[release-5.0] OCPBUGS-127447: Nodes created with 4.13 base images cannot join the cluster

pkg/daemon/daemon.go

@@ -1232,7 +1232,10 @@ func (dn *Daemon) RunFirstbootCompleteMachineconfig(machineConfigFile string) er
 	// This currently will incur a double reboot; see https://github.com/coreos/rpm-ostree/issues/4018
 	// If skopeo is < 1.22.2 on a multi-arch image, run as a privileged container which has updated skopeo.
 	// See https://redhat.atlassian.net/browse/OCPBUGS-83826 and https://redhat.atlassian.net/browse/OCPBUGS-81187
-	if !newEnough || !skopeoSupportsMultiArchSigstore(mc.Spec.OSImageURL) {
+	// If rpm-ostree is < 2023.5, it has a skopeo-proxy sandboxing bug when rebasing
+	// from containers-storage or registry sources; run as a privileged container instead.
+	// See https://issues.redhat.com/browse/OCPBUGS-86768 (temporary until 4.13/4.14 boot images are unsupported).
+	if !newEnough || !skopeoSupportsMultiArchSigstore(mc.Spec.OSImageURL) || !dn.NodeUpdaterClient.SupportsContainerStorageRebase() {
 		logSystem("rpm-ostree or skopeo is not new enough for new-format image; forcing an update via container and queuing immediate reboot")
 		if err := dn.InplaceUpdateViaNewContainer(mc.Spec.OSImageURL); err != nil {
 			return err

pkg/daemon/rpm-ostree.go

@@ -7,8 +7,10 @@ import (
 	"path/filepath"
 	"reflect"
 	"strings"
+	"sync"
 
 	"github.com/containers/image/v5/signature"
+	"github.com/coreos/go-semver/semver"
 	rpmostreeclient "github.com/coreos/rpmostree-client-go/pkg/client"
 	"gopkg.in/yaml.v2"
 	"k8s.io/klog/v2"
@@ -19,6 +21,24 @@ const imagePolicyFilePath = "/etc/containers/policy.json"
 const rpmOstreeTemporalDropinFile = "/run/systemd/system/rpm-ostreed.service.d/temporal-policy-binding.conf"
 const rpmOstreeTemporalPolicyFile = "/run/tmp-rpm-ostree-policy.json"
 
+// minRpmOstreeVersionForContainerStorageRebase is the first upstream rpm-ostree
+// release that contains the fix for
+// https://github.com/coreos/rpm-ostree/issues/4283 (merged via
+// https://github.com/coreos/rpm-ostree/pull/4466), which resolved a skopeo-proxy
+// sandboxing bug affecting rebases from both containers-storage and registry
+// sources. Hosts running an older rpm-ostree hit the following issue:
+//
+// https://redhat.atlassian.net/browse/OCPBUGS-86768
+//
+// Expressed in dotted-tri form for comparison with go-semver, since rpm-ostree's
+// own version strings (e.g. "2023.3") only carry two components.
+const minRpmOstreeVersionForContainerStorageRebase = "2023.5.0"
+
+var (
+	rpmOstreeContainerStorageRebaseChecked   sync.Once
+	rpmOstreeContainerStorageRebaseSupported bool
+)
+
 // RpmOstreeClient provides all RpmOstree related methods in one structure.
 // This structure implements DeploymentClient
 //
@@ -192,6 +212,37 @@ func (r *RpmOstreeClient) IsNewEnoughForLayering() (bool, error) {
 	return false, nil
 }
 
+// normalizeCalVerForSemver adapts rpm-ostree's CalVer-style version strings
+// (e.g. "2023.3") to the dotted-tri format go-semver requires (e.g. "2023.3.0")
+// by appending a zero patch component when only two are present. Strings that
+// already have three (or more) dot-separated components are left untouched,
+// and any resulting parse failure is handled by the caller.
+func normalizeCalVerForSemver(version string) string {
+	if strings.Count(version, ".") == 1 {
+		return version + ".0"
+	}
+	return version
+}
+
+// NOTE: This is a temporary workaround for boot images predating rpm-ostree
+// v2023.5. Remove once 4.13 and 4.14 boot images are no longer supported.
+func (r *RpmOstreeClient) SupportsContainerStorageRebase() bool {
+	rpmOstreeContainerStorageRebaseChecked.Do(func() {
+		verdata, err := r.rpmOstreeVersion()
+		if err != nil {
+			klog.Errorf("failed to get rpm-ostree version: %v", err)
+			return
+		}
+		hostVersion, err := semver.NewVersion(normalizeCalVerForSemver(verdata.Version))
+		if err != nil {
+			klog.Errorf("failed to parse rpm-ostree version %q: %v", verdata.Version, err)
+			return
+		}
+		rpmOstreeContainerStorageRebaseSupported = hostVersion.Compare(*semver.New(minRpmOstreeVersionForContainerStorageRebase)) >= 0
+	})
+	return rpmOstreeContainerStorageRebaseSupported
+}
+
 // RebaseLayered rebases system or errors if already rebased.
 func (r *RpmOstreeClient) RebaseLayered(imgURL string) error {
 	// Try to re-link the merged pull secrets if they exist, since it could have been populated without a daemon reboot

pkg/daemon/rpm-ostree_test.go

@@ -1,6 +1,9 @@
 package daemon
 
 import (
+	"errors"
+	"fmt"
+	"sync"
 	"testing"
 
