mirror of
https://github.com/fosrl/olm.git
synced 2026-08-07 16:37:14 -05:00
Compare commits
| Author | SHA1 | Date | |
|---|---|---|---|
|
|
4f54e27b22 |
+7
-39
@@ -62,15 +62,6 @@ type OlmError struct {
|
||||
Message string `json:"message"`
|
||||
}
|
||||
|
||||
// ExitNodeStatus represents the connectivity status of the client's own exit
|
||||
// node connection (used for site resources hosted on the exit node).
|
||||
type ExitNodeStatus struct {
|
||||
Connected bool `json:"connected"`
|
||||
RTT time.Duration `json:"rtt"`
|
||||
LastSeen time.Time `json:"lastSeen"`
|
||||
Endpoint string `json:"endpoint,omitempty"`
|
||||
}
|
||||
|
||||
// StatusResponse is returned by the status endpoint
|
||||
type StatusResponse struct {
|
||||
Connected bool `json:"connected"`
|
||||
@@ -82,7 +73,6 @@ type StatusResponse struct {
|
||||
OrgID string `json:"orgId,omitempty"`
|
||||
PeerStatuses map[int]*PeerStatus `json:"peers,omitempty"`
|
||||
NetworkSettings network.NetworkSettings `json:"networkSettings,omitempty"`
|
||||
ExitNodeStatus *ExitNodeStatus `json:"exitNode,omitempty"`
|
||||
}
|
||||
|
||||
type MetadataChangeRequest struct {
|
||||
@@ -113,14 +103,13 @@ type API struct {
|
||||
onPowerMode func(PowerModeRequest) error
|
||||
onJITConnect func(JITConnectionRequest) error
|
||||
|
||||
statusMu sync.RWMutex
|
||||
peerStatuses map[int]*PeerStatus
|
||||
exitNodeStatus *ExitNodeStatus
|
||||
connectedAt time.Time
|
||||
isConnected bool
|
||||
isRegistered bool
|
||||
isTerminated bool
|
||||
olmError *OlmError
|
||||
statusMu sync.RWMutex
|
||||
peerStatuses map[int]*PeerStatus
|
||||
connectedAt time.Time
|
||||
isConnected bool
|
||||
isRegistered bool
|
||||
isTerminated bool
|
||||
olmError *OlmError
|
||||
|
||||
version string
|
||||
agent string
|
||||
@@ -420,25 +409,6 @@ func (s *API) UpdatePeerHolepunchStatus(siteID int, holepunchConnected bool) {
|
||||
status.HolepunchConnected = holepunchConnected
|
||||
}
|
||||
|
||||
// SetExitNodeStatus sets the connectivity status of the client's own exit node connection
|
||||
func (s *API) SetExitNodeStatus(connected bool, rtt time.Duration, endpoint string) {
|
||||
s.statusMu.Lock()
|
||||
defer s.statusMu.Unlock()
|
||||
s.exitNodeStatus = &ExitNodeStatus{
|
||||
Connected: connected,
|
||||
RTT: rtt,
|
||||
LastSeen: time.Now(),
|
||||
Endpoint: endpoint,
|
||||
}
|
||||
}
|
||||
|
||||
// ClearExitNodeStatus removes the exit node status, e.g. when disconnecting from it
|
||||
func (s *API) ClearExitNodeStatus() {
|
||||
s.statusMu.Lock()
|
||||
defer s.statusMu.Unlock()
|
||||
s.exitNodeStatus = nil
|
||||
}
|
||||
|
||||
// handleConnect handles the /connect endpoint
|
||||
func (s *API) handleConnect(w http.ResponseWriter, r *http.Request) {
|
||||
if r.Method != http.MethodPost {
|
||||
@@ -503,7 +473,6 @@ func (s *API) handleStatus(w http.ResponseWriter, r *http.Request) {
|
||||
OrgID: s.orgID,
|
||||
PeerStatuses: s.peerStatuses,
|
||||
NetworkSettings: network.GetSettings(),
|
||||
ExitNodeStatus: s.exitNodeStatus,
|
||||
}
|
||||
|
||||
s.statusMu.RUnlock()
|
||||
@@ -671,7 +640,6 @@ func (s *API) GetStatus() StatusResponse {
|
||||
OrgID: s.orgID,
|
||||
PeerStatuses: s.peerStatuses,
|
||||
NetworkSettings: network.GetSettings(),
|
||||
ExitNodeStatus: s.exitNodeStatus,
|
||||
}
|
||||
}
|
||||
|
||||
|
||||
@@ -8,7 +8,6 @@ require (
|
||||
github.com/godbus/dbus/v5 v5.2.2
|
||||
github.com/gorilla/websocket v1.5.3
|
||||
github.com/miekg/dns v1.1.70
|
||||
golang.org/x/net v0.56.0
|
||||
golang.org/x/sys v0.46.0
|
||||
golang.zx2c4.com/wireguard v0.0.0-20250521234502-f333402bd9cb
|
||||
golang.zx2c4.com/wireguard/wgctrl v0.0.0-20241231184526-a9ab2273dd10
|
||||
@@ -24,6 +23,7 @@ require (
|
||||
golang.org/x/crypto v0.53.0 // indirect
|
||||
golang.org/x/exp v0.0.0-20251113190631-e25ba8c21ef6 // indirect
|
||||
golang.org/x/mod v0.34.0 // indirect
|
||||
golang.org/x/net v0.56.0 // indirect
|
||||
golang.org/x/sync v0.20.0 // indirect
|
||||
golang.org/x/time v0.12.0 // indirect
|
||||
golang.org/x/tools v0.43.0 // indirect
|
||||
@@ -32,4 +32,4 @@ require (
|
||||
)
|
||||
|
||||
// To be used ONLY for local development
|
||||
replace github.com/fosrl/newt => ../newt
|
||||
// replace github.com/fosrl/newt => ../newt
|
||||
|
||||
@@ -1,5 +1,7 @@
|
||||
github.com/Microsoft/go-winio v0.6.2 h1:F2VQgta7ecxGYO8k3ZZz3RS8fVIXVxONVUPlNERoyfY=
|
||||
github.com/Microsoft/go-winio v0.6.2/go.mod h1:yd8OoFMLzJbo9gZq8j5qaps8bJ9aShtEA8Ipt1oGCvU=
|
||||
github.com/fosrl/newt v1.15.0 h1:WpL0whZM1FMjUe2Vy5jSH1bgbxm1O9k1qCyF/mqZT+s=
|
||||
github.com/fosrl/newt v1.15.0/go.mod h1:l6kWoZPSaXT+ZRUjiyPgwflRqZWYaXpUj9oQ0sOPh4o=
|
||||
github.com/godbus/dbus/v5 v5.2.2 h1:TUR3TgtSVDmjiXOgAAyaZbYmIeP3DPkld3jgKGV8mXQ=
|
||||
github.com/godbus/dbus/v5 v5.2.2/go.mod h1:3AAv2+hPq5rdnr5txxxRwiGjPXamgoIHgz9FPBfOp3c=
|
||||
github.com/google/btree v1.1.3 h1:CVpQJjYgC4VbzxeGVHfvZrv1ctoYCAI8vbl07Fcxlyg=
|
||||
|
||||
+1
-34
@@ -11,7 +11,6 @@ import (
|
||||
|
||||
"github.com/fosrl/newt/logger"
|
||||
"github.com/fosrl/newt/network"
|
||||
"github.com/fosrl/newt/util"
|
||||
olmDevice "github.com/fosrl/olm/device"
|
||||
"github.com/fosrl/olm/dns"
|
||||
dnsOverride "github.com/fosrl/olm/dns/override"
|
||||
@@ -52,11 +51,6 @@ func (o *Olm) handleConnect(msg websocket.WSMessage) {
|
||||
o.updateRegister = nil
|
||||
}
|
||||
|
||||
if o.stopPingRequest != nil {
|
||||
o.stopPingRequest()
|
||||
o.stopPingRequest = nil
|
||||
}
|
||||
|
||||
// if there is an existing tunnel then close it
|
||||
if o.dev != nil {
|
||||
logger.Info("Got new message. Closing existing tunnel!")
|
||||
@@ -74,17 +68,6 @@ func (o *Olm) handleConnect(msg websocket.WSMessage) {
|
||||
return
|
||||
}
|
||||
|
||||
// When handed an already-open FD (mobile/NetworkExtension platforms), the
|
||||
// TUN device's addresses and routes are owned and reconciled by the host
|
||||
// platform from NetworkSettings (e.g. Apple's NEPacketTunnelProvider via
|
||||
// setTunnelNetworkSettings) - our own ifconfig/route subprocess calls must
|
||||
// not also run against the same interface, or the two end up installing
|
||||
// competing routes to the same destination. On macOS specifically this
|
||||
// package's darwin code paths would otherwise run for real here (the NE
|
||||
// build shares GOOS=darwin with the CLI), unlike iOS where they're already
|
||||
// no-ops via a GOOS check.
