X-Git-Url: http://demsky.eecs.uci.edu/git/?a=blobdiff_plain;f=lib%2FSupport%2FMemoryBuffer.cpp;h=ea72720b7f01c53e0cb7c799391cb0ab2c6b521a;hb=88c924c67e894981e3e92ecd95e4d9b46ac92e65;hp=b1a767c35a21822c194ded8be2204c7eab35cb6a;hpb=60e6f3d4123a01babeb2c1a0e00d0a2b109008e5;p=oota-llvm.git diff --git a/lib/Support/MemoryBuffer.cpp b/lib/Support/MemoryBuffer.cpp index b1a767c35a2..ea72720b7f0 100644 --- a/lib/Support/MemoryBuffer.cpp +++ b/lib/Support/MemoryBuffer.cpp @@ -14,14 +14,17 @@ #include "llvm/Support/MemoryBuffer.h" #include "llvm/ADT/OwningPtr.h" #include "llvm/ADT/SmallString.h" -#include "llvm/System/Errno.h" -#include "llvm/System/Path.h" -#include "llvm/System/Process.h" -#include "llvm/System/Program.h" +#include "llvm/Support/MathExtras.h" +#include "llvm/Support/Errno.h" +#include "llvm/Support/Path.h" +#include "llvm/Support/Process.h" +#include "llvm/Support/Program.h" +#include "llvm/Support/system_error.h" #include #include #include #include +#include #include #include #if !defined(_MSC_VER) && !defined(__MINGW32__) @@ -33,54 +36,55 @@ #include using namespace llvm; +namespace { const llvm::error_code success; } + //===----------------------------------------------------------------------===// // MemoryBuffer implementation itself. //===----------------------------------------------------------------------===// -MemoryBuffer::~MemoryBuffer() { - if (MustDeleteBuffer) - free((void*)BufferStart); -} - -/// initCopyOf - Initialize this source buffer with a copy of the specified -/// memory range. We make the copy so that we can null terminate it -/// successfully. -void MemoryBuffer::initCopyOf(const char *BufStart, const char *BufEnd) { - size_t Size = BufEnd-BufStart; - BufferStart = (char *)malloc(Size+1); - BufferEnd = BufferStart+Size; - memcpy(const_cast(BufferStart), BufStart, Size); - *const_cast(BufferEnd) = 0; // Null terminate buffer. - MustDeleteBuffer = true; -} +MemoryBuffer::~MemoryBuffer() { } /// init - Initialize this MemoryBuffer as a reference to externally allocated /// memory, memory that we know is already null terminated. -void MemoryBuffer::init(const char *BufStart, const char *BufEnd) { - assert(BufEnd[0] == 0 && "Buffer is not null terminated!"); +void MemoryBuffer::init(const char *BufStart, const char *BufEnd, + bool RequiresNullTerminator) { + assert((!RequiresNullTerminator || BufEnd[0] == 0) && + "Buffer is not null terminated!"); BufferStart = BufStart; BufferEnd = BufEnd; - MustDeleteBuffer = false; } //===----------------------------------------------------------------------===// // MemoryBufferMem implementation. //===----------------------------------------------------------------------===// +/// CopyStringRef - Copies contents of a StringRef into a block of memory and +/// null-terminates it. +static void CopyStringRef(char *Memory, StringRef Data) { + memcpy(Memory, Data.data(), Data.size()); + Memory[Data.size()] = 0; // Null terminate string. +} + +/// GetNamedBuffer - Allocates a new MemoryBuffer with Name copied after it. +template +static T* GetNamedBuffer(StringRef Buffer, StringRef Name, + bool RequiresNullTerminator) { + char *Mem = static_cast(operator new(sizeof(T) + Name.size() + 1)); + CopyStringRef(Mem + sizeof(T), Name); + return new (Mem) T(Buffer, RequiresNullTerminator); +} + namespace { +/// MemoryBufferMem - Named MemoryBuffer pointing to a block of memory. class MemoryBufferMem : public MemoryBuffer { - std::string FileID; public: - MemoryBufferMem(StringRef InputData, StringRef FID, bool Copy = false) - : FileID(FID) { - if (!Copy) - init(InputData.data(), InputData.data()+InputData.size()); - else - initCopyOf(InputData.data(), InputData.data()+InputData.size()); + MemoryBufferMem(StringRef InputData, bool RequiresNullTerminator) { + init(InputData.begin(), InputData.end(), RequiresNullTerminator); } - + virtual const char *getBufferIdentifier() const { - return FileID.c_str(); + // The name is stored after the class itself. + return reinterpret_cast(this + 1); } }; } @@ -88,42 +92,57 @@ public: /// getMemBuffer - Open the specified memory range as a MemoryBuffer. Note /// that EndPtr[0] must be a null byte and be accessible! MemoryBuffer *MemoryBuffer::getMemBuffer(StringRef InputData, - const