internal/cri: simplify netns setup with pinned userns

Motivation:

For pod-level user namespaces, it's impossible to force the container runtime
to join an existing network namespace after creating a new user namespace.

According to the capabilities section in [user_namespaces(7)][1], a network
namespace created by containerd is owned by the root user namespace. When the
container runtime (like runc or crun) creates a new user namespace, it becomes
a child of the root user namespace. Processes within this child user namespace
are not permitted to access resources owned by the parent user namespace.

If the network namespace is not owned by the new user namespace, the container
runtime will fail to mount /sys due to the [sysfs: Restrict mounting sysfs][2]
patch.

Referencing the [cap_capable][3] function in Linux, a process can access a
resource if:

* The resource is owned by the process's user namespace, and the process has
the required capability.

* The resource is owned by a child of the process's user namespace, and the
owner's user namespace was created by the process's UID.

In the context of pod-level user namespaces, the CRI plugin delegates the
creation of the network namespace to the container runtime when running the
pause container. After the pause container is initialized, the CRI plugin pins
the pause container's network namespace into `/run/netns` and then executes
the `CNI_ADD` command over it.

However, if the pause container is terminated during the pinning process, the
CRI plugin might encounter a PID cycle, leading to the `CNI_ADD` command
operating on an incorrect network namespace.

Moreover, rolling back the `RunPodSandbox` API is complex due to the delegation
of network namespace creation. As highlighted in issue #10363, the CRI plugin
can lose IP information after a containerd restart, making it challenging to
maintain robustness in the RunPodSandbox API.

Solution:

Allow containerd to create a new user namespace and then create the network
namespace within that user namespace. This way, the CRI plugin can force the
container runtime to join both the user namespace and the network namespace.
Since the network namespace is owned by the newly created user namespace,
the container runtime will have the necessary permissions to mount `/sys` on
the container's root filesystem. As a result, delegation of network namespace
creation is no longer needed.

NOTE:

* The CRI plugin does not need to pin the newly created user namespace as it
does with the network namespace, because the kernel allows retrieving a user
namespace reference via [ioctl_ns(2)][4]. As a result, the podsandbox
implementation can obtain the user namespace using the `netnsPath` parameter.

[1]: <https://man7.org/linux/man-pages/man7/user_namespaces.7.html>
[2]: <7dc5dbc879>
[3]: <2c85ebc57b/security/commoncap.c (L65)>
[4]: <https://man7.org/linux/man-pages/man2/ioctl_ns.2.html>

Signed-off-by: Wei Fu <fuweid89@gmail.com>

internal/cri/server/podsandbox/helpers_linux.go

@@ -40,6 +40,7 @@ import (
 	"github.com/containerd/containerd/v2/core/snapshots"
 	"github.com/containerd/containerd/v2/internal/cri/seutil"
 	"github.com/containerd/containerd/v2/pkg/seccomp"
+	"github.com/containerd/containerd/v2/pkg/sys"
 )
 
 const (
@@ -88,6 +89,50 @@ func (c *Controller) getSandboxDevShm(id string) string {
 	return filepath.Join(c.getVolatileSandboxRootDir(id), "shm")
 }
 
