add uncloud machine token command to print machine token to add machine to cluster

internal/machine/config.go

@@ -41,12 +41,16 @@ func ConfigPath(dataDir string) string {
 func ParseConfig(path string) (*Config, error) {
 	data, err := os.ReadFile(path)
 	if err != nil {
-		return nil, fmt.Errorf("read config file %q: %w", path, err)
+		return nil, fmt.Errorf("read config file: %w", err)
 	}
 	var config Config
 	if err = json.Unmarshal(data, &config); err != nil {
 		return nil, fmt.Errorf("parse config file %q: %w", path, err)
 	}
+
+	if config.Network == nil {
+		return nil, fmt.Errorf("missing network configuration in config file %q", path)
+	}
 	return &config, nil
 }
 

internal/machine/daemon/daemon.go

@@ -38,7 +38,7 @@ func InitCluster(dataDir, machineName string, netPrefix netip.Prefix, users []*p
 	}
 
 	// Use all routable addresses as endpoints.
-	addrs, err := network.ListRoutableAddresses()
+	addrs, err := network.ListRoutableIPs()
 	if err != nil {
 		return fmt.Errorf("list routable addresses: %w", err)
 	}
@@ -112,9 +112,30 @@ type Daemon struct {
 }
 
 func New(dataDir string) (*Daemon, error) {
-	cfg, err := machine.ParseConfig(machine.ConfigPath(dataDir))
+	cfgPath := machine.ConfigPath(dataDir)
+	cfg, err := machine.ParseConfig(cfgPath)
 	if err != nil {
-		return nil, fmt.Errorf("load machine config: %w", err)
+		if !errors.Is(err, os.ErrNotExist) {
+			return nil, fmt.Errorf("load machine config: %w", err)
+		}
+		// Generate an empty machine config with a new key pair.
+		slog.Info("Machine config not found, creating a new one.", "path", cfgPath)
+		privKey, pubKey, kErr := network.NewMachineKeys()
+		if kErr != nil {
+			return nil, fmt.Errorf("generate machine keys: %w", kErr)
+		}
+		slog.Info("Generated machine key pair.", "pubkey", pubKey)
+
+		cfg = &machine.Config{
+			Network: &network.Config{
+				PrivateKey: privKey,
+				PublicKey:  pubKey,
+			},
+		}
+		cfg.SetPath(cfgPath)
+		if err = cfg.Save(); err != nil {
+			return nil, fmt.Errorf("save machine config: %w", err)
+		}
 	}
 
 	statePath := cluster.StatePath(dataDir)
@@ -123,59 +144,74 @@ func New(dataDir string) (*Daemon, error) {
 		if !errors.Is(err, os.ErrNotExist) {
 			return nil, fmt.Errorf("load cluster state: %w", err)
 		}
-		slog.Info("No cluster state found, creating a new one.", "path", statePath)
+		slog.Info("Cluster state not found, creating a new one.", "path", statePath)
 		if err = state.Save(); err != nil {
 			return nil, fmt.Errorf("save cluster state: %w", err)
 		}
 	}
 
-	apiAddr := net.JoinHostPort(cfg.Network.ManagementIP.String(), strconv.Itoa(machine.APIPort))
-	c := cluster.NewCluster(state, apiAddr)
+	d := &Daemon{
+		config: cfg,
+	}
+	if cfg.Network.IsConfigured() {
+		apiAddr := net.JoinHostPort(cfg.Network.ManagementIP.String(), strconv.Itoa(machine.APIPort))
+		d.cluster = cluster.NewCluster(state, apiAddr)
+	}
 
