Skip to content

Commit

Permalink
Fix estargz compression loses the original tar metadata
Browse files Browse the repository at this point in the history
Currently, eStargz compression doesn't preserve the original tar metadata
(header bytes and their order). This causes failure of `TestGetRemote` because
an uncompressed blob converted from a gzip blob provides different digset
against the one converted from eStargz blob even if their original tar (computed
by differ) are the same.
This commit solves this issue by fixing eStargz to preserve original tar's
metadata that is modified by eStargz.

Signed-off-by: Kohei Tokunaga <[email protected]>
  • Loading branch information
ktock committed Sep 8, 2021
1 parent ea773f6 commit d90d0fa
Show file tree
Hide file tree
Showing 142 changed files with 6,572 additions and 4,690 deletions.
8 changes: 4 additions & 4 deletions cache/blobs.go
Original file line number Diff line number Diff line change
Expand Up @@ -78,7 +78,7 @@ func computeBlobChain(ctx context.Context, sr *immutableRef, createIfNeeded bool
case compression.Gzip:
mediaType = ocispecs.MediaTypeImageLayerGzip
case compression.EStargz:
compressorFunc, finalize = writeEStargz()
compressorFunc, finalize = compressEStargz()
mediaType = ocispecs.MediaTypeImageLayerGzip
case compression.Zstd:
compressorFunc = zstdWriter
Expand Down Expand Up @@ -314,7 +314,7 @@ func ensureCompression(ctx context.Context, ref *immutableRef, compressionType c
}

// Resolve converters
layerConvertFunc, err := getConverter(desc, compressionType)
layerConvertFunc, err := getConverter(ctx, ref.cm.ContentStore, desc, compressionType)
if err != nil {
return nil, err
} else if layerConvertFunc == nil {
Expand Down Expand Up @@ -343,12 +343,12 @@ func ensureCompression(ctx context.Context, ref *immutableRef, compressionType c
}
newDesc, err := layerConvertFunc(ctx, ref.cm.ContentStore, desc)
if err != nil {
return nil, err
return nil, errors.Wrapf(err, "failed to convert")
}

// Start to track converted layer
if err := ref.addCompressionBlob(ctx, *newDesc, compressionType); err != nil {
return nil, err
return nil, errors.Wrapf(err, "failed to add compression blob")
}
return nil, nil
})
Expand Down
85 changes: 57 additions & 28 deletions cache/converter.go
Original file line number Diff line number Diff line change
Expand Up @@ -13,30 +13,32 @@ import (
"github.com/containerd/containerd/images/converter"
"github.com/containerd/containerd/labels"
"github.com/klauspost/compress/zstd"
"github.com/moby/buildkit/identity"
"github.com/moby/buildkit/util/compression"
digest "github.com/opencontainers/go-digest"
ocispecs "github.com/opencontainers/image-spec/specs-go/v1"
"github.com/pkg/errors"
)

// needsConversion indicates whether a conversion is needed for the specified mediatype to
// needsConversion indicates whether a conversion is needed for the specified descriptor to
// be the compressionType.
func needsConversion(mediaType string, compressionType compression.Type) (bool, error) {
func needsConversion(ctx context.Context, cs content.Store, desc ocispecs.Descriptor, compressionType compression.Type) (bool, error) {
mediaType := desc.MediaType
switch compressionType {
case compression.Uncompressed:
if !images.IsLayerType(mediaType) || compression.FromMediaType(mediaType) == compression.Uncompressed {
return false, nil
}
case compression.Gzip:
if !images.IsLayerType(mediaType) || compression.FromMediaType(mediaType) == compression.Gzip {
if (!images.IsLayerType(mediaType) || compression.FromMediaType(mediaType) == compression.Gzip) && !isEStargz(ctx, cs, desc.Digest) {
return false, nil
}
case compression.Zstd:
if !images.IsLayerType(mediaType) || compression.FromMediaType(mediaType) == compression.Zstd {
return false, nil
}
case compression.EStargz:
if !images.IsLayerType(mediaType) {
if !images.IsLayerType(mediaType) || isEStargz(ctx, cs, desc.Digest) {
return false, nil
}
default:
Expand All @@ -47,9 +49,9 @@ func needsConversion(mediaType string, compressionType compression.Type) (bool,

