package distribution import ( "errors" "fmt" "io" "runtime" "sort" "strings" "sync" "golang.org/x/net/context" "github.com/Sirupsen/logrus" "github.com/docker/distribution" "github.com/docker/distribution/digest" "github.com/docker/distribution/manifest/schema1" "github.com/docker/distribution/manifest/schema2" distreference "github.com/docker/distribution/reference" "github.com/docker/distribution/registry/client" apitypes "github.com/docker/docker/api/types" "github.com/docker/docker/distribution/metadata" "github.com/docker/docker/distribution/xfer" "github.com/docker/docker/layer" "github.com/docker/docker/pkg/ioutils" "github.com/docker/docker/pkg/progress" "github.com/docker/docker/pkg/stringid" "github.com/docker/docker/reference" "github.com/docker/docker/registry" ) const ( smallLayerMaximumSize = 100 * (1 << 10) // 100KB middleLayerMaximumSize = 10 * (1 << 20) // 10MB ) type v2Pusher struct { v2MetadataService metadata.V2MetadataService ref reference.Named endpoint registry.APIEndpoint repoInfo *registry.RepositoryInfo config *ImagePushConfig repo distribution.Repository // pushState is state built by the Upload functions. pushState pushState } type pushState struct { sync.Mutex // remoteLayers is the set of layers known to exist on the remote side. // This avoids redundant queries when pushing multiple tags that // involve the same layers. It is also used to fill in digest and size // information when building the manifest. remoteLayers map[layer.DiffID]distribution.Descriptor // confirmedV2 is set to true if we confirm we're talking to a v2 // registry. This is used to limit fallbacks to the v1 protocol. confirmedV2 bool } func (p *v2Pusher) Push(ctx context.Context) (err error) { p.pushState.remoteLayers = make(map[layer.DiffID]distribution.Descriptor) p.repo, p.pushState.confirmedV2, err = NewV2Repository(ctx, p.repoInfo, p.endpoint, p.config.MetaHeaders, p.config.AuthConfig, "push", "pull") if err != nil { logrus.Debugf("Error getting v2 registry: %v", err) return err } if err = p.pushV2Repository(ctx); err != nil { if continueOnError(err) { return fallbackError{ err: err, confirmedV2: p.pushState.confirmedV2, transportOK: true, } } } return err } func (p *v2Pusher) pushV2Repository(ctx context.Context) (err error) { if namedTagged, isNamedTagged := p.ref.(reference.NamedTagged); isNamedTagged { imageID, err := p.config.ReferenceStore.Get(p.ref) if err != nil { return fmt.Errorf("tag does not exist: %s", p.ref.String()) } return p.pushV2Tag(ctx, namedTagged, imageID) } if !reference.IsNameOnly(p.ref) { return errors.New("cannot push a digest reference") } // Pull all tags pushed := 0 for _, association := range p.config.ReferenceStore.ReferencesByName(p.ref) { if namedTagged, isNamedTagged := association.Ref.(reference.NamedTagged); isNamedTagged { pushed++ if err := p.pushV2Tag(ctx, namedTagged, association.ID); err != nil { return err } } } if pushed == 0 { return fmt.Errorf("no tags to push for %s", p.repoInfo.Name()) } return nil } func (p *v2Pusher) pushV2Tag(ctx context.Context, ref reference.NamedTagged, id digest.Digest) error { logrus.Debugf("Pushing repository: %s", ref.String()) imgConfig, err := p.config.ImageStore.Get(id) if err != nil { return fmt.Errorf("could not find image from tag %s: %v", ref.String(), err) } rootfs, err := p.config.ImageStore.RootFSFromConfig(imgConfig) if err != nil { return fmt.Errorf("unable to get rootfs for image %s: %s", ref.String(), err) } l, err := p.config.LayerStore.Get(rootfs.ChainID()) if err != nil { return fmt.Errorf("failed to get top layer from image: %v", err) } defer