+// getSandboxPinnedNamespaces returns the pinned namespaces directory inside the
+// sandbox state directory.
+func (c *Controller) getSandboxPinnedNamespaces(id string) string {
+	return filepath.Join(c.getVolatileSandboxRootDir(id), "pinned-namespaces")
+}
+
+// getSandboxPinnedUserNamespace returns the pinned user namespace file.
+func (c *Controller) getSandboxPinnedUserNamespace(id string) string {
+	return filepath.Join(c.getSandboxPinnedNamespaces(id), "user")
+}
+
+// pinUserNamespace persists user namespace in namespace filesystem.
+func (c *Controller) pinUserNamespace(sandboxID string, netnsPath string) error {
+	nsPath := c.getSandboxPinnedUserNamespace(sandboxID)
+
+	baseDir := filepath.Dir(nsPath)
+	if err := os.MkdirAll(baseDir, 0755); err != nil {
+		return fmt.Errorf("failed to init pinned-namespaces directory %s: %w", baseDir, err)
+	}
+
+	emptyFd, err := os.OpenFile(nsPath, os.O_RDWR|os.O_CREATE|os.O_EXCL, 0666)
+	if err != nil {
+		return fmt.Errorf("failed to create empty file %s: %w", nsPath, err)
+	}
+	emptyFd.Close()
+
+	netnsFd, err := os.Open(netnsPath)
+	if err != nil {
+		return fmt.Errorf("failed to open netns(%s): %w", netnsPath, err)
+	}
+	defer netnsFd.Close()
+
+	usernsFd, err := sys.GetUsernsForNamespace(netnsFd.Fd())
+	if err != nil {
+		return fmt.Errorf("failed to get user namespace for netns(%s): %w", netnsPath, err)
+	}
+	defer usernsFd.Close()
+
+	if err = unix.Mount(usernsFd.Name(), nsPath, "none", unix.MS_BIND, ""); err != nil {
+		return fmt.Errorf("failed to bind mount ns src: %v at %s: %w", usernsFd.Name(), nsPath, err)
+	}
+	return nil
+}
+
 func toLabel(selinuxOptions *runtime.SELinuxOption) ([]string, error) {
 	var labels []string
 

internal/cri/server/podsandbox/sandbox_run.go

@@ -95,6 +95,39 @@ func (c *Controller) Start(ctx context.Context, id string) (cin sandbox.Controll
 
 	labels["oci_runtime_type"] = ociRuntime.Type
 
+	// Create sandbox container root directories.
+	sandboxRootDir := c.getSandboxRootDir(id)
+	if err := c.os.MkdirAll(sandboxRootDir, 0755); err != nil {
+		return cin, fmt.Errorf("failed to create sandbox root directory %q: %w",
+			sandboxRootDir, err)
+	}
+	defer func() {
+		if retErr != nil && cleanupErr == nil {
+			// Cleanup the sandbox root directory.
+			if cleanupErr = c.os.RemoveAll(sandboxRootDir); cleanupErr != nil {
+				log.G(ctx).WithError(cleanupErr).Errorf("Failed to remove sandbox root directory %q",
+					sandboxRootDir)
+			}
+		}
+	}()
+
+	volatileSandboxRootDir := c.getVolatileSandboxRootDir(id)
+	if err := c.os.MkdirAll(volatileSandboxRootDir, 0755); err != nil {
+		return cin, fmt.Errorf("failed to create volatile sandbox root directory %q: %w",
+			volatileSandboxRootDir, err)
+	}
+	defer func() {
+		if retErr != nil && cleanupErr == nil {
+			deferCtx, deferCancel := ctrdutil.DeferContext()
+			defer deferCancel()
+			// Cleanup the volatile sandbox root directory.
+			if cleanupErr = ensureRemoveAll(deferCtx, volatileSandboxRootDir); cleanupErr != nil {
+				log.G(ctx).WithError(cleanupErr).Errorf("Failed to remove volatile sandbox root directory %q",
+					volatileSandboxRootDir)
+			}
+		}
+	}()
+
 	// Create sandbox container.
 	// NOTE: sandboxContainerSpec SHOULD NOT have side
 	// effect, e.g. accessing/creating files, so that we can test
@@ -164,37 +197,6 @@ func (c *Controller) Start(ctx context.Context, id string) (cin sandbox.Controll
 		}
 	}()
 