|
||||
network.NativeConfigDisabled = o.tunnelConfig.FileDescriptorTun != 0
|
||||
|
||||
o.tdev, err = func() (tun.Device, error) {
|
||||
if o.tunnelConfig.FileDescriptorTun != 0 {
|
||||
return olmDevice.CreateTUNFromFD(o.tunnelConfig.FileDescriptorTun, o.tunnelConfig.MTU)
|
||||
@@ -155,14 +138,6 @@ func (o *Olm) handleConnect(msg websocket.WSMessage) {
|
||||
logger.Error("Failed to bring up WireGuard device: %v", err)
|
||||
}
|
||||
|
||||
// Set the private key unconditionally, since it's otherwise only ever set as a
|
||||
// side effect of configuring a site peer (see peers.ConfigurePeer) - if there are
|
||||
// no sites (e.g. an exit-node-only connection), the interface would otherwise be
|
||||
// brought up with no private key configured at all.
|
||||
if err := o.dev.IpcSet(fmt.Sprintf("private_key=%s\n", util.FixKey(o.privateKey.String()))); err != nil {
|
||||
logger.Error("Failed to set private key on WireGuard device: %v", err)
|
||||
}
|
||||
|
||||
// Extract interface IP (strip CIDR notation if present)
|
||||
interfaceIP := wgData.TunnelIP
|
||||
if strings.Contains(interfaceIP, "/") {
|
||||
@@ -187,7 +162,7 @@ func (o *Olm) handleConnect(msg websocket.WSMessage) {
|
||||
logger.Error("Failed to o.tunnelConfigure interface: %v", err)
|
||||
}
|
||||
|
||||
if err := network.AddRoutesWithSource([]string{wgData.UtilitySubnet}, o.tunnelConfig.InterfaceName, interfaceIP); err != nil { // also route the utility subnet
|
||||
if network.AddRoutes([]string{wgData.UtilitySubnet}, o.tunnelConfig.InterfaceName); err != nil { // also route the utility subnet
|
||||
logger.Error("Failed to add route for utility subnet: %v", err)
|
||||
}
|
||||
|
||||
@@ -282,14 +257,6 @@ func (o *Olm) handleConnect(msg websocket.WSMessage) {
|
||||
network.SetDNSServers([]string{o.dnsProxy.GetProxyIP().String()})
|
||||
}
|
||||
|
||||
if wgData.ExitNode != nil && wgData.ExitNode.Connect {
|
||||
if err := o.connectExitNode(*wgData.ExitNode); err != nil {
|
||||
logger.Error("Failed to connect to exit node: %v", err)
|
||||
}
|
||||
} else {
|
||||
logger.Debug("No exit node to connect to (not provided, or connect flag is false)")
|
||||
}
|
||||
|
||||
o.apiServer.SetRegistered(true)
|
||||
|
||||
o.registered = true
|
||||
|
||||
@@ -202,10 +202,6 @@ func (o *Olm) handleSync(msg websocket.WSMessage) {
|
||||
// Sync exit nodes for hole punching
|
||||
o.syncExitNodes(syncData.ExitNodes)
|
||||
|
||||
// Reconcile the client's own exit node connection (connect/switch/update
|
||||
// aliases/disconnect), same as what the initial olm/wg/connect message does
|
||||
o.syncExitNodeConnection(syncData.ExitNode)
|
||||
|
||||
// Build a map of expected peers from the incoming data
|
||||
expectedPeers := make(map[int]peers.SiteConfig)
|
||||
for _, site := range syncData.Sites {
|
||||
|
||||
-386
@@ -1,386 +0,0 @@
|
||||
package olm
|
||||
|
||||
import (
|
||||
"encoding/json"
|
||||
"fmt"
|
||||
"net"
|
||||
"strings"
|
||||
|
||||
"github.com/fosrl/newt/logger"
|
||||
"github.com/fosrl/newt/network"
|
||||
"github.com/fosrl/newt/util"
|
||||
"github.com/fosrl/olm/peers"
|
||||
"github.com/fosrl/olm/websocket"
|
||||
)
|
||||
|
||||
// exitNodeAliasSiteId is the sentinel siteId used when registering exit node
|
||||
// aliases with the DNS proxy. It is not a real site, and the JIT handler
|
||||
// treats siteId 0 as "no JIT lookup", which is correct here since the exit
|
||||
// node is connected directly rather than on demand.
|
||||
const exitNodeAliasSiteId = 0
|
||||
|
||||
// connectExitNode configures a WireGuard peer connection to an exit node, on the
|
||||
// same interface and WireGuard device already used for site peers. The exit node
|
||||
// lives in a different address space than the site tunnel, so a secondary address
|
||||
// (ExitNodeConfig.TunnelIP) is added to the interface for it - the exit node's own
|
||||
// WireGuard peer entry only accepts traffic sourced from that address. Nothing here
|
||||
// is persisted; it's purely in-memory WireGuard/routing state, same as site peers.
|
||||
func (o *Olm) connectExitNode(cfg ExitNodeConfig) error {
|
||||
if !o.tunnelRunning {
|
||||
return fmt.Errorf("tunnel not running")
|
||||
}
|
||||
if cfg.PublicKey == "" || cfg.Endpoint == "" || cfg.ServerIP == "" || cfg.TunnelIP == "" {
|
||||
return fmt.Errorf("incomplete exit node configuration")
|
||||
}
|
||||
|
||||
o.exitNodeMu.Lock()
|
||||
defer o.exitNodeMu.Unlock()
|
||||
|
||||
dev := o.dev
|
||||
if dev == nil {
|
||||
return fmt.Errorf("wireguard device not initialized")
|
||||
}
|
||||
|
||||
if o.exitNode != nil && o.exitNode.PublicKey == cfg.PublicKey &&
|
||||
o.exitNode.Endpoint == cfg.Endpoint && o.exitNode.ServerIP == cfg.ServerIP &&
|
||||
o.exitNode.TunnelIP == cfg.TunnelIP {
|
||||
if !slicesEqual(o.exitNode.Aliases, cfg.Aliases) {
|
||||
logger.Info("Already connected to exit node %s, updating aliases", cfg.PublicKey)
|
||||
o.updateExitNodeAliasesLocked(cfg.Aliases)
|
||||
} else {
|
||||
logger.Info("Already connected to exit node %s, ignoring duplicate connect message", cfg.PublicKey)
|
||||
}
|
||||
return nil
|
||||
}
|
||||
|
||||
if o.exitNode != nil && o.exitNode.PublicKey != cfg.PublicKey {
|
||||
logger.Info("Switching exit nodes, removing previous exit node peer")
|
||||
if err := o.removeExitNodePeerLocked(); err != nil {
|
||||
logger.Warn("Failed to remove previous exit node peer: %v", err)
|
||||
}
|
||||
}
|
||||
|
||||
endpoint := cfg.Endpoint
|
||||
if !strings.Contains(endpoint, ":") {
|
||||
relayPort := cfg.RelayPort
|
||||
if relayPort == 0 {
|
||||
relayPort = 21820
|
||||
}
|
||||
endpoint = fmt.Sprintf("%s:%d", endpoint, relayPort)
|
||||
}
|
||||
|
||||
resolvedEndpoint, err := util.ResolveDomain(endpoint)
|
||||
if err != nil {
|
||||
return fmt.Errorf("failed to resolve exit node endpoint: %w", err)
|
||||
}
|
||||
|
||||
persistentKeepalive := 0
|
||||
if pm := o.getPeerManager(); pm != nil {
|
||||
persistentKeepalive = pm.PersistentKeepalive
|
||||
}
|
||||
|
||||
allowedIP := strings.Split(cfg.ServerIP, "/")[0] + "/32"
|
||||
|
||||
wgConfig := fmt.Sprintf(`public_key=%s
|
||||
allowed_ip=%s
|
||||
endpoint=%s
|
||||
persistent_keepalive_interval=%d`, util.FixKey(cfg.PublicKey), allowedIP, resolvedEndpoint, persistentKeepalive)
|
||||
|
||||
if err := dev.IpcSet(wgConfig); err != nil {
|
||||
return fmt.Errorf("failed to configure exit node peer: %w", err)
|
||||
}
|
||||
|
||||
interfaceName := o.tunnelConfig.InterfaceName
|
||||
tunnelIP := cfg.TunnelIP
|
||||
if !strings.Contains(tunnelIP, "/") {
|
||||
tunnelIP += "/32"
|
||||
}
|
||||
if err := network.AddSecondaryAddress(interfaceName, tunnelIP); err != nil {
|
||||
logger.Warn("Failed to add secondary address %s for exit node: %v", tunnelIP, err)
|
||||
}
|
||||
|
||||
// ServerIP arrives as a bare IP with no CIDR suffix, but AddRouteForServerIP
|
||||
// parses it as a CIDR on darwin (to explicitly route the subnet up the tunnel,
|
||||
// since unlike Linux, adding the address to the interface does not implicitly
|
||||
// create a route for it) - without a mask that parse fails and the route (and
|
||||
// its corresponding NetworkSettings entry, which is what surfaces it via the
|
||||
// API) is silently never added.
