docker-library · tianon · Nov 3, 2022 · Oct 22, 2022
@@ -17,11 +17,20 @@ jobs:
     steps:
       - uses: actions/checkout@v3
       - run: docker build --pull --tag oisupport/perl-bashbrew .
-      - uses: actions/checkout@v3
-        with:
-          repository: docker-library/official-images
-          ref: b851f53b765e622928a94663d425ddbe37aceb3b
-          path: oi
-      - run: echo "BASHBREW_LIBRARY=$PWD/oi/library" >> "$GITHUB_ENV"
-      - run: docker run -dit --name registry --restart always -p 5000:5000 registry
-      - run: ./test-localhost.sh
+      - name: Install "bashbrew"
+        run: |
+          git clone --depth 1 https://github.com/docker-library/bashbrew.git -b master bashbrew
+          ./bashbrew/bashbrew.sh --version > /dev/null
+          export PATH="$PWD/bashbrew/bin:$PATH"
+          echo "PATH=$PATH" >> "$GITHUB_ENV"
+          bashbrew --version
+      - name: Clone DOI
+        run: |
+          git clone --depth 1 https://github.com/docker-library/official-images.git oi
+          export BASHBREW_LIBRARY="$PWD/oi/library"
+          echo "BASHBREW_LIBRARY=$BASHBREW_LIBRARY" >> "$GITHUB_ENV"
+          bashbrew from hello-world:latest > /dev/null
+      - run: docker run -dit --name registry --restart always -p 5000:5000 --env REGISTRY_VALIDATION_MANIFESTS_URLS_ALLOW='["^.*$"]' --env REGISTRY_VALIDATION_MANIFESTS_URLS_DENY='[]' registry
+      - run: ./test-localhost.sh hello-world:latest # includes Windows images
+      - run: ./test-localhost.sh hello-world:nanoserver-ltsc2022 # forces an "os.version from the config blob" lookup
+      - run: ./test-localhost.sh busybox:latest # includes three separate "linux/arm" variants
@@ -31,39 +31,106 @@
 sub get_arch_p ($targetRef, $arch, $archRef) {
 	return $ua->get_manifest_p($archRef)->then(sub ($manifestData = undef) {
 		return unless $manifestData;
-		my ($digest, $manifest, $size) = ($manifestData->{digest}, $manifestData->{manifest}, $manifestData->{size});
+		my ($mediaType, $digest, $size, $manifest) = (
+			$manifestData->{mediaType},
+			$manifestData->{digest},
+			$manifestData->{size},
+			$manifestData->{manifest},
+		);
 