-	// Create sandbox container root directories.
-	sandboxRootDir := c.getSandboxRootDir(id)
-	if err := c.os.MkdirAll(sandboxRootDir, 0755); err != nil {
-		return cin, fmt.Errorf("failed to create sandbox root directory %q: %w",
-			sandboxRootDir, err)
-	}
-	defer func() {
-		if retErr != nil && cleanupErr == nil {
-			// Cleanup the sandbox root directory.
-			if cleanupErr = c.os.RemoveAll(sandboxRootDir); cleanupErr != nil {
-				log.G(ctx).WithError(cleanupErr).Errorf("Failed to remove sandbox root directory %q",
-					sandboxRootDir)
-			}
-		}
-	}()
-
-	volatileSandboxRootDir := c.getVolatileSandboxRootDir(id)
-	if err := c.os.MkdirAll(volatileSandboxRootDir, 0755); err != nil {
-		return cin, fmt.Errorf("failed to create volatile sandbox root directory %q: %w",
-			volatileSandboxRootDir, err)
-	}
-	defer func() {
-		if retErr != nil && cleanupErr == nil {
-			// Cleanup the volatile sandbox root directory.
-			if cleanupErr = c.os.RemoveAll(volatileSandboxRootDir); cleanupErr != nil {
-				log.G(ctx).WithError(cleanupErr).Errorf("Failed to remove volatile sandbox root directory %q",
-					volatileSandboxRootDir)
-			}
-		}
-	}()
-
 	// Setup files required for the sandbox.
 	if err = c.setupSandboxFiles(id, config); err != nil {
 		return cin, fmt.Errorf("failed to setup sandbox files: %w", err)

internal/cri/server/podsandbox/sandbox_run_linux.go

@@ -103,6 +103,11 @@ func (c *Controller) sandboxContainerSpec(id string, config *runtime.PodSandboxC
 		case runtime.NamespaceMode_POD:
 			specOpts = append(specOpts, oci.WithUserNamespace(uids, gids))
 			usernsEnabled = true
+
+			if err := c.pinUserNamespace(id, nsPath); err != nil {
+				return nil, fmt.Errorf("failed to pin user namespace: %w", err)
+			}
+			specOpts = append(specOpts, customopts.WithNamespacePath(runtimespec.UserNamespace, c.getSandboxPinnedUserNamespace(id)))
 		default:
 			return nil, fmt.Errorf("unsupported user namespace mode: %q", mode)
 		}

internal/cri/server/podsandbox/sandbox_run_linux_test.go

@@ -17,9 +17,11 @@
 package podsandbox
 
 import (
+	"context"
 	"os"
 	"path/filepath"
 	"strconv"
+	"syscall"
 	"testing"
 
 	"github.com/moby/sys/userns"
@@ -32,11 +34,15 @@ import (
 	v1 "k8s.io/cri-api/pkg/apis/runtime/v1"
 
 	"github.com/containerd/containerd/v2/internal/cri/annotations"
+	criconfig "github.com/containerd/containerd/v2/internal/cri/config"
 	"github.com/containerd/containerd/v2/internal/cri/opts"
+	"github.com/containerd/containerd/v2/pkg/netns"
 	ostesting "github.com/containerd/containerd/v2/pkg/os/testing"
+	"github.com/containerd/containerd/v2/pkg/sys"
+	"github.com/containerd/containerd/v2/pkg/testutil"
 )
 
-func getRunPodSandboxTestData() (*runtime.PodSandboxConfig, *imagespec.ImageConfig, func(*testing.T, string, *runtimespec.Spec)) {
+func getRunPodSandboxTestData(criCfg criconfig.Config) (*runtime.PodSandboxConfig, *imagespec.ImageConfig, func(*testing.T, string, *runtimespec.Spec)) {
 	config := &runtime.PodSandboxConfig{
 		Metadata: &runtime.PodSandboxMetadata{
 			Name:      "test-name",
@@ -94,7 +100,7 @@ func getRunPodSandboxTestData() (*runtime.PodSandboxConfig, *imagespec.ImageConf
 		}
 
 		assert.Contains(t, spec.Mounts, runtimespec.Mount{
-			Source:      "/test/root/sandboxes/test-id/resolv.conf",
+			Source:      filepath.Join(criCfg.RootDir, "sandboxes/test-id/resolv.conf"),
 			Destination: resolvConfPath,
 			Type:        "bind",
 			Options:     []string{"rbind", "ro", "nosuid", "nodev", "noexec"},
@@ -105,8 +111,10 @@ func getRunPodSandboxTestData() (*runtime.PodSandboxConfig, *imagespec.ImageConf
 }
 
 func TestLinuxSandboxContainerSpec(t *testing.T) {
+	testutil.RequiresRoot(t)
+
 	testID := "test-id"
-	nsPath := "test-cni"
+
 	idMap := runtime.IDMapping{
 		HostId:      1000,
 		ContainerId: 1000,
@@ -118,15 +126,30 @@ func TestLinuxSandboxContainerSpec(t *testing.T) {
 		Size:        10,
 	}
 
