Files
lmvpn_client/internal/vpn/session.go
T
kevin ebe082500f
Release / build-macos (push) Canceled after 0s
Release / build-windows (push) Canceled after 0s
Release / release (push) Canceled after 0s
perf: 路由批量执行+CIDR聚合+并行脚本+连接步骤显示+CIDR刷新按钮
路由性能优化:
- 修复 macOS route 命令语法错误(去掉 -net 参数)
- 路由应用拆分为 Essential(ready前,4-6条命令<1s)+ Deferred(ready后,批量脚本)
- 新增 CIDR 聚合算法(cidrmerge.go),合并相邻 CIDR 块减少路由数量(10826->7500)
- 批量脚本执行:3平台各实现8个批量函数,临时脚本单次执行替代逐条进程创建
- 并行批量:拆为4个子shell并行执行,总时间降为1/4

连接步骤显示:
- stats 新增 connectStep + cidrError 原子变量
- 连接生命周期中设置步骤(fetch_cidrs/connecting/load_routes),UI 实时显示
- after-proxy CIDR 获取期间显示加载状态,获取失败显示错误提示

CIDR 刷新按钮:
- ipc 新增 CmdRefreshCIDR 命令
- daemon 处理刷新请求,session.RefreshCIDRs() 重新获取+添加路由
- UI 状态卡新增刷新按钮,proxy/bypass 模式已连接时显示

README:
- 补充 URL 获取时机注意事项(bypass模式after-proxy可能失败)
- 补充 CIDR 聚合与批量执行说明
2026-07-09 09:44:49 +08:00

1125 lines
33 KiB
Go

// Package vpn orchestrates the full VPN session lifecycle: transport
// connection, TUN device setup, route management, the bidirectional
// packet pump, and automatic reconnection with exponential backoff.
//
// A SessionManager runs in its own goroutine once Connect is called.
// State changes and periodic stats are reported via callbacks, making
// it suitable for both the daemon (IPC) and headless use.
package vpn
import (
"context"
"crypto/tls"
"errors"
"fmt"
"net"
"net/http"
"strings"
"sync"
"sync/atomic"
"time"
"lmvpn/internal/auth"
"lmvpn/internal/cidrsource"
"lmvpn/internal/log"
"lmvpn/internal/model"
"lmvpn/internal/protocol"
"lmvpn/internal/route"
"lmvpn/internal/stats"
"lmvpn/internal/tlsconfig"
"lmvpn/internal/transport"
"lmvpn/internal/tun"
)
// SessionConfig describes how to connect to a VPN server.
type SessionConfig struct {
ServerURL string
SNIHost string // TLS SNI hostname for CDN
ServerIPs []string // CDN edge IPs for failover
Username string
Password string
AuthMode model.AuthMode
Token string // pre-obtained JWT (empty = fetch via HTTP login)
RoutingMode route.Mode
CIDRV4 []string // static IPv4 CIDRs (proxy/bypass mode)
CIDRV6 []string // static IPv6 CIDRs (proxy/bypass mode)
CIDRV4URLs []model.CIDRURLSource // IPv4 CIDR URL sources
CIDRV6URLs []model.CIDRURLSource // IPv6 CIDR URL sources
MTUOverride int // 0 = use server MTU
TLSCACert string // inline CA cert PEM (wss only)
TLSCAPath string // CA cert file path (wss only)
TLSInsecure bool // skip cert verification (wss only)
TLSPinnedHash string // SHA-256 cert pin (wss only)
}
// SessionManager manages a single VPN session with auto-reconnect.
type SessionManager struct {
stats *stats.Stats
onState func(stats.State)
onStats func(stats.Snapshot)
onError func(code string, msg string)
mu sync.Mutex
running bool
cancel context.CancelFunc
dev tun.Device
routeMgr *route.Manager
conn *transport.Conn
done chan struct{}
// CIDR hit tracking. Set during setupTUN, cleared on cleanup.
cidrTracker *cidrTracker
// lastCfg stores the session config for RefreshCIDRs.
lastCfg SessionConfig
// EWMA speed smoothing state. Only touched by reportStats (single
// goroutine), so no lock needed. ewma* fields hold the smoothed
// bytes/sec; prev* hold the last snapshot's cumulative counters and
// tick time for delta computation.
ewmaRxV4 float64
ewmaTxV4 float64
ewmaRxV6 float64
ewmaTxV6 float64
prevSnap stats.Snapshot
prevTick time.Time
speedReady bool
}
// New creates a SessionManager. The onState callback (if non-nil) is
// invoked on every state transition. The onStats callback (if non-nil)
// is invoked periodically while connected. The onError callback (if
// non-nil) is invoked once when a fatal, non-retryable error (such as
// an authentication failure) terminates the session.
func New(onState func(stats.State), onStats func(stats.Snapshot), onError func(string, string)) *SessionManager {
return &SessionManager{
stats: stats.New(),
onState: onState,
onStats: onStats,
onError: onError,
}
}
// Stats returns the live stats handle.
