fix(proxy): prevent deleting wrong tunnel in defer cleanup

Add pointer check before delete to handle race where UpdateLocalSNIs
removes our tunnel and a new one is created for the same hostname.

proxy/proxy.go

@@ -18,6 +18,7 @@ import (
 
 	"github.com/fosrl/gerbil/logger"
 	"github.com/patrickmn/go-cache"
+	"golang.org/x/sync/errgroup"
 )
 
 // RouteRecord represents a routing configuration
@@ -72,7 +73,9 @@ type SNIProxy struct {
 }
 
 type activeTunnel struct {
-	conns map[net.Conn]struct{}
+	ctx    context.Context
+	cancel context.CancelFunc
+	count  int // protected by activeTunnelsLock
 }
 
 // readOnlyConn is a wrapper for io.Reader that implements net.Conn
@@ -588,30 +591,32 @@ func (p *SNIProxy) handleConnection(clientConn net.Conn) {
 		}
 	}
 
-	// Track this tunnel by SNI
+	// Track this tunnel by SNI using context for cancellation
 	p.activeTunnelsLock.Lock()
 	tunnel, ok := p.activeTunnels[hostname]
 	if !ok {
-		tunnel = &activeTunnel{conns: make(map[net.Conn]struct{})}
+		ctx, cancel := context.WithCancel(p.ctx)
+		tunnel = &activeTunnel{ctx: ctx, cancel: cancel}
 		p.activeTunnels[hostname] = tunnel
 	}
-	tunnel.conns[actualClientConn] = struct{}{}
+	tunnel.count++
+	tunnelCtx := tunnel.ctx
 	p.activeTunnelsLock.Unlock()
 
 	defer func() {
-		// Remove this conn from active tunnels - O(1) with map
 		p.activeTunnelsLock.Lock()
-		if tunnel, ok := p.activeTunnels[hostname]; ok {
-			delete(tunnel.conns, actualClientConn)
-			if len(tunnel.conns) == 0 {
+		tunnel.count--
+		if tunnel.count == 0 {
+			tunnel.cancel()
+			if p.activeTunnels[hostname] == tunnel {
 				delete(p.activeTunnels, hostname)
 			}
 		}
 		p.activeTunnelsLock.Unlock()
 	}()
 
-	// Start bidirectional data transfer
-	p.pipe(actualClientConn, targetConn, clientReader)
+	// Start bidirectional data transfer with tunnel context
+	p.pipe(tunnelCtx, actualClientConn, targetConn, clientReader)
 }
 
 // getRoute retrieves routing information for a hostname
@@ -747,47 +752,36 @@ func (p *SNIProxy) selectStickyEndpoint(clientAddr string, endpoints []string) s
 }
 
 // pipe handles bidirectional data transfer between connections
-func (p *SNIProxy) pipe(clientConn, targetConn net.Conn, clientReader io.Reader) {
-	var wg sync.WaitGroup
-	wg.Add(2)
+func (p *SNIProxy) pipe(ctx context.Context, clientConn, targetConn net.Conn, clientReader io.Reader) {
+	g, gCtx := errgroup.WithContext(ctx)
 
-	// closeOnce ensures we only close connections once
-	var closeOnce sync.Once
-	closeConns := func() {
-		closeOnce.Do(func() {
-			// Close both connections to unblock any pending reads
-			clientConn.Close()
-			targetConn.Close()
-		})
-	}
+	// Close connections when context cancels to unblock io.Copy operations
+	context.AfterFunc(gCtx, func() {
+		clientConn.Close()
+		targetConn.Close()
+	})
 
-	// Copy data from client to target (using the buffered reader)
-	go func() {
-		defer wg.Done()
-		defer closeConns()
-
-		// Use a large buffer for better performance
+	// Copy data from client to target
+	g.Go(func() error {
 		buf := make([]byte, 32*1024)
 		_, err := io.CopyBuffer(targetConn, clientReader, buf)
 		if err != nil && err != io.EOF {
 			logger.Debug("Copy client->target error: %v", err)
 		}
-	}()
+		return err
+	})
 
 	// Copy data from target to client
-	go func() {
-		defer wg.Done()
-		defer closeConns()
-
-		// Use a large buffer for better performance
+	g.Go(func() error {
 		buf := make([]byte, 32*1024)
 		_, err := io.CopyBuffer(clientConn, targetConn, buf)
 		if err != nil && err != io.EOF {
 			logger.Debug("Copy target->client error: %v", err)
 		}
-	}()
+		return err
+	})
 
-	wg.Wait()
+	g.Wait()
 }
 
 // GetCacheStats returns cache statistics
@@ -803,46 +797,34 @@ func (p *SNIProxy) ClearCache() {
 
 // UpdateLocalSNIs updates the local SNIs and invalidates cache for changed domains
 func (p *SNIProxy) UpdateLocalSNIs(fullDomains []string) {
-	newSNIs := make(map[string]struct{}, len(fullDomains))
+	newSNIs := make(map[string]struct{})
 	for _, domain := range fullDomains {
 		newSNIs[domain] = struct{}{}
+		// Invalidate any cached route for this domain
+		p.cache.Delete(domain)
 	}
 
-	// Get old SNIs with read lock to compute diff outside write lock
-	p.localSNIsLock.RLock()
-	oldSNIs := p.localSNIs
-	p.localSNIsLock.RUnlock()
-
-	// Compute removed SNIs outside the lock
+	// Update localSNIs
+	p.localSNIsLock.Lock()
 	removed := make([]string, 0)
-	for sni := range oldSNIs {
+	for sni := range p.localSNIs {
 		if _, stillLocal := newSNIs[sni]; !stillLocal {
 			removed = append(removed, sni)
 		}
 	}
-
-	// Swap with minimal write lock hold time
-	p.localSNIsLock.Lock()
 	p.localSNIs = newSNIs
 	p.localSNIsLock.Unlock()
 
-	// Invalidate cache for new domains (cache is thread-safe)
-	for domain := range newSNIs {
-		p.cache.Delete(domain)
-	}
-
 	logger.Debug("Updated local SNIs, added %d, removed %d", len(newSNIs), len(removed))
 
-	// Terminate tunnels for removed SNIs
+	// Terminate tunnels for removed SNIs via context cancellation
 	if len(removed) > 0 {
 		p.activeTunnelsLock.Lock()
 		for _, sni := range removed {
 			if tunnel, ok := p.activeTunnels[sni]; ok {
-				for conn := range tunnel.conns {
-					conn.Close()
-				}
+				tunnel.cancel()
 				delete(p.activeTunnels, sni)
-				logger.Debug("Closed tunnels for SNI target change: %s", sni)
+				logger.Debug("Cancelled tunnel context for SNI target change: %s", sni)
 			}
 		}
 		p.activeTunnelsLock.Unlock()