control/controlclient,ipn/ipnlocal: wire tka enable/disable

Signed-off-by: Tom DNetto <tom@tailscale.com>

ipn/ipnlocal/local.go

@@ -684,6 +684,9 @@ func (b *LocalBackend) setClientStatus(st controlclient.Status) {
 		}
 	}
 	if st.NetMap != nil {
+		if err := b.tkaSyncIfNeededLocked(st.NetMap); err != nil {
+			b.logf("[v1] TKA sync error: %v", err)
+		}
 		if b.findExitNodeIDLocked(st.NetMap) {
 			prefsChanged = true
 		}

ipn/ipnlocal/network-lock.go

@@ -1,4 +1,4 @@
-// Copyright (c) 2021 Tailscale Inc & AUTHORS All rights reserved.
+// Copyright (c) 2022 Tailscale Inc & AUTHORS All rights reserved.
 // Use of this source code is governed by a BSD-style
 // license that can be found in the LICENSE file.
 
@@ -12,6 +12,8 @@ import (
 	"fmt"
 	"io"
 	"net/http"
+	"os"
+	"path/filepath"
 	"time"
 
 	"tailscale.com/envknob"
@@ -31,6 +33,118 @@ type tkaState struct {
 	storage   *tka.FS
 }
 
+// tkaSyncIfNeededLocked examines TKA info reported from the control plane,
+// performing the steps necessary to synchronize local tka state.
+//
+// There are 4 scenarios handled here:
+//   - Enablement: nm.TKAEnabled but b.tka == nil
+//     ∴ reach out to /machine/tka/boostrap to get the genesis AUM, then
+//     initialize TKA.
+//   - Disablement: !nm.TKAEnabled but b.tka != nil
+//     ∴ reach out to /machine/tka/boostrap to read the disablement secret,
+//     then verify and clear tka local state.
+//   - Sync needed: b.tka.Head != nm.TKAHead
+//     ∴ complete multi-step synchronization flow.
+//   - Everything up to date: All other cases.
+//     ∴ no action necessary.
+//
+// b.mu must be held. b.mu will be stepped out of (and back in) during network
+// RPCs.
+func (b *LocalBackend) tkaSyncIfNeededLocked(nm *netmap.NetworkMap) error {
+	if !networkLockAvailable() {
+		// If the feature flag is not enabled, pretend we don't exist.
+		return nil
+	}
+	if nm.SelfNode == nil {
+		return errors.New("SelfNode missing")
+	}
+
+	isEnabled := b.tka != nil
+	wantEnabled := nm.TKAEnabled
+	if isEnabled != wantEnabled {
+		var ourHead tka.AUMHash
+		if b.tka != nil {
+			ourHead = b.tka.authority.Head()
+		}
+
+		// Regardless of whether we are moving to disabled or enabled, we
+		// need information from the tka bootstrap endpoint.
+		b.mu.Unlock()
+		bs, err := b.tkaFetchBootstrap(nm.SelfNode.ID, ourHead)
+		b.mu.Lock()
+		if err != nil {
+			return fmt.Errorf("fetching bootstrap: %v", err)
+		}
+
+		if wantEnabled && !isEnabled {
+			if err := b.tkaBootstrapFromGenesisLocked(bs.GenesisAUM); err != nil {
+				return fmt.Errorf("bootstrap: %v", err)
+			}
+			isEnabled = true
+		} else if !wantEnabled && isEnabled {
+			if b.tka.authority.ValidDisablement(bs.DisablementSecret) {
+				b.tka = nil
+				isEnabled = false
+
+				if err := os.RemoveAll(b.chonkPath()); err != nil {
+					return fmt.Errorf("os.RemoveAll: %v", err)
+				}
+			} else {
+				b.logf("Disablement secret did not verify, leaving TKA enabled.")
+			}
+		} else {
+			return fmt.Errorf("[bug] unreachable invariant of wantEnabled /w isEnabled")
+		}
+	}
+
