...
Run Format

Source file src/syscall/exec_linux_test.go

Documentation: syscall

  // Copyright 2015 The Go Authors. All rights reserved.
  // Use of this source code is governed by a BSD-style
  // license that can be found in the LICENSE file.
  
  // +build linux
  
  package syscall_test
  
  import (
  	"flag"
  	"fmt"
  	"internal/testenv"
  	"io"
  	"io/ioutil"
  	"os"
  	"os/exec"
  	"os/user"
  	"path/filepath"
  	"strconv"
  	"strings"
  	"syscall"
  	"testing"
  	"unsafe"
  )
  
  // Check if we are in a chroot by checking if the inode of / is
  // different from 2 (there is no better test available to non-root on
  // linux).
  func isChrooted(t *testing.T) bool {
  	root, err := os.Stat("/")
  	if err != nil {
  		t.Fatalf("cannot stat /: %v", err)
  	}
  	return root.Sys().(*syscall.Stat_t).Ino != 2
  }
  
  func checkUserNS(t *testing.T) {
  	if _, err := os.Stat("/proc/self/ns/user"); err != nil {
  		if os.IsNotExist(err) {
  			t.Skip("kernel doesn't support user namespaces")
  		}
  		if os.IsPermission(err) {
  			t.Skip("unable to test user namespaces due to permissions")
  		}
  		t.Fatalf("Failed to stat /proc/self/ns/user: %v", err)
  	}
  	if isChrooted(t) {
  		// create_user_ns in the kernel (see
  		// https://git.kernel.org/cgit/linux/kernel/git/torvalds/linux.git/tree/kernel/user_namespace.c)
  		// forbids the creation of user namespaces when chrooted.
  		t.Skip("cannot create user namespaces when chrooted")
  	}
  	// On some systems, there is a sysctl setting.
  	if os.Getuid() != 0 {
  		data, errRead := ioutil.ReadFile("/proc/sys/kernel/unprivileged_userns_clone")
  		if errRead == nil && data[0] == '0' {
  			t.Skip("kernel prohibits user namespace in unprivileged process")
  		}
  	}
  	// On Centos 7 make sure they set the kernel parameter user_namespace=1
  	// See issue 16283 and 20796.
  	if _, err := os.Stat("/sys/module/user_namespace/parameters/enable"); err == nil {
  		buf, _ := ioutil.ReadFile("/sys/module/user_namespace/parameters/enabled")
  		if !strings.HasPrefix(string(buf), "Y") {
  			t.Skip("kernel doesn't support user namespaces")
  		}
  	}
  	// 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")
  	}
  }
  
  func whoamiCmd(t *testing.T, uid, gid int, setgroups bool) *exec.Cmd {
  	checkUserNS(t)
  	cmd := exec.Command("whoami")
  	cmd.SysProcAttr = &syscall.SysProcAttr{
  		Cloneflags: syscall.CLONE_NEWUSER,
  		UidMappings: []syscall.SysProcIDMap{
  			{ContainerID: 0, HostID: uid, Size: 1},
  		},
  		GidMappings: []syscall.SysProcIDMap{
  			{ContainerID: 0, HostID: gid, Size: 1},
  		},
  		GidMappingsEnableSetgroups: setgroups,
  	}
  	return cmd
  }
  
  func testNEWUSERRemap(t *testing.T, uid, gid int, setgroups bool) {
  	cmd := whoamiCmd(t, uid, gid, setgroups)
  	out, err := cmd.CombinedOutput()
  	if err != nil {
  		t.Fatalf("Cmd failed with err %v, output: %s", err, out)
  	}
  	sout := strings.TrimSpace(string(out))
  	want := "root"
  	if sout != want {
  		t.Fatalf("whoami = %q; want %q", out, want)
  	}
  }
  
  func TestCloneNEWUSERAndRemapRootDisableSetgroups(t *testing.T) {
  	if os.Getuid() != 0 {
  		t.Skip("skipping root only test")
  	}
  	testNEWUSERRemap(t, 0, 0, false)
  }
  
  func TestCloneNEWUSERAndRemapRootEnableSetgroups(t *testing.T) {
  	if os.Getuid() != 0 {
  		t.Skip("skipping root only test")
  	}
  	testNEWUSERRemap(t, 0, 0, false)
  }
  
  func TestCloneNEWUSERAndRemapNoRootDisableSetgroups(t *testing.T) {
  	if os.Getuid() == 0 {
  		t.Skip("skipping unprivileged user only test")
  	}
  	testNEWUSERRemap(t, os.Getuid(), os.Getgid(), false)
  }
  
