X-Git-Url: http://demsky.eecs.uci.edu/git/?a=blobdiff_plain;f=lib%2FSupport%2FMemoryBuffer.cpp;h=5f4b7daae53d162affc9dd71d9cccaba140b4312;hb=3cf9f3731222d5c86173c83b6dec9783f40b305c;hp=8cea07fdb5451f8a95d437c507b200670a48a6fe;hpb=bdbd2d710c665bcdf31cbed4b44cf3f94ba746e7;p=oota-llvm.git diff --git a/lib/Support/MemoryBuffer.cpp b/lib/Support/MemoryBuffer.cpp index 8cea07fdb54..5f4b7daae53 100644 --- a/lib/Support/MemoryBuffer.cpp +++ b/lib/Support/MemoryBuffer.cpp @@ -12,281 +12,400 @@ //===----------------------------------------------------------------------===// #include "llvm/Support/MemoryBuffer.h" -#include "llvm/System/MappedFile.h" -#include "llvm/System/Process.h" -#include "llvm/System/Program.h" +#include "llvm/ADT/SmallString.h" +#include "llvm/Config/config.h" +#include "llvm/Support/Errc.h" +#include "llvm/Support/Errno.h" +#include "llvm/Support/FileSystem.h" +#include "llvm/Support/MathExtras.h" +#include "llvm/Support/Path.h" +#include "llvm/Support/Process.h" +#include "llvm/Support/Program.h" #include +#include #include #include -#include +#include +#include +#include +#if !defined(_MSC_VER) && !defined(__MINGW32__) +#include +#else +#include +#endif using namespace llvm; //===----------------------------------------------------------------------===// // MemoryBuffer implementation itself. //===----------------------------------------------------------------------===// -MemoryBuffer::~MemoryBuffer() { - if (MustDeleteBuffer) - delete [] 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 = new char[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. +} + namespace { +struct NamedBufferAlloc { + StringRef Name; + NamedBufferAlloc(StringRef Name) : Name(Name) {} +}; +} + +void *operator new(size_t N, const NamedBufferAlloc &Alloc) { + char *Mem = static_cast(operator new(N + Alloc.Name.size() + 1)); + CopyStringRef(Mem + N, Alloc.Name); + return Mem; +} + +namespace { +/// MemoryBufferMem - Named MemoryBuffer pointing to a block of memory. class MemoryBufferMem : public MemoryBuffer { - std::string FileID; public: - MemoryBufferMem(const char *Start, const char *End, const char *FID, - bool Copy = false) - : FileID(FID) { - if (!Copy) - init(Start, End); - else - initCopyOf(Start, End); + MemoryBufferMem(StringRef InputData, bool RequiresNullTerminator) { + init(InputData.begin(), InputData.end(), RequiresNullTerminator); } - - virtual const char *getBufferIdentifier() const { - return FileID.c_str(); + + const char *getBufferIdentifier() const override { + // The name is stored after the class itself. + return reinterpret_cast(this + 1); + } + + BufferKind getBufferKind() const override { + return MemoryBuffer_Malloc; } }; } /// getMemBuffer - Open the specified memory range as a MemoryBuffer. Note -/// that EndPtr[0] must be a null byte and be accessible! -MemoryBuffer *MemoryBuffer::getMemBuffer(const char *StartPtr, - const char *EndPtr, - const char *BufferName) { - return new MemoryBufferMem(StartPtr, EndPtr, BufferName); +/// that InputData must be a null terminated if RequiresNullTerminator is true! +MemoryBuffer *MemoryBuffer::getMemBuffer(StringRef InputData, + StringRef BufferName, + bool RequiresNullTerminator) { + return new (NamedBufferAlloc(BufferName)) + MemoryBufferMem(InputData, 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(const char *StartPtr, - const char *EndPtr, - const char *BufferName) { - return new MemoryBufferMem(StartPtr, EndPtr, BufferName, true); +MemoryBuffer *MemoryBuffer::getMemBufferCopy(StringRef InputData, + StringRef BufferName) { + MemoryBuffer *Buf = getNewUninitMemBuffer(InputData.size(), BufferName); + if (!Buf) return nullptr; + 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. -MemoryBuffer *MemoryBuffer::getNewUninitMemBuffer(unsigned Size, - const char *BufferName) { - char *Buf = new char[Size+1]; - Buf[Size] = 0; - MemoryBufferMem *SB = new MemoryBufferMem(Buf, Buf+Size, BufferName); - // The memory for this buffer is owned by the MemoryBuffer. - SB->MustDeleteBuffer = true; - return SB; +/// 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) { + // Allocate space for