 	"github.com/stretchr/testify/assert"
@@ -25,3 +28,89 @@ func TestParseVersion(t *testing.T) {
 	assert.Contains(t, q.Root.Features, "rust")
 	assert.NotContains(t, q.Root.Features, "container")
 }
+
+// rpmOstreeVersionYAML builds a fake `rpm-ostree --version` YAML payload for
+// the given Version field, mirroring real-world output.
+func rpmOstreeVersionYAML(version string) []byte {
+	return []byte(fmt.Sprintf(`rpm-ostree:
+  Version: '%s'
+  Git: 6b302116c969397fd71899e3b9bb3b8c100d1af9
+  Features:
+   - container
+`, version))
+}
+
+func TestSupportsContainerStorageRebase(t *testing.T) {
+	tests := []struct {
+		name     string
+		output   []byte
+		cmdErr   error
+		expected bool
+	}{
+		{
+			// The exact version shipped in both the 4.13 and 4.14 RHCOS boot
+			// images (confirmed via their commitmeta.json rpmdb.pkglist:
+			// rpm-ostree-2023.3-1.el9_2 and rpm-ostree-2023.3-2.el9_2
+			// respectively), which reproduces OCPBUGS-86768.
+			name:     "4.13/4.14 shipped version is not new enough",
+			output:   rpmOstreeVersionYAML("2023.3"),
+			expected: false,
+		},
+		{
+			name:     "well below the fixed version",
+			output:   rpmOstreeVersionYAML("2022.10"),
+			expected: false,
+		},
+		{
+			name:     "exactly the fixed version",
+			output:   rpmOstreeVersionYAML("2023.5"),
+			expected: true,
+		},
+		{
+			name:     "above the fixed version",
+			output:   rpmOstreeVersionYAML("2023.6"),
+			expected: true,
+		},
+		{
+			name:     "unparseable version defaults to not supported",
+			output:   rpmOstreeVersionYAML("not-a-version"),
+			expected: false,
+		},
+		{
+			name:     "command execution failure defaults to not supported",
+			cmdErr:   errors.New("rpm-ostree: command not found"),
+			expected: false,
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			// SupportsContainerStorageRebase caches its result behind a
+			// package-level sync.Once (matching podmanSupportsSigstore and
+			// skopeoVersionSupportsMultiArchSigstore); reset both the guard
+			// and the cached value so each case is independently evaluated.
+			rpmOstreeContainerStorageRebaseChecked = sync.Once{}
+			rpmOstreeContainerStorageRebaseSupported = false
+
+			mock := &MockCommandRunner{
+				outputs: map[string][]byte{},
+				errors:  map[string]error{},
+			}
+			if tt.cmdErr != nil {
+				mock.errors["rpm-ostree --version"] = tt.cmdErr
+			} else {
+				mock.outputs["rpm-ostree --version"] = tt.output
+			}
+
+			r := &RpmOstreeClient{commandRunner: mock}
+			assert.Equal(t, tt.expected, r.SupportsContainerStorageRebase())
+		})
+	}
+}
+
+func TestNormalizeCalVerForSemver(t *testing.T) {
+	assert.Equal(t, "2023.3.0", normalizeCalVerForSemver("2023.3"))
+	assert.Equal(t, "2023.5.0", normalizeCalVerForSemver("2023.5"))
+	// Already dotted-tri (or with more components): left untouched.
+	assert.Equal(t, "2023.5.1", normalizeCalVerForSemver("2023.5.1"))
+}

pkg/daemon/update.go

@@ -2872,9 +2872,14 @@ func (dn *Daemon) updateLayeredOS(config *mcfgv1.MachineConfig) error {
 	}
 	// If the host isn't new enough to understand the new container model natively, run as a privileged container.
 	// See https://github.com/coreos/rpm-ostree/pull/3961 and https://issues.redhat.com/browse/MCO-356
+	//
 	// If skopeo is < 1.22.2 on a multi-arch image, run as a privileged container which has updated skopeo.
 	// See https://redhat.atlassian.net/browse/OCPBUGS-83826 and https://redhat.atlassian.net/browse/OCPBUGS-81187
-	if !newEnough || !skopeoSupportsMultiArchSigstore(newURL) {
+	//
+	// If rpm-ostree is < 2023.5, it has a skopeo-proxy sandboxing bug when rebasing
+	// from containers-storage or registry sources; run as a privileged container instead.
+	// See https://redhat.atlassian.net/browse/OCPBUGS-86768 (temporary until 4.13/4.14 boot images are unsupported).
+	if !newEnough || !skopeoSupportsMultiArchSigstore(newURL) || !dn.NodeUpdaterClient.SupportsContainerStorageRebase() {
 		logSystem("rpm-ostree or skopeo is not new enough for layering; forcing an update via container")
 		return dn.InplaceUpdateViaNewContainer(newURL)
 	}