Cleanup

README.md

@@ -1,11 +1,5 @@
 # vppn: Virtual Potentially Private Network
 
-## TO DO
-
-* peer - write status to file instead of using sockets
-* peer - improve relay selection
-* Double buffering in IFReader and ConnReader ?
-
 ## Hub Server Configuration
 
 ```
@@ -59,15 +53,17 @@ Sign-in and configure.
 
 Install the binary somewhere, for example `~/bin/vppn`.
 
+Add the API key for your network name in `~/.vppn/<netname>/apikey`.
+
 Create systemd file in `/etc/systemd/system/vppn.service`.
 
 ```
 [Service]
-AmbientCapabilities=CAP_NET_BIND_SERVICE CAP_NET_ADMIN
+AmbientCapabilities=AP_NET_ADMIN CAP_DAC_OVERRIDE CAP_CHOWN
 Type=simple
 User=user
 WorkingDirectory=/home/user/
-ExecStart=/home/user/vppn run my_net_name https://my.hub my_api_key
+ExecStart=/home/user/bin/vppn -name my_net_name -hub https://my.hub
 Restart=always
 RestartSec=8
 TimeoutStopSec=24

cmd/vppn/main.go

@@ -1,11 +1,72 @@
 package main
 
 import (
+	"flag"
 	"log"
+	"os"
+	"path/filepath"
+	"strings"
+
 	"vppn/peer"
+
+	"git.crumpington.com/lib/go/flock"
 )
 
 func main() {
 	log.SetFlags(0)
-	peer.Main2()
+
+	name := flag.String("name", "", "network name (required)")
+	hub := flag.String("hub", "", "hub base URL (required)")
+	flag.Parse()
+
+	if *name == "" || *hub == "" {
+		flag.Usage()
+		os.Exit(1)
+	}
+
+	apiKey, err := loadAPIKey(*name)
+	if err != nil {
+		log.Fatalf("api key: %v", err)
+	}
+
+	// Directory existence is guaranteed by the apikey file read above.
+	lockFile, err := flock.TryLock(vppnPath(*name, "lock"))
+	if err != nil {
+		log.Fatalf("lock: %v", err)
+	}
+	if lockFile == nil {
+		log.Fatalf("already running for network %q", *name)
+	}
+	defer flock.Unlock(lockFile)
+
+	state, err := peer.LoadOrInit(vppnPath(*name, "state.json"), *hub, apiKey)
+	if err != nil {
+		log.Fatalf("init: %v", err)
+	}
+
+	ifaceName := strings.TrimSuffix(state.LocalDomain, ".local")
+	app, err := peer.New(state, *hub, apiKey, ifaceName, state.LocalDomain, vppnPath(*name, "network.json"))
+	if err != nil {
+		log.Fatalf("start: %v", err)
+	}
+
+	if err := app.Run(); err != nil {
+		log.Fatalf("run: %v", err)
+	}
+}
+
+func loadAPIKey(name string) (string, error) {
+	data, err := os.ReadFile(vppnPath(name, "apikey"))
+	if err != nil {
+		return "", err
+	}
+	return strings.TrimSpace(string(data)), nil
+}
+
+func vppnPath(name, file string) string {
+	home, err := os.UserHomeDir()
+	if err != nil {
+		return filepath.Join(".vppn", name, file)
+	}
+	return filepath.Join(home, ".vppn", name, file)
 }

go.mod

@@ -6,6 +6,7 @@ require (
 	git.crumpington.com/lib/go v0.9.1
 	golang.org/x/crypto v0.42.0
 	golang.org/x/sys v0.36.0
+	golang.zx2c4.com/wireguard/wgctrl v0.0.0-20241231184526-a9ab2273dd10
 )
 
 require (
@@ -15,11 +16,8 @@ require (
 	github.com/mdlayher/genetlink v1.3.2 // indirect
 	github.com/mdlayher/netlink v1.7.2 // indirect
 	github.com/mdlayher/socket v0.5.1 // indirect
-	github.com/vishvananda/netlink v1.3.1 // indirect
-	github.com/vishvananda/netns v0.0.5 // indirect
 	golang.org/x/net v0.44.0 // indirect
 	golang.org/x/sync v0.17.0 // indirect
 	golang.org/x/text v0.29.0 // indirect
 	golang.zx2c4.com/wireguard v0.0.0-20231211153847-12269c276173 // indirect
-	golang.zx2c4.com/wireguard/wgctrl v0.0.0-20241231184526-a9ab2273dd10 // indirect
 )

go.sum

@@ -12,18 +12,14 @@ github.com/mdlayher/netlink v1.7.2 h1:/UtM3ofJap7Vl4QWCPDGXY8d3GIY2UGSDbK+QWmY8/
 github.com/mdlayher/netlink v1.7.2/go.mod h1:xraEF7uJbxLhc5fpHL4cPe221LI2bdttWlU+ZGLfQSw=
 github.com/mdlayher/socket v0.5.1 h1:VZaqt6RkGkt2OE9l3GcC6nZkqD3xKeQLyfleW/uBcos=
 github.com/mdlayher/socket v0.5.1/go.mod h1:TjPLHI1UgwEv5J1B5q0zTZq12A/6H7nKmtTanQE37IQ=
-github.com/vishvananda/netlink v1.3.1 h1:3AEMt62VKqz90r0tmNhog0r/PpWKmrEShJU0wJW6bV0=
-github.com/vishvananda/netlink v1.3.1/go.mod h1:ARtKouGSTGchR8aMwmkzC0qiNPrrWO5JS/XMVl45+b4=
-github.com/vishvananda/netns v0.0.5 h1:DfiHV+j8bA32MFM7bfEunvT8IAqQ/NzSJHtcmW5zdEY=
-github.com/vishvananda/netns v0.0.5/go.mod h1:SpkAiCQRtJ6TvvxPnOSyH3BMl6unz3xZlaprSwhNNJM=
+github.com/mikioh/ipaddr v0.0.0-20190404000644-d465c8ab6721 h1:RlZweED6sbSArvlE924+mUcZuXKLBHA35U7LN621Bws=
+github.com/mikioh/ipaddr v0.0.0-20190404000644-d465c8ab6721/go.mod h1:Ickgr2WtCLZ2MDGd4Gr0geeCH5HybhRJbonOgQpvSxc=
 golang.org/x/crypto v0.42.0 h1:chiH31gIWm57EkTXpwnqf8qeuMUi0yekh6mT2AvFlqI=
 golang.org/x/crypto v0.42.0/go.mod h1:4+rDnOTJhQCx2q7/j6rAN5XDw8kPjeaXEUR2eL94ix8=
 golang.org/x/net v0.44.0 h1:evd8IRDyfNBMBTTY5XRF1vaZlD+EmWx6x8PkhR04H/I=
 golang.org/x/net v0.44.0/go.mod h1:ECOoLqd5U3Lhyeyo/QDCEVQ4sNgYsqvCZ722XogGieY=
 golang.org/x/sync v0.17.0 h1:l60nONMj9l5drqw6jlhIELNv9I0A4OFgRsG9k2oT9Ug=
 golang.org/x/sync v0.17.0/go.mod h1:9KTHXmSnoGruLpwFjVSX0lNNA75CykiMECbovNTZqGI=
-golang.org/x/sys v0.2.0/go.mod h1:oPkhp1MJrh7nUepCBck5+mAzfO9JrbApNNgaTdGDITg=
-golang.org/x/sys v0.10.0/go.mod h1:oPkhp1MJrh7nUepCBck5+mAzfO9JrbApNNgaTdGDITg=
 golang.org/x/sys v0.36.0 h1:KVRy2GtZBrk1cBYA7MKu5bEZFxQk4NIDV6RLVcC8o0k=
 golang.org/x/sys v0.36.0/go.mod h1:OgkHotnGiDImocRcuBABYBEXf8A9a87e/uXjp9XT3ks=
 golang.org/x/text v0.29.0 h1:1neNs90w9YzJ9BocxfsQNHKuAT4pkghyXc4nhZ6sJvk=

hub/api/api.go

@@ -40,7 +40,13 @@ func New(dbPath string) (*API, error) {
 		sessions: make(map[string]*Session),
 	}
 
-	return a, a.ensurePassword()
+	if err := a.ensurePassword(); err != nil {
+		return nil, err
+	}
+
+	go a.sweepSessions()
+
+	return a, nil
 }
 
 func (a *API) ensurePassword() error {
@@ -65,12 +71,8 @@ func (a *API) ensurePassword() error {
 	return db.Config_Insert(a.db, conf)
 }
 
-func (a *API) Config_Get() *Config {
-	conf, err := db.Config_Get(a.db, 1)
-	if err != nil {
-		panic(err)
-	}
-	return conf
+func (a *API) Config_Get() (*Config, error) {
+	return db.Config_Get(a.db, 1)
 }
 
 func (a *API) Config_Update(conf *Config) error {
@@ -84,46 +86,72 @@ func (a *API) Session_Delete(sessionID string) error {
 	return nil
 }
 
-func (a *API) Session_Get(sessionID string) (*Session, error) {
+const (
+	sessionTTLSecs    = 86400 * 21 // sessions expire 21 days after last use
+	sessionSweepEvery = time.Hour  // cadence of expired-session eviction
+)
+
+// Session_Get returns a snapshot copy of the signed-in session for sessionID,
+// or the zero Session if the cookie is missing/unknown/expired. It never
+// creates a session, so anonymous requests cost no memory — a session is minted
+// only by Session_SignIn. Returning a value (not the stored pointer) keeps
+// callers from racing on the shared struct.
+func (a *API) Session_Get(sessionID string) (Session, error) {
 	a.sessionsMu.Lock()
 	defer a.sessionsMu.Unlock()
 
-	if sessionID != "" {
-		s, ok := a.sessions[sessionID]
-		if ok {
-			if timeSince(s.LastSeenAt) <= 86400*21 {
-				if timeSince(s.LastSeenAt) > 86400*7 {
-					s.LastSeenAt = time.Now().Unix()
-				}
-				return s, nil
-			}
-			delete(a.sessions, sessionID)
-		}
+	s, ok := a.sessions[sessionID]
+
+	if sessionID == "" || !ok {
+		return Session{}, nil
 	}
 
-	return a.session_Create(), nil
+	if timeSince(s.LastSeenAt) > sessionTTLSecs {
+		delete(a.sessions, sessionID)
+		return Session{}, nil
+	}
+
+	s.LastSeenAt = time.Now().Unix()
+	return *s, nil
 }
 
-// caller must hold sessionsMu
-func (a *API) session_Create() *Session {
+// Session_SignIn verifies pwd and, on success, mints a fresh signed-in session,
+// returning it so the caller can set the cookie. A new ID per sign-in rotates
+// the session at the privilege boundary (session-fixation resistance).
+func (a *API) Session_SignIn(pwd string) (Session, error) {
+	conf, err := a.Config_Get()
+	if err != nil {
+		return Session{}, err
+	}
+	if err := bcrypt.CompareHashAndPassword(conf.Password, []byte(pwd)); err != nil {
+		return Session{}, ErrNotAuthorized
+	}
+
+	a.sessionsMu.Lock()
+	defer a.sessionsMu.Unlock()
 	s := &Session{
 		SessionID:  idgen.NewToken(),
+		SignedIn:   true,
 		CreatedAt:  time.Now().Unix(),
 		LastSeenAt: time.Now().Unix(),
 	}
 	a.sessions[s.SessionID] = s
-	return s
+	return *s, nil
 }
 
-func (a *API) Session_SignIn(s *Session, pwd string) error {
-	conf := a.Config_Get()
-	if err := bcrypt.CompareHashAndPassword(conf.Password, []byte(pwd)); err != nil {
-		return ErrNotAuthorized
+// sweepSessions periodically evicts sessions past their TTL. Without it, a
+// signed-in session whose ID is never presented again would linger forever
+// (Session_Get only evicts on a lookup of that same ID).
+func (a *API) sweepSessions() {
+	for range time.Tick(sessionSweepEvery) {
+		a.sessionsMu.Lock()
+		for id, s := range a.sessions {
+			if timeSince(s.LastSeenAt) > sessionTTLSecs {
+				delete(a.sessions, id)
+			}
+		}
+		a.sessionsMu.Unlock()
 	}
-	a.sessionsMu.Lock()
-	s.SignedIn = true
-	a.sessionsMu.Unlock()
-	return nil
 }
 
 func (a *API) Network_Create(n *Network) error {
@@ -140,13 +168,13 @@ func (a *API) Network_Get(id int64) (*Network, error) {
 }
 
 func (a *API) Network_List() ([]*Network, error) {
-	const query = db.Network_SelectQuery + ` ORDER BY Name ASC`
+	const query = db.Network_SelectQuery + ` ORDER BY LocalDomain ASC`
 	return db.Network_List(a.db, query)
 }
 
 func (a *API) Peer_CreateNew(p *Peer) error {
-	p.Version = idgen.NextID(0)
 	p.WGPubKey = []byte{}
+	p.SignPubKey = []byte{}
 	p.APIKey = idgen.NewToken()
 
 	return db.Peer_Insert(a.db, p)
@@ -156,20 +184,22 @@ func (a *API) Peer_Init(peer *Peer, args m.PeerInitArgs) error {
 	a.lock.Lock()
 	defer a.lock.Unlock()
 
-	peer.Version = idgen.NextID(0)
+	// Re-read from DB inside the lock — the caller's copy was fetched before
+	// we held the lock, so it may be stale under concurrent requests.
+	current, err := db.Peer_Get(a.db, peer.NetworkID, peer.PeerIP)
+	if err != nil {
+		return err
+	}
+	if len(current.WGPubKey) != 0 {
+		return errors.New("peer already initialized")
+	}
+
 	peer.WGPubKey = args.WGPubKey
+	peer.SignPubKey = args.SignPubKey
 
 	return db.Peer_UpdateFull(a.db, peer)
 }
 
-func (a *API) Peer_Update(p *Peer) error {
-	a.lock.Lock()
-	defer a.lock.Unlock()
-
-	p.Version = idgen.NextID(0)
-	return db.Peer_Update(a.db, p)
-}
-
 func (a *API) Peer_Delete(networkID int64, peerIP byte) error {
 	return db.Peer_Delete(a.db, networkID, peerIP)
 }

hub/api/db/generated.go

@@ -191,12 +191,12 @@ func Config_List(
 // ----------------------------------------------------------------------------
 
 type Network struct {
-	NetworkID int64
-	Name      string
-	Network   []byte
+	NetworkID   int64
+	LocalDomain string
+	Network     []byte
 }
 
-const Network_SelectQuery = "SELECT NetworkID,Name,Network FROM networks"
+const Network_SelectQuery = "SELECT NetworkID,LocalDomain,Network FROM networks"
 
 func Network_Insert(
 	tx TX,
@@ -207,7 +207,7 @@ func Network_Insert(
 		return err
 	}
 
-	_, err = tx.Exec("INSERT INTO networks(NetworkID,Name,Network) VALUES(?,?,?)", row.NetworkID, row.Name, row.Network)
+	_, err = tx.Exec("INSERT INTO networks(NetworkID,LocalDomain,Network) VALUES(?,?,?)", row.NetworkID, row.LocalDomain, row.Network)
 	return err
 }
 
@@ -220,7 +220,7 @@ func Network_UpdateFull(
 		return err
 	}
 
-	result, err := tx.Exec("UPDATE networks SET Name=?,Network=? WHERE NetworkID=?", row.Name, row.Network, row.NetworkID)
+	result, err := tx.Exec("UPDATE networks SET LocalDomain=?,Network=? WHERE NetworkID=?", row.LocalDomain, row.Network, row.NetworkID)
 	if err != nil {
 		return err
 	}
@@ -270,8 +270,8 @@ func Network_Get(
 	err error,
 ) {
 	row = &Network{}
-	r := tx.QueryRow("SELECT NetworkID,Name,Network FROM networks WHERE NetworkID=?", NetworkID)
-	if err = r.Scan(&row.NetworkID, &row.Name, &row.Network); err != nil {
+	r := tx.QueryRow("SELECT NetworkID,LocalDomain,Network FROM networks WHERE NetworkID=?", NetworkID)
+	if err = r.Scan(&row.NetworkID, &row.LocalDomain, &row.Network); err != nil {
 		row = nil
 	}
 	return
@@ -287,7 +287,7 @@ func Network_GetWhere(
 ) {
 	row = &Network{}
 	r := tx.QueryRow(query, args...)
-	if err = r.Scan(&row.NetworkID, &row.Name, &row.Network); err != nil {
+	if err = r.Scan(&row.NetworkID, &row.LocalDomain, &row.Network); err != nil {
 		row = nil
 	}
 	return
@@ -309,7 +309,7 @@ func Network_Iterate(
 		defer rows.Close()
 		for rows.Next() {
 			row := &Network{}
-			err := rows.Scan(&row.NetworkID, &row.Name, &row.Network)
+			err := rows.Scan(&row.NetworkID, &row.LocalDomain, &row.Network)
 			if !yield(row, err) {
 				return
 			}
@@ -339,20 +339,19 @@ func Network_List(
 // ----------------------------------------------------------------------------
 
 type Peer struct {
-	NetworkID int64
-	PeerIP    byte
-	Version   int64
-	APIKey    string
-	Name      string
-	PublicIP1 []byte
-	Port1     uint16
-	PublicIP2 []byte
-	Port2     uint16
-	Relay     bool
-	WGPubKey  []byte
+	NetworkID  int64
+	PeerIP     byte
+	APIKey     string
+	Name       string
+	Addr4      []byte
+	Addr6      []byte
+	Port       uint16
+	Relay      bool
+	WGPubKey   []byte
+	SignPubKey []byte
 }
 
-const Peer_SelectQuery = "SELECT NetworkID,PeerIP,Version,APIKey,Name,PublicIP1,Port1,PublicIP2,Port2,Relay,WGPubKey FROM peers"
+const Peer_SelectQuery = "SELECT NetworkID,PeerIP,APIKey,Name,Addr4,Addr6,Port,Relay,WGPubKey,SignPubKey FROM peers"
 
 func Peer_Insert(
 	tx TX,
@@ -363,38 +362,10 @@ func Peer_Insert(
 		return err
 	}
 
-	_, err = tx.Exec("INSERT INTO peers(NetworkID,PeerIP,Version,APIKey,Name,PublicIP1,Port1,PublicIP2,Port2,Relay,WGPubKey) VALUES(?,?,?,?,?,?,?,?,?,?,?)", row.NetworkID, row.PeerIP, row.Version, row.APIKey, row.Name, row.PublicIP1, row.Port1, row.PublicIP2, row.Port2, row.Relay, row.WGPubKey)
+	_, err = tx.Exec("INSERT INTO peers(NetworkID,PeerIP,APIKey,Name,Addr4,Addr6,Port,Relay,WGPubKey,SignPubKey) VALUES(?,?,?,?,?,?,?,?,?,?)", row.NetworkID, row.PeerIP, row.APIKey, row.Name, row.Addr4, row.Addr6, row.Port, row.Relay, row.WGPubKey, row.SignPubKey)
 	return err
 }
 
-func Peer_Update(
-	tx TX,
-	row *Peer,
-) (err error) {
-	Peer_Sanitize(row)
-	if err = Peer_Validate(row); err != nil {
-		return err
-	}
-
-	result, err := tx.Exec("UPDATE peers SET Version=?,Name=?,PublicIP1=?,Port1=?,PublicIP2=?,Port2=?,Relay=? WHERE NetworkID=? AND PeerIP=?", row.Version, row.Name, row.PublicIP1, row.Port1, row.PublicIP2, row.Port2, row.Relay, row.NetworkID, row.PeerIP)
-	if err != nil {
-		return err
-	}
-
-	n, err := result.RowsAffected()
-	if err != nil {
-		panic(err)
-	}
-	switch n {
-	case 0:
-		return sql.ErrNoRows
-	case 1:
-		return nil
-	default:
-		panic("multiple rows updated")
-	}
-}
-
 func Peer_UpdateFull(
 	tx TX,
 	row *Peer,
@@ -404,7 +375,7 @@ func Peer_UpdateFull(
 		return err
 	}
 
-	result, err := tx.Exec("UPDATE peers SET Version=?,APIKey=?,Name=?,PublicIP1=?,Port1=?,PublicIP2=?,Port2=?,Relay=?,WGPubKey=? WHERE NetworkID=? AND PeerIP=?", row.Version, row.APIKey, row.Name, row.PublicIP1, row.Port1, row.PublicIP2, row.Port2, row.Relay, row.WGPubKey, row.NetworkID, row.PeerIP)
+	result, err := tx.Exec("UPDATE peers SET APIKey=?,Name=?,Addr4=?,Addr6=?,Port=?,Relay=?,WGPubKey=?,SignPubKey=? WHERE NetworkID=? AND PeerIP=?", row.APIKey, row.Name, row.Addr4, row.Addr6, row.Port, row.Relay, row.WGPubKey, row.SignPubKey, row.NetworkID, row.PeerIP)
 	if err != nil {
 		return err
 	}
@@ -456,8 +427,8 @@ func Peer_Get(
 	err error,
 ) {
 	row = &Peer{}
-	r := tx.QueryRow("SELECT NetworkID,PeerIP,Version,APIKey,Name,PublicIP1,Port1,PublicIP2,Port2,Relay,WGPubKey FROM peers WHERE NetworkID=? AND PeerIP=?", NetworkID, PeerIP)
-	if err = r.Scan(&row.NetworkID, &row.PeerIP, &row.Version, &row.APIKey, &row.Name, &row.PublicIP1, &row.Port1, &row.PublicIP2, &row.Port2, &row.Relay, &row.WGPubKey); err != nil {
+	r := tx.QueryRow("SELECT NetworkID,PeerIP,APIKey,Name,Addr4,Addr6,Port,Relay,WGPubKey,SignPubKey FROM peers WHERE NetworkID=? AND PeerIP=?", NetworkID, PeerIP)
+	if err = r.Scan(&row.NetworkID, &row.PeerIP, &row.APIKey, &row.Name, &row.Addr4, &row.Addr6, &row.Port, &row.Relay, &row.WGPubKey, &row.SignPubKey); err != nil {
 		row = nil
 	}
 	return
@@ -473,7 +444,7 @@ func Peer_GetWhere(
 ) {
 	row = &Peer{}
 	r := tx.QueryRow(query, args...)
-	if err = r.Scan(&row.NetworkID, &row.PeerIP, &row.Version, &row.APIKey, &row.Name, &row.PublicIP1, &row.Port1, &row.PublicIP2, &row.Port2, &row.Relay, &row.WGPubKey); err != nil {
+	if err = r.Scan(&row.NetworkID, &row.PeerIP, &row.APIKey, &row.Name, &row.Addr4, &row.Addr6, &row.Port, &row.Relay, &row.WGPubKey, &row.SignPubKey); err != nil {
 		row = nil
 	}
 	return
@@ -495,7 +466,7 @@ func Peer_Iterate(
 		defer rows.Close()
 		for rows.Next() {
 			row := &Peer{}
-			err := rows.Scan(&row.NetworkID, &row.PeerIP, &row.Version, &row.APIKey, &row.Name, &row.PublicIP1, &row.Port1, &row.PublicIP2, &row.Port2, &row.Relay, &row.WGPubKey)
+			err := rows.Scan(&row.NetworkID, &row.PeerIP, &row.APIKey, &row.Name, &row.Addr4, &row.Addr6, &row.Port, &row.Relay, &row.WGPubKey, &row.SignPubKey)
 			if !yield(row, err) {
 				return
 			}

hub/api/db/sanitize-validate.go

@@ -8,9 +8,11 @@ import (
 
 var (
 	ErrInvalidIP       = errors.New("invalid IP")
+	ErrInvalidPeerIP   = errors.New("invalid peer IP")
 	ErrNonPrivateIP    = errors.New("non-private IP")
 	ErrInvalidPort     = errors.New("invalid port")
 	ErrInvalidNetName  = errors.New("invalid network name")
+	ErrNetNameNotLocal = errors.New("network name must end with .local")
 	ErrInvalidPeerName = errors.New("invalid peer name")
 )
 
