fix(agent): retire replaced tunnel sessions (#541)

go-gost/x/chain/chain.go

@@ -2,6 +2,8 @@ package chain
 
 import (
 	"context"
+	"errors"
+	"io"
 
 	"github.com/go-gost/core/chain"
 	"github.com/go-gost/core/hop"
@@ -38,11 +40,12 @@ type chainNamer interface {
 }
 
 type Chain struct {
-	name     string
-	hops     []hop.Hop
-	marker   selector.Marker
-	metadata metadata.Metadata
-	logger   logger.Logger
+	name      string
+	hops      []hop.Hop
+	ownedHops []hop.Hop
+	marker    selector.Marker
+	metadata  metadata.Metadata
+	logger    logger.Logger
 }
 
 func NewChain(name string, opts ...ChainOption) *Chain {
@@ -61,8 +64,15 @@ func NewChain(name string, opts ...ChainOption) *Chain {
 	}
 }
 
-func (c *Chain) AddHop(hop hop.Hop) {
+func (c *Chain) AddHop(hop hop.Hop, owned ...bool) {
 	c.hops = append(c.hops, hop)
+	isOwned := true
+	if len(owned) > 0 {
+		isOwned = owned[0]
+	}
+	if isOwned {
+		c.ownedHops = append(c.ownedHops, hop)
+	}
 }
 
 // Metadata implements metadata.Metadatable interface.
@@ -112,6 +122,36 @@ func (c *Chain) Route(ctx context.Context, network, address string, opts ...chai
 	return rt
 }
 
+// Retire gracefully drains resources owned by a chain that has been replaced.
+func (c *Chain) Retire() {
+	if c == nil {
+		return
+	}
+	for _, h := range c.ownedHops {
+		if retirer, ok := h.(interface{ Retire() }); ok {
+			retirer.Retire()
+			continue
+		}
+		if closer, ok := h.(io.Closer); ok {
+			_ = closer.Close()
+		}
+	}
+}
+
+// Close immediately releases all resources owned by the chain.
+func (c *Chain) Close() error {
+	if c == nil {
+		return nil
+	}
+	var errs []error
+	for _, h := range c.ownedHops {
+		if closer, ok := h.(io.Closer); ok {
+			errs = append(errs, closer.Close())
+		}
+	}
+	return errors.Join(errs...)
+}
+
 type chainGroup struct {
 	chains   []chain.Chainer
 	selector selector.Selector[chain.Chainer]

go-gost/x/chain/chain_test.go

@@ -0,0 +1,64 @@
+package chain
+
+import (
+	"context"
+	"testing"
+
+	corechain "github.com/go-gost/core/chain"
+	corehop "github.com/go-gost/core/hop"
+)
+
+type lifecycleTestHop struct {
+	selected int
+	retired  int
+	closed   int
+}
+
+func (h *lifecycleTestHop) Select(context.Context, ...corehop.SelectOption) *corechain.Node {
+	h.selected++
+	return nil
+}
+
+func (h *lifecycleTestHop) Retire() {
+	h.retired++
+}
+
+func (h *lifecycleTestHop) Close() error {
+	h.closed++
+	return nil
+}
+
+func TestChainRoutesThroughSharedHopWithoutOwningLifecycle(t *testing.T) {
+	hop := &lifecycleTestHop{}
+	chain := NewChain("shared-hop")
+	chain.AddHop(hop, false)
+
+	if route := chain.Route(context.Background(), "tcp", "example.com:443"); route == nil {
+		t.Fatal("route is nil")
+	}
+	if hop.selected != 1 {
+		t.Fatalf("shared hop selected %d times, want 1", hop.selected)
+	}
+
+	chain.Retire()
+	if err := chain.Close(); err != nil {
+		t.Fatalf("close chain: %v", err)
+	}
+	if hop.retired != 0 || hop.closed != 0 {
+		t.Fatalf("shared hop lifecycle changed: retired=%d closed=%d", hop.retired, hop.closed)
+	}
+}
+
+func TestChainRetiresAndClosesOwnedHop(t *testing.T) {
+	hop := &lifecycleTestHop{}
+	chain := NewChain("owned-hop")
+	chain.AddHop(hop)
+
+	chain.Retire()
+	if err := chain.Close(); err != nil {
+		t.Fatalf("close chain: %v", err)
+	}
+	if hop.retired != 1 || hop.closed != 1 {
+		t.Fatalf("owned hop lifecycle: retired=%d closed=%d, want 1/1", hop.retired, hop.closed)
+	}
+}

go-gost/x/chain/transport.go

@@ -2,6 +2,8 @@ package chain
 
 import (
 	"context"
+	"errors"
+	"io"
 	"net"
 
 	"github.com/go-gost/core/chain"
@@ -102,5 +104,34 @@ func (tr *Transport) Options() *chain.TransportOptions {
 func (tr *Transport) Copy() chain.Transporter {
 	tr2 := &Transport{}
 	*tr2 = *tr
-	return tr
+	return tr2
+}
+
+// Retire prevents long-lived dialer sessions owned by an obsolete chain from
+// accepting new streams while allowing existing streams to drain.
+func (tr *Transport) Retire() {
+	if tr == nil {
+		return
+	}
+	if retirer, ok := tr.dialer.(interface{ Retire() }); ok {
+		retirer.Retire()
+	}
+	if retirer, ok := tr.connector.(interface{ Retire() }); ok {
+		retirer.Retire()
+	}
+}
+
+// Close immediately releases transport-owned dialer and connector resources.
+func (tr *Transport) Close() error {
+	if tr == nil {
+		return nil
+	}
+	var errs []error
+	if closer, ok := tr.dialer.(io.Closer); ok {
+		errs = append(errs, closer.Close())
+	}
+	if closer, ok := tr.connector.(io.Closer); ok {
+		errs = append(errs, closer.Close())
+	}
+	return errors.Join(errs...)
 }

go-gost/x/chain/transport_test.go

@@ -0,0 +1,45 @@
+package chain
+
+import (
+	"context"
+	"net"
+	"testing"
+
+	corechain "github.com/go-gost/core/chain"
+)
+
+type copyTestRoute struct{}
+
+func (copyTestRoute) Dial(context.Context, string, string, ...corechain.DialOption) (net.Conn, error) {
+	return nil, nil
+}
+
+func (copyTestRoute) Bind(context.Context, string, string, ...corechain.BindOption) (net.Listener, error) {
+	return nil, nil
+}
+
+func (copyTestRoute) Nodes() []*corechain.Node {
+	return nil
+}
+
+func TestTransportCopyReturnsIndependentTransport(t *testing.T) {
+	originalRoute := copyTestRoute{}
+	replacementRoute := &copyTestRoute{}
+	original := NewTransport(nil, nil, corechain.RouteTransportOption(originalRoute))
+
+	copied, ok := original.Copy().(*Transport)
+	if !ok {
+		t.Fatalf("copy type = %T, want *Transport", original.Copy())
+	}
+	if copied == original {
+		t.Fatal("Copy returned the original transport")
+	}
+
+	copied.Options().Route = replacementRoute
+	if original.Options().Route != originalRoute {
+		t.Fatal("mutating copied transport changed original route")
+	}
+	if copied.Options().Route != replacementRoute {
+		t.Fatal("copied transport did not retain its independent route")
+	}
+}