char *BufferName) { - return new MemoryBufferMem(InputData, BufferName); + StringRef BufferName, + bool RequiresNullTerminator) { + return GetNamedBuffer(InputData, BufferName, + RequiresNullTerminator); } /// getMemBufferCopy - Open the specified memory range as a MemoryBuffer, /// copying the contents and taking ownership of it. This has no requirements /// on EndPtr[0]. MemoryBuffer *MemoryBuffer::getMemBufferCopy(StringRef InputData, - const char *BufferName) { - return new MemoryBufferMem(InputData, BufferName, true); + StringRef BufferName) { + MemoryBuffer *Buf = getNewUninitMemBuffer(InputData.size(), BufferName); + if (!Buf) return 0; + memcpy(const_cast(Buf->getBufferStart()), InputData.data(), + InputData.size()); + return Buf; } /// getNewUninitMemBuffer - Allocate a new MemoryBuffer of the specified size -/// that is completely initialized to zeros. Note that the caller should -/// initialize the memory allocated by this method. The memory is owned by -/// the MemoryBuffer object. +/// that is not initialized. Note that the caller should initialize the +/// memory allocated by this method. The memory is owned by the MemoryBuffer +/// object. MemoryBuffer *MemoryBuffer::getNewUninitMemBuffer(size_t Size, StringRef BufferName) { - char *Buf = (char *)malloc(Size+1); - if (!Buf) return 0; - Buf[Size] = 0; - MemoryBufferMem *SB = new MemoryBufferMem(StringRef(Buf, Size), BufferName); - // The memory for this buffer is owned by the MemoryBuffer. - SB->MustDeleteBuffer = true; - return SB; + // Allocate space for the MemoryBuffer, the data and the name. It is important + // that MemoryBuffer and data are aligned so PointerIntPair works with them. + size_t AlignedStringLen = + RoundUpToAlignment(sizeof(MemoryBufferMem) + BufferName.size() + 1, + sizeof(void*)); // TODO: Is sizeof(void*) enough? + size_t RealLen = AlignedStringLen + Size + 1; + char *Mem = static_cast(operator new(RealLen, std::nothrow)); + if (!Mem) return 0; + + // The name is stored after the class itself. + CopyStringRef(Mem + sizeof(MemoryBufferMem), BufferName); + + // The buffer begins after the name and must be aligned. + char *Buf = Mem + AlignedStringLen; + Buf[Size] = 0; // Null terminate buffer. + + return new (Mem) MemoryBufferMem(StringRef(Buf, Size), true); } /// getNewMemBuffer - Allocate a new MemoryBuffer of the specified size that /// is completely initialized to zeros. Note that the caller should /// initialize the memory allocated by this method. The memory is owned by /// the MemoryBuffer object. -MemoryBuffer *MemoryBuffer::getNewMemBuffer(size_t Size, - const char *BufferName) { +MemoryBuffer *MemoryBuffer::getNewMemBuffer(size_t Size, StringRef BufferName) { MemoryBuffer *SB = getNewUninitMemBuffer(Size, BufferName); if (!SB) return 0; - memset(const_cast(SB->getBufferStart()), 0, Size+1); + memset(const_cast(SB->getBufferStart()), 0, Size); return SB; } @@ -132,22 +151,20 @@ MemoryBuffer *MemoryBuffer::getNewMemBuffer(size_t Size, /// if the Filename is "-". If an error occurs, this returns null and fills /// in *ErrStr with a reason. If stdin is empty, this API (unlike getSTDIN) /// returns an empty buffer. -MemoryBuffer *MemoryBuffer::getFileOrSTDIN(StringRef Filename, - std::string *ErrStr, - int64_t FileSize, - struct stat *FileInfo) { +error_code MemoryBuffer::getFileOrSTDIN(StringRef Filename, + OwningPtr &result, + int64_t FileSize) { if (Filename == "-") - return getSTDIN(ErrStr); - return getFile(Filename, ErrStr, FileSize, FileInfo); + return getSTDIN(result); + return getFile(Filename, result, FileSize); } -MemoryBuffer *MemoryBuffer::getFileOrSTDIN(const char *Filename, - std::string *ErrStr, - int64_t FileSize, - struct stat *FileInfo) { +error_code MemoryBuffer::getFileOrSTDIN(const char *Filename, + OwningPtr &result, + int64_t FileSize) { if (strcmp(Filename, "-") == 0) - return getSTDIN(ErrStr); - return getFile(Filename, ErrStr, FileSize, FileInfo); + return getSTDIN(result); + return getFile(Filename, result, FileSize); } //===----------------------------------------------------------------------===// @@ -158,149 +175,194 @@ namespace { /// MemoryBufferMMapFile - This represents a file that was mapped in with the /// sys::Path::MapInFilePages