//===----------------------------------------------------------------------===//
#include "llvm/Support/MemoryBuffer.h"
-#include "llvm/System/MappedFile.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 <cassert>
#include <cstdio>
#include <cstring>
#include <cerrno>
+#include <sys/types.h>
+#include <sys/stat.h>
+#if !defined(_MSC_VER) && !defined(__MINGW32__)
+#include <unistd.h>
+#include <sys/uio.h>
+#else
+#include <io.h>
+#endif
+#include <fcntl.h>
using namespace llvm;
//===----------------------------------------------------------------------===//
MemoryBuffer::~MemoryBuffer() {
if (MustDeleteBuffer)
- delete [] BufferStart;
+ free((void*)BufferStart);
}
/// initCopyOf - Initialize this source buffer with a copy of the specified
/// successfully.
void MemoryBuffer::initCopyOf(const char *BufStart, const char *BufEnd) {
size_t Size = BufEnd-BufStart;
- BufferStart = new char[Size+1];
+ BufferStart = (char *)malloc(Size+1);
BufferEnd = BufferStart+Size;
memcpy(const_cast<char*>(BufferStart), BufStart, Size);
*const_cast<char*>(BufferEnd) = 0; // Null terminate buffer.
class MemoryBufferMem : public MemoryBuffer {
std::string FileID;
public:
- MemoryBufferMem(const char *Start, const char *End, const char *FID,
- bool Copy = false)
+ MemoryBufferMem(StringRef InputData, StringRef FID, bool Copy = false)
: FileID(FID) {
if (!Copy)
- init(Start, End);
+ init(InputData.data(), InputData.data()+InputData.size());
else
- initCopyOf(Start, End);
+ initCopyOf(InputData.data(), InputData.data()+InputData.size());
}
virtual const char *getBufferIdentifier() const {
/// 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,
+MemoryBuffer *MemoryBuffer::getMemBuffer(StringRef InputData,
const char *BufferName) {
- return new MemoryBufferMem(StartPtr, EndPtr, BufferName);
+ return new MemoryBufferMem(InputData, BufferName);
}
/// 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,
+MemoryBuffer *MemoryBuffer::getMemBufferCopy(StringRef InputData,
const char *BufferName) {
- return new MemoryBufferMem(StartPtr, EndPtr, BufferName, true);
+ return new MemoryBufferMem(InputData, BufferName, true);
}
/// 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];
+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(Buf, Buf+Size, BufferName);
+ MemoryBufferMem *SB = new MemoryBufferMem(StringRef(Buf, Size), BufferName);
// The memory for this buffer is owned by the MemoryBuffer.
SB->MustDeleteBuffer = true;
return SB;
/// 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,
+MemoryBuffer *MemoryBuffer::getNewMemBuffer(size_t Size,
const char *BufferName) {
MemoryBuffer *SB = getNewUninitMemBuffer(Size, BufferName);
+ if (!SB) return 0;
memset(const_cast<char*>(SB->getBufferStart()), 0, Size+1);
return SB;
}
/// 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,
+MemoryBuffer *MemoryBuffer::getFileOrSTDIN(StringRef Filename,
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, "<stdin>");
+ int64_t FileSize,
+ struct stat *FileInfo) {
+ if (Filename == "-")
+ return getSTDIN();
+ return getFile(Filename, ErrStr, FileSize, FileInfo);
}
//===----------------------------------------------------------------------===//
-// MemoryBufferMMapFile implementation.
+// MemoryBuffer::getFile implementation.