  func TestCloneNEWUSERAndRemapNoRootSetgroupsEnableSetgroups(t *testing.T) {
  	if os.Getuid() == 0 {
  		t.Skip("skipping unprivileged user only test")
  	}
  	cmd := whoamiCmd(t, os.Getuid(), os.Getgid(), true)
  	err := cmd.Run()
  	if err == nil {
  		t.Skip("probably old kernel without security fix")
  	}
  	if !os.IsPermission(err) {
  		t.Fatalf("Unprivileged gid_map rewriting with GidMappingsEnableSetgroups must fail")
  	}
  }
  
  func TestEmptyCredGroupsDisableSetgroups(t *testing.T) {
  	cmd := whoamiCmd(t, os.Getuid(), os.Getgid(), false)
  	cmd.SysProcAttr.Credential = &syscall.Credential{}
  	if err := cmd.Run(); err != nil {
  		t.Fatal(err)
  	}
  }
  
  func TestUnshare(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")
  	}
  
  	path := "/proc/net/dev"
  	if _, err := os.Stat(path); err != nil {
  		if os.IsNotExist(err) {
  			t.Skip("kernel doesn't support proc filesystem")
  		}
  		if os.IsPermission(err) {
  			t.Skip("unable to test proc filesystem due to permissions")
  		}
  		t.Fatal(err)
  	}
  	if _, err := os.Stat("/proc/self/ns/net"); err != nil {
  		if os.IsNotExist(err) {
  			t.Skip("kernel doesn't support net namespace")
  		}
  		t.Fatal(err)
  	}
  
  	orig, err := ioutil.ReadFile(path)
  	if err != nil {
  		t.Fatal(err)
  	}
  	origLines := strings.Split(strings.TrimSpace(string(orig)), "\n")
  
  	cmd := exec.Command("cat", path)
  	cmd.SysProcAttr = &syscall.SysProcAttr{
  		Unshareflags: syscall.CLONE_NEWNET,
  	}
  	out, err := cmd.CombinedOutput()
  	if err != nil {
  		if strings.Contains(err.Error(), "operation not permitted") {
  			// Issue 17206: despite all the checks above,
  			// this still reportedly fails for some users.
  			// (older kernels?). Just skip.
  			t.Skip("skipping due to permission error")
  		}
  		t.Fatalf("Cmd failed with err %v, output: %s", err, out)
  	}
  
  	// Check there is only the local network interface
  	sout := strings.TrimSpace(string(out))
  	if !strings.Contains(sout, "lo:") {
  		t.Fatalf("Expected lo network interface to exist, got %s", sout)
  	}
  
  	lines := strings.Split(sout, "\n")
  	if len(lines) >= len(origLines) {
  		t.Fatalf("Got %d lines of output, want <%d", len(lines), len(origLines))
  	}
  }
  
  func TestGroupCleanup(t *testing.T) {
  	if os.Getuid() != 0 {
  		t.Skip("we need root for credential")
  	}
  	cmd := exec.Command("id")
  	cmd.SysProcAttr = &syscall.SysProcAttr{
  		Credential: &syscall.Credential{
  			Uid: 0,
  			Gid: 0,
  		},
  	}
  	out, err := cmd.CombinedOutput()
  	if err != nil {
  		t.Fatalf("Cmd failed with err %v, output: %s", err, out)
  	}
  	strOut := strings.TrimSpace(string(out))
  	expected := "uid=0(root) gid=0(root)"
  	// Just check prefix because some distros reportedly output a
  	// context parameter; see https://golang.org/issue/16224.
  	// Alpine does not output groups; see https://golang.org/issue/19938.
  	if !strings.HasPrefix(strOut, expected) {
  		t.Errorf("id command output: %q, expected prefix: %q", strOut, expected)
  	}
  }
  
  func TestGroupCleanupUserNamespace(t *testing.T) {
  	if os.Getuid() != 0 {
  		t.Skip("we need root for credential")
  	}
  	checkUserNS(t)
  	cmd := exec.Command("id")
  	uid, gid := os.Getuid(), os.Getgid()
  	cmd.SysProcAttr = &syscall.SysProcAttr{
  		Cloneflags: syscall.CLONE_NEWUSER,
  		Credential: &syscall.Credential{
  			Uid: uint32(uid),
  			Gid: uint32(gid),
  		},
  		UidMappings: []syscall.SysProcIDMap{
  			{ContainerID: 0, HostID: uid, Size: 1},
  		},
  		GidMappings: []syscall.SysProcIDMap{
  			{ContainerID: 0, HostID: gid, Size: 1},
  		},
  	}
  	out, err := cmd.CombinedOutput()
  	if err != nil {
  		t.Fatalf("Cmd failed with err %v, output: %s", err, out)
  	}
  	strOut := strings.TrimSpace(string(out))
  