|
||||
serverIPForRoute := strings.Split(cfg.ServerIP, "/")[0] + "/32"
|
||||
// The route must also be pinned to our exit node tunnel address as its source
|
||||
// (darwin route(8) -ifa): the interface carries a second address for the site
|
||||
// tunnel too, and without an explicit source darwin picks that one instead,
|
||||
// which the exit node's WireGuard AllowedIPs filtering then silently drops.
|
||||
tunnelIPForRoute := strings.Split(cfg.TunnelIP, "/")[0]
|
||||
if err := network.AddRouteForServerIPWithSource(serverIPForRoute, interfaceName, tunnelIPForRoute); err != nil {
|
||||
logger.Warn("Failed to add route for exit node server IP: %v", err)
|
||||
}
|
||||
|
||||
cfgCopy := cfg
|
||||
o.exitNode = &cfgCopy
|
||||
|
||||
if o.dnsProxy != nil {
|
||||
serverIP := net.ParseIP(cfg.ServerIP)
|
||||
if serverIP != nil {
|
||||
for _, alias := range cfg.Aliases {
|
||||
logger.Debug("Adding alias %s to the edit node", alias)
|
||||
if err := o.dnsProxy.AddDNSRecord(alias, serverIP, exitNodeAliasSiteId); err != nil {
|
||||
logger.Warn("Failed to add DNS record for exit node alias %s: %v", alias, err)
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
if pm := o.getPeerManager(); pm != nil {
|
||||
pm.SetExitNode(strings.Split(cfg.ServerIP, "/")[0], strings.Split(cfg.TunnelIP, "/")[0])
|
||||
}
|
||||
|
||||
logger.Info("Connected to exit node at %s", resolvedEndpoint)
|
||||
return nil
|
||||
}
|
||||
|
||||
// disconnectExitNode tears down the current exit node peer connection, if any.
|
||||
func (o *Olm) disconnectExitNode() error {
|
||||
o.exitNodeMu.Lock()
|
||||
defer o.exitNodeMu.Unlock()
|
||||
|
||||
return o.removeExitNodePeerLocked()
|
||||
}
|
||||
|
||||
// removeExitNodePeerLocked removes the current exit node peer, its secondary
|
||||
// interface address, and its server IP route. Must be called with exitNodeMu held.
|
||||
func (o *Olm) removeExitNodePeerLocked() error {
|
||||
if o.exitNode == nil {
|
||||
return nil
|
||||
}
|
||||
cfg := o.exitNode
|
||||
o.exitNode = nil
|
||||
|
||||
if pm := o.getPeerManager(); pm != nil {
|
||||
pm.ClearExitNode()
|
||||
}
|
||||
|
||||
if o.dnsProxy != nil {
|
||||
serverIP := net.ParseIP(cfg.ServerIP)
|
||||
if serverIP != nil {
|
||||
for _, alias := range cfg.Aliases {
|
||||
o.dnsProxy.RemoveDNSRecordForSite(alias, serverIP, exitNodeAliasSiteId)
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
if o.dev != nil {
|
||||
if err := peers.RemovePeer(o.dev, 0, cfg.PublicKey); err != nil {
|
||||
logger.Warn("Failed to remove exit node peer: %v", err)
|
||||
}
|
||||
}
|
||||
|
||||
interfaceName := o.tunnelConfig.InterfaceName
|
||||
serverIPForRoute := strings.Split(cfg.ServerIP, "/")[0] + "/32"
|
||||
tunnelIPForRoute := strings.Split(cfg.TunnelIP, "/")[0]
|
||||
if err := network.RemoveRouteForServerIPWithSource(serverIPForRoute, interfaceName, tunnelIPForRoute); err != nil {
|
||||
logger.Warn("Failed to remove route for exit node server IP: %v", err)
|
||||
}
|
||||
|
||||
tunnelIP := cfg.TunnelIP
|
||||
if !strings.Contains(tunnelIP, "/") {
|
||||
tunnelIP += "/32"
|
||||
}
|
||||
if err := network.RemoveSecondaryAddress(interfaceName, tunnelIP); err != nil {
|
||||
logger.Warn("Failed to remove secondary address %s for exit node: %v", tunnelIP, err)
|
||||
}
|
||||
|
||||
logger.Info("Disconnected from exit node")
|
||||
return nil
|
||||
}
|
||||
|
||||
// syncExitNodeConnection reconciles the client's own exit node connection (used
|
||||
// for site resources hosted on the exit node) with the desired state sent in a
|
||||
// sync message - connecting, switching, updating aliases, or disconnecting as
|
||||
// needed. This mirrors what the initial "olm/wg/connect" message does, so a
|
||||
// client that reconnects with a stale exit node assignment (or none at all)
|
||||
// converges without needing to fully re-register.
|
||||
func (o *Olm) syncExitNodeConnection(cfg *ExitNodeConfig) {
|
||||
if !o.tunnelRunning {
|
||||
logger.Debug("Tunnel stopped, ignoring exit node sync")
|
||||
return
|
||||
}
|
||||
|
||||
if cfg == nil || !cfg.Connect {
|
||||
if err := o.disconnectExitNode(); err != nil {
|
||||
logger.Error("Sync: Failed to disconnect from exit node: %v", err)
|
||||
}
|
||||
return
|
||||
}
|
||||
|
||||
if err := o.connectExitNode(*cfg); err != nil {
|
||||
logger.Error("Sync: Failed to connect to exit node: %v", err)
|
||||
}
|
||||
}
|
||||
|
||||
// updateExitNodeAliasesLocked reconciles the currently connected exit node's
|
||||
// aliases with newAliases, adding new ones before removing stale ones so a
|
||||
// rename never has a gap in resolution. Must be called with exitNodeMu held.
|
||||
func (o *Olm) updateExitNodeAliasesLocked(newAliases []string) {
|
||||
if o.exitNode == nil {
|
||||
return
|
||||
}
|
||||
|
||||
added := stringSliceDiff(newAliases, o.exitNode.Aliases)
|
||||
removed := stringSliceDiff(o.exitNode.Aliases, newAliases)
|
||||
|
||||
serverIP := net.ParseIP(o.exitNode.ServerIP)
|
||||
if o.dnsProxy != nil && serverIP != nil {
|
||||
for _, alias := range added {
|
||||
if err := o.dnsProxy.AddDNSRecord(alias, serverIP, exitNodeAliasSiteId); err != nil {
|
||||
logger.Warn("Failed to add DNS record for exit node alias %s: %v", alias, err)
|
||||
}
|
||||
}
|
||||
for _, alias := range removed {
|
||||
o.dnsProxy.RemoveDNSRecordForSite(alias, serverIP, exitNodeAliasSiteId)
|
||||
}
|
||||
}
|
||||
|
||||
o.exitNode.Aliases = applyStringListUpdate(o.exitNode.Aliases, removed, added)
|
||||
}
|
||||
|
||||
// stringSliceDiff returns the elements of a that are not present in b.
|
||||
func stringSliceDiff(a, b []string) []string {
|
||||
inB := make(map[string]struct{}, len(b))
|
||||
for _, s := range b {
|
||||
inB[s] = struct{}{}
|
||||
}
|
||||
diff := make([]string, 0, len(a))
|
||||
for _, s := range a {
|
||||
if _, ok := inB[s]; !ok {
|
||||
diff = append(diff, s)
|
||||
}
|
||||
}
|
||||
return diff
|
||||
}
|
||||
|
||||
// handleExitNodeConnect handles a server-initiated request to connect to (or switch to)
|
||||
// an exit node, delivered as a full ExitNodeConfig payload.