//===----------------------------------------------------------------------===//
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 {
- sys::MappedFile File;
+ std::string Filename;
public:
- MemoryBufferMMapFile() {}
-
- bool open(const sys::Path &Filename, std::string *ErrStr);
+ MemoryBufferMMapFile(StringRef filename, const char *Pages, uint64_t Size)
+ : Filename(filename) {
+ init(Pages, Pages+Size);
+ }
virtual const char *getBufferIdentifier() const {
- return File.path().c_str();
+ return Filename.c_str();
}
- ~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))
- 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();
+ ~MemoryBufferMMapFile() {
+ sys::Path::UnMapFilePages(getBufferStart(), getBufferSize());
}
- return false;
-}
+};
-MemoryBufferMMapFile::~MemoryBufferMMapFile() {
- if (File.isMapped())
- File.unmap();
+/// FileCloser - RAII object to make sure an FD gets closed properly.
+class FileCloser {
+ int FD;
+public:
+ FileCloser(int FD) : FD(FD) {}
+ ~FileCloser() { ::close(FD); }
+};
}
-//===----------------------------------------------------------------------===//
-// MemoryBuffer::getFile implementation.
-//===----------------------------------------------------------------------===//
-
-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;
+MemoryBuffer *MemoryBuffer::getFile(StringRef Filename, std::string *ErrStr,
+ int64_t FileSize, struct stat *FileInfo) {
+ int OpenFlags = 0;
+#ifdef O_BINARY
+ OpenFlags |= O_BINARY; // Open input file in binary mode on win32.
+#endif
+ SmallString<256> PathBuf(Filename.begin(), Filename.end());
+ int FD = ::open(PathBuf.c_str(), O_RDONLY|OpenFlags);
+ if (FD == -1) {
+ if (ErrStr) *ErrStr = sys::StrError();
+ return 0;
}
+ FileCloser FC(FD); // Close FD on return.
- // 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 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;
+
+ // TODO: This should use fstat64 when available.
+ if (fstat(FD, FileInfoPtr) == -1) {
+ if (ErrStr) *ErrStr = sys::StrError();
+ return 0;
+ }
+ FileSize = FileInfoPtr->st_size;
}
- MemoryBuffer *SB = getNewUninitMemBuffer(FileSize, FilenameStart);
- char *BufPtr = const_cast<char*>(SB->getBufferStart());
- int FD = ::open(FilenameStart, O_RDONLY);
- if (FD == -1) {
- delete SB;
+ // 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);
+ }
+ }
+
+ MemoryBuffer *Buf = MemoryBuffer::getNewUninitMemBuffer(FileSize, Filename);
+ if (!Buf) {
+ // Failed to create a buffer.
+ if (ErrStr) *ErrStr = "could not allocate buffer";
return 0;
}
-
- unsigned BytesLeft = FileSize;
+
+ OwningPtr<MemoryBuffer> SB(Buf);
+ char *BufPtr = const_cast<char*>(SB->getBufferStart());
+
+ size_t BytesLeft = FileSize;
while (BytesLeft) {
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;
+ if (NumRead == -1) {
+ if (errno == EINTR)
+ continue;
+ // Error while reading.
+ if (ErrStr) *ErrStr = sys::StrError();
return 0;
+ } else if (NumRead == 0) {
+ // We hit EOF early, truncate and terminate buffer.
+ Buf->BufferEnd = BufPtr;
+ *BufPtr = 0;
+ return SB.take();
}
+ BytesLeft -= NumRead;
+ BufPtr += NumRead;
}
- close(FD);
-
- return SB;
-#endif
-}
+ return SB.take();
+}
//===----------------------------------------------------------------------===//
// MemoryBuffer::getSTDIN implementation.
MemoryBuffer *MemoryBuffer::getSTDIN() {
char Buffer[4096*4];
-
+
std::vector<char> FileData;
-
+
// 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();
- while (size_t ReadBytes = fread(Buffer, sizeof(char), 4096*4, stdin))
+ size_t ReadBytes;
+ do {
+ ReadBytes = fread(Buffer, sizeof(char), sizeof(Buffer), stdin);
FileData.insert(FileData.end(), Buffer, Buffer+ReadBytes);
+ } while (ReadBytes == sizeof(Buffer));
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;