func (sm *SessionManager) Stats() *stats.Stats { return sm.stats }
// State returns the current session state.
func (sm *SessionManager) State() stats.State { return sm.stats.State() }
// Connect starts the VPN session. It returns immediately; the session
// runs in a background goroutine until Disconnect is called or the
// context is cancelled. If already running, it returns an error.
func (sm *SessionManager) Connect(ctx context.Context, cfg SessionConfig) error {
sm.mu.Lock()
if sm.running {
sm.mu.Unlock()
return errors.New("session already running")
}
ctx, cancel := context.WithCancel(ctx)
sm.cancel = cancel
sm.running = true
sm.done = make(chan struct{})
sm.lastCfg = cfg
sm.mu.Unlock()
go sm.run(ctx, cfg)
return nil
}
// Disconnect stops the session and cleans up resources. It blocks
// until the session has fully shut down.
func (sm *SessionManager) Disconnect() {
sm.mu.Lock()
if !sm.running {
sm.mu.Unlock()
return
}
sm.running = false
cancel := sm.cancel
// Extract resources so we can clean them up outside the lock.
// Setting them to nil here also prevents the run goroutine's
// cleanup from double-cleaning (it will see nil and skip).
routeMgr := sm.routeMgr
conn := sm.conn
dev := sm.dev
sm.routeMgr = nil
sm.conn = nil
sm.dev = nil
sm.cidrTracker = nil
sm.mu.Unlock()
if cancel != nil {
cancel()
}
// CRITICAL: remove routes BEFORE closing the TUN device. If the TUN
// is closed first, the /1 cover routes still point at the dead TUN
// and all traffic is blackholed until routeMgr.Cleanup() runs -
// this causes a brief network outage (browsers, DNS, etc.).
if routeMgr != nil {
if err := routeMgr.Cleanup(); err != nil {
log.L().Error("route cleanup error during disconnect", "error", err)
}
}
// Now safe to close the transport and TUN device.
if conn != nil {
conn.Close()
}
if dev != nil {
dev.Close()
}
// Wait for the run goroutine to fully exit. By now it should be
// unblocked (ctx cancelled, conn/dev closed) and will see nil
// routeMgr/conn/dev in cleanup, so it won't double-clean.
if sm.done != nil {
<-sm.done
}
}
// run is the main session loop with exponential-backoff reconnection
// and CDN IP failover.
func (sm *SessionManager) run(ctx context.Context, cfg SessionConfig) {
fatal := false
defer close(sm.done)
defer func() {
if !fatal && ctx.Err() == nil {
sm.setState(stats.StateDisconnected)
}
}()
// Fetch "before-proxy" CIDR lists ONCE before the reconnect loop.
// These HTTP requests go through the physical NIC (routes are
// clean). The result is reused across reconnection attempts so we
// don't re-fetch on every retry. This runs outside the handshake
// to avoid consuming the server's 30s ReadyTimeout budget.
var beforeCIDRs []string
allURLSources := append(append([]model.CIDRURLSource{}, cfg.CIDRV4URLs...), cfg.CIDRV6URLs...)
if len(allURLSources) > 0 {
sm.stats.SetConnectStep("fetch_cidrs")
sm.setState(stats.StateConnecting)
log.L().Info("fetching before-proxy CIDR lists", "url_count", len(allURLSources))
fetched, err := cidrsource.FetchBeforeProxy(ctx, allURLSources)
if err != nil {
log.L().Error("fetch before-proxy CIDR lists failed (continuing)", "error", err)
}
beforeCIDRs = fetched
log.L().Info("before-proxy CIDR lists ready", "total_cidrs", len(beforeCIDRs))
sm.stats.SetConnectStep("")
}
backoff := time.Second
maxBackoff := 60 * time.Second
// Build the full target list: original host first, then CDN IPs.
targets := append([]string{""}, cfg.ServerIPs...) // "" = use base URL
ipIndex := 0
for {
if ctx.Err() != nil {
sm.cleanup()
return
}
targetIP := ""
if ipIndex > 0 && ipIndex < len(targets) {
targetIP = targets[ipIndex]
}
err := sm.connectOnce(ctx, cfg, targetIP, beforeCIDRs)
if ctx.Err() != nil {
sm.cleanup()
return
}
if err != nil {
log.L().Error("VPN connection failed", "error", err)
// Safety net: ensure no TUN/routes leak from a failed
// attempt. connectOnce should have already cleaned up, but
// this guards against any path that returned early.
sm.cleanupResources()
// A TLS certificate verification failure on the original
// hostname (ipIndex == 0) is not retryable: the cert won't
// change between attempts, so stop the loop and surface the
// reason to the user. On a CDN edge IP (ipIndex > 0) the
// TLS error likely means that IP points to a different
// server; skip it and try the next target.