|
||||
func (o *Olm) handleExitNodeConnect(msg websocket.WSMessage) {
|
||||
logger.Debug("Received exit node connect message: %v", msg.Data)
|
||||
|
||||
if !o.tunnelRunning {
|
||||
logger.Debug("Tunnel stopped, ignoring exit node connect message")
|
||||
return
|
||||
}
|
||||
|
||||
jsonData, err := json.Marshal(msg.Data)
|
||||
if err != nil {
|
||||
logger.Error("Error marshaling exit node connect data: %v", err)
|
||||
return
|
||||
}
|
||||
|
||||
var cfg ExitNodeConfig
|
||||
if err := json.Unmarshal(jsonData, &cfg); err != nil {
|
||||
logger.Error("Error unmarshaling exit node connect data: %v", err)
|
||||
return
|
||||
}
|
||||
|
||||
if !cfg.Connect {
|
||||
logger.Debug("Exit node connect message has connect=false, disconnecting instead")
|
||||
if err := o.disconnectExitNode(); err != nil {
|
||||
logger.Error("Failed to disconnect from exit node: %v", err)
|
||||
}
|
||||
return
|
||||
}
|
||||
|
||||
if err := o.connectExitNode(cfg); err != nil {
|
||||
logger.Error("Failed to connect to exit node: %v", err)
|
||||
}
|
||||
}
|
||||
|
||||
// handleExitNodeDisconnect handles a server-initiated request to disconnect from the
|
||||
// currently connected exit node.
|
||||
func (o *Olm) handleExitNodeDisconnect(msg websocket.WSMessage) {
|
||||
logger.Debug("Received exit node disconnect message: %v", msg.Data)
|
||||
|
||||
if !o.tunnelRunning {
|
||||
logger.Debug("Tunnel stopped, ignoring exit node disconnect message")
|
||||
return
|
||||
}
|
||||
|
||||
if err := o.disconnectExitNode(); err != nil {
|
||||
logger.Error("Failed to disconnect from exit node: %v", err)
|
||||
}
|
||||
}
|
||||
|
||||
// handleExitNodeUpdateData handles a server-initiated request to change data
|
||||
// associated with the currently connected exit node, such as its aliases (e.g. a
|
||||
// resource was renamed). Unlike site aliases, there is no per-alias address to
|
||||
// track since every exit node alias resolves to the exit node's own ServerIP.
|
||||
func (o *Olm) handleExitNodeUpdateData(msg websocket.WSMessage) {
|
||||
logger.Debug("Received exit node update data message: %v", msg.Data)
|
||||
|
||||
if !o.tunnelRunning {
|
||||
logger.Debug("Tunnel stopped, ignoring exit node update data message")
|
||||
return
|
||||
}
|
||||
|
||||
jsonData, err := json.Marshal(msg.Data)
|
||||
if err != nil {
|
||||
logger.Error("Error marshaling exit node update data: %v", err)
|
||||
return
|
||||
}
|
||||
|
||||
var update ExitNodeUpdateData
|
||||
if err := json.Unmarshal(jsonData, &update); err != nil {
|
||||
logger.Error("Error unmarshaling exit node update data: %v", err)
|
||||
return
|
||||
}
|
||||
|
||||
o.exitNodeMu.Lock()
|
||||
defer o.exitNodeMu.Unlock()
|
||||
|
||||
if o.exitNode == nil {
|
||||
logger.Debug("Ignoring exit node update data message: no exit node connected")
|
||||
return
|
||||
}
|
||||
|
||||
serverIP := net.ParseIP(o.exitNode.ServerIP)
|
||||
|
||||
// Add new aliases BEFORE removing old ones, same as site aliases, so a rename
|
||||
// that keeps the same underlying address never has a gap in resolution.
|
||||
if o.dnsProxy != nil && serverIP != nil {
|
||||
for _, alias := range update.NewAliases {
|
||||
if err := o.dnsProxy.AddDNSRecord(alias, serverIP, exitNodeAliasSiteId); err != nil {
|
||||
logger.Warn("Failed to add DNS record for exit node alias %s: %v", alias, err)
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
if o.dnsProxy != nil && serverIP != nil {
|
||||
for _, alias := range update.OldAliases {
|
||||
o.dnsProxy.RemoveDNSRecordForSite(alias, serverIP, exitNodeAliasSiteId)
|
||||
}
|
||||
}
|
||||
|
||||
o.exitNode.Aliases = applyStringListUpdate(o.exitNode.Aliases, update.OldAliases, update.NewAliases)
|
||||
|
||||
logger.Info("Successfully updated exit node data")
|
||||
}
|
||||
|
||||
// applyStringListUpdate returns list with every entry in removed dropped and every
|
||||
// entry in added appended, preserving the add-before-remove semantics of the caller.
|
||||
func applyStringListUpdate(list, removed, added []string) []string {
|
||||
next := make([]string, 0, len(list)+len(added))
|
||||
next = append(next, list...)
|
||||
next = append(next, added...)
|
||||
|
||||
removedSet := make(map[string]struct{}, len(removed))
|
||||
for _, alias := range removed {
|
||||
removedSet[alias] = struct{}{}
|
||||
}
|
||||
|
||||
filtered := next[:0]
|
||||
for _, alias := range next {
|
||||
if _, ok := removedSet[alias]; ok {
|
||||
continue
|
||||
}
|
||||
filtered = append(filtered, alias)
|
||||
}
|
||||
return filtered
|
||||
}
|
||||
+15
-115
@@ -4,7 +4,6 @@ import (
|
||||
"context"
|
||||
"crypto/rand"
|
||||
"encoding/hex"
|
||||
"encoding/json"
|
||||
"fmt"
|
||||
"net"
|
||||
"net/http"
|
||||
@@ -16,7 +15,6 @@ import (
|
||||
|
||||
"github.com/fosrl/newt/bind"
|
||||
"github.com/fosrl/newt/clients/permissions"
|
||||
"github.com/fosrl/newt/exitnode"
|
||||
"github.com/fosrl/newt/holepunch"
|
||||
"github.com/fosrl/newt/logger"
|
||||
"github.com/fosrl/newt/network"
|
||||
@@ -57,11 +55,6 @@ type Olm struct {
|
||||
holePunchManager *holepunch.Manager
|
||||
peerManager *peers.PeerManager
|
||||
peerManagerMu sync.RWMutex
|
||||
|
||||
// exitNode tracks the currently connected exit node peer, if any. It lives on a
|
||||
// secondary address on the same interface/WireGuard device as the site peers.
|
||||
exitNode *ExitNodeConfig
|
||||
exitNodeMu sync.Mutex
|
||||
// Power mode management
|
||||
currentPowerMode string
|
||||
powerModeMu sync.Mutex
|
||||
@@ -82,11 +75,6 @@ type Olm struct {
|
||||
stopRegister func()
|
||||
updateRegister func(newData any)
|
||||
|
||||
// Exit node ping dance, run before registration so the server can pick
|
||||
// the best exit node (mirrors newt's newt/ping/request flow).
