Let a group share the info API too, and serve it on windows

The info API is for other programs on the host, and on unix the only
setting between "the user nebula runs as" and "everyone" was a manual
chgrp that the next bind undid; on windows it was not served at all,
because which accounts a pipe admits had not been decided. The control
socket's `group` decides it, and the info API takes the same key.

- `local_api.info.group`: the socket is handed to the group at 0660
  unless `socket_mode` says otherwise, and a mode that gives a named
  group nothing is refused with the mode that would
- windows: served on a named pipe, the group an ACE on it as on the
  control pipe; the e2e test's helpers dial unix sockets, so it still
  skips there
- a changed group or mode on a reload is applied to the socket in place

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>

localapi/access_unix_test.go

@@ -4,6 +4,7 @@ package localapi
 
 import (
 	"context"
+	"net/http"
 	"os"
 	"os/user"
 	"strconv"
@@ -142,6 +143,66 @@ func TestControlSocketWithAGroupTrustsTheKernel(t *testing.T) {
 	require.ErrorContains(t, err, "not permitted")
 }
 
+// TestInfoAccessFromConfig: a group with no mode is 0660; a mode that gives the group
+// nothing is refused, with the mode that would; an explicit mode with group bits is
+// kept as given.
+func TestInfoAccessFromConfig(t *testing.T) {
+	name, gid := ownGroup(t)
+	load := func(body string) (infoConfig, error) {
+		c := config.NewC(test.NewLogger())
+		require.NoError(t, c.LoadString(body))
+		return infoConfigFrom(c)
+	}
+	section := func(extra string) string {
+		return "local_api:\n  info:\n    enabled: true\n    socket: /run/nebula/info.sock\n" + extra
+	}
+
+	cfg, err := load(section("    group: " + name + "\n"))
+	require.NoError(t, err)
+	assert.Equal(t, os.FileMode(0660), cfg.Access.Mode)
+	require.NotNil(t, cfg.Access.Group)
+	assert.Equal(t, gid, cfg.Access.Group.ID)
+
+	cfg, err = load(section("    group: " + name + "\n    socket_mode: \"0664\"\n"))
+	require.NoError(t, err)
+	assert.Equal(t, os.FileMode(0664), cfg.Access.Mode)
+
+	_, err = load(section("    group: " + name + "\n    socket_mode: \"0600\"\n"))
+	require.ErrorContains(t, err, "gives group "+name+" no access")
+	require.ErrorContains(t, err, `"0660"`)
+
+	cfg, err = load(section(""))
+	require.NoError(t, err)
+	assert.Equal(t, os.FileMode(0600), cfg.Access.Mode)
+	assert.Nil(t, cfg.Access.Group)
+}
+
+// TestSocketIsHandedToTheGroup: the socket is created with the group's gid and the
+// mode, and a changed group or mode on a reload is applied to the socket in place.
+func TestSocketIsHandedToTheGroup(t *testing.T) {
+	name, gid := ownGroup(t)
+	group := &Group{Name: name, ID: gid}
+	want, err := strconv.ParseUint(gid, 10, 32)
+	require.NoError(t, err)
+
+	s, path := serveInfo(t, infoRegistry(), 0600)
+	require.NoError(t, s.Rebind(context.Background(), path, Access{Mode: 0660, Group: group}))
+	fi, err := os.Lstat(path)
+	require.NoError(t, err)
+	assert.Equal(t, os.FileMode(0660), fi.Mode().Perm())
+	assert.Equal(t, uint32(want), socketGroup(t, path))
+
+	status, _ := test.HTTPGet(t, path, "/v1/whoami")
+	assert.Equal(t, http.StatusOK, status, "the same listener is still serving")
+
+	// The mode is the operator's to widen further.
+	require.NoError(t, s.Rebind(context.Background(), path, Access{Mode: 0664, Group: group}))
+	fi, err = os.Lstat(path)
+	require.NoError(t, err)
+	assert.Equal(t, os.FileMode(0664), fi.Mode().Perm())
+	assert.Equal(t, path, s.Path())
+}
+
 // TestRemovingTheGroupHandsTheSocketBack: a reload that drops the group key leaves the
 // mode the operator set but returns the socket to the daemon's own group, so the old
 // group's members are no longer admitted. The socket is first handed to a group this