-		my $mediaType = $manifestData->{mediaType};
-		if ($mediaType eq Bashbrew::RegistryUserAgent::MEDIA_OCI_INDEX_V1 || $mediaType eq Bashbrew::RegistryUserAgent::MEDIA_MANIFEST_LIST) {
-			# jackpot -- if it's already a manifest list, the hard work is done!
-			return ($archRef, $manifest->{manifests});
-			# TODO we should validate the "platform" values here to make sure we're not violating the security boundary
+		my @manifests;
+		if (Bashbrew::RegistryUserAgent::is_media_image_list($mediaType)) {
+			push @manifests, @{ $manifest->{manifests} };
 		}
-		if ($mediaType eq Bashbrew::RegistryUserAgent::MEDIA_OCI_MANIFEST_V1 || $mediaType eq Bashbrew::RegistryUserAgent::MEDIA_MANIFEST_V1 || $mediaType eq Bashbrew::RegistryUserAgent::MEDIA_MANIFEST_V2) {
-			my $manifestListItem = {
+		elsif (Bashbrew::RegistryUserAgent::is_media_image_manifest($mediaType)) {
+			push @manifests, {
 				mediaType => $mediaType,
 				size => $size,
 				digest => $digest,
-				platform => {
-					arch_to_platform($arch),
-					($manifest->{'os.version'} ? ('os.version' => $manifest->{'os.version'}) : ()),
-				},
 			};
-			if ($manifestListItem->{platform}{os} eq 'windows' && !$manifestListItem->{platform}{'os.version'} && $mediaType eq Bashbrew::RegistryUserAgent::MEDIA_MANIFEST_V2) {
-				# if we're on Windows, we need to make an effort to fetch the "os.version" value from the config for the platform object
-				return $ua->get_blob_p($archRef->clone->digest($manifest->{config}{digest}))->then(sub ($config = undef) {
-					if ($config && $config->{'os.version'}) {
-						$manifestListItem->{platform}{'os.version'} = $config->{'os.version'};
-					}
-					return ($archRef, [ $manifestListItem ]);
-				});
+		}
+		else {
+			die "unknown mediaType '$mediaType' for '$archRef'";
+		}
+
+		# filter out objects we know we don't want (not a hashref, missing required fields)
+		@manifests = grep {
+			# https://specs.opencontainers.org/image-spec/descriptor/?v=v1.0.1
+			'HASH' eq ref($_) && $_->{mediaType} && $_->{digest} && $_->{size}
+		} @manifests;
+
+		# filter objects down to just fields we care about
+		@manifests = map {
+			# https://specs.opencontainers.org/image-spec/descriptor/?v=v1.0.1
+			{ %{ $_ }{qw{
+				mediaType
+				digest
+				size
+				annotations
+				platform
+			}} }
+		} @manifests;
+
+		my %platform = arch_to_platform($arch);
+
+		# normalize the result a bit (delete empty annotations and make sure platform is an object and that every platform has at least "os" and "architecture")
+		@manifests = map {
+			$_->{platform} //= {};
+			for my $key (qw( os architecture )) {
+				$_->{platform}{$key} //= $platform{$key};
 			}
-			else {
-				return ($archRef, [ $manifestListItem ]);
+			delete $_->{annotations} unless defined $_->{annotations};
+			$_
+		} @manifests;
+
+		# now that we have a list of potential manifests, let's filter it based on %platform's "os" and "architecture" (avoids "riscv64" from being able to poison us with anything other than riscv64-tagged manifests)
+		my @filteredManifests = grep {
+			Bashbrew::RegistryUserAgent::is_media_image_manifest($_->{mediaType})
+			&& $_->{platform}{os} eq $platform{os}
+			&& $_->{platform}{architecture} eq $platform{architecture}
+		} @manifests;
+		# normalize "platform" objects (esp. for "variant")
+		for my $item (@filteredManifests) {
+			for my $key (keys %platform) {
+				$item->{platform}{$key} = $platform{$key};
 			}
 		}
 
-		die "unknown mediaType '$mediaType' for '$archRef'";
+		# include any relevant "Docker-style" attachments (https://github.com/moby/buildkit/pull/2983, https://github.com/moby/buildkit/pull/3129, etc)
+		my %digests = map { $_ => 1 } map { $_->{digest} } @filteredManifests;
+		push @filteredManifests, grep {
+			$_->{mediaType} eq Bashbrew::RegistryUserAgent::MEDIA_OCI_MANIFEST_V1
+			&& $_->{platform}{os} eq 'unknown'
+			&& $_->{platform}{architecture} eq 'unknown'
+			&& $_->{annotations}
+			&& $digests{$_->{annotations}{'vnd.docker.reference.digest'} // ''}
+			&& $_->{annotations}{'vnd.docker.reference.type'}
+		} @manifests;
+
+		# if we're looking at Windows, we need to make an effort to fetch the "os.version" value from the config for the platform object
+		return Mojo::Promise->map({ concurrency => 3 }, sub ($item) {
+			unless (
+				Bashbrew::RegistryUserAgent::is_media_image_manifest($item->{mediaType})
+				&& $item->{platform}{os} eq 'windows'
+				&& !$item->{platform}{'os.version'}
+			) {
+				return Mojo::Promise->resolve($item);
+			}
+			return $ua->get_manifest_p($archRef->clone->digest($item->{digest}))->then(sub ($manifestData = undef) {
+				return $item unless $manifestData;
+				my $manifest = $manifestData->{manifest};
+				return $item unless $manifest->{config} and $manifest->{config}{digest};
+				return $ua->get_blob_p($archRef->clone->digest($manifest->{config}{digest}))->then(sub ($config = undef) {
+					if ($config && $config->{'os.version'}) {
+						$item->{platform}{'os.version'} = $config->{'os.version'};
+					}
+					return $item;
+				});
+			});
+		}, @filteredManifests)->then(sub (@manifests) {
+			@manifests = map { @$_ } @manifests;
+			return ($archRef, \@manifests);
+		});
 	});
 }
 