|
||||
stopPingRequest func()
|
||||
pendingPingChainId string
|
||||
|
||||
stopPeerSends map[string]func()
|
||||
stopPeerInits map[string]func()
|
||||
jitPendingSites map[int]string // siteId -> chainId for in-flight JIT requests
|
||||
@@ -562,66 +550,6 @@ func (o *Olm) StartTunnel(config TunnelConfig) {
|
||||
o.websocket.RegisterHandler("olm/wg/peer/chain/cancel", o.handleCancelChain)
|
||||
o.websocket.RegisterHandler("olm/sync", o.handleSync)
|
||||
|
||||
// Handlers for the server to direct connecting/disconnecting an exit node after registration
|
||||
o.websocket.RegisterHandler("olm/wg/exitnode/connect", o.handleExitNodeConnect)
|
||||
o.websocket.RegisterHandler("olm/wg/exitnode/disconnect", o.handleExitNodeDisconnect)
|
||||
o.websocket.RegisterHandler("olm/wg/exitnode/data/update", o.handleExitNodeUpdateData)
|
||||
|
||||
o.websocket.RegisterHandler("olm/ping/exitNodes", func(msg websocket.WSMessage) {
|
||||
logger.Debug("Received exit node ping request")
|
||||
|
||||
if o.stopPingRequest != nil {
|
||||
o.stopPingRequest()
|
||||
o.stopPingRequest = nil
|
||||
}
|
||||
|
||||
if !o.tunnelRunning {
|
||||
logger.Debug("Tunnel is no longer running, skipping exit node ping")
|
||||
return
|
||||
}
|
||||
|
||||
var exitNodeData exitnode.ExitNodeData
|
||||
jsonData, err := json.Marshal(msg.Data)
|
||||
if err != nil {
|
||||
logger.Error("Error marshaling exit node data: %v", err)
|
||||
return
|
||||
}
|
||||
if err := json.Unmarshal(jsonData, &exitNodeData); err != nil {
|
||||
logger.Error("Error unmarshaling exit node data: %v", err)
|
||||
return
|
||||
}
|
||||
|
||||
if exitNodeData.ChainId != "" {
|
||||
if exitNodeData.ChainId != o.pendingPingChainId {
|
||||
logger.Debug("Discarding duplicate/stale olm/ping/exitNodes (chainId=%s, expected=%s)", exitNodeData.ChainId, o.pendingPingChainId)
|
||||
return
|
||||
}
|
||||
o.pendingPingChainId = ""
|
||||
}
|
||||
|
||||
if len(exitNodeData.ExitNodes) == 0 {
|
||||
logger.Info("No exit nodes provided")
|
||||
return
|
||||
}
|
||||
|
||||
pingResults := exitnode.PingExitNodes(exitNodeData.ExitNodes, "", false)
|
||||
|
||||
publicKey := o.privateKey.PublicKey()
|
||||
logger.Debug("Sending registration message to server with public key: %s, relay: %v, pingResults: %+v", publicKey, !config.Holepunch, pingResults)
|
||||
o.stopRegister, o.updateRegister = o.websocket.SendMessageInterval("olm/wg/register", map[string]any{
|
||||
"publicKey": publicKey.String(),
|
||||
"relay": !config.Holepunch,
|
||||
"olmVersion": o.olmConfig.Version,
|
||||
"olmAgent": o.olmConfig.Agent,
|
||||
"orgId": config.OrgID,
|
||||
"userToken": userToken,
|
||||
"fingerprint": o.fingerprint,
|
||||
"postures": o.postures,
|
||||
"pingResults": pingResults,
|
||||
"chainId": generateChainId(), // use a random chainId for registration updates - it won't be used for cancellation since registration is a one-time message but for tracking the session
|
||||
}, 2*time.Second, 20) // after 18 tries on the server side we send the error so dont change this without changing that
|
||||
})
|
||||
|
||||
o.websocket.OnConnect(func() error {
|
||||
logger.Info("Websocket Connected")
|
||||
|
||||
@@ -640,6 +568,8 @@ func (o *Olm) StartTunnel(config TunnelConfig) {
|
||||
return nil
|
||||
}
|
||||
|
||||
publicKey := o.privateKey.PublicKey()
|
||||
|
||||
// delay for 500ms to allow for time for the hp to get processed
|
||||
time.Sleep(500 * time.Millisecond)
|
||||
|
||||
@@ -649,36 +579,19 @@ func (o *Olm) StartTunnel(config TunnelConfig) {
|
||||
return nil
|
||||
}
|
||||
|
||||
if o.stopRegister == nil && o.stopPingRequest == nil {
|
||||
publicKey := o.privateKey.PublicKey()
|
||||
|
||||
pingChainId := generateChainId()
|
||||
o.pendingPingChainId = pingChainId
|
||||
logger.Debug("Requesting exit nodes from server for ping selection")
|
||||
o.stopPingRequest, _ = o.websocket.SendMessageInterval("olm/ping/request", map[string]any{
|
||||
"chainId": pingChainId,
|
||||
}, 3*time.Second, 10)
|
||||
|
||||
// Backwards-compatible one-shot registration, with no pingResults,
|
||||
// for servers that predate the exit node ping dance. Servers that
|
||||
// support it ignore backwardsCompatible register messages (see
|
||||
// handleOlmRegisterMessage server-side) and wait for the real
|
||||
// registration sent from the olm/ping/exitNodes handler above.
|
||||
bcChainId := generateChainId()
|
||||
if err := o.websocket.SendMessage("olm/wg/register", map[string]any{
|
||||
"publicKey": publicKey.String(),
|
||||
"relay": !config.Holepunch,
|
||||
"olmVersion": o.olmConfig.Version,
|
||||
"olmAgent": o.olmConfig.Agent,
|
||||
"orgId": config.OrgID,
|
||||
"userToken": userToken,
|
||||
"fingerprint": o.fingerprint,
|
||||
"postures": o.postures,
|
||||
"backwardsCompatible": true,
|
||||
"chainId": bcChainId,
|
||||
}); err != nil {
|
||||
logger.Error("Failed to send registration message: %v", err)
|
||||
}
|
||||
if o.stopRegister == nil {
|
||||
logger.Debug("Sending registration message to server with public key: %s and relay: %v", publicKey, !config.Holepunch)
|
||||
o.stopRegister, o.updateRegister = o.websocket.SendMessageInterval("olm/wg/register", map[string]any{
|
||||
"publicKey": publicKey.String(),
|
||||
"relay": !config.Holepunch,
|
||||
"olmVersion": o.olmConfig.Version,
|
||||
"olmAgent": o.olmConfig.Agent,
|
||||
"orgId": config.OrgID,
|
||||
"userToken": userToken,
|
||||
"fingerprint": o.fingerprint,
|
||||
"postures": o.postures,
|
||||
"chainId": generateChainId(), // use a random chainId for registration updates - it won't be used for cancellation since registration is a one-time message but for tracking the session
|
||||
}, 2*time.Second, 20) // after 18 tries on the server side we send the error so dont change this without changing that
|
||||
|
||||
// Invoke onRegistered callback if configured
|
||||
if o.olmConfig.OnRegistered != nil {
|
||||
@@ -775,12 +688,6 @@ func (o *Olm) Close() {
|
||||
o.stopRegister = nil
|
||||
}
|
||||
|
||||
if o.stopPingRequest != nil {
|
||||
logger.Debug("Stopping exit node ping request interval")
|
||||
o.stopPingRequest()
|
||||
o.stopPingRequest = nil
|
||||
}
|
||||
|
||||
// Stop all pending peer init and send senders before closing websocket
|
||||
o.peerSendMu.Lock()
|
||||
for _, stop := range o.stopPeerInits {
|
||||
@@ -837,13 +744,6 @@ func (o *Olm) Close() {
|
||||
}
|
||||
o.peerManagerMu.Unlock()
|
||||
|
||||
// The WireGuard device and TUN interface are being torn down below, which takes
|
||||
// the exit node peer and its secondary address with them - just clear the
|
||||
// in-memory record so a stale config isn't reused on the next connect.
|
||||
o.exitNodeMu.Lock()
|
||||
o.exitNode = nil
|
||||
o.exitNodeMu.Unlock()
|
||||
|
||||
if o.uapiListener != nil {
|
||||
_ = o.uapiListener.Close()
|
||||
o.uapiListener = nil
|
||||
|
||||
@@ -10,44 +10,11 @@ type WgData struct {
|
||||
Sites []peers.SiteConfig `json:"sites"`
|
||||
TunnelIP string `json:"tunnelIP"`
|
||||
UtilitySubnet string `json:"utilitySubnet"` // this is for things like the DNS server, and alias addresses
|
||||
ExitNode *ExitNodeConfig `json:"exitNode,omitempty"`
|
||||
}
|
||||
|
||||
// ExitNodeConfig describes an exit node the olm client can connect to for
|
||||
// resources (e.g. inference) hosted on that node, separate from the site
|
||||
// peers. It lives in a different address space than the site tunnel - the
|
||||
// client is assigned TunnelIP (within the exit node's subnet) to reach the
|
||||
// node at ServerIP. It arrives on the initial "olm/wg/connect" message and can
|
||||
// also be sent later via "olm/wg/exitnode/connect" / "olm/wg/exitnode/disconnect"
|
||||
// so the server can direct a client to connect/disconnect after registration.
|
||||
type ExitNodeConfig struct {
|
||||
Connect bool `json:"connect"`
|
||||
Endpoint string `json:"endpoint"`
|
||||
RelayPort uint16 `json:"relayPort"`
|
||||
PublicKey string `json:"publicKey"`
|
||||
ServerIP string `json:"serverIP"`
|
||||
TunnelIP string `json:"tunnelIP"`
|
||||
Aliases []string `json:"aliases,omitempty"`
|
||||
}
|
||||
|
||||
// ExitNodeUpdateData describes a change to data associated with the currently
|
||||
// connected exit node, e.g. when a resource's alias is renamed on the server.
|
||||
// Aliases have no per-alias address here since every exit node alias resolves
|
||||
// to the exit node's own ServerIP. More fields can be added here in the
|
||||
// future as other exit node data becomes updatable.
|
||||
type ExitNodeUpdateData struct {
|
||||
OldAliases []string `json:"oldAliases,omitempty"`
|
||||
NewAliases []string `json:"newAliases,omitempty"`
|
||||
}
|
||||
|
||||
type SyncData struct {
|
||||
Sites []peers.SiteConfig `json:"sites"`
|
||||
ExitNodes []SyncExitNode `json:"exitNodes"`
|
||||
// ExitNode is the exit node the client itself is assigned to (for site
|
||||
// resources hosted on it, e.g. inference), mirroring the ExitNode field
|
||||
// on WgData sent at registration. It is separate from ExitNodes above,
|
||||
// which is only the set of exit nodes used for hole punching.