-	return &Daemon{
-		config:  cfg,
-		cluster: c,
-	}, nil
+	return d, nil
 }
 
 func (d *Daemon) Run(ctx context.Context) error {
-	wgnet, err := network.NewWireGuardNetwork()
-	if err != nil {
-		return fmt.Errorf("create WireGuard network: %w", err)
-	}
-	if err = wgnet.Configure(*d.config.Network); err != nil {
-		return fmt.Errorf("configure WireGuard network: %w", err)
-	}
-	//ctx, cancel := context.WithCancel(context.Background())
-	//go wgnet.WatchEndpoints(ctx, peerEndpointChangeNotifier)
-
-	//addrs, err := network.ListRoutableAddresses()
-	//if err != nil {
-	//	return err
-	//}
-	//fmt.Println("Addresses:", addrs)
-
 	// Use an errgroup to coordinate error handling and graceful shutdown of multiple daemon components.
 	errGroup, ctx := errgroup.WithContext(ctx)
-	errGroup.Go(func() error {
-		slog.Info("Starting cluster.")
-		if err = d.cluster.Run(); err != nil {
-			return fmt.Errorf("cluster failed: %w", err)
+
+	// Start the network only if it is configured.
+	if d.config.Network.IsConfigured() {
+		wgnet, err := network.NewWireGuardNetwork()
+		if err != nil {
+			return fmt.Errorf("create WireGuard network: %w", err)
 		}
-		return nil
-	})
-	errGroup.Go(func() error {
-		if err = wgnet.Run(ctx); err != nil {
-			return fmt.Errorf("WireGuard network failed: %w", err)
+		if err = wgnet.Configure(*d.config.Network); err != nil {
+			return fmt.Errorf("configure WireGuard network: %w", err)
 		}
-		return nil
-	})
+
+		//ctx, cancel := context.WithCancel(context.Background())
+		//go wgnet.WatchEndpoints(ctx, peerEndpointChangeNotifier)
+
+		//addrs, err := network.ListRoutableIPs()
+		//if err != nil {
+		//	return err
+		//}
+		//fmt.Println("Addresses:", addrs)
+
+		errGroup.Go(func() error {
+			if err = wgnet.Run(ctx); err != nil {
+				return fmt.Errorf("WireGuard network failed: %w", err)
+			}
+			return nil
+		})
+	} else {
+		slog.Info("Waiting for network configuration to start WireGuard network.")
+	}
+
+	if d.cluster != nil {
+		errGroup.Go(func() error {
+			slog.Info("Starting cluster.")
+			if err := d.cluster.Run(); err != nil {
+				return fmt.Errorf("cluster failed: %w", err)
+			}
+			return nil
+		})
+	}
+
 	// Shutdown goroutine.
 	errGroup.Go(func() error {
 		<-ctx.Done()
-		slog.Info("Stopping cluster.")
-		d.cluster.Stop()
-		slog.Info("Cluster stopped.")
+		if d.cluster != nil {
+			slog.Info("Stopping cluster.")
+			d.cluster.Stop()
+			slog.Info("Cluster stopped.")
+		}
 		return nil
 	})
 

internal/machine/daemon/token.go

@@ -0,0 +1,41 @@
+package daemon
+
+import (
+	"errors"
+	"fmt"
+	"net/netip"
+	"os"
+	"uncloud/internal/machine"
+	"uncloud/internal/machine/network"
+)
+
+// MachineToken returns the local machine's token that can be used for adding the machine to a cluster.
+// TODO: ideally, this should be an RPC call to the daemon API to ensure the config is created and up-to-date.
+func MachineToken(dataDir string) (machine.Token, error) {
+	cfg, err := machine.ParseConfig(machine.ConfigPath(dataDir))
+	if err != nil {
+		if errors.Is(err, os.ErrNotExist) {
+			return machine.Token{}, fmt.Errorf("load machine config (is uncloudd daemon running?): %w", err)
+		}
+		return machine.Token{}, fmt.Errorf("load machine config: %w", err)
+	}
+	if len(cfg.Network.PublicKey) == 0 {
+		return machine.Token{}, errors.New("public key is not set in machine config")
+	}
+
+	ips, err := network.ListRoutableIPs()
+	if err != nil {
+		return machine.Token{}, fmt.Errorf("list routable addresses: %w", err)
+	}
+	publicIP, err := network.GetPublicIP()
+	// Ignore the error if failed to get the public IP using API services.
+	if err == nil {
+		ips = append([]netip.Addr{publicIP}, ips...)
+	}
+
+	endpoints := make([]netip.AddrPort, len(ips))
+	for i, ip := range ips {
+		endpoints[i] = netip.AddrPortFrom(ip, network.WireGuardPort)
+	}
+	return machine.NewToken(cfg.Network.PublicKey, endpoints), nil
+}