@@ -22,7 +24,7 @@ func Config_Validate(c *Config) error {
 }
 
 func Network_Sanitize(n *Network) {
-	n.Name = strings.TrimSpace(n.Name)
+	n.LocalDomain = strings.TrimSpace(n.LocalDomain)
 
 	if addr, ok := netip.AddrFromSlice(n.Network); ok {
 		n.Network = addr.AsSlice()
@@ -30,12 +32,17 @@ func Network_Sanitize(n *Network) {
 }
 
 func Network_Validate(c *Network) error {
-	// 16 bytes is linux limit for network interface names.
-	if len(c.Name) == 0 || len(c.Name) > 16 {
+	// 15 bytes is linux limit for network interface names. With ending .local,
+	// max length is 21.
+	if len(c.LocalDomain) == 0 || len(c.LocalDomain) > 21 {
 		return ErrInvalidNetName
 	}
 
-	for _, c := range c.Name {
+	if !strings.HasSuffix(c.LocalDomain, ".local") {
+		return ErrNetNameNotLocal
+	}
+
+	for _, c := range strings.TrimSuffix(c.LocalDomain, ".local") {
 		if c >= 'a' && c <= 'z' {
 			continue
 		}
@@ -59,42 +66,45 @@ func Network_Validate(c *Network) error {
 
 func Peer_Sanitize(p *Peer) {
 	p.Name = strings.TrimSpace(p.Name)
-	if len(p.PublicIP1) != 0 {
-		if addr, ok := netip.AddrFromSlice(p.PublicIP1); ok {
-			p.PublicIP1 = addr.AsSlice()
+	if len(p.Addr4) != 0 {
+		if addr, ok := netip.AddrFromSlice(p.Addr4); ok {
+			// Unmap so an IPv4-mapped form is stored canonically as 4 bytes.
+			p.Addr4 = addr.Unmap().AsSlice()
 		}
 	}
-	if len(p.PublicIP2) != 0 {
-		if addr, ok := netip.AddrFromSlice(p.PublicIP2); ok {
-			p.PublicIP2 = addr.AsSlice()
+	if len(p.Addr6) != 0 {
+		if addr, ok := netip.AddrFromSlice(p.Addr6); ok {
+			p.Addr6 = addr.AsSlice()
 		}
 	}
-	if p.Port1 == 0 {
-		p.Port1 = 456
-	}
-	if len(p.PublicIP2) != 0 && p.Port2 == 0 {
-		p.Port2 = 456
+	if p.Port == 0 {
+		p.Port = 51820
 	}
 }
 
 func Peer_Validate(p *Peer) error {
-	if len(p.PublicIP1) > 0 {
-		if _, ok := netip.AddrFromSlice(p.PublicIP1); !ok {
+	if p.PeerIP < 1 || p.PeerIP > 254 {
+		return ErrInvalidPeerIP
+	}
+	if len(p.Addr4) > 0 {
+		// Must be a genuine IPv4 address (reject an IPv6 in the v4 field).
+		if addr, ok := netip.AddrFromSlice(p.Addr4); !ok || !addr.Is4() {
 			return ErrInvalidIP
 		}
 	}
-	if len(p.PublicIP2) > 0 {
-		if _, ok := netip.AddrFromSlice(p.PublicIP2); !ok {
+	if len(p.Addr6) > 0 {
+		// Must be a genuine IPv6 address (reject IPv4 / IPv4-mapped in the v6 field).
+		if addr, ok := netip.AddrFromSlice(p.Addr6); !ok || !addr.Is6() || addr.Is4In6() {
 			return ErrInvalidIP
 		}
-		if p.Port2 == 0 {
-			return ErrInvalidPort
-		}
 	}
-	if p.Port1 == 0 {
+	if p.Port == 0 {
 		return ErrInvalidPort
 	}
 
+	if len(p.Name) == 0 {
+		return ErrInvalidPeerName
+	}
 	for _, c := range p.Name {
 		if c >= 'a' && c <= 'z' {
 			continue
@@ -102,7 +112,7 @@ func Peer_Validate(p *Peer) error {
 		if c >= '0' && c <= '9' {
 			continue
 		}
-		if c == '.' || c == '-' || c == '_' {
+		if c == '-' {
 			continue
 		}
 		return ErrInvalidPeerName

hub/api/db/tables.defs

@@ -4,21 +4,20 @@ TABLE config OF Config (
 );
 
 TABLE networks OF Network (
-  NetworkID int64 PK,
-  Name      string NoUpdate,
-  Network   []byte NoUpdate
+  NetworkID   int64 PK,
+  LocalDomain string NoUpdate,
+  Network     []byte NoUpdate
 );
 
 TABLE peers OF Peer (
   NetworkID  int64 PK,
   PeerIP     byte PK,
-  Version    int64,
   APIKey     string NoUpdate,
-  Name       string,
-  PublicIP1  []byte,
-  Port1      uint16,
-  PublicIP2  []byte,
-  Port2      uint16,
-  Relay      bool,
-  WGPubKey   []byte NoUpdate
+  Name       string NoUpdate,
+  Addr4      []byte NoUpdate,
+  Addr6      []byte NoUpdate,
+  Port       uint16 NoUpdate,
+  Relay      bool   NoUpdate,
+  WGPubKey   []byte NoUpdate,
+  SignPubKey []byte NoUpdate
 );

hub/api/db/written.go

@@ -11,9 +11,3 @@ func Peer_GetByAPIKey(tx TX, apiKey string) (*Peer, error) {
 		Peer_SelectQuery+` WHERE APIKey=?`,
 		apiKey)
 }
-
-func Peer_Exists(tx TX, networkID int64, ip byte) (exists bool, err error) {
-	const query = `SELECT EXISTS(SELECT 1 FROM peers WHERE NetworkID=? AND PeerIP=?)`
-	err = tx.QueryRow(query, networkID, ip).Scan(&exists)
-	return
-}

hub/api/errors.go

@@ -7,7 +7,6 @@ import (
 
 var (
 	ErrNotAuthorized = errors.New("not authorized")
-	ErrNoIPAvailable = errors.New("no IP address available")
 	ErrInvalidIP     = db.ErrInvalidIP
 	ErrInvalidPort   = db.ErrInvalidPort
 )

hub/api/migrations/2024-11-30-init.sql

@@ -5,21 +5,21 @@ CREATE TABLE config (
 
 CREATE TABLE networks (
   NetworkID   INTEGER NOT NULL PRIMARY KEY,
-  Name        TEXT    NOT NULL UNIQUE, -- Network/interface name.
+  LocalDomain TEXT    NOT NULL UNIQUE, -- Network/interface name.
   Network     BLOB    NOT NULL UNIQUE  -- Network (/24), example 10.51.50.0
 ) WITHOUT ROWID;
 
 CREATE TABLE peers (
   NetworkID  INTEGER NOT NULL,
   PeerIP     INTEGER NOT NULL,           -- Final byte of IP.
-  Version    INTEGER NOT NULL,           -- Changes when updated.
   APIKey     TEXT    NOT NULL UNIQUE,    -- Peer's secret API key.
-  Name       TEXT    NOT NULL UNIQUE,    -- For humans.
-  PublicIP1  BLOB    NOT NULL,
-  Port1      INTEGER NOT NULL,
-  PublicIP2  BLOB    NOT NULL,
-  Port2      INTEGER NOT NULL,
+  Name       TEXT    NOT NULL,           -- For humans.
+  Addr4      BLOB    NOT NULL,
+  Addr6      BLOB    NOT NULL,
+  Port       INTEGER NOT NULL,
   Relay      INTEGER NOT NULL DEFAULT 0, -- Boolean if peer will forward packets.
   WGPubKey   BLOB    NOT NULL,
+  SignPubKey BLOB    NOT NULL,
+  UNIQUE(NetworkID, Name),
   PRIMARY KEY(NetworkID, PeerIP)
 ) WITHOUT ROWID;

hub/cookie.go

@@ -2,7 +2,6 @@ package hub
 
 import (
 	"net/http"
-	"time"
 )
 
 func (a *App) getCookie(r *http.Request, name string) string {
@@ -26,9 +25,12 @@ func (a *App) setCookie(w http.ResponseWriter, name, value string) {
 
 func (a *App) deleteCookie(w http.ResponseWriter, name string) {
 	http.SetCookie(w, &http.Cookie{
-		Name:    name,
-		Value:   "",
-		Path:    "/",
-		Expires: time.Unix(0, 0),
+		Name:     name,
+		Value:    "",
+		Path:     "/",
+		Secure:   !a.insecure,
+		SameSite: http.SameSiteStrictMode,
+		HttpOnly: true,
+		MaxAge:   -1, // delete now
 	})
 }

hub/global.go

@@ -1,5 +1,5 @@
 package hub
 
 const (
-	SESSION_ID_COOKIE_NAME = "SessionID"
+	sessionIDCookieName = "SessionID"
 )

hub/handler.go

@@ -12,7 +12,7 @@ type handlerFunc func(s *api.Session, w http.ResponseWriter, r *http.Request) er
 
 func (app *App) handlePub(pattern string, fn handlerFunc) {
 	wrapped := func(w http.ResponseWriter, r *http.Request) {
-		sessionID := app.getCookie(r, SESSION_ID_COOKIE_NAME)
+		sessionID := app.getCookie(r, sessionIDCookieName)
 		s, err := app.api.Session_Get(sessionID)
 		if err != nil {
 			log.Printf("Failed to get session: %v", err)
@@ -20,17 +20,13 @@ func (app *App) handlePub(pattern string, fn handlerFunc) {
 			return
 		}
 
-		if s.SessionID != sessionID {
-			app.setCookie(w, SESSION_ID_COOKIE_NAME, s.SessionID)
-		}
-
 		if r.Method == http.MethodPost {
 			r.ParseMultipartForm(64 * 1024)
 		} else {
 			r.ParseForm()
 		}
 
-		if err := fn(s, w, r); err != nil {
+		if err := fn(&s, w, r); err != nil {
 			http.Error(w, err.Error(), http.StatusInternalServerError)
 		}
 	}

hub/handlers.go

@@ -10,6 +10,7 @@ import (
 
 	"git.crumpington.com/lib/go/webutil"
 	"golang.org/x/crypto/bcrypt"
+	"golang.zx2c4.com/wireguard/wgctrl/wgtypes"
 )
 
 func (a *App) _root(s *api.Session, w http.ResponseWriter, r *http.Request) error {
@@ -33,9 +34,11 @@ func (a *App) _signinSubmit(s *api.Session, w http.ResponseWriter, r *http.Reque
 		return err
 	}
 
-	if err := a.api.Session_SignIn(s, pwd); err != nil {
+	sess, err := a.api.Session_SignIn(pwd)
+	if err != nil {
 		return err
 	}
+	a.setCookie(w, sessionIDCookieName, sess.SessionID)
 
 	return a.redirect(w, r, "/")
 }
@@ -48,7 +51,7 @@ func (a *App) _adminSignOutSubmit(s *api.Session, w http.ResponseWriter, r *http
 	if err := a.api.Session_Delete(s.SessionID); err != nil {
 		log.Printf("Failed to delete session cookie %s: %v", s.SessionID, err)
 	}
-	a.deleteCookie(w, SESSION_ID_COOKIE_NAME)
+	a.deleteCookie(w, sessionIDCookieName)
 	return a.redirect(w, r, "/")
 }
 
@@ -72,7 +75,7 @@ func (a *App) _adminNetworkCreateSubmit(s *api.Session, w http.ResponseWriter, r
 	var netStr string
 
 	err := webutil.NewFormScanner(r.Form).
-		Scan("Name", &n.Name).
+		Scan("LocalDomain", &n.LocalDomain).
 		Scan("Network", &netStr).
 		Error()
 	if err != nil {
@@ -142,27 +145,26 @@ func (a *App) _adminPeerCreate(s *api.Session, w http.ResponseWriter, r *http.Re
 }
 
 func (a *App) _adminPeerCreateSubmit(s *api.Session, w http.ResponseWriter, r *http.Request) error {
-	var ip1Str, ip2Str string
+	var addr4Str, addr6Str string
 
 	p := &api.Peer{}
 	err := webutil.NewFormScanner(r.Form).
 		Scan("NetworkID", &p.NetworkID).
 		Scan("IP", &p.PeerIP).
 		Scan("Name", &p.Name).
-		Scan("PublicIP1", &ip1Str).
-		Scan("Port1", &p.Port1).
-		Scan("PublicIP2", &ip2Str).
-		Scan("Port2", &p.Port2).
+		Scan("Addr4", &addr4Str).
+		Scan("Addr6", &addr6Str).
+		Scan("Port", &p.Port).
 		Scan("Relay", &p.Relay).
 		Error()
 	if err != nil {
 		return err
 	}
 
-	if p.PublicIP1, err = stringToIP(ip1Str); err != nil {
+	if p.Addr4, err = stringToIP(addr4Str); err != nil {
 		return err
 	}
-	if p.PublicIP2, err = stringToIP(ip2Str); err != nil {
+	if p.Addr6, err = stringToIP(addr6Str); err != nil {
 		return err
 	}
 
@@ -185,53 +187,6 @@ func (a *App) _adminPeerView(s *api.Session, w http.ResponseWriter, r *http.Requ
 	}{s, net, peer})
 }
 
-func (a *App) _adminPeerEdit(s *api.Session, w http.ResponseWriter, r *http.Request) error {
-	net, peer, err := a.formGetPeer(r.Form)
-	if err != nil {
-		return err
-	}
-
-	return a.render("/network/peer-edit.html", w, struct {
-		Session *api.Session
-		Network *api.Network
-		Peer    *api.Peer
-	}{s, net, peer})
-}
-
-func (a *App) _adminPeerEditSubmit(s *api.Session, w http.ResponseWriter, r *http.Request) error {
-	_, peer, err := a.formGetPeer(r.Form)
-	if err != nil {
-		return err
-	}
-
-	var ip1Str, ip2Str string
-
-	err = webutil.NewFormScanner(r.Form).
-		Scan("Name", &peer.Name).
-		Scan("PublicIP1", &ip1Str).
-		Scan("Port1", &peer.Port1).
-		Scan("PublicIP2", &ip2Str).
-		Scan("Port2", &peer.Port2).
-		Scan("Relay", &peer.Relay).
-		Error()
-	if err != nil {
-		return err
-	}
-
-	if peer.PublicIP1, err = stringToIP(ip1Str); err != nil {
-		return err
-	}
-	if peer.PublicIP2, err = stringToIP(ip2Str); err != nil {
-		return err
-	}
-
-	if err = a.api.Peer_Update(peer); err != nil {
-		return err
-	}
-
-	return a.redirect(w, r, "/admin/peer/view/?NetworkID=%d&PeerIP=%d", peer.NetworkID, peer.PeerIP)
-}
-
 func (a *App) _adminPeerDelete(s *api.Session, w http.ResponseWriter, r *http.Request) error {
 	n, peer, err := a.formGetPeer(r.Form)
 	if err != nil {
@@ -262,13 +217,17 @@ func (a *App) _adminPasswordEdit(s *api.Session, w http.ResponseWriter, r *http.
 
 func (a *App) _adminPasswordSubmit(s *api.Session, w http.ResponseWriter, r *http.Request) error {
 	var (
-		conf    = a.api.Config_Get()
 		curPwd  string
 		newPwd  string
 		newPwd2 string
 	)
 
-	err := webutil.NewFormScanner(r.Form).
+	conf, err := a.api.Config_Get()
+	if err != nil {
+		return err
+	}
+
+	err = webutil.NewFormScanner(r.Form).
 		Scan("CurrentPassword", &curPwd).
 		Scan("NewPassword", &newPwd).
 		Scan("NewPassword2", &newPwd2).
@@ -315,6 +274,15 @@ func (a *App) _peerInit(peer *api.Peer, w http.ResponseWriter, r *http.Request)
 		return err
 	}
 
+	if len(args.WGPubKey) != 32 {
+		http.Error(w, "invalid WGPubKey", http.StatusBadRequest)
+		return nil
+	}
+	if len(args.SignPubKey) != 32 {
+		http.Error(w, "invalid SignPubKey", http.StatusBadRequest)
+		return nil
+	}
+
 	net, err := a.api.Network_Get(peer.NetworkID)
 	if err != nil {
 		return err
@@ -325,11 +293,12 @@ func (a *App) _peerInit(peer *api.Peer, w http.ResponseWriter, r *http.Request)
 	}
 
 	resp := m.PeerInitResp{
-		PeerIP:  peer.PeerIP,
-		Network: net.Network,
+		PeerIP:      peer.PeerIP,
+		Network:     net.Network,
+		LocalDomain: net.LocalDomain,
 	}
 
-	resp.NetworkState.Peers, err = a.peersArray(net.NetworkID)
+	resp.NetworkState.Peers, err = a.peersList(net.NetworkID)
 	if err != nil {
 		return err
 	}
@@ -338,35 +307,42 @@ func (a *App) _peerInit(peer *api.Peer, w http.ResponseWriter, r *http.Request)
 }
 
 func (a *App) _peerFetchState(peer *api.Peer, w http.ResponseWriter, r *http.Request) error {
-
-	peers, err := a.peersArray(peer.NetworkID)
+	peers, err := a.peersList(peer.NetworkID)
 	if err != nil {
 		return err
 	}
 	return a.sendJSON(w, m.NetworkState{Peers: peers})
 }
 
-func (a *App) peersArray(networkID int64) (peers [256]*m.Peer, err error) {
+func (a *App) peersList(networkID int64) (peers []m.Peer, err error) {
 	l, err := a.api.Peer_List(networkID)
 	if err != nil {
-		return peers, err
+		return nil, err
 	}
 
+	peers = make([]m.Peer, 0, len(l))
+
 	for _, p := range l {
-		if len(p.WGPubKey) != 0 {
-			peers[p.PeerIP] = &m.Peer{
-				PeerIP:    p.PeerIP,
-				Version:   p.Version,
-				Name:      p.Name,
-				PublicIP1: p.PublicIP1,
-				Port1:     p.Port1,
-				PublicIP2: p.PublicIP2,
-				Port2:     p.Port2,
-				Relay:     p.Relay,
-				WGPubKey:  p.WGPubKey,
-			}
+		if len(p.WGPubKey) == 0 {
+			continue
 		}
+		wgKey, err := wgtypes.NewKey(p.WGPubKey)
+		if err != nil {
+			continue // malformed key; skip rather than serve garbage
+		}
+		var signKey [32]byte
+		copy(signKey[:], p.SignPubKey)
+		peers = append(peers, m.Peer{
+			PeerIP:     p.PeerIP,
+			Name:       p.Name,
+			Addr4:      addrFromBytes(p.Addr4),
+			Addr6:      addrFromBytes(p.Addr6),
+			Port:       p.Port,
+			Relay:      p.Relay,
+			WGPubKey:   wgKey,
+			SignPubKey: signKey,
+		})
 	}
 
-	return
+	return peers, nil
 }

hub/routes.go

@@ -22,8 +22,6 @@ func (a *App) registerRoutes() {
 	a.handleSignedIn("GET  /admin/peer/create/", a._adminPeerCreate)
 	a.handleSignedIn("POST /admin/peer/create/", a._adminPeerCreateSubmit)
 	a.handleSignedIn("GET  /admin/peer/view/", a._adminPeerView)
-	a.handleSignedIn("GET  /admin/peer/edit/", a._adminPeerEdit)
-	a.handleSignedIn("POST /admin/peer/edit/", a._adminPeerEditSubmit)
 	a.handleSignedIn("GET  /admin/peer/delete/", a._adminPeerDelete)
 	a.handleSignedIn("POST /admin/peer/delete/", a._adminPeerDeleteSubmit)
 

hub/templates/admin-network-create.html

@@ -3,8 +3,8 @@
 
 <form method="POST">
 <p>
-    <label>Name</label><br>
-    <input type="text" name="Name">
+    <label>Local Domain (ending with .local)</label><br>
+    <input type="text" name="LocalDomain">
   </p>
   <p>
     <label>Network /24</label><br>

hub/templates/admin-network-list.html

@@ -9,7 +9,7 @@
 <table>
   <thead>
     <tr>
-      <th>Name</th>
+      <th>Local Domain</th>
       <th>Network</th>
     </tr>
   </thead>
@@ -18,7 +18,7 @@
   <tr>
     <td>
       <a href="/admin/network/view/?NetworkID={{.NetworkID}}">
-        {{.Name}}
+        {{.LocalDomain}}
       </a>
     </td>
     <td>{{ipToString .Network}}</td>

hub/templates/network/base.html

@@ -17,7 +17,7 @@
   </header>
   <h2>
     Network:
-    <a href="/admin/network/view/?NetworkID={{.Network.NetworkID}}">{{.Network.Name}}</a>
+    <a href="/admin/network/view/?NetworkID={{.Network.NetworkID}}">{{.Network.LocalDomain}}</a>
   </h2>
 