// getConverter returns converter function according to the specified compression type.
// If no conversion is needed, this returns nil without error.
func getConverter(desc ocispecs.Descriptor, compressionType compression.Type) (converter.ConvertFunc, error) {
if needs, err := needsConversion(desc.MediaType, compressionType); err != nil {
return nil, err
func getConverter(ctx context.Context, cs content.Store, desc ocispecs.Descriptor, compressionType compression.Type) (converter.ConvertFunc, error) {
if needs, err := needsConversion(ctx, cs, desc, compressionType); err != nil {
return nil, errors.Wrapf(err, "failed to determine conversion needs")
} else if !needs {
// No conversion. No need to return an error here.
return nil, nil
Expand All @@ -61,7 +63,24 @@ func getConverter(desc ocispecs.Descriptor, compressionType compression.Type) (c
switch from {
case compression.Uncompressed:
case compression.Gzip, compression.Zstd:
c.decompress = cdcompression.DecompressStream
c.decompress = func(ctx context.Context, desc ocispecs.Descriptor) (r io.ReadCloser, err error) {
ra, err := cs.ReaderAt(ctx, desc)
if err != nil {
return nil, err
}
if isEStargz(ctx, cs, desc.Digest) {
r, err = decompressEStargz(io.NewSectionReader(ra, 0, ra.Size()))
if err != nil {
return nil, err
}
} else {
r, err = cdcompression.DecompressStream(io.NewSectionReader(ra, 0, ra.Size()))
if err != nil {
return nil, err
}
}
return &readCloser{r, ra.Close}, nil
}
default:
return nil, errors.Errorf("unsupported source compression type %q from mediatype %q", from, desc.MediaType)
}
Expand All @@ -77,7 +96,7 @@ func getConverter(desc ocispecs.Descriptor, compressionType compression.Type) (c
return zstd.NewWriter(w)
}
case compression.EStargz:
compressorFunc, finalize := writeEStargz()
compressorFunc, finalize := compressEStargz()
c.compress = func(w io.Writer) (io.WriteCloser, error) {
return compressorFunc(w, ocispecs.MediaTypeImageLayerGzip)
}
Expand All @@ -91,27 +110,15 @@ func getConverter(desc ocispecs.Descriptor, compressionType compression.Type) (c

type conversion struct {
target compression.Type
decompress func(io.Reader) (cdcompression.DecompressReadCloser, error)
decompress func(context.Context, ocispecs.Descriptor) (io.ReadCloser, error)
compress func(w io.Writer) (io.WriteCloser, error)
finalize func(context.Context, content.Store) (map[string]string, error)
}

func (c *conversion) convert(ctx context.Context, cs content.Store, desc ocispecs.Descriptor) (*ocispecs.Descriptor, error) {
// prepare the source and destination
info, err := cs.Info(ctx, desc.Digest)
if err != nil {
return nil, err
}
labelz := info.Labels
if labelz == nil {
labelz = make(map[string]string)
}
ra, err := cs.ReaderAt(ctx, desc)
if err != nil {
return nil, err
}
defer ra.Close()
ref := fmt.Sprintf("convert-from-%s-to-%s", desc.Digest, c.target.String())
labelz := make(map[string]string)
ref := fmt.Sprintf("convert-from-%s-to-%s-%s", desc.Digest, c.target.String(), identity.NewID())
w, err := cs.Writer(ctx, content.WithRef(ref))
if err != nil {
return nil, err
Expand All @@ -133,9 +140,16 @@ func (c *conversion) convert(ctx context.Context, cs content.Store, desc ocispec