  	// Strings we've seen in the wild.
  	expected := []string{
  		"uid=0(root) gid=0(root) groups=0(root)",
  		"uid=0(root) gid=0(root) groups=0(root),65534(nobody)",
  		"uid=0(root) gid=0(root) groups=0(root),65534(nogroup)",
  		"uid=0(root) gid=0(root) groups=0(root),65534",
  		"uid=0(root) gid=0(root) groups=0(root),65534(nobody),65534(nobody),65534(nobody),65534(nobody),65534(nobody),65534(nobody),65534(nobody),65534(nobody),65534(nobody),65534(nobody)", // Alpine; see https://golang.org/issue/19938
  	}
  	for _, e := range expected {
  		if strOut == e {
  			return
  		}
  	}
  	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 {
  		if strings.Contains(err.Error(), ": permission denied") {
  			t.Skipf("Skipping test (golang.org/issue/19698); unshare failed due to permissions: %s, %v", o, err)
  		}
  		t.Fatalf("unshare failed: %s, %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)
  		}
  	}
  }
  
  // Test for Issue 20103: unshare fails when chroot is used
  func TestUnshareMountNameSpaceChroot(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)
  	}
  
  	// Since we are doing a chroot, we need the binary there,
  	// and it must be statically linked.
  	x := filepath.Join(d, "syscall.test")
  	cmd := exec.Command(testenv.GoToolPath(t), "test", "-c", "-o", x, "syscall")
  	cmd.Env = append(os.Environ(), "CGO_ENABLED=0")
  	if o, err := cmd.CombinedOutput(); err != nil {
  		t.Fatalf("Build of syscall in chroot failed, output %v, err %v", o, err)
  	}
  
  	cmd = exec.Command("/syscall.test", "-test.run=TestUnshareMountNameSpaceHelper", "/")
  	cmd.Env = []string{"GO_WANT_HELPER_PROCESS=1"}
  	cmd.SysProcAttr = &syscall.SysProcAttr{Chroot: d, Unshareflags: syscall.CLONE_NEWNS}
  
  	o, err := cmd.CombinedOutput()
  	if err != nil {
  		if strings.Contains(err.Error(), ": permission denied") {
  			t.Skipf("Skipping test (golang.org/issue/19698); unshare failed due to permissions: %s, %v", o, err)
  		}
  		t.Fatalf("unshare failed: %s, %v", o, err)
  	}
  
  	// How do we tell if the namespace was really unshared? It turns out
  	// to be simple: just try to remove the executable. If it's still mounted
  	// on, the rm will fail. Then we have some cleanup to do:
  	// we must force unmount it, then try to remove it again.
  
  	if err := os.Remove(x); err != nil {
  		t.Errorf("rm failed on %v: %v", x, err)
  		if err := syscall.Unmount(d, syscall.MNT_FORCE); err != nil {
  			t.Fatalf("Can't unmount %v: %v", d, err)
  		}
  		if err := os.Remove(x); err != nil {
  			t.Fatalf("rm failed on %v: %v", x, err)
  		}
  	}
  
  	if err := os.Remove(d); err != nil {
  		t.Errorf("rmdir failed on %v: %v", d, err)
  	}
  }
  
  type capHeader struct {
  	version uint32
  	pid     int
  }
  
  type capData struct {
  	effective   uint32
  	permitted   uint32
  	inheritable uint32
  }
  
  const CAP_SYS_TIME = 25
  
  type caps struct {
  	hdr  capHeader
  	data [2]capData
  }
  
  func getCaps() (caps, error) {
  	var c caps
  
  	// Get capability version
  	if _, _, errno := syscall.Syscall(syscall.SYS_CAPGET, uintptr(unsafe.Pointer(&c.hdr)), uintptr(unsafe.Pointer(nil)), 0); errno != 0 {
  		return c, fmt.Errorf("SYS_CAPGET: %v", errno)
  	}
  