+	if isEnabled && b.tka.authority.Head() != nm.TKAHead {
+		// TODO(tom): Implement sync
+	}
+
+	return nil
+}
+
+// chonkPath returns the absolute path to the directory in which TKA
+// state (the 'tailchonk') is stored.
+func (b *LocalBackend) chonkPath() string {
+	return filepath.Join(b.TailscaleVarRoot(), "tka")
+}
+
+// tkaBootstrapFromGenesisLocked initializes the local (on-disk) state of the
+// tailnet key authority, based on the given genesis AUM.
+//
+// b.mu must be held.
+func (b *LocalBackend) tkaBootstrapFromGenesisLocked(g tkatype.MarshaledAUM) error {
+	if !b.CanSupportNetworkLock() {
+		return errors.New("network lock not supported in this configuration")
+	}
+
+	var genesis tka.AUM
+	if err := genesis.Unserialize(g); err != nil {
+		return fmt.Errorf("reading genesis: %v", err)
+	}
+
+	chonkDir := b.chonkPath()
+	if err := os.Mkdir(chonkDir, 0755); err != nil && !os.IsExist(err) {
+		return fmt.Errorf("mkdir: %v", err)
+	}
+
+	chonk, err := tka.ChonkDir(chonkDir)
+	if err != nil {
+		return fmt.Errorf("chonk: %v", err)
+	}
+	authority, err := tka.Bootstrap(chonk, genesis)
+	if err != nil {
+		return fmt.Errorf("tka bootstrap: %v", err)
+	}
+
+	b.tka = &tkaState{
+		authority: authority,
+		storage:   chonk,
+	}
+	return nil
+}
+
 // CanSupportNetworkLock returns true if tailscaled is able to operate
 // a local tailnet key authority (and hence enforce network lock).
 func (b *LocalBackend) CanSupportNetworkLock() bool {
@@ -237,3 +351,50 @@ func (b *LocalBackend) tkaInitFinish(nm *netmap.NetworkMap, nks map[tailcfg.Node
 		return a, nil
 	}
 }
+
+// tkaFetchBootstrap sends a /machine/tka/bootstrap RPC to the control plane
+// over noise. This is used to get values necessary to enable or disable TKA.
+func (b *LocalBackend) tkaFetchBootstrap(nodeID tailcfg.NodeID, head tka.AUMHash) (*tailcfg.TKABootstrapResponse, error) {
+	bootstrapReq := tailcfg.TKABootstrapRequest{
+		NodeID: nodeID,
+	}
+	if !head.IsZero() {
+		head, err := head.MarshalText()
+		if err != nil {
+			return nil, fmt.Errorf("head.MarshalText failed: %v", err)
+		}
+		bootstrapReq.Head = string(head)
+	}
+
+	var req bytes.Buffer
+	if err := json.NewEncoder(&req).Encode(bootstrapReq); err != nil {
+		return nil, fmt.Errorf("encoding request: %v", err)
+	}
+
+	ctx, cancel := context.WithTimeout(context.Background(), time.Minute)
+	defer cancel()
+	if err := ctx.Err(); err != nil {
+		return nil, fmt.Errorf("ctx: %w", err)
+	}
+	req2, err := http.NewRequestWithContext(ctx, "GET", "https://unused/machine/tka/bootstrap", &req)
+	if err != nil {
+		return nil, fmt.Errorf("req: %w", err)
+	}
+	res, err := b.DoNoiseRequest(req2)
+	if err != nil {
+		return nil, fmt.Errorf("resp: %w", err)
+	}
+	if res.StatusCode != 200 {
+		body, _ := io.ReadAll(res.Body)
+		res.Body.Close()
+		return nil, fmt.Errorf("request returned (%d): %s", res.StatusCode, string(body))
+	}
+	a := new(tailcfg.TKABootstrapResponse)
+	err = json.NewDecoder(res.Body).Decode(a)
+	res.Body.Close()
+	if err != nil {
+		return nil, fmt.Errorf("decoding JSON: %w", err)
+	}
+
+	return a, nil
+}