@@ -36,6 +36,13 @@ use constant MEDIA_FOREIGN_LAYER => 'application/vnd.docker.image.rootfs.foreign
 use constant MEDIA_OCI_INDEX_V1    => 'application/vnd.oci.image.index.v1+json';
 use constant MEDIA_OCI_MANIFEST_V1 => 'application/vnd.oci.image.manifest.v1+json';
 
+sub is_media_image_manifest ($mediaType) {
+	return $mediaType eq MEDIA_OCI_MANIFEST_V1 || $mediaType eq MEDIA_MANIFEST_V2 || $mediaType eq MEDIA_MANIFEST_V1;
+}
+sub is_media_image_list ($mediaType) {
+	return $mediaType eq MEDIA_OCI_INDEX_V1 || $mediaType eq MEDIA_MANIFEST_LIST;
+}
+
 # this is "normally" handled for us by https://github.com/tianon/dockerhub-public-proxy but is necessary for alternative registries
 my $acceptHeader = [
 	MEDIA_MANIFEST_LIST,

@@ -3,17 +3,49 @@ set -Eeuo pipefail
 
 # docker run -dit --name registry --restart always -p 5000:5000 registry
 
-arches=( amd64 arm32v5 arm32v6 arm32v7 arm64v8 i386 mips64le ppc64le riscv64 s390x )
-image='busybox:latest'
+# docker run -dit --name registry --restart always -p 5000:5000 --env REGISTRY_VALIDATION_MANIFESTS_URLS_ALLOW='["^.*$"]' --env REGISTRY_VALIDATION_MANIFESTS_URLS_DENY='[]' registry
+
+image="${1:-'hello-world:latest'}"
+registry='docker.io'
 target='localhost:5000'
 
+arches="$(bashbrew cat --format '{{- range .Entries }}{{ json .Architectures -}}{{- end -}}' "$image")"
+arches="$(jq <<<"$arches" -sr 'flatten | unique | map(@sh) | join(" ")')"
+eval "arches=( $arches )"
+
+if ! command -v crane &> /dev/null; then
+	if ! docker image inspect --format '.' gcr.io/go-containerregistry/crane &> /dev/null; then
+		docker pull gcr.io/go-containerregistry/crane
+	fi
+	crane() {
+		local args=(
+			--interactive --rm
+			--user "$RANDOM:$RANDOM"
+			--network host
+			--security-opt no-new-privileges
+		)
+		if [ -t 0 ] && [ -t 1 ]; then
+			args+=( --tty )
+		fi
+		docker run "${args[@]}" gcr.io/go-containerregistry/crane "$@"
+	}
+fi
+
 BASHBREW_ARCH_NAMESPACES=
 for arch in "${arches[@]}"; do
-	docker image inspect "$arch/$image" &> /dev/null || docker pull "$arch/$image"
-	docker tag "$arch/$image" "$target/$arch/$image"
-	docker push "$target/$arch/$image"
-	#skopeo copy --format oci "docker://docker.io/$arch/$image" --dest-tls-verify=false "docker://$target/$arch/$image"
-	#skopeo copy --format v2s2 "docker://docker.io/$arch/$image" --dest-tls-verify=false "docker://$target/$arch/$image"
+	if [ "$arch" = 'windows-amd64' ]; then
+		src="$registry/winamd64/$image"
+	else
+		src="$registry/$arch/$image"
+	fi
+	trg="$target/$arch/$image"
+
+	crane copy "$src" --insecure "$trg"
+
+	# skopeo appears to be the only one of these "registry copy" tools willing to do format conversions between Docker and OCI 👀
+	#skopeo copy --multi-arch all --format oci "docker://$src" --dest-tls-verify=false "docker://$trg"
+	#skopeo copy --multi-arch all --format v2s2 "docker://$src" --dest-tls-verify=false "docker://$trg"
+
 	[ -z "$BASHBREW_ARCH_NAMESPACES" ] || BASHBREW_ARCH_NAMESPACES+=', '
 	BASHBREW_ARCH_NAMESPACES+="$arch = $target/$arch"
 done