   {{block "body" .}}There's nothing here.{{end}}

hub/templates/network/network-view.html

@@ -22,10 +22,9 @@
     <tr>
       <th>PeerIP</th>
       <th>Name</th>
-      <th>Public IP 1</th>
-      <th>Port 1</th>
-      <th>Public IP 2</th>
-      <th>Port 2</th>
+      <th>IPv4</th>
+      <th>IPv6</th>
+      <th>Port</th>
       <th>Relay</th>
     </tr>
   </thead>
@@ -38,10 +37,9 @@
       </a>
     </td>
     <td>{{.Name}}</td>
-    <td>{{ipToString .PublicIP1}}</td>
-    <td>{{.Port1}}</td>
-    <td>{{ipToString .PublicIP2}}</td>
-    <td>{{.Port2}}</td>
+    <td>{{ipToString .Addr4}}</td>
+    <td>{{ipToString .Addr6}}</td>
+    <td>{{.Port}}</td>
     <td>{{if .Relay}}T{{else}}F{{end}}</td>
   </tr>
   </tbody>

hub/templates/network/peer-create.html

@@ -12,20 +12,16 @@
     <input type="text" name="Name">
   </p>
   <p>
-    <label>Public IP 1</label><br>
-    <input type="text" name="PublicIP1">
+    <label>IPv4 Address (optional)</label><br>
+    <input type="text" name="Addr4">
   </p>
   <p>
-    <label>Port 1</label><br>
-    <input type="number" name="Port1" value="456">
+    <label>IPv6 Address (optional)</label><br>
+    <input type="text" name="Addr6">
   </p>
   <p>
-    <label>Public IP 2 (optional)</label><br>
-    <input type="text" name="PublicIP2">
-  </p>
-  <p>
-    <label>Port 2</label><br>
-    <input type="number" name="Port2" value="0">
+    <label>WireGuard Port</label><br>
+    <input type="number" name="Port" value="51820">
   </p>
   <p>
     <label>

hub/templates/network/peer-delete.html

@@ -4,7 +4,7 @@
 {{with .Peer -}}
 <form method="POST">
 <input type="hidden" name="NetworkID" value="{{.NetworkID}}">
-  <input type="hidden" name="NetworkID" value="{{.PeerIP}}">
+  <input type="hidden" name="PeerIP" value="{{.PeerIP}}">
   <p>
     <button type="submit">Delete</button>
     <a href="/admin/peer/view/?PeerIP={{.PeerIP}}&NetworkID={{.NetworkID}}">Cancel</a>

hub/templates/network/peer-edit.html

@@ -1,42 +0,0 @@
-{{define "body" -}}
-<h2>Edit Peer</h2>
-
-{{with .Peer -}}
-<form method="POST">
-  <p>
-    <label>Peer IP</label><br>
-    <input type="text" value="{{.PeerIP}}" disabled>
-  </p>
-  <p>
-    <label>Name</label><br>
-    <input type="text" name="Name" value="{{.Name}}">
-  </p>
-  <p>
-    <label>Public IP 1</label><br>
-    <input type="text" name="PublicIP1" value="{{ipToString .PublicIP1}}">
-  </p>
-  <p>
-    <label>Port 1</label><br>
-    <input type="number" name="Port1" value="{{.Port1}}">
-  </p>
-  <p>
-    <label>Public IP 2 (optional)</label><br>
-    <input type="text" name="PublicIP2" value="{{ipToString .PublicIP2}}">
-  </p>
-  <p>
-    <label>Port 2</label><br>
-    <input type="number" name="Port2" value="{{.Port2}}">
-  </p>
-  <p>
-    <label>
-      <input type="checkbox" name="Relay" {{if .Relay}}checked{{end}}>
-      Relay
-    </label>
-  </p>
-  <p>
-    <button type="submit">Save</button>
-    <a href="/admin/peer/view/?NetworkID={{$.Network.NetworkID}}&PeerIP={{.PeerIP}}">Cancel</a>
-  </p>
-</form>
-{{- end}}
-{{- end}}

hub/templates/network/peer-view.html

@@ -1,17 +1,15 @@
 {{define "body" -}}
 <h3>{{.Peer.Name}}</h3>
 <p>
-  <a href="/admin/peer/edit/?NetworkID={{.Network.NetworkID}}&PeerIP={{.Peer.PeerIP}}">Edit</a> /
   <a href="/admin/peer/delete/?NetworkID={{.Network.NetworkID}}&PeerIP={{.Peer.PeerIP}}">Delete</a>
 </p>
 
 {{with .Peer -}}
 <table class="def-list">
   <tr><td>Peer IP</td><td>{{.PeerIP}}</td></tr>
-  <tr><td>Public IP 1</td><td>{{ipToString .PublicIP1}}</td></tr>
-  <tr><td>Port 1</td><td>{{.Port1}}</td></tr>
-  <tr><td>Public IP 2</td><td>{{ipToString .PublicIP2}}</td></tr>
-  <tr><td>Port 2</td><td>{{.Port2}}</td></tr>
+  <tr><td>IPv4 Address</td><td>{{ipToString .Addr4}}</td></tr>
+  <tr><td>IPv6 Address</td><td>{{ipToString .Addr6}}</td></tr>
+  <tr><td>WireGuard Port</td><td>{{.Port}}</td></tr>
   <tr><td>Relay</td><td>{{if .Relay}}T{{else}}F{{end}}</td></tr>
   <tr><td>WG Public Key</td><td>{{wgKeyString .WGPubKey}}</td></tr>
 </table>

hub/util.go

@@ -38,6 +38,19 @@ func (app *App) sendJSON(w http.ResponseWriter, data any) error {
 	return nil
 }
 
+// addrFromBytes parses raw IP bytes (4 or 16) into a netip.Addr, unmapping
+// IPv4-in-IPv6, returning the zero Addr for empty/invalid input.
+func addrFromBytes(b []byte) netip.Addr {
+	if len(b) == 0 {
+		return netip.Addr{}
+	}
+	addr, ok := netip.AddrFromSlice(b)
+	if !ok {
+		return netip.Addr{}
+	}
+	return addr.Unmap()
+}
+
 func stringToIP(in string) ([]byte, error) {
 	in = strings.TrimSpace(in)
 	if len(in) == 0 {

m/models.go

@@ -1,28 +1,133 @@
 // The package `m` contains models shared between the hub and peer programs.
 package m
 
+import (
+	"encoding/base64"
+	"encoding/json"
+	"fmt"
+	"net/netip"
+
+	"golang.zx2c4.com/wireguard/wgctrl/wgtypes"
+)
+
 type PeerInitArgs struct {
-	WGPubKey []byte
+	WGPubKey   []byte
+	SignPubKey []byte
 }
 
 type PeerInitResp struct {
 	PeerIP       byte
 	Network      []byte
+	LocalDomain  string
 	NetworkState NetworkState
 }
 
+// Peer is the network membership record for a single peer, exchanged between
+// the hub and peers. Addr4/Addr6 are the peer's public endpoint addresses (zero
+// if it has none); Port is its WireGuard listen port, meaningful even for a
+// non-public peer (it is the peer's own bind/beacon port).
 type Peer struct {
-	PeerIP    byte
-	Version   int64
-	Name      string
-	PublicIP1 []byte
-	Port1     uint16
-	PublicIP2 []byte
-	Port2     uint16
-	Relay     bool
-	WGPubKey  []byte
+	PeerIP     byte
+	Name       string
+	Addr4      netip.Addr // zero if none
+	Addr6      netip.Addr // zero if none
+	Port       uint16
+	Relay      bool
+	WGPubKey   wgtypes.Key
+	SignPubKey [32]byte
+}
+
+// IsPublic reports whether the peer advertises at least one reachable endpoint.
+func (p Peer) IsPublic() bool {
+	return p.Addr4.IsValid() || p.Addr6.IsValid()
+}
+
+// Endpoint4 returns the IPv4 endpoint (addr+port), or the zero AddrPort if the
+// peer has no IPv4 address.
+func (p Peer) Endpoint4() netip.AddrPort {
+	if !p.Addr4.IsValid() {
+		return netip.AddrPort{}
+	}
+	return netip.AddrPortFrom(p.Addr4, p.Port)
+}
+
+// Endpoint6 returns the IPv6 endpoint (addr+port), or the zero AddrPort if the
+// peer has no IPv6 address.
+func (p Peer) Endpoint6() netip.AddrPort {
+	if !p.Addr6.IsValid() {
+		return netip.AddrPort{}
+	}
+	return netip.AddrPortFrom(p.Addr6, p.Port)
+}
+
+// PreferredEndpoint returns the IPv4 endpoint if present, else IPv6.
+func (p Peer) PreferredEndpoint() netip.AddrPort {
+	if ep := p.Endpoint4(); ep.IsValid() {
+		return ep
+	}
+	return p.Endpoint6()
+}
+
+// peerJSON is the wire representation. netip.Addr fields round-trip as text
+// strings automatically; only the fixed-size key arrays need base64 (otherwise
+// encoding/json would emit them as arrays of numbers).
+type peerJSON struct {
+	PeerIP     byte
+	Name       string
+	Addr4      netip.Addr
+	Addr6      netip.Addr
+	Port       uint16
+	Relay      bool
+	WGPubKey   string
+	SignPubKey string
+}
+
+func (p Peer) MarshalJSON() ([]byte, error) {
+	return json.Marshal(peerJSON{
+		PeerIP:     p.PeerIP,
+		Name:       p.Name,
+		Addr4:      p.Addr4,
+		Addr6:      p.Addr6,
+		Port:       p.Port,
+		Relay:      p.Relay,
+		WGPubKey:   base64.StdEncoding.EncodeToString(p.WGPubKey[:]),
+		SignPubKey: base64.StdEncoding.EncodeToString(p.SignPubKey[:]),
+	})
+}
+
+func (p *Peer) UnmarshalJSON(data []byte) error {
+	var j peerJSON
+	if err := json.Unmarshal(data, &j); err != nil {
+		return err
+	}
+	wg, err := base64.StdEncoding.DecodeString(j.WGPubKey)
+	if err != nil {
+		return fmt.Errorf("decode WGPubKey: %w", err)
+	}
+	key, err := wgtypes.NewKey(wg)
+	if err != nil {
+		return fmt.Errorf("invalid WGPubKey: %w", err)
+	}
+	sign, err := base64.StdEncoding.DecodeString(j.SignPubKey)
+	if err != nil {
+		return fmt.Errorf("decode SignPubKey: %w", err)
+	}
+	if len(sign) != 32 {
+		return fmt.Errorf("invalid SignPubKey length: %d", len(sign))
+	}
+	*p = Peer{
+		PeerIP:     j.PeerIP,
+		Name:       j.Name,
+		Addr4:      j.Addr4,
+		Addr6:      j.Addr6,
+		Port:       j.Port,
+		Relay:      j.Relay,
+		WGPubKey:   key,
+		SignPubKey: [32]byte(sign),
+	}
+	return nil
 }
 
 type NetworkState struct {
-	Peers [256]*Peer
+	Peers []Peer
 }

peer/app.go

@@ -0,0 +1,105 @@
+package peer
+
+import (
+	"net/netip"
+	"os"
+	"os/signal"
+	"syscall"
+	"time"
+
+	"golang.zx2c4.com/wireguard/wgctrl/wgtypes"
+
+	"vppn/m"
+	"vppn/peer/control"
+	"vppn/peer/multicast"
+	"vppn/peer/wginterface"
+)
+
+var _ WGDevice = (*wginterface.Device)(nil) // compile-time check: Device satisfies WGDevice
+
+const (
+	ControlPort     = 4561
+	PingInterval    = 8 * time.Second
+	TimeoutInterval = 30 * time.Second
+)
+
+// scratchSize is large enough for the biggest buffer either the ping or the
+// multicast path serializes through the shared App scratch.
+const scratchSize = max(control.Size, multicast.SignedPacketSize)
+
+type PingEvent struct {
+	srcVPNIP netip.Addr
+	ping     control.Ping
+}
+
+// App is the peer application. All mutable state lives here and is
+// accessed only from the Run goroutine.
+type App struct {
+	// Identity
+	vpnIP       netip.Addr
+	vpnNet      netip.Prefix
+	privKey     wgtypes.Key
+	pubKey      wgtypes.Key
+	isRelay     bool
+	isPublic    bool
+	localDomain string
+
+	// Infrastructure
+	dev         WGDevice
+	controlConn ControlConn
+
+	// Peer state
+	relay      *Peer
+	peersByKey map[wgtypes.Key]*Peer
+	peersByIP  map[netip.Addr]*Peer
+
+	// Our own external endpoints, learned from Dst fields in incoming pings
+	selfV4 netip.AddrPort
+	selfV6 netip.AddrPort
+
+	// Reusable serialization scratch for outgoing pings and multicast signature
+	// verification. Only touched from the Run goroutine.
+	scratch []byte
+
+	// Event channels fed by background goroutines
+	hubAddCh    <-chan m.Peer
+	hubRemoveCh <-chan wgtypes.Key
+	pingCh      <-chan PingEvent
+	multicastCh <-chan multicast.Packet
+}
+
+// Run is the main event loop. It runs until SIGTERM/SIGINT.
+func (a *App) Run() error {
+	// Establish a clean hosts section before the first poll lands, clearing
+	// any stale entries left by a prior run (e.g. crash, or peers removed
+	// while we were down).
+	a.updateHosts()
+
+	ticker := time.NewTicker(PingInterval)
+	defer ticker.Stop()
+
+	sig := make(chan os.Signal, 1)
+	signal.Notify(sig, syscall.SIGTERM, syscall.SIGINT)
+	defer signal.Stop(sig)
+
+	for {
+		select {
+		case p := <-a.hubAddCh:
+			a.onAddPeer(p)
+		case key := <-a.hubRemoveCh:
+			a.onRemovePeer(key)
+		case e := <-a.pingCh:
+			a.onPing(e)
+		case e := <-a.multicastCh:
+			a.onMulticastDiscovery(e)
+		case <-ticker.C:
+			a.onTick()
+		case <-sig:
+			return a.onShutdown()
+		}
+	}
+}
+
+func (a *App) onShutdown() error {
+	return wginterface.Delete(a.dev.Name())
+}

peer/app_test.go

@@ -0,0 +1,62 @@
+package peer
+
+import (
+	"net/netip"
+	"testing"
+	"time"
+
+	"golang.zx2c4.com/wireguard/wgctrl/wgtypes"
+
+	"vppn/m"
+	"vppn/peer/multicast"
+)
+
+// addRelayPeer adds a public relay peer and marks it Up so it satisfies
+// CanRelay. It does not set a.relay — callers do that explicitly.
+func addRelayPeer(t *testing.T, a *App, vpnIP string, ep netip.AddrPort) *Peer {
+	t.Helper()
+	key := mustKey(t)
+	ip := netip.MustParseAddr(vpnIP)
+	a.onAddPeer(m.Peer{
+		WGPubKey: key,
+		PeerIP:   ip.As4()[3],
+		Addr4:    ep.Addr(),
+		Port:     ep.Port(),
+		Relay:    true,
+	})
+	p := a.peersByKey[key]
+	p.wgPeer.LastHandshakeTime = time.Now()
+	return p
+}
+
+// newTestApp returns a minimal App wired to a fakeWGDevice and fakeControlConn.
+// vpnIP is the local VPN address (e.g. "10.0.0.1").
+// isPublic / isRelay describe the local node's role.
+func newTestApp(t *testing.T, vpnIP string, isPublic, isRelay bool) (*App, *fakeWGDevice, *fakeControlConn) {
+	t.Helper()
+	privKey, err := wgtypes.GeneratePrivateKey()
+	if err != nil {
+		t.Fatalf("generate key: %v", err)
+	}
+	ip := netip.MustParseAddr(vpnIP)
+	dev := &fakeWGDevice{}
+	cc := &fakeControlConn{}
+	a := &App{
+		vpnIP:       ip,
+		vpnNet:      netip.MustParsePrefix("10.0.0.0/24"),
+		privKey:     privKey,
+		pubKey:      privKey.PublicKey(),
+		isPublic:    isPublic,
+		isRelay:     isRelay,
+		dev:         dev,
+		controlConn: cc,
+		peersByKey:  make(map[wgtypes.Key]*Peer),
+		peersByIP:   make(map[netip.Addr]*Peer),
+		scratch:     make([]byte, scratchSize),
+		hubAddCh:    make(chan m.Peer),
+		hubRemoveCh: make(chan wgtypes.Key),
+		pingCh:      make(chan PingEvent),
+		multicastCh: make(chan multicast.Packet),
+	}
+	return a, dev, cc
+}

peer/control/ping.go

@@ -0,0 +1,76 @@
+// Package control implements the VPN-internal peer control protocol.
+// Peers exchange Ping packets over UDP on the VPN control port to maintain
+// liveness and discover external endpoints for direct connection attempts.
+package control
+
+import (
+	"encoding/binary"
+	"fmt"
+	"net/netip"
+)
+
+const (
+	version = 1
+	Size    = 51 // 1 version + 8 PingTS + 6 SrcV4 + 18 SrcV6 + 18 Dst
+)
+
+// Ping is the single control packet type exchanged between VPN peers.
+//
+// In each peer pair, the peer with the lower VPN IP is the client: it sets
+// PingTS and sends pings on a timer. The server echoes PingTS back in its
+// response, allowing the client to compute RTT = now - PingTS.
+//
+// Both client and server populate SrcV4, SrcV6, and Dst on every packet so
+// endpoint information flows in both directions.
+//
+// Dst is the recipient's external endpoint as observed by the sender from the
+// WireGuard handshake source. Zero if the sender has not observed a handshake
+// from the recipient.
+type Ping struct {
+	PingTS int64          // Client ping send time in nanoseconds.
+	SrcV4  netip.AddrPort // Sender's discovered IPv4 address and port.
+	SrcV6  netip.AddrPort // Sender's discovered IPv6 address and port.
+	Dst    netip.AddrPort
+}
+
+// Marshal encodes p into buf (which must be at least Size bytes) and returns
+// buf[:Size]. Taking the buffer lets callers reuse one across sends; every
+// field is written unconditionally so a reused buffer needs no pre-zeroing.
+func (p Ping) Marshal(buf []byte) []byte {
+	buf[0] = version
+	binary.BigEndian.PutUint64(buf[1:9], uint64(p.PingTS))
+	if p.SrcV4.IsValid() {
+		a4 := p.SrcV4.Addr().As4()
+		copy(buf[9:13], a4[:])
+		binary.BigEndian.PutUint16(buf[13:15], p.SrcV4.Port())
+	} else {
+		clear(buf[9:15])
+	}
+	a16 := p.SrcV6.Addr().As16()
+	copy(buf[15:31], a16[:])
+	binary.BigEndian.PutUint16(buf[31:33], p.SrcV6.Port())
+	a16 = p.Dst.Addr().As16()
+	copy(buf[33:49], a16[:])
+	binary.BigEndian.PutUint16(buf[49:51], p.Dst.Port())
+	return buf[:Size]
+}
+
+// Unmarshal decodes a Ping from a fixed-size 51-byte array.
+func Unmarshal(buf [Size]byte) (Ping, error) {
+	if buf[0] != version {
+		return Ping{}, fmt.Errorf("unknown ping version %d", buf[0])
+	}
+	p := Ping{
+		PingTS: int64(binary.BigEndian.Uint64(buf[1:9])),
+	}
+	if addr := netip.AddrFrom4([4]byte(buf[9:13])); !addr.IsUnspecified() {
+		p.SrcV4 = netip.AddrPortFrom(addr, binary.BigEndian.Uint16(buf[13:15]))
+	}
+	if addr := netip.AddrFrom16([16]byte(buf[15:31])); !addr.IsUnspecified() {
+		p.SrcV6 = netip.AddrPortFrom(addr, binary.BigEndian.Uint16(buf[31:33]))
+	}
+	if addr := netip.AddrFrom16([16]byte(buf[33:49])).Unmap(); !addr.IsUnspecified() {
+		p.Dst = netip.AddrPortFrom(addr, binary.BigEndian.Uint16(buf[49:51]))
+	}
+	return p, nil
+}

peer/control/ping_test.go

@@ -0,0 +1,106 @@
+package control_test
+
+import (
+	"net/netip"
+	"testing"
+
+	"vppn/peer/control"
+)
+
+func TestRoundTrip(t *testing.T) {
+	cases := []struct {
+		name string
+		ping control.Ping
+	}{
+		{
+			name: "zero",
+			ping: control.Ping{},
+		},
+		{
+			name: "client ping",
+			ping: control.Ping{
+				PingTS: 1234567890,
+				SrcV4:  netip.MustParseAddrPort("1.2.3.4:51820"),
+				Dst:    netip.MustParseAddrPort("5.6.7.8:51820"),
+			},
+		},
+		{
+			name: "server response",
+			ping: control.Ping{
+				PingTS: 1234567890,
+				SrcV4:  netip.MustParseAddrPort("5.6.7.8:51820"),
+				Dst:    netip.MustParseAddrPort("1.2.3.4:9999"),
+			},
+		},
+		{
+			name: "IPv6 only",
+			ping: control.Ping{
+				PingTS: 999,
+				SrcV6:  netip.MustParseAddrPort("[2001:db8::1]:51820"),
+				Dst:    netip.MustParseAddrPort("[2001:db8::2]:51820"),
+			},
+		},
+		{
+			name: "dual stack",
+			ping: control.Ping{
+				PingTS: 555,
+				SrcV4:  netip.MustParseAddrPort("1.2.3.4:51820"),
+				SrcV6:  netip.MustParseAddrPort("[2001:db8::1]:51820"),
+				Dst:    netip.MustParseAddrPort("5.6.7.8:9999"),
+			},
+		},
+		{
+			name: "no src known",
+			ping: control.Ping{
+				Dst: netip.MustParseAddrPort("5.6.7.8:51820"),
+			},
+		},
+	}
+
+	for _, tc := range cases {
+		t.Run(tc.name, func(t *testing.T) {
+			var buf [control.Size]byte
+			tc.ping.Marshal(buf[:])
+			got, err := control.Unmarshal(buf)
+			if err != nil {
+				t.Fatalf("Unmarshal: %v", err)
+			}
+			if got != tc.ping {
+				t.Fatalf("round-trip mismatch:\n  got  %+v\n  want %+v", got, tc.ping)
+			}
+		})
+	}
+}
+
+func TestUnmarshalBadVersion(t *testing.T) {
+	var buf [control.Size]byte
+	buf[0] = 99
+	if _, err := control.Unmarshal(buf); err == nil {
+		t.Fatal("expected error for unknown version, got nil")
+	}
+}
+
+func TestZeroEncoding(t *testing.T) {
+	var buf [control.Size]byte
+	(control.Ping{}).Marshal(buf[:])
+	for i, b := range buf {
+		if i == 0 {
+			continue // version byte
+		}
+		if b != 0 {
+			t.Fatalf("expected zero encoding at byte %d, got %d", i, b)
+		}
+	}
+}
+
+func TestRoleFor(t *testing.T) {
+	lo := netip.MustParseAddr("10.0.0.1")
+	hi := netip.MustParseAddr("10.0.0.2")
+
+	if control.RoleFor(lo, hi) != control.Client {
+		t.Error("lower IP should be client")
+	}
+	if control.RoleFor(hi, lo) != control.Server {
+		t.Error("higher IP should be server")
+	}
+}