  	// Get current capabilities
  	if _, _, errno := syscall.Syscall(syscall.SYS_CAPGET, uintptr(unsafe.Pointer(&c.hdr)), uintptr(unsafe.Pointer(&c.data[0])), 0); errno != 0 {
  		return c, fmt.Errorf("SYS_CAPGET: %v", errno)
  	}
  
  	return c, nil
  }
  
  func mustSupportAmbientCaps(t *testing.T) {
  	var uname syscall.Utsname
  	if err := syscall.Uname(&uname); err != nil {
  		t.Fatalf("Uname: %v", err)
  	}
  	var buf [65]byte
  	for i, b := range uname.Release {
  		buf[i] = byte(b)
  	}
  	ver := string(buf[:])
  	if i := strings.Index(ver, "\x00"); i != -1 {
  		ver = ver[:i]
  	}
  	if strings.HasPrefix(ver, "2.") ||
  		strings.HasPrefix(ver, "3.") ||
  		strings.HasPrefix(ver, "4.1.") ||
  		strings.HasPrefix(ver, "4.2.") {
  		t.Skipf("kernel version %q predates required 4.3; skipping test", ver)
  	}
  }
  
  // TestAmbientCapsHelper isn't a real test. It's used as a helper process for
  // TestAmbientCaps.
  func TestAmbientCapsHelper(*testing.T) {
  	if os.Getenv("GO_WANT_HELPER_PROCESS") != "1" {
  		return
  	}
  	defer os.Exit(0)
  
  	caps, err := getCaps()
  	if err != nil {
  		fmt.Fprintln(os.Stderr, err)
  		os.Exit(2)
  	}
  	if caps.data[0].effective&(1<<uint(CAP_SYS_TIME)) == 0 {
  		fmt.Fprintln(os.Stderr, "CAP_SYS_TIME unexpectedly not in the effective capability mask")
  		os.Exit(2)
  	}
  }
  
  func TestAmbientCaps(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")
  	}
  	mustSupportAmbientCaps(t)
  
  	// 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")
  	}
  
  	caps, err := getCaps()
  	if err != nil {
  		t.Fatal(err)
  	}
  
  	// Add CAP_SYS_TIME to the permitted and inheritable capability mask,
  	// otherwise we will not be able to add it to the ambient capability mask.
  	caps.data[0].permitted |= 1 << uint(CAP_SYS_TIME)
  	caps.data[0].inheritable |= 1 << uint(CAP_SYS_TIME)
  
  	if _, _, errno := syscall.Syscall(syscall.SYS_CAPSET, uintptr(unsafe.Pointer(&caps.hdr)), uintptr(unsafe.Pointer(&caps.data[0])), 0); errno != 0 {
  		t.Fatalf("SYS_CAPSET: %v", errno)
  	}
  
  	u, err := user.Lookup("nobody")
  	if err != nil {
  		t.Fatal(err)
  	}
  	uid, err := strconv.ParseInt(u.Uid, 0, 32)
  	if err != nil {
  		t.Fatal(err)
  	}
  	gid, err := strconv.ParseInt(u.Gid, 0, 32)
  	if err != nil {
  		t.Fatal(err)
  	}
  
  	// Copy the test binary to a temporary location which is readable by nobody.
  	f, err := ioutil.TempFile("", "gotest")
  	if err != nil {
  		t.Fatal(err)
  	}
  	defer os.Remove(f.Name())
  	defer f.Close()
  	e, err := os.Open(os.Args[0])
  	if err != nil {
  		t.Fatal(err)
  	}
  	defer e.Close()
  	if _, err := io.Copy(f, e); err != nil {
  		t.Fatal(err)
  	}
  	if err := f.Chmod(0755); err != nil {
  		t.Fatal(err)
  	}
  	if err := f.Close(); err != nil {
  		t.Fatal(err)
  	}
  
  	cmd := exec.Command(f.Name(), "-test.run=TestAmbientCapsHelper")
  	cmd.Env = []string{"GO_WANT_HELPER_PROCESS=1"}
  	cmd.Stdout = os.Stdout
  	cmd.Stderr = os.Stderr
  	cmd.SysProcAttr = &syscall.SysProcAttr{
  		Credential: &syscall.Credential{
  			Uid: uint32(uid),
  			Gid: uint32(gid),
  		},
  		AmbientCaps: []uintptr{CAP_SYS_TIME},
  	}
  	if err := cmd.Run(); err != nil {
  		t.Fatal(err.Error())
  	}
  }
  

View as plain text