getSTDIN) 166 /// returns an empty buffer. 167 error_code MemoryBuffer::getFileOrSTDIN(StringRef Filename, 168 OwningPtr<MemoryBuffer> &result, 169 int64_t FileSize) { 170 if (Filename == "-") 171 return getSTDIN(result); 172 return getFile(Filename, result, FileSize); 173 } 174 175 //===----------------------------------------------------------------------===// 176 // MemoryBuffer::getFile implementation. 177 //===----------------------------------------------------------------------===// 178 179 namespace { 180 /// \brief Memorry maps a file descriptor using sys::fs::mapped_file_region. 181 /// 182 /// This handles converting the offset into a legal offset on the platform. 183 class MemoryBufferMMapFile : public MemoryBuffer { 184 sys::fs::mapped_file_region MFR; 185 186 static uint64_t getLegalMapOffset(uint64_t Offset) { 187 return Offset & ~(sys::fs::mapped_file_region::alignment() - 1); 188 } 189 190 static uint64_t getLegalMapSize(uint64_t Len, uint64_t Offset) { 191 return Len + (Offset - getLegalMapOffset(Offset)); 192 } 193 194 const char *getStart(uint64_t Len, uint64_t Offset) { 195 return MFR.const_data() + (Offset - getLegalMapOffset(Offset)); 196 } 197 198 public: 199 MemoryBufferMMapFile(bool RequiresNullTerminator, int FD, uint64_t Len, 200 uint64_t Offset, error_code EC) 201 : MFR(FD, false, sys::fs::mapped_file_region::readonly, 202 getLegalMapSize(Len, Offset), getLegalMapOffset(Offset), EC) { 203 if (!EC) { 204 const char *Start = getStart(Len, Offset); 205 init(Start, Start + Len, RequiresNullTerminator); 206 } 207 } 208 209 virtual const char *getBufferIdentifier() const LLVM_OVERRIDE { 210 // The name is stored after the class itself. 211 return reinterpret_cast<const char *>(this + 1); 212 } 213 214 virtual BufferKind getBufferKind() const LLVM_OVERRIDE { 215 return MemoryBuffer_MMap; 216 } 217 }; 218 } 219 220 static error_code getMemoryBufferForStream(int FD, 221 StringRef BufferName, 222 OwningPtr<MemoryBuffer> &result) { 223 const ssize_t ChunkSize = 4096*4; 224 SmallString<ChunkSize> Buffer; 225 ssize_t ReadBytes; 226 // Read into Buffer until we hit EOF. 227 do { 228 Buffer.reserve(Buffer.size() + ChunkSize); 229 ReadBytes = read(FD, Buffer.end(), ChunkSize); 230 if (ReadBytes == -1) { 231 if (errno == EINTR) continue; 232 return error_code(errno, posix_category()); 233 } 234 Buffer.set_size(Buffer.size() + ReadBytes); 235 } while (ReadBytes != 0); 236 237 result.reset(MemoryBuffer::getMemBufferCopy(Buffer, BufferName)); 238 return error_code::success(); 239 } 240 241 error_code MemoryBuffer::getFile(StringRef Filename, 242 OwningPtr<MemoryBuffer> &result, 243 int64_t FileSize, 244 bool RequiresNullTerminator) { 245 // Ensure the path is null terminated. 246 SmallString<256> PathBuf(Filename.begin(), Filename.end()); 247 return MemoryBuffer::getFile(PathBuf.c_str(), result, FileSize, 248 RequiresNullTerminator); 249 } 250 251 static error_code getOpenFileImpl(int FD, const char *Filename, 252 OwningPtr<MemoryBuffer> &Result, 253 uint64_t FileSize, uint64_t MapSize, 254 int64_t Offset, bool RequiresNullTerminator); 255 256 error_code MemoryBuffer::getFile(const char *Filename, 257 OwningPtr<MemoryBuffer> &result, 258 int64_t FileSize, 259 bool RequiresNullTerminator) { 260 int FD; 261 error_code EC = sys::fs::openFileForRead(Filename, FD); 262 if (EC) 263 return EC; 264 265 error_code ret = getOpenFileImpl(FD, Filename, result, FileSize, FileSize, 0, 266 RequiresNullTerminator); 267 close(FD); 268 return ret; 269 } 270 271 static bool shouldUseMmap(int FD, 272 size_t FileSize, 273 size_t MapSize, 274 off_t Offset, 275 bool RequiresNullTerminator, 276 int PageSize) { 277 // We don't use mmap for small files because this can severely fragment our 278 // address space. 279 if (MapSize < 4096*4) 280 return false; 281 282 if (!RequiresNullTerminator) 283 return true; 284 285 286 // If we don't know the file size, use fstat to find out. fstat on an open 287 // file descriptor is cheaper than stat on a random path. 288 // FIXME: this chunk of code is duplicated, but it avoids a fstat when 289 // RequiresNullTerminator = false and MapSize != -1. 