func (s *Storage) ReceiveBlob(br blob.Ref, source io.Reader) (blob.SizedRef, error) {
var buf bytes.Buffer
size, err := io.Copy(&buf, source)
if err != nil {
return blob.SizedRef{}, err
}
b := bytes.NewReader(buf.Bytes())
fi, err := s.b.Upload(b, s.dirPrefix+br.String(), "")
if err != nil {
return blob.SizedRef{}, err
}
if int64(fi.ContentLength) != size {
return blob.SizedRef{}, fmt.Errorf("b2: expected ContentLength %d, got %d", size, fi.ContentLength)
}
if br.HashName() == "sha1" && fi.ContentSHA1 != br.Digest() {
return blob.SizedRef{}, fmt.Errorf("b2: expected ContentSHA1 %s, got %s", br.Digest(), fi.ContentSHA1)
}
if s.cache != nil {
// NoHash because it's already verified if we read it without
// errors from the source, and uploaded it without mismatch.
blobserver.ReceiveNoHash(s.cache, br, &buf)
}
return blob.SizedRef{Ref: br, Size: uint32(size)}, nil
}
func (sto *replicaStorage) ReceiveBlob(br blob.Ref, src io.Reader) (_ blob.SizedRef, err error) {
// Slurp the whole blob before replicating. Bounded by 16 MB anyway.
var buf bytes.Buffer
size, err := io.Copy(&buf, src)
if err != nil {
return
}
nReplicas := len(sto.replicas)
resc := make(chan sizedBlobAndError, nReplicas)
uploadToReplica := func(idx int, dst blobserver.BlobReceiver) {
// Using ReceiveNoHash because it's already been
// verified implicitly by the io.Copy above:
sb, err := blobserver.ReceiveNoHash(dst, br, bytes.NewReader(buf.Bytes()))
resc <- sizedBlobAndError{idx, sb, err}
}
for idx, replica := range sto.replicas {
go uploadToReplica(idx, replica)
}
nSuccess := 0
var fails []sizedBlobAndError
for range sto.replicas {
res := <-resc
switch {
case res.err == nil && int64(res.sb.Size) == size:
nSuccess++
if nSuccess == sto.minWritesForSuccess {
return res.sb, nil
}
case res.err == nil:
err = fmt.Errorf("replica: upload shard reported size %d, expected %d", res.sb.Size, size)
res.err = err
fails = append(fails, res)
default:
err = res.err
fails = append(fails, res)
}
}
for _, res := range fails {
log.Printf("replica: receiving blob %v, %d successes, %d failures; backend %s reported: %v",
br,
nSuccess, len(fails),
sto.replicaPrefixes[res.idx], res.err)
}
return
}
func uploadString(bs blobserver.StatReceiver, br blob.Ref, s string) (blob.Ref, error) {
if !br.Valid() {
panic("invalid blobref")
}
hasIt, err := serverHasBlob(bs, br)
if err != nil {
return blob.Ref{}, err
}
if hasIt {
return br, nil
}
_, err = blobserver.ReceiveNoHash(bs, br, strings.NewReader(s))
if err != nil {
return blob.Ref{}, err
}
return br, nil
}
func (s *Storage) ReceiveBlob(br blob.Ref, source io.Reader) (blob.SizedRef, error) {
var buf bytes.Buffer
size, err := io.Copy(&buf, source)
if err != nil {
return blob.SizedRef{}, err
}
err = s.client.PutObject(
&googlestorage.Object{Bucket: s.bucket, Key: s.dirPrefix + br.String()},
ioutil.NopCloser(bytes.NewReader(buf.Bytes())))
if err != nil {
return blob.SizedRef{}, err
}
if s.cache != nil {
// NoHash because it's already verified if we read it
// without errors on the io.Copy above.
blobserver.ReceiveNoHash(s.cache, br, bytes.NewReader(buf.Bytes()))
}
return blob.SizedRef{Ref: br, Size: uint32(size)}, nil
}
func (s *storage) ReceiveBlob(plainBR blob.Ref, source io.Reader) (sb blob.SizedRef, err error) {
iv := s.randIV()
stream := cipher.NewCTR(s.block, iv)
hash := plainBR.Hash()
var buf bytes.Buffer
// TODO: compress before encrypting?
buf.Write(iv) // TODO: write more structured header w/ version & IV length? or does that weaken it?