ipn/ipnlocal/network-lock_test.go

@@ -0,0 +1,243 @@
+// Copyright (c) 2022 Tailscale Inc & AUTHORS All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+package ipnlocal
+
+import (
+	"bytes"
+	"context"
+	"encoding/json"
+	"net"
+	"net/http"
+	"net/http/httptest"
+	"os"
+	"path/filepath"
+	"testing"
+
+	"tailscale.com/control/controlclient"
+	"tailscale.com/hostinfo"
+	"tailscale.com/tailcfg"
+	"tailscale.com/tka"
+	"tailscale.com/types/key"
+	"tailscale.com/types/netmap"
+)
+
+func fakeControlClient(t *testing.T, c *http.Client) *controlclient.Auto {
+	hi := hostinfo.New()
+	ni := tailcfg.NetInfo{LinkType: "wired"}
+	hi.NetInfo = &ni
+
+	k := key.NewMachine()
+	opts := controlclient.Options{
+		ServerURL: "https://example.com",
+		Hostinfo:  hi,
+		GetMachinePrivateKey: func() (key.MachinePrivate, error) {
+			return k, nil
+		},
+		HTTPTestClient:  c,
+		NoiseTestClient: c,
+		Status:          func(controlclient.Status) {},
+	}
+
+	cc, err := controlclient.NewNoStart(opts)
+	if err != nil {
+		t.Fatal(err)
+	}
+	return cc
+}
+
+// NOTE: URLs must have a https scheme and example.com domain to work with the underlying
+// httptest plumbing, despite the domain being unused in the actual noise request transport.
+func fakeNoiseServer(t *testing.T, handler http.HandlerFunc) (*httptest.Server, *http.Client) {
+	ts := httptest.NewUnstartedServer(handler)
+	ts.StartTLS()
+	client := ts.Client()
+	client.Transport.(*http.Transport).TLSClientConfig.InsecureSkipVerify = true
+	client.Transport.(*http.Transport).DialContext = func(ctx context.Context, network, addr string) (net.Conn, error) {
+		return (&net.Dialer{}).DialContext(ctx, network, ts.Listener.Addr().String())
+	}
+	return ts, client
+}
+
+func TestTKAEnablementFlow(t *testing.T) {
+	networkLockAvailable = func() bool { return true } // Enable the feature flag
+
+	// Make a fake TKA authority, getting a usable genesis AUM which
+	// our mock server can communicate.
+	nlPriv := key.NewNLPrivate()
+	key := tka.Key{Kind: tka.Key25519, Public: nlPriv.Public().Verifier(), Votes: 2}
+	a1, genesisAUM, err := tka.Create(&tka.Mem{}, tka.State{
+		Keys:               []tka.Key{key},
+		DisablementSecrets: [][]byte{bytes.Repeat([]byte{0xa5}, 32)},
+	}, nlPriv)
+	if err != nil {
+		t.Fatalf("tka.Create() failed: %v", err)
+	}
+
+	ts, client := fakeNoiseServer(t, http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		defer r.Body.Close()
+		switch r.URL.Path {
+		case "/machine/tka/bootstrap":
+			body := new(tailcfg.TKABootstrapRequest)
+			if err := json.NewDecoder(r.Body).Decode(body); err != nil {
+				t.Fatal(err)
+			}
+			if body.NodeID != 420 {
+				t.Errorf("bootstrap nodeID=%v, want 420", body.NodeID)
+			}
+			if body.Head != "" {
+				t.Errorf("bootstrap head=%s, want empty hash", body.Head)
+			}
+
+			w.WriteHeader(200)
+			out := tailcfg.TKABootstrapResponse{
+				GenesisAUM: genesisAUM.Serialize(),
+			}
+			if err := json.NewEncoder(w).Encode(out); err != nil {
+				t.Fatal(err)
+			}
+
+		default:
+			t.Errorf("unhandled endpoint path: %v", r.URL.Path)
+			w.WriteHeader(404)
+		}
+	}))
+	defer ts.Close()
+	temp := t.TempDir()
+
+	cc := fakeControlClient(t, client)
+	b := LocalBackend{
+		varRoot: temp,
+		cc:      cc,
+		ccAuto:  cc,
+		logf:    t.Logf,
+	}
+
+	b.mu.Lock()
+	err = b.tkaSyncIfNeededLocked(&netmap.NetworkMap{
+		SelfNode:   &tailcfg.Node{ID: 420},
+		TKAEnabled: true,
+		TKAHead:    tka.AUMHash{},
+	})
+	b.mu.Unlock()
+	if err != nil {
+		t.Errorf("tkaSyncIfNeededLocked() failed: %v", err)
+	}
+	if b.tka == nil {
+		t.Fatal("tka was not initialized")
+	}
+	if b.tka.authority.Head() != a1.Head() {
