[cllient] Don't track ebpf traffic in conntrack

client/firewall/iptables/manager_linux.go

@@ -83,6 +83,10 @@ func (m *Manager) Init(stateManager *statemanager.Manager) error {
 		return fmt.Errorf("acl manager init: %w", err)
 	}
 
+	if err := m.initNoTrackChain(); err != nil {
+		return fmt.Errorf("init notrack chain: %w", err)
+	}
+
 	// persist early to ensure cleanup of chains
 	go func() {
 		if err := stateManager.PersistState(context.Background()); err != nil {
@@ -177,6 +181,10 @@ func (m *Manager) Close(stateManager *statemanager.Manager) error {
 
 	var merr *multierror.Error
 
+	if err := m.cleanupNoTrackChain(); err != nil {
+		merr = multierror.Append(merr, fmt.Errorf("cleanup notrack chain: %w", err))
+	}
+
 	if err := m.aclMgr.Reset(); err != nil {
 		merr = multierror.Append(merr, fmt.Errorf("reset acl manager: %w", err))
 	}
@@ -277,6 +285,125 @@ func (m *Manager) RemoveInboundDNAT(localAddr netip.Addr, protocol firewall.Prot
 	return m.router.RemoveInboundDNAT(localAddr, protocol, sourcePort, targetPort)
 }
 