if tlsconfig.IsTLSError(err) {
if ipIndex == 0 {
log.L().Warn("fatal TLS error, stopping reconnect", "error", err)
sm.setState(stats.StateError)
if sm.onError != nil {
sm.onError("tls_error", err.Error())
}
fatal = true
sm.cleanup()
return
}
log.L().Warn("TLS error on CDN IP, skipping",
"index", ipIndex, "ip", targets[ipIndex], "error", err)
}
// A fatal authentication failure (wrong password, disabled
// account, expired token, rate limit) on the original
// hostname is not retryable. On a CDN edge IP it likely
// means the IP points to a different server that returned
// 401/403, so skip it instead of stopping the loop.
if code, msg, isFatal := fatalAuthError(err); isFatal {
if ipIndex == 0 {
log.L().Warn("fatal auth error, stopping reconnect", "code", code, "message", msg)
sm.setState(stats.StateError)
if sm.onError != nil {
sm.onError(string(code), msg)
}
fatal = true
sm.cleanup()
return
}
log.L().Warn("auth error on CDN IP, skipping",
"index", ipIndex, "ip", targets[ipIndex], "code", code)
}
sm.setState(stats.StateReconnecting)
// Try next CDN IP immediately.
ipIndex++
if ipIndex < len(targets) {
log.L().Info("trying next CDN IP", "index", ipIndex, "ip", targets[ipIndex])
continue
}
// All targets exhausted; reset and wait with backoff.
ipIndex = 0
} else {
sm.setState(stats.StateReconnecting)
ipIndex = 0
}
select {
case <-ctx.Done():
sm.cleanup()
return
case <-time.After(backoff):
backoff *= 2
if backoff > maxBackoff {
backoff = maxBackoff
}
}
}
}
// fatalAuthError inspects err and, if it represents a permanent
// authentication failure that should not be retried, returns the
// stable error code, the raw server message, and true. It recognises
// all three auth-failure shapes produced by the transport/auth layers:
// - *transport.AuthError (WebSocket auth_err at the auth stage)
// - *transport.ServerError (auth_err at the init stage, JWT path)
// - *auth.LoginError (HTTP /api/login failure, JWT path)
//
// errors.As transparently unwraps fmt.Errorf("...: %w", err) chains,
// so the wrapped LoginError returned by connectOnce is matched too.
// A non-empty code is required: an auth_err with an unrecognized
// message is treated as non-fatal so the loop falls back to retrying
// (the server still closed the connection, but we lack a categorical
// reason to give up).
func fatalAuthError(err error) (protocol.AuthErrorCode, string, bool) {
var authErr *transport.AuthError
if errors.As(err, &authErr) {
return authErr.Code, authErr.Message, authErr.Code != ""
}
var serverErr *transport.ServerError
if errors.As(err, &serverErr) && serverErr.Type == protocol.TypeAuthErr {
return serverErr.Code, serverErr.Message, serverErr.Code != ""
}
var loginErr *auth.LoginError
if errors.As(err, &loginErr) {
switch loginErr.Code {
case http.StatusTooManyRequests:
return protocol.AuthCodeRateLimited, loginErr.Message, true
case http.StatusUnauthorized, http.StatusForbidden:
return protocol.AuthCodeWrongCredentials, loginErr.Message, true
}
}
return "", "", false
}
// connectOnce performs a single connection lifecycle: authenticate,
// handshake, configure TUN, apply routes, pump packets until failure.
// beforeCIDRs contains CIDRs pre-fetched from "before" timing URL
// sources (fetched once in run(), reused across reconnection attempts).
func (sm *SessionManager) connectOnce(ctx context.Context, cfg SessionConfig, targetIP string, beforeCIDRs []string) error {
sm.setState(stats.StateConnecting)
sm.stats.SetConnectStep("connecting")
// Build URL for this attempt. If targetIP is set (CDN failover),
// build a URL with that IP. Otherwise use base ServerURL.
serverURL := cfg.ServerURL
if targetIP != "" {
serverURL = replaceHost(cfg.ServerURL, targetIP)
}
// Build TLS config for wss:// connections. For ws:// there is no
// TLS layer, so tlsCfg remains nil and both the HTTP client and
// the WebSocket dialer use their default (plaintext) behaviour.
var tlsCfg *tls.Config
if strings.HasPrefix(serverURL, "wss://") {
serverName := cfg.SNIHost
if serverName == "" {
serverName = serverHostFromURL(cfg.ServerURL)
}
var err error
tlsCfg, err = tlsconfig.Build(tlsconfig.Config{
ServerName: serverName,
CACertPEM: cfg.TLSCACert,
CACertPath: cfg.TLSCAPath,
InsecureSkipVerify: cfg.TLSInsecure,
PinnedCertHash: cfg.TLSPinnedHash,
})
if err != nil {
return fmt.Errorf("tls config: %w", err)
}
}
// Determine auth strategy and obtain JWT if needed.
token := cfg.Token
if token == "" && (cfg.AuthMode == model.AuthModeJWT || cfg.AuthMode == model.AuthModeBoth) {
httpBase, err := wsURLToHTTP(serverURL)
if err != nil {
return fmt.Errorf("parse server URL: %w", err)
}
result, err := auth.Login(ctx, httpBase, cfg.Username, cfg.Password, tlsCfg)
if err != nil {
if cfg.AuthMode == model.AuthModeBoth {
// Fall back to password auth.