l.Release() hmacKey, err := metadata.ComputeV2MetadataHMACKey(p.config.AuthConfig) if err != nil { return fmt.Errorf("failed to compute hmac key of auth config: %v", err) } var descriptors []xfer.UploadDescriptor descriptorTemplate := v2PushDescriptor{ v2MetadataService: p.v2MetadataService, hmacKey: hmacKey, repoInfo: p.repoInfo, ref: p.ref, repo: p.repo, pushState: &p.pushState, } // Loop bounds condition is to avoid pushing the base layer on Windows. for i := 0; i < len(rootfs.DiffIDs); i++ { descriptor := descriptorTemplate descriptor.layer = l descriptor.checkedDigests = make(map[digest.Digest]struct{}) descriptors = append(descriptors, &descriptor) l = l.Parent() } if err := p.config.UploadManager.Upload(ctx, descriptors, p.config.ProgressOutput); err != nil { return err } // Try schema2 first builder := schema2.NewManifestBuilder(p.repo.Blobs(ctx), p.config.ConfigMediaType, imgConfig) manifest, err := manifestFromBuilder(ctx, builder, descriptors) if err != nil { return err } manSvc, err := p.repo.Manifests(ctx) if err != nil { return err } putOptions := []distribution.ManifestServiceOption{distribution.WithTag(ref.Tag())} if _, err = manSvc.Put(ctx, manifest, putOptions...); err != nil { if runtime.GOOS == "windows" || p.config.TrustKey == nil || p.config.RequireSchema2 { logrus.Warnf("failed to upload schema2 manifest: %v", err) return err } logrus.Warnf("failed to upload schema2 manifest: %v - falling back to schema1", err) manifestRef, err := distreference.WithTag(p.repo.Named(), ref.Tag()) if err != nil { return err } builder = schema1.NewConfigManifestBuilder(p.repo.Blobs(ctx), p.config.TrustKey, manifestRef, imgConfig) manifest, err = manifestFromBuilder(ctx, builder, descriptors) if err != nil { return err } if _, err = manSvc.Put(ctx, manifest, putOptions...); err != nil { return err } } var canonicalManifest []byte switch v := manifest.(type) { case *schema1.SignedManifest: canonicalManifest = v.Canonical case *schema2.DeserializedManifest: _, canonicalManifest, err = v.Payload() if err != nil { return err } } manifestDigest := digest.FromBytes(canonicalManifest) progress.Messagef(p.config.ProgressOutput, "", "%s: digest: %s size: %d", ref.Tag(), manifestDigest, len(canonicalManifest)) if err := addDigestReference(p.config.ReferenceStore, ref, manifestDigest, id); err != nil { return err } // Signal digest to the trust client so it can sign the // push, if appropriate. progress.Aux(p.config.ProgressOutput, apitypes.PushResult{Tag: ref.Tag(), Digest: manifestDigest.String(), Size: len(canonicalManifest)}) return nil } func manifestFromBuilder(ctx context.Context, builder distribution.ManifestBuilder, descriptors []xfer.UploadDescriptor) (distribution.Manifest, error) { // descriptors is in reverse order; iterate backwards to get references // appended in the right order. for i := len(descriptors) - 1; i >= 0; i-- { if err := builder.AppendReference(descriptors[i].(*v2PushDescriptor)); err != nil { return nil, err } } return builder.Build(ctx) } type v2PushDescriptor struct { layer PushLayer v2MetadataService metadata.V2MetadataService hmacKey []byte repoInfo reference.Named ref reference.Named repo distribution.Repository pushState *pushState remoteDescriptor distribution.Descriptor // a set of digests whose presence has been checked in a target repository checkedDigests map[digest.Digest]struct{} } func (pd *v2PushDescriptor) Key() string { return "v2push:" + pd.ref.FullName() + " " + pd.layer.DiffID().String() } func (pd *v2PushDescriptor) ID() string { return stringid.TruncateID(pd.layer.DiffID().String()) } func (pd *v2PushDescriptor) DiffID() layer.DiffID { return pd.layer.DiffID() } func (pd *v2PushDescriptor) Upload(ctx context.Context, progressOutput progress.Output) (distribution.Descriptor, error) { if fs, ok := pd.layer.