|
||||
ExitNode *ExitNodeConfig `json:"exitNode,omitempty"`
|
||||
}
|
||||
|
||||
type SyncExitNode struct {
|
||||
|
||||
+6
-32
@@ -44,11 +44,7 @@ type PeerManager struct {
|
||||
peerMonitor *monitor.PeerMonitor
|
||||
dnsProxy *dns.DNSProxy
|
||||
interfaceName string
|
||||
// localIP is our own address on the site tunnel (as opposed to any exit
|
||||
// node's secondary address that may also be present on the interface).
|
||||
// Routes for site peers are pinned to it on darwin - see AddRoutesWithSource.
|
||||
localIP string
|
||||
privateKey wgtypes.Key
|
||||
privateKey wgtypes.Key
|
||||
// allowedIPOwners tracks which peer currently "owns" each allowed IP in WireGuard
|
||||
// key is the CIDR string, value is the siteId that has it configured in WG
|
||||
allowedIPOwners map[string]int
|
||||
@@ -92,7 +88,6 @@ func NewPeerManager(config PeerManagerConfig) *PeerManager {
|
||||
peers: make(map[int]SiteConfig),
|
||||
dnsProxy: config.DNSProxy,
|
||||
interfaceName: config.InterfaceName,
|
||||
localIP: config.LocalIP,
|
||||
privateKey: config.PrivateKey,
|
||||
allowedIPOwners: make(map[string]int),
|
||||
allowedIPClaims: make(map[string]map[int]bool),
|
||||
@@ -132,27 +127,6 @@ func (pm *PeerManager) GetPeerMonitor() *monitor.PeerMonitor {
|
||||
return pm.peerMonitor
|
||||
}
|
||||
|
||||
// SetExitNode starts (or updates) ICMP connectivity monitoring of the given exit node.
|
||||
// tunnelIP is the secondary address assigned to us for this exit node, which the ping
|
||||
// probe must be sourced from since the exit node's WireGuard peer entry only accepts
|
||||
// traffic from that address.
|
||||
func (pm *PeerManager) SetExitNode(serverIP, tunnelIP string) {
|
||||
pm.mu.RLock()
|
||||
defer pm.mu.RUnlock()
|
||||
if pm.peerMonitor != nil {
|
||||
pm.peerMonitor.SetExitNode(serverIP, tunnelIP)
|
||||
}
|
||||
}
|
||||
|
||||
// ClearExitNode stops ICMP connectivity monitoring of the exit node
|
||||
func (pm *PeerManager) ClearExitNode() {
|
||||
pm.mu.RLock()
|
||||
defer pm.mu.RUnlock()
|
||||
if pm.peerMonitor != nil {
|
||||
pm.peerMonitor.ClearExitNode()
|
||||
}
|
||||
}
|
||||
|
||||
// SetPublicDNS replaces the DNS servers used to resolve WireGuard peer
|
||||
// endpoints and hole-punch targets. The servers must be in "host:port" format
|
||||
// (e.g. "8.8.8.8:53"). The change takes effect for all future peer
|
||||
@@ -221,10 +195,10 @@ func (pm *PeerManager) AddPeer(siteConfig SiteConfig) error {
|
||||
return err
|
||||
}
|
||||
|
||||
if err := network.AddRouteForServerIPWithSource(siteConfig.ServerIP, pm.interfaceName, pm.localIP); err != nil {
|
||||
if err := network.AddRouteForServerIP(siteConfig.ServerIP, pm.interfaceName); err != nil {
|
||||
logger.Error("Failed to add route for server IP: %v", err)
|
||||
}
|
||||
if err := network.AddRoutesWithSource(siteConfig.RemoteSubnets, pm.interfaceName, pm.localIP); err != nil {
|
||||
if err := network.AddRoutes(siteConfig.RemoteSubnets, pm.interfaceName); err != nil {
|
||||
logger.Error("Failed to add routes for remote subnets: %v", err)
|
||||
}
|
||||
|
||||
@@ -285,7 +259,7 @@ func (pm *PeerManager) RemovePeer(siteId int) error {
|
||||
return err
|
||||
}
|
||||
|
||||
if err := network.RemoveRouteForServerIPWithSource(peer.ServerIP, pm.interfaceName, pm.localIP); err != nil {
|
||||
if err := network.RemoveRouteForServerIP(peer.ServerIP, pm.interfaceName); err != nil {
|
||||
logger.Error("Failed to remove route for server IP: %v", err)
|
||||
}
|
||||
|
||||
@@ -521,7 +495,7 @@ func (pm *PeerManager) UpdatePeer(siteConfig SiteConfig) error {
|
||||
|
||||
// Add routes for added subnets
|
||||
if len(addedSubnets) > 0 {
|
||||
if err := network.AddRoutesWithSource(addedSubnets, pm.interfaceName, pm.localIP); err != nil {
|
||||
if err := network.AddRoutes(addedSubnets, pm.interfaceName); err != nil {
|
||||
logger.Error("Failed to add routes: %v", err)
|
||||
}
|
||||
}
|
||||
@@ -722,7 +696,7 @@ func (pm *PeerManager) AddRemoteSubnet(siteId int, cidr string) error {
|
||||
}
|
||||
|
||||
// Add route
|
||||
if err := network.AddRoutesWithSource([]string{cidr}, pm.interfaceName, pm.localIP); err != nil {
|
||||
if err := network.AddRoutes([]string{cidr}, pm.interfaceName); err != nil {
|
||||
return err
|
||||
}
|
||||
|
||||
|
||||
@@ -1,279 +0,0 @@
|
||||
package monitor
|
||||
|
||||
import (
|
||||
"bytes"
|
||||
"context"
|
||||
"fmt"
|
||||
"net/netip"
|
||||
"time"
|
||||
|
||||
"github.com/fosrl/newt/logger"
|
||||
"golang.org/x/net/icmp"
|
||||
xipv4 "golang.org/x/net/ipv4"
|
||||
"gvisor.dev/gvisor/pkg/tcpip"
|
||||
gipv4 "gvisor.dev/gvisor/pkg/tcpip/network/ipv4"
|
||||
gstack "gvisor.dev/gvisor/pkg/tcpip/stack"
|
||||
gicmp "gvisor.dev/gvisor/pkg/tcpip/transport/icmp"
|
||||
"gvisor.dev/gvisor/pkg/waiter"
|
||||
)
|
||||
|
||||
const (
|
||||
exitNodePingInterval = 3 * time.Second
|
||||
exitNodePingTimeout = 1 * time.Second
|
||||
exitNodePingMaxAttempts = 3
|
||||
)
|
||||
|
||||
// SetExitNode starts (or, if the exit node changed, restarts) ICMP
|
||||
// connectivity monitoring of the exit node at serverIP. tunnelIP is the
|
||||
// secondary address assigned to us for this exit node (ExitNodeConfig.TunnelIP) -
|
||||
// the exit node's WireGuard peer entry only accepts traffic sourced from that
|
||||
// address, so probes must be sourced from it rather than the site tunnel IP.
|
||||
// Both serverIP and tunnelIP must be bare IP addresses (no CIDR suffix).
|
||||
func (pm *PeerMonitor) SetExitNode(serverIP, tunnelIP string) {
|
||||
pm.exitNodeMu.Lock()
|
||||
if pm.exitNodeCancel != nil && pm.exitNodeServerIP == serverIP && pm.exitNodeTunnelIP == tunnelIP {
|
||||
pm.exitNodeMu.Unlock()
|
||||
return
|
||||
}
|
||||
prevTunnelIP := pm.exitNodeTunnelIP
|
||||
if pm.exitNodeCancel != nil {
|
||||
pm.exitNodeCancel()
|
||||
}
|
||||
pm.exitNodeServerIP = serverIP
|
||||
pm.exitNodeTunnelIP = tunnelIP
|
||||
ctx, cancel := context.WithCancel(context.Background())
|
||||
pm.exitNodeCancel = cancel
|
||||
pm.exitNodeMu.Unlock()
|
||||
|
||||
if prevTunnelIP != "" && prevTunnelIP != tunnelIP {
|
||||
pm.removeExitNodeAddress(prevTunnelIP)
|
||||
}
|
||||
if tunnelIP != prevTunnelIP {
|
||||
if err := pm.addExitNodeAddress(tunnelIP); err != nil {
|
||||
logger.Error("Failed to register exit node tunnel address %s: %v", tunnelIP, err)
|
||||
}
|
||||
}
|
||||
|
||||
logger.Info("Started exit node connectivity monitor for %s (via %s)", serverIP, tunnelIP)
|
||||
go pm.runExitNodeMonitor(ctx, serverIP, tunnelIP)
|
||||
}
|
||||
|
||||
// ClearExitNode stops ICMP monitoring of the exit node and clears its status
|
||||
// from the API.