internal/machine/network/address.go

@@ -0,0 +1,106 @@
+package network
+
+import (
+	"context"
+	"fmt"
+	"io"
+	"net"
+	"net/http"
+	"net/netip"
+	"strings"
+	"time"
+)
+
+// ListRoutableIPs returns a list of routable unicast IP addresses.
+func ListRoutableIPs() ([]netip.Addr, error) {
+	interfaces, err := net.Interfaces()
+	if err != nil {
+		return nil, fmt.Errorf("list network interfaces: %w", err)
+	}
+
+	var routable []netip.Addr
+	for _, iface := range interfaces {
+		if iface.Name == WireGuardInterfaceName || strings.HasPrefix(iface.Name, "docker") {
+			// Skip the Uncloud WireGuard and Docker interfaces.
+			continue
+		}
+		if iface.Flags&net.FlagUp == 0 || iface.Flags&net.FlagRunning == 0 || iface.Flags&net.FlagLoopback != 0 {
+			// Skip interfaces:
+			//  * Not administratively UP.
+			//  * The operational status is not RUNNING. This is the closest equivalent to checking for NO-CARRIER.
+			//  * Loopback.
+			continue
+		}
+		// TODO: check for link/ether ifaces?
+
+		addrs, aErr := iface.Addrs()
+		if aErr != nil {
+			return nil, fmt.Errorf("list unicast addresses for interface %q: %w", iface.Name, err)
+		}
+		for _, addr := range addrs {
+			ipNet, ok := addr.(*net.IPNet)
+			if !ok {
+				continue
+			}
+			// Includes IPv4 private address space and local IPv6 unicast address space.
+			if ipNet.IP.IsGlobalUnicast() {
+				ip, pErr := netip.ParseAddr(ipNet.IP.String())
+				if pErr != nil {
+					return nil, fmt.Errorf("parse IP address %q: %w", ipNet.IP, err)
+				}
+				routable = append(routable, ip)
+			}
+		}
+	}
+	return routable, nil
+}
+
+func GetPublicIP() (netip.Addr, error) {
+	services := []struct {
+		URL    string
+		Parser func([]byte) (netip.Addr, error)
+	}{
+		{"https://api.ipify.org", parsePlaintextIP},
+		{"https://ipinfo.io/ip", parsePlaintextIP},
+		{"http://ip-api.com/line/?fields=query", parsePlaintextIP},
+	}
+
+	for _, service := range services {
+		if ip, err := queryIP(service.URL, service.Parser); err == nil {
+			return ip, nil
+		}
+	}
+
+	return netip.Addr{}, fmt.Errorf("failed to get public IP from all services")
+}
+
+func queryIP(service string, parser func([]byte) (netip.Addr, error)) (netip.Addr, error) {
+	ctx, cancel := context.WithTimeout(context.Background(), 5*time.Second)
+	defer cancel()
+
+	req, err := http.NewRequestWithContext(ctx, http.MethodGet, service, nil)
+	if err != nil {
+		return netip.Addr{}, fmt.Errorf("create request: %w", err)
+	}
+
+	resp, err := http.DefaultClient.Do(req)
+	if err != nil {
+		return netip.Addr{}, fmt.Errorf("send request: %w", err)
+	}
+	defer func() {
+		_ = resp.Body.Close()
+	}()
+	if resp.StatusCode != http.StatusOK {
+		return netip.Addr{}, fmt.Errorf("unexpected status code: %d", resp.StatusCode)
+	}
+
+	body, err := io.ReadAll(resp.Body)
+	if err != nil {
+		return netip.Addr{}, fmt.Errorf("read response body: %w", err)
+	}
+	return parser(body)
+}
+
+func parsePlaintextIP(data []byte) (netip.Addr, error) {
+	return netip.ParseAddr(string(data))
+}

internal/machine/network/config.go

@@ -35,6 +35,11 @@ type PeerConfig struct {
 	PublicKey    secret.Secret
 }
 
+func (c Config) IsConfigured() bool {
+	return c.Subnet != (netip.Prefix{}) && c.ManagementIP != (netip.Addr{}) &&
+		c.PrivateKey != nil && c.PublicKey != nil
+}
+
 func (c Config) toDeviceConfig() (wgtypes.Config, error) {
 	privateKey, err := wgtypes.NewKey(c.PrivateKey)
 	if err != nil {

internal/machine/token.go

@@ -0,0 +1,54 @@
+package machine
+
+import (
+	"encoding/base64"
+	"encoding/json"
+	"fmt"
+	"net/netip"
+	"strings"
+	"uncloud/internal/secret"
+)
+
+const (
+	TokenPrefix = "mtkn:"
+)
+
+// Token represents the machine's token for joining a cluster.
+type Token struct {
+	PublicKey secret.Secret
+	Endpoints []netip.AddrPort
+}
+
+// NewToken creates a new machine token with the given public key and endpoints.
+func NewToken(publicKey secret.Secret, endpoints []netip.AddrPort) Token {
+	return Token{
+		PublicKey: publicKey,
+		Endpoints: endpoints,
+	}
+}
+
+// ParseToken decodes a machine token from the given string.
+func ParseToken(s string) (Token, error) {
+	if strings.HasPrefix(s, TokenPrefix) {
+		return Token{}, fmt.Errorf("invalid token prefix")
+	}
+	decoded, err := base64.StdEncoding.DecodeString(s[len(TokenPrefix):])
+	if err != nil {
+		return Token{}, fmt.Errorf("decode token: %w", err)
+	}
+	var token Token
+	if err = json.Unmarshal(decoded, &token); err != nil {
+		return Token{}, fmt.Errorf("unmarshal token: %w", err)
+	}
+	return token, nil
+}
+
+// String returns the machine token encoded as a string.
+func (t Token) String() (string, error) {
+	js, err := json.Marshal(t)
+	if err != nil {
+		return "", fmt.Errorf("marshal token: %w", err)
+	}
+	encoded := base64.StdEncoding.EncodeToString(js)
+	return TokenPrefix + encoded, nil
+}