peer/control/role.go

@@ -0,0 +1,22 @@
+package control
+
+import "net/netip"
+
+// Role identifies a peer's role in a ping exchange with a specific remote peer.
+type Role string
+
+const (
+	// Client initiates pings and measures RTT.
+	Client Role = "CLIENT"
+	// Server responds to pings.
+	Server Role = "SERVER"
+)
+
+// RoleFor returns the Role of local relative to remote.
+// The peer with the lower VPN IP is the client.
+func RoleFor(local, remote netip.Addr) Role {
+	if local.Compare(remote) < 0 {
+		return Client
+	}
+	return Server
+}

peer/control_conn.go

@@ -0,0 +1,64 @@
+package peer
+
+import (
+	"log"
+	"net"
+	"net/netip"
+
+	"vppn/peer/control"
+)
+
+var _ ControlConn = (*udpControlConn)(nil)
+
+type udpControlConn struct {
+	conn *net.UDPConn
+}
+
+// newUDPControlConn opens a UDP socket bound to localIP:port.
+func newUDPControlConn(localIP netip.Addr, port uint16) (*udpControlConn, error) {
+	addr := net.UDPAddrFromAddrPort(netip.AddrPortFrom(localIP, port))
+	conn, err := net.ListenUDP("udp4", addr)
+	if err != nil {
+		return nil, err
+	}
+	return &udpControlConn{conn: conn}, nil
+}
+
+func (c *udpControlConn) SendPing(dst netip.AddrPort, ping control.Ping, buf []byte) error {
+	_, err := c.conn.WriteToUDP(ping.Marshal(buf), net.UDPAddrFromAddrPort(dst))
+	return err
+}
+
+// run reads incoming ping packets and forwards them to ch until ctx is done.
+// Call this in a goroutine before starting the App event loop.
+func (c *udpControlConn) run(ch chan<- PingEvent) {
+	var buf [control.Size]byte
+	for {
+		n, src, err := c.conn.ReadFromUDP(buf[:])
+		if err != nil {
+			log.Printf("control read: %v", err)
+			continue
+		}
+
+		if n != control.Size {
+			continue
+		}
+
+		ping, err := control.Unmarshal(buf)
+		if err != nil {
+			log.Printf("control unmarshal: %v", err)
+			continue
+		}
+
+		srcIP, ok := netip.AddrFromSlice(src.IP)
+		if !ok {
+			continue
+		}
+
+		ch <- PingEvent{srcVPNIP: srcIP.Unmap(), ping: ping}
+	}
+}
+
+func (c *udpControlConn) Close() error {
+	return c.conn.Close()
+}

peer/crypto.go

@@ -1,15 +0,0 @@
-package peer
-
-import (
-	"log"
-
-	"golang.zx2c4.com/wireguard/wgctrl/wgtypes"
-)
-
-func generateWGKey() wgtypes.Key {
-	key, err := wgtypes.GeneratePrivateKey()
-	if err != nil {
-		log.Fatalf("Failed to generate WireGuard private key: %v", err)
-	}
-	return key
-}

peer/device.go

@@ -0,0 +1,79 @@
+package peer
+
+import (
+	"errors"
+	"log"
+	"net/netip"
+	"syscall"
+	"time"
+
+	"golang.zx2c4.com/wireguard/wgctrl/wgtypes"
+)
+
+// devRetry calls fn up to 6 times with exponential backoff, retrying on EBUSY
+// (transient netlink contention during WireGuard handshake/rekey). Fatal on
+// any other error.
+func devRetry(vpnIP netip.Addr, op string, fn func() error) {
+	const attempts = 6
+	timeout := 10 * time.Millisecond
+	for i := range attempts {
+		err := fn()
+		if err == nil {
+			return
+		}
+		if errors.Is(err, syscall.EBUSY) && i < attempts-1 {
+			time.Sleep(timeout)
+			timeout *= 2
+			continue
+		}
+		log.Fatalf("%s %v: %v", op, vpnIP, err)
+	}
+}
+
+func (a *App) devPeers() []wgtypes.Peer {
+	peers, err := a.dev.Peers()
+	if err != nil {
+		log.Fatalf("Failed to get peers %v: %v", a.vpnIP, err)
+	}
+	return peers
+}
+
+func (a *App) devAddPeer(p *Peer) {
+	log.Printf("RELAYED: %s - %s ", p.Name, p.VPNIP.String())
+	devRetry(p.VPNIP, "AddPeer", func() error { return a.dev.AddPeer(p.PubKey()) })
+	p.State = StateRelayed
+}
+
+func (a *App) devAddDirect(p *Peer, endpoint netip.AddrPort) {
+	log.Printf("DIRECT: %s - %s @ %s", p.Name, p.VPNIP.String(), endpoint.String())
+	devRetry(p.VPNIP, "AddDirect", func() error { return a.dev.AddDirect(p.PubKey(), endpoint, p.VPNIP) })
+	p.State = StateDirect
+}
+
+func (a *App) devSetRelay(p *Peer, endpoint netip.AddrPort) {
+	log.Printf("RELAY: %s - %s @ %s", p.Name, p.VPNIP.String(), endpoint.String())
+	devRetry(p.VPNIP, "SetRelay", func() error { return a.dev.SetRelay(p.PubKey(), endpoint, a.vpnNet) })
+	p.State = StateDirect // Dirrect connection. The app marks peer as relay.
+}
+
+func (a *App) devPromote(p *Peer) {
+	ep := p.WGEndpoint()
+	if ep.IsValid() {
+		log.Printf("PROMOTED: %s - %s @ %s", p.Name, p.VPNIP.String(), p.WGEndpoint().String())
+	} else {
+		log.Printf("PROMOTED: %s - %s (no IP)", p.Name, p.VPNIP.String())
+	}
+	devRetry(p.VPNIP, "Promote", func() error { return a.dev.Promote(p.PubKey(), p.VPNIP) })
+	p.State = StateDirect
+}
+
+func (a *App) devAddProbe(p *Peer, endpoint netip.AddrPort) {
+	log.Printf("PROBE: %s - %s @ %s", p.Name, p.VPNIP.String(), endpoint.String())
+	devRetry(p.VPNIP, "AddProbe", func() error { return a.dev.AddProbe(p.PubKey(), endpoint) })
+	p.State = StateProbing
+}
+
+func (a *App) devRemove(p *Peer) {
+	log.Printf("REMOVED: %s - %s", p.Name, p.VPNIP.String())
+	devRetry(p.VPNIP, "RemovePeer", func() error { return a.dev.RemovePeer(p.PubKey()) })
+}

peer/fake_control_conn_test.go

@@ -0,0 +1,43 @@
+package peer
+
+import (
+	"net/netip"
+	"testing"
+
+	"vppn/peer/control"
+)
+
+type sentPing struct {
+	Dst  netip.AddrPort
+	Ping control.Ping
+}
+
+type fakeControlConn struct {
+	Sent []sentPing
+}
+
+func (f *fakeControlConn) SendPing(dst netip.AddrPort, ping control.Ping, _ []byte) error {
+	f.Sent = append(f.Sent, sentPing{Dst: dst, Ping: ping})
+	return nil
+}
+
+func (f *fakeControlConn) AssertNone(t *testing.T) {
+	t.Helper()
+	if len(f.Sent) != 0 {
+		t.Fatalf("expected no pings sent, got %d: %v", len(f.Sent), f.Sent)
+	}
+}
+
+func (f *fakeControlConn) AssertSent(t *testing.T, i int, dst netip.AddrPort, ping control.Ping) {
+	t.Helper()
+	if i >= len(f.Sent) {
+		t.Fatalf("no ping at index %d (have %d)", i, len(f.Sent))
+	}
+	got := f.Sent[i]
+	if got.Dst != dst {
+		t.Errorf("ping[%d].Dst = %v, want %v", i, got.Dst, dst)
+	}
+	if got.Ping != ping {
+		t.Errorf("ping[%d].Ping = %+v, want %+v", i, got.Ping, ping)
+	}
+}

peer/fake_wgdevice_test.go

@@ -0,0 +1,123 @@
+package peer
+
+import (
+	"net/netip"
+	"sync"
+	"testing"
+
+	"golang.zx2c4.com/wireguard/wgctrl/wgtypes"
+)
+
+// fakeWGDevice records every call made to it. It is safe to read Calls after
+// the event loop has processed the event under test (single-threaded loop
+// means no extra synchronisation needed, but the mutex guards concurrent test
+// helpers if needed).
+type fakeWGDevice struct {
+	mu    sync.Mutex
+	Calls []fakeCall
+	peers []wgtypes.Peer
+}
+
+type fakeCall struct {
+	Method   string
+	PubKey   wgtypes.Key
+	Endpoint netip.AddrPort
+	VPNiP    netip.Addr
+	Network  netip.Prefix
+}
+
+func (f *fakeWGDevice) record(c fakeCall) {
+	f.mu.Lock()
+	f.Calls = append(f.Calls, c)
+	f.mu.Unlock()
+}
+
+func (f *fakeWGDevice) Name() string { return "wg-test" }
+
+func (f *fakeWGDevice) Peers() ([]wgtypes.Peer, error) {
+	f.mu.Lock()
+	defer f.mu.Unlock()
+	out := make([]wgtypes.Peer, len(f.peers))
+	copy(out, f.peers)
+	return out, nil
+}
+
+func (f *fakeWGDevice) AddPeer(pubKey wgtypes.Key) error {
+	f.record(fakeCall{Method: "AddPeer", PubKey: pubKey})
+	return nil
+}
+
+func (f *fakeWGDevice) AddDirect(pubKey wgtypes.Key, endpoint netip.AddrPort, vpnIP netip.Addr) error {
+	f.record(fakeCall{Method: "AddDirect", PubKey: pubKey, Endpoint: endpoint, VPNiP: vpnIP})
+	return nil
+}
+
+func (f *fakeWGDevice) SetRelay(pubKey wgtypes.Key, endpoint netip.AddrPort, network netip.Prefix) error {
+	f.record(fakeCall{Method: "SetRelay", PubKey: pubKey, Endpoint: endpoint, Network: network})
+	return nil
+}
+
+func (f *fakeWGDevice) AddProbe(pubKey wgtypes.Key, endpoint netip.AddrPort) error {
+	f.record(fakeCall{Method: "AddProbe", PubKey: pubKey, Endpoint: endpoint})
+	return nil
+}
+
+func (f *fakeWGDevice) Promote(pubKey wgtypes.Key, vpnIP netip.Addr) error {
+	f.record(fakeCall{Method: "Promote", PubKey: pubKey, VPNiP: vpnIP})
+	return nil
+}
+
+func (f *fakeWGDevice) RemovePeer(pubKey wgtypes.Key) error {
+	f.record(fakeCall{Method: "RemovePeer", PubKey: pubKey})
+	return nil
+}
+
+// AssertNoCalls fails the test if any dev calls were recorded.
+func (f *fakeWGDevice) AssertNoCalls(t *testing.T) {
+	t.Helper()
+	f.mu.Lock()
+	defer f.mu.Unlock()
+	if len(f.Calls) != 0 {
+		t.Fatalf("unexpected dev calls: %v", f.Calls)
+	}
+}
+
+func (f *fakeWGDevice) AssertAddPeer(t *testing.T, i int, pubKey wgtypes.Key) {
+	t.Helper()
+	f.assertCall(t, i, fakeCall{Method: "AddPeer", PubKey: pubKey})
+}
+
+func (f *fakeWGDevice) AssertAddDirect(t *testing.T, i int, pubKey wgtypes.Key, endpoint netip.AddrPort, vpnIP netip.Addr) {
+	t.Helper()
+	f.assertCall(t, i, fakeCall{Method: "AddDirect", PubKey: pubKey, Endpoint: endpoint, VPNiP: vpnIP})
+}
+
+func (f *fakeWGDevice) AssertSetRelay(t *testing.T, i int, pubKey wgtypes.Key, endpoint netip.AddrPort, network netip.Prefix) {
+	t.Helper()
+	f.assertCall(t, i, fakeCall{Method: "SetRelay", PubKey: pubKey, Endpoint: endpoint, Network: network})
+}
+
+func (f *fakeWGDevice) AssertAddProbe(t *testing.T, i int, pubKey wgtypes.Key, endpoint netip.AddrPort) {
+	t.Helper()
+	f.assertCall(t, i, fakeCall{Method: "AddProbe", PubKey: pubKey, Endpoint: endpoint})
+}
+
+func (f *fakeWGDevice) AssertPromote(t *testing.T, i int, pubKey wgtypes.Key, vpnIP netip.Addr) {
+	t.Helper()
+	f.assertCall(t, i, fakeCall{Method: "Promote", PubKey: pubKey, VPNiP: vpnIP})
+}
+
+func (f *fakeWGDevice) AssertRemovePeer(t *testing.T, i int, pubKey wgtypes.Key) {
+	t.Helper()
+	f.assertCall(t, i, fakeCall{Method: "RemovePeer", PubKey: pubKey})
+}
+
+func (f *fakeWGDevice) assertCall(t *testing.T, i int, c fakeCall) {
+	t.Helper()
+	if len(f.Calls) <= i {
+		t.Fatalf("no call at index %d: %v", i, c)
+	}
+	if c != f.Calls[i] {
+		t.Fatalf("call[%d]: got %v, want %v", i, f.Calls[i], c)
+	}
+}

peer/files.go

@@ -1,97 +0,0 @@
-package peer
-
-import (
-	"encoding/json"
-	"log"
-	"os"
-	"path/filepath"
-	"vppn/m"
-)
-
-type LocalConfig struct {
-	LocalPeerIP byte
-	Network     []byte
-	WGPrivKey   string
-}
-
-func configDir(netName string) string {
-	d, err := os.UserHomeDir()
-	if err != nil {
-		log.Fatalf("Failed to get user home directory: %v", err)
-	}
-	return filepath.Join(d, ".vppn", netName)
-}
-
-func lockFilePath(netName string) string {
-	return filepath.Join(configDir(netName), "__lock__")
-}
-
-func peerConfigPath(netName string) string {
-	return filepath.Join(configDir(netName), "config.json")
-}
-
-func peerStatePath(netName string) string {
-	return filepath.Join(configDir(netName), "state.json")
-}
-
-func statusSocketPath(netName string) string {
-	return filepath.Join(configDir(netName), "status.sock")
-}
-
-func storeJson(x any, outPath string) error {
-	outDir := filepath.Dir(outPath)
-	_ = os.MkdirAll(outDir, 0700)
-
-	tmpPath := outPath + ".tmp"
-	buf, err := json.Marshal(x)
-	if err != nil {
-		return err
-	}
-
-	f, err := os.Create(tmpPath)
-	if err != nil {
-		return err
-	}
-
-	if _, err := f.Write(buf); err != nil {
-		f.Close()
-		return err
-	}
-
-	if err := f.Sync(); err != nil {
-		f.Close()
-		return err
-	}
-
-	if err := f.Close(); err != nil {
-		return err
-	}
-
-	return os.Rename(tmpPath, outPath)
-}
-
-func storePeerConfig(netName string, pc LocalConfig) error {
-	return storeJson(pc, peerConfigPath(netName))
-}
-
-func storeNetworkState(netName string, ps m.NetworkState) error {
-	return storeJson(ps, peerStatePath(netName))
-}
-
-func loadJson(dataPath string, ptr any) error {
-	data, err := os.ReadFile(dataPath)
-	if err != nil {
-		return err
-	}
-
-	return json.Unmarshal(data, ptr)
-}
-
-func loadPeerConfig(netName string) (pc LocalConfig, err error) {
-	return pc, loadJson(peerConfigPath(netName), &pc)
-}
-
-func loadNetworkState(netName string) (ps m.NetworkState, err error) {
-	return ps, loadJson(peerStatePath(netName), &ps)
-}
-

peer/files_test.go

@@ -1,57 +0,0 @@
-package peer
-
-import (
-	"path/filepath"
-	"reflect"
-	"testing"
-)
-
-func TestFilePaths(t *testing.T) {
-	confDir := configDir("netName")
-	if filepath.Base(confDir) != "netName" {
-		t.Fatal(confDir)
-	}
-	if filepath.Base(filepath.Dir(confDir)) != ".vppn" {
-		t.Fatal(confDir)
-	}
-
-	path := peerConfigPath("netName")
-	if path != filepath.Join(confDir, "config.json") {
-		t.Fatal(path)
-	}
-
-	path = peerStatePath("netName")
-	if path != filepath.Join(confDir, "state.json") {
-		t.Fatal(path)
-	}
-}
-
-func TestStoreLoadJson(t *testing.T) {
-	type Object struct {
-		Name  string
-		Age   int
-		Price float64
-	}
-
-	tmpDir := t.TempDir()
-	outPath := filepath.Join(tmpDir, "object.json")
-
-	obj := Object{
-		Name:  "Jason",
-		Age:   22,
-		Price: 123.534,
-	}
-
-	if err := storeJson(obj, outPath); err != nil {
-		t.Fatal(err)
-	}
-
-	obj2 := Object{}
-	if err := loadJson(outPath, &obj2); err != nil {
-		t.Fatal(err)
-	}
-
-	if !reflect.DeepEqual(obj, obj2) {
-		t.Fatal(obj, obj2)
-	}
-}

peer/globals.go

@@ -1,39 +0,0 @@
-package peer
-
-import (
-	"net"
-	"net/netip"
-	"time"
-
-	"golang.zx2c4.com/wireguard/wgctrl"
-	"golang.zx2c4.com/wireguard/wgctrl/wgtypes"
-)
-
-const (
-	broadcastInterval             = 16 * time.Second
-	broadcastErrorTimeoutInterval = 8 * time.Second
-)
-
-var multicastAddr = net.UDPAddrFromAddrPort(netip.AddrPortFrom(
-	netip.AddrFrom4([4]byte{224, 0, 0, 157}),
-	4560))
-
-type Globals struct {
-	LocalConfig // Embed, immutable.
-
-	// WireGuard private key, client, and device name. Immutable after init.
-	WGPrivKey wgtypes.Key
-	WGClient  *wgctrl.Client
-	WGDevName string
-
-	// Local public address (if available). Immutable.
-	LocalAddr      netip.AddrPort
-	LocalAddrValid bool
-}
-
-func NewGlobals(localConfig LocalConfig, localAddr netip.AddrPort) (g Globals) {
-	g.LocalConfig = localConfig
-	g.LocalAddr = localAddr
-	g.LocalAddrValid = localAddr.IsValid()
-	return g
-}

peer/hosts.go

@@ -0,0 +1,128 @@
+package peer
+
+import (
+	"fmt"
+	"log"
+	"net/netip"
+	"os"
+	"sort"
+	"strings"
+	"syscall"
+
+	"git.crumpington.com/lib/go/flock"
+)
+
+const (
+	hostsFile  = "/etc/hosts"
+	hostsBegin = "# BEGIN vppn"
+	hostsEnd   = "# END vppn"
+)
+
+// hostMarkers returns the begin/end marker lines that delimit the managed
+// section for localDomain. The domain is wrapped in parentheses so one domain's
+// marker can never be a prefix of another's (e.g. "net" vs "net2") when
+// multiple vppn instances share /etc/hosts.
+func hostMarkers(localDomain string) (begin, end string) {
+	return hostsBegin + "(" + localDomain + ")", hostsEnd + "(" + localDomain + ")"
+}
+
+// updateHosts rewrites the managed vppn section in /etc/hosts using the
+// current peersByIP map. Peers without a Name are skipped.
+func (a *App) updateHosts() {
+	if a.localDomain == "" {
+		return
+	}
+	if err := updateHosts(hostsFile, a.localDomain, a.peersByIP); err != nil {
+		log.Printf("Failed to update hosts file: %v", err)
+	}
+}
+
+func updateHosts(hostsPath, localDomain string, peers map[netip.Addr]*Peer) error {
+	lockFile, err := flock.Lock(hostsPath + ".vppn.lock")
+	if err != nil {
+		return err
+	}
+	defer lockFile.Close()
+
+	begin, end := hostMarkers(localDomain)
+
+	info, err := os.Stat(hostsPath)
+	if err != nil {
+		return err
+	}
+
+	raw, err := os.ReadFile(hostsPath)
+	if err != nil {
+		return err
+	}
+
+	data := string(raw)
+
+	before := strings.TrimSpace(data)
+	after := ""
+
+	if idxBegin := strings.Index(data, begin); idxBegin != -1 {
+		idxEnd := strings.Index(data[idxBegin:], end)
+		if idxEnd != -1 {
+			after = strings.TrimSpace(data[idxBegin+idxEnd+len(end):])
+		}
+		before = strings.TrimSpace(data[:idxBegin])
+	}
+
+	b := strings.Builder{}
+	b.WriteString(before)
+	b.WriteRune('\n')
+	b.WriteString(after)
+	b.WriteRune('\n')
+	b.WriteRune('\n')
+
+	b.WriteString(begin)
+	b.WriteRune('\n')
+
+	// Collect entries so we can sort by IP for stable output. Pad the IP
+	// column to the width of the widest possible address ("255.255.255.255")
+	// for readability.
+	type entry struct {
+		ip   netip.Addr
+		host string
+	}
+	var entries []entry
+	for ip, p := range peers {
+		if p.Name == "" {
+			continue
+		}
+		entries = append(entries, entry{ip: ip, host: p.Name + "." + localDomain})
+	}
+	sort.Slice(entries, func(i, j int) bool {
+		return entries[i].ip.Less(entries[j].ip)
+	})
+
+	for _, e := range entries {
+		b.WriteString(fmt.Sprintf("%-15s %s\n", e.ip.String(), e.host))
+	}
+
+	b.WriteString(end)
+	b.WriteRune('\n')
+
+	// Write to a temp file in the same directory, then rename over the
+	// original so readers never observe a partial file. Preserve the
+	// original's mode and ownership, since rename replaces the inode.
+	tmpPath := hostsPath + ".vppn.tmp"
+	if err := os.WriteFile(tmpPath, []byte(b.String()), info.Mode().Perm()); err != nil {
+		return err
+	}
+
+	if st, ok := info.Sys().(*syscall.Stat_t); ok {
+		if err := os.Chown(tmpPath, int(st.Uid), int(st.Gid)); err != nil {
+			os.Remove(tmpPath)
+			return err
+		}
+	}
+
+	if err := os.Rename(tmpPath, hostsPath); err != nil {
+		os.Remove(tmpPath)
+		return err
+	}
+
+	return nil
+}