+const (
+	chainNameRaw = "NETBIRD-RAW"
+	chainOUTPUT  = "OUTPUT"
+	tableRaw     = "raw"
+)
+
+// SetupEBPFProxyNoTrack creates notrack rules for eBPF proxy loopback traffic.
+// This prevents conntrack from tracking WireGuard proxy traffic on loopback, which
+// can interfere with MASQUERADE rules (e.g., from container runtimes like Podman/netavark).
+//
+// Traffic flows that need NOTRACK:
+//
+//  1. Egress: WireGuard -> fake endpoint (before eBPF rewrite)
+//     src=127.0.0.1:wgPort -> dst=127.0.0.1:fakePort
+//     Matched by: sport=wgPort
+//
+//  2. Egress: Proxy -> WireGuard (via raw socket)
+//     src=127.0.0.1:fakePort -> dst=127.0.0.1:wgPort
+//     Matched by: dport=wgPort
+//
+//  3. Ingress: Packets to WireGuard
+//     dst=127.0.0.1:wgPort
+//     Matched by: dport=wgPort
+//
+//  4. Ingress: Packets to proxy (after eBPF rewrite)
+//     dst=127.0.0.1:proxyPort
+//     Matched by: dport=proxyPort
+//
+// Rules are cleaned up when the firewall manager is closed.
+func (m *Manager) SetupEBPFProxyNoTrack(proxyPort, wgPort uint16) error {
+	m.mutex.Lock()
+	defer m.mutex.Unlock()
+
+	wgPortStr := fmt.Sprintf("%d", wgPort)
+	proxyPortStr := fmt.Sprintf("%d", proxyPort)
+
+	// Egress rules: match outgoing loopback UDP packets
+	outputRuleSport := []string{"-o", "lo", "-s", "127.0.0.1", "-d", "127.0.0.1", "-p", "udp", "--sport", wgPortStr, "-j", "NOTRACK"}
+	if err := m.ipv4Client.AppendUnique(tableRaw, chainNameRaw, outputRuleSport...); err != nil {
+		return fmt.Errorf("add output sport notrack rule: %w", err)
+	}
+
+	outputRuleDport := []string{"-o", "lo", "-s", "127.0.0.1", "-d", "127.0.0.1", "-p", "udp", "--dport", wgPortStr, "-j", "NOTRACK"}
+	if err := m.ipv4Client.AppendUnique(tableRaw, chainNameRaw, outputRuleDport...); err != nil {
+		return fmt.Errorf("add output dport notrack rule: %w", err)
+	}
+
+	// Ingress rules: match incoming loopback UDP packets
+	preroutingRuleWg := []string{"-i", "lo", "-s", "127.0.0.1", "-d", "127.0.0.1", "-p", "udp", "--dport", wgPortStr, "-j", "NOTRACK"}
+	if err := m.ipv4Client.AppendUnique(tableRaw, chainNameRaw, preroutingRuleWg...); err != nil {
+		return fmt.Errorf("add prerouting wg notrack rule: %w", err)
+	}
+
+	preroutingRuleProxy := []string{"-i", "lo", "-s", "127.0.0.1", "-d", "127.0.0.1", "-p", "udp", "--dport", proxyPortStr, "-j", "NOTRACK"}
+	if err := m.ipv4Client.AppendUnique(tableRaw, chainNameRaw, preroutingRuleProxy...); err != nil {
+		return fmt.Errorf("add prerouting proxy notrack rule: %w", err)
+	}
+
+	log.Debugf("set up ebpf proxy notrack rules for ports %d,%d", proxyPort, wgPort)
+	return nil
+}
+
+func (m *Manager) initNoTrackChain() error {
+	if err := m.cleanupNoTrackChain(); err != nil {
+		log.Debugf("cleanup notrack chain: %v", err)
+	}
+
+	if err := m.ipv4Client.NewChain(tableRaw, chainNameRaw); err != nil {
+		return fmt.Errorf("create chain: %w", err)
+	}
+
+	jumpRule := []string{"-j", chainNameRaw}
+
+	if err := m.ipv4Client.InsertUnique(tableRaw, chainOUTPUT, 1, jumpRule...); err != nil {
+		if delErr := m.ipv4Client.DeleteChain(tableRaw, chainNameRaw); delErr != nil {
+			log.Debugf("delete orphan chain: %v", delErr)
+		}
+		return fmt.Errorf("add output jump rule: %w", err)
+	}
+
+	if err := m.ipv4Client.InsertUnique(tableRaw, chainPREROUTING, 1, jumpRule...); err != nil {
+		if delErr := m.ipv4Client.DeleteIfExists(tableRaw, chainOUTPUT, jumpRule...); delErr != nil {
+			log.Debugf("delete output jump rule: %v", delErr)
+		}
+		if delErr := m.ipv4Client.DeleteChain(tableRaw, chainNameRaw); delErr != nil {
+			log.Debugf("delete orphan chain: %v", delErr)
+		}
+		return fmt.Errorf("add prerouting jump rule: %w", err)
+	}
+
+	return nil
+}
+
+func (m *Manager) cleanupNoTrackChain() error {
+	exists, err := m.ipv4Client.ChainExists(tableRaw, chainNameRaw)
+	if err != nil {
+		return fmt.Errorf("check chain exists: %w", err)
+	}
+	if !exists {
+		return nil
+	}
+
+	jumpRule := []string{"-j", chainNameRaw}
+
+	if err := m.ipv4Client.DeleteIfExists(tableRaw, chainOUTPUT, jumpRule...); err != nil {
+		return fmt.Errorf("remove output jump rule: %w", err)
+	}
+
+	if err := m.ipv4Client.DeleteIfExists(tableRaw, chainPREROUTING, jumpRule...); err != nil {
+		return fmt.Errorf("remove prerouting jump rule: %w", err)
+	}
+
+	if err := m.ipv4Client.ClearAndDeleteChain(tableRaw, chainNameRaw); err != nil {
+		return fmt.Errorf("clear and delete chain: %w", err)
+	}
+
+	return nil
+}
+
 func getConntrackEstablished() []string {
 	return []string{"-m", "conntrack", "--ctstate", "RELATED,ESTABLISHED", "-j", "ACCEPT"}
 }

client/firewall/manager/firewall.go

@@ -168,6 +168,10 @@ type Manager interface {
 
 	// RemoveInboundDNAT removes inbound DNAT rule
 	RemoveInboundDNAT(localAddr netip.Addr, protocol Protocol, sourcePort, targetPort uint16) error
+
+	// SetupEBPFProxyNoTrack creates static notrack rules for eBPF proxy loopback traffic.
+	// This prevents conntrack from interfering with WireGuard proxy communication.
+	SetupEBPFProxyNoTrack(proxyPort, wgPort uint16) error
 }
 
 func GenKey(format string, pair RouterPair) string {

client/firewall/nftables/manager_linux.go

@@ -12,6 +12,7 @@ import (
 	"github.com/google/nftables/binaryutil"
 	"github.com/google/nftables/expr"
 	log "github.com/sirupsen/logrus"
+	"golang.org/x/sys/unix"
 
 	firewall "github.com/netbirdio/netbird/client/firewall/manager"
 	"github.com/netbirdio/netbird/client/iface/wgaddr"
@@ -48,8 +49,10 @@ type Manager struct {
 	rConn   *nftables.Conn
 	wgIface iFaceMapper
 