the MemoryBuffer, the data and the name. It is important + // that MemoryBuffer and data are aligned so PointerIntPair works with them. + // TODO: Is 16-byte alignment enough? We copy small object files with large + // alignment expectations into this buffer. + size_t AlignedStringLen = + RoundUpToAlignment(sizeof(MemoryBufferMem) + BufferName.size() + 1, 16); + size_t RealLen = AlignedStringLen + Size + 1; + char *Mem = static_cast(operator new(RealLen, std::nothrow)); + if (!Mem) return nullptr; + + // 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(unsigned Size, - const char *BufferName) { +MemoryBuffer *MemoryBuffer::getNewMemBuffer(size_t Size, StringRef BufferName) { MemoryBuffer *SB = getNewUninitMemBuffer(Size, BufferName); - memset(const_cast(SB->getBufferStart()), 0, Size+1); + if (!SB) return nullptr; + memset(const_cast(SB->getBufferStart()), 0, Size); return SB; } - -/// getFileOrSTDIN - Open the specified file as a MemoryBuffer, or open stdin -/// 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(const char *FilenameStart, - unsigned FnSize, - std::string *ErrStr, - int64_t FileSize) { - if (FnSize != 1 || FilenameStart[0] != '-') - return getFile(FilenameStart, FnSize, ErrStr, FileSize); - MemoryBuffer *M = getSTDIN(); - if (M) return M; - - // If stdin was empty, M is null. Cons up an empty memory buffer now. - const char *EmptyStr = ""; - return MemoryBuffer::getMemBuffer(EmptyStr, EmptyStr, ""); +ErrorOr> +MemoryBuffer::getFileOrSTDIN(StringRef Filename, int64_t FileSize) { + if (Filename == "-") + return getSTDIN(); + return getFile(Filename, FileSize); } + //===----------------------------------------------------------------------===// -// MemoryBufferMMapFile implementation. +// MemoryBuffer::getFile implementation. //===----------------------------------------------------------------------===// namespace { +/// \brief Memory maps a file descriptor using sys::fs::mapped_file_region. +/// +/// This handles converting the offset into a legal offset on the platform. class MemoryBufferMMapFile : public MemoryBuffer { - sys::MappedFile File; + sys::fs::mapped_file_region MFR; + + static uint64_t getLegalMapOffset(uint64_t Offset) { + return Offset & ~(sys::fs::mapped_file_region::alignment() - 1); + } + + static uint64_t getLegalMapSize(uint64_t Len, uint64_t Offset) { + return Len + (Offset - getLegalMapOffset(Offset)); + } + + const char *getStart(uint64_t Len, uint64_t Offset) { + return MFR.const_data() + (Offset - getLegalMapOffset(Offset)); + } + public: - MemoryBufferMMapFile() {} - - bool open(const sys::Path &Filename, std::string *ErrStr); - - virtual const char *getBufferIdentifier() const { - return File.path().c_str(); + MemoryBufferMMapFile(bool RequiresNullTerminator, int FD, uint64_t Len, + uint64_t Offset, std::error_code EC) + : MFR(FD, false, sys::fs::mapped_file_region::readonly, + getLegalMapSize(Len, Offset), getLegalMapOffset(Offset), EC) { + if (!EC) { + const char *Start = getStart(Len, Offset); + init(Start, Start + Len, RequiresNullTerminator); + } + } + + const char *getBufferIdentifier() const override { + // The name is stored after the class itself. + return reinterpret_cast(this + 1); + } + + BufferKind getBufferKind() const override { + return MemoryBuffer_MMap; } - - ~MemoryBufferMMapFile(); }; } -bool MemoryBufferMMapFile::open(const sys::Path &Filename, - std::string *ErrStr) { - // FIXME: This does an extra stat syscall to figure out the size, but we - // already know the size! - bool Failure = File.open(Filename, ErrStr); - if (Failure) return true; - - if (!File.map(ErrStr)) +static ErrorOr> +getMemoryBufferForStream(int FD, StringRef BufferName) { + 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(FD, Buffer.end(), ChunkSize); + if (ReadBytes == -1) { + if (errno == EINTR) continue; + return std::error_code(errno, std::generic_category()); + } + Buffer.set_size(Buffer.size() + ReadBytes); + } while (ReadBytes != 0); + + std::unique_ptr Ret( + MemoryBuffer::getMemBufferCopy(Buffer, BufferName)); + return std::move(Ret); +} + +static ErrorOr> +getFileAux(const char *Filename, int64_t