token = ""
} else {
return fmt.Errorf("login: %w", err)
}
} else {
token = result.Token
}
}
// Prepare the TUN + route setup callback (called during handshake,
// between receiving init and sending ready).
handshake := transport.HandshakeConfig{
ServerURL: serverURL,
SNIHost: cfg.SNIHost,
Token: token,
Username: cfg.Username,
Password: cfg.Password,
OnInit: func(init protocol.InitMessage) error {
return sm.setupTUN(init, cfg, beforeCIDRs)
},
TLSConfig: tlsCfg,
}
// Attempt JWT connection first; fall back to password on auth error.
conn, err := transport.Connect(ctx, handshake)
if err != nil {
if cfg.AuthMode == model.AuthModeBoth && token != "" {
// JWT failure can surface as AuthError (password-auth path)
// or ServerError with Type=auth_err (JWT rejected at /ws).
var authErr *transport.AuthError
var serverErr *transport.ServerError
if errors.As(err, &authErr) ||
(errors.As(err, &serverErr) && serverErr.Type == protocol.TypeAuthErr) {
log.L().Info("JWT auth failed, falling back to password auth", "error", err)
// Clean up any resources created by the failed attempt's
// setupTUN before retrying.
sm.cleanupResources()
handshake.Token = ""
conn, err = transport.Connect(ctx, handshake)
}
}
if err != nil {
// Clean up TUN/routes if setupTUN ran but the handshake
// failed (e.g. server closed the connection after
// ReadyTimeout). Without this, leaked /1 cover routes
// blackhole all traffic and prevent reconnection.
sm.cleanupResources()
return err
}
}
sm.mu.Lock()
sm.conn = conn
sm.mu.Unlock()
sm.stats.SetConnected(conn.Init().IP, conn.Init().IP6)
sm.stats.SetConnectStep("load_routes")
sm.setState(stats.StateConnected)
log.L().Info("VPN connected",
"ip", conn.Init().IP, "server_ip", conn.Init().ServerIP,
"ip6", conn.Init().IP6, "server_ip6", conn.Init().ServerIP6,
"mtu", conn.Init().MTU)
// Start stats reporter.
statsDone := make(chan struct{})
go sm.reportStats(statsDone, ctx)
// Apply deferred routes (user CIDRs) and fetch after-proxy CIDR
// lists in a background goroutine. For proxy/bypass modes with
// thousands of CIDRs this uses batch script execution (~3-5s).
go func() {
sm.mu.Lock()
mgr := sm.routeMgr
sm.mu.Unlock()
if mgr != nil {
sm.stats.SetRouteLoading(true)
sm.stats.SetConnectStep("load_routes")
log.L().Info("applying deferred routes")
if err := mgr.ApplyDeferred(); err != nil {
log.L().Error("apply deferred routes failed (continuing)", "error", err)
}
sm.stats.SetRouteLoading(false)
sm.stats.SetConnectStep("")
log.L().Info("deferred routes applied")
}
sm.fetchAfterProxyCIDRs(ctx, cfg)
}()
// Run the packet pump (blocks until connection breaks).
sm.pumpPackets(ctx, conn)
close(statsDone)
sm.cleanup()
return nil
}
// setupTUN creates and configures the TUN device and applies routes.
// This is called by the transport during the handshake, between init
// and ready. It performs NO network calls (HTTP/DNS) so it completes
// in milliseconds and never exceeds the server's ReadyTimeout.
//
// beforeCIDRs contains CIDRs fetched from "before" timing URL sources,
// pre-fetched by connectOnce before the handshake began.
func (sm *SessionManager) setupTUN(init protocol.InitMessage, cfg SessionConfig, beforeCIDRs []string) error {
dev, err := tun.Create("")
if err != nil {
return fmt.Errorf("create tun: %w", err)
}
sm.mu.Lock()
sm.dev = dev
sm.mu.Unlock()
localIP := net.ParseIP(init.IP)
peerIP := net.ParseIP(init.ServerIP)
if localIP == nil || peerIP == nil {
return fmt.Errorf("invalid init IPs: %s / %s", init.IP, init.ServerIP)
}
if err := dev.Configure(localIP, init.Prefix, peerIP); err != nil {
return fmt.Errorf("configure tun: %w", err)
}
// Configure IPv6 address when the server assigned one (dual-stack).
hasV6 := init.IP6 != ""
if hasV6 {
ip6 := net.ParseIP(init.IP6)
if ip6 == nil {
return fmt.Errorf("invalid init IPv6: %s", init.IP6)
}
if err := dev.ConfigureIPv6(ip6, init.Prefix6); err != nil {
return fmt.Errorf("configure tun ipv6: %w", err)
}
}
mtu := init.MTU
if cfg.MTUOverride > 0 {
mtu = cfg.MTUOverride
}
if err := dev.SetMTU(mtu); err != nil {
return fmt.Errorf("set mtu: %w", err)
}
// Merge static CIDRs with pre-fetched before-proxy CIDRs.
var cidrs []string
cidrs = append(cidrs, cfg.CIDRV4...)
cidrs = append(cidrs, cfg.CIDRV6...)
cidrs = append(cidrs, beforeCIDRs...)
