userns: use the intermediate mountns for volumes

when --uidmap is used, the user won't be able to access
/var/lib/containers/storage/volumes.  Use the intermediate mount
namespace, that is accessible to root in the container, for mounting
the volumes inside the container.

Closes: https://github.com/containers/libpod/issues/2713

Signed-off-by: Giuseppe Scrivano <gscrivan@redhat.com>

libpod/container_internal_linux.go

@@ -203,7 +203,8 @@ func (c *Container) generateSpec(ctx context.Context) (*spec.Spec, error) {
 	}
 	// Check if the spec file mounts contain the label Relabel flags z or Z.
 	// If they do, relabel the source directory and then remove the option.
-	for _, m := range g.Mounts() {
+	for i := range g.Config.Mounts {
+		m := &g.Config.Mounts[i]
 		var options []string
 		for _, o := range m.Options {
 			switch o {
@@ -219,6 +220,13 @@ func (c *Container) generateSpec(ctx context.Context) (*spec.Spec, error) {
 			}
 		}
 		m.Options = options
+
+		// If we are using a user namespace, we will use an intermediate
+		// directory to bind mount volumes
+		if c.state.UserNSRoot != "" && strings.HasPrefix(m.Source, c.runtime.config.VolumePath) {
+			newSourceDir := filepath.Join(c.state.UserNSRoot, "volumes")
+			m.Source = strings.Replace(m.Source, c.runtime.config.VolumePath, newSourceDir, 1)
+		}
 	}
 
 	g.SetProcessSelinuxLabel(c.ProcessLabel())

libpod/oci_linux.go

@@ -106,6 +106,23 @@ func (r *OCIRuntime) createContainer(ctr *Container, cgroupParent string, restor
 		if err != nil {
 			return
 		}
+
+		if ctr.state.UserNSRoot != "" {
+			_, err := os.Stat(ctr.runtime.config.VolumePath)
+			if err != nil && !os.IsNotExist(err) {
+				return
+			}
+			if err == nil {
+				volumesTarget := filepath.Join(ctr.state.UserNSRoot, "volumes")
+				if err := idtools.MkdirAs(volumesTarget, 0700, ctr.RootUID(), ctr.RootGID()); err != nil {
+					return
+				}
+				if err = unix.Mount(ctr.runtime.config.VolumePath, volumesTarget, "none", unix.MS_BIND, ""); err != nil {
+					return
+				}
+			}
+		}
+
 		err = r.createOCIContainer(ctr, cgroupParent, restoreOptions)
 	}()
 	wg.Wait()

test/e2e/run_userns_test.go

@@ -69,6 +69,21 @@ var _ = Describe("Podman UserNS support", func() {
 		Expect(ok).To(BeTrue())
 	})
 
+	It("podman uidmapping and gidmapping with a volume", func() {
+		if os.Getenv("SKIP_USERNS") != "" {
+			Skip("Skip userns tests.")
+		}
+		if _, err := os.Stat("/proc/self/uid_map"); err != nil {
+			Skip("User namespaces not supported.")
+		}
+
+		session := podmanTest.Podman([]string{"run", "--uidmap=0:1:70000", "--gidmap=0:20000:70000", "-v", "my-foo-volume:/foo:Z", "busybox", "echo", "hello"})
+		session.WaitWithDefaultTimeout()
+		Expect(session.ExitCode()).To(Equal(0))
+		ok, _ := session.GrepString("hello")
+		Expect(ok).To(BeTrue())
+	})
+
 	It("podman uidmapping and gidmapping --net=host", func() {
 		if os.Getenv("SKIP_USERNS") != "" {
 			Skip("Skip userns tests.")