+	netnsBasedir := t.TempDir()
+	t.Cleanup(func() {
+		assert.NoError(t, unmountRecursive(context.Background(), netnsBasedir))
+	})
+
+	var netNs *netns.NetNS
+	uerr := sys.UnshareAfterEnterUserns("1000:1000:10", "1000:1000:10", syscall.CLONE_NEWNET, func(pid int) error {
+		var err error
+		netNs, err = netns.NewNetNSFromPID(netnsBasedir, uint32(pid))
+		return err
+	})
+	require.NoError(t, uerr)
+
+	nsPath := netNs.GetPath()
+
 	for _, test := range []struct {
 		desc         string
 		configChange func(*runtime.PodSandboxConfig)
-		specCheck    func(*testing.T, *runtimespec.Spec)
+		specCheck    func(*testing.T, *Controller, *runtimespec.Spec)
 		expectErr    bool
 	}{
 		{
 			desc: "spec should reflect original config",
-			specCheck: func(t *testing.T, spec *runtimespec.Spec) {
+			specCheck: func(t *testing.T, _ *Controller, spec *runtimespec.Spec) {
 				// runtime spec should have expected namespaces enabled by default.
 				require.NotNil(t, spec.Linux)
 				assert.Contains(t, spec.Linux.Namespaces, runtimespec.LinuxNamespace{
@@ -162,10 +185,11 @@ func TestLinuxSandboxContainerSpec(t *testing.T) {
 					},
 				}
 			},
-			specCheck: func(t *testing.T, spec *runtimespec.Spec) {
+			specCheck: func(t *testing.T, c *Controller, spec *runtimespec.Spec) {
 				require.NotNil(t, spec.Linux)
 				assert.Contains(t, spec.Linux.Namespaces, runtimespec.LinuxNamespace{
 					Type: runtimespec.UserNamespace,
+					Path: filepath.Join(c.config.StateDir, "sandboxes", testID, "pinned-namespaces", "user"),
 				})
 				assert.NotContains(t, spec.Linux.Sysctl["net.ipv4.ping_group_range"], "0 2147483647")
 			},
@@ -181,7 +205,7 @@ func TestLinuxSandboxContainerSpec(t *testing.T) {
 					},
 				}
 			},
-			specCheck: func(t *testing.T, spec *runtimespec.Spec) {
+			specCheck: func(t *testing.T, _ *Controller, spec *runtimespec.Spec) {
 				// runtime spec should disable expected namespaces in host mode.
 				require.NotNil(t, spec.Linux)
 				assert.NotContains(t, spec.Linux.Namespaces, runtimespec.LinuxNamespace{
@@ -213,10 +237,11 @@ func TestLinuxSandboxContainerSpec(t *testing.T) {
 					},
 				}
 			},
-			specCheck: func(t *testing.T, spec *runtimespec.Spec) {
+			specCheck: func(t *testing.T, c *Controller, spec *runtimespec.Spec) {
 				require.NotNil(t, spec.Linux)
 				assert.Contains(t, spec.Linux.Namespaces, runtimespec.LinuxNamespace{
 					Type: runtimespec.UserNamespace,
+					Path: filepath.Join(c.config.StateDir, "sandboxes", testID, "pinned-namespaces", "user"),
 				})
 				require.Equal(t, spec.Linux.UIDMappings, []runtimespec.LinuxIDMapping{expIDMap})
 				require.Equal(t, spec.Linux.GIDMappings, []runtimespec.LinuxIDMapping{expIDMap})
@@ -314,7 +339,7 @@ func TestLinuxSandboxContainerSpec(t *testing.T) {
 					SupplementalGroups: []int64{1111, 2222},
 				}
 			},
-			specCheck: func(t *testing.T, spec *runtimespec.Spec) {
+			specCheck: func(t *testing.T, _ *Controller, spec *runtimespec.Spec) {
 				require.NotNil(t, spec.Process)