localapi/info.go

@@ -14,6 +14,7 @@ import (
 	"net/url"
 	"os"
 	"path"
+	"runtime"
 	"runtime/debug"
 	"slices"
 	"strconv"
@@ -31,12 +32,13 @@ const (
 	InfoEnabledKey = "local_api.info.enabled"
 	InfoSocketKey  = "local_api.info.socket"
 	InfoModeKey    = "local_api.info.socket_mode"
+	InfoGroupKey   = "local_api.info.group"
 )
 
 const (
-	// defaultInfoMode is the socket's mode when the config names none: the daemon's
-	// user alone, so that enabling the API does not open it to anyone until the operator
-	// says who.
+	// defaultInfoMode is the socket's mode when the config names neither a mode nor a
+	// group: the daemon's user alone, so that enabling the API does not open it to
+	// anyone until the operator says who. Naming a group makes it groupSocketMode.
 	defaultInfoMode os.FileMode = 0600
 
 	// infoMaxConns bounds the connections held open at once. Beyond it the kernel
@@ -394,7 +396,10 @@ type infoConfig struct {
 }
 
 // infoConfigFrom reads the info API section. A disabled section needs nothing else to be
-// right; an enabled one needs a socket path and, if it names a mode, one that parses.
+// right; an enabled one needs a socket path, a group that exists if it names one, and
+// a mode that parses if it names one. A group with no mode means groupSocketMode, and a
+// mode that gives a named group nothing is refused: it would be a group that cannot
+// connect, which is not what naming one asks for.
 func infoConfigFrom(c *config.C) (infoConfig, error) {
 	cfg := infoConfig{Enabled: c.GetBool(InfoEnabledKey, false)}
 	if !cfg.Enabled {
@@ -410,11 +415,26 @@ func infoConfigFrom(c *config.C) (infoConfig, error) {
 	}
 	cfg.Path = p
 
-	mode, err := parseSocketMode(c.Get(InfoModeKey))
+	group, err := groupFromConfig(c, InfoGroupKey)
+	if err != nil {
+		return cfg, err
+	}
+	def := defaultInfoMode
+	if group != nil {
+		def = groupSocketMode
+	}
+
+	mode, err := parseSocketMode(c.Get(InfoModeKey), def)
 	if err != nil {
 		return cfg, fmt.Errorf("%s: %w", InfoModeKey, err)
 	}
-	cfg.Access = Access{Mode: mode}
+	// On unix the mode is what admits the group, so one that shuts it out is a config
+	// error; on windows the group is an entry in the pipe's security descriptor and the
+	// mode means nothing to a pipe, so it is not held against the group there.
+	if runtime.GOOS != "windows" && group != nil && mode&0070 == 0 {
+		return cfg, fmt.Errorf("%s %s gives group %s no access to the socket; use \"%04o\" or leave the mode unset", InfoModeKey, modeString(mode), group.Name, mode|0060)
+	}
+	cfg.Access = Access{Mode: mode, Group: group}
 
 	return cfg, nil
 }
@@ -423,10 +443,10 @@ func infoConfigFrom(c *config.C) (infoConfig, error) {
 // integer, and which integer depends on the parser (octal 432 under YAML 1.1, decimal
 // 660 under 1.2), so a number is refused with the fix rather than quietly applied as
 // whichever mode its digits happen to spell in octal.
-func parseSocketMode(raw any) (os.FileMode, error) {
+func parseSocketMode(raw any, def os.FileMode) (os.FileMode, error) {
 	switch v := raw.(type) {
 	case nil:
-		return defaultInfoMode, nil
+		return def, nil
 	case string:
 		mode, err := strconv.ParseUint(v, 8, 32)
 		if err != nil || mode&^0777 != 0 {

localapi/info_windows_test.go

@@ -4,20 +4,55 @@ package localapi
 
 import (
 	"context"
+	"io"
+	"net"
+	"net/http"
 	"testing"
+	"time"
 
+	"github.com/Microsoft/go-winio"
 	"github.com/slackhq/nebula/commands"
 	"github.com/slackhq/nebula/test"
 	"github.com/stretchr/testify/assert"
 	"github.com/stretchr/testify/require"
+	"golang.org/x/sys/windows"
 )
 