FileSize, bool RequiresNullTerminator, + bool IsVolatileSize); + +ErrorOr> +MemoryBuffer::getFile(Twine Filename, int64_t FileSize, + bool RequiresNullTerminator, bool IsVolatileSize) { + // Ensure the path is null terminated. + SmallString<256> PathBuf; + StringRef NullTerminatedName = Filename.toNullTerminatedStringRef(PathBuf); + return getFileAux(NullTerminatedName.data(), FileSize, RequiresNullTerminator, + IsVolatileSize); +} + +static ErrorOr> +getOpenFileImpl(int FD, const char *Filename, uint64_t FileSize, + uint64_t MapSize, int64_t Offset, bool RequiresNullTerminator, + bool IsVolatileSize); + +static ErrorOr> +getFileAux(const char *Filename, int64_t FileSize, bool RequiresNullTerminator, + bool IsVolatileSize) { + int FD; + std::error_code EC = sys::fs::openFileForRead(Filename, FD); + if (EC) + return EC; + + ErrorOr> Ret = + getOpenFileImpl(FD, Filename, FileSize, FileSize, 0, + RequiresNullTerminator, IsVolatileSize); + close(FD); + return Ret; +} + +static bool shouldUseMmap(int FD, + size_t FileSize, + size_t MapSize, + off_t Offset, + bool RequiresNullTerminator, + int PageSize, + bool IsVolatileSize) { + // mmap may leave the buffer without null terminator if the file size changed + // by the time the last page is mapped in, so avoid it if the file size is + // likely to change. + if (IsVolatileSize) + return false; + + // We don't use mmap for small files because this can severely fragment our + // address space. + if (MapSize < 4 * 4096 || MapSize < (unsigned)PageSize) + return false; + + if (!RequiresNullTerminator) return true; - - size_t Size = File.size(); - - static unsigned PageSize = sys::Process::GetPageSize(); - assert(((PageSize & (PageSize-1)) == 0) && PageSize && - "Page size is not a power of 2!"); - - // If this file is not an exact multiple of the system page size (common - // case), then the OS has zero terminated the buffer for us. - if ((Size & (PageSize-1))) { - init(File.charBase(), File.charBase()+Size); - } else { - // Otherwise, we allocate a new memory buffer and copy the data over - initCopyOf(File.charBase(), File.charBase()+Size); - - // No need to keep the file mapped any longer. - File.unmap(); + + + // 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. + // FIXME: this chunk of code is duplicated, but it avoids a fstat when + // RequiresNullTerminator = false and MapSize != -1. + if (FileSize == size_t(-1)) { + sys::fs::file_status Status; + if (sys::fs::status(FD, Status)) + return false; + FileSize = Status.getSize(); } - return false; -} -MemoryBufferMMapFile::~MemoryBufferMMapFile() { - if (File.isMapped()) - File.unmap(); + // 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; } -//===----------------------------------------------------------------------===// -// MemoryBuffer::getFile implementation. -//===----------------------------------------------------------------------===// +static ErrorOr> +getOpenFileImpl(int FD, const char *Filename, uint64_t FileSize, + uint64_t MapSize, int64_t Offset, bool RequiresNullTerminator, + bool IsVolatileSize) { + static int PageSize = sys::process::get_self()->page_size(); -MemoryBuffer *MemoryBuffer::getFile(const char *FilenameStart, unsigned FnSize, - std::string *ErrStr, int64_t FileSize){ - // FIXME: it would be nice if PathWithStatus didn't copy the filename into a - // temporary string. :( - sys::PathWithStatus P(FilenameStart, FnSize); -#if 1 - MemoryBufferMMapFile *M = new MemoryBufferMMapFile(); - if (!M->open(P, ErrStr)) - return M; - delete M; - return 0; -#else - // FIXME: We need an efficient and portable method to open a file and then use - // 'read' to copy the bits out. The unix implementation is below. This is - // an important optimization for clients that want to open large numbers of - // small files (using mmap on everything can easily exhaust address space!). - - // If the user didn't specify a filesize, do a stat to find it. - if (FileSize == -1) { - const sys::FileStatus *FS = P.getFileStatus(); - if (FS == 0) return 0; // Error stat'ing file. - - FileSize = FS->fileSize; + // Default is to map the full file. + if (MapSize == uint64_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 == uint64_t(-1)) { + sys::fs::file_status Status; + std::error_code EC = sys::fs::status(FD, Status); + if (EC) + return EC; + + // If this not a file or a block device (e.g. it's a named pipe + // or character device), we can't trust the size. Create the memory + // buffer by copying off the stream. + sys::fs::file_type Type = Status.type(); + if (Type != sys::fs::file_type::regular_file && + Type != sys::fs::file_type::block_file) + return getMemoryBufferForStream(FD, Filename); + + FileSize = Status.getSize(); + } + MapSize = FileSize; } - - // If the file is larger than some threshold, use mmap, otherwise use 'read'. - if (FileSize >= 4096*4) { - MemoryBufferMMapFile *M = new MemoryBufferMMapFile(); - if (!M->open(P, ErrStr)) - return M; - delete M; - return 0; + + if (shouldUseMmap(FD, FileSize, MapSize, Offset, RequiresNullTerminator, + PageSize, IsVolatileSize)) { + std::error_code EC; + std::unique_ptr Result( + new (NamedBufferAlloc(Filename)) + MemoryBufferMMapFile(RequiresNullTerminator, FD, MapSize, Offset, EC)); + if (!EC) + return std::move(Result); } - - MemoryBuffer *SB = getNewUninitMemBuffer(FileSize, FilenameStart); - char *BufPtr = const_cast(SB->getBufferStart()); - - int FD = ::open(FilenameStart, O_RDONLY); - if (FD == -1) { - delete SB; - return 0; + + MemoryBuffer *Buf = MemoryBuffer::getNewUninitMemBuffer(MapSize, Filename); + if (!Buf) { + // 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); } - - unsigned BytesLeft = FileSize; + + std::unique_ptr SB(Buf); + char *BufPtr = const_cast(SB->getBufferStart()); + + size_t BytesLeft = MapSize; +#ifndef HAVE_PREAD + if (lseek(FD, Offset, SEEK_SET) == -1) + return std::error_code(errno, std::generic_category()); +#endif + while (BytesLeft) { +#ifdef HAVE_PREAD + ssize_t NumRead = ::pread(FD, BufPtr, BytesLeft, MapSize-BytesLeft+Offset); +#else ssize_t NumRead = ::read(FD, BufPtr, BytesLeft); - if (NumRead != -1) { - BytesLeft -= NumRead; - BufPtr += NumRead; - } else if (errno == EINTR) { - // try again - } else { - // error reading. - close(FD); - delete SB; - return 0; +#endif + if (NumRead == -1) { + if (errno == EINTR) + continue; + // Error while reading. + return std::error_code(errno, std::generic_category()); + } + if (NumRead == 0) { + memset(BufPtr, 0, BytesLeft); // zero-initialize rest of the buffer. + break; } + BytesLeft -= NumRead; + BufPtr += NumRead; } - close(FD); - - return SB; -#endif -} + return std::move(SB); +} -//===----------------------------------------------------------------------===// -// MemoryBuffer::getSTDIN implementation. -//===----------------------------------------------------------------------===// +ErrorOr> +MemoryBuffer::getOpenFile(int FD, const char *Filename, uint64_t FileSize, + bool RequiresNullTerminator, bool IsVolatileSize) { + return getOpenFileImpl(FD, Filename, FileSize, FileSize, 0, + RequiresNullTerminator, IsVolatileSize); +} -namespace { -class STDINBufferFile : public MemoryBuffer { -public: - virtual const char *getBufferIdentifier() const { - return ""; - } -}; +ErrorOr> +MemoryBuffer::getOpenFileSlice(int FD, const char *Filename, uint64_t MapSize, + int64_t Offset, bool IsVolatileSize) { + return getOpenFileImpl(FD, Filename, -1, MapSize, Offset, false, + IsVolatileSize); } -MemoryBuffer *MemoryBuffer::getSTDIN() { - char Buffer[4096*4]; - - std::vector FileData; - +ErrorOr> MemoryBuffer::getSTDIN() { // Read in all of the data from stdin, we cannot mmap stdin. - sys::Program::ChangeStdinToBinary(); - while (size_t ReadBytes = fread(Buffer, sizeof(char), 4096*4, stdin)) - FileData.insert(FileData.end(), Buffer, Buffer+ReadBytes); - - FileData.push_back(0); // &FileData[Size] is invalid. So is &*FileData.end(). - size_t Size = FileData.size(); - if (Size <= 1) - return 0; - MemoryBuffer *B = new STDINBufferFile(); - B->initCopyOf(&FileData[0], &FileData[Size-1]); - return B; + // + // FIXME: That isn't necessarily true, we should try to mmap stdin and + // fallback if it fails. + sys::ChangeStdinToBinary(); + + return getMemoryBufferForStream(0, ""); }