|
||||
func (pm *PeerMonitor) ClearExitNode() {
|
||||
pm.exitNodeMu.Lock()
|
||||
if pm.exitNodeCancel != nil {
|
||||
pm.exitNodeCancel()
|
||||
pm.exitNodeCancel = nil
|
||||
}
|
||||
tunnelIP := pm.exitNodeTunnelIP
|
||||
pm.exitNodeServerIP = ""
|
||||
pm.exitNodeTunnelIP = ""
|
||||
pm.exitNodeMu.Unlock()
|
||||
|
||||
if tunnelIP != "" {
|
||||
pm.removeExitNodeAddress(tunnelIP)
|
||||
}
|
||||
|
||||
if pm.apiServer != nil {
|
||||
pm.apiServer.ClearExitNodeStatus()
|
||||
}
|
||||
|
||||
logger.Info("Stopped exit node connectivity monitor")
|
||||
}
|
||||
|
||||
// addExitNodeAddress registers tunnelIP as a protocol address on the peer
|
||||
// monitor's netstack NIC and adds a MiddleDevice rule so ICMP replies destined
|
||||
// to it are intercepted and redirected into the netstack instead of being
|
||||
// delivered to the host TUN device.
|
||||
func (pm *PeerMonitor) addExitNodeAddress(tunnelIP string) error {
|
||||
pm.mutex.Lock()
|
||||
st := pm.stack
|
||||
pm.mutex.Unlock()
|
||||
|
||||
if st == nil {
|
||||
return fmt.Errorf("netstack not initialized")
|
||||
}
|
||||
|
||||
addr, err := netip.ParseAddr(tunnelIP)
|
||||
if err != nil {
|
||||
return fmt.Errorf("invalid tunnel IP: %w", err)
|
||||
}
|
||||
|
||||
protoAddr := tcpip.ProtocolAddress{
|
||||
Protocol: gipv4.ProtocolNumber,
|
||||
AddressWithPrefix: tcpip.AddrFrom4(addr.As4()).WithPrefix(),
|
||||
}
|
||||
if tcpipErr := st.AddProtocolAddress(1, protoAddr, gstack.AddressProperties{}); tcpipErr != nil {
|
||||
return fmt.Errorf("failed to add protocol address: %s", tcpipErr)
|
||||
}
|
||||
|
||||
pm.middleDev.AddRule(addr, pm.handlePacket)
|
||||
return nil
|
||||
}
|
||||
|
||||
// removeExitNodeAddress undoes addExitNodeAddress.
|
||||
func (pm *PeerMonitor) removeExitNodeAddress(tunnelIP string) {
|
||||
addr, err := netip.ParseAddr(tunnelIP)
|
||||
if err != nil {
|
||||
return
|
||||
}
|
||||
|
||||
pm.middleDev.RemoveRule(addr)
|
||||
|
||||
pm.mutex.Lock()
|
||||
st := pm.stack
|
||||
pm.mutex.Unlock()
|
||||
|
||||
if st != nil {
|
||||
st.RemoveAddress(1, tcpip.AddrFrom4(addr.As4()))
|
||||
}
|
||||
}
|
||||
|
||||
// runExitNodeMonitor periodically pings the exit node and reports its status
|
||||
// to the API server until ctx is cancelled.
|
||||
func (pm *PeerMonitor) runExitNodeMonitor(ctx context.Context, serverIP, tunnelIP string) {
|
||||
check := func() {
|
||||
var (
|
||||
connected bool
|
||||
rtt time.Duration
|
||||
)
|
||||
for attempt := 0; attempt < exitNodePingMaxAttempts; attempt++ {
|
||||
if d, err := pm.pingExitNode(serverIP, tunnelIP, exitNodePingTimeout); err == nil {
|
||||
connected = true
|
||||
rtt = d
|
||||
break
|
||||
} else {
|
||||
logger.Debug("Exit node ping attempt %d/%d to %s failed: %v", attempt+1, exitNodePingMaxAttempts, serverIP, err)
|
||||
}
|
||||
select {
|
||||
case <-ctx.Done():
|
||||
return
|
||||
default:
|
||||
}
|
||||
}
|
||||
if pm.apiServer != nil {
|
||||
pm.apiServer.SetExitNodeStatus(connected, rtt, serverIP)
|
||||
}
|
||||
}
|
||||
|
||||
check()
|
||||
|
||||
ticker := time.NewTicker(exitNodePingInterval)
|
||||
defer ticker.Stop()
|
||||
for {
|
||||
select {
|
||||
case <-ctx.Done():
|
||||
return
|
||||
case <-ticker.C:
|
||||
check()
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
// pingExitNode sends a single ICMP echo request from localTunnelIP to dst and
|
||||
// waits up to timeout for the matching reply. The request is built and read
|
||||
// directly on the peer monitor's gvisor netstack, so it's injected into (and
|
||||
// intercepted from) the WireGuard device via MiddleDevice - it never touches
|
||||
// the host's real network stack, matching how the UDP peer tests above work.
|
||||
func (pm *PeerMonitor) pingExitNode(dst, localTunnelIP string, timeout time.Duration) (time.Duration, error) {
|
||||
pm.mutex.Lock()
|
||||
st := pm.stack
|
||||
pm.mutex.Unlock()
|
||||
|
||||
if st == nil {
|
||||
return 0, fmt.Errorf("netstack not initialized")
|
||||
}
|
||||
|
||||
dstAddr, err := netip.ParseAddr(dst)
|
||||
if err != nil {
|
||||
return 0, fmt.Errorf("invalid destination address: %w", err)
|
||||
}
|
||||
localAddr, err := netip.ParseAddr(localTunnelIP)
|
||||
if err != nil {
|
||||
return 0, fmt.Errorf("invalid local address: %w", err)
|
||||
}
|
||||
|
||||
var wq waiter.Queue
|
||||
ep, tcpipErr := st.NewEndpoint(gicmp.ProtocolNumber4, gipv4.ProtocolNumber, &wq)
|
||||
if tcpipErr != nil {
|
||||
return 0, fmt.Errorf("failed to create ICMP endpoint: %s", tcpipErr)
|
||||
}
|
||||
defer ep.Close()
|
||||
|
||||
if tcpipErr := ep.Bind(tcpip.FullAddress{NIC: 1, Addr: tcpip.AddrFromSlice(localAddr.AsSlice())}); tcpipErr != nil {
|
||||
return 0, fmt.Errorf("failed to bind ICMP endpoint: %s", tcpipErr)
|
||||
}
|
||||
|
||||
// gvisor's ICMP endpoint overwrites whatever Identifier we put in the outgoing
|
||||
// echo with its own bound "port" (assigned above by Bind), and demuxes incoming
|
||||
// Echo Replies by that same value - so we must use it, not one we generate
|
||||
// ourselves, both for the outgoing message and to register with handlePacket's
|
||||
// filter below.