(distribution.Describable); ok { if d := fs.Descriptor(); len(d.URLs) > 0 { progress.Update(progressOutput, pd.ID(), "Skipped foreign layer") return d, nil } } diffID := pd.DiffID() pd.pushState.Lock() if descriptor, ok := pd.pushState.remoteLayers[diffID]; ok { // it is already known that the push is not needed and // therefore doing a stat is unnecessary pd.pushState.Unlock() progress.Update(progressOutput, pd.ID(), "Layer already exists") return descriptor, nil } pd.pushState.Unlock() maxMountAttempts, maxExistenceChecks, checkOtherRepositories := getMaxMountAndExistenceCheckAttempts(pd.layer) // Do we have any metadata associated with this layer's DiffID? v2Metadata, err := pd.v2MetadataService.GetMetadata(diffID) if err == nil { // check for blob existence in the target repository if we have a mapping with it descriptor, exists, err := pd.layerAlreadyExists(ctx, progressOutput, diffID, false, 1, v2Metadata) if exists || err != nil { return descriptor, err } } // if digest was empty or not saved, or if blob does not exist on the remote repository, // then push the blob. bs := pd.repo.Blobs(ctx) var layerUpload distribution.BlobWriter // Attempt to find another repository in the same registry to mount the layer from to avoid an unnecessary upload candidates := getRepositoryMountCandidates(pd.repoInfo, pd.hmacKey, maxMountAttempts, v2Metadata) for _, mountCandidate := range candidates { logrus.Debugf("attempting to mount layer %s (%s) from %s", diffID, mountCandidate.Digest, mountCandidate.SourceRepository) createOpts := []distribution.BlobCreateOption{} if len(mountCandidate.SourceRepository) > 0 { namedRef, err := reference.WithName(mountCandidate.SourceRepository) if err != nil { logrus.Errorf("failed to parse source repository reference %v: %v", namedRef.String(), err) pd.v2MetadataService.Remove(mountCandidate) continue } // TODO (brianbland): We need to construct a reference where the Name is // only the full remote name, so clean this up when distribution has a // richer reference package remoteRef, err := distreference.WithName(namedRef.RemoteName()) if err != nil { logrus.Errorf("failed to make remote reference out of %q: %v", namedRef.RemoteName(), namedRef.RemoteName()) continue } canonicalRef, err := distreference.WithDigest(distreference.TrimNamed(remoteRef), mountCandidate.Digest) if err != nil { logrus.Errorf("failed to make canonical reference: %v", err) continue } createOpts = append(createOpts, client.WithMountFrom(canonicalRef)) } // send the layer lu, err := bs.Create(ctx, createOpts...) switch err := err.(type) { case nil: // noop case distribution.ErrBlobMounted: progress.Updatef(progressOutput, pd.ID(), "Mounted from %s", err.From.Name()) err.Descriptor.MediaType = schema2.MediaTypeLayer pd.pushState.Lock() pd.pushState.confirmedV2 = true pd.pushState.remoteLayers[diffID] = err.Descriptor pd.pushState.Unlock() // Cache mapping from this layer's DiffID to the blobsum if err := pd.v2MetadataService.TagAndAdd(diffID, pd.hmacKey, metadata.V2Metadata{ Digest: err.Descriptor.Digest, SourceRepository: pd.repoInfo.FullName(), }); err != nil { return distribution.Descriptor{}, xfer.DoNotRetry{Err: err} } return err.Descriptor, nil default: logrus.Infof("failed to mount layer %s (%s) from %s: %v", diffID, mountCandidate.Digest, mountCandidate.SourceRepository, err) } if len(mountCandidate.SourceRepository) > 0 && (metadata.CheckV2MetadataHMAC(&mountCandidate, pd.hmacKey) || len(mountCandidate.HMAC) == 0) { cause := "blob mount failure" if err != nil { cause = fmt.Sprintf("an error: %v", err.Error()) } logrus.Debugf("removing association between layer %s and %s due to %s", mountCandidate.Digest, mountCandidate.SourceRepository, cause) pd.v2MetadataService.Remove(mountCandidate) } if lu != nil { // cancel previous upload cancelLayerUpload(ctx, mountCandidate.Digest, layerUpload) layerUpload = lu } } if maxExistenceChecks-len(pd.checkedDigests) > 0 { // do additional layer existence checks with other known digests if any descriptor, exists, err := pd.layerAlreadyExists(ctx, progressOutput, diffID, checkOtherRepositories, maxExistenceChecks-len(pd.checkedDigests), v2Metadata) if exists || err != nil { return descriptor, err } } logrus.Debugf("Pushing layer: %s", diffID) if layerUpload == nil { layerUpload, err = bs.Create(ctx) if err != nil { return distribution.Descriptor{}, retryOnError(err) } } defer layerUpload.Close() // upload the blob desc, err := pd.uploadUsingSession(ctx, progressOutput, diffID, layerUpload) if err != nil { return desc, err } return desc, nil } func (pd *v2PushDescriptor) SetRemoteDescriptor(descriptor distribution.Descriptor) { pd.remoteDescriptor = descriptor } func (pd *v2PushDescriptor) Descriptor() distribution.Descriptor { return pd.remoteDescriptor } func (pd *v2PushDescriptor) uploadUsingSession( ctx context.Context, progressOutput progress.Output, diffID layer.DiffID, layerUpload distribution.BlobWriter, ) (distribution.Descriptor, error) { var reader io.ReadCloser contentReader, err := pd.layer.Open() size, _ := pd.layer.Size() reader = progress.NewProgressReader(ioutils.NewCancelReadCloser(ctx, contentReader), progressOutput, size, pd.ID(), "Pushing") switch m := pd.layer.MediaType(); m { case schema2.MediaTypeUncompressedLayer: compressedReader, compressionDone := compress(reader) defer func(closer io.Closer) { closer.Close() <-compressionDone }(reader) reader = compressedReader case schema2.MediaTypeLayer: default: reader.Close() return distribution.Descriptor{}, fmt.Errorf("unsupported layer media type %s", m) } digester := digest.Canonical.New() tee := io.TeeReader(reader, digester.Hash()) nn, err := layerUpload.ReadFrom(tee) reader.Close() if err != nil { return distribution.Descriptor{}, retryOnError(err) } pushDigest := digester.Digest() if _, err := layerUpload.Commit(ctx, distribution.Descriptor{Digest: pushDigest}); err != nil { return distribution.Descriptor{}, retryOnError(err) } logrus.Debugf("uploaded layer %s (%s), %d bytes", diffID, pushDigest, nn) progress.Update(progressOutput, pd.ID(), "Pushed") // Cache mapping from this layer's DiffID to the blobsum if err := pd.v2MetadataService.TagAndAdd(diffID, pd.hmacKey, metadata.V2Metadata{ Digest: pushDigest, SourceRepository: pd.repoInfo.FullName(), }); err != nil { return distribution.Descriptor{}, xfer.DoNotRetry{Err: err} } desc := distribution.Descriptor{ Digest: pushDigest, MediaType: schema2.MediaTypeLayer, Size: nn, } pd.pushState.Lock() // If Commit succeeded, that's an indication that the remote registry speaks the v2 protocol. pd.pushState.confirmedV2 = true pd.pushState.remoteLayers[diffID] = desc pd.pushState.Unlock() return desc, nil } // layerAlreadyExists checks if the registry already knows about any of the metadata passed in the "metadata" // slice. If it finds one that the registry knows about, it returns the known digest and "true". If // "checkOtherRepositories" is true, stat will be performed also with digests mapped to any other repository // (not just the