sw := cipher.StreamWriter{S: stream, W: &buf}
plainSize, err := io.Copy(io.MultiWriter(sw, hash), source)
if err != nil {
return sb, err
}
if !plainBR.HashMatches(hash) {
return sb, blobserver.ErrCorruptBlob
}
encBR := blob.SHA1FromBytes(buf.Bytes())
_, err = blobserver.Receive(s.blobs, encBR, bytes.NewReader(buf.Bytes()))
if err != nil {
log.Printf("encrypt: error writing encrypted blob %v (plaintext %v): %v", encBR, plainBR, err)
return sb, errors.New("encrypt: error writing encrypted blob")
}
meta := encodeMetaValue(uint32(plainSize), iv, encBR, buf.Len())
metaBlob := s.makeSingleMetaBlob(plainBR, meta)
_, err = blobserver.ReceiveNoHash(s.meta, blob.SHA1FromBytes(metaBlob), bytes.NewReader(metaBlob))
if err != nil {
log.Printf("encrypt: error writing encrypted meta for plaintext %v (encrypted blob %v): %v", plainBR, encBR, err)
return sb, errors.New("encrypt: error writing encrypted meta")
}
err = s.index.Set(plainBR.String(), meta)
if err != nil {
return sb, fmt.Errorf("encrypt: error updating index for encrypted %v (plaintext %v): %v", encBR, plainBR, err)
}
return blob.SizedRef{plainBR, uint32(plainSize)}, nil
}
func (s *Storage) ReceiveBlob(br blob.Ref, source io.Reader) (blob.SizedRef, error) {
var buf bytes.Buffer
size, err := io.Copy(&buf, source)
if err != nil {
return blob.SizedRef{}, err
}
// TODO(mpl): use context from caller, once one is available (issue 733)
w := s.client.Bucket(s.bucket).Object(s.dirPrefix + br.String()).NewWriter(context.TODO())
if _, err := io.Copy(w, ioutil.NopCloser(bytes.NewReader(buf.Bytes()))); err != nil {
return blob.SizedRef{}, err
}
if err := w.Close(); err != nil {
return blob.SizedRef{}, err
}
if s.cache != nil {
// NoHash because it's already verified if we read it
// without errors on the io.Copy above.
blobserver.ReceiveNoHash(s.cache, br, bytes.NewReader(buf.Bytes()))
}
return blob.SizedRef{Ref: br, Size: uint32(size)}, nil
}
func (sto *s3Storage) ReceiveBlob(b blob.Ref, source io.Reader) (sr blob.SizedRef, err error) {
var buf bytes.Buffer
md5h := md5.New()
size, err := io.Copy(io.MultiWriter(&buf, md5h), source)
if err != nil {
return sr, err
}
if faultReceive.FailErr(&err) {
return
}
err = sto.s3Client.PutObject(sto.dirPrefix+b.String(), sto.bucket, md5h, size, &buf)
if err != nil {
return sr, err
}
if sto.cache != nil {
// NoHash because it's already verified if we read it
// without errors on the io.Copy above.
blobserver.ReceiveNoHash(sto.cache, b, bytes.NewReader(buf.Bytes()))
}
return blob.SizedRef{Ref: b, Size: uint32(size)}, nil
}
// vivify verifies that all the chunks for the file described by fileblob are on the blobserver.
// It makes a planned permanode, signs it, and uploads it. It finally makes a camliContent claim
// on that permanode for fileblob, signs it, and uploads it to the blobserver.