// Apply routing.
routeCfg := route.Config{
Mode: cfg.RoutingMode,
InterfaceName: dev.Name(),
VPNIP: init.IP,
VPNPrefix: init.Prefix,
VPNIP6: init.IP6,
VPNPrefix6: init.Prefix6,
ServerHost: serverHostFromURL(cfg.ServerURL),
CIDRs: cidrs,
}
sm.routeMgr = route.NewManager(routeCfg)
if err := sm.routeMgr.Apply(); err != nil {
log.L().Error("route apply failed (continuing)", "error", err)
}
// Initialise the CIDR hit tracker for proxy/bypass modes.
sm.cidrTracker = newCIDRTracker(cfg.RoutingMode, cidrs)
// Log CIDR breakdown by family for diagnostics.
v4Total, _, v6Total, _ := sm.cidrTracker.Stats()
log.L().Info("TUN configured",
"dev", dev.Name(), "ip", init.IP, "prefix", init.Prefix,
"ip6", init.IP6, "prefix6", init.Prefix6, "mtu", mtu,
"routing_mode", cfg.RoutingMode,
"cidr_v4", v4Total, "cidr_v6", v6Total,
"before_proxy_cidrs", len(beforeCIDRs))
return nil
}
// fetchAfterProxyCIDRs fetches CIDR lists from URLs with "after" timing
// (via the tunnel) and dynamically adds their routes to the route
// manager. This is called in a goroutine after the data plane is up.
func (sm *SessionManager) fetchAfterProxyCIDRs(ctx context.Context, cfg SessionConfig) {
allURLSources := append(append([]model.CIDRURLSource{}, cfg.CIDRV4URLs...), cfg.CIDRV6URLs...)
// Count only "after" sources for logging.
afterCount := 0
for _, s := range allURLSources {
if s.FetchTiming == model.FetchAfter {
afterCount++
}
}
if afterCount == 0 {
return
}
sm.stats.SetRouteLoading(true)
sm.stats.SetConnectStep("fetch_cidrs")
log.L().Info("fetching after-proxy CIDR lists", "url_count", afterCount)
fetched, err := cidrsource.FetchAfterProxy(ctx, allURLSources)
sm.stats.SetConnectStep("load_routes")
if err != nil {
log.L().Error("fetch after-proxy CIDR lists completed with errors", "error", err)
sm.stats.SetCIDRError(err.Error())
} else {
sm.stats.SetCIDRError("")
}
if len(fetched) == 0 {
sm.stats.SetRouteLoading(false)
sm.stats.SetConnectStep("")
return
}
merged := fetched // AddRoutes already calls mergeCIDRs internally
added := sm.addCIDRRoutes(fetched)
if added > 0 {
log.L().Info("added after-proxy routes", "fetched", len(fetched), "added", added, "merged", len(merged))
}
sm.stats.SetRouteLoading(false)
sm.stats.SetConnectStep("")
}
// addCIDRRoutes adds routes and updates the CIDR tracker. Returns the
// number of CIDRs successfully added (after merge).
func (sm *SessionManager) addCIDRRoutes(cidrs []string) int {
sm.mu.Lock()
mgr := sm.routeMgr
tracker := sm.cidrTracker
sm.mu.Unlock()
if mgr == nil {
return 0
}
if err := mgr.AddRoutes(cidrs); err != nil {
log.L().Error("add CIDR routes failed (continuing)", "error", err)
}
if tracker != nil {
tracker.AddCIDRs(cidrs)
}
return len(cidrs)
}
// RefreshCIDRs re-fetches all CIDR URL sources (both before and after
// timing) via the current tunnel and dynamically adds their routes.
// This is called when the user clicks the "Refresh CIDR" button.
func (sm *SessionManager) RefreshCIDRs() {
sm.mu.Lock()
ctx := sm.cancel
cfg := sm.lastCfg
sm.mu.Unlock()
if ctx == nil {
return
}
// Use a background context with 30s timeout so the refresh works
// even if the session context is in a weird state.
refreshCtx, cancel := context.WithTimeout(context.Background(), 30*time.Second)
defer cancel()
_ = refreshCtx
allURLSources := append(append([]model.CIDRURLSource{}, cfg.CIDRV4URLs...), cfg.CIDRV6URLs...)
if len(allURLSources) == 0 {
return
}
sm.stats.SetRouteLoading(true)
sm.stats.SetCIDRError("")
log.L().Info("refreshing CIDR lists", "url_count", len(allURLSources))
fetched, err := cidrsource.FetchAfterProxy(refreshCtx, allURLSources)
if err != nil {
log.L().Error("refresh CIDR lists completed with errors", "error", err)
sm.stats.SetCIDRError(err.Error())
} else {
sm.stats.SetCIDRError("")
}
if len(fetched) == 0 {
sm.stats.SetRouteLoading(false)
return
}
added := sm.addCIDRRoutes(fetched)
log.L().Info("refreshed CIDR routes", "fetched", len(fetched), "added", added)
sm.stats.SetRouteLoading(false)
}
// pumpPackets runs the bidirectional packet loop until the connection
// breaks.