peer/hosts_test.go

@@ -0,0 +1,205 @@
+package peer
+
+import (
+	"net/netip"
+	"os"
+	"path/filepath"
+	"sort"
+	"strings"
+	"testing"
+)
+
+// writeTempHosts creates a temp hosts file with the given content and returns
+// its path.
+func writeTempHosts(t *testing.T, content string) string {
+	t.Helper()
+	path := filepath.Join(t.TempDir(), "hosts")
+	if err := os.WriteFile(path, []byte(content), 0600); err != nil {
+		t.Fatal(err)
+	}
+	return path
+}
+
+// readManagedSection returns the lines between the begin/end markers for the
+// given localDomain, plus everything outside the section ("outside").
+func readManagedSection(t *testing.T, path, localDomain string) (inside, outside []string) {
+	t.Helper()
+	raw, err := os.ReadFile(path)
+	if err != nil {
+		t.Fatal(err)
+	}
+	begin, end := hostMarkers(localDomain)
+
+	inSection := false
+	for _, line := range strings.Split(string(raw), "\n") {
+		switch {
+		case strings.HasPrefix(line, begin):
+			inSection = true
+		case strings.HasPrefix(line, end):
+			inSection = false
+		case inSection:
+			if f := strings.Join(strings.Fields(line), " "); f != "" {
+				inside = append(inside, f)
+			}
+		default:
+			if f := strings.Join(strings.Fields(line), " "); f != "" {
+				outside = append(outside, f)
+			}
+		}
+	}
+	return inside, outside
+}
+
+func peer(name string) *Peer {
+	return &Peer{Name: name}
+}
+
+func TestUpdateHosts_AddsSection(t *testing.T) {
+	path := writeTempHosts(t, "127.0.0.1 localhost\n")
+
+	peers := map[netip.Addr]*Peer{
+		netip.MustParseAddr("10.11.12.1"):  peer("hub"),
+		netip.MustParseAddr("10.11.12.10"): peer("laptop"),
+	}
+
+	if err := updateHosts(path, "mynet.local", peers); err != nil {
+		t.Fatal(err)
+	}
+
+	inside, outside := readManagedSection(t, path, "mynet.local")
+
+	sort.Strings(inside)
+	want := []string{
+		"10.11.12.1 hub.mynet.local",
+		"10.11.12.10 laptop.mynet.local",
+	}
+	if strings.Join(inside, "\n") != strings.Join(want, "\n") {
+		t.Errorf("managed section = %v, want %v", inside, want)
+	}
+
+	if !contains(outside, "127.0.0.1 localhost") {
+		t.Errorf("original content lost; outside = %v", outside)
+	}
+}
+
+func TestUpdateHosts_ReplacesExistingSection(t *testing.T) {
+	path := writeTempHosts(t, "127.0.0.1 localhost\n")
+
+	// First write.
+	first := map[netip.Addr]*Peer{
+		netip.MustParseAddr("10.11.12.1"): peer("hub"),
+	}
+	if err := updateHosts(path, "mynet.local", first); err != nil {
+		t.Fatal(err)
+	}
+
+	// Second write with a different set of peers.
+	second := map[netip.Addr]*Peer{
+		netip.MustParseAddr("10.11.12.20"): peer("phone"),
+	}
+	if err := updateHosts(path, "mynet.local", second); err != nil {
+		t.Fatal(err)
+	}
+
+	inside, outside := readManagedSection(t, path, "mynet.local")
+
+	if len(inside) != 1 || inside[0] != "10.11.12.20 phone.mynet.local" {
+		t.Errorf("section not replaced; inside = %v", inside)
+	}
+	if contains(inside, "10.11.12.1 hub.mynet.local") {
+		t.Errorf("stale entry remained; inside = %v", inside)
+	}
+	if !contains(outside, "127.0.0.1 localhost") {
+		t.Errorf("original content lost; outside = %v", outside)
+	}
+}
+
+func TestUpdateHosts_SkipsEmptyNames(t *testing.T) {
+	path := writeTempHosts(t, "127.0.0.1 localhost\n")
+
+	peers := map[netip.Addr]*Peer{
+		netip.MustParseAddr("10.11.12.1"):  peer("hub"),
+		netip.MustParseAddr("10.11.12.99"): peer(""), // no name
+	}
+	if err := updateHosts(path, "mynet.local", peers); err != nil {
+		t.Fatal(err)
+	}
+
+	inside, _ := readManagedSection(t, path, "mynet.local")
+	if len(inside) != 1 || inside[0] != "10.11.12.1 hub.mynet.local" {
+		t.Errorf("expected only named peer; inside = %v", inside)
+	}
+}
+
+func TestUpdateHosts_Idempotent(t *testing.T) {
+	path := writeTempHosts(t, "127.0.0.1 localhost\n")
+
+	peers := map[netip.Addr]*Peer{
+		netip.MustParseAddr("10.11.12.1"): peer("hub"),
+	}
+	if err := updateHosts(path, "mynet.local", peers); err != nil {
+		t.Fatal(err)
+	}
+	first, err := os.ReadFile(path)
+	if err != nil {
+		t.Fatal(err)
+	}
+	if err := updateHosts(path, "mynet.local", peers); err != nil {
+		t.Fatal(err)
+	}
+	second, err := os.ReadFile(path)
+	if err != nil {
+		t.Fatal(err)
+	}
+	if string(first) != string(second) {
+		t.Errorf("repeated update changed file:\nfirst:\n%s\nsecond:\n%s", first, second)
+	}
+}
+
+// TestUpdateHosts_PrefixDomainsCoexist guards finding 4.4: two domains where
+// one label is a prefix of the other ("net" vs "net2") must each manage their
+// own section without clobbering the other's, even sharing one hosts file.
+func TestUpdateHosts_PrefixDomainsCoexist(t *testing.T) {
+	path := writeTempHosts(t, "127.0.0.1 localhost\n")
+
+	if err := updateHosts(path, "net2.local", map[netip.Addr]*Peer{
+		netip.MustParseAddr("10.0.2.1"): peer("a"),
+	}); err != nil {
+		t.Fatal(err)
+	}
+	if err := updateHosts(path, "net.local", map[netip.Addr]*Peer{
+		netip.MustParseAddr("10.0.1.1"): peer("b"),
+	}); err != nil {
+		t.Fatal(err)
+	}
+
+	// Both sections coexist after writing the prefix domain.
+	if in, _ := readManagedSection(t, path, "net2.local"); len(in) != 1 || in[0] != "10.0.2.1 a.net2.local" {
+		t.Errorf("net2 section clobbered: %v", in)
+	}
+	if in, _ := readManagedSection(t, path, "net.local"); len(in) != 1 || in[0] != "10.0.1.1 b.net.local" {
+		t.Errorf("net section wrong: %v", in)
+	}
+
+	// Re-updating net2 must not disturb the net section.
+	if err := updateHosts(path, "net2.local", map[netip.Addr]*Peer{
+		netip.MustParseAddr("10.0.2.2"): peer("c"),
+	}); err != nil {
+		t.Fatal(err)
+	}
+	if in, _ := readManagedSection(t, path, "net.local"); len(in) != 1 || in[0] != "10.0.1.1 b.net.local" {
+		t.Errorf("net section disturbed by net2 update: %v", in)
+	}
+	if in, _ := readManagedSection(t, path, "net2.local"); len(in) != 1 || in[0] != "10.0.2.2 c.net2.local" {
+		t.Errorf("net2 section not updated: %v", in)
+	}
+}
+
+func contains(ss []string, s string) bool {
+	for _, x := range ss {
+		if x == s {
+			return true
+		}
+	}
+	return false
+}

peer/hub_poller.go

@@ -0,0 +1,153 @@
+package peer
+
+import (
+	"encoding/json"
+	"io"
+	"log"
+	"net/http"
+	"net/netip"
+	"net/url"
+	"time"
+
+	"golang.zx2c4.com/wireguard/wgctrl/wgtypes"
+
+	"vppn/m"
+)
+
+const hubPollInterval = 64 * time.Second
+
+type HubPoller struct {
+	selfVPNIP netip.Addr
+	vpnNet    netip.Prefix
+	hubURL    string
+	apiKey    string
+	statePath string // where the network state cache is persisted
+	addCh     chan<- m.Peer
+	removeCh  chan<- wgtypes.Key
+	known     map[wgtypes.Key]struct{} // pubKeys currently configured
+}
+
+func NewHubPoller(
+	selfVPNIP netip.Addr,
+	vpnNet netip.Prefix,
+	hubURL, apiKey string,
+	statePath string,
+	addCh chan<- m.Peer,
+	removeCh chan<- wgtypes.Key,
+) (*HubPoller, error) {
+	u, err := url.Parse(hubURL)
+	if err != nil {
+		return nil, err
+	}
+	u.Path = "/peer/fetch-state/"
+
+	return &HubPoller{
+		selfVPNIP: selfVPNIP,
+		vpnNet:    vpnNet,
+		hubURL:    u.String(),
+		apiKey:    apiKey,
+		statePath: statePath,
+		addCh:     addCh,
+		removeCh:  removeCh,
+		known:     make(map[wgtypes.Key]struct{}),
+	}, nil
+}
+
+func (hp *HubPoller) Run() {
+	// Prime from the on-disk cache before reaching the hub, so the peer
+	// configures WireGuard from its last known state even if the hub is down.
+	// known starts empty, so this emits every cached peer as an add; the first
+	// real poll then emits only deltas (adds for new peers, removes for gone).
+	if state, err := loadNetworkState(hp.statePath); err == nil {
+		hp.apply(state)
+	}
+
+	hp.poll()
+	for range time.Tick(hubPollInterval) {
+		hp.poll()
+	}
+}
+
+func (hp *HubPoller) poll() {
+	req, err := http.NewRequest(http.MethodGet, hp.hubURL, nil)
+	if err != nil {
+		log.Printf("[HubPoller] build request: %v", err)
+		return
+	}
+	req.SetBasicAuth("", hp.apiKey)
+
+	client := &http.Client{Timeout: 32 * time.Second}
+	resp, err := client.Do(req)
+	if err != nil {
+		log.Printf("[HubPoller] fetch: %v", err)
+		return
+	}
+	defer resp.Body.Close()
+
+	if resp.StatusCode != http.StatusOK {
+		log.Printf("[HubPoller] unexpected status %d", resp.StatusCode)
+		return
+	}
+
+	body, err := io.ReadAll(resp.Body)
+	if err != nil {
+		log.Printf("[HubPoller] read body: %v", err)
+		return
+	}
+
+	var state m.NetworkState
+	if err := json.Unmarshal(body, &state); err != nil {
+		log.Printf("[HubPoller] unmarshal: %v", err)
+		return
+	}
+
+	// Persist only when the state actually changed, to avoid needless writes
+	// on every poll.
+	if hp.apply(state) {
+		if err := saveNetworkState(hp.statePath, state); err != nil {
+			log.Printf("[HubPoller] save state: %v", err)
+		}
+	}
+}
+
+// apply diffs state against the set of known peers, emitting an add for each
+// newly-seen peer and a remove for each that disappeared. It returns true if
+// anything changed. A peer's config is immutable under a stable WG key (the hub
+// has no peer-edit path), so a key already in known needs no re-emit.
+func (hp *HubPoller) apply(state m.NetworkState) (changed bool) {
+	seen := make(map[wgtypes.Key]struct{}, len(hp.known))
+
+	netAddr := hp.vpnNet.Addr().As4()
+
+	for _, p := range state.Peers {
+		if p.WGPubKey == (wgtypes.Key{}) {
+			continue
+		}
+
+		octets := netAddr
+		octets[3] = p.PeerIP
+		vpnIP := netip.AddrFrom4(octets)
+		if vpnIP == hp.selfVPNIP {
+			continue
+		}
+
+		seen[p.WGPubKey] = struct{}{}
+
+		if _, ok := hp.known[p.WGPubKey]; ok {
+			continue
+		}
+		hp.known[p.WGPubKey] = struct{}{}
+		hp.addCh <- p
+		changed = true
+	}
+
+	for key := range hp.known {
+		if _, ok := seen[key]; !ok {
+			delete(hp.known, key)
+			hp.removeCh <- key
+			changed = true
+		}
+	}
+
+	return changed
+}

peer/hub_poller_test.go

@@ -0,0 +1,80 @@
+package peer
+
+import (
+	"net/netip"
+	"testing"
+
+	"golang.zx2c4.com/wireguard/wgctrl/wgtypes"
+
+	"vppn/m"
+)
+
+func testPoller(t *testing.T) (*HubPoller, chan m.Peer, chan wgtypes.Key) {
+	t.Helper()
+	addCh := make(chan m.Peer, 8)
+	removeCh := make(chan wgtypes.Key, 8)
+	hp := &HubPoller{
+		selfVPNIP: netip.MustParseAddr("10.0.0.1"),
+		vpnNet:    netip.MustParsePrefix("10.0.0.0/24"),
+		addCh:     addCh,
+		removeCh:  removeCh,
+		known:     make(map[wgtypes.Key]struct{}),
+	}
+	return hp, addCh, removeCh
+}
+
+func stateWith(key wgtypes.Key, peerIP byte) m.NetworkState {
+	return m.NetworkState{Peers: []m.Peer{{
+		PeerIP:   peerIP,
+		WGPubKey: key,
+	}}}
+}
+
+func TestApply_EmitsAddsAndReportsChange(t *testing.T) {
+	hp, addCh, _ := testPoller(t)
+	key := mustKey(t)
+
+	if changed := hp.apply(stateWith(key, 2)); !changed {
+		t.Fatal("expected changed=true on first apply")
+	}
+	if len(addCh) != 1 {
+		t.Fatalf("expected 1 add, got %d", len(addCh))
+	}
+	if got := <-addCh; got.WGPubKey != key {
+		t.Errorf("add pubkey mismatch")
+	}
+}
+
+func TestApply_NoChangeWhenKnown(t *testing.T) {
+	hp, addCh, _ := testPoller(t)
+	key := mustKey(t)
+
+	hp.apply(stateWith(key, 2))
+	<-addCh // drain initial add
+
+	if changed := hp.apply(stateWith(key, 2)); changed {
+		t.Fatal("expected changed=false when peer already known")
+	}
+	if len(addCh) != 0 {
+		t.Fatalf("expected no re-emit, got %d adds", len(addCh))
+	}
+}
+
+func TestApply_RemovesVanishedPeer(t *testing.T) {
+	hp, addCh, removeCh := testPoller(t)
+	key := mustKey(t)
+
+	hp.apply(stateWith(key, 2))
+	<-addCh
+
+	// Empty state: the peer is gone.
+	if changed := hp.apply(m.NetworkState{}); !changed {
+		t.Fatal("expected changed=true when peer vanishes")
+	}
+	if len(removeCh) != 1 {
+		t.Fatalf("expected 1 remove, got %d", len(removeCh))
+	}
+	if got := <-removeCh; got != key {
+		t.Errorf("remove key mismatch")
+	}
+}

peer/hubpoller.go

@@ -1,148 +0,0 @@
-package peer
-
-import (
-	"encoding/json"
-	"io"
-	"log"
-	"net/http"
-	"net/netip"
-	"net/url"
-	"time"
-	"vppn/m"
-
-	"golang.zx2c4.com/wireguard/wgctrl/wgtypes"
-)
-
-type HubPoller struct {
-	Globals
-	holePunch *HolePunch
-	client    *http.Client
-	req       *http.Request
-	versions  [256]int64
-	netName   string
-}
-
-func NewHubPoller(
-	g Globals,
-	hp *HolePunch,
-	netName,
-	hubURL,
-	apiKey string,
-) (*HubPoller, error) {
-	u, err := url.Parse(hubURL)
-	if err != nil {
-		return nil, err
-	}
-	u.Path = "/peer/fetch-state/"
-
-	client := &http.Client{Timeout: 8 * time.Second}
-
-	req := &http.Request{
-		Method: http.MethodGet,
-		URL:    u,
-		Header: http.Header{},
-	}
-	req.SetBasicAuth("", apiKey)
-
-	return &HubPoller{
-		Globals:   g,
-		holePunch: hp,
-		client:    client,
-		req:       req,
-		netName:   netName,
-	}, nil
-}
-
-func (hp *HubPoller) logf(s string, args ...any) {
-	log.Printf("[HubPoller] "+s, args...)
-}
-
-func (hp *HubPoller) Run() {
-	state, err := loadNetworkState(hp.netName)
-	if err != nil {
-		hp.logf("Failed to load network state: %v", err)
-		hp.logf("Polling hub...")
-		hp.pollHub()
-	} else {
-		hp.applyNetworkState(state)
-	}
-
-	for range time.Tick(64 * time.Second) {
-		hp.pollHub()
-	}
-}
-
-func (hp *HubPoller) pollHub() {
-	var state m.NetworkState
-
-	resp, err := hp.client.Do(hp.req)
-	if err != nil {
-		hp.logf("Failed to fetch peer state: %v", err)
-		return
-	}
-	body, err := io.ReadAll(resp.Body)
-	_ = resp.Body.Close()
-	if err != nil {
-		hp.logf("Failed to read body from hub: %v", err)
-		return
-	}
-
-	if err := json.Unmarshal(body, &state); err != nil {
-		hp.logf("Failed to unmarshal response from hub: %v\n%s", err, body)
-		return
-	}
-
-	if err := storeNetworkState(hp.netName, state); err != nil {
-		hp.logf("Failed to store network state: %v", err)
-	}
-
-	hp.applyNetworkState(state)
-}
-
-func (hp *HubPoller) applyNetworkState(state m.NetworkState) {
-	for i, peer := range state.Peers {
-		if i == int(hp.LocalPeerIP) {
-			continue
-		}
-		if peer != nil && peer.Version == hp.versions[i] {
-			continue
-		}
-		hp.applyPeerConfig(peer)
-		if peer != nil {
-			hp.versions[i] = peer.Version
-		}
-	}
-}
-
-func (hp *HubPoller) applyPeerConfig(peer *m.Peer) {
-	if peer == nil || len(peer.WGPubKey) != wgtypes.KeyLen {
-		return
-	}
-	if len(peer.PublicIP1) == 0 || peer.Port1 == 0 {
-		return
-	}
-
-	pubKey, err := wgtypes.NewKey(peer.WGPubKey)
-	if err != nil {
-		hp.logf("Invalid WG key for peer %d: %v", peer.PeerIP, err)
-		return
-	}
-
-	ip, ok := netip.AddrFromSlice(peer.PublicIP1)
-	if !ok {
-		hp.logf("Invalid public IP for peer %d", peer.PeerIP)
-		return
-	}
-	endpoint := netip.AddrPortFrom(ip.Unmap(), peer.Port1)
-
-	if peer.Relay {
-		if err := applyBaseConfig(hp.WGClient, hp.WGDevName, pubKey, endpoint, hp.Network); err != nil {
-			hp.logf("Failed to update relay config: %v", err)
-		}
-		return
-	}
-
-	if hp.holePunch != nil {
-		hp.holePunch.OnEndpointLearned(peer.PeerIP, pubKey, endpoint, true)
-	}
-}