func vivify(blobReceiver blobserver.BlobReceiveConfiger, fileblob blob.SizedRef) error {
sf, ok := blobReceiver.(blob.StreamingFetcher)
if !ok {
return fmt.Errorf("BlobReceiver is not a StreamingFetcher")
}
fetcher := blob.SeekerFromStreamingFetcher(sf)
fr, err := schema.NewFileReader(fetcher, fileblob.Ref)
if err != nil {
return fmt.Errorf("Filereader error for blobref %v: %v", fileblob.Ref.String(), err)
}
defer fr.Close()
h := sha1.New()
n, err := io.Copy(h, fr)
if err != nil {
return fmt.Errorf("Could not read all file of blobref %v: %v", fileblob.Ref.String(), err)
}
if n != fr.Size() {
return fmt.Errorf("Could not read all file of blobref %v. Wanted %v, got %v", fileblob.Ref.String(), fr.Size(), n)
}
config := blobReceiver.Config()
if config == nil {
return errors.New("blobReceiver has no config")
}
hf := config.HandlerFinder
if hf == nil {
return errors.New("blobReceiver config has no HandlerFinder")
}
JSONSignRoot, sh, err := hf.FindHandlerByType("jsonsign")
if err != nil || sh == nil {
return errors.New("jsonsign handler not found")
}
sigHelper, ok := sh.(*signhandler.Handler)
if !ok {
return errors.New("handler is not a JSON signhandler")
}
discoMap := sigHelper.DiscoveryMap(JSONSignRoot)
publicKeyBlobRef, ok := discoMap["publicKeyBlobRef"].(string)
if !ok {
return fmt.Errorf("Discovery: json decoding error: %v", err)
}
// The file schema must have a modtime to vivify, as the modtime is used for all three of:
// 1) the permanode's signature
// 2) the camliContent attribute claim's "claimDate"
// 3) the signature time of 2)
claimDate, err := time.Parse(time.RFC3339, fr.FileSchema().UnixMtime)
if err != nil {
return fmt.Errorf("While parsing modtime for file %v: %v", fr.FileSchema().FileName, err)
}
permanodeBB := schema.NewHashPlannedPermanode(h)
permanodeBB.SetSigner(blob.MustParse(publicKeyBlobRef))
permanodeBB.SetClaimDate(claimDate)
permanodeSigned, err := sigHelper.Sign(permanodeBB)
if err != nil {
return fmt.Errorf("Signing permanode %v: %v", permanodeSigned, err)
}
permanodeRef := blob.SHA1FromString(permanodeSigned)
_, err = blobserver.ReceiveNoHash(blobReceiver, permanodeRef, strings.NewReader(permanodeSigned))
if err != nil {
return fmt.Errorf("While uploading signed permanode %v, %v: %v", permanodeRef, permanodeSigned, err)
}
contentClaimBB := schema.NewSetAttributeClaim(permanodeRef, "camliContent", fileblob.Ref.String())
contentClaimBB.SetSigner(blob.MustParse(publicKeyBlobRef))
contentClaimBB.SetClaimDate(claimDate)
contentClaimSigned, err := sigHelper.Sign(contentClaimBB)
if err != nil {
return fmt.Errorf("Signing camliContent claim: %v", err)
}
contentClaimRef := blob.SHA1FromString(contentClaimSigned)
_, err = blobserver.ReceiveNoHash(blobReceiver, contentClaimRef, strings.NewReader(contentClaimSigned))
if err != nil {
return fmt.Errorf("While uploading signed camliContent claim %v, %v: %v", contentClaimRef, contentClaimSigned, err)
}
return nil
}
// trunc is a hint about which blob to truncate after. It may be zero.
// If the returned error is of type 'needsTruncatedAfterError', then
// the zip should be attempted to be written again, but truncating the
// data after the listed blob.
func (pk *packer) writeAZip(trunc blob.Ref) (err error) {
defer func() {
if e := recover(); e != nil {
if v, ok := e.(error); ok && err == nil {
err = v
} else {
panic(e)
}
}
}()
mf := Manifest{
WholeRef: pk.wholeRef,
WholeSize: pk.wholeSize,
WholePartIndex: len(pk.zips),
}
var zbuf bytes.Buffer
cw := &countWriter{w: &zbuf}
zw := zip.NewWriter(cw)
var approxSize = zipFixedOverhead // can't use zbuf.Len because zw buffers
var dataRefsWritten []blob.Ref
var dataBytesWritten int64
var schemaBlobSeen = map[blob.Ref]bool{}
var schemaBlobs []blob.Ref // to add after the main file
baseFileName := pk.fr.FileName()
if strings.Contains(baseFileName, "/") || strings.Contains(baseFileName, "\\") {
return fmt.Errorf("File schema blob %v filename had a slash in it: %q", pk.fr.SchemaBlobRef(), baseFileName)
}
fh := &zip.FileHeader{
Name: baseFileName,
Method: zip.Store, // uncompressed
}
fh.SetModTime(pk.fr.ModTime())
fh.SetMode(0644)
fw, err := zw.CreateHeader(fh)
check(err)
check(zw.Flush())
dataStart := cw.n
approxSize += zipPerEntryOverhead // for the first FileHeader w/ the data
zipMax := pk.s.maxZipBlobSize()
chunks := pk.chunksRemain
chunkWholeHash := blob.NewHash()
for len(chunks) > 0 {
dr := chunks[0] // the next chunk to maybe write
if trunc.Valid() && trunc == dr {
if approxSize == 0 {
return errors.New("first blob is too large to pack, once you add the zip overhead")
}
break
}
schemaBlobsSave := schemaBlobs
for _, parent := range pk.schemaParent[dr] {
if !schemaBlobSeen[parent] {
schemaBlobSeen[parent] = true
schemaBlobs = append(schemaBlobs, parent)
approxSize += int(pk.schemaBlob[parent].Size()) + zipPerEntryOverhead
}
}
thisSize := pk.dataSize[dr]
approxSize += int(thisSize)
if approxSize+mf.approxSerializedSize() > zipMax {
if fn := testHookStopBeforeOverflowing; fn != nil {
fn()
}
schemaBlobs = schemaBlobsSave // restore it
break
}
// Copy the data to the zip.