 				assert.Contains(t, spec.Process.User.AdditionalGids, uint32(1111))
 				assert.Contains(t, spec.Process.User.AdditionalGids, uint32(2222))
@@ -328,7 +353,7 @@ func TestLinuxSandboxContainerSpec(t *testing.T) {
 					"net.ipv4.ping_group_range":           "1 1000",
 				}
 			},
-			specCheck: func(t *testing.T, spec *runtimespec.Spec) {
+			specCheck: func(t *testing.T, _ *Controller, spec *runtimespec.Spec) {
 				require.NotNil(t, spec.Process)
 				assert.Contains(t, spec.Linux.Sysctl["net.ipv4.ip_unprivileged_port_start"], "500")
 				assert.Contains(t, spec.Linux.Sysctl["net.ipv4.ping_group_range"], "1 1000")
@@ -344,7 +369,7 @@ func TestLinuxSandboxContainerSpec(t *testing.T) {
 					MemoryLimitInBytes: 1024,
 				}
 			},
-			specCheck: func(t *testing.T, spec *runtimespec.Spec) {
+			specCheck: func(t *testing.T, _ *Controller, spec *runtimespec.Spec) {
 				value, ok := spec.Annotations[annotations.SandboxCPUPeriod]
 				assert.True(t, ok)
 				assert.EqualValues(t, strconv.FormatInt(100, 10), value)
@@ -365,7 +390,7 @@ func TestLinuxSandboxContainerSpec(t *testing.T) {
 		},
 		{
 			desc: "sandbox sizing annotations should not be set if LinuxContainerResources were not provided",
-			specCheck: func(t *testing.T, spec *runtimespec.Spec) {
+			specCheck: func(t *testing.T, _ *Controller, spec *runtimespec.Spec) {
 				_, ok := spec.Annotations[annotations.SandboxCPUPeriod]
 				assert.False(t, ok)
 				_, ok = spec.Annotations[annotations.SandboxCPUQuota]
@@ -381,7 +406,7 @@ func TestLinuxSandboxContainerSpec(t *testing.T) {
 			configChange: func(c *runtime.PodSandboxConfig) {
 				c.Linux.Resources = &v1.LinuxContainerResources{}
 			},
-			specCheck: func(t *testing.T, spec *runtimespec.Spec) {
+			specCheck: func(t *testing.T, _ *Controller, spec *runtimespec.Spec) {
 				value, ok := spec.Annotations[annotations.SandboxCPUPeriod]
 				assert.True(t, ok)
 				assert.EqualValues(t, "0", value)
@@ -400,9 +425,17 @@ func TestLinuxSandboxContainerSpec(t *testing.T) {
 		test := test
 		t.Run(test.desc, func(t *testing.T) {
 			c := newControllerService()
+			c.config.RootDir = t.TempDir()
+			c.config.StateDir = t.TempDir()
+
+			defer func() {
+				assert.NoError(t, unmountRecursive(context.Background(), c.config.StateDir))
+			}()
+
 			c.config.EnableUnprivilegedICMP = true
 			c.config.EnableUnprivilegedPorts = true
-			config, imageConfig, specCheck := getRunPodSandboxTestData()
+
+			config, imageConfig, specCheck := getRunPodSandboxTestData(c.config)
 			if test.configChange != nil {
 				test.configChange(config)
 			}
@@ -416,7 +449,7 @@ func TestLinuxSandboxContainerSpec(t *testing.T) {
 			assert.NotNil(t, spec)
 			specCheck(t, testID, spec)
 			if test.specCheck != nil {
-				test.specCheck(t, spec)
+				test.specCheck(t, c, spec)
 			}
 		})
 	}
@@ -757,6 +790,3 @@ options timeout:1
 		})
 	}
 }
-
-// TODO(random-liu): [P1] Add unit test for different error cases to make sure
-// the function cleans up on error properly.