290 if (FileSize == size_t(-1)) { 291 sys::fs::file_status Status; 292 error_code EC = sys::fs::status(FD, Status); 293 if (EC) 294 return EC; 295 FileSize = Status.getSize(); 296 } 297 298 // If we need a null terminator and the end of the map is inside the file, 299 // we cannot use mmap. 300 size_t End = Offset + MapSize; 301 assert(End <= FileSize); 302 if (End != FileSize) 303 return false; 304 305 // Don't try to map files that are exactly a multiple of the system page size 306 // if we need a null terminator. 307 if ((FileSize & (PageSize -1)) == 0) 308 return false; 309 310 return true; 311 } 312 313 static error_code getOpenFileImpl(int FD, const char *Filename, 314 OwningPtr<MemoryBuffer> &result, 315 uint64_t FileSize, uint64_t MapSize, 316 int64_t Offset, bool RequiresNullTerminator) { 317 static int PageSize = sys::process::get_self()->page_size(); 318 319 // Default is to map the full file. 320 if (MapSize == uint64_t(-1)) { 321 // If we don't know the file size, use fstat to find out. fstat on an open 322 // file descriptor is cheaper than stat on a random path. 323 if (FileSize == uint64_t(-1)) { 324 sys::fs::file_status Status; 325 error_code EC = sys::fs::status(FD, Status); 326 if (EC) 327 return EC; 328 329 // If this not a file or a block device (e.g. it's a named pipe 330 // or character device), we can't trust the size. Create the memory 331 // buffer by copying off the stream. 332 sys::fs::file_type Type = Status.type(); 333 if (Type != sys::fs::file_type::regular_file && 334 Type != sys::fs::file_type::block_file) 335 return getMemoryBufferForStream(FD, Filename, result); 336 337 FileSize = Status.getSize(); 338 } 339 MapSize = FileSize; 340 } 341 342 if (shouldUseMmap(FD, FileSize, MapSize, Offset, RequiresNullTerminator, 343 PageSize)) { 344 error_code EC; 345 result.reset(new (NamedBufferAlloc(Filename)) MemoryBufferMMapFile( 346 RequiresNullTerminator, FD, MapSize, Offset, EC)); 347 if (!EC) 348 return error_code::success(); 349 } 350 351 MemoryBuffer *Buf = MemoryBuffer::getNewUninitMemBuffer(MapSize, Filename); 352 if (!Buf) { 353 // Failed to create a buffer. The only way it can fail is if 354 // new(std::nothrow) returns 0. 355 return make_error_code(errc::not_enough_memory); 356 } 357 358 OwningPtr<MemoryBuffer> SB(Buf); 359 char *BufPtr = const_cast<char*>(SB->getBufferStart()); 360 361 size_t BytesLeft = MapSize; 362 #ifndef HAVE_PREAD 363 if (lseek(FD, Offset, SEEK_SET) == -1) 364 return error_code(errno, posix_category()); 365 #endif 366 367 while (BytesLeft) { 368 #ifdef HAVE_PREAD 369 ssize_t NumRead = ::pread(FD, BufPtr, BytesLeft, MapSize-BytesLeft+Offset); 370 #else 371 ssize_t NumRead = ::read(FD, BufPtr, BytesLeft); 372 #endif 373 if (NumRead == -1) { 374 if (errno == EINTR) 375 continue; 376 // Error while reading. 377 return error_code(errno, posix_category()); 378 } 379 if (NumRead == 0) { 380 assert(0 && "We got inaccurate FileSize value or fstat reported an " 381 "invalid file size."); 382 *BufPtr = '\0'; // null-terminate at the actual size. 383 break; 384 } 385 BytesLeft -= NumRead; 386 BufPtr += NumRead; 387 } 388 389 result.swap(SB); 390 return error_code::success(); 391 } 392 393 error_code MemoryBuffer::getOpenFile(int FD, const char *Filename, 394 OwningPtr<MemoryBuffer> &Result, 395 uint64_t FileSize, 396 bool RequiresNullTerminator) { 397 return getOpenFileImpl(FD, Filename, Result, FileSize, FileSize, 0, 398 RequiresNullTerminator); 399 } 400 401 error_code MemoryBuffer::getOpenFileSlice(int FD, const char *Filename, 402 OwningPtr<MemoryBuffer> &Result, 403 uint64_t MapSize, int64_t Offset) { 404 return getOpenFileImpl(FD, Filename, Result, -1, MapSize, Offset, false); 405 } 406 407 //===----------------------------------------------------------------------===// 408 // MemoryBuffer::getSTDIN implementation. 409 //===----------------------------------------------------------------------===// 410 411 error_code MemoryBuffer::getSTDIN(OwningPtr<MemoryBuffer> &result) { 412 // Read in all of the data from stdin, we cannot mmap stdin. 413 // 414 // FIXME: That isn't necessarily true, we should try to mmap stdin and 415 // fallback if it fails. 416 sys::ChangeStdinToBinary(); 417 418 return getMemoryBufferForStream(0, "<stdin>", result); 419 } 420