func (sm *SessionManager) pumpPackets(ctx context.Context, conn *transport.Conn) {
var wg sync.WaitGroup
wg.Add(2)
// TUN → WebSocket
go func() {
defer wg.Done()
buf := make([]byte, 65536)
// Cache the cidrTracker pointer once; it is stable for the
// lifetime of pumpPackets (set in setupTUN, cleared in cleanup
// which only runs after pumpPackets returns).
sm.mu.Lock()
tracker := sm.cidrTracker
sm.mu.Unlock()
for {
select {
case <-ctx.Done():
return
default:
}
n, err := sm.readTUN(buf)
if err != nil {
if ctx.Err() == nil {
log.L().Error("tun read error", "error", err)
}
conn.Close()
return
}
if n == 0 {
continue
}
if err := conn.WritePacket(buf[:n]); err != nil {
if ctx.Err() == nil {
log.L().Error("ws write error", "error", err)
}
return
}
sm.stats.AddTx(buf[:n])
if tracker != nil {
tracker.Record(buf[:n])
}
}
}()
// WebSocket → TUN
go func() {
defer wg.Done()
for {
select {
case <-ctx.Done():
return
default:
}
data, err := conn.ReadPacket()
if err != nil {
if ctx.Err() == nil {
log.L().Error("ws read error", "error", err)
}
return
}
if _, err := sm.writeTUN(data); err != nil {
if ctx.Err() == nil {
log.L().Error("tun write error", "error", err)
}
conn.Close()
return
}
sm.stats.AddRx(data)
}
}()
wg.Wait()
}
// cleanupResources tears down the TUN device, routes, and transport
// connection WITHOUT changing the session state. This is used on
// handshake-failure paths where the caller will set the appropriate
// state (e.g. StateReconnecting). Returns without error if nothing
// was set up.
func (sm *SessionManager) cleanupResources() {
sm.mu.Lock()
dev := sm.dev
routeMgr := sm.routeMgr
conn := sm.conn
sm.dev = nil
sm.routeMgr = nil
sm.conn = nil
sm.cidrTracker = nil
sm.mu.Unlock()
if routeMgr != nil {
if err := routeMgr.Cleanup(); err != nil {
log.L().Error("route cleanup error", "error", err)
}
}
if dev != nil {
dev.Close()
}
if conn != nil {
conn.Close()
}
}
// cleanup tears down the TUN device, routes, and transport, then marks
// the session as disconnected. Used on normal session termination.
func (sm *SessionManager) cleanup() {
sm.cleanupResources()
sm.stats.SetDisconnected()
}
func (sm *SessionManager) readTUN(p []byte) (int, error) {
sm.mu.Lock()
dev := sm.dev
sm.mu.Unlock()
if dev == nil {
return 0, errors.New("tun device not available")
}
return dev.Read(p)
}
func (sm *SessionManager) writeTUN(p []byte) (int, error) {
sm.mu.Lock()
dev := sm.dev
sm.mu.Unlock()
if dev == nil {
return 0, errors.New("tun device not available")
}
return dev.Write(p)
}
func (sm *SessionManager) setState(s stats.State) {
sm.stats.SetState(s)
if sm.onState != nil {
sm.onState(s)
}
}
// reportStats periodically calls the onStats callback while connected.
// On each tick it derives per-family speeds (bytes/sec) from the
// delta between the current and previous cumulative counters, then
// applies EWMA smoothing (0.7 old + 0.3 new) so the displayed rates
// don't jitter. The combined speeds are the sum of the per-family
// smoothed values.
func (sm *SessionManager) reportStats(done <-chan struct{}, ctx context.Context) {
ticker := time.NewTicker(time.Second)
defer ticker.Stop()
const ewmaAlpha = 0.3
for {
select {
case <-done:
return
case <-ctx.Done():
return
case <-ticker.C:
if sm.onStats == nil {
continue
}
snap := sm.stats.Snapshot()
now := time.Now()
if sm.speedReady {
elapsed := now.Sub(sm.prevTick).Seconds()
if elapsed <= 0 {
elapsed = 1
}
// Per-second deltas (bytes/sec), clamped to >= 0 in case
// of counter resets between reconnects within the same
// SessionManager lifetime.
rxV4 := max(0.0, float64(snap.RxBytesV4-sm.prevSnap.RxBytesV4)/elapsed)
txV4 := max(0.0, float64(snap.TxBytesV4-sm.prevSnap.TxBytesV4)/elapsed)
rxV6 := max(0.0, float64(snap.RxBytesV6-sm.prevSnap.RxBytesV6)/elapsed)
txV6 := max(0.0, float64(snap.TxBytesV6-sm.prevSnap.TxBytesV6)/elapsed)
if sm.ewmaRxV4 == 0 && sm.ewmaTxV4 == 0 && sm.ewmaRxV6 == 0 && sm.ewmaTxV6 == 0 {
// First real sample: seed instead of ramping from 0.
