in backuptar/tar.go [127:319]
func WriteTarFileFromBackupStream(t *tar.Writer, r io.Reader, name string, size int64, fileInfo *winio.FileBasicInfo) error {
name = filepath.ToSlash(name)
hdr := BasicInfoHeader(name, size, fileInfo)
// If r can be seeked, then this function is two-pass: pass 1 collects the
// tar header data, and pass 2 copies the data stream. If r cannot be
// seeked, then some header data (in particular EAs) will be silently lost.
var (
restartPos int64
err error
)
sr, readTwice := r.(io.Seeker)
if readTwice {
if restartPos, err = sr.Seek(0, io.SeekCurrent); err != nil {
readTwice = false
}
}
br := winio.NewBackupStreamReader(r)
var dataHdr *winio.BackupHeader
for dataHdr == nil {
bhdr, err := br.Next()
if err == io.EOF {
break
}
if err != nil {
return err
}
switch bhdr.Id {
case winio.BackupData:
hdr.Mode |= c_ISREG
if !readTwice {
dataHdr = bhdr
}
case winio.BackupSecurity:
sd, err := ioutil.ReadAll(br)
if err != nil {
return err
}
hdr.PAXRecords[hdrRawSecurityDescriptor] = base64.StdEncoding.EncodeToString(sd)
case winio.BackupReparseData:
hdr.Mode |= c_ISLNK
hdr.Typeflag = tar.TypeSymlink
reparseBuffer, err := ioutil.ReadAll(br)
rp, err := winio.DecodeReparsePoint(reparseBuffer)
if err != nil {
return err
}
if rp.IsMountPoint {
hdr.PAXRecords[hdrMountPoint] = "1"
}
hdr.Linkname = rp.Target
case winio.BackupEaData:
eab, err := ioutil.ReadAll(br)
if err != nil {
return err
}
eas, err := winio.DecodeExtendedAttributes(eab)
if err != nil {
return err
}
for _, ea := range eas {
// Use base64 encoding for the binary value. Note that there
// is no way to encode the EA's flags, since their use doesn't
// make any sense for persisted EAs.
hdr.PAXRecords[hdrEaPrefix+ea.Name] = base64.StdEncoding.EncodeToString(ea.Value)
}
case winio.BackupAlternateData, winio.BackupLink, winio.BackupPropertyData, winio.BackupObjectId, winio.BackupTxfsData:
// ignore these streams
default:
return fmt.Errorf("%s: unknown stream ID %d", name, bhdr.Id)
}
}
err = t.WriteHeader(hdr)
if err != nil {
return err
}
if readTwice {
// Get back to the data stream.
if _, err = sr.Seek(restartPos, io.SeekStart); err != nil {
return err
}
for dataHdr == nil {
bhdr, err := br.Next()
if err == io.EOF {
break
}
if err != nil {
return err
}
if bhdr.Id == winio.BackupData {
dataHdr = bhdr
}
}
}
// The logic for copying file contents is fairly complicated due to the need for handling sparse files,
// and the weird ways they are represented by BackupRead. A normal file will always either have a data stream
// with size and content, or no data stream at all (if empty). However, for a sparse file, the content can also
// be represented using a series of sparse block streams following the data stream. Additionally, the way sparse
// files are handled by BackupRead has changed in the OS recently. The specifics of the representation are described
// in the list at the bottom of this block comment.
//
// Sparse files can be represented in four different ways, based on the specifics of the file.
// - Size = 0:
// Previously: BackupRead yields no data stream and no sparse block streams.
// Recently: BackupRead yields a data stream with size = 0. There are no following sparse block streams.
// - Size > 0, no allocated ranges:
// BackupRead yields a data stream with size = 0. Following is a single sparse block stream with
// size = 0 and offset = <file size>.
// - Size > 0, one allocated range:
// BackupRead yields a data stream with size = <file size> containing the file contents. There are no
// sparse block streams. This is the case if you take a normal file with contents and simply set the
// sparse flag on it.
// - Size > 0, multiple allocated ranges:
// BackupRead yields a data stream with size = 0. Following are sparse block streams for each allocated
// range of the file containing the range contents. Finally there is a sparse block stream with
// size = 0 and offset = <file size>.
if dataHdr != nil {
// A data stream was found. Copy the data.
// We assume that we will either have a data stream size > 0 XOR have sparse block streams.
if dataHdr.Size > 0 || (dataHdr.Attributes&winio.StreamSparseAttributes) == 0 {
if size != dataHdr.Size {
return fmt.Errorf("%s: mismatch between file size %d and header size %d", name, size, dataHdr.Size)
}
if _, err = io.Copy(t, br); err != nil {
return fmt.Errorf("%s: copying contents from data stream: %s", name, err)
}
} else if size > 0 {
// As of a recent OS change, BackupRead now returns a data stream for empty sparse files.
// These files have no sparse block streams, so skip the copySparse call if file size = 0.
if err = copySparse(t, br); err != nil {
return fmt.Errorf("%s: copying contents from sparse block stream: %s", name, err)
}
}
}
// Look for streams after the data stream. The only ones we handle are alternate data streams.
// Other streams may have metadata that could be serialized, but the tar header has already
// been written. In practice, this means that we don't get EA or TXF metadata.
for {
bhdr, err := br.Next()
if err == io.EOF {
break
}
if err != nil {
return err
}
switch bhdr.Id {
case winio.BackupAlternateData:
altName := bhdr.Name
if strings.HasSuffix(altName, ":$DATA") {
altName = altName[:len(altName)-len(":$DATA")]
}
if (bhdr.Attributes & winio.StreamSparseAttributes) == 0 {
hdr = &tar.Header{
Format: hdr.Format,
Name: name + altName,
Mode: hdr.Mode,
Typeflag: tar.TypeReg,
Size: bhdr.Size,
ModTime: hdr.ModTime,
AccessTime: hdr.AccessTime,
ChangeTime: hdr.ChangeTime,
}
err = t.WriteHeader(hdr)
if err != nil {
return err
}
_, err = io.Copy(t, br)
if err != nil {
return err
}
} else {
// Unsupported for now, since the size of the alternate stream is not present
// in the backup stream until after the data has been read.
return fmt.Errorf("%s: tar of sparse alternate data streams is unsupported", name)
}
case winio.BackupEaData, winio.BackupLink, winio.BackupPropertyData, winio.BackupObjectId, winio.BackupTxfsData:
// ignore these streams
default:
return fmt.Errorf("%s: unknown stream ID %d after data", name, bhdr.Id)
}
}
return nil
}