package containerd import ( "context" "fmt" "strings" "sync/atomic" "time" containerd "github.com/containerd/containerd/v2/client" c8dimages "github.com/containerd/containerd/v2/core/images" "github.com/containerd/containerd/v2/core/remotes/docker" "github.com/containerd/containerd/v2/pkg/snapshotters" cerrdefs "github.com/containerd/errdefs" "github.com/containerd/log" "github.com/containerd/platforms" "github.com/distribution/reference" "github.com/moby/moby/api/types/events" registrytypes "github.com/moby/moby/api/types/registry" "github.com/moby/moby/v2/daemon/internal/distribution" "github.com/moby/moby/v2/daemon/internal/metrics" "github.com/moby/moby/v2/daemon/internal/progress" "github.com/moby/moby/v2/daemon/internal/streamformatter" "github.com/moby/moby/v2/daemon/internal/stringid" "github.com/moby/moby/v2/daemon/server/imagebackend" "github.com/moby/moby/v2/errdefs" ocispec "github.com/opencontainers/image-spec/specs-go/v1" "github.com/pkg/errors" ) // PullImage initiates a pull operation. baseRef is the image to pull. // If reference is not tagged, all tags are pulled. func (i *ImageService) PullImage(ctx context.Context, baseRef reference.Named, options imagebackend.PullOptions) (retErr error) { if len(options.Platforms) > 1 { // TODO(thaJeztah): add support for pulling multiple platforms return cerrdefs.ErrInvalidArgument.WithMessage("multiple platforms is not supported") } start := time.Now() defer func() { if retErr == nil { metrics.ImageActions.WithValues("pull").UpdateSince(start) } }() out := streamformatter.NewJSONProgressOutput(options.OutStream, false) ctx, done, err := i.withLease(ctx, true) if err != nil { return err } defer done() var platform *ocispec.Platform if len(options.Platforms) > 0 { p := options.Platforms[0] platform = &p } if !reference.IsNameOnly(baseRef) { return i.pullTag(ctx, baseRef, platform, options.MetaHeaders, options.AuthConfig, out) } tags, err := distribution.Tags(ctx, baseRef, &distribution.Config{ RegistryService: i.registryService, MetaHeaders: options.MetaHeaders, AuthConfig: options.AuthConfig, }) if err != nil { return err } for _, tag := range tags { ref, err := reference.WithTag(baseRef, tag) if err != nil { log.G(ctx).WithFields(log.Fields{ "tag": tag, "baseRef": baseRef, }).Warn("invalid tag, won't pull") continue } if err := i.pullTag(ctx, ref, platform, options.MetaHeaders, options.AuthConfig, out); err != nil { return fmt.Errorf("error pulling %s: %w", ref, err) } } return nil } func (i *ImageService) pullTag(ctx context.Context, ref reference.Named, platform *ocispec.Platform, metaHeaders map[string][]string, authConfig *registrytypes.AuthConfig, out progress.Output) error { var opts []containerd.RemoteOpt if platform != nil { opts = append(opts, containerd.WithPlatform(platforms.FormatAll(*platform))) } resolver, _ := i.newResolverFromAuthConfig(ctx, authConfig, ref, metaHeaders) opts = append(opts, containerd.WithResolver(resolver)) oldImage, err := i.resolveImage(ctx, ref.String()) if err != nil && !cerrdefs.IsNotFound(err) { return err } // Will be set to the new image after pull succeeds. var outNewImg containerd.Image if oldImage.Target.Digest != "" { err = i.leaseContent(ctx, i.content, oldImage.Target) if err != nil { return errdefs.System(fmt.Errorf("failed to lease content: %w", err)) } // If the pulled image is different than the old image, we will keep the old image as a dangling image. defer func() { if outNewImg != nil { if outNewImg.Target().Digest != oldImage.Target.Digest { if err := i.ensureDanglingImage(ctx, oldImage); err != nil { log.G(ctx).WithError(err).Warn("failed to keep the previous image as dangling") } } } }() } p := platforms.Default() if platform != nil { p = platforms.Only(*platform) } pullJobs := newJobs() opts = append(opts, containerd.WithImageHandler(c8dimages.HandlerFunc(func(ctx context.Context, desc ocispec.Descriptor) ([]ocispec.Descriptor, error) { if showBlobProgress(desc) { pullJobs.Add(desc) } return nil, nil }))) pp := &pullProgress{ store: i.content, snapshotter: i.snapshotterService(i.snapshotter), showExists: true, } finishProgress := pullJobs.showProgress(ctx, out, pp) defer func() { finishProgress() // Send final status message after the progress updater has finished. // Otherwise the layer/manifest progress messages may arrive AFTER the // status message have been sent, so they won't update the previous // progress leaving stale progress like: // 70f5ac315c5a: Downloading [> ] 0B/3.19kB // Digest: sha256:4f53e2564790c8e7856ec08e384732aa38dc43c52f02952483e3f003afbf23db // 70f5ac315c5a: Download complete // Status: Downloaded newer image for hello-world:latest // docker.io/library/hello-world:latest if outNewImg != nil { img := outNewImg progress.Message(out, "", "Digest: "+img.Target().Digest.String()) newer := oldImage.Target.Digest != img.Target().Digest writeStatus(out, reference.FamiliarString(ref), newer) } }() var sentPullingFrom, sentModelNotSupported atomic.Bool ah := c8dimages.HandlerFunc(func(ctx context.Context, desc ocispec.Descriptor) ([]ocispec.Descriptor, error) { if desc.MediaType == c8dimages.MediaTypeDockerSchema1Manifest { return nil, distribution.DeprecatedSchema1ImageError(ref) } ociAiArtifactManifest := c8dimages.IsManifestType(desc.MediaType) && isModelMediaType(desc.ArtifactType) aiMediaType := isModelMediaType(desc.MediaType) if ociAiArtifactManifest || aiMediaType { if !sentModelNotSupported.Load() { sentModelNotSupported.Store(true) progress.Message(out, "", `WARNING: AI models are not supported by the Engine yet, did you mean to use "docker model pull/run" instead?`) } } if c8dimages.IsLayerType(desc.MediaType) { id := stringid.TruncateID(desc.Digest.String()) progress.Update(out, id, "Pulling fs layer") } if c8dimages.IsManifestType(desc.MediaType) { if !sentPullingFrom.Load() { var tagOrDigest string if tagged, ok := ref.(reference.Tagged); ok { tagOrDigest = tagged.Tag() } else { tagOrDigest = ref.String() } progress.Message(out, tagOrDigest, "Pulling from "+reference.Path(ref)) sentPullingFrom.Store(true) } available, _, _, missing, err := c8dimages.Check(ctx, i.content, desc, p) if err != nil { return nil, err } // If we already have all the contents pull shouldn't show any layer // download progress, not even a "Already present" message. if available && len(missing) == 0 { pp.hideLayers = true } } return nil, nil }) opts = append(opts, containerd.WithImageHandler(ah)) opts = append(opts, containerd.WithPullUnpack) // TODO(thaJeztah): we may have to pass the snapshotter to use if the pull is part of a "docker run" (container create -> pull image if missing). See https://github.com/moby/moby/issues/45273 opts = append(opts, containerd.WithPullSnapshotter(i.snapshotter)) // AppendInfoHandlerWrapper will annotate the image with basic information like manifest and layer digests as labels; // this information is used to enable remote snapshotters like nydus and stargz to query a registry. // This is also needed for the pull progress to detect the `Extracting` status. infoHandler := snapshotters.AppendInfoHandlerWrapper(ref.String()) opts = append(opts, containerd.WithImageHandlerWrapper(infoHandler)) img, err := i.client.Pull(ctx, ref.String(), opts...) if err != nil { if errors.Is(err, docker.ErrInvalidAuthorization) { // Match error returned by containerd. // https://github.com/containerd/containerd/blob/v2.1.1/core/remotes/docker/authorizer.go#L201-L203 if strings.Contains(err.Error(), "no basic auth credentials") { return err } return errdefs.NotFound(fmt.Errorf("pull access denied for %s, repository does not exist or may require 'docker login'", reference.FamiliarName(ref))) } if cerrdefs.IsNotFound(err) { // Transform "no match for platform in manifest" error returned by containerd into // the same message as the graphdrivers backend. // The one returned by containerd doesn't contain the platform and is much less informative. if strings.Contains(err.Error(), "platform") { platformStr := platforms.DefaultString() if platform != nil { platformStr = platforms.FormatAll(*platform) } return errdefs.NotFound(fmt.Errorf("no matching manifest for %s in the manifest list entries: %w", platformStr, err)) } } return translateRegistryError(ctx, err) } logger := log.G(ctx).WithFields(log.Fields{ "digest": img.Target().Digest, "remote": ref.String(), }) logger.Info("image pulled") // The pull succeeded, so try to remove any dangling image we have for this target err = i.images.Delete(context.WithoutCancel(ctx), danglingImageName(img.Target().Digest)) if err != nil && !cerrdefs.IsNotFound(err) { // Image pull succeeded, but cleaning up the dangling image failed. Ignore the // error to not mark the pull as failed. logger.WithError(err).Warn("unexpected error while removing outdated dangling image reference") } i.LogImageEvent(ctx, reference.FamiliarString(ref), reference.FamiliarName(ref), events.ActionPull) outNewImg = img return nil } // writeStatus writes a status message to out. If newerDownloaded is true, the // status message indicates that a newer image was downloaded. Otherwise, it // indicates that the image is up to date. requestedTag is the tag the message // will refer to. func writeStatus(out progress.Output, requestedTag string, newerDownloaded bool) { if newerDownloaded { progress.Message(out, "", "Status: Downloaded newer image for "+requestedTag) } else { progress.Message(out, "", "Status: Image is up to date for "+requestedTag) } } func isModelMediaType(mediaType string) bool { return strings.HasPrefix(strings.ToLower(mediaType), "application/vnd.docker.ai.") }