sm.ewmaRxV4, sm.ewmaTxV4, sm.ewmaRxV6, sm.ewmaTxV6 = rxV4, txV4, rxV6, txV6
} else {
sm.ewmaRxV4 = sm.ewmaRxV4*(1-ewmaAlpha) + rxV4*ewmaAlpha
sm.ewmaTxV4 = sm.ewmaTxV4*(1-ewmaAlpha) + txV4*ewmaAlpha
sm.ewmaRxV6 = sm.ewmaRxV6*(1-ewmaAlpha) + rxV6*ewmaAlpha
sm.ewmaTxV6 = sm.ewmaTxV6*(1-ewmaAlpha) + txV6*ewmaAlpha
}
snap.RxSpeedV4 = int64(sm.ewmaRxV4)
snap.TxSpeedV4 = int64(sm.ewmaTxV4)
snap.RxSpeedV6 = int64(sm.ewmaRxV6)
snap.TxSpeedV6 = int64(sm.ewmaTxV6)
snap.RxSpeed = snap.RxSpeedV4 + snap.RxSpeedV6
snap.TxSpeed = snap.TxSpeedV4 + snap.TxSpeedV6
}
sm.prevSnap = snap
// Clear speed fields on the stored prev copy so we don't
// accidentally carry stale speed into the next delta base
// (only cumulative bytes matter for deltas).
sm.prevSnap.RxSpeedV4, sm.prevSnap.TxSpeedV4 = 0, 0
sm.prevSnap.RxSpeedV6, sm.prevSnap.TxSpeedV6 = 0, 0
sm.prevSnap.RxSpeed, sm.prevSnap.TxSpeed = 0, 0
sm.prevTick = now
sm.speedReady = true
// Fill in CIDR hit statistics.
sm.mu.Lock()
tracker := sm.cidrTracker
sm.mu.Unlock()
if tracker != nil {
v4Total, v4Hits, v6Total, v6Hits := tracker.Stats()
snap.RoutingMode = string(tracker.mode)
snap.CIDRV4Total = v4Total
snap.CIDRV4Hits = v4Hits
snap.CIDRV6Total = v6Total
snap.CIDRV6Hits = v6Hits
}
sm.onStats(snap)
}
}
}
// serverHostFromURL extracts the host portion from a WebSocket URL.
func serverHostFromURL(wsURL string) string {
u := wsURL
for _, p := range []string{"wss://", "ws://"} {
if len(u) > len(p) && u[:len(p)] == p {
u = u[len(p):]
break
}
}
// Strip port.
for i := 0; i < len(u); i++ {
if u[i] == ':' {
u = u[:i]
break
}
}
// Strip path.
for i := 0; i < len(u); i++ {
if u[i] == '/' {
u = u[:i]
break
}
}
return u
}
// wsURLToHTTP converts a WebSocket URL to HTTP origin.
func wsURLToHTTP(wsURL string) (string, error) {
return auth.WSURLToHTTP(wsURL)
}
// replaceHost substitutes the host portion of a URL string.
// e.g. wss://host:443/ws with 1.2.3.4 → wss://1.2.3.4:443/ws
// Bare IPv6 addresses are automatically bracketed:
// wss://host:443/ws with 2001:db8::1 → wss://[2001:db8::1]:443/ws
func replaceHost(rawURL, newHost string) string {
// Auto-bracket bare IPv6 addresses so the colons in the address
// are not confused with the port separator.
if ip := net.ParseIP(newHost); ip != nil && ip.To4() == nil && !strings.HasPrefix(newHost, "[") {
newHost = "[" + newHost + "]"
}
u := rawURL
for _, prefix := range []string{"wss://", "ws://"} {
if len(u) > len(prefix) && u[:len(prefix)] == prefix {
rest := u[len(prefix):]
// Find end of host (either port or path).
end := 0
for end < len(rest) && rest[end] != ':' && rest[end] != '/' {
end++
}
return prefix + newHost + rest[end:]
}
}
return rawURL
}
// cidrTracker tracks which configured CIDRs have been "hit" by
// outbound traffic (TUN -> WebSocket). The behaviour differs by mode:
//
// - Proxy mode: each outbound packet's destination IP is matched
// against the CIDR list; the first matching CIDR is marked as hit.
// Stats() returns the count of hit CIDRs vs total CIDRs.
// - Bypass mode: outbound traffic through TUN is, by definition,
// traffic that did NOT match any bypass CIDR (bypassed traffic
// goes via the physical NIC). We count distinct destination
// /24 (v4) or /48 (v6) prefixes seen on TUN as "unmatched
// destinations". Stats() returns the total bypass CIDR count
// and the unmatched destination count.
type cidrTracker struct {
mode route.Mode
mu sync.RWMutex
// Proxy mode: pre-parsed CIDR nets + per-CIDR hit flags.