internal/cri/server/podsandbox/sandbox_run_other_test.go

@@ -21,12 +21,13 @@ package podsandbox
 import (
 	"testing"
 
+	criconfig "github.com/containerd/containerd/v2/internal/cri/config"
 	imagespec "github.com/opencontainers/image-spec/specs-go/v1"
 	runtimespec "github.com/opencontainers/runtime-spec/specs-go"
 	runtime "k8s.io/cri-api/pkg/apis/runtime/v1"
 )
 
-func getRunPodSandboxTestData() (*runtime.PodSandboxConfig, *imagespec.ImageConfig, func(*testing.T, string, *runtimespec.Spec)) {
+func getRunPodSandboxTestData(_ criconfig.Config) (*runtime.PodSandboxConfig, *imagespec.ImageConfig, func(*testing.T, string, *runtimespec.Spec)) {
 	config := &runtime.PodSandboxConfig{}
 	imageConfig := &imagespec.ImageConfig{}
 	specCheck := func(t *testing.T, id string, spec *runtimespec.Spec) {

internal/cri/server/podsandbox/sandbox_run_test.go

@@ -27,8 +27,14 @@ import (
 	runtime "k8s.io/cri-api/pkg/apis/runtime/v1"
 
 	sandboxstore "github.com/containerd/containerd/v2/internal/cri/store/sandbox"
+	"github.com/containerd/containerd/v2/pkg/testutil"
 )
 
+func TestEmpty(t *testing.T) {
+	// NOTE: It's used to register -test.root for all platforms.
+	testutil.RequiresRoot(t)
+}
+
 func TestSandboxContainerSpec(t *testing.T) {
 	switch goruntime.GOOS {
 	case "darwin":
@@ -97,7 +103,7 @@ func TestSandboxContainerSpec(t *testing.T) {
 		test := test
 		t.Run(test.desc, func(t *testing.T) {
 			c := newControllerService()
-			config, imageConfig, specCheck := getRunPodSandboxTestData()
+			config, imageConfig, specCheck := getRunPodSandboxTestData(c.config)
 			if test.configChange != nil {
 				test.configChange(config)
 			}
@@ -154,7 +160,9 @@ func TestTypeurlMarshalUnmarshalSandboxMeta(t *testing.T) {
 				Name:      "sandbox_1",
 				NetNSPath: "/home/cloud",
 			}
-			meta.Config, _, _ = getRunPodSandboxTestData()
+
+			c := newControllerService()
+			meta.Config, _, _ = getRunPodSandboxTestData(c.config)
 			if test.configChange != nil {
 				test.configChange(meta.Config)
 			}

internal/cri/server/podsandbox/sandbox_run_windows_test.go

@@ -25,10 +25,11 @@ import (
 	runtime "k8s.io/cri-api/pkg/apis/runtime/v1"
 
 	"github.com/containerd/containerd/v2/internal/cri/annotations"
+	criconfig "github.com/containerd/containerd/v2/internal/cri/config"
 	"github.com/containerd/containerd/v2/internal/cri/opts"
 )
 
-func getRunPodSandboxTestData() (*runtime.PodSandboxConfig, *imagespec.ImageConfig, func(*testing.T, string, *runtimespec.Spec)) {
+func getRunPodSandboxTestData(criCfg criconfig.Config) (*runtime.PodSandboxConfig, *imagespec.ImageConfig, func(*testing.T, string, *runtimespec.Spec)) {
 	config := &runtime.PodSandboxConfig{
 		Metadata: &runtime.PodSandboxMetadata{
 			Name:      "test-name",
@@ -100,7 +101,7 @@ func TestSandboxWindowsNetworkNamespace(t *testing.T) {
 	nsPath := "test-cni"
 	c := newControllerService()
 
-	config, imageConfig, specCheck := getRunPodSandboxTestData()
+	config, imageConfig, specCheck := getRunPodSandboxTestData(c.config)
 	spec, err := c.sandboxContainerSpec(testID, config, imageConfig, nsPath, nil)
 	assert.NoError(t, err)
 	assert.NotNil(t, spec)