rc, size, err := pk.s.Fetch(dr)
check(err)
if size != thisSize {
rc.Close()
return errors.New("unexpected size")
}
if n, err := io.Copy(io.MultiWriter(fw, chunkWholeHash), rc); err != nil || n != int64(size) {
rc.Close()
return fmt.Errorf("copy to zip = %v, %v; want %v bytes", n, err, size)
}
rc.Close()
dataRefsWritten = append(dataRefsWritten, dr)
dataBytesWritten += int64(size)
chunks = chunks[1:]
}
mf.DataBlobsOrigin = blob.RefFromHash(chunkWholeHash)
// zipBlobs is where a schema or data blob is relative to the beginning
// of the zip file.
var zipBlobs []BlobAndPos
var dataOffset int64
for _, br := range dataRefsWritten {
size := pk.dataSize[br]
mf.DataBlobs = append(mf.DataBlobs, BlobAndPos{blob.SizedRef{Ref: br, Size: size}, dataOffset})
zipBlobs = append(zipBlobs, BlobAndPos{blob.SizedRef{Ref: br, Size: size}, dataStart + dataOffset})
dataOffset += int64(size)
}
for _, br := range schemaBlobs {
fw, err := zw.CreateHeader(&zip.FileHeader{
Name: "camlistore/" + br.String() + ".json",
Method: zip.Store, // uncompressed
})
check(err)
check(zw.Flush())
b := pk.schemaBlob[br]
zipBlobs = append(zipBlobs, BlobAndPos{blob.SizedRef{Ref: br, Size: b.Size()}, cw.n})
rc := b.Open()
n, err := io.Copy(fw, rc)
rc.Close()
check(err)
if n != int64(b.Size()) {
return fmt.Errorf("failed to write all of schema blob %v: %d bytes, not wanted %d", br, n, b.Size())
}
}
// Manifest file
fw, err = zw.Create(zipManifestPath)
check(err)
enc, err := json.MarshalIndent(mf, "", " ")
check(err)
_, err = fw.Write(enc)
check(err)
err = zw.Close()
check(err)
if zbuf.Len() > zipMax {
// We guessed wrong. Back up. Find out how many blobs we went over.
overage := zbuf.Len() - zipMax
for i := len(dataRefsWritten) - 1; i >= 0; i-- {
dr := dataRefsWritten[i]
if overage <= 0 {
return needsTruncatedAfterError{dr}
}
overage -= int(pk.dataSize[dr])
}
return errors.New("file is unpackable; first blob is too big to fit")
}
zipRef := blob.SHA1FromBytes(zbuf.Bytes())
zipSB, err := blobserver.ReceiveNoHash(pk.s.large, zipRef, bytes.NewReader(zbuf.Bytes()))
if err != nil {
return err
}
bm := pk.s.meta.BeginBatch()
bm.Set(fmt.Sprintf("%s%s:%d", wholeMetaPrefix, pk.wholeRef, len(pk.zips)),
fmt.Sprintf("%s %d %d %d",
zipRef,
dataStart,
pk.wholeBytesWritten,
dataBytesWritten))
pk.wholeBytesWritten += dataBytesWritten
pk.zips = append(pk.zips, writtenZip{
SizedRef: zipSB,
dataRefs: dataRefsWritten,
})
for _, zb := range zipBlobs {
bm.Set(blobMetaPrefix+zb.Ref.String(), fmt.Sprintf("%d %v %d", zb.Size, zipRef, zb.Offset))
}
if err := pk.s.meta.CommitBatch(bm); err != nil {
return err
}
// Delete from small
if !pk.s.skipDelete {
toDelete := make([]blob.Ref, 0, len(dataRefsWritten)+len(schemaBlobs))
toDelete = append(toDelete, dataRefsWritten...)
toDelete = append(toDelete, schemaBlobs...)
if err := pk.s.small.RemoveBlobs(toDelete); err != nil {
// Can't really do anything about it and doesn't really matter, so
// just log for now.
pk.s.Logf("Error removing blobs from %s: %v", pk.s.small, err)
}
}
// On success, consume the chunks we wrote from pk.chunksRemain.
pk.chunksRemain = pk.chunksRemain[len(dataRefsWritten):]
return nil
}