// Protected by mu for concurrent AddCIDRs (write) vs Record/Stats
// (read). The atomic.Bool hit flags are individually atomic, but
// the slices themselves need the lock.
v4CIDRs []*net.IPNet
v6CIDRs []*net.IPNet
v4Hits []atomic.Bool
v6Hits []atomic.Bool
// Bypass mode: distinct destination prefix sets.
// v4 key = first 3 bytes of dest IP (a /24 prefix).
// v6 key = first 6 bytes of dest IP (a /48 prefix).
// sync.Map is already goroutine-safe; no lock needed.
v4Prefixes sync.Map
v6Prefixes sync.Map
v4Count atomic.Int64
v6Count atomic.Int64
}
// newCIDRTracker creates a tracker for the given routing mode and CIDR
// list. For full-tunnel mode, a tracker is still created but will
// report zero totals (no CIDRs configured).
func newCIDRTracker(mode route.Mode, cidrs []string) *cidrTracker {
t := &cidrTracker{mode: mode}
if mode == route.ModeFull {
return t
}
t.addCIDRs(cidrs)
return t
}
// AddCIDRs appends additional CIDRs to the tracker (used for
// after-proxy fetched CIDRs). Existing hit flags are preserved.
func (t *cidrTracker) AddCIDRs(cidrs []string) {
t.addCIDRs(cidrs)
}
func (t *cidrTracker) addCIDRs(cidrs []string) {
t.mu.Lock()
defer t.mu.Unlock()
for _, cidrStr := range cidrs {
cidrStr = strings.TrimSpace(cidrStr)
if cidrStr == "" {
continue
}
_, ipNet, err := net.ParseCIDR(cidrStr)
if err != nil {
continue
}
if ipNet.IP.To4() != nil {
t.v4CIDRs = append(t.v4CIDRs, ipNet)
t.v4Hits = append(t.v4Hits, atomic.Bool{})
} else {
t.v6CIDRs = append(t.v6CIDRs, ipNet)
t.v6Hits = append(t.v6Hits, atomic.Bool{})
}
}
}
// Record inspects an outbound IP packet (from TUN) and updates hit
// counters. It is called from the TUN -> WebSocket goroutine for every
// packet. Packets too short to contain an IP header or with an unknown
// version are silently ignored.
func (t *cidrTracker) Record(p []byte) {
if len(p) < 1 {
return
}
switch p[0] >> 4 {
case 4:
if len(p) < 20 {
return
}
dst := net.IP(p[16:20])
t.recordV4(dst)
case 6:
if len(p) < 40 {
return
}
dst := net.IP(p[24:40])
t.recordV6(dst)
}
}
func (t *cidrTracker) recordV4(dst net.IP) {
switch t.mode {
case route.ModeProxy:
t.mu.RLock()
defer t.mu.RUnlock()
for i, cidr := range t.v4CIDRs {
if !t.v4Hits[i].Load() && cidr.Contains(dst) {
t.v4Hits[i].Store(true)
}
}
case route.ModeBypass:
key := string(dst.To4()[:3])
if _, loaded := t.v4Prefixes.LoadOrStore(key, struct{}{}); !loaded {
t.v4Count.Add(1)
}
}
}
func (t *cidrTracker) recordV6(dst net.IP) {
switch t.mode {
case route.ModeProxy:
t.mu.RLock()
defer t.mu.RUnlock()
for i, cidr := range t.v6CIDRs {
if !t.v6Hits[i].Load() && cidr.Contains(dst) {
t.v6Hits[i].Store(true)
}
}
case route.ModeBypass:
v6 := dst.To16()
if len(v6) < 6 {
return
}
key := string(v6[:6])
if _, loaded := t.v6Prefixes.LoadOrStore(key, struct{}{}); !loaded {
t.v6Count.Add(1)
}
}
}
// Stats returns the total and hit counts for IPv4 and IPv6 CIDRs.
//
// For proxy mode: "hits" = number of CIDRs that have been matched by
// at least one outbound packet.
// For bypass mode: "hits" = number of distinct destination prefixes
// seen on TUN (i.e. unmatched destinations that went through the
// tunnel because they didn't match any bypass CIDR).
func (t *cidrTracker) Stats() (v4Total, v4Hits, v6Total, v6Hits int) {
t.mu.RLock()
defer t.mu.RUnlock()
switch t.mode {
case route.ModeProxy:
v4Total = len(t.v4CIDRs)
for i := range t.v4Hits {
if t.v4Hits[i].Load() {
v4Hits++
}
}
v6Total = len(t.v6CIDRs)
for i := range t.v6Hits {
if t.v6Hits[i].Load() {
v6Hits++
}
}
case route.ModeBypass:
v4Total = len(t.v4CIDRs)
v4Hits = int(t.v4Count.Load())
v6Total = len(t.v6CIDRs)
v6Hits = int(t.v6Count.Load())
}
return
}