+		t.Errorf("authority.Head() = %x, want %x", b.tka.authority.Head(), a1.Head())
+	}
+}
+
+func TestTKADisablementFlow(t *testing.T) {
+	networkLockAvailable = func() bool { return true } // Enable the feature flag
+	temp := t.TempDir()
+	os.Mkdir(filepath.Join(temp, "tka"), 0755)
+
+	// Make a fake TKA authority, to seed local state.
+	disablementSecret := bytes.Repeat([]byte{0xa5}, 32)
+	nlPriv := key.NewNLPrivate()
+	key := tka.Key{Kind: tka.Key25519, Public: nlPriv.Public().Verifier(), Votes: 2}
+	chonk, err := tka.ChonkDir(filepath.Join(temp, "tka"))
+	if err != nil {
+		t.Fatal(err)
+	}
+	authority, _, err := tka.Create(chonk, tka.State{
+		Keys:               []tka.Key{key},
+		DisablementSecrets: [][]byte{tka.DisablementKDF(disablementSecret)},
+	}, nlPriv)
+	if err != nil {
+		t.Fatalf("tka.Create() failed: %v", err)
+	}
+
+	ts, client := fakeNoiseServer(t, http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		defer r.Body.Close()
+		switch r.URL.Path {
+		case "/machine/tka/bootstrap":
+			body := new(tailcfg.TKABootstrapRequest)
+			if err := json.NewDecoder(r.Body).Decode(body); err != nil {
+				t.Fatal(err)
+			}
+			var disablement []byte
+			switch body.NodeID {
+			case 42:
+				disablement = bytes.Repeat([]byte{0x42}, 32) // wrong secret
+			case 420:
+				disablement = disablementSecret
+			default:
+				t.Errorf("bootstrap nodeID=%v, wanted 42 or 420", body.NodeID)
+			}
+			var head tka.AUMHash
+			if err := head.UnmarshalText([]byte(body.Head)); err != nil {
+				t.Fatalf("failed unmarshal of body.Head: %v", err)
+			}
+			if head != authority.Head() {
+				t.Errorf("reported head = %x, want %x", head, authority.Head())
+			}
+
+			w.WriteHeader(200)
+			out := tailcfg.TKABootstrapResponse{
+				DisablementSecret: disablement,
+			}
+			if err := json.NewEncoder(w).Encode(out); err != nil {
+				t.Fatal(err)
+			}
+
+		default:
+			t.Errorf("unhandled endpoint path: %v", r.URL.Path)
+			w.WriteHeader(404)
+		}
+	}))
+	defer ts.Close()
+
+	cc := fakeControlClient(t, client)
+	b := LocalBackend{
+		varRoot: temp,
+		cc:      cc,
+		ccAuto:  cc,
+		logf:    t.Logf,
+		tka: &tkaState{
+			authority: authority,
+			storage:   chonk,
+		},
+	}
+
+	// Test that the wrong disablement secret does not shut down the authority.
+	// NodeID == 42 indicates this scenario to our mock server.
+	b.mu.Lock()
+	err = b.tkaSyncIfNeededLocked(&netmap.NetworkMap{
+		SelfNode:   &tailcfg.Node{ID: 42},
+		TKAEnabled: false,
+		TKAHead:    authority.Head(),
+	})
+	b.mu.Unlock()
+	if err != nil {
+		t.Errorf("tkaSyncIfNeededLocked() failed: %v", err)
+	}
+	if b.tka == nil {
+		t.Error("TKA was disabled despite incorrect disablement secret")
+	}
+
+	// Test the correct disablement secret shuts down the authority.
+	// NodeID == 420 indicates this scenario to our mock server.
+	b.mu.Lock()
+	err = b.tkaSyncIfNeededLocked(&netmap.NetworkMap{
+		SelfNode:   &tailcfg.Node{ID: 420},
+		TKAEnabled: false,
+		TKAHead:    authority.Head(),
+	})
+	b.mu.Unlock()
+	if err != nil {
+		t.Errorf("tkaSyncIfNeededLocked() failed: %v", err)
+	}
+
+	if b.tka != nil {
+		t.Fatal("tka was not shut down")
+	}
+	if _, err := os.Stat(b.chonkPath()); err == nil || !os.IsNotExist(err) {
+		t.Errorf("os.Stat(chonkDir) = %v, want ErrNotExist", err)
+	}
+}

ipn/ipnserver/server.go

@@ -772,7 +772,7 @@ func New(logf logger.Logf, logid string, store ipn.StateStore, eng wgengine.Engi
 	})
 
 	if root := b.TailscaleVarRoot(); root != "" {
-		chonkDir := filepath.Join(root, "chonk")
+		chonkDir := filepath.Join(root, "tka")
 		if _, err := os.Stat(chonkDir); err == nil {
 			// The directory exists, which means network-lock has been initialized.
 			storage, err := tka.ChonkDir(chonkDir)