method. When destroyed, it calls the /// sys::Path::UnMapFilePages method. -class MemoryBufferMMapFile : public MemoryBuffer { - std::string Filename; +class MemoryBufferMMapFile : public MemoryBufferMem { public: - MemoryBufferMMapFile(StringRef filename, const char *Pages, uint64_t Size) - : Filename(filename) { - init(Pages, Pages+Size); - } - - virtual const char *getBufferIdentifier() const { - return Filename.c_str(); - } - + MemoryBufferMMapFile(StringRef Buffer, bool RequiresNullTerminator) + : MemoryBufferMem(Buffer, RequiresNullTerminator) { } + ~MemoryBufferMMapFile() { - sys::Path::UnMapFilePages(getBufferStart(), getBufferSize()); - } -}; + static int PageSize = sys::Process::GetPageSize(); -/// FileCloser - RAII object to make sure an FD gets closed properly. -class FileCloser { - int FD; -public: - explicit FileCloser(int FD) : FD(FD) {} - ~FileCloser() { ::close(FD); } + uintptr_t Start = reinterpret_cast(getBufferStart()); + size_t Size = getBufferSize(); + uintptr_t RealStart = Start & ~(PageSize - 1); + size_t RealSize = Size + (Start - RealStart); + + sys::Path::UnMapFilePages(reinterpret_cast(RealStart), + RealSize); + } }; } -MemoryBuffer *MemoryBuffer::getFile(StringRef Filename, std::string *ErrStr, - int64_t FileSize, struct stat *FileInfo) { +error_code MemoryBuffer::getFile(StringRef Filename, + OwningPtr &result, + int64_t FileSize, + bool RequiresNullTerminator) { + // Ensure the path is null terminated. SmallString<256> PathBuf(Filename.begin(), Filename.end()); - return MemoryBuffer::getFile(PathBuf.c_str(), ErrStr, FileSize, FileInfo); + return MemoryBuffer::getFile(PathBuf.c_str(), result, FileSize, + RequiresNullTerminator); } -MemoryBuffer *MemoryBuffer::getFile(const char *Filename, std::string *ErrStr, - int64_t FileSize, struct stat *FileInfo) { +error_code MemoryBuffer::getFile(const char *Filename, + OwningPtr &result, + int64_t FileSize, + bool RequiresNullTerminator) { int OpenFlags = O_RDONLY; #ifdef O_BINARY OpenFlags |= O_BINARY; // Open input file in binary mode on win32. #endif int FD = ::open(Filename, OpenFlags); if (FD == -1) { - if (ErrStr) *ErrStr = sys::StrError(); - return 0; + return error_code(errno, posix_category()); } - FileCloser FC(FD); // Close FD on return. - + error_code ret = getOpenFile(FD, Filename, result, FileSize, FileSize, + 0, RequiresNullTerminator); + close(FD); + return ret; +} + +static bool shouldUseMmap(int FD, + size_t FileSize, + size_t MapSize, + off_t Offset, + bool RequiresNullTerminator, + int PageSize) { + // We don't use mmap for small files because this can severely fragment our + // address space. + if (MapSize < 4096*4) + return false; + + if (!RequiresNullTerminator) + return true; + + // If we don't know the file size, use fstat to find out. fstat on an open // file descriptor is cheaper than stat on a random path. - if (FileSize == -1 || FileInfo) { - struct stat MyFileInfo; - struct stat *FileInfoPtr = FileInfo? FileInfo : &MyFileInfo; - + // FIXME: this chunk of code is duplicated, but it avoids a fstat when + // RequiresNullTerminator = false and MapSize != -1. + if (FileSize == size_t(-1)) { + struct stat FileInfo; // TODO: This should use fstat64 when available. - if (fstat(FD, FileInfoPtr) == -1) { - if (ErrStr) *ErrStr = sys::StrError(); - return 0; + if (fstat(FD, &FileInfo) == -1) { + return error_code(errno, posix_category()); } - FileSize = FileInfoPtr->st_size; + FileSize = FileInfo.st_size; } - - - // If the file is large, try to use mmap to read it in. We don't use mmap - // for small files, because this can severely fragment our address space. Also - // don't try to map files that are exactly a multiple of the system page size, - // as the file would not have the required null terminator. - // - // FIXME: Can we just mmap an extra page in the latter case? - if (FileSize >= 4096*4 && - (FileSize & (sys::Process::GetPageSize()-1)) != 0) { - if (const char *Pages = sys::Path::MapInFilePages(FD, FileSize)) { - // Close the file descriptor, now that the whole file is in memory. - return new MemoryBufferMMapFile(Filename, Pages, FileSize); + + // If we need a null terminator and the end of the map is inside the file, + // we cannot use mmap. + size_t End = Offset + MapSize; + assert(End <= FileSize); + if (End != FileSize) + return false; + + // Don't try to map files that are exactly a multiple of the system page size + // if we need a null terminator. + if ((FileSize & (PageSize -1)) == 0) + return false; + + return true; +} + +error_code MemoryBuffer::getOpenFile(int FD, const char *Filename, + OwningPtr &result, + size_t FileSize, size_t MapSize, + off_t Offset, + bool RequiresNullTerminator) { + static int PageSize = sys::Process::GetPageSize(); + + // Default is to map the full file. + if (MapSize == size_t(-1)) { + // If we don't know the file size, use fstat to find out. fstat on an open + // file descriptor is cheaper than stat on a random path. + if (FileSize == size_t(-1)) { + struct stat FileInfo; + // TODO: This should use fstat64 when available. + if (fstat(FD, &FileInfo) == -1) { + return error_code(errno, posix_category()); + } + FileSize = FileInfo.st_size; + } + MapSize = FileSize; + } + + if (shouldUseMmap(FD, FileSize, MapSize, Offset, RequiresNullTerminator, + PageSize)) { + off_t RealMapOffset = Offset & ~(PageSize - 1); + off_t Delta = Offset - RealMapOffset; + size_t RealMapSize = MapSize + Delta; + + if (const char *Pages = sys::Path::MapInFilePages(FD, + RealMapSize, + RealMapOffset)) { + result.reset(GetNamedBuffer( + StringRef(Pages + Delta, MapSize), Filename, RequiresNullTerminator)); + return success; } } - MemoryBuffer *Buf = MemoryBuffer::getNewUninitMemBuffer(FileSize, Filename); + MemoryBuffer *Buf = MemoryBuffer::getNewUninitMemBuffer(MapSize, Filename); if (!Buf) { - // Failed to create a buffer. - if (ErrStr) *ErrStr = "could not allocate buffer"; - return 0; + // Failed to create a buffer. The only way it can fail is if + // new(std::nothrow) returns 0. + return make_error_code(errc::not_enough_memory); } OwningPtr SB(Buf); char *BufPtr = const_cast(SB->getBufferStart()); - size_t BytesLeft = FileSize; + size_t BytesLeft = MapSize; + if (lseek(FD, Offset, SEEK_SET) == -1) + return error_code(errno, posix_category()); + while (BytesLeft) { ssize_t NumRead = ::read(FD, BufPtr, BytesLeft); if (NumRead == -1) { if (errno == EINTR) continue; // Error while reading. - if (ErrStr) *ErrStr = sys::StrError(); - return 0; + return error_code(errno, posix_category()); } else if (NumRead == 0) { // We hit EOF early, truncate and terminate buffer. Buf->BufferEnd = BufPtr; *BufPtr = 0; - return SB.take(); + result.swap(SB); + return success; } BytesLeft -= NumRead; BufPtr += NumRead; } - return SB.take(); + result.swap(SB); + return success; } //===----------------------------------------------------------------------===// // MemoryBuffer::getSTDIN implementation. //===----------------------------------------------------------------------===// -namespace { -class STDINBufferFile : public MemoryBuffer { -public: - virtual const char *getBufferIdentifier() const { - return ""; - } -}; -} - -MemoryBuffer *MemoryBuffer::getSTDIN(std::string *ErrStr) { - char Buffer[4096*4]; - - std::vector FileData; - +error_code MemoryBuffer::getSTDIN(OwningPtr &result) { // Read in all of the data from stdin, we cannot mmap stdin. // // FIXME: That isn't necessarily true, we should try to mmap stdin and // fallback if it fails. sys::Program::ChangeStdinToBinary(); - size_t ReadBytes; - do { - ReadBytes = fread(Buffer, sizeof(char), sizeof(Buffer), stdin); - FileData.insert(FileData.end(), Buffer, Buffer+ReadBytes); - } while (ReadBytes == sizeof(Buffer)); - if (!feof(stdin)) { - if (ErrStr) *ErrStr = "error reading from stdin"; - return 0; - } + const ssize_t ChunkSize = 4096*4; + SmallString Buffer; + ssize_t ReadBytes; + // Read into Buffer until we hit EOF. + do { + Buffer.reserve(Buffer.size() + ChunkSize); + ReadBytes = read(0, Buffer.end(), ChunkSize); + if (ReadBytes == -1) { + if (errno == EINTR) continue; + return error_code(errno, posix_category()); + } + Buffer.set_size(Buffer.size() + ReadBytes); + } while (ReadBytes != 0); - FileData.push_back(0); // &FileData[Size] is invalid. So is &*FileData.end(). - size_t Size = FileData.size(); - MemoryBuffer *B = new STDINBufferFile(); - B->initCopyOf(&FileData[0], &FileData[Size-1]); - return B; + result.reset(getMemBufferCopy(Buffer, "")); + return success; }