target one). func (pd *v2PushDescriptor) layerAlreadyExists( ctx context.Context, progressOutput progress.Output, diffID layer.DiffID, checkOtherRepositories bool, maxExistenceCheckAttempts int, v2Metadata []metadata.V2Metadata, ) (desc distribution.Descriptor, exists bool, err error) { // filter the metadata candidates := []metadata.V2Metadata{} for _, meta := range v2Metadata { if len(meta.SourceRepository) > 0 && !checkOtherRepositories && meta.SourceRepository != pd.repoInfo.FullName() { continue } candidates = append(candidates, meta) } // sort the candidates by similarity sortV2MetadataByLikenessAndAge(pd.repoInfo, pd.hmacKey, candidates) digestToMetadata := make(map[digest.Digest]*metadata.V2Metadata) // an array of unique blob digests ordered from the best mount candidates to worst layerDigests := []digest.Digest{} for i := 0; i < len(candidates); i++ { if len(layerDigests) >= maxExistenceCheckAttempts { break } meta := &candidates[i] if _, exists := digestToMetadata[meta.Digest]; exists { // keep reference just to the first mapping (the best mount candidate) continue } if _, exists := pd.checkedDigests[meta.Digest]; exists { // existence of this digest has already been tested continue } digestToMetadata[meta.Digest] = meta layerDigests = append(layerDigests, meta.Digest) } attempts: for _, dgst := range layerDigests { meta := digestToMetadata[dgst] logrus.Debugf("Checking for presence of layer %s (%s) in %s", diffID, dgst, pd.repoInfo.FullName()) desc, err = pd.repo.Blobs(ctx).Stat(ctx, dgst) pd.checkedDigests[meta.Digest] = struct{}{} switch err { case nil: if m, ok := digestToMetadata[desc.Digest]; !ok || m.SourceRepository != pd.repoInfo.FullName() || !metadata.CheckV2MetadataHMAC(m, pd.hmacKey) { // cache mapping from this layer's DiffID to the blobsum if err := pd.v2MetadataService.TagAndAdd(diffID, pd.hmacKey, metadata.V2Metadata{ Digest: desc.Digest, SourceRepository: pd.repoInfo.FullName(), }); err != nil { return distribution.Descriptor{}, false, xfer.DoNotRetry{Err: err} } } desc.MediaType = schema2.MediaTypeLayer exists = true break attempts case distribution.ErrBlobUnknown: if meta.SourceRepository == pd.repoInfo.FullName() { // remove the mapping to the target repository pd.v2MetadataService.Remove(*meta) } default: logrus.WithError(err).Debugf("Failed to check for presence of layer %s (%s) in %s", diffID, dgst, pd.repoInfo.FullName()) } } if exists { progress.Update(progressOutput, pd.ID(), "Layer already exists") pd.pushState.Lock() pd.pushState.remoteLayers[diffID] = desc pd.pushState.Unlock() } return desc, exists, nil } // getMaxMountAndExistenceCheckAttempts returns a maximum number of cross repository mount attempts from // source repositories of target registry, maximum number of layer existence checks performed on the target // repository and whether the check shall be done also with digests mapped to different repositories. The // decision is based on layer size. The smaller the layer, the fewer attempts shall be made because the cost // of upload does not outweigh a latency. func getMaxMountAndExistenceCheckAttempts(layer PushLayer) (maxMountAttempts, maxExistenceCheckAttempts int, checkOtherRepositories bool) { size, err := layer.Size() switch { // big blob case size > middleLayerMaximumSize: // 1st attempt to mount the blob few times // 2nd few existence checks with digests associated to any repository // then fallback to upload return 4, 3, true // middle sized blobs; if we could not get the size, assume we deal with middle sized blob case size > smallLayerMaximumSize, err != nil: // 1st attempt to mount blobs