-// TestInfoIsNotServedOnWindows: a pipe has no mode to open to other accounts, so Rebind
-// says so and leaves nothing listening, whatever the config asked for.
-func TestInfoIsNotServedOnWindows(t *testing.T) {
+// TestInfoIsServedOnAPipe: the info API serves HTTP on a named pipe, whose security
+// descriptor is its access control the way a socket's mode is on unix; a group named
+// in the config is admitted by it.
+func TestInfoIsServedOnAPipe(t *testing.T) {
+	users, err := windows.CreateWellKnownSid(windows.WinBuiltinUsersSid)
+	require.NoError(t, err)
+	group := &Group{Name: "Users", ID: users.String()}
+
 	s := NewInfoServer(test.NewLogger(), commands.New())
-	err := s.Rebind(context.Background(), pipePrefix+"nebula-info-test", Access{Mode: 0600})
-	require.ErrorIs(t, err, errInfoUnsupported)
-	assert.Empty(t, s.Path())
-	s.Stop()
+	path := pipePrefix + "nebula-info-test"
+	require.NoError(t, s.Rebind(context.Background(), path, Access{Mode: groupSocketMode, Group: group}))
+	t.Cleanup(s.Stop)
+	assert.Equal(t, path, s.Path())
+
+	client := &http.Client{Transport: &http.Transport{
+		// What a program on the host has to do: ask for the data rights alone, since the
+		// group is granted no more; see pipeClientAccess.
+		DialContext: func(ctx context.Context, _, _ string) (net.Conn, error) {
+			ctx, cancel := context.WithTimeout(ctx, 2*time.Second)
+			defer cancel()
+			return winio.DialPipeAccess(ctx, path, pipeClientAccess)
+		},
+	}}
+	res, err := client.Get("http://nebula/v1")
+	require.NoError(t, err)
+	defer res.Body.Close()
+	body, err := io.ReadAll(res.Body)
+	require.NoError(t, err)
+	assert.Equal(t, http.StatusOK, res.StatusCode)
+	assert.Contains(t, string(body), `"help"`)
+
+	// A change of group means a new pipe under the same name, still serving.
+	require.NoError(t, s.Rebind(context.Background(), path, Access{Mode: groupSocketMode}))
+	res, err = client.Get("http://nebula/v1")
+	require.NoError(t, err)
+	res.Body.Close()
+	assert.Equal(t, http.StatusOK, res.StatusCode)
 }

localapi/platform_windows.go

@@ -33,11 +33,10 @@ const network = "pipe"
 // exists so a Server is built the same way on every platform.
 var currentUID = func() uint32 { return 0 }
 
-// infoSupported is false here. A pipe has no mode to open to other accounts, and the
-// security descriptor listen builds admits the daemon's account alone, which is the
-// opposite of what the info API is for. Serving it on windows means designing which
-// accounts a pipe admits, and that has not been done.
-const infoSupported = false
+// infoSupported is true here: the pipe's security descriptor is what a mode is on unix,
+// and the group an operator names in the config is the way to open it to other
+// accounts, as it is for the control pipe. The mode itself means nothing to a pipe.
+const infoSupported = true
 
 // listen creates the named pipe at path with a security descriptor that admits the user
 // nebula runs as, SYSTEM and Administrators, and the group acc names, if any, for

localapi/wire.go

@@ -253,8 +253,9 @@ func wireInfo(ctx context.Context, l *slog.Logger, c *config.C, reg *commands.Re
 
 // errEndpointHeld is returned by listen when the endpoint's name is already held by
 // some other process, which nebula will not serve alongside: whoever holds it would be
-// answering nebula ctl. It is only raised on windows, where the default name can be
-// held by another administrator's process and an override outside the protected
-// namespace by any account's; on unix a path that is still being served is an earlier
-// nebula, which clearStale reports as in use.
-var errEndpointHeld = errors.New("the local control socket's name is held by another process")
+// answering the endpoint's clients. It is only raised on windows, where the default
+// name can be held by another administrator's process and an override outside the
+// protected namespace by any account's; on unix a path that is still being served is an
+// earlier nebula, which clearStale reports as in use. The log line that carries it names
+// the endpoint, so the sentence does not.
+var errEndpointHeld = errors.New("the pipe's name is held by another process")