-	router     *router
-	aclManager *AclManager
+	router                 *router
+	aclManager             *AclManager
+	notrackOutputChain     *nftables.Chain
+	notrackPreroutingChain *nftables.Chain
 }
 
 // Create nftables firewall manager
@@ -91,6 +94,10 @@ func (m *Manager) Init(stateManager *statemanager.Manager) error {
 		return fmt.Errorf("acl manager init: %w", err)
 	}
 
+	if err := m.initNoTrackChains(workTable); err != nil {
+		return fmt.Errorf("init notrack chains: %w", err)
+	}
+
 	stateManager.RegisterState(&ShutdownState{})
 
 	// We only need to record minimal interface state for potential recreation.
@@ -288,7 +295,15 @@ func (m *Manager) Flush() error {
 	m.mutex.Lock()
 	defer m.mutex.Unlock()
 
-	return m.aclManager.Flush()
+	if err := m.aclManager.Flush(); err != nil {
+		return err
+	}
+
+	if err := m.refreshNoTrackChains(); err != nil {
+		log.Errorf("failed to refresh notrack chains: %v", err)
+	}
+
+	return nil
 }
 
 // AddDNATRule adds a DNAT rule
@@ -331,6 +346,176 @@ func (m *Manager) RemoveInboundDNAT(localAddr netip.Addr, protocol firewall.Prot
 	return m.router.RemoveInboundDNAT(localAddr, protocol, sourcePort, targetPort)
 }
 