of average size few times // 2nd try at most 1 existence check if there's an existing mapping to the target repository // then fallback to upload return 3, 1, false // small blobs, do a minimum number of checks default: return 1, 1, false } } // getRepositoryMountCandidates returns an array of v2 metadata items belonging to the given registry. The // array is sorted from youngest to oldest. If requireReigstryMatch is true, the resulting array will contain // only metadata entries having registry part of SourceRepository matching the part of repoInfo. func getRepositoryMountCandidates( repoInfo reference.Named, hmacKey []byte, max int, v2Metadata []metadata.V2Metadata, ) []metadata.V2Metadata { candidates := []metadata.V2Metadata{} for _, meta := range v2Metadata { sourceRepo, err := reference.ParseNamed(meta.SourceRepository) if err != nil || repoInfo.Hostname() != sourceRepo.Hostname() { continue } // target repository is not a viable candidate if meta.SourceRepository == repoInfo.FullName() { continue } candidates = append(candidates, meta) } sortV2MetadataByLikenessAndAge(repoInfo, hmacKey, candidates) if max >= 0 && len(candidates) > max { // select the youngest metadata candidates = candidates[:max] } return candidates } // byLikeness is a sorting container for v2 metadata candidates for cross repository mount. The // candidate "a" is preferred over "b": // // 1. if it was hashed using the same AuthConfig as the one used to authenticate to target repository and the // "b" was not // 2. if a number of its repository path components exactly matching path components of target repository is higher type byLikeness struct { arr []metadata.V2Metadata hmacKey []byte pathComponents []string } func (bla byLikeness) Less(i, j int) bool { aMacMatch := metadata.CheckV2MetadataHMAC(&bla.arr[i], bla.hmacKey) bMacMatch := metadata.CheckV2MetadataHMAC(&bla.arr[j], bla.hmacKey) if aMacMatch != bMacMatch { return aMacMatch } aMatch := numOfMatchingPathComponents(bla.arr[i].SourceRepository, bla.pathComponents) bMatch := numOfMatchingPathComponents(bla.arr[j].SourceRepository, bla.pathComponents) return aMatch > bMatch } func (bla byLikeness) Swap(i, j int) { bla.arr[i], bla.arr[j] = bla.arr[j], bla.arr[i] } func (bla byLikeness) Len() int { return len(bla.arr) } func sortV2MetadataByLikenessAndAge(repoInfo reference.Named, hmacKey []byte, marr []metadata.V2Metadata) { // reverse the metadata array to shift the newest entries to the beginning for i := 0; i < len(marr)/2; i++ { marr[i], marr[len(marr)-i-1] = marr[len(marr)-i-1], marr[i] } // keep equal entries ordered from the youngest to the oldest sort.Stable(byLikeness{ arr: marr, hmacKey: hmacKey, pathComponents: getPathComponents(repoInfo.FullName()), }) } // numOfMatchingPathComponents returns a number of path components in "pth" that exactly match "matchComponents". func numOfMatchingPathComponents(pth string, matchComponents []string) int { pthComponents := getPathComponents(pth) i := 0 for ; i < len(pthComponents) && i < len(matchComponents); i++ { if matchComponents[i] != pthComponents[i] { return i } } return i } func getPathComponents(path string) []string { // make sure to add docker.io/ prefix to the path named, err := reference.ParseNamed(path) if err == nil { path = named.FullName() } return strings.Split(path, "/") } func cancelLayerUpload(ctx context.Context, dgst digest.Digest, layerUpload distribution.BlobWriter) { if layerUpload != nil { logrus.Debugf("cancelling upload of blob %s", dgst) err := layerUpload.Cancel(ctx) if err != nil { logrus.Warnf("failed to cancel upload: %v", err) } } }