os/exec: handle Unshareflags with CLONE_NEWNS

In some newer Linux distros, systemd forces
all mount namespaces to be shared, starting
at /. This disables the CLONE_NEWNS
flag in unshare(2) and clone(2).
While this problem is most commonly seen
on systems with systemd, it can happen anywhere,
due to how Linux namespaces now work.

Hence, to create a private mount namespace,
it is not sufficient to just set
CLONE_NEWS; you have to call mount(2) to change
the behavior of namespaces, i.e.
mount("none", "/", NULL, MS_REC|MS_PRIVATE, NULL)

This is tested and working and we can now correctly
start child process with private namespaces on Linux
distros that use systemd.

The new test works correctly on Ubuntu 16.04.2 LTS.
It fails if I comment out the new Mount, and
succeeds otherwise. In each case it correctly
cleans up after itself.

Fixes #19661

Change-Id: I52240b59628e3772b529d9bbef7166606b0c157d
Reviewed-on: https://go-review.googlesource.com/38471
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

src/syscall/exec_linux.go

@@ -42,6 +42,11 @@ type SysProcAttr struct {
 	GidMappingsEnableSetgroups bool
 }
 
+var (
+	none  = [...]byte{'n', 'o', 'n', 'e', 0}
+	slash = [...]byte{'/', 0}
+)
+
 // Implemented in runtime package.
 func runtime_BeforeFork()
 func runtime_AfterFork()
@@ -204,6 +209,19 @@ func forkAndExecInChild(argv0 *byte, argv, envv []*byte, chroot, dir *byte, attr
 		if err1 != 0 {
 			goto childerror
 		}
+		// The unshare system call in Linux doesn't unshare mount points
+		// mounted with --shared. Systemd mounts / with --shared. For a
+		// long discussion of the pros and cons of this see debian bug 739593.
+		// The Go model of unsharing is more like Plan 9, where you ask
+		// to unshare and the namespaces are unconditionally unshared.
+		// To make this model work we must further mark / as MS_PRIVATE.
+		// This is what the standard unshare command does.
+		if sys.Unshareflags&CLONE_NEWNS == CLONE_NEWNS {
+			_, _, err1 = RawSyscall6(SYS_MOUNT, uintptr(unsafe.Pointer(&none[0])), uintptr(unsafe.Pointer(&slash[0])), 0, MS_REC|MS_PRIVATE, 0, 0)
+			if err1 != 0 {
+				goto childerror
+			}
+		}
 	}
 
 	// User and groups

src/syscall/exec_linux_test.go

@@ -7,6 +7,8 @@
 package syscall_test
 
 import (
+	"flag"
+	"fmt"
 	"io/ioutil"
 	"os"
 	"os/exec"
@@ -253,3 +255,63 @@ func TestGroupCleanupUserNamespace(t *testing.T) {
 	}
 	t.Errorf("id command output: %q, expected one of %q", strOut, expected)
 }
+
+// TestUnshareHelperProcess isn't a real test. It's used as a helper process
+// for TestUnshareMountNameSpace.
+func TestUnshareMountNameSpaceHelper(*testing.T) {
+	if os.Getenv("GO_WANT_HELPER_PROCESS") != "1" {
+		return
+	}
+	defer os.Exit(0)
+
+	if err := syscall.Mount("none", flag.Args()[0], "proc", 0, ""); err != nil {
+		fmt.Fprintf(os.Stderr, "unshare: mount %v failed: %v", os.Args, err)
+		os.Exit(2)
+	}
+}
+
+// Test for Issue 38471: unshare fails because systemd has forced / to be shared
+func TestUnshareMountNameSpace(t *testing.T) {
+	// Make sure we are running as root so we have permissions to use unshare
+	// and create a network namespace.
+	if os.Getuid() != 0 {
+		t.Skip("kernel prohibits unshare in unprivileged process, unless using user namespace")
+	}
+
+	// When running under the Go continuous build, skip tests for
+	// now when under Kubernetes. (where things are root but not quite)
+	// Both of these are our own environment variables.
+	// See Issue 12815.
+	if os.Getenv("GO_BUILDER_NAME") != "" && os.Getenv("IN_KUBERNETES") == "1" {
+		t.Skip("skipping test on Kubernetes-based builders; see Issue 12815")
+	}
+
+	d, err := ioutil.TempDir("", "unshare")
+	if err != nil {
+		t.Fatalf("tempdir: %v", err)
+	}
+
+	cmd := exec.Command(os.Args[0], "-test.run=TestUnshareMountNameSpaceHelper", d)
+	cmd.Env = []string{"GO_WANT_HELPER_PROCESS=1"}
+	cmd.SysProcAttr = &syscall.SysProcAttr{Unshareflags: syscall.CLONE_NEWNS}
+
+	o, err := cmd.CombinedOutput()
+	if err != nil {
+		t.Fatalf("unshare failed: %v, %v", o, err)
+	}
+
+	// How do we tell if the namespace was really unshared? It turns out
+	// to be simple: just try to remove the directory. If it's still mounted
+	// on the rm will fail with EBUSY. Then we have some cleanup to do:
+	// we must unmount it, then try to remove it again.
+
+	if err := os.Remove(d); err != nil {
+		t.Errorf("rmdir failed on %v: %v", d, err)
+		if err := syscall.Unmount(d, syscall.MNT_FORCE); err != nil {
+			t.Errorf("Can't unmount %v: %v", d, err)
+		}
+		if err := os.Remove(d); err != nil {
+			t.Errorf("rmdir after unmount failed on %v: %v", d, err)
+		}
+	}
+}