// convert this layer
diffID := digest.Canonical.Digester()
var rdr io.Reader = io.NewSectionReader(ra, 0, ra.Size())
if c.decompress != nil {
rc, err := c.decompress(rdr)
var rdr io.Reader
if c.decompress == nil {
ra, err := cs.ReaderAt(ctx, desc)
if err != nil {
return nil, err
}
defer ra.Close()
rdr = io.NewSectionReader(ra, 0, ra.Size())
} else {
rc, err := c.decompress(ctx, desc)
if err != nil {
return nil, err
}
Expand All @@ -157,7 +171,7 @@ func (c *conversion) convert(ctx context.Context, cs content.Store, desc ocispec
if err := w.Close(); err != nil {
return nil, err
}
info, err = cs.Info(ctx, w.Digest())
info, err := cs.Info(ctx, w.Digest())
if err != nil {
return nil, err
}
Expand All @@ -166,6 +180,7 @@ func (c *conversion) convert(ctx context.Context, cs content.Store, desc ocispec
newDesc.MediaType = c.target.DefaultMediaType()
newDesc.Digest = info.Digest
newDesc.Size = info.Size
newDesc.Annotations = nil
if c.finalize != nil {
a, err := c.finalize(ctx, cs)
if err != nil {
Expand All @@ -180,3 +195,17 @@ func (c *conversion) convert(ctx context.Context, cs content.Store, desc ocispec
}
return &newDesc, nil
}

type readCloser struct {
io.ReadCloser
closeFunc func() error
}

func (rc *readCloser) Close() error {
err1 := rc.ReadCloser.Close()
err2 := rc.closeFunc()
if err1 != nil {
return errors.Wrapf(err1, "failed to close: %v", err2)
}
return err2
}
106 changes: 78 additions & 28 deletions cache/estargz.go
Original file line number Diff line number Diff line change
Expand Up @@ -11,76 +11,122 @@ import (
"github.com/containerd/stargz-snapshotter/estargz"
"github.com/moby/buildkit/util/compression"
digest "github.com/opencontainers/go-digest"
ocispecs "github.com/opencontainers/image-spec/specs-go/v1"
"github.com/pkg/errors"
)

var eStargzAnnotations = []string{estargz.TOCJSONDigestAnnotation, estargz.StoreUncompressedSizeAnnotation}

// writeEStargz writes the passed blobs stream as an eStargz-compressed blob.
// getAnnotations function returns all eStargz annotations.
func writeEStargz() (compressorFunc compressor, finalize func(context.Context, content.Store) (map[string]string, error)) {
annotations := make(map[string]string)
var bInfo blobInfo
// compressEStargz writes the passed blobs stream as an eStargz-compressed blob.
// finalize function finalizes the written blob metadata and returns all eStargz annotations.
func compressEStargz() (compressorFunc compressor, finalize func(context.Context, content.Store) (map[string]string, error)) {
var cInfo *compressionInfo
var writeErr error
var mu sync.Mutex
return func(dest io.Writer, requiredMediaType string) (io.WriteCloser, error) {
if compression.FromMediaType(requiredMediaType) != compression.Gzip {
return nil, errors.Errorf("unsupported media type for estargz compressor %q", requiredMediaType)
return nil, fmt.Errorf("unsupported media type for estargz compressor %q", requiredMediaType)
}
done := make(chan struct{})

pr, pw := io.Pipe()
go func() {
go func() (retErr error) {
defer close(done)
defer pr.Close()
cw, bInfoCh := calculateBlob()
defer cw.Close()
w := estargz.NewWriter(io.MultiWriter(dest, cw))
if err := w.AppendTar(pr); err != nil {
defer func() {
if retErr != nil {
mu.Lock()
writeErr = retErr
mu.Unlock()
}
}()

blobInfoW, bInfoCh := calculateBlobInfo()
defer blobInfoW.Close()
w := estargz.NewWriter(io.MultiWriter(dest, blobInfoW))

// Using lossless API here to make sure that decompressEStargz provides the exact
// same tar as the original.
//
// Note that we don't support eStragz compression for tar that contains a file named
// `stargz.index.json` because we cannot create eStargz in loseless way for such blob
// (we must overwrite stargz.index.json file).
if err := w.AppendTarLossLess(pr); err != nil {
pr.CloseWithError(err)
return
return err
}
tocDgst, err := w.Close()
if err != nil {
pr.CloseWithError(err)
return
return err
}
if err := cw.Close(); err != nil {
if err := blobInfoW.Close(); err != nil {
pr.CloseWithError(err)
return err
}
if err != nil {
pr.CloseWithError(err)
return
return err
}
bInfo := <-bInfoCh
mu.Lock()
bInfo = <-bInfoCh
annotations[estargz.TOCJSONDigestAnnotation] = tocDgst.String()
annotations[estargz.StoreUncompressedSizeAnnotation] = fmt.Sprintf("%d", bInfo.uncompressedSize)
cInfo = &compressionInfo{bInfo, tocDgst}
mu.Unlock()
return nil
}()
return &writeCloser{pw, func() error {
<-done // wait until the write completes
return nil
}}, nil
}, func(ctx context.Context, cs content.Store) (map[string]string, error) {
a := make(map[string]string)
mu.Lock()
bInfo := bInfo
for k, v := range annotations {
a[k] = v
}
cInfo, writeErr := cInfo, writeErr
mu.Unlock()
info, err := cs.Info(ctx, bInfo.compressedDigest)
if cInfo == nil {
if writeErr != nil {
return nil, errors.Wrapf(writeErr, "cannot finalize due to write error")
}
return nil, errors.Errorf("cannot finalize (reason unknown)")
}

// Fill necessary labels
info, err := cs.Info(ctx, cInfo.compressedDigest)
if err != nil {
return nil, errors.Wrap(err, "failed to get info from content store")
}
if info.Labels == nil {
info.Labels = make(map[string]string)
}
info.Labels[containerdUncompressed] = bInfo.uncompressedDigest.String()
info.Labels[containerdUncompressed] = cInfo.uncompressedDigest.String()
if _, err := cs.Update(ctx, info, "labels."+containerdUncompressed); err != nil {
return nil, err
}
a[containerdUncompressed] = bInfo.uncompressedDigest.String()

// Fill annotations
a := make(map[string]string)
a[estargz.TOCJSONDigestAnnotation] = cInfo.tocDigest.String()
a[estargz.StoreUncompressedSizeAnnotation] = fmt.Sprintf("%d", cInfo.uncompressedSize)
a[containerdUncompressed] = cInfo.uncompressedDigest.String()
return a, nil
}
}

// isEStargz returns true when the specified digest of content exists in
// the content store as eStargz.
func isEStargz(ctx context.Context, cs content.Store, dgst digest.Digest) bool {
r, err := cs.ReaderAt(ctx, ocispecs.Descriptor{Digest: dgst})
if err != nil {
return false
}
defer r.Close()
_, _, err = estargz.OpenFooter(io.NewSectionReader(r, 0, r.Size()))
return err == nil
}

func decompressEStargz(r *io.SectionReader) (io.ReadCloser, error) {
return estargz.Unpack(r, new(estargz.GzipDecompressor))
}

type writeCloser struct {
io.WriteCloser
closeFunc func() error
Expand Down Expand Up @@ -114,13 +160,18 @@ func (c *counter) size() (n int64) {
return
}

type compressionInfo struct {
blobInfo
tocDigest digest.Digest
}

type blobInfo struct {
compressedDigest digest.Digest
uncompressedDigest digest.Digest
uncompressedSize int64
}

func calculateBlob() (io.WriteCloser, chan blobInfo) {
func calculateBlobInfo() (io.WriteCloser, chan blobInfo) {
res := make(chan blobInfo)
pr, pw := io.Pipe()
go func() {
Expand All @@ -143,7 +194,6 @@ func calculateBlob() (io.WriteCloser, chan blobInfo) {
return
}
res <- blobInfo{dgstr.Digest(), diffID.Digest(), c.size()}

}()
return pw, res
}
Loading

0 comments on commit d90d0fa

Please sign in to comment.