+const (
+	chainNameRawOutput     = "netbird-raw-out"
+	chainNameRawPrerouting = "netbird-raw-pre"
+)
+
+// SetupEBPFProxyNoTrack creates notrack rules for eBPF proxy loopback traffic.
+// This prevents conntrack from tracking WireGuard proxy traffic on loopback, which
+// can interfere with MASQUERADE rules (e.g., from container runtimes like Podman/netavark).
+//
+// Traffic flows that need NOTRACK:
+//
+//  1. Egress: WireGuard -> fake endpoint (before eBPF rewrite)
+//     src=127.0.0.1:wgPort -> dst=127.0.0.1:fakePort
+//     Matched by: sport=wgPort
+//
+//  2. Egress: Proxy -> WireGuard (via raw socket)
+//     src=127.0.0.1:fakePort -> dst=127.0.0.1:wgPort
+//     Matched by: dport=wgPort
+//
+//  3. Ingress: Packets to WireGuard
+//     dst=127.0.0.1:wgPort
+//     Matched by: dport=wgPort
+//
+//  4. Ingress: Packets to proxy (after eBPF rewrite)
+//     dst=127.0.0.1:proxyPort
+//     Matched by: dport=proxyPort
+//
+// Rules are cleaned up when the firewall manager is closed.
+func (m *Manager) SetupEBPFProxyNoTrack(proxyPort, wgPort uint16) error {
+	m.mutex.Lock()
+	defer m.mutex.Unlock()
+
+	if m.notrackOutputChain == nil || m.notrackPreroutingChain == nil {
+		return fmt.Errorf("notrack chains not initialized")
+	}
+
+	proxyPortBytes := binaryutil.BigEndian.PutUint16(proxyPort)
+	wgPortBytes := binaryutil.BigEndian.PutUint16(wgPort)
+	loopback := []byte{127, 0, 0, 1}
+
+	// Egress rules: match outgoing loopback UDP packets
+	m.rConn.AddRule(&nftables.Rule{
+		Table: m.notrackOutputChain.Table,
+		Chain: m.notrackOutputChain,
+		Exprs: []expr.Any{
+			&expr.Meta{Key: expr.MetaKeyOIFNAME, Register: 1},
+			&expr.Cmp{Op: expr.CmpOpEq, Register: 1, Data: ifname("lo")},
+			&expr.Payload{DestRegister: 1, Base: expr.PayloadBaseNetworkHeader, Offset: 12, Len: 4}, // saddr
+			&expr.Cmp{Op: expr.CmpOpEq, Register: 1, Data: loopback},
+			&expr.Payload{DestRegister: 1, Base: expr.PayloadBaseNetworkHeader, Offset: 16, Len: 4}, // daddr
+			&expr.Cmp{Op: expr.CmpOpEq, Register: 1, Data: loopback},
+			&expr.Meta{Key: expr.MetaKeyL4PROTO, Register: 1},
+			&expr.Cmp{Op: expr.CmpOpEq, Register: 1, Data: []byte{unix.IPPROTO_UDP}},
+			&expr.Payload{DestRegister: 1, Base: expr.PayloadBaseTransportHeader, Offset: 0, Len: 2},
+			&expr.Cmp{Op: expr.CmpOpEq, Register: 1, Data: wgPortBytes}, // sport=wgPort
+			&expr.Counter{},
+			&expr.Notrack{},
+		},
+	})
+	m.rConn.AddRule(&nftables.Rule{
+		Table: m.notrackOutputChain.Table,
+		Chain: m.notrackOutputChain,
+		Exprs: []expr.Any{
+			&expr.Meta{Key: expr.MetaKeyOIFNAME, Register: 1},
+			&expr.Cmp{Op: expr.CmpOpEq, Register: 1, Data: ifname("lo")},
+			&expr.Payload{DestRegister: 1, Base: expr.PayloadBaseNetworkHeader, Offset: 12, Len: 4}, // saddr
+			&expr.Cmp{Op: expr.CmpOpEq, Register: 1, Data: loopback},
+			&expr.Payload{DestRegister: 1, Base: expr.PayloadBaseNetworkHeader, Offset: 16, Len: 4}, // daddr
+			&expr.Cmp{Op: expr.CmpOpEq, Register: 1, Data: loopback},
+			&expr.Meta{Key: expr.MetaKeyL4PROTO, Register: 1},
+			&expr.Cmp{Op: expr.CmpOpEq, Register: 1, Data: []byte{unix.IPPROTO_UDP}},
+			&expr.Payload{DestRegister: 1, Base: expr.PayloadBaseTransportHeader, Offset: 2, Len: 2},
+			&expr.Cmp{Op: expr.CmpOpEq, Register: 1, Data: wgPortBytes}, // dport=wgPort
+			&expr.Counter{},
+			&expr.Notrack{},
+		},
+	})
+
+	// Ingress rules: match incoming loopback UDP packets
+	m.rConn.AddRule(&nftables.Rule{
+		Table: m.notrackPreroutingChain.Table,
+		Chain: m.notrackPreroutingChain,
+		Exprs: []expr.Any{
+			&expr.Meta{Key: expr.MetaKeyIIFNAME, Register: 1},
+			&expr.Cmp{Op: expr.CmpOpEq, Register: 1, Data: ifname("lo")},
+			&expr.Payload{DestRegister: 1, Base: expr.PayloadBaseNetworkHeader, Offset: 12, Len: 4}, // saddr
+			&expr.Cmp{Op: expr.CmpOpEq, Register: 1, Data: loopback},
+			&expr.Payload{DestRegister: 1, Base: expr.PayloadBaseNetworkHeader, Offset: 16, Len: 4}, // daddr
+			&expr.Cmp{Op: expr.CmpOpEq, Register: 1, Data: loopback},
+			&expr.Meta{Key: expr.MetaKeyL4PROTO, Register: 1},
+			&expr.Cmp{Op: expr.CmpOpEq, Register: 1, Data: []byte{unix.IPPROTO_UDP}},
+			&expr.Payload{DestRegister: 1, Base: expr.PayloadBaseTransportHeader, Offset: 2, Len: 2},
+			&expr.Cmp{Op: expr.CmpOpEq, Register: 1, Data: wgPortBytes}, // dport=wgPort
+			&expr.Counter{},
+			&expr.Notrack{},
+		},
+	})
+	m.rConn.AddRule(&nftables.Rule{
+		Table: m.notrackPreroutingChain.Table,
+		Chain: m.notrackPreroutingChain,
+		Exprs: []expr.Any{
+			&expr.Meta{Key: