vendor: update c/common to latest

Includes some netns cleanup fixes.

Signed-off-by: Paul Holzinger <pholzing@redhat.com>
This commit is contained in:
Paul Holzinger
2024-08-09 13:08:59 +02:00
parent 4aaa5cb6f0
commit be41ee4131
22 changed files with 676 additions and 363 deletions

View File

@ -181,7 +181,7 @@ func (n *Netns) cleanup() error {
if err := n.cleanupRootlessNetns(); err != nil {
multiErr = multierror.Append(multiErr, wrapError("kill network process", err))
}
if err := os.RemoveAll(n.dir); err != nil {
if err := os.RemoveAll(n.dir); err != nil && !errors.Is(err, fs.ErrNotExist) {
multiErr = multierror.Append(multiErr, wrapError("remove rootless netns dir", err))
}
@ -280,6 +280,11 @@ func (n *Netns) setupSlirp4netns(nsPath string) error {
func (n *Netns) cleanupRootlessNetns() error {
pidFile := n.getPath(rootlessNetNsConnPidFile)
pid, err := readPidFile(pidFile)
// do not hard error if the file dos not exists, cleanup should be idempotent
if errors.Is(err, fs.ErrNotExist) {
logrus.Debugf("Rootless netns conn pid file does not exists %s", pidFile)
return nil
}
if err == nil {
// kill the slirp/pasta process so we do not leak it
err = unix.Kill(pid, unix.SIGTERM)
@ -512,14 +517,14 @@ func (n *Netns) mountCNIVarDir() error {
return nil
}
func (n *Netns) runInner(toRun func() error) (err error) {
func (n *Netns) runInner(toRun func() error, cleanup bool) (err error) {
nsRef, newNs, err := n.getOrCreateNetns()
if err != nil {
return err
}
defer nsRef.Close()
// If a new netns was created make sure to clean it up again on an error to not leak it.
if newNs {
// If a new netns was created make sure to clean it up again on an error to not leak it if requested.
if newNs && cleanup {
defer func() {
if err != nil {
if err := n.cleanup(); err != nil {
@ -555,7 +560,7 @@ func (n *Netns) runInner(toRun func() error) (err error) {
}
func (n *Netns) Setup(nets int, toRun func() error) error {
err := n.runInner(toRun)
err := n.runInner(toRun, true)
if err != nil {
return err
}
@ -564,25 +569,22 @@ func (n *Netns) Setup(nets int, toRun func() error) error {
}
func (n *Netns) Teardown(nets int, toRun func() error) error {
var multiErr *multierror.Error
count, countErr := refCount(n.dir, -nets)
if countErr != nil {
multiErr = multierror.Append(multiErr, countErr)
}
err := n.runInner(toRun)
err := n.runInner(toRun, true)
if err != nil {
multiErr = multierror.Append(multiErr, err)
return err
}
// decrement only if teardown didn't fail, podman will call us again on errors so we should not double decrement
count, err := refCount(n.dir, -nets)
if err != nil {
return err
}
// only cleanup if the ref count did not throw an error
if count == 0 && countErr == nil {
err = n.cleanup()
if err != nil {
multiErr = multierror.Append(multiErr, wrapError("cleanup", err))
}
// cleanup when ref count is 0
if count == 0 {
return n.cleanup()
}
return multiErr.ErrorOrNil()
return nil
}
// Run any long running function in the userns.
@ -604,7 +606,7 @@ func (n *Netns) Run(lock *lockfile.LockFile, toRun func() error) error {
return err
}
inErr := n.runInner(inner)
inErr := n.runInner(inner, false)
// make sure to always reset the ref counter afterwards
count, err := refCount(n.dir, -1)
if err != nil {
@ -614,9 +616,8 @@ func (n *Netns) Run(lock *lockfile.LockFile, toRun func() error) error {
logrus.Errorf("Failed to decrement ref count: %v", err)
return inErr
}
// runInner() already cleans up the netns when it created a new one on errors
// so we only need to do that if there was no error.
if inErr == nil && count == 0 {
if count == 0 {
err = n.cleanup()
if err != nil {
return wrapError("cleanup", err)

View File

@ -28,10 +28,12 @@ import (
"runtime"
"strings"
"sync"
"time"
"github.com/containernetworking/plugins/pkg/ns"
"github.com/containers/storage/pkg/homedir"
"github.com/containers/storage/pkg/unshare"
"github.com/sirupsen/logrus"
"golang.org/x/sys/unix"
)
@ -177,26 +179,35 @@ func newNSPath(nsPath string) (ns.NetNS, error) {
// UnmountNS unmounts the given netns path
func UnmountNS(nsPath string) error {
var rErr error
// Only unmount if it's been bind-mounted (don't touch namespaces in /proc...)
if !strings.HasPrefix(nsPath, "/proc/") {
if err := unix.Unmount(nsPath, unix.MNT_DETACH); err != nil {
// Do not return here, always try to remove below.
// This is important in case podman now is in a new userns compared to
// when the netns was created. The umount will fail EINVAL but removing
// the file will work and the kernel will destroy the bind mount in the
// other ns because of this. We also need it so pasta doesn't leak.
rErr = fmt.Errorf("failed to unmount NS: at %s: %w", nsPath, err)
// EINVAL means the path exists but is not mounted, just try to remove the path below
if err := unix.Unmount(nsPath, unix.MNT_DETACH); err != nil && !errors.Is(err, unix.EINVAL) {
// If path does not exists we can return without error as we have nothing to do.
if errors.Is(err, unix.ENOENT) {
return nil
}
return fmt.Errorf("failed to unmount NS: at %s: %w", nsPath, err)
}
if err := os.Remove(nsPath); err != nil {
err := fmt.Errorf("failed to remove ns path: %w", err)
if rErr != nil {
err = fmt.Errorf("%v, %w", err, rErr)
for {
if err := os.Remove(nsPath); err != nil {
if errors.Is(err, unix.EBUSY) {
// mount is still busy, sleep a moment and try again to remove
logrus.Debugf("Netns %s still busy, try removing it again in 10ms", nsPath)
time.Sleep(10 * time.Millisecond)
continue
}
// If path does not exists we can return without error.
if errors.Is(err, unix.ENOENT) {
break
}
return fmt.Errorf("failed to remove ns path: %w", err)
}
rErr = err
break
}
}
return rErr
return nil
}

View File

@ -11,6 +11,7 @@ import (
"github.com/containers/image/v5/pkg/compression"
compressiontypes "github.com/containers/image/v5/pkg/compression/types"
"github.com/containers/image/v5/types"
chunkedToc "github.com/containers/storage/pkg/chunked/toc"
imgspecv1 "github.com/opencontainers/image-spec/specs-go/v1"
"github.com/sirupsen/logrus"
)
@ -34,10 +35,10 @@ var (
// bpDetectCompressionStepData contains data that the copy pipeline needs about the “detect compression” step.
type bpDetectCompressionStepData struct {
isCompressed bool
format compressiontypes.Algorithm // Valid if isCompressed
decompressor compressiontypes.DecompressorFunc // Valid if isCompressed
srcCompressorName string // Compressor name to possibly record in the blob info cache for the source blob.
isCompressed bool
format compressiontypes.Algorithm // Valid if isCompressed
decompressor compressiontypes.DecompressorFunc // Valid if isCompressed
srcCompressorBaseVariantName string // Compressor name to possibly record in the blob info cache for the source blob.
}
// blobPipelineDetectCompressionStep updates *stream to detect its current compression format.
@ -57,9 +58,9 @@ func blobPipelineDetectCompressionStep(stream *sourceStream, srcInfo types.BlobI
decompressor: decompressor,
}
if res.isCompressed {
res.srcCompressorName = format.Name()
res.srcCompressorBaseVariantName = format.BaseVariantName()
} else {
res.srcCompressorName = internalblobinfocache.Uncompressed
res.srcCompressorBaseVariantName = internalblobinfocache.Uncompressed
}
if expectedBaseFormat, known := expectedBaseCompressionFormats[stream.info.MediaType]; known && res.isCompressed && format.BaseVariantName() != expectedBaseFormat.Name() {
@ -70,13 +71,13 @@ func blobPipelineDetectCompressionStep(stream *sourceStream, srcInfo types.BlobI
// bpCompressionStepData contains data that the copy pipeline needs about the compression step.
type bpCompressionStepData struct {
operation bpcOperation // What we are actually doing
uploadedOperation types.LayerCompression // Operation to use for updating the blob metadata (matching the end state, not necessarily what we do)
uploadedAlgorithm *compressiontypes.Algorithm // An algorithm parameter for the compressionOperation edits.
uploadedAnnotations map[string]string // Compression-related annotations that should be set on the uploaded blob. WARNING: This is only set after the srcStream.reader is fully consumed.
srcCompressorName string // Compressor name to record in the blob info cache for the source blob.
uploadedCompressorName string // Compressor name to record in the blob info cache for the uploaded blob.
closers []io.Closer // Objects to close after the upload is done, if any.
operation bpcOperation // What we are actually doing
uploadedOperation types.LayerCompression // Operation to use for updating the blob metadata (matching the end state, not necessarily what we do)
uploadedAlgorithm *compressiontypes.Algorithm // An algorithm parameter for the compressionOperation edits.
uploadedAnnotations map[string]string // Compression-related annotations that should be set on the uploaded blob. WARNING: This is only set after the srcStream.reader is fully consumed.
srcCompressorBaseVariantName string // Compressor base variant name to record in the blob info cache for the source blob.
uploadedCompressorName string // Compressor name to record in the blob info cache for the uploaded blob.
closers []io.Closer // Objects to close after the upload is done, if any.
}
type bpcOperation int
@ -128,11 +129,11 @@ func (ic *imageCopier) bpcPreserveEncrypted(stream *sourceStream, _ bpDetectComp
// We cant do anything with an encrypted blob unless decrypted.
logrus.Debugf("Using original blob without modification for encrypted blob")
return &bpCompressionStepData{
operation: bpcOpPreserveOpaque,
uploadedOperation: types.PreserveOriginal,
uploadedAlgorithm: nil,
srcCompressorName: internalblobinfocache.UnknownCompression,
uploadedCompressorName: internalblobinfocache.UnknownCompression,
operation: bpcOpPreserveOpaque,
uploadedOperation: types.PreserveOriginal,
uploadedAlgorithm: nil,
srcCompressorBaseVariantName: internalblobinfocache.UnknownCompression,
uploadedCompressorName: internalblobinfocache.UnknownCompression,
}, nil
}
return nil, nil
@ -157,13 +158,13 @@ func (ic *imageCopier) bpcCompressUncompressed(stream *sourceStream, detected bp
Size: -1,
}
return &bpCompressionStepData{
operation: bpcOpCompressUncompressed,
uploadedOperation: types.Compress,
uploadedAlgorithm: uploadedAlgorithm,
uploadedAnnotations: annotations,
srcCompressorName: detected.srcCompressorName,
uploadedCompressorName: uploadedAlgorithm.Name(),
closers: []io.Closer{reader},
operation: bpcOpCompressUncompressed,
uploadedOperation: types.Compress,
uploadedAlgorithm: uploadedAlgorithm,
uploadedAnnotations: annotations,
srcCompressorBaseVariantName: detected.srcCompressorBaseVariantName,
uploadedCompressorName: uploadedAlgorithm.Name(),
closers: []io.Closer{reader},
}, nil
}
return nil, nil
@ -198,13 +199,13 @@ func (ic *imageCopier) bpcRecompressCompressed(stream *sourceStream, detected bp
}
succeeded = true
return &bpCompressionStepData{
operation: bpcOpRecompressCompressed,
uploadedOperation: types.PreserveOriginal,
uploadedAlgorithm: ic.compressionFormat,
uploadedAnnotations: annotations,
srcCompressorName: detected.srcCompressorName,
uploadedCompressorName: ic.compressionFormat.Name(),
closers: []io.Closer{decompressed, recompressed},
operation: bpcOpRecompressCompressed,
uploadedOperation: types.PreserveOriginal,
uploadedAlgorithm: ic.compressionFormat,
uploadedAnnotations: annotations,
srcCompressorBaseVariantName: detected.srcCompressorBaseVariantName,
uploadedCompressorName: ic.compressionFormat.Name(),
closers: []io.Closer{decompressed, recompressed},
}, nil
}
return nil, nil
@ -225,12 +226,12 @@ func (ic *imageCopier) bpcDecompressCompressed(stream *sourceStream, detected bp
Size: -1,
}
return &bpCompressionStepData{
operation: bpcOpDecompressCompressed,
uploadedOperation: types.Decompress,
uploadedAlgorithm: nil,
srcCompressorName: detected.srcCompressorName,
uploadedCompressorName: internalblobinfocache.Uncompressed,
closers: []io.Closer{s},
operation: bpcOpDecompressCompressed,
uploadedOperation: types.Decompress,
uploadedAlgorithm: nil,
srcCompressorBaseVariantName: detected.srcCompressorBaseVariantName,
uploadedCompressorName: internalblobinfocache.Uncompressed,
closers: []io.Closer{s},
}, nil
}
return nil, nil
@ -268,11 +269,14 @@ func (ic *imageCopier) bpcPreserveOriginal(_ *sourceStream, detected bpDetectCom
algorithm = nil
}
return &bpCompressionStepData{
operation: bpcOp,
uploadedOperation: uploadedOp,
uploadedAlgorithm: algorithm,
srcCompressorName: detected.srcCompressorName,
uploadedCompressorName: detected.srcCompressorName,
operation: bpcOp,
uploadedOperation: uploadedOp,
uploadedAlgorithm: algorithm,
srcCompressorBaseVariantName: detected.srcCompressorBaseVariantName,
// We only record the base variant of the format on upload; we didnt do anything with
// the TOC, we dont know whether it matches the blob digest, so we dont want to trigger
// reuse of any kind between the blob digest and the TOC digest.
uploadedCompressorName: detected.srcCompressorBaseVariantName,
}
}
@ -308,6 +312,15 @@ func (d *bpCompressionStepData) recordValidatedDigestData(c *copier, uploadedInf
// No useful information
case bpcOpCompressUncompressed:
c.blobInfoCache.RecordDigestUncompressedPair(uploadedInfo.Digest, srcInfo.Digest)
if d.uploadedAnnotations != nil {
tocDigest, err := chunkedToc.GetTOCDigest(d.uploadedAnnotations)
if err != nil {
return fmt.Errorf("parsing just-created compression annotations: %w", err)
}
if tocDigest != nil {
c.blobInfoCache.RecordTOCUncompressedPair(*tocDigest, srcInfo.Digest)
}
}
case bpcOpDecompressCompressed:
c.blobInfoCache.RecordDigestUncompressedPair(srcInfo.Digest, uploadedInfo.Digest)
case bpcOpRecompressCompressed, bpcOpPreserveCompressed:
@ -323,9 +336,9 @@ func (d *bpCompressionStepData) recordValidatedDigestData(c *copier, uploadedInf
return fmt.Errorf("Internal error: Unexpected d.operation value %#v", d.operation)
}
}
if d.srcCompressorName == "" || d.uploadedCompressorName == "" {
return fmt.Errorf("internal error: missing compressor names (src: %q, uploaded: %q)",
d.srcCompressorName, d.uploadedCompressorName)
if d.srcCompressorBaseVariantName == "" || d.uploadedCompressorName == "" {
return fmt.Errorf("internal error: missing compressor names (src base: %q, uploaded: %q)",
d.srcCompressorBaseVariantName, d.uploadedCompressorName)
}
if d.uploadedCompressorName != internalblobinfocache.UnknownCompression {
if d.uploadedCompressorName != compressiontypes.ZstdChunkedAlgorithmName {
@ -337,15 +350,19 @@ func (d *bpCompressionStepData) recordValidatedDigestData(c *copier, uploadedInf
// between zstd and zstd:chunked; so we could, in varying situations over time, call RecordDigestCompressorName
// with the same digest and both ZstdAlgorithmName and ZstdChunkedAlgorithmName , which causes warnings about
// inconsistent data to be logged.
c.blobInfoCache.RecordDigestCompressorName(uploadedInfo.Digest, d.uploadedCompressorName)
c.blobInfoCache.RecordDigestCompressorData(uploadedInfo.Digest, internalblobinfocache.DigestCompressorData{
BaseVariantCompressor: d.uploadedCompressorName,
})
}
}
if srcInfo.Digest != "" && srcInfo.Digest != uploadedInfo.Digest &&
d.srcCompressorName != internalblobinfocache.UnknownCompression {
if d.srcCompressorName != compressiontypes.ZstdChunkedAlgorithmName {
// HACK: Dont record zstd:chunked algorithms, see above.
c.blobInfoCache.RecordDigestCompressorName(srcInfo.Digest, d.srcCompressorName)
}
d.srcCompressorBaseVariantName != internalblobinfocache.UnknownCompression {
// If the source is already using some TOC-dependent variant, we either copied the
// blob as is, or perhaps decompressed it; either way we dont trust the TOC digest,
// so record neither the variant name, nor the TOC digest.
c.blobInfoCache.RecordDigestCompressorData(srcInfo.Digest, internalblobinfocache.DigestCompressorData{
BaseVariantCompressor: d.srcCompressorBaseVariantName,
})
}
return nil
}

View File

@ -48,7 +48,7 @@ func (ic *imageCopier) blobPipelineDecryptionStep(stream *sourceStream, srcInfo
Annotations: stream.info.Annotations,
}
// DecryptLayer supposedly returns a digest of the decrypted stream.
// In pratice, that value is never set in the current implementation.
// In practice, that value is never set in the current implementation.
// And we shouldnt use it anyway, because it is not trusted: encryption can be made to a public key,
// i.e. it doesnt authenticate the origin of the metadata in any way.
reader, _, err := ocicrypt.DecryptLayer(ic.c.options.OciDecryptConfig, stream.reader, desc, false)

View File

@ -121,7 +121,7 @@ func (c *copier) printCopyInfo(kind string, info types.BlobInfo) {
}
}
// mark100PercentComplete marks the progres bars as 100% complete;
// mark100PercentComplete marks the progress bars as 100% complete;
// it may do so by possibly advancing the current state if it is below the known total.
func (bar *progressBar) mark100PercentComplete() {
if bar.originalSize > 0 {

View File

@ -149,6 +149,28 @@ func (c *copier) copySingleImage(ctx context.Context, unparsedImage *image.Unpar
ic.compressionFormat = c.options.DestinationCtx.CompressionFormat
ic.compressionLevel = c.options.DestinationCtx.CompressionLevel
}
// HACK: Dont combine zstd:chunked and encryption.
// zstd:chunked can only usefully be consumed using range requests of parts of the layer, which would require the encryption
// to support decrypting arbitrary subsets of the stream. Thats plausible but not supported using the encryption API we have.
// Also, the chunked metadata is exposed in annotations unencrypted, which reveals the TOC digest = layer identity without
// encryption. (That can be determined from the unencrypted config anyway, but, still...)
//
// Ideally this should query a well-defined property of the compression algorithm (and $somehow determine the right fallback) instead of
// hard-coding zstd:chunked / zstd.
if ic.c.options.OciEncryptLayers != nil {
format := ic.compressionFormat
if format == nil {
format = defaultCompressionFormat
}
if format.Name() == compression.ZstdChunked.Name() {
if ic.requireCompressionFormatMatch {
return copySingleImageResult{}, errors.New("explicitly requested to combine zstd:chunked with encryption, which is not beneficial; use plain zstd instead")
}
logrus.Warnf("Compression using zstd:chunked is not beneficial for encrypted layers, using plain zstd instead")
ic.compressionFormat = &compression.Zstd
}
}
// Decide whether we can substitute blobs with semantic equivalents:
// - Dont do that if we cant modify the manifest at all
// - Ensure _this_ copy sees exactly the intended data when either processing a signed image or signing it.
@ -192,7 +214,7 @@ func (c *copier) copySingleImage(ctx context.Context, unparsedImage *image.Unpar
shouldUpdateSigs := len(sigs) > 0 || len(c.signers) != 0 // TODO: Consider allowing signatures updates only and skipping the image's layers/manifest copy if possible
noPendingManifestUpdates := ic.noPendingManifestUpdates()
logrus.Debugf("Checking if we can skip copying: has signatures=%t, OCI encryption=%t, no manifest updates=%t, compression match required for resuing blobs=%t", shouldUpdateSigs, destRequiresOciEncryption, noPendingManifestUpdates, opts.requireCompressionFormatMatch)
logrus.Debugf("Checking if we can skip copying: has signatures=%t, OCI encryption=%t, no manifest updates=%t, compression match required for reusing blobs=%t", shouldUpdateSigs, destRequiresOciEncryption, noPendingManifestUpdates, opts.requireCompressionFormatMatch)
if !shouldUpdateSigs && !destRequiresOciEncryption && noPendingManifestUpdates && !ic.requireCompressionFormatMatch {
matchedResult, err := ic.compareImageDestinationManifestEqual(ctx, targetInstance)
if err != nil {

View File

@ -27,7 +27,14 @@ func (bic *v1OnlyBlobInfoCache) Open() {
func (bic *v1OnlyBlobInfoCache) Close() {
}
func (bic *v1OnlyBlobInfoCache) RecordDigestCompressorName(anyDigest digest.Digest, compressorName string) {
func (bic *v1OnlyBlobInfoCache) UncompressedDigestForTOC(tocDigest digest.Digest) digest.Digest {
return ""
}
func (bic *v1OnlyBlobInfoCache) RecordTOCUncompressedPair(tocDigest digest.Digest, uncompressed digest.Digest) {
}
func (bic *v1OnlyBlobInfoCache) RecordDigestCompressorData(anyDigest digest.Digest, data DigestCompressorData) {
}
func (bic *v1OnlyBlobInfoCache) CandidateLocations2(transport types.ImageTransport, scope types.BICTransportScope, digest digest.Digest, options CandidateLocations2Options) []BICReplacementCandidate2 {

View File

@ -26,19 +26,34 @@ type BlobInfoCache2 interface {
// Close destroys state created by Open().
Close()
// RecordDigestCompressorName records a compressor for the blob with the specified digest,
// or Uncompressed or UnknownCompression.
// WARNING: Only call this with LOCALLY VERIFIED data; dont record a compressor for a
// digest just because some remote author claims so (e.g. because a manifest says so);
// UncompressedDigestForTOC returns an uncompressed digest corresponding to anyDigest.
// Returns "" if the uncompressed digest is unknown.
UncompressedDigestForTOC(tocDigest digest.Digest) digest.Digest
// RecordTOCUncompressedPair records that the tocDigest corresponds to uncompressed.
// WARNING: Only call this for LOCALLY VERIFIED data; dont record a digest pair just because some remote author claims so (e.g.
// because a manifest/config pair exists); otherwise the cache could be poisoned and allow substituting unexpected blobs.
// (Eventually, the DiffIDs in image config could detect the substitution, but that may be too late, and not all image formats contain that data.)
RecordTOCUncompressedPair(tocDigest digest.Digest, uncompressed digest.Digest)
// RecordDigestCompressorData records data for the blob with the specified digest.
// WARNING: Only call this with LOCALLY VERIFIED data:
// - dont record a compressor for a digest just because some remote author claims so
// (e.g. because a manifest says so);
// otherwise the cache could be poisoned and cause us to make incorrect edits to type
// information in a manifest.
RecordDigestCompressorName(anyDigest digest.Digest, compressorName string)
RecordDigestCompressorData(anyDigest digest.Digest, data DigestCompressorData)
// CandidateLocations2 returns a prioritized, limited, number of blobs and their locations (if known)
// that could possibly be reused within the specified (transport scope) (if they still
// exist, which is not guaranteed).
CandidateLocations2(transport types.ImageTransport, scope types.BICTransportScope, digest digest.Digest, options CandidateLocations2Options) []BICReplacementCandidate2
}
// DigestCompressorData is information known about how a blob is compressed.
// (This is worded generically, but basically targeted at the zstd / zstd:chunked situation.)
type DigestCompressorData struct {
BaseVariantCompressor string // A compressors base variant name, or Uncompressed or UnknownCompression.
}
// CandidateLocations2Options are used in CandidateLocations2.
type CandidateLocations2Options struct {
// If !CanSubstitute, the returned candidates will match the submitted digest exactly; if

View File

@ -202,7 +202,7 @@ func (m *Schema2) ConfigInfo() types.BlobInfo {
// The Digest field is guaranteed to be provided; Size may be -1.
// WARNING: The list may contain duplicates, and they are semantically relevant.
func (m *Schema2) LayerInfos() []LayerInfo {
blobs := []LayerInfo{}
blobs := make([]LayerInfo, 0, len(m.LayersDescriptors))
for _, layer := range m.LayersDescriptors {
blobs = append(blobs, LayerInfo{
BlobInfo: BlobInfoFromSchema2Descriptor(layer),

View File

@ -95,7 +95,7 @@ func (m *OCI1) ConfigInfo() types.BlobInfo {
// The Digest field is guaranteed to be provided; Size may be -1.
// WARNING: The list may contain duplicates, and they are semantically relevant.
func (m *OCI1) LayerInfos() []LayerInfo {
blobs := []LayerInfo{}
blobs := make([]LayerInfo, 0, len(m.Layers))
for _, layer := range m.Layers {
blobs = append(blobs, LayerInfo{
BlobInfo: BlobInfoFromOCI1Descriptor(layer),

View File

@ -151,9 +151,9 @@ func openRepo(path string) (*C.struct_OstreeRepo, error) {
var cerr *C.GError
cpath := C.CString(path)
defer C.free(unsafe.Pointer(cpath))
pathc := C.g_file_new_for_path(cpath)
defer C.g_object_unref(C.gpointer(pathc))
repo := C.ostree_repo_new(pathc)
file := C.g_file_new_for_path(cpath)
defer C.g_object_unref(C.gpointer(file))
repo := C.ostree_repo_new(file)
r := glib.GoBool(glib.GBoolean(C.ostree_repo_open(repo, nil, &cerr)))
if !r {
C.g_object_unref(C.gpointer(repo))

View File

@ -25,57 +25,103 @@ const replacementAttempts = 5
// This is a heuristic/guess, and could well use a different value.
const replacementUnknownLocationAttempts = 2
// CandidateCompression returns (true, compressionOp, compressionAlgo) if a blob
// with compressionName (which can be Uncompressed or UnknownCompression) is acceptable for a CandidateLocations* call with v2Options.
// CandidateTemplate is a subset of BICReplacementCandidate2 with data related to a specific digest,
// which can be later combined with information about a location.
type CandidateTemplate struct {
digest digest.Digest
compressionOperation types.LayerCompression // Either types.Decompress for uncompressed, or types.Compress for compressed
compressionAlgorithm *compression.Algorithm // An algorithm when the candidate is compressed, or nil when it is uncompressed
}
// CandidateTemplateWithCompression returns a CandidateTemplate if a blob with data is acceptable
// for a CandidateLocations* call with v2Options.
//
// v2Options can be set to nil if the call is CandidateLocations (i.e. compression is not required to be known);
// if not nil, the call is assumed to be CandidateLocations2.
//
// The (compressionOp, compressionAlgo) values are suitable for BICReplacementCandidate2
func CandidateCompression(v2Options *blobinfocache.CandidateLocations2Options, digest digest.Digest, compressorName string) (bool, types.LayerCompression, *compression.Algorithm) {
func CandidateTemplateWithCompression(v2Options *blobinfocache.CandidateLocations2Options, digest digest.Digest, data blobinfocache.DigestCompressorData) *CandidateTemplate {
if v2Options == nil {
return true, types.PreserveOriginal, nil // Anything goes. The (compressionOp, compressionAlgo) values are not used.
return &CandidateTemplate{ // Anything goes. The compressionOperation, compressionAlgorithm values are not used.
digest: digest,
}
}
var op types.LayerCompression
var algo *compression.Algorithm
switch compressorName {
requiredCompression := "nil"
if v2Options.RequiredCompression != nil {
requiredCompression = v2Options.RequiredCompression.Name()
}
switch data.BaseVariantCompressor {
case blobinfocache.Uncompressed:
op = types.Decompress
algo = nil
if !manifest.CandidateCompressionMatchesReuseConditions(manifest.ReuseConditions{
PossibleManifestFormats: v2Options.PossibleManifestFormats,
RequiredCompression: v2Options.RequiredCompression,
}, nil) {
logrus.Debugf("Ignoring BlobInfoCache record of digest %q, uncompressed format does not match required %s or MIME types %#v",
digest.String(), requiredCompression, v2Options.PossibleManifestFormats)
return nil
}
return &CandidateTemplate{
digest: digest,
compressionOperation: types.Decompress,
compressionAlgorithm: nil,
}
case blobinfocache.UnknownCompression:
logrus.Debugf("Ignoring BlobInfoCache record of digest %q with unknown compression", digest.String())
return false, types.PreserveOriginal, nil // Not allowed with CandidateLocations2
return nil // Not allowed with CandidateLocations2
default:
op = types.Compress
algo_, err := compression.AlgorithmByName(compressorName)
algo, err := compression.AlgorithmByName(data.BaseVariantCompressor)
if err != nil {
logrus.Debugf("Ignoring BlobInfoCache record of digest %q with unrecognized compression %q: %v",
digest.String(), compressorName, err)
return false, types.PreserveOriginal, nil // The BICReplacementCandidate2.CompressionAlgorithm field is required
digest.String(), data.BaseVariantCompressor, err)
return nil // The BICReplacementCandidate2.CompressionAlgorithm field is required
}
algo = &algo_
}
if !manifest.CandidateCompressionMatchesReuseConditions(manifest.ReuseConditions{
PossibleManifestFormats: v2Options.PossibleManifestFormats,
RequiredCompression: v2Options.RequiredCompression,
}, algo) {
requiredCompresssion := "nil"
if v2Options.RequiredCompression != nil {
requiredCompresssion = v2Options.RequiredCompression.Name()
if !manifest.CandidateCompressionMatchesReuseConditions(manifest.ReuseConditions{
PossibleManifestFormats: v2Options.PossibleManifestFormats,
RequiredCompression: v2Options.RequiredCompression,
}, &algo) {
logrus.Debugf("Ignoring BlobInfoCache record of digest %q, compression %q does not match required %s or MIME types %#v",
digest.String(), data.BaseVariantCompressor, requiredCompression, v2Options.PossibleManifestFormats)
return nil
}
return &CandidateTemplate{
digest: digest,
compressionOperation: types.Compress,
compressionAlgorithm: &algo,
}
logrus.Debugf("Ignoring BlobInfoCache record of digest %q, compression %q does not match required %s or MIME types %#v",
digest.String(), compressorName, requiredCompresssion, v2Options.PossibleManifestFormats)
return false, types.PreserveOriginal, nil
}
return true, op, algo
}
// CandidateWithTime is the input to types.BICReplacementCandidate prioritization.
type CandidateWithTime struct {
Candidate blobinfocache.BICReplacementCandidate2 // The replacement candidate
LastSeen time.Time // Time the candidate was last known to exist (either read or written) (not set for Candidate.UnknownLocation)
candidate blobinfocache.BICReplacementCandidate2 // The replacement candidate
lastSeen time.Time // Time the candidate was last known to exist (either read or written) (not set for Candidate.UnknownLocation)
}
// CandidateWithLocation returns a complete CandidateWithTime combining (template from CandidateTemplateWithCompression, location, lastSeen)
func (template CandidateTemplate) CandidateWithLocation(location types.BICLocationReference, lastSeen time.Time) CandidateWithTime {
return CandidateWithTime{
candidate: blobinfocache.BICReplacementCandidate2{
Digest: template.digest,
CompressionOperation: template.compressionOperation,
CompressionAlgorithm: template.compressionAlgorithm,
UnknownLocation: false,
Location: location,
},
lastSeen: lastSeen,
}
}
// CandidateWithUnknownLocation returns a complete CandidateWithTime for a template from CandidateTemplateWithCompression and an unknown location.
func (template CandidateTemplate) CandidateWithUnknownLocation() CandidateWithTime {
return CandidateWithTime{
candidate: blobinfocache.BICReplacementCandidate2{
Digest: template.digest,
CompressionOperation: template.compressionOperation,
CompressionAlgorithm: template.compressionAlgorithm,
UnknownLocation: true,
Location: types.BICLocationReference{Opaque: ""},
},
lastSeen: time.Time{},
}
}
// candidateSortState is a closure for a comparison used by slices.SortFunc on candidates to prioritize,
@ -91,35 +137,35 @@ func (css *candidateSortState) compare(xi, xj CandidateWithTime) int {
// Other digest values are primarily sorted by time (more recent first), secondarily by digest (to provide a deterministic order)
// First, deal with the primaryDigest/uncompressedDigest cases:
if xi.Candidate.Digest != xj.Candidate.Digest {
if xi.candidate.Digest != xj.candidate.Digest {
// - The two digests are different, and one (or both) of the digests is primaryDigest or uncompressedDigest: time does not matter
if xi.Candidate.Digest == css.primaryDigest {
if xi.candidate.Digest == css.primaryDigest {
return -1
}
if xj.Candidate.Digest == css.primaryDigest {
if xj.candidate.Digest == css.primaryDigest {
return 1
}
if css.uncompressedDigest != "" {
if xi.Candidate.Digest == css.uncompressedDigest {
if xi.candidate.Digest == css.uncompressedDigest {
return 1
}
if xj.Candidate.Digest == css.uncompressedDigest {
if xj.candidate.Digest == css.uncompressedDigest {
return -1
}
}
} else { // xi.Candidate.Digest == xj.Candidate.Digest
// The two digests are the same, and are either primaryDigest or uncompressedDigest: order by time
if xi.Candidate.Digest == css.primaryDigest || (css.uncompressedDigest != "" && xi.Candidate.Digest == css.uncompressedDigest) {
return -xi.LastSeen.Compare(xj.LastSeen)
if xi.candidate.Digest == css.primaryDigest || (css.uncompressedDigest != "" && xi.candidate.Digest == css.uncompressedDigest) {
return -xi.lastSeen.Compare(xj.lastSeen)
}
}
// Neither of the digests are primaryDigest/uncompressedDigest:
if cmp := xi.LastSeen.Compare(xj.LastSeen); cmp != 0 { // Order primarily by time
if cmp := xi.lastSeen.Compare(xj.lastSeen); cmp != 0 { // Order primarily by time
return -cmp
}
// Fall back to digest, if timestamps end up _exactly_ the same (how?!)
return cmp.Compare(xi.Candidate.Digest, xj.Candidate.Digest)
return cmp.Compare(xi.candidate.Digest, xj.candidate.Digest)
}
// destructivelyPrioritizeReplacementCandidatesWithMax is destructivelyPrioritizeReplacementCandidates with parameters for the
@ -138,7 +184,7 @@ func destructivelyPrioritizeReplacementCandidatesWithMax(cs []CandidateWithTime,
uncompressedDigest: uncompressedDigest,
}).compare)
for _, candidate := range cs {
if candidate.Candidate.UnknownLocation {
if candidate.candidate.UnknownLocation {
unknownLocationCandidates = append(unknownLocationCandidates, candidate)
} else {
knownLocationCandidates = append(knownLocationCandidates, candidate)
@ -150,11 +196,11 @@ func destructivelyPrioritizeReplacementCandidatesWithMax(cs []CandidateWithTime,
unknownLocationCandidatesUsed := min(noLocationLimit, remainingCapacity, len(unknownLocationCandidates))
res := make([]blobinfocache.BICReplacementCandidate2, knownLocationCandidatesUsed)
for i := 0; i < knownLocationCandidatesUsed; i++ {
res[i] = knownLocationCandidates[i].Candidate
res[i] = knownLocationCandidates[i].candidate
}
// If candidates with unknown location are found, lets add them to final list
for i := 0; i < unknownLocationCandidatesUsed; i++ {
res = append(res, unknownLocationCandidates[i].Candidate)
res = append(res, unknownLocationCandidates[i].candidate)
}
return res
}

View File

@ -24,10 +24,11 @@ type locationKey struct {
type cache struct {
mutex sync.Mutex
// The following fields can only be accessed with mutex held.
uncompressedDigests map[digest.Digest]digest.Digest
digestsByUncompressed map[digest.Digest]*set.Set[digest.Digest] // stores a set of digests for each uncompressed digest
knownLocations map[locationKey]map[types.BICLocationReference]time.Time // stores last known existence time for each location reference
compressors map[digest.Digest]string // stores a compressor name, or blobinfocache.Uncompressed (not blobinfocache.UnknownCompression), for each digest
uncompressedDigests map[digest.Digest]digest.Digest
uncompressedDigestsByTOC map[digest.Digest]digest.Digest
digestsByUncompressed map[digest.Digest]*set.Set[digest.Digest] // stores a set of digests for each uncompressed digest
knownLocations map[locationKey]map[types.BICLocationReference]time.Time // stores last known existence time for each location reference
compressors map[digest.Digest]string // stores a compressor name, or blobinfocache.Uncompressed (not blobinfocache.UnknownCompression), for each digest
}
// New returns a BlobInfoCache implementation which is in-memory only.
@ -44,10 +45,11 @@ func New() types.BlobInfoCache {
func new2() *cache {
return &cache{
uncompressedDigests: map[digest.Digest]digest.Digest{},
digestsByUncompressed: map[digest.Digest]*set.Set[digest.Digest]{},
knownLocations: map[locationKey]map[types.BICLocationReference]time.Time{},
compressors: map[digest.Digest]string{},
uncompressedDigests: map[digest.Digest]digest.Digest{},
uncompressedDigestsByTOC: map[digest.Digest]digest.Digest{},
digestsByUncompressed: map[digest.Digest]*set.Set[digest.Digest]{},
knownLocations: map[locationKey]map[types.BICLocationReference]time.Time{},
compressors: map[digest.Digest]string{},
}
}
@ -104,6 +106,30 @@ func (mem *cache) RecordDigestUncompressedPair(anyDigest digest.Digest, uncompre
anyDigestSet.Add(anyDigest)
}
// UncompressedDigestForTOC returns an uncompressed digest corresponding to anyDigest.
// Returns "" if the uncompressed digest is unknown.
func (mem *cache) UncompressedDigestForTOC(tocDigest digest.Digest) digest.Digest {
mem.mutex.Lock()
defer mem.mutex.Unlock()
if d, ok := mem.uncompressedDigestsByTOC[tocDigest]; ok {
return d
}
return ""
}
// RecordTOCUncompressedPair records that the tocDigest corresponds to uncompressed.
// WARNING: Only call this for LOCALLY VERIFIED data; dont record a digest pair just because some remote author claims so (e.g.
// because a manifest/config pair exists); otherwise the cache could be poisoned and allow substituting unexpected blobs.
// (Eventually, the DiffIDs in image config could detect the substitution, but that may be too late, and not all image formats contain that data.)
func (mem *cache) RecordTOCUncompressedPair(tocDigest digest.Digest, uncompressed digest.Digest) {
mem.mutex.Lock()
defer mem.mutex.Unlock()
if previous, ok := mem.uncompressedDigestsByTOC[tocDigest]; ok && previous != uncompressed {
logrus.Warnf("Uncompressed digest for blob with TOC %q previously recorded as %q, now %q", tocDigest, previous, uncompressed)
}
mem.uncompressedDigestsByTOC[tocDigest] = uncompressed
}
// RecordKnownLocation records that a blob with the specified digest exists within the specified (transport, scope) scope,
// and can be reused given the opaque location data.
func (mem *cache) RecordKnownLocation(transport types.ImageTransport, scope types.BICTransportScope, blobDigest digest.Digest, location types.BICLocationReference) {
@ -118,19 +144,24 @@ func (mem *cache) RecordKnownLocation(transport types.ImageTransport, scope type
locationScope[location] = time.Now() // Possibly overwriting an older entry.
}
// RecordDigestCompressorName records that the blob with the specified digest is either compressed with the specified
// algorithm, or uncompressed, or that we no longer know.
func (mem *cache) RecordDigestCompressorName(blobDigest digest.Digest, compressorName string) {
// RecordDigestCompressorData records data for the blob with the specified digest.
// WARNING: Only call this with LOCALLY VERIFIED data:
// - dont record a compressor for a digest just because some remote author claims so
// (e.g. because a manifest says so);
//
// otherwise the cache could be poisoned and cause us to make incorrect edits to type
// information in a manifest.
func (mem *cache) RecordDigestCompressorData(anyDigest digest.Digest, data blobinfocache.DigestCompressorData) {
mem.mutex.Lock()
defer mem.mutex.Unlock()
if previous, ok := mem.compressors[blobDigest]; ok && previous != compressorName {
logrus.Warnf("Compressor for blob with digest %s previously recorded as %s, now %s", blobDigest, previous, compressorName)
if previous, ok := mem.compressors[anyDigest]; ok && previous != data.BaseVariantCompressor {
logrus.Warnf("Compressor for blob with digest %s previously recorded as %s, now %s", anyDigest, previous, data.BaseVariantCompressor)
}
if compressorName == blobinfocache.UnknownCompression {
delete(mem.compressors, blobDigest)
if data.BaseVariantCompressor == blobinfocache.UnknownCompression {
delete(mem.compressors, anyDigest)
return
}
mem.compressors[blobDigest] = compressorName
mem.compressors[anyDigest] = data.BaseVariantCompressor
}
// appendReplacementCandidates creates prioritize.CandidateWithTime values for digest in memory
@ -144,34 +175,19 @@ func (mem *cache) appendReplacementCandidates(candidates []prioritize.CandidateW
if v, ok := mem.compressors[digest]; ok {
compressorName = v
}
ok, compressionOp, compressionAlgo := prioritize.CandidateCompression(v2Options, digest, compressorName)
if !ok {
template := prioritize.CandidateTemplateWithCompression(v2Options, digest, blobinfocache.DigestCompressorData{
BaseVariantCompressor: compressorName,
})
if template == nil {
return candidates
}
locations := mem.knownLocations[locationKey{transport: transport.Name(), scope: scope, blobDigest: digest}] // nil if not present
if len(locations) > 0 {
for l, t := range locations {
candidates = append(candidates, prioritize.CandidateWithTime{
Candidate: blobinfocache.BICReplacementCandidate2{
Digest: digest,
CompressionOperation: compressionOp,
CompressionAlgorithm: compressionAlgo,
Location: l,
},
LastSeen: t,
})
candidates = append(candidates, template.CandidateWithLocation(l, t))
}
} else if v2Options != nil {
candidates = append(candidates, prioritize.CandidateWithTime{
Candidate: blobinfocache.BICReplacementCandidate2{
Digest: digest,
CompressionOperation: compressionOp,
CompressionAlgorithm: compressionAlgo,
UnknownLocation: true,
Location: types.BICLocationReference{Opaque: ""},
},
LastSeen: time.Time{},
})
candidates = append(candidates, template.CandidateWithUnknownLocation())
}
return candidates
}

View File

@ -34,6 +34,19 @@ func (noCache) UncompressedDigest(anyDigest digest.Digest) digest.Digest {
func (noCache) RecordDigestUncompressedPair(anyDigest digest.Digest, uncompressed digest.Digest) {
}
// UncompressedDigestForTOC returns an uncompressed digest corresponding to anyDigest.
// Returns "" if the uncompressed digest is unknown.
func (noCache) UncompressedDigestForTOC(tocDigest digest.Digest) digest.Digest {
return ""
}
// RecordTOCUncompressedPair records that the tocDigest corresponds to uncompressed.
// WARNING: Only call this for LOCALLY VERIFIED data; dont record a digest pair just because some remote author claims so (e.g.
// because a manifest/config pair exists); otherwise the cache could be poisoned and allow substituting unexpected blobs.
// (Eventually, the DiffIDs in image config could detect the substitution, but that may be too late, and not all image formats contain that data.)
func (noCache) RecordTOCUncompressedPair(tocDigest digest.Digest, uncompressed digest.Digest) {
}
// RecordKnownLocation records that a blob with the specified digest exists within the specified (transport, scope) scope,
// and can be reused given the opaque location data.
func (noCache) RecordKnownLocation(transport types.ImageTransport, scope types.BICTransportScope, blobDigest digest.Digest, location types.BICLocationReference) {

View File

@ -295,6 +295,14 @@ func ensureDBHasCurrentSchema(db *sql.DB) error {
`PRIMARY KEY (transport, scope, digest, location)
)`,
},
{
"DigestTOCUncompressedPairs",
`CREATE TABLE IF NOT EXISTS DigestTOCUncompressedPairs(` +
// index implied by PRIMARY KEY
`tocDigest TEXT PRIMARY KEY NOT NULL,` +
`uncompressedDigest TEXT NOT NULL
)`,
},
}
_, err := dbTransaction(db, func(tx *sql.Tx) (void, error) {
@ -385,6 +393,57 @@ func (sqc *cache) RecordDigestUncompressedPair(anyDigest digest.Digest, uncompre
}) // FIXME? Log error (but throttle the log volume on repeated accesses)?
}
// UncompressedDigestForTOC returns an uncompressed digest corresponding to anyDigest.
// Returns "" if the uncompressed digest is unknown.
func (sqc *cache) UncompressedDigestForTOC(tocDigest digest.Digest) digest.Digest {
res, err := transaction(sqc, func(tx *sql.Tx) (digest.Digest, error) {
uncompressedString, found, err := querySingleValue[string](tx, "SELECT uncompressedDigest FROM DigestTOCUncompressedPairs WHERE tocDigest = ?", tocDigest.String())
if err != nil {
return "", err
}
if found {
d, err := digest.Parse(uncompressedString)
if err != nil {
return "", err
}
return d, nil
}
return "", nil
})
if err != nil {
return "" // FIXME? Log err (but throttle the log volume on repeated accesses)?
}
return res
}
// RecordTOCUncompressedPair records that the tocDigest corresponds to uncompressed.
// WARNING: Only call this for LOCALLY VERIFIED data; dont record a digest pair just because some remote author claims so (e.g.
// because a manifest/config pair exists); otherwise the cache could be poisoned and allow substituting unexpected blobs.
// (Eventually, the DiffIDs in image config could detect the substitution, but that may be too late, and not all image formats contain that data.)
func (sqc *cache) RecordTOCUncompressedPair(tocDigest digest.Digest, uncompressed digest.Digest) {
_, _ = transaction(sqc, func(tx *sql.Tx) (void, error) {
previousString, gotPrevious, err := querySingleValue[string](tx, "SELECT uncompressedDigest FROM DigestTOCUncompressedPairs WHERE tocDigest = ?", tocDigest.String())
if err != nil {
return void{}, fmt.Errorf("looking for uncompressed digest for blob with TOC %q", tocDigest)
}
if gotPrevious {
previous, err := digest.Parse(previousString)
if err != nil {
return void{}, err
}
if previous != uncompressed {
logrus.Warnf("Uncompressed digest for blob with TOC %q previously recorded as %q, now %q", tocDigest, previous, uncompressed)
}
}
if _, err := tx.Exec("INSERT OR REPLACE INTO DigestTOCUncompressedPairs(tocDigest, uncompressedDigest) VALUES (?, ?)",
tocDigest.String(), uncompressed.String()); err != nil {
return void{}, fmt.Errorf("recording uncompressed digest %q for blob with TOC %q: %w", uncompressed, tocDigest, err)
}
return void{}, nil
}) // FIXME? Log error (but throttle the log volume on repeated accesses)?
}
// RecordKnownLocation records that a blob with the specified digest exists within the specified (transport, scope) scope,
// and can be reused given the opaque location data.
func (sqc *cache) RecordKnownLocation(transport types.ImageTransport, scope types.BICTransportScope, digest digest.Digest, location types.BICLocationReference) {
@ -398,29 +457,30 @@ func (sqc *cache) RecordKnownLocation(transport types.ImageTransport, scope type
}) // FIXME? Log error (but throttle the log volume on repeated accesses)?
}
// RecordDigestCompressorName records a compressor for the blob with the specified digest,
// or Uncompressed or UnknownCompression.
// WARNING: Only call this with LOCALLY VERIFIED data; dont record a compressor for a
// digest just because some remote author claims so (e.g. because a manifest says so);
// RecordDigestCompressorData records data for the blob with the specified digest.
// WARNING: Only call this with LOCALLY VERIFIED data:
// - dont record a compressor for a digest just because some remote author claims so
// (e.g. because a manifest says so);
//
// otherwise the cache could be poisoned and cause us to make incorrect edits to type
// information in a manifest.
func (sqc *cache) RecordDigestCompressorName(anyDigest digest.Digest, compressorName string) {
func (sqc *cache) RecordDigestCompressorData(anyDigest digest.Digest, data blobinfocache.DigestCompressorData) {
_, _ = transaction(sqc, func(tx *sql.Tx) (void, error) {
previous, gotPrevious, err := querySingleValue[string](tx, "SELECT compressor FROM DigestCompressors WHERE digest = ?", anyDigest.String())
if err != nil {
return void{}, fmt.Errorf("looking for compressor of for %q", anyDigest)
}
if gotPrevious && previous != compressorName {
logrus.Warnf("Compressor for blob with digest %s previously recorded as %s, now %s", anyDigest, previous, compressorName)
if gotPrevious && previous != data.BaseVariantCompressor {
logrus.Warnf("Compressor for blob with digest %s previously recorded as %s, now %s", anyDigest, previous, data.BaseVariantCompressor)
}
if compressorName == blobinfocache.UnknownCompression {
if data.BaseVariantCompressor == blobinfocache.UnknownCompression {
if _, err := tx.Exec("DELETE FROM DigestCompressors WHERE digest = ?", anyDigest.String()); err != nil {
return void{}, fmt.Errorf("deleting compressor for digest %q: %w", anyDigest, err)
}
} else {
if _, err := tx.Exec("INSERT OR REPLACE INTO DigestCompressors(digest, compressor) VALUES (?, ?)",
anyDigest.String(), compressorName); err != nil {
return void{}, fmt.Errorf("recording compressor %q for %q: %w", compressorName, anyDigest, err)
anyDigest.String(), data.BaseVariantCompressor); err != nil {
return void{}, fmt.Errorf("recording compressor %q for %q: %w", data.BaseVariantCompressor, anyDigest, err)
}
}
return void{}, nil
@ -443,8 +503,10 @@ func (sqc *cache) appendReplacementCandidates(candidates []prioritize.CandidateW
compressorName = compressor
}
}
ok, compressionOp, compressionAlgo := prioritize.CandidateCompression(v2Options, digest, compressorName)
if !ok {
template := prioritize.CandidateTemplateWithCompression(v2Options, digest, blobinfocache.DigestCompressorData{
BaseVariantCompressor: compressorName,
})
if template == nil {
return candidates, nil
}
@ -463,15 +525,7 @@ func (sqc *cache) appendReplacementCandidates(candidates []prioritize.CandidateW
if err := rows.Scan(&location, &time); err != nil {
return nil, fmt.Errorf("scanning candidate: %w", err)
}
candidates = append(candidates, prioritize.CandidateWithTime{
Candidate: blobinfocache.BICReplacementCandidate2{
Digest: digest,
CompressionOperation: compressionOp,
CompressionAlgorithm: compressionAlgo,
Location: types.BICLocationReference{Opaque: location},
},
LastSeen: time,
})
candidates = append(candidates, template.CandidateWithLocation(types.BICLocationReference{Opaque: location}, time))
rowAdded = true
}
if err := rows.Err(); err != nil {
@ -479,16 +533,7 @@ func (sqc *cache) appendReplacementCandidates(candidates []prioritize.CandidateW
}
if !rowAdded && v2Options != nil {
candidates = append(candidates, prioritize.CandidateWithTime{
Candidate: blobinfocache.BICReplacementCandidate2{
Digest: digest,
CompressionOperation: compressionOp,
CompressionAlgorithm: compressionAlgo,
UnknownLocation: true,
Location: types.BICLocationReference{Opaque: ""},
},
LastSeen: time.Time{},
})
candidates = append(candidates, template.CandidateWithUnknownLocation())
}
return candidates, nil
}

View File

@ -61,14 +61,12 @@ func parseUnnormalizedShortName(input string) (bool, reference.Named, error) {
// the tag or digest and stores it in the return values so that both can be
// re-added to a possible resolved alias' or USRs at a later point.
func splitUserInput(named reference.Named) (isTagged bool, isDigested bool, normalized reference.Named, tag string, digest digest.Digest) {
tagged, isT := named.(reference.NamedTagged)
if isT {
if tagged, ok := named.(reference.NamedTagged); ok {
isTagged = true
tag = tagged.Tag()
}
digested, isD := named.(reference.Digested)
if isD {
if digested, ok := named.(reference.Digested); ok {
isDigested = true
digest = digested.Digest()
}

View File

@ -84,18 +84,36 @@ type storageImageDestinationLockProtected struct {
currentIndex int // The index of the layer to be committed (i.e., lower indices have already been committed)
indexToAddedLayerInfo map[int]addedLayerInfo // Mapping from layer (by index) to blob to add to the image
// In general, a layer is identified either by (compressed) digest, or by TOC digest.
// Externally, a layer is identified either by (compressed) digest, or by TOC digest
// (and we assume the TOC digest also uniquely identifies the contents, i.e. there arent two
// different formats/ways to parse a single TOC); internally, we use uncompressed digest (“DiffID”) or a TOC digest.
// We may or may not know the relationships between these three values.
//
// When creating a layer, the c/storage layer metadata and image IDs must _only_ be based on trusted values
// we have computed ourselves. (Layer reuse can then look up against such trusted values, but it might not
// recompute those values for incomding layers — the point of the reuse is that we dont need to consume the incoming layer.)
// Layer identification: For a layer, at least one of indexToTOCDigest and blobDiffIDs must be available before commitLayer is called.
// The presence of an indexToTOCDigest is what decides how the layer is identified, i.e. which fields must be trusted.
blobDiffIDs map[digest.Digest]digest.Digest // Mapping from layer blobsums to their corresponding DiffIDs
indexToTOCDigest map[int]digest.Digest // Mapping from layer index to a TOC Digest, IFF the layer was created/found/reused by TOC digest
// recompute those values for incoming layers — the point of the reuse is that we dont need to consume the incoming layer.)
//
// Layer identification: For a layer, at least one of (indexToDiffID, indexToTOCDigest, blobDiffIDs) must be available
// before commitLayer is called.
// The layer is identified by the first of the three fields which exists, in that order (and the value must be trusted).
//
// WARNING: All values in indexToDiffID, indexToTOCDigest, and blobDiffIDs are _individually_ trusted, but blobDiffIDs is more subtle.
// The values in indexTo* are all consistent, because the code writing them processed them all at once, and consistently.
// But it is possible for a layers indexToDiffID an indexToTOCDigest to be based on a TOC, without setting blobDiffIDs
// for the compressed digest of that index, and for blobDiffIDs[compressedDigest] to be set _separately_ while processing some
// other layer entry. In particular it is possible for indexToDiffID[index] and blobDiffIDs[compressedDigestAtIndex]] to refer
// to mismatching contents.
// Users of these fields should use trustedLayerIdentityDataLocked, which centralizes the validity logic,
// instead of interpreting these fields, especially blobDiffIDs, directly.
//
// Ideally we wouldnt have blobDiffIDs, and we would just keep records by index, but the public API does not require the caller
// to provide layer indices; and configs dont have layer indices. blobDiffIDs needs to exist for those cases.
indexToDiffID map[int]digest.Digest // Mapping from layer index to DiffID
indexToTOCDigest map[int]digest.Digest // Mapping from layer index to a TOC Digest
blobDiffIDs map[digest.Digest]digest.Digest // Mapping from layer blobsums to their corresponding DiffIDs. CAREFUL: See the WARNING above.
// Layer data: Before commitLayer is called, either at least one of (diffOutputs, indexToAdditionalLayer, filenames)
// should be available; or indexToTOCDigest/blobDiffIDs should be enough to locate an existing c/storage layer.
// should be available; or indexToDiffID/indexToTOCDigest/blobDiffIDs should be enough to locate an existing c/storage layer.
// They are looked up in the order they are mentioned above.
diffOutputs map[int]*graphdriver.DriverWithDifferOutput // Mapping from layer index to a partially-pulled layer intermediate data
indexToAdditionalLayer map[int]storage.AdditionalLayer // Mapping from layer index to their corresponding additional layer
@ -145,9 +163,12 @@ func newImageDestination(sys *types.SystemContext, imageRef storageReference) (*
},
indexToStorageID: make(map[int]string),
lockProtected: storageImageDestinationLockProtected{
indexToAddedLayerInfo: make(map[int]addedLayerInfo),
blobDiffIDs: make(map[digest.Digest]digest.Digest),
indexToTOCDigest: make(map[int]digest.Digest),
indexToAddedLayerInfo: make(map[int]addedLayerInfo),
indexToDiffID: make(map[int]digest.Digest),
indexToTOCDigest: make(map[int]digest.Digest),
blobDiffIDs: make(map[digest.Digest]digest.Digest),
diffOutputs: make(map[int]*graphdriver.DriverWithDifferOutput),
indexToAdditionalLayer: make(map[int]storage.AdditionalLayer),
filenames: make(map[digest.Digest]string),
@ -323,20 +344,30 @@ func (s *storageImageDestination) PutBlobPartial(ctx context.Context, chunkAcces
s.lock.Lock()
if out.UncompressedDigest != "" {
s.lockProtected.indexToDiffID[options.LayerIndex] = out.UncompressedDigest
if out.TOCDigest != "" {
options.Cache.RecordTOCUncompressedPair(out.TOCDigest, out.UncompressedDigest)
}
// Dont set indexToTOCDigest on this path:
// - Using UncompressedDigest allows image reuse with non-partially-pulled layers, so we want to set indexToDiffID.
// - If UncompressedDigest has been computed, that means the layer was read completely, and the TOC has been created from scratch.
// That TOC is quite unlikely to match any other TOC value.
// The computation of UncompressedDigest means the whole layer has been consumed; while doing that, chunked.GetDiffer is
// responsible for ensuring blobDigest has been validated.
if out.CompressedDigest != blobDigest {
return private.UploadedBlob{}, fmt.Errorf("internal error: ApplyDiffWithDiffer returned CompressedDigest %q not matching expected %q",
out.CompressedDigest, blobDigest)
}
s.lockProtected.blobDiffIDs[blobDigest] = out.UncompressedDigest
// So, record also information about blobDigest, that might benefit reuse.
// We trust ApplyDiffWithDiffer to validate or create both values correctly.
s.lockProtected.blobDiffIDs[blobDigest] = out.UncompressedDigest
options.Cache.RecordDigestUncompressedPair(out.CompressedDigest, out.UncompressedDigest)
} else {
// Dont identify layers by TOC if UncompressedDigest is available.
// - Using UncompressedDigest allows image reuse with non-partially-pulled layers
// - If UncompressedDigest has been computed, that means the layer was read completely, and the TOC has been created from scratch.
// That TOC is quite unlikely to match with any other TOC value.
// Use diffID for layer identity if it is known.
if uncompressedDigest := options.Cache.UncompressedDigestForTOC(out.TOCDigest); uncompressedDigest != "" {
s.lockProtected.indexToDiffID[options.LayerIndex] = uncompressedDigest
}
s.lockProtected.indexToTOCDigest[options.LayerIndex] = out.TOCDigest
}
s.lockProtected.diffOutputs[options.LayerIndex] = out
@ -465,49 +496,40 @@ func (s *storageImageDestination) tryReusingBlobAsPending(blobDigest digest.Dige
if err != nil && !errors.Is(err, storage.ErrLayerUnknown) {
return false, private.ReusedBlob{}, fmt.Errorf(`looking for layers with digest %q: %w`, uncompressedDigest, err)
}
if len(layers) > 0 {
if size != -1 {
s.lockProtected.blobDiffIDs[blobDigest] = uncompressedDigest
return true, private.ReusedBlob{
Digest: blobDigest,
Size: size,
}, nil
}
if !options.CanSubstitute {
return false, private.ReusedBlob{}, fmt.Errorf("Internal error: options.CanSubstitute was expected to be true for blob with digest %s", blobDigest)
}
s.lockProtected.blobDiffIDs[uncompressedDigest] = uncompressedDigest
return true, private.ReusedBlob{
Digest: uncompressedDigest,
Size: layers[0].UncompressedSize,
}, nil
if found, reused := reusedBlobFromLayerLookup(layers, blobDigest, size, options); found {
s.lockProtected.blobDiffIDs[blobDigest] = uncompressedDigest
return true, reused, nil
}
}
}
if options.TOCDigest != "" && options.LayerIndex != nil {
// Check if we know which which UncompressedDigest the TOC digest resolves to, and we have a match for that.
// Prefer this over LayersByTOCDigest because we can identify the layer using UncompressedDigest, maximizing reuse.
uncompressedDigest := options.Cache.UncompressedDigestForTOC(options.TOCDigest)
if uncompressedDigest != "" {
layers, err = s.imageRef.transport.store.LayersByUncompressedDigest(uncompressedDigest)
if err != nil && !errors.Is(err, storage.ErrLayerUnknown) {
return false, private.ReusedBlob{}, fmt.Errorf(`looking for layers with digest %q: %w`, uncompressedDigest, err)
}
if found, reused := reusedBlobFromLayerLookup(layers, blobDigest, size, options); found {
s.lockProtected.indexToDiffID[*options.LayerIndex] = uncompressedDigest
reused.MatchedByTOCDigest = true
return true, reused, nil
}
}
// Check if we have a chunked layer in storage with the same TOC digest.
layers, err := s.imageRef.transport.store.LayersByTOCDigest(options.TOCDigest)
if err != nil && !errors.Is(err, storage.ErrLayerUnknown) {
return false, private.ReusedBlob{}, fmt.Errorf(`looking for layers with TOC digest %q: %w`, options.TOCDigest, err)
}
if len(layers) > 0 {
if size != -1 {
s.lockProtected.indexToTOCDigest[*options.LayerIndex] = options.TOCDigest
return true, private.ReusedBlob{
Digest: blobDigest,
Size: size,
MatchedByTOCDigest: true,
}, nil
} else if options.CanSubstitute && layers[0].UncompressedDigest != "" {
s.lockProtected.indexToTOCDigest[*options.LayerIndex] = options.TOCDigest
return true, private.ReusedBlob{
Digest: layers[0].UncompressedDigest,
Size: layers[0].UncompressedSize,
MatchedByTOCDigest: true,
}, nil
if found, reused := reusedBlobFromLayerLookup(layers, blobDigest, size, options); found {
if uncompressedDigest != "" {
s.lockProtected.indexToDiffID[*options.LayerIndex] = uncompressedDigest
}
s.lockProtected.indexToTOCDigest[*options.LayerIndex] = options.TOCDigest
reused.MatchedByTOCDigest = true
return true, reused, nil
}
}
@ -515,49 +537,137 @@ func (s *storageImageDestination) tryReusingBlobAsPending(blobDigest digest.Dige
return false, private.ReusedBlob{}, nil
}
// reusedBlobFromLayerLookup returns (true, ReusedBlob) if layers contain a usable match; or (false, ...) if not.
// The caller is still responsible for setting the layer identification fields, to allow the layer to be found again.
func reusedBlobFromLayerLookup(layers []storage.Layer, blobDigest digest.Digest, blobSize int64, options *private.TryReusingBlobOptions) (bool, private.ReusedBlob) {
if len(layers) > 0 {
if blobSize != -1 {
return true, private.ReusedBlob{
Digest: blobDigest,
Size: blobSize,
}
} else if options.CanSubstitute && layers[0].UncompressedDigest != "" {
return true, private.ReusedBlob{
Digest: layers[0].UncompressedDigest,
Size: layers[0].UncompressedSize,
CompressionOperation: types.Decompress,
CompressionAlgorithm: nil,
}
}
}
return false, private.ReusedBlob{}
}
// trustedLayerIdentityData is a _consistent_ set of information known about a single layer.
type trustedLayerIdentityData struct {
layerIdentifiedByTOC bool // true if we decided the layer should be identified by tocDigest, false if by diffID
diffID digest.Digest // A digest of the uncompressed full contents of the layer, or "" if unknown; must be set if !layerIdentifiedByTOC
tocDigest digest.Digest // A digest of the TOC digest, or "" if unknown; must be set if layerIdentifiedByTOC
blobDigest digest.Digest // A digest of the (possibly-compressed) layer as presented, or "" if unknown/untrusted.
}
// trustedLayerIdentityDataLocked returns a _consistent_ set of information for a layer with (layerIndex, blobDigest).
// blobDigest is the (possibly-compressed) layer digest referenced in the manifest.
// It returns (trusted, true) if the layer was found, or (_, false) if insufficient data is available.
//
// The caller must hold s.lock.
func (s *storageImageDestination) trustedLayerIdentityDataLocked(layerIndex int, blobDigest digest.Digest) (trustedLayerIdentityData, bool) {
// The decision about layerIdentifiedByTOC must be _stable_ once the data for layerIndex is set,
// even if s.lockProtected.blobDiffIDs changes later and we can subsequently find an entry that wasnt originally available.
//
// If we previously didn't have a blobDigest match and decided to use the TOC, but _later_ we happen to find
// a blobDigest match, we might in principle want to reconsider, set layerIdentifiedByTOC to false, and use the file:
// but the layer in question, and possibly child layers, might already have been committed to storage.
// A late-arriving addition to s.lockProtected.blobDiffIDs would mean that we would want to set
// new layer IDs for potentially the whole parent chain = throw away the just-created layers and create them all again.
//
// Such a within-image layer reuse is expected to be pretty rare; instead, ignore the unexpected file match
// and proceed to the originally-planned TOC match.
res := trustedLayerIdentityData{}
diffID, layerIdentifiedByDiffID := s.lockProtected.indexToDiffID[layerIndex]
if layerIdentifiedByDiffID {
res.layerIdentifiedByTOC = false
res.diffID = diffID
}
if tocDigest, ok := s.lockProtected.indexToTOCDigest[layerIndex]; ok {
res.tocDigest = tocDigest
if !layerIdentifiedByDiffID {
res.layerIdentifiedByTOC = true
}
}
if otherDiffID, ok := s.lockProtected.blobDiffIDs[blobDigest]; ok {
if !layerIdentifiedByDiffID && !res.layerIdentifiedByTOC {
// This is the only data we have, so it is clearly self-consistent.
res.layerIdentifiedByTOC = false
res.diffID = otherDiffID
res.blobDigest = blobDigest
layerIdentifiedByDiffID = true
} else {
// We have set up the layer identity without referring to blobDigest:
// an attacker might have used a manifest with non-matching tocDigest and blobDigest.
// But, if we know a trusted diffID value from other sources, and it matches the one for blobDigest,
// we know blobDigest is fine as well.
if res.diffID != "" && otherDiffID == res.diffID {
res.blobDigest = blobDigest
}
}
}
if !layerIdentifiedByDiffID && !res.layerIdentifiedByTOC {
return trustedLayerIdentityData{}, false // We found nothing at all
}
return res, true
}
// computeID computes a recommended image ID based on information we have so far. If
// the manifest is not of a type that we recognize, we return an empty value, indicating
// that since we don't have a recommendation, a random ID should be used if one needs
// to be allocated.
func (s *storageImageDestination) computeID(m manifest.Manifest) string {
func (s *storageImageDestination) computeID(m manifest.Manifest) (string, error) {
// This is outside of the scope of HasThreadSafePutBlob, so we dont need to hold s.lock.
layerInfos := m.LayerInfos()
// Build the diffID list. We need the decompressed sums that we've been calculating to
// fill in the DiffIDs. It's expected (but not enforced by us) that the number of
// diffIDs corresponds to the number of non-EmptyLayer entries in the history.
var diffIDs []digest.Digest
switch m := m.(type) {
switch m.(type) {
case *manifest.Schema1:
// Build a list of the diffIDs we've generated for the non-throwaway FS layers,
// in reverse of the order in which they were originally listed.
for i, compat := range m.ExtractedV1Compatibility {
if compat.ThrowAway {
// Build a list of the diffIDs we've generated for the non-throwaway FS layers
for i, li := range layerInfos {
if li.EmptyLayer {
continue
}
blobSum := m.FSLayers[i].BlobSum
diffID, ok := s.lockProtected.blobDiffIDs[blobSum]
if !ok {
// this can, in principle, legitimately happen when a layer is reused by TOC.
logrus.Infof("error looking up diffID for layer %q", blobSum.String())
return ""
trusted, ok := s.trustedLayerIdentityDataLocked(i, li.Digest)
if !ok { // We have already committed all layers if we get to this point, so the data must have been available.
return "", fmt.Errorf("internal inconsistency: layer (%d, %q) not found", i, li.Digest)
}
diffIDs = append([]digest.Digest{diffID}, diffIDs...)
if trusted.diffID == "" {
if trusted.layerIdentifiedByTOC {
logrus.Infof("v2s1 image uses a layer identified by TOC with unknown diffID; choosing a random image ID")
return "", nil
}
return "", fmt.Errorf("internal inconsistency: layer (%d, %q) is not identified by TOC and has no diffID", i, li.Digest)
}
diffIDs = append(diffIDs, trusted.diffID)
}
case *manifest.Schema2, *manifest.OCI1:
// We know the ID calculation doesn't actually use the diffIDs, so we don't need to populate
// the diffID list.
default:
return ""
return "", nil
}
// We want to use the same ID for “the same” images, but without risking unwanted sharing / malicious image corruption.
//
// Traditionally that means the same ~config digest, as computed by m.ImageID;
// but if we pull a layer by TOC, we verify the layer against neither the (compressed) blob digest in the manifest,
// but if we identify a layer by TOC, we verify the layer against neither the (compressed) blob digest in the manifest,
// nor against the configs RootFS.DiffIDs. We dont really want to do either, to allow partial layer pulls where we never see
// most of the data.
//
// So, if a layer is pulled by TOC (and we do validate against the TOC), the fact that we used the TOC, and the value of the TOC,
// So, if a layer is identified by TOC (and we do validate against the TOC), the fact that we used the TOC, and the value of the TOC,
// must enter into the image ID computation.
// But for images where no TOC was used, continue to use IDs computed the traditional way, to maximize image reuse on upgrades,
// and to introduce the changed behavior only when partial pulls are used.
@ -566,28 +676,31 @@ func (s *storageImageDestination) computeID(m manifest.Manifest) string {
// (skopeo copy --format v2s2 docker://…/zstd-chunked-image containers-storage:… ). So this is not happening only in the OCI case above.
ordinaryImageID, err := m.ImageID(diffIDs)
if err != nil {
return ""
return "", err
}
tocIDInput := ""
hasLayerPulledByTOC := false
for i := range m.LayerInfos() {
layerValue := "" // An empty string is not a valid digest, so this is unambiguous with the TOC case.
tocDigest, ok := s.lockProtected.indexToTOCDigest[i] // "" if not a TOC
if ok {
for i, li := range layerInfos {
trusted, ok := s.trustedLayerIdentityDataLocked(i, li.Digest)
if !ok { // We have already committed all layers if we get to this point, so the data must have been available.
return "", fmt.Errorf("internal inconsistency: layer (%d, %q) not found", i, li.Digest)
}
layerValue := "" // An empty string is not a valid digest, so this is unambiguous with the TOC case.
if trusted.layerIdentifiedByTOC {
hasLayerPulledByTOC = true
layerValue = tocDigest.String()
layerValue = trusted.tocDigest.String()
}
tocIDInput += layerValue + "|" // "|" can not be present in a TOC digest, so this is an unambiguous separator.
}
if !hasLayerPulledByTOC {
return ordinaryImageID
return ordinaryImageID, nil
}
// ordinaryImageID is a digest of a config, which is a JSON value.
// To avoid the risk of collisions, start the input with @ so that the input is not a valid JSON.
tocImageID := digest.FromString("@With TOC:" + tocIDInput).Encoded()
logrus.Debugf("Ordinary storage image ID %s; a layer was looked up by TOC, so using image ID %s", ordinaryImageID, tocImageID)
return tocImageID
return tocImageID, nil
}
// getConfigBlob exists only to let us retrieve the configuration blob so that the manifest package can dig
@ -671,14 +784,14 @@ func (s *storageImageDestination) singleLayerIDComponent(layerIndex int, blobDig
s.lock.Lock()
defer s.lock.Unlock()
if d, found := s.lockProtected.indexToTOCDigest[layerIndex]; found {
return "@TOC=" + d.Encoded(), false // "@" is not a valid start of a digest.Digest, so this is unambiguous.
trusted, ok := s.trustedLayerIdentityDataLocked(layerIndex, blobDigest)
if !ok {
return "", false
}
if d, found := s.lockProtected.blobDiffIDs[blobDigest]; found {
return d.Encoded(), true // This looks like chain IDs, and it uses the traditional value.
if trusted.layerIdentifiedByTOC {
return "@TOC=" + trusted.tocDigest.Encoded(), false // "@" is not a valid start of a digest.Digest, so this is unambiguous.
}
return "", false
return trusted.diffID.Encoded(), true // This looks like chain IDs, and it uses the traditional value.
}
// commitLayer commits the specified layer with the given index to the storage.
@ -778,6 +891,16 @@ func (s *storageImageDestination) createNewLayer(index int, layerDigest digest.D
diffOutput, ok := s.lockProtected.diffOutputs[index]
s.lock.Unlock()
if ok {
// If we know a trusted DiffID value (e.g. from a BlobInfoCache), set it in diffOutput.
// That way it will be persisted in storage even if the cache is deleted; also
// we can use the value below to avoid the untrustedUncompressedDigest logic (and notably
// the costly commit delay until a manifest is available).
s.lock.Lock()
if d, ok := s.lockProtected.indexToDiffID[index]; ok {
diffOutput.UncompressedDigest = d
}
s.lock.Unlock()
var untrustedUncompressedDigest digest.Digest
if diffOutput.UncompressedDigest == "" {
d, err := s.untrustedLayerDiffID(index)
@ -832,47 +955,43 @@ func (s *storageImageDestination) createNewLayer(index int, layerDigest digest.D
// Check if we previously cached a file with that blob's contents. If we didn't,
// then we need to read the desired contents from a layer.
var trustedUncompressedDigest, trustedOriginalDigest digest.Digest // For storage.LayerOptions
var filename string
var gotFilename bool
s.lock.Lock()
tocDigest := s.lockProtected.indexToTOCDigest[index] // "" if not set
optionalDiffID := s.lockProtected.blobDiffIDs[layerDigest] // "" if not set
filename, gotFilename := s.lockProtected.filenames[layerDigest]
trusted, ok := s.trustedLayerIdentityDataLocked(index, layerDigest)
if ok && trusted.blobDigest != "" {
filename, gotFilename = s.lockProtected.filenames[trusted.blobDigest]
}
s.lock.Unlock()
if gotFilename && tocDigest == "" {
// If tocDigest != "", if we now happen to find a layerDigest match, the newLayerID has already been computed as TOC-based,
// and we don't know the relationship of the layerDigest and TOC digest.
// We could recompute newLayerID to be DiffID-based and use the file, but such a within-image layer
// reuse is expected to be pretty rare; instead, ignore the unexpected file match and proceed to the
// originally-planned TOC match.
// Because tocDigest == "", optionaldiffID must have been set; and even if it werent, PutLayer will recompute the digest from the stream.
trustedUncompressedDigest = optionalDiffID
trustedOriginalDigest = layerDigest // The code setting .filenames[layerDigest] is responsible for the contents matching.
if !ok { // We have already determined newLayerID, so the data must have been available.
return nil, fmt.Errorf("internal inconsistency: layer (%d, %q) not found", index, layerDigest)
}
var trustedOriginalDigest digest.Digest // For storage.LayerOptions
if gotFilename {
// The code setting .filenames[trusted.blobDigest] is responsible for ensuring that the file contents match trusted.blobDigest.
trustedOriginalDigest = trusted.blobDigest
} else {
// Try to find the layer with contents matching the data we use.
var layer *storage.Layer // = nil
if tocDigest != "" {
layers, err2 := s.imageRef.transport.store.LayersByTOCDigest(tocDigest)
if err2 == nil && len(layers) > 0 {
if trusted.diffID != "" {
if layers, err2 := s.imageRef.transport.store.LayersByUncompressedDigest(trusted.diffID); err2 == nil && len(layers) > 0 {
layer = &layers[0]
} else {
return nil, fmt.Errorf("locating layer for TOC digest %q: %w", tocDigest, err2)
}
} else {
// Because tocDigest == "", optionaldiffID must have been set
layers, err2 := s.imageRef.transport.store.LayersByUncompressedDigest(optionalDiffID)
if err2 == nil && len(layers) > 0 {
layer = &layers[0]
} else {
layers, err2 = s.imageRef.transport.store.LayersByCompressedDigest(layerDigest)
if err2 == nil && len(layers) > 0 {
layer = &layers[0]
}
}
if layer == nil {
return nil, fmt.Errorf("locating layer for blob %q: %w", layerDigest, err2)
}
}
if layer == nil && trusted.tocDigest != "" {
if layers, err2 := s.imageRef.transport.store.LayersByTOCDigest(trusted.tocDigest); err2 == nil && len(layers) > 0 {
layer = &layers[0]
}
}
if layer == nil && trusted.blobDigest != "" {
if layers, err2 := s.imageRef.transport.store.LayersByCompressedDigest(trusted.blobDigest); err2 == nil && len(layers) > 0 {
layer = &layers[0]
}
}
if layer == nil {
return nil, fmt.Errorf("layer for blob %q/%q/%q not found", trusted.blobDigest, trusted.tocDigest, trusted.diffID)
}
// Read the layer's contents.
noCompression := archive.Uncompressed
diffOptions := &storage.DiffOptions{
@ -880,7 +999,7 @@ func (s *storageImageDestination) createNewLayer(index int, layerDigest digest.D
}
diff, err2 := s.imageRef.transport.store.Diff("", layer.ID, diffOptions)
if err2 != nil {
return nil, fmt.Errorf("reading layer %q for blob %q: %w", layer.ID, layerDigest, err2)
return nil, fmt.Errorf("reading layer %q for blob %q/%q/%q: %w", layer.ID, trusted.blobDigest, trusted.tocDigest, trusted.diffID, err2)
}
// Copy the layer diff to a file. Diff() takes a lock that it holds
// until the ReadCloser that it returns is closed, and PutLayer() wants
@ -902,20 +1021,19 @@ func (s *storageImageDestination) createNewLayer(index int, layerDigest digest.D
return nil, fmt.Errorf("storing blob to file %q: %w", filename, err)
}
if optionalDiffID == "" && layer.UncompressedDigest != "" {
optionalDiffID = layer.UncompressedDigest
if trusted.diffID == "" && layer.UncompressedDigest != "" {
trusted.diffID = layer.UncompressedDigest // This data might have been unavailable in tryReusingBlobAsPending, and is only known now.
}
// The stream we have is uncompressed, this matches contents of the stream.
// If tocDigest != "", trustedUncompressedDigest might still be ""; in that case PutLayer will compute the value from the stream.
trustedUncompressedDigest = optionalDiffID
// FIXME? trustedOriginalDigest could be set to layerDigest IF tocDigest == "" (otherwise layerDigest is untrusted).
// The stream we have is uncompressed, and it matches trusted.diffID (if known).
//
// FIXME? trustedOriginalDigest could be set to trusted.blobDigest if known, to allow more layer reuse.
// But for c/storage to reasonably use it (as a CompressedDigest value), we should also ensure the CompressedSize of the created
// layer is correct, and the API does not currently make it possible (.CompressedSize is set from the input stream).
//
// We can legitimately set storage.LayerOptions.OriginalDigest to "",
// but that would just result in PutLayer computing the digest of the input stream == optionalDiffID.
// but that would just result in PutLayer computing the digest of the input stream == trusted.diffID.
// So, instead, set .OriginalDigest to the value we know already, to avoid that digest computation.
trustedOriginalDigest = optionalDiffID
trustedOriginalDigest = trusted.diffID
// Allow using the already-collected layer contents without extracting the layer again.
//
@ -923,11 +1041,11 @@ func (s *storageImageDestination) createNewLayer(index int, layerDigest digest.D
// We dont have the original compressed data here to trivially set filenames[layerDigest].
// In particular we cant achieve the correct Layer.CompressedSize value with the current c/storage API.
// Within-image layer reuse is probably very rare, for now we prefer to avoid that complexity.
if trustedUncompressedDigest != "" {
if trusted.diffID != "" {
s.lock.Lock()
s.lockProtected.blobDiffIDs[trustedUncompressedDigest] = trustedUncompressedDigest
s.lockProtected.filenames[trustedUncompressedDigest] = filename
s.lockProtected.fileSizes[trustedUncompressedDigest] = fileSize
s.lockProtected.blobDiffIDs[trusted.diffID] = trusted.diffID
s.lockProtected.filenames[trusted.diffID] = filename
s.lockProtected.fileSizes[trusted.diffID] = fileSize
s.lock.Unlock()
}
}
@ -940,11 +1058,12 @@ func (s *storageImageDestination) createNewLayer(index int, layerDigest digest.D
// Build the new layer using the diff, regardless of where it came from.
// TODO: This can take quite some time, and should ideally be cancellable using ctx.Done().
layer, _, err := s.imageRef.transport.store.PutLayer(newLayerID, parentLayer, nil, "", false, &storage.LayerOptions{
OriginalDigest: trustedOriginalDigest,
UncompressedDigest: trustedUncompressedDigest,
OriginalDigest: trustedOriginalDigest,
// This might be "" if trusted.layerIdentifiedByTOC; in that case PutLayer will compute the value from the stream.
UncompressedDigest: trusted.diffID,
}, file)
if err != nil && !errors.Is(err, storage.ErrDuplicateID) {
return nil, fmt.Errorf("adding layer with blob %q: %w", layerDigest, err)
return nil, fmt.Errorf("adding layer with blob %q/%q/%q: %w", trusted.blobDigest, trusted.tocDigest, trusted.diffID, err)
}
return layer, nil
}
@ -1155,7 +1274,10 @@ func (s *storageImageDestination) Commit(ctx context.Context, unparsedToplevel t
// Create the image record, pointing to the most-recently added layer.
intendedID := s.imageRef.id
if intendedID == "" {
intendedID = s.computeID(man)
intendedID, err = s.computeID(man)
if err != nil {
return err
}
}
oldNames := []string{}
img, err := s.imageRef.transport.store.CreateImage(intendedID, nil, lastLayer, "", options)

View File

@ -6,12 +6,12 @@ const (
// VersionMajor is for an API incompatible changes
VersionMajor = 5
// VersionMinor is for functionality in a backwards-compatible manner
VersionMinor = 32
VersionMinor = 33
// VersionPatch is for backwards-compatible bug fixes
VersionPatch = 0
// VersionDev indicates development branch. Releases will be empty string.
VersionDev = ""
VersionDev = "-dev"
)
// Version is the specification version that the package types support.