peer/init.go

@@ -0,0 +1,190 @@
+package peer
+
+import (
+	"bytes"
+	"crypto/rand"
+	"encoding/base64"
+	"encoding/json"
+	"fmt"
+	"net/http"
+	"net/netip"
+	"os"
+
+	"golang.org/x/crypto/nacl/sign"
+	"golang.zx2c4.com/wireguard/wgctrl/wgtypes"
+
+	"vppn/m"
+)
+
+// LocalState is the persisted identity for this peer, written on first run and
+// loaded on every subsequent run.
+type LocalState struct {
+	PrivKey     wgtypes.Key
+	SignKey     [64]byte // nacl/sign Ed25519 private key
+	VPNIP       netip.Addr
+	VPNNet      netip.Prefix
+	WGPort      uint16
+	IsRelay     bool
+	IsPublic    bool
+	LocalDomain string
+}
+
+// localStateJSON is the on-disk representation.
+type localStateJSON struct {
+	PrivKey     string
+	SignKey     string
+	VPNIP       netip.Addr
+	VPNNet      netip.Prefix
+	WGPort      uint16
+	IsRelay     bool
+	IsPublic    bool
+	LocalDomain string
+}
+
+// LoadOrInit loads LocalState from path, or registers with the hub and creates
+// the file if it doesn't exist.
+func LoadOrInit(statePath, hubURL, apiKey string) (LocalState, error) {
+	var state LocalState
+	switch err := loadJSON(statePath, &state); {
+	case err == nil:
+		return state, nil
+	case !os.IsNotExist(err):
+		// File exists but is unreadable/corrupt: surface it rather than
+		// silently regenerating a new identity and re-registering.
+		return LocalState{}, fmt.Errorf("load state: %w", err)
+	}
+
+	privKey, err := wgtypes.GeneratePrivateKey()
+	if err != nil {
+		return LocalState{}, fmt.Errorf("generate key: %w", err)
+	}
+
+	state, err = initFromHub(hubURL, apiKey, privKey)
+	if err != nil {
+		return LocalState{}, err
+	}
+
+	if err := storeJSON(statePath, state); err != nil {
+		return LocalState{}, fmt.Errorf("save state: %w", err)
+	}
+	return state, nil
+}
+
+func initFromHub(hubURL, apiKey string, privKey wgtypes.Key) (LocalState, error) {
+	wgPubKey := privKey.PublicKey()
+
+	signPubKey, signPrivKey, err := sign.GenerateKey(rand.Reader)
+	if err != nil {
+		return LocalState{}, fmt.Errorf("generate sign key: %w", err)
+	}
+
+	body, err := json.Marshal(m.PeerInitArgs{
+		WGPubKey:   wgPubKey[:],
+		SignPubKey: signPubKey[:],
+	})
+	if err != nil {
+		return LocalState{}, fmt.Errorf("json error: %w", err)
+	}
+
+	req, err := http.NewRequest(http.MethodPost, hubURL+"/peer/init/", bytes.NewReader(body))
+	if err != nil {
+		return LocalState{}, err
+	}
+	req.SetBasicAuth("", apiKey)
+	req.Header.Set("Content-Type", "application/json")
+
+	resp, err := http.DefaultClient.Do(req)
+	if err != nil {
+		return LocalState{}, fmt.Errorf("hub init: %w", err)
+	}
+	defer resp.Body.Close()
+
+	if resp.StatusCode != http.StatusOK {
+		return LocalState{}, fmt.Errorf("hub init: HTTP %d", resp.StatusCode)
+	}
+
+	var r m.PeerInitResp
+	if err := json.NewDecoder(resp.Body).Decode(&r); err != nil {
+		return LocalState{}, fmt.Errorf("hub init decode: %w", err)
+	}
+
+	if len(r.Network) != 4 {
+		return LocalState{}, fmt.Errorf("hub init: invalid network %v", r.Network)
+	}
+
+	netAddr := netip.AddrFrom4([4]byte(r.Network))
+	octets := netAddr.As4()
+	octets[3] = r.PeerIP
+	vpnIP := netip.AddrFrom4(octets)
+	vpnNet := netip.PrefixFrom(netAddr, 24)
+
+	var self *m.Peer
+	for i := range r.NetworkState.Peers {
+		if r.NetworkState.Peers[i].PeerIP == r.PeerIP {
+			self = &r.NetworkState.Peers[i]
+			break
+		}
+	}
+	if self == nil {
+		return LocalState{}, fmt.Errorf("hub init: no peer for own IP: %d", r.PeerIP)
+	}
+
+	public := self.IsPublic()
+
+	return LocalState{
+		PrivKey:     privKey,
+		SignKey:     *signPrivKey,
+		VPNIP:       vpnIP,
+		VPNNet:      vpnNet,
+		WGPort:      self.Port,
+		IsRelay:     self.Relay && public,
+		IsPublic:    public,
+		LocalDomain: r.LocalDomain,
+	}, nil
+}
+
+func (s LocalState) MarshalJSON() ([]byte, error) {
+	return json.Marshal(localStateJSON{
+		PrivKey:     base64.StdEncoding.EncodeToString(s.PrivKey[:]),
+		SignKey:     base64.StdEncoding.EncodeToString(s.SignKey[:]),
+		VPNIP:       s.VPNIP,
+		VPNNet:      s.VPNNet,
+		WGPort:      s.WGPort,
+		IsRelay:     s.IsRelay,
+		IsPublic:    s.IsPublic,
+		LocalDomain: s.LocalDomain,
+	})
+}
+
+func (s *LocalState) UnmarshalJSON(data []byte) error {
+	var j localStateJSON
+	if err := json.Unmarshal(data, &j); err != nil {
+		return err
+	}
+	keyBytes, err := base64.StdEncoding.DecodeString(j.PrivKey)
+	if err != nil {
+		return fmt.Errorf("decode key: %w", err)
+	}
+	key, err := wgtypes.NewKey(keyBytes)
+	if err != nil {
+		return fmt.Errorf("invalid key: %w", err)
+	}
+	signKeyBytes, err := base64.StdEncoding.DecodeString(j.SignKey)
+	if err != nil {
+		return fmt.Errorf("decode sign key: %w", err)
+	}
+	if len(signKeyBytes) != 64 {
+		return fmt.Errorf("invalid sign key length: %d", len(signKeyBytes))
+	}
+	*s = LocalState{
+		PrivKey:     key,
+		SignKey:     [64]byte(signKeyBytes),
+		VPNIP:       j.VPNIP,
+		VPNNet:      j.VPNNet,
+		WGPort:      j.WGPort,
+		IsRelay:     j.IsRelay,
+		IsPublic:    j.IsPublic,
+		LocalDomain: j.LocalDomain,
+	}
+	return nil
+}

peer/interfaces.go

@@ -0,0 +1,28 @@
+package peer
+
+import (
+	"net/netip"
+	"vppn/peer/control"
+
+	"golang.zx2c4.com/wireguard/wgctrl/wgtypes"
+)
+
+// WGDevice is the subset of wginterface.Device used by App.
+type WGDevice interface {
+	Name() string
+	Peers() ([]wgtypes.Peer, error)
+	AddPeer(pubKey wgtypes.Key) error
+	AddDirect(pubKey wgtypes.Key, endpoint netip.AddrPort, vpnIP netip.Addr) error
+	SetRelay(pubKey wgtypes.Key, endpoint netip.AddrPort, network netip.Prefix) error
+	AddProbe(pubKey wgtypes.Key, endpoint netip.AddrPort) error
+	Promote(pubKey wgtypes.Key, vpnIP netip.Addr) error
+	RemovePeer(pubKey wgtypes.Key) error
+}
+
+// ControlConn sends pings to peers over the VPN control port.
+// Reading is handled separately via run, which feeds the App's pingCh.
+// buf is a caller-provided scratch buffer (at least control.Size bytes) used to
+// marshal the ping; the caller reuses one across sends.
+type ControlConn interface {
+	SendPing(dst netip.AddrPort, ping control.Ping, buf []byte) error
+}

peer/json.go

@@ -0,0 +1,36 @@
+package peer
+
+import (
+	"encoding/json"
+	"os"
+	"path/filepath"
+)
+
+func loadJSON(path string, target any) error {
+	data, err := os.ReadFile(path)
+	if err != nil {
+		return err
+	}
+	return json.Unmarshal(data, target)
+}
+
+func storeJSON(path string, obj any) error {
+	if err := os.MkdirAll(filepath.Dir(path), 0700); err != nil {
+		return err
+	}
+	data, err := json.MarshalIndent(obj, "", "  ")
+	if err != nil {
+		return err
+	}
+
+	tmpPath := path + ".tmp"
+	if err := os.WriteFile(tmpPath, data, 0600); err != nil {
+		return err
+	}
+
+	if err := os.Rename(tmpPath, path); err != nil {
+		os.Remove(tmpPath)
+		return err
+	}
+	return nil
+}

peer/main.go

@@ -1,209 +0,0 @@
-package peer
-
-import (
-	"encoding/json"
-	"fmt"
-	"log"
-	"net"
-	"net/http"
-	"net/netip"
-	"os"
-	"time"
-)
-
-// Usage:
-//
-// vppn netName run
-// vppn netName status
-func Main2() {
-	printUsage := func() {
-		fmt.Fprintf(os.Stderr, `%s COMMAND [ARGUMENTS...]
-
-Available commands:
-    run
-    status
-    hosts
-`, os.Args[0])
-		os.Exit(1)
-	}
-
-	if len(os.Args) < 2 {
-		printUsage()
-	}
-
-	command := os.Args[1]
-
-	switch command {
-	case "run":
-		main_run()
-	case "status":
-		main_status()
-	case "hosts":
-		main_hosts()
-	default:
-		printUsage()
-	}
-}
-
-// ----------------------------------------------------------------------------
-
-type mainArgs struct {
-	NetName    string
-	HubAddress string
-	APIKey     string
-}
-
-func main_run() {
-	printUsage := func() {
-		fmt.Fprintf(os.Stderr, `Usage: %s run NETWORK_NAME HUB_ADDRESS API_KEY
-
-    NETWORK_NAME
-        Unique name of the network interface created. The network name
-        shouldn't change between invocations of the application.
-
-    HUB_ADDRESS
-        The address of the hub server. This should also contain the scheme, for
-        example https://hub.domain.com/.
-
-    API_KEY
-        The API key assigned to this peer by the hub.
-
-`, os.Args[0])
-		os.Exit(1)
-	}
-
-	if len(os.Args) != 5 {
-		printUsage()
-	}
-
-	args := mainArgs{
-		NetName:    os.Args[2],
-		HubAddress: os.Args[3],
-		APIKey:     os.Args[4],
-	}
-
-	newPeerMain(args).Run()
-}
-
-// ----------------------------------------------------------------------------
-
-func main_status() {
-	printUsage := func() {
-		fmt.Fprintf(os.Stderr, `Usage: %s status NETWORK_NAME
-
-    NETWORK_NAME
-        Unique name of the network interface created.
-
-`, os.Args[0])
-		os.Exit(1)
-	}
-
-	if len(os.Args) != 3 {
-		printUsage()
-	}
-
-	netName := os.Args[2]
-	report := fetchStatusReport(netName)
-
-	fmt.Printf("\n%s Status\n\n", netName)
-
-	if len(report.Network) != 4 {
-		fmt.Println("ERROR: Network isn't 4 bytes.")
-		fmt.Printf("Network: %v\n\n", report.Network)
-	} else {
-		nw := report.Network
-		fmt.Printf("%-8s %d.%d.%d.%d\n", "IP", nw[0], nw[1], nw[2], report.LocalPeerIP)
-		fmt.Printf("%-8s %d.%d.%d.%d/24\n", "Network", nw[0], nw[1], nw[2], nw[3])
-	}
-
-	if report.RelayPeerIP != 0 {
-		fmt.Printf("%-8s %d\n\n", "Relay", report.RelayPeerIP)
-	} else {
-		fmt.Printf("%-8s -\n\n", "Relay")
-	}
-
-	for _, status := range report.Remotes {
-		fmt.Printf("%3d %s\n", status.PeerIP, status.Name)
-		fmt.Printf("    %-11s %v\n", "Up", status.Up)
-
-		pubIP, ok := netip.AddrFromSlice(status.PublicIP)
-		if ok {
-			fmt.Printf("    %-11s %v\n", "Public IP", pubIP)
-		} else {
-			fmt.Printf("    %-11s\n", "Public IP")
-		}
-		fmt.Printf("    %-11s %d\n", "Port", status.Port)
-		fmt.Printf("    %-11s %v\n", "Relay", status.Relay)
-		fmt.Printf("    %-11s %v\n", "Server", status.Server)
-		fmt.Printf("    %-11s %v\n", "Direct", status.Direct)
-		if status.DirectAddr.IsValid() {
-			fmt.Printf("    %-11s %v\n", "Address", status.DirectAddr)
-		}
-		fmt.Println("")
-	}
-}
-
-// ----------------------------------------------------------------------------
-
-func main_hosts() {
-	printUsage := func() {
-		fmt.Fprintf(os.Stderr, `Usage: %s hosts NETWORK_NAME
-
-    NETWORK_NAME
-        Unique name of the network interface created.
-
-`, os.Args[0])
-		os.Exit(1)
-	}
-
-	if len(os.Args) != 3 {
-		printUsage()
-	}
-
-	netName := os.Args[2]
-	state, err := loadNetworkState(netName)
-	if err != nil {
-		log.Fatalf("Failed to load network state: %v", err)
-	}
-
-	config, err := loadPeerConfig(netName)
-	if err != nil {
-		log.Fatalf("Failed to load config: %v", err)
-	}
-
-	nw := config.Network
-	for _, peer := range state.Peers {
-		if peer == nil {
-			continue
-		}
-		fmt.Printf("%d.%d.%d.%d %s\n",
-			nw[0], nw[1], nw[2], peer.PeerIP, peer.Name)
-	}
-	fmt.Println("")
-}
-
-// ----------------------------------------------------------------------------
-
-func fetchStatusReport(netName string) StatusReport {
-	client := http.Client{
-		Transport: &http.Transport{
-			Dial: func(_, _ string) (net.Conn, error) {
-				return net.Dial("unix", statusSocketPath(netName))
-			},
-		},
-		Timeout: 8 * time.Second,
-	}
-
-	getURL := "http://unix" + statusSocketPath(netName)
-	resp, err := client.Get(getURL)
-	if err != nil {
-		log.Fatalf("Failed to get response: %v", err)
-	}
-
-	report := StatusReport{}
-	if err := json.NewDecoder(resp.Body).Decode(&report); err != nil {
-		log.Fatalf("Failed to decode status report: %v", err)
-	}
-
-	return report
-}

peer/mcreader.go

@@ -1,66 +0,0 @@
-package peer
-
-import (
-	"encoding/binary"
-	"fmt"
-	"log"
-	"net"
-	"net/netip"
-	"time"
-
-	"golang.zx2c4.com/wireguard/wgctrl/wgtypes"
-)
-
-func RunMCReader(g Globals, hp *HolePunch, netName string) {
-	for {
-		if err := runMCReaderInner(g, hp, netName); err != nil {
-			log.Printf("[MCReader] %v", err)
-		}
-		time.Sleep(broadcastErrorTimeoutInterval)
-	}
-}
-
-func runMCReaderInner(g Globals, hp *HolePunch, netName string) error {
-	conn, err := net.ListenMulticastUDP("udp", nil, multicastAddr)
-	if err != nil {
-		return fmt.Errorf("bind: %w", err)
-	}
-	defer conn.Close()
-
-	buf := make([]byte, 64)
-	for {
-		conn.SetReadDeadline(time.Now().Add(32 * time.Second))
-		n, src, err := conn.ReadFromUDPAddrPort(buf)
-		if err != nil {
-			return fmt.Errorf("read: %w", err)
-		}
-		if n != beaconLen {
-			continue
-		}
-		handleBeacon(g, hp, netName, buf[:n], src)
-	}
-}
-
-func handleBeacon(g Globals, hp *HolePunch, netName string, beacon []byte, src netip.AddrPort) {
-	peerIPByte := beacon[0]
-	if peerIPByte == g.LocalPeerIP {
-		return
-	}
-
-	pubKey, err := wgtypes.NewKey(beacon[1:33])
-	if err != nil {
-		return
-	}
-
-	// Skip relay peers: probing would replace their /24 AllowedIPs with empty.
-	if state, err := loadNetworkState(netName); err == nil {
-		if p := state.Peers[peerIPByte]; p != nil && p.Relay {
-			return
-		}
-	}
-
-	wgPort := binary.BigEndian.Uint16(beacon[33:35])
-	endpoint := netip.AddrPortFrom(src.Addr().Unmap(), wgPort)
-
-	hp.OnEndpointLearned(peerIPByte, pubKey, endpoint, false)
-}

peer/mcwriter.go

@@ -1,43 +0,0 @@
-package peer
-
-import (
-	"encoding/binary"
-	"fmt"
-	"log"
-	"net"
-	"time"
-)
-
-const beaconLen = 35 // 1 VPN IP byte + 32 WG pubkey + 2 WG listen port
-
-func RunMCWriter(g Globals) {
-	conn, err := net.ListenMulticastUDP("udp", nil, multicastAddr)
-	if err != nil {
-		log.Fatalf("[MCWriter] bind: %v", err)
-	}
-
-	for range time.Tick(broadcastInterval) {
-		beacon, err := buildBeacon(g)
-		if err != nil {
-			log.Printf("[MCWriter] build beacon: %v", err)
-			continue
-		}
-		log.Printf("[MCWriter] Broadcasting on %v...", multicastAddr)
-		if _, err := conn.WriteToUDP(beacon, multicastAddr); err != nil {
-			log.Printf("[MCWriter] write: %v", err)
-		}
-	}
-}
-
-func buildBeacon(g Globals) ([]byte, error) {
-	dev, err := g.WGClient.Device(g.WGDevName)
-	if err != nil {
-		return nil, fmt.Errorf("get WG device: %w", err)
-	}
-	beacon := make([]byte, beaconLen)
-	beacon[0] = g.LocalPeerIP
-	pubKey := g.WGPrivKey.PublicKey()
-	copy(beacon[1:33], pubKey[:])
-	binary.BigEndian.PutUint16(beacon[33:35], uint16(dev.ListenPort))
-	return beacon, nil
-}

peer/multicast/broadcaster.go

@@ -0,0 +1,62 @@
+package multicast
+
+import (
+	"log"
+	"net"
+	"net/netip"
+	"time"
+
+	"golang.zx2c4.com/wireguard/wgctrl/wgtypes"
+)
+
+var addr = net.UDPAddrFromAddrPort(netip.AddrPortFrom(
+	netip.AddrFrom4([4]byte{224, 0, 0, 157}),
+	4560))
+
+func Broadcast(
+	selfVPNIP netip.Addr,
+	pubKey wgtypes.Key,
+	wgPort uint16,
+	signKey *[64]byte,
+) {
+	for {
+		broadcastInner(selfVPNIP, pubKey, wgPort, signKey)
+		time.Sleep(errorTimeout)
+	}
+}
+
+func broadcastInner(selfVPNIP netip.Addr, pubKey wgtypes.Key, wgPort uint16, signKey *[64]byte) {
+	conn, err := net.ListenMulticastUDP("udp", nil, addr)
+	if err != nil {
+		log.Printf("[MCBroadcast] bind: %v", err)
+		return
+	}
+	defer conn.Close()
+
+	buf := make([]byte, BufferSize)
+	packet := Packet{
+		PeerIP:   selfVPNIP.As4()[3],
+		WGPubKey: pubKey,
+		WGPort:   wgPort,
+	}
+
+	// Re-sign on each send so the timestamp is fresh; a stale timestamp would be
+	// dropped by receivers' freshness gate.
+	send := func() error {
+		packet.Timestamp = time.Now().Unix()
+		payload := packet.Marshal(buf, signKey)
+		_, err := conn.WriteToUDP(payload, addr)
+		return err
+	}
+
+	if err := send(); err != nil {
+		log.Printf("[MCBroadcast] write: %v", err)
+	}
+
+	for range time.Tick(broadcastInterval) {
+		if err := send(); err != nil {
+			log.Printf("[MCBroadcast] write: %v", err)
+			return
+		}
+	}
+}

peer/multicast/global.go

@@ -0,0 +1,9 @@
+package multicast
+
+import "time"
+
+const (
+	errorTimeout      = 16 * time.Second
+	broadcastInterval = 16 * time.Second
+	maxPacketAge      = time.Minute
+)

peer/multicast/packet.go

@@ -0,0 +1,54 @@
+package multicast
+
+import (
+	"encoding/binary"
+	"net/netip"
+
+	"golang.org/x/crypto/nacl/sign"
+)
+
+const (
+	BufferSize       = packetSize + SignedPacketSize
+	SignedPacketSize = packetSize + signSize
+	packetSize       = 43
+	signSize         = 64
+)
+
+// Layout:
+//
+// [0]      final octet of the sender's VPN IP
+// [1:33]   WG public key
+// [33:35]  WG listen port (big-endian uint16)
+// [35:43]  send time, Unix seconds (big-endian int64) — freshness/replay gate
+type Packet struct {
+	PeerIP    byte       // Final octet of the sender's VPN IP.
+	WGPubKey  [32]byte   // WG public key.
+	WGPort    uint16     // WG listen port.
+	Timestamp int64      // Unix timestamp.
+	Src       netip.Addr // Source of packet.
+	Signed    []byte     // Raw signed message for verification (incoming packet).
+}
+
+// Marshal the packet into a buffer with prefixed signature.
+func (p Packet) Marshal(buf []byte, signKey *[64]byte) []byte {
+	buf[0] = p.PeerIP
+	copy(buf[1:33], p.WGPubKey[:])
+	binary.BigEndian.PutUint16(buf[33:35], p.WGPort)
+	binary.BigEndian.PutUint64(buf[35:43], uint64(p.Timestamp))
+	return sign.Sign(buf[packetSize:packetSize], buf[:packetSize], signKey)
+}
+
+func (p Packet) Verify(buf []byte, pubKey *[32]byte) bool {
+	_, ok := sign.Open(buf, p.Signed, pubKey)
+	return ok
+}
+
+func Unmarshal(signed []byte) (p Packet) {
+	buf := signed[signSize:]
+	p.PeerIP = buf[0]
+	copy(p.WGPubKey[:], buf[1:33])
+	p.WGPort = binary.BigEndian.Uint16(buf[33:35])
+	p.Timestamp = int64(binary.BigEndian.Uint64(buf[35:43]))
+	p.Signed = signed
+	return
+}

peer/multicast/packet_test.go

@@ -0,0 +1,38 @@
+package multicast
+
+import (
+	"crypto/rand"
+	"testing"
+
+	"golang.org/x/crypto/nacl/sign"
+)
+
+func TestPacket(t *testing.T) {
+	pub, priv, err := sign.GenerateKey(rand.Reader)
+	if err != nil {
+		t.Fatal(err)
+	}
+
+	p := Packet{
+		PeerIP:    10,
+		WGPubKey:  [32]byte{1, 2, 3, 4, 5, 6, 7, 8, 9, 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 0, 1, 2},
+		WGPort:    44,
+		Timestamp: 12948893,
+	}
+
+	buf := make([]byte, BufferSize)
+	signed := p.Marshal(buf, priv)
+	if len(signed) != SignedPacketSize {
+		t.Fatalf("signed length = %d, want %d", len(signed), SignedPacketSize)
+	}
+
+	got := Unmarshal(signed)
+	if got.PeerIP != p.PeerIP || got.WGPubKey != p.WGPubKey ||
+		got.WGPort != p.WGPort || got.Timestamp != p.Timestamp {
+		t.Fatalf("round-trip mismatch:\n got %+v\nwant %+v", got, p)
+	}
+
+	if !got.Verify(nil, pub) {
+		t.Error("signature did not verify")
+	}
+}

peer/multicast/receiver.go

@@ -0,0 +1,61 @@
+package multicast
+
+import (
+	"bytes"
+	"fmt"
+	"log"
+	"net"
+	"net/netip"
+	"time"
+)
+
+func Receiver(vpnNet netip.Prefix, selfVPNIP netip.Addr, ch chan<- Packet) {
+	for {
+		if err := receiver(vpnNet, selfVPNIP, ch); err != nil {
+			log.Printf("[MCReader] %v", err)
+		}
+		time.Sleep(errorTimeout)
+	}
+}
+
+func receiver(vpnNet netip.Prefix, selfVPNIP netip.Addr, ch chan<- Packet) error {
+	selfIP := selfVPNIP.As4()[3]
+
+	conn, err := net.ListenMulticastUDP("udp", nil, addr)
+	if err != nil {
+		return fmt.Errorf("bind: %w", err)
+	}
+	defer conn.Close()
+
+	buf := make([]byte, BufferSize+1) // +1 to detect oversized packets
+
+	for {
+		conn.SetReadDeadline(time.Now().Add(32 * time.Second))
+		n, src, err := conn.ReadFromUDPAddrPort(buf)
+		if err != nil {
+			if ne, ok := err.(net.Error); ok && ne.Timeout() {
+				continue
+			}
+			return fmt.Errorf("read: %w", err)
+		}
+
+		if n != SignedPacketSize {
+			continue
+		}
+
+		packet := Unmarshal(buf[:n])
+
+		if packet.PeerIP == selfIP {
+			continue
+		}
+
+		age := time.Since(time.Unix(packet.Timestamp, 0))
+		if age > maxPacketAge || age < -maxPacketAge {
+			continue
+		}
+
+		packet.Signed = bytes.Clone(packet.Signed)
+		packet.Src = src.Addr().Unmap()
+		ch <- packet
+	}
+}

peer/network_state.go

@@ -0,0 +1,18 @@
+package peer
+
+import "vppn/m"
+
+// loadNetworkState reads a cached network state from disk. Any error (most
+// commonly a missing file on first run) is returned to the caller, which
+// treats it as "no cache available".
+func loadNetworkState(path string) (m.NetworkState, error) {
+	var state m.NetworkState
+	err := loadJSON(path, &state)
+	return state, err
+}
+
+// saveNetworkState writes state to path atomically (see storeJSON), so a crash
+// mid-write cannot leave a corrupt cache.
+func saveNetworkState(path string, state m.NetworkState) error {
+	return storeJSON(path, state)
+}

peer/network_state_test.go

@@ -0,0 +1,56 @@
+package peer
+
+import (
+	"net/netip"
+	"path/filepath"
+	"reflect"
+	"testing"
+
+	"vppn/m"
+)
+
+func TestNetworkState_RoundTrip(t *testing.T) {
+	path := filepath.Join(t.TempDir(), "network.json")
+
+	var sign1 [32]byte
+	copy(sign1[:], []byte("0123456789abcdef0123456789abcdef"))
+
+	state := m.NetworkState{Peers: []m.Peer{
+		{
+			PeerIP:     1,
+			Name:       "hub",
+			Addr4:      netip.MustParseAddr("10.11.12.1"),
+			Port:       51820,
+			Relay:      true,
+			WGPubKey:   mustKey(t),
+			SignPubKey: sign1,
+		},
+		{
+			PeerIP:   10,
+			Name:     "laptop",
+			Addr4:    netip.MustParseAddr("10.11.12.10"),
+			Port:     51820,
+			WGPubKey: mustKey(t),
+		},
+	}}
+
+	if err := saveNetworkState(path, state); err != nil {
+		t.Fatal(err)
+	}
+
+	got, err := loadNetworkState(path)
+	if err != nil {
+		t.Fatal(err)
+	}
+
+	if !reflect.DeepEqual(got, state) {
+		t.Errorf("round-trip mismatch:\n got: %+v\nwant: %+v", got.Peers[1], state.Peers[1])
+	}
+}
+
+func TestNetworkState_LoadMissing(t *testing.T) {
+	path := filepath.Join(t.TempDir(), "does-not-exist.json")
+	if _, err := loadNetworkState(path); err == nil {
+		t.Fatal("expected error loading missing cache, got nil")
+	}
+}

peer/new.go

@@ -0,0 +1,109 @@
+package peer
+
+import (
+	"fmt"
+	"net/netip"
+
+	"golang.zx2c4.com/wireguard/wgctrl/wgtypes"
+
+	"vppn/m"
+	"vppn/peer/multicast"
+	"vppn/peer/wginterface"
+)
+
+// New constructs an App, creates the WireGuard interface, and starts the
+// background goroutines (hub poller, multicast, control conn reader).
+// The caller should invoke Run() to start the event loop.
+func New(
+	state LocalState,
+	hubURL, apiKey string,
+	ifaceName string,
+	localDomain string,
+	networkStatePath string,
+) (*App, error) {
+
+	a4 := state.VPNIP.As4()
+	if err := wginterface.Create(ifaceName, a4[:], 24); err != nil {
+		return nil, fmt.Errorf("create WG interface: %w", err)
+	}
+
+	dev, err := wginterface.Open(ifaceName)
+	if err != nil {
+		_ = wginterface.Delete(ifaceName)
+		return nil, fmt.Errorf("open WG device: %w", err)
+	}
+
+	cc, err := newUDPControlConn(state.VPNIP, ControlPort)
+	if err != nil {
+		_ = dev.Close()
+		_ = wginterface.Delete(ifaceName)
+		return nil, fmt.Errorf("control conn: %w", err)
+	}
+
+	cleanup := func() {
+		_ = cc.Close()
+		_ = dev.Close()
+		_ = wginterface.Delete(ifaceName)
+	}
+
+	if err := dev.Configure(state.PrivKey, int(state.WGPort)); err != nil {
+		cleanup()
+		return nil, fmt.Errorf("configure WG device: %w", err)
+	}
+
+	if state.IsRelay {
+		if err := dev.EnableForwarding(); err != nil {
+			cleanup()
+			return nil, fmt.Errorf("enable forwarding: %w", err)
+		}
+	}
+
+	pingCh := make(chan PingEvent)
+	hubAddCh := make(chan m.Peer)
+	hubRemoveCh := make(chan wgtypes.Key)
+	multicastCh := make(chan multicast.Packet)
+
+	poller, err := NewHubPoller(
+		state.VPNIP,
+		state.VPNNet,
+		hubURL,
+		apiKey,
+		networkStatePath,
+		hubAddCh,
+		hubRemoveCh)
+	if err != nil {
+		cleanup()
+		return nil, fmt.Errorf("hub poller: %w", err)
+	}
+
+	go cc.run(pingCh)
+	go poller.Run()
+
+	if !state.IsPublic {
+		go multicast.Broadcast(state.VPNIP, state.PrivKey.PublicKey(), state.WGPort, &state.SignKey)
+		go multicast.Receiver(state.VPNNet, state.VPNIP, multicastCh)
+	}
+
+	return &App{
+		vpnIP:       state.VPNIP,
+		vpnNet:      state.VPNNet,
+		privKey:     state.PrivKey,
+		pubKey:      state.PrivKey.PublicKey(),
+		isRelay:     state.IsRelay,
+		isPublic:    state.IsPublic,
+		localDomain: localDomain,
+
+		dev:         dev,
+		controlConn: cc,
+
+		peersByKey: make(map[wgtypes.Key]*Peer),
+		peersByIP:  make(map[netip.Addr]*Peer),
+
+		scratch: make([]byte, scratchSize),
+
+		hubAddCh:    hubAddCh,
+		hubRemoveCh: hubRemoveCh,
+		pingCh:      pingCh,
+		multicastCh: multicastCh,
+	}, nil
+}

peer/on_hub.go

@@ -0,0 +1,114 @@
+package peer
+
+import (
+	"log"
+	"math"
+	"net/netip"
+	"time"
+
+	"golang.zx2c4.com/wireguard/wgctrl/wgtypes"
+
+	"vppn/m"
+	"vppn/peer/control"
+)
+
+func (a *App) onAddPeer(p m.Peer) {
+	a.onRemovePeer(p.WGPubKey)
+
+	octets := a.vpnNet.Addr().As4()
+	octets[3] = p.PeerIP
+	vpnIP := netip.AddrFrom4(octets)
+
+	peer := &Peer{
+		wgPeer:     wgtypes.Peer{PublicKey: p.WGPubKey},
+		VPNIP:      vpnIP,
+		Name:       p.Name,
+		IsRelay:    p.Relay,
+		IsPublic:   p.IsPublic(),
+		EndpointV4: p.Endpoint4(),
+		EndpointV6: p.Endpoint6(),
+		RTT:        time.Duration(math.MaxInt64) * time.Nanosecond,
+		Role:       roleFor(a.isPublic, a.vpnIP, p.IsPublic(), vpnIP),
+		SignPubKey: p.SignPubKey,
+	}
+
+	a.peersByKey[p.WGPubKey] = peer
+	a.peersByIP[peer.VPNIP] = peer
+	defer a.updateHosts()
+
+	if !peer.IsPublic {
+		if a.isPublic {
+			// Public nodes accept traffic from non-public peers as soon as they
+			// initiate a handshake. Set /32 AllowedIPs now; WireGuard learns the
+			// endpoint from the incoming handshake automatically.
+			a.devPromote(peer)
+		} else {
+			a.devAddPeer(peer)
+		}
+		return
+	}
+
+	a.devAddDirect(peer, peer.PreferredEndpoint())
+}
+
+func (a *App) onRemovePeer(key wgtypes.Key) {
+	peer, exists := a.peersByKey[key]
+	if !exists {
+		return
+	}
+	a.devRemove(peer)
+	delete(a.peersByKey, key)
+	delete(a.peersByIP, peer.VPNIP)
+	a.updateHosts()
+
+	if peer == a.relay {
+		a.relay = nil
+		a.switchActiveRelay()
+	}
+}
+
+// switchActiveRelay promotes the lowest-latency relay peer to active.
+func (a *App) switchActiveRelay() {
+	if a.relay != nil {
+		// If we have a relay, it's public, so should go back to being a direct
+		// peer - this will convert it's /24 to a /32.
+		a.devAddDirect(a.relay, a.relay.PreferredEndpoint())
+		a.relay = nil
+	}
+
+	var best *Peer
+	for _, p := range a.peersByKey {
+		if !p.CanRelay() {
+			continue
+		}
+
+		if best == nil || p.RTT < best.RTT {
+			best = p
+		}
+	}
+	if best == nil {
+		log.Printf("no relay available")
+		return
+	}
+
+	a.devSetRelay(best, best.PreferredEndpoint())
+	a.relay = best
+}
+
+func preferredEndpoint(v4, v6 netip.AddrPort) netip.AddrPort {
+	// We always prefer v4 since all peers can connect to IPv4 addresses.
+	if v4.IsValid() {
+		return v4
+	}
+	return v6
+}
+
+func roleFor(selfIsPublic bool, selfIP netip.Addr, peerIsPublic bool, peerVPNIP netip.Addr) control.Role {
+	if !selfIsPublic && peerIsPublic {
+		return control.Client
+	}
+	if selfIsPublic && !peerIsPublic {
+		return control.Server
+	}
+	return control.RoleFor(selfIP, peerVPNIP)
+}

peer/on_hub_test.go

@@ -0,0 +1,299 @@
+package peer
+
+import (
+	"net/netip"
+	"testing"
+	"time"
+
+	"golang.zx2c4.com/wireguard/wgctrl/wgtypes"
+
+	"vppn/m"
+)
+
+func mustKey(t *testing.T) wgtypes.Key {
+	t.Helper()
+	k, err := wgtypes.GeneratePrivateKey()
+	if err != nil {
+		t.Fatalf("generate key: %v", err)
+	}
+	return k.PublicKey()
+}
+
+func TestOnAddPeer(t *testing.T) {
+	ep1 := netip.MustParseAddrPort("1.2.3.4:51820")
+	ep2 := netip.MustParseAddrPort("5.6.7.8:51820")
+	peerVPNIP := netip.MustParseAddr("10.0.0.2")
+
+	testCases := []struct {
+		name  string
+		setup func(a *App, key wgtypes.Key)
+		peer  func(key wgtypes.Key) m.Peer
+		check func(t *testing.T, a *App, dev *fakeWGDevice, key wgtypes.Key)
+	}{
+		{
+			name: "non-public peer registered in WG via AddPeer",
+			peer: func(k wgtypes.Key) m.Peer {
+				return m.Peer{WGPubKey: k, PeerIP: 2}
+			},
+			check: func(t *testing.T, a *App, dev *fakeWGDevice, key wgtypes.Key) {
+				p := a.peersByKey[key]
+				if p == nil {
+					t.Fatal("not in peersByKey")
+				}
+				if a.peersByIP[peerVPNIP] == nil {
+					t.Fatal("not in peersByIP")
+				}
+				if p.State != StateRelayed {
+					t.Fatalf("state = %v, want StateRelayed", p.State)
+				}
+				dev.AssertAddPeer(t, 0, key)
+			},
+		},
+		{
+			name: "public peer with endpoint registered via AddDirect",
+			peer: func(k wgtypes.Key) m.Peer {
+				return m.Peer{WGPubKey: k, PeerIP: 2, Addr4: ep1.Addr(), Port: ep1.Port()}
+			},
+			check: func(t *testing.T, a *App, dev *fakeWGDevice, key wgtypes.Key) {
+				p := a.peersByKey[key]
+				if p == nil {
+					t.Fatal("not in peersByKey")
+				}
+				dev.AssertAddDirect(t, 0, p.PubKey(), ep1, p.VPNIP)
+			},
+		},
+		{
+			name: "re-add removes old WG entry before adding new one",
+			setup: func(a *App, key wgtypes.Key) {
+				a.onAddPeer(m.Peer{WGPubKey: key, PeerIP: 2, Addr4: ep1.Addr(), Port: ep1.Port()})
+			},
+			peer: func(k wgtypes.Key) m.Peer {
+				return m.Peer{WGPubKey: k, PeerIP: 2, Addr4: ep2.Addr(), Port: ep2.Port()}
+			},
+			check: func(t *testing.T, a *App, dev *fakeWGDevice, key wgtypes.Key) {
+				if len(dev.Calls) != 2 {
+					t.Fatalf("dev calls = %v, want [RemovePeer, AddDirect]", dev.Calls)
+				}
+				dev.AssertRemovePeer(t, 0, key)
+				dev.AssertAddDirect(t, 1, key, ep2, peerVPNIP)
+				if len(a.peersByKey) != 1 || len(a.peersByIP) != 1 {
+					t.Errorf("maps: peersByKey=%d peersByIP=%d, want 1 each", len(a.peersByKey), len(a.peersByIP))
+				}
+			},
+		},
+	}
+
+	for _, tc := range testCases {
+		t.Run(tc.name, func(t *testing.T) {
+			a, dev, _ := newTestApp(t, "10.0.0.1", false, false)
+			key := mustKey(t)
+			if tc.setup != nil {
+				tc.setup(a, key)
+				dev.Calls = nil
+			}
+			a.onAddPeer(tc.peer(key))
+			tc.check(t, a, dev, key)
+		})
+	}
+}
+
+func TestOnRemovePeer(t *testing.T) {
+	ep1 := netip.MustParseAddrPort("1.2.3.4:51820")
+	ep2 := netip.MustParseAddrPort("5.6.7.8:51820")
+
+	testCases := []struct {
+		name  string
+		setup func(t *testing.T, a *App) wgtypes.Key // returns the key to remove
+		check func(t *testing.T, a *App, dev *fakeWGDevice)
+	}{
+		{
+			name: "unknown key is a no-op",
+			setup: func(t *testing.T, a *App) wgtypes.Key {
+				return mustKey(t)
+			},
+			check: func(t *testing.T, a *App, dev *fakeWGDevice) {
+				dev.AssertNoCalls(t)
+				if len(a.peersByKey) != 0 {
+					t.Errorf("peersByKey should be empty")
+				}
+			},
+		},
+		{
+			name: "StateRelayed peer removed from maps with RemovePeer",
+			setup: func(t *testing.T, a *App) wgtypes.Key {
+				key := mustKey(t)
+				a.onAddPeer(m.Peer{WGPubKey: key, PeerIP: 2})
+				return key
+			},
+			check: func(t *testing.T, a *App, dev *fakeWGDevice) {
+				if len(dev.Calls) != 1 {
+					t.Fatalf("dev calls = %v, want [RemovePeer]", dev.Calls)
+				}
+				dev.AssertRemovePeer(t, 0, dev.Calls[0].PubKey)
+				if len(a.peersByKey) != 0 || len(a.peersByIP) != 0 {
+					t.Errorf("maps should be empty after remove")
+				}
+			},
+		},
+		{
+			name: "StateDirect peer removed from maps with RemovePeer",
+			setup: func(t *testing.T, a *App) wgtypes.Key {
+				key := mustKey(t)
+				a.onAddPeer(m.Peer{WGPubKey: key, PeerIP: 2, Addr4: ep1.Addr(), Port: ep1.Port()})
+				return key
+			},
+			check: func(t *testing.T, a *App, dev *fakeWGDevice) {
+				if len(dev.Calls) != 1 {
+					t.Fatalf("dev calls = %v, want [RemovePeer]", dev.Calls)
+				}
+				dev.AssertRemovePeer(t, 0, dev.Calls[0].PubKey)
+				if len(a.peersByKey) != 0 || len(a.peersByIP) != 0 {
+					t.Errorf("maps should be empty after remove")
+				}
+			},
+		},
+		{
+			name: "removing active relay with no backup clears relay field",
+			setup: func(t *testing.T, a *App) wgtypes.Key {
+				relay := addRelayPeer(t, a, "10.0.0.10", ep1)
+				a.relay = relay
+				return relay.PubKey()
+			},
+			check: func(t *testing.T, a *App, dev *fakeWGDevice) {
+				if len(dev.Calls) != 1 {
+					t.Fatalf("dev calls = %v, want [RemovePeer]", dev.Calls)
+				}
+				dev.AssertRemovePeer(t, 0, dev.Calls[0].PubKey)
+				if a.relay != nil {
+					t.Errorf("relay should be nil after removing only relay")
+				}
+			},
+		},
+		{
+			name: "removing active relay elects backup via SetRelay",
+			setup: func(t *testing.T, a *App) wgtypes.Key {
+				relay1 := addRelayPeer(t, a, "10.0.0.10", ep1)
+				addRelayPeer(t, a, "10.0.0.11", ep2)
+				a.relay = relay1
+				return relay1.PubKey()
+			},
+			check: func(t *testing.T, a *App, dev *fakeWGDevice) {
+				if len(dev.Calls) != 2 {
+					t.Fatalf("dev calls = %v, want [RemovePeer, SetRelay]", dev.Calls)
+				}
+				dev.AssertRemovePeer(t, 0, dev.Calls[0].PubKey)
+				dev.AssertSetRelay(t, 1, dev.Calls[1].PubKey, ep2, a.vpnNet)
+				if a.relay == nil {
+					t.Errorf("relay should be set to backup after failover")
+				}
+			},
+		},
+	}
+
+	for _, tc := range testCases {
+		t.Run(tc.name, func(t *testing.T) {
+			a, dev, _ := newTestApp(t, "10.0.0.1", false, false)
+			key := tc.setup(t, a)
+			dev.Calls = nil
+			a.onRemovePeer(key)
+			tc.check(t, a, dev)
+		})
+	}
+}
+
+func TestSwitchActiveRelay(t *testing.T) {
+	ep1 := netip.MustParseAddrPort("1.2.3.4:51820")
+	ep2 := netip.MustParseAddrPort("5.6.7.8:51820")
+
+	testCases := []struct {
+		name  string
+		setup func(t *testing.T, a *App)
+		check func(t *testing.T, a *App, dev *fakeWGDevice)
+	}{
+		{
+			name:  "no candidates leaves relay nil",
+			setup: func(t *testing.T, a *App) {},
+			check: func(t *testing.T, a *App, dev *fakeWGDevice) {
+				dev.AssertNoCalls(t)
+				if a.relay != nil {
+					t.Error("relay should be nil")
+				}
+			},
+		},
+		{
+			name: "single candidate elected via SetRelay",
+			setup: func(t *testing.T, a *App) {
+				addRelayPeer(t, a, "10.0.0.10", ep1)
+			},
+			check: func(t *testing.T, a *App, dev *fakeWGDevice) {
+				if len(dev.Calls) != 1 {
+					t.Fatalf("dev calls = %v, want [SetRelay]", dev.Calls)
+				}
+				dev.AssertSetRelay(t, 0, dev.Calls[0].PubKey, ep1, a.vpnNet)
+				if a.relay == nil {
+					t.Error("relay should be set")
+				}
+			},
+		},
+		{
+			name: "measured RTT beats zero RTT",
+			setup: func(t *testing.T, a *App) {
+				r1 := addRelayPeer(t, a, "10.0.0.10", ep1)
+				r1.RTT = 10 * time.Millisecond
+				addRelayPeer(t, a, "10.0.0.11", ep2) // RTT stays MaxInt64 (unmeaured)
+			},
+			check: func(t *testing.T, a *App, dev *fakeWGDevice) {
+				if len(dev.Calls) != 1 {
+					t.Fatalf("dev calls = %v, want [SetRelay]", dev.Calls)
+				}
+				dev.AssertSetRelay(t, 0, dev.Calls[0].PubKey, ep1, a.vpnNet)
+			},
+		},
+		{
+			name: "lower RTT wins",
+			setup: func(t *testing.T, a *App) {
+				r1 := addRelayPeer(t, a, "10.0.0.10", ep1)
+				r1.RTT = 5 * time.Millisecond
+				r2 := addRelayPeer(t, a, "10.0.0.11", ep2)
+				r2.RTT = 20 * time.Millisecond
+			},
+			check: func(t *testing.T, a *App, dev *fakeWGDevice) {
+				if len(dev.Calls) != 1 {
+					t.Fatalf("dev calls = %v, want [SetRelay]", dev.Calls)
+				}
+				dev.AssertSetRelay(t, 0, dev.Calls[0].PubKey, ep1, a.vpnNet)
+			},
+		},
+		{
+			name: "stale relay demoted to direct before backup elected",
+			setup: func(t *testing.T, a *App) {
+				old := addRelayPeer(t, a, "10.0.0.10", ep1)
+				old.wgPeer.LastHandshakeTime = time.Time{} // stale — triggers switch from onTick
+				a.relay = old
+				addRelayPeer(t, a, "10.0.0.11", ep2)
+			},
+			check: func(t *testing.T, a *App, dev *fakeWGDevice) {
+				if len(dev.Calls) != 2 {
+					t.Fatalf("dev calls = %v, want [AddDirect, SetRelay]", dev.Calls)
+				}
+				if dev.Calls[0].Method != "AddDirect" || dev.Calls[0].Endpoint != ep1 {
+					t.Errorf("call[0]: got %v, want AddDirect with ep1", dev.Calls[0])
+				}
+				dev.AssertSetRelay(t, 1, dev.Calls[1].PubKey, ep2, a.vpnNet)
+				if a.relay == nil || a.relay.EndpointV4 != ep2 {
+					t.Error("relay should be the backup peer")
+				}
+			},
+		},
+	}
+
+	for _, tc := range testCases {
+		t.Run(tc.name, func(t *testing.T) {
+			a, dev, _ := newTestApp(t, "10.0.0.1", false, false)
+			tc.setup(t, a)
+			dev.Calls = nil
+			a.switchActiveRelay()
+			tc.check(t, a, dev)
+		})
+	}
+}

peer/on_multicast.go

@@ -0,0 +1,51 @@
+package peer
+
+import (
+	"net/netip"
+
+	"golang.zx2c4.com/wireguard/wgctrl/wgtypes"
+
+	"vppn/peer/multicast"
+)
+
+func (a *App) onMulticastDiscovery(pkt multicast.Packet) {
+	if a.isPublic {
+		return
+	}
+
+	// Locate the sender peer by its VPN IP (final octet carried in the beacon).
+	octets := a.vpnNet.Addr().As4()
+	octets[3] = pkt.PeerIP
+	vpnIP := netip.AddrFrom4(octets)
+
+	peer, ok := a.peersByIP[vpnIP]
+	if !ok || peer.IsPublic || peer.State == StateDirect {
+		return
+	}
+
+	// Authenticate the beacon against the peer's known sign key. scratch[:0]
+	// gives sign.Open an empty-but-capacity buffer to decode into.
+	if !pkt.Verify(a.scratch[:0], &peer.SignPubKey) {
+		return
+	}
+
+	// The beacon is authentic but must also advertise the WG key the hub gave
+	// us for this peer; otherwise it's inconsistent — drop it.
+	if wgtypes.Key(pkt.WGPubKey) != peer.PubKey() {
+		return
+	}
+
+	endpoint := netip.AddrPortFrom(pkt.Src, pkt.WGPort)
+	if !endpoint.IsValid() {
+		return
+	}
+
+	var v4, v6 netip.AddrPort
+	if pkt.Src.Is4() {
+		v4 = endpoint
+	} else {
+		v6 = endpoint
+	}
+
+	a.addProbe(peer, v4, v6)
+}

peer/on_ping.go

@@ -0,0 +1,58 @@
+package peer
+
+import (
+	"net/netip"
+	"time"
+
+	"vppn/peer/control"
+)
+
+func (a *App) onPing(e PingEvent) {
+	peer, ok := a.peersByIP[e.srcVPNIP]
+	if !ok {
+		// TODO: Log here.
+		return
+	}
+
+	now := time.Now()
+
+	// If we're the server, respond - this is always necessary as it's used to
+	// know if peers are up or down.
+	if peer.Role == control.Server {
+		a.sendPing(peer, e.ping.PingTS)
+	}
+
+	// Compute RTT from server echo.
+	if peer.Role == control.Client {
+		peer.RTT = now.Sub(time.Unix(0, e.ping.PingTS))
+	}
+
+	// If we're public, nothing more to do.
+	if a.isPublic {
+		return
+	}
+
+	// We can only learn our own endpoint from directly-connected peers — Dst
+	// is the sender's observation of our WG handshake source.
+	if peer.State == StateDirect {
+		if dst := e.ping.Dst; dst.IsValid() {
+			if dst.Addr().Is4() {
+				a.selfV4 = dst
+			} else {
+				a.selfV6 = dst
+			}
+		}
+		return
+	}
+
+	a.addProbe(peer, e.ping.SrcV4, e.ping.SrcV6)
+}
+
+func (a *App) addProbe(peer *Peer, v4, v6 netip.AddrPort) {
+	endpoint := preferredEndpoint(v4, v6)
+	if !endpoint.IsValid() || endpoint == peer.PreferredEndpoint() {
+		return
+	}
+	peer.UpdateEndpoints(v4, v6)
+	a.devAddProbe(peer, endpoint)
+}

peer/on_tick.go

@@ -0,0 +1,52 @@
+package peer
+
+import (
+	"log"
+	"time"
+
+	"vppn/peer/control"
+	"vppn/peer/wginterface"
+)
+
+func (a *App) onTick() {
+	wgPeers := a.devPeers()
+
+	now := time.Now().UnixNano()
+
+	for _, wgPeer := range wgPeers {
+		p, ok := a.peersByKey[wgPeer.PublicKey]
+		if !ok {
+			log.Printf("Wireguard peer not in index, removing: %v", wgPeer)
+			a.devRemove(&Peer{wgPeer: wgPeer})
+			continue
+		}
+		p.wgPeer = wgPeer
+
+		// Send pings to peers where we're the client.
+		if p.Role == control.Client {
+			a.sendPing(p, now)
+		}
+
+		switch p.State {
+		case StateProbing:
+			// Promote probing peers to direct once alive (direct path confirmed
+			// working).
+			if time.Since(p.LastHandshakeTime()) < 2*wginterface.ProbeKeepalive {
+				a.devPromote(p)
+			}
+
+		case StateDirect:
+			if p.IsPublic || a.isPublic || p.Up() {
+				break
+			}
+			// Stale non-public direct peer: demote to probing so WireGuard
+			// resumes handshake attempts on the direct path.
+			a.devAddProbe(p, p.WGEndpoint())
+		}
+	}
+
+	// Ensure we have a live relay (if we're not public).
+	if !a.isPublic && (a.relay == nil || !a.relay.Up()) {
+		a.switchActiveRelay()
+	}
+}

peer/peer.go

@@ -1,220 +0,0 @@
-package peer
-
-import (
-	"bytes"
-	"encoding/json"
-	"io"
-	"log"
-	"net/http"
-	"net/netip"
-	"net/url"
-	"os"
-	"vppn/m"
-
-	"git.crumpington.com/lib/go/flock"
-	"golang.zx2c4.com/wireguard/wgctrl/wgtypes"
-)
-
-type peerMain struct {
-	Globals
-	netName          string
-	holePunch        *HolePunch
-	controlServer    *ControlServer
-	endpointReporter *EndpointReporter // non-nil on relay peers only
-	hubPoller        *HubPoller
-	lockFile         *os.File
-}
-
-func newPeerMain(args mainArgs) *peerMain {
-	logf := func(s string, args ...any) {
-		log.Printf("[Main] "+s, args...)
-	}
-
-	if err := os.MkdirAll(configDir(args.NetName), 0700); err != nil {
-		log.Fatalf("Failed to create config directory: %v", err)
-	}
-
-	lockFile, err := flock.TryLock(lockFilePath(args.NetName))
-	if err != nil {
-		log.Fatalf("Failed to open lock file: %v", err)
-	}
-	if lockFile == nil {
-		log.Fatalf("Failed to obtain file lock.")
-	}
-
-	config, err := loadPeerConfig(args.NetName)
-	if err != nil {
-		logf("Failed to load configuration: %v", err)
-		logf("Initializing...")
-		initPeerWithHub(args)
-
-		config, err = loadPeerConfig(args.NetName)
-		if err != nil {
-			log.Fatalf("Failed to load configuration: %v", err)
-		}
-	}
-
-	state, err := loadNetworkState(args.NetName)
-	if err != nil {
-		log.Fatalf("Failed to load network state: %v", err)
-	}
-
-	wgPrivKey, err := wgtypes.ParseKey(config.WGPrivKey)
-	if err != nil {
-		log.Fatalf("Failed to parse WireGuard private key: %v", err)
-	}
-
-	localPeer := state.Peers[config.LocalPeerIP]
-	var listenPort int
-	if localPeer != nil {
-		listenPort = int(localPeer.Port1)
-	}
-
-	vpnIP := netip.AddrFrom4([4]byte{
-		config.Network[0],
-		config.Network[1],
-		config.Network[2],
-		config.LocalPeerIP,
-	})
-
-	wgClient, err := createWGDevice(args.NetName, wgPrivKey, listenPort, vpnIP, config.Network)
-	if err != nil {
-		log.Fatalf("Failed to create WireGuard device: %v", err)
-	}
-
-	for _, p := range state.Peers {
-		if p == nil || !p.Relay || p.PeerIP == config.LocalPeerIP {
-			continue
-		}
-		if len(p.WGPubKey) != wgtypes.KeyLen || len(p.PublicIP1) == 0 || p.Port1 == 0 {
-			continue
-		}
-		relayPubKey, err := wgtypes.NewKey(p.WGPubKey)
-		if err != nil {
-			logf("Invalid relay WG key: %v", err)
-			continue
-		}
-		relayIP, ok := netip.AddrFromSlice(p.PublicIP1)
-		if !ok {
-			continue
-		}
-		relayEndpoint := netip.AddrPortFrom(relayIP.Unmap(), p.Port1)
-		if err := applyBaseConfig(wgClient, args.NetName, relayPubKey, relayEndpoint, config.Network); err != nil {
-			logf("Failed to apply relay base config: %v", err)
-		}
-		break
-	}
-
-	g := NewGlobals(config, netip.AddrPort{})
-	g.WGPrivKey = wgPrivKey
-	g.WGClient = wgClient
-	g.WGDevName = args.NetName
-
-	holePunch := NewHolePunch(g)
-
-	controlServer, err := NewControlServer(g, holePunch, args.NetName)
-	if err != nil {
-		log.Fatalf("Failed to create control server: %v", err)
-	}
-
-	var endpointReporter *EndpointReporter
-	if localPeer != nil && localPeer.Relay {
-		if err := enableForwarding(args.NetName); err != nil {
-			log.Fatalf("Failed to enable IP forwarding: %v", err)
-		}
-		endpointReporter = NewEndpointReporter(g, controlServer, args.NetName)
-	}
-
-	hubPoller, err := NewHubPoller(g, holePunch, args.NetName, args.HubAddress, args.APIKey)
-	if err != nil {
-		log.Fatalf("Failed to create hub poller: %v", err)
-	}
-
-	go runStatusServer(g, statusSocketPath(args.NetName))
-
-	return &peerMain{
-		Globals:          g,
-		netName:          args.NetName,
-		holePunch:        holePunch,
-		controlServer:    controlServer,
-		endpointReporter: endpointReporter,
-		hubPoller:        hubPoller,
-		lockFile:         lockFile,
-	}
-}
-
-func (p *peerMain) Run() {
-	go p.controlServer.Run()
-	if p.endpointReporter != nil {
-		go p.endpointReporter.Run()
-	}
-	go RunMCWriter(p.Globals)
-	go RunMCReader(p.Globals, p.holePunch, p.netName)
-	go p.hubPoller.Run()
-	select {}
-}
-
-func initPeerWithHub(args mainArgs) {
-	privKey := generateWGKey()
-	pubKey := privKey.PublicKey()
-
-	initURL, err := url.Parse(args.HubAddress)
-	if err != nil {
-		log.Fatalf("Failed to parse hub URL: %v", err)
-	}
-	initURL.Path = "/peer/init/"
-
-	initArgs := m.PeerInitArgs{
-		WGPubKey: pubKey[:],
-	}
-
-	buf := &bytes.Buffer{}
-	if err := json.NewEncoder(buf).Encode(initArgs); err != nil {
-		log.Fatalf("Failed to encode init args: %v", err)
-	}
-
-	req, err := http.NewRequest(http.MethodPost, initURL.String(), buf)
-	if err != nil {
-		log.Fatalf("Failed to construct request: %v", err)
-	}
-	req.SetBasicAuth("", args.APIKey)
-
-	resp, err := http.DefaultClient.Do(req)
-	if err != nil {
-		log.Fatalf("Failed to init with hub: %v", err)
-	}
-	defer resp.Body.Close()
-
-	data, err := io.ReadAll(resp.Body)
-	if err != nil {
-		log.Fatalf("Failed to read response body: %v", err)
-	}
-
-	if resp.StatusCode == http.StatusConflict {
-		log.Fatalf("WireGuard key already registered (HTTP 409). Delete and re-create the peer to re-register.")
-	}
-	if resp.StatusCode != http.StatusOK {
-		log.Fatalf("Hub returned unexpected status %d: %s", resp.StatusCode, data)
-	}
-
-	initResp := m.PeerInitResp{}
-	if err := json.Unmarshal(data, &initResp); err != nil {
-		log.Fatalf("Failed to parse configuration: %v\n%s", err, data)
-	}
-
-	config := LocalConfig{
-		LocalPeerIP: initResp.PeerIP,
-		Network:     initResp.Network,
-		WGPrivKey:   privKey.String(),
-	}
-
-	if err := storeNetworkState(args.NetName, initResp.NetworkState); err != nil {
-		log.Fatalf("Failed to store network state: %v", err)
-	}
-
-	if err := storePeerConfig(args.NetName, config); err != nil {
-		log.Fatalf("Failed to store configuration: %v", err)
-	}
-
-	log.Print("Initialization successful.")
-}

peer/ping.go

@@ -0,0 +1,21 @@
+package peer
+
+import (
+	"log"
+	"net/netip"
+
+	"vppn/peer/control"
+)
+
+func (a *App) sendPing(p *Peer, ts int64) {
+	ping := control.Ping{
+		PingTS: ts,
+		SrcV4:  a.selfV4,
+		SrcV6:  a.selfV6,
+		Dst:    p.WGEndpoint(),
+	}
+	dst := netip.AddrPortFrom(p.VPNIP, ControlPort)
+	if err := a.controlConn.SendPing(dst, ping, a.scratch); err != nil {
+		log.Printf("sendPing %v: %v", p.VPNIP, err)
+	}
+}

peer/remote.go

@@ -0,0 +1,75 @@
+package peer
+
+import (
+	"net/netip"
+	"time"
+
+	"golang.zx2c4.com/wireguard/wgctrl/wgtypes"
+
+	"vppn/peer/control"
+	"vppn/peer/wginterface"
+)
+
+type PeerState string
+
+const (
+	StateRelayed = PeerState("RELAY")
+	StateProbing = PeerState("PROBE")
+	StateDirect  = PeerState("DIRECT")
+)
+
+type Peer struct {
+	wgPeer     wgtypes.Peer
+	VPNIP      netip.Addr     // VPN IP address.
+	Name       string         // Human-readable DNS label.
+	IsRelay    bool           // Peer is a relay.
+	IsPublic   bool           // Peer has a public IP.
+	EndpointV4 netip.AddrPort // Reported IPv4 endpoint.
+	EndpointV6 netip.AddrPort // Reported IPv6 endpoint.
+	RTT        time.Duration  // Round-trip time.
+	State      PeerState      // Current routing state; updated on each devXxx call.
+	Role       control.Role   // Client initiates pings; server responds.
+	SignPubKey [32]byte       // nacl/sign public key for verifying multicast beacons.
+}
+
+// PubKey is the wireguard public key.
+func (p *Peer) PubKey() wgtypes.Key {
+	return p.wgPeer.PublicKey
+}
+
+func (p *Peer) WGEndpoint() netip.AddrPort {
+	ep := p.wgPeer.Endpoint
+	if ep == nil {
+		return netip.AddrPort{}
+	}
+	addr, ok := netip.AddrFromSlice(ep.IP)
+	if !ok {
+		return netip.AddrPort{}
+	}
+	return netip.AddrPortFrom(addr.Unmap(), uint16(ep.Port))
+}
+
+func (p *Peer) LastHandshakeTime() time.Time {
+	return p.wgPeer.LastHandshakeTime
+}
+
+func (p *Peer) Up() bool {
+	return time.Since(p.wgPeer.LastHandshakeTime) < wginterface.SessionTimeout
+}
+
+func (p *Peer) CanRelay() bool {
+	return p.IsRelay && p.Up()
+}
+
+func (p *Peer) PreferredEndpoint() netip.AddrPort {
+	return preferredEndpoint(p.EndpointV4, p.EndpointV6)
+}
+
+func (p *Peer) UpdateEndpoints(v4, v6 netip.AddrPort) {
+	if v4.IsValid() {
+		p.EndpointV4 = v4
+	}
+	if v6.IsValid() {
+		p.EndpointV6 = v6
+	}
+}

peer/statusserver.go

@@ -1,50 +0,0 @@
-package peer
-
-import (
-	"encoding/json"
-	"log"
-	"net"
-	"net/http"
-	"net/netip"
-	"os"
-)
-
-type StatusReport struct {
-	LocalPeerIP byte
-	Network     []byte
-	RelayPeerIP byte
-	Remotes     []RemoteStatus
-}
-
-type RemoteStatus struct {
-	PeerIP     byte
-	Up         bool
-	Name       string
-	PublicIP   []byte
-	Port       uint16
-	Relay      bool
-	Server     bool
-	Direct     bool
-	DirectAddr netip.AddrPort
-}
-
-func runStatusServer(g Globals, socketPath string) {
-	_ = os.RemoveAll(socketPath)
-
-	handler := func(w http.ResponseWriter, r *http.Request) {
-		report := StatusReport{
-			LocalPeerIP: g.LocalPeerIP,
-			Network:     g.Network,
-		}
-		json.NewEncoder(w).Encode(report)
-	}
-
-	server := http.Server{Handler: http.HandlerFunc(handler)}
-	unixListener, err := net.Listen("unix", socketPath)
-	if err != nil {
-		log.Fatalf("Failed to bind to unix socket: %v", err)
-	}
-	if err := server.Serve(unixListener); err != nil {
-		log.Fatalf("Failed to serve on unix socket: %v", err)
-	}
-}

peer/wginterface/interface.go

@@ -21,6 +21,7 @@ import (
 // Create creates a WireGuard interface named name, assigns vpnIP/prefixLen to
 // it, and brings it up.
 func Create(name string, vpnIP net.IP, prefixLen int) error {
+	_ = Delete(name) // remove any stale interface left by a previous run
 	if err := nlNewLink(name); err != nil {
 		return fmt.Errorf("failed to create wireguard link: %w", err)
 	}

peer/wginterface/manage.go

@@ -23,10 +23,9 @@ const (
 	SessionTimeout = 180 * time.Second
 )
 
-var (
-	probeKeepalive = 5 * time.Second
-	zeroKeepalive  = time.Duration(0)
-)
+const ProbeKeepalive = 8 * time.Second
+
+var zeroKeepalive = time.Duration(0)
 
 // Device wraps a wgctrl client bound to a named WireGuard interface.
 type Device struct {
@@ -48,6 +47,11 @@ func (d *Device) Close() error {
 	return d.client.Close()
 }
 
+// Name returns the interface name.
+func (d *Device) Name() string {
+	return d.name
+}
+
 // Configure sets the device's private key and UDP listen port.
 func (d *Device) Configure(privKey wgtypes.Key, listenPort int) error {
 	return d.client.ConfigureDevice(d.name, wgtypes.Config{
@@ -79,6 +83,17 @@ func (d *Device) Peer(pubKey wgtypes.Key) (wgtypes.Peer, error) {
 	return wgtypes.Peer{}, fmt.Errorf("peer %v not found in %q", pubKey, d.name)
 }
 
+// AddPeer registers a peer with no AllowedIPs and no endpoint. WireGuard will
+// accept handshakes from this peer but route no traffic to it yet.
+func (d *Device) AddPeer(pubKey wgtypes.Key) error {
+	return d.client.ConfigureDevice(d.name, wgtypes.Config{
+		Peers: []wgtypes.PeerConfig{{
+			PublicKey:         pubKey,
+			ReplaceAllowedIPs: true,
+		}},
+	})
+}
+
 // SetRelay configures the relay peer with AllowedIPs covering the entire VPN
 // network prefix. This is the fallback route for all VPN traffic.
 func (d *Device) SetRelay(pubKey wgtypes.Key, endpoint netip.AddrPort, network netip.Prefix) error {
@@ -100,13 +115,14 @@ func (d *Device) SetRelay(pubKey wgtypes.Key, endpoint netip.AddrPort, network n
 // AddProbe adds a peer with no AllowedIPs and a 5s keepalive. WireGuard will
 // attempt handshakes without routing any traffic through this peer yet.
 func (d *Device) AddProbe(pubKey wgtypes.Key, endpoint netip.AddrPort) error {
+	keepalive := ProbeKeepalive
 	return d.client.ConfigureDevice(d.name, wgtypes.Config{
 		Peers: []wgtypes.PeerConfig{{
 			PublicKey:                   pubKey,
 			Endpoint:                    net.UDPAddrFromAddrPort(endpoint),
 			AllowedIPs:                  []net.IPNet{},
 			ReplaceAllowedIPs:           true,
-			PersistentKeepaliveInterval: &probeKeepalive,
+			PersistentKeepaliveInterval: &keepalive,
 		}},
 	})
 }
@@ -157,9 +173,12 @@ func (d *Device) RemovePeer(pubKey wgtypes.Key) error {
 	})
 }
 
-// EnableForwarding enables IPv4 forwarding on the interface, required for
-// relay peers that forward traffic between VPN peers.
+// EnableForwarding enables IPv4 forwarding globally and on the interface,
+// required for relay peers that forward traffic between VPN peers.
 func (d *Device) EnableForwarding() error {
+	if err := os.WriteFile("/proc/sys/net/ipv4/ip_forward", []byte("1\n"), 0644); err != nil {
+		return err
+	}
 	path := fmt.Sprintf("/proc/sys/net/ipv4/conf/%s/forwarding", d.name)
 	return os.WriteFile(path, []byte("1\n"), 0644)
 }