|
||||
laddr, tcpipErr := ep.GetLocalAddress()
|
||||
if tcpipErr != nil {
|
||||
return 0, fmt.Errorf("failed to get local ICMP endpoint address: %s", tcpipErr)
|
||||
}
|
||||
echoID := int(laddr.Port)
|
||||
|
||||
pm.portsLock.Lock()
|
||||
pm.activeICMPIdents[laddr.Port] = true
|
||||
pm.portsLock.Unlock()
|
||||
defer func() {
|
||||
pm.portsLock.Lock()
|
||||
delete(pm.activeICMPIdents, laddr.Port)
|
||||
pm.portsLock.Unlock()
|
||||
}()
|
||||
|
||||
if tcpipErr := ep.Connect(tcpip.FullAddress{NIC: 1, Addr: tcpip.AddrFromSlice(dstAddr.AsSlice())}); tcpipErr != nil {
|
||||
return 0, fmt.Errorf("failed to connect ICMP endpoint: %s", tcpipErr)
|
||||
}
|
||||
|
||||
requestPing := icmp.Echo{
|
||||
ID: echoID,
|
||||
Seq: 1,
|
||||
Data: []byte("olmping"),
|
||||
}
|
||||
icmpBytes, err := (&icmp.Message{Type: xipv4.ICMPTypeEcho, Code: 0, Body: &requestPing}).Marshal(nil)
|
||||
if err != nil {
|
||||
return 0, fmt.Errorf("failed to marshal ICMP message: %w", err)
|
||||
}
|
||||
|
||||
waitEntry, notifyCh := waiter.NewChannelEntry(waiter.EventIn)
|
||||
wq.EventRegister(&waitEntry)
|
||||
defer wq.EventUnregister(&waitEntry)
|
||||
|
||||
start := time.Now()
|
||||
|
||||
if _, tcpipErr := ep.Write(bytes.NewReader(icmpBytes), tcpip.WriteOptions{}); tcpipErr != nil {
|
||||
return 0, fmt.Errorf("failed to write ICMP echo request: %s", tcpipErr)
|
||||
}
|
||||
|
||||
deadline := time.NewTimer(timeout)
|
||||
defer deadline.Stop()
|
||||
|
||||
readBuf := make([]byte, 1500)
|
||||
for {
|
||||
select {
|
||||
case <-deadline.C:
|
||||
return 0, fmt.Errorf("ping to %s timed out", dst)
|
||||
case <-notifyCh:
|
||||
w := tcpip.SliceWriter(readBuf)
|
||||
res, tcpipErr := ep.Read(&w, tcpip.ReadOptions{})
|
||||
if tcpipErr != nil {
|
||||
continue
|
||||
}
|
||||
|
||||
reply, err := icmp.ParseMessage(1, readBuf[:res.Count])
|
||||
if err != nil {
|
||||
continue
|
||||
}
|
||||
replyEcho, ok := reply.Body.(*icmp.Echo)
|
||||
if !ok || replyEcho.ID != echoID || replyEcho.Seq != requestPing.Seq {
|
||||
continue
|
||||
}
|
||||
|
||||
return time.Since(start), nil
|
||||
}
|
||||
}
|
||||
}
|
||||
+13
-76
@@ -3,7 +3,6 @@ package monitor
|
||||
import (
|
||||
"context"
|
||||
"crypto/rand"
|
||||
"encoding/binary"
|
||||
"encoding/hex"
|
||||
"fmt"
|
||||
"net"
|
||||
@@ -26,7 +25,6 @@ import (
|
||||
"gvisor.dev/gvisor/pkg/tcpip/network/ipv4"
|
||||
"gvisor.dev/gvisor/pkg/tcpip/network/ipv6"
|
||||
"gvisor.dev/gvisor/pkg/tcpip/stack"
|
||||
"gvisor.dev/gvisor/pkg/tcpip/transport/icmp"
|
||||
"gvisor.dev/gvisor/pkg/tcpip/transport/udp"
|
||||
)
|
||||
|
||||
@@ -107,20 +105,6 @@ type PeerMonitor struct {
|
||||
wgConnectionStatus map[int]bool // siteID -> WG connected status
|
||||
wgConnectionRTT map[int]time.Duration // siteID -> last known RTT
|
||||
statusChangeCallback func(siteId int) // called when any peer's connection status changes
|
||||
|
||||
// Exit node ICMP monitoring fields. The exit node is a single peer (not a
|
||||
// site), pinged over the same gvisor netstack used for the peer UDP tests
|
||||
// above, so the probe never touches the host's real network stack - it's
|
||||
// injected directly into the WireGuard device via MiddleDevice.
|
||||
exitNodeMu sync.Mutex
|
||||
exitNodeServerIP string
|
||||
exitNodeTunnelIP string
|
||||
exitNodeCancel context.CancelFunc
|
||||
|
||||
// activeICMPIdents tracks the ICMP identifiers of our own in-flight exit-node
|
||||
// ping probes (guarded by portsLock, alongside activePorts), so handlePacket
|
||||
// only intercepts Echo Replies that are actually ours.
|
||||
activeICMPIdents map[uint16]bool
|
||||
}
|
||||
|
||||
// NewPeerMonitor creates a new peer monitor with the given callback
|
||||
@@ -141,7 +125,6 @@ func NewPeerMonitor(wsClient *websocket.Client, middleDev *middleDevice.MiddleDe
|
||||
localIP: localIP,
|
||||
publicDNS: publicDNS,
|
||||
activePorts: make(map[uint16]bool),
|
||||
activeICMPIdents: make(map[uint16]bool),
|
||||
nsCtx: ctx,
|
||||
nsCancel: cancel,
|
||||
sharedBind: sharedBind,
|
||||
@@ -1169,14 +1152,6 @@ func (pm *PeerMonitor) Close() {
|
||||
// Stop holepunch monitor first (outside of mutex to avoid deadlock)
|
||||
pm.stopHolepunchMonitor()
|
||||
|
||||
// Stop exit node ICMP monitor, if running
|
||||
pm.exitNodeMu.Lock()
|
||||
if pm.exitNodeCancel != nil {
|
||||
pm.exitNodeCancel()
|
||||
pm.exitNodeCancel = nil
|
||||
}
|
||||
pm.exitNodeMu.Unlock()
|
||||
|
||||
// Stop all pending relay senders
|
||||
pm.relaySendMu.Lock()
|
||||
for chainId, stop := range pm.relaySends {
|
||||
@@ -1313,7 +1288,7 @@ func (pm *PeerMonitor) initNetstack() error {
|
||||
// Create gvisor netstack
|
||||
stackOpts := stack.Options{
|
||||
NetworkProtocols: []stack.NetworkProtocolFactory{ipv4.NewProtocol, ipv6.NewProtocol},
|
||||
TransportProtocols: []stack.TransportProtocolFactory{udp.NewProtocol, icmp.NewProtocol4, icmp.NewProtocol6},
|
||||
TransportProtocols: []stack.TransportProtocolFactory{udp.NewProtocol},
|
||||
HandleLocal: true,
|
||||
}
|
||||
|
||||
@@ -1354,64 +1329,26 @@ func (pm *PeerMonitor) initNetstack() error {
|
||||
return nil
|
||||
}
|
||||
|
||||
// icmpv4EchoReplyIdent returns the ICMP identifier of packet if it is an IPv4
|
||||
// ICMP Echo Reply (type 0), so it can be matched against our own in-flight
|
||||
// exit-node ping probes before being pulled off the host's real traffic path.
|
||||
func icmpv4EchoReplyIdent(packet []byte) (uint16, bool) {
|
||||
if len(packet) < 20 || packet[0]>>4 != 4 {
|
||||
return 0, false
|
||||
}
|
||||
ihl := int(packet[0]&0x0f) * 4
|
||||
if ihl < 20 || len(packet) < ihl+8 {
|
||||
return 0, false
|
||||
}
|
||||
const icmpEchoReply = 0
|
||||
if packet[ihl] != icmpEchoReply {
|
||||
return 0, false
|
||||
}
|
||||
return binary.BigEndian.Uint16(packet[ihl+4 : ihl+6]), true
|
||||
}
|
||||
|
||||
// handlePacket is called by MiddleDevice when a packet arrives for our IP
|
||||
func (pm *PeerMonitor) handlePacket(packet []byte) bool {
|
||||
// Check if it's UDP
|
||||
proto, ok := util.GetProtocol(packet)
|
||||
if !ok || proto != 17 { // UDP
|
||||
return false
|
||||
}
|
||||
|
||||
// Check destination port
|
||||
port, ok := util.GetDestPort(packet)
|
||||
if !ok {
|
||||
return false
|
||||
}
|
||||
|
||||
switch proto {
|
||||
case 1: // ICMPv4 - only intercept Echo Replies matching one of our own active
|
||||
// exit-node ping probes, identified by the ICMP identifier field. Anything
|
||||
// else (including real ICMP traffic to/from the host, e.g. `ping`) must be
|
||||
// left alone so it reaches the host TUN normally.
|
||||
ident, ok := icmpv4EchoReplyIdent(packet)
|
||||
if !ok {
|
||||
return false
|
||||
}
|
||||
// Check if we are listening on this port
|
||||
pm.portsLock.RLock()
|
||||
active := pm.activePorts[uint16(port)]
|
||||
pm.portsLock.RUnlock()
|
||||
|
||||
pm.portsLock.RLock()
|
||||
active := pm.activeICMPIdents[ident]
|
||||
pm.portsLock.RUnlock()
|
||||
|
||||
if !active {
|
||||
return false
|
||||
}
|
||||
case 17: // UDP
|
||||
// Check destination port
|
||||
port, ok := util.GetDestPort(packet)
|
||||
if !ok {
|
||||
return false
|
||||
}
|
||||
|
||||
// Check if we are listening on this port
|
||||
pm.portsLock.RLock()
|
||||
active := pm.activePorts[uint16(port)]
|
||||
pm.portsLock.RUnlock()
|
||||
|
||||
if !active {
|
||||
return false
|
||||
}
|
||||
default:
|
||||
if !active {
|
||||
return false
|
||||
}
|
||||
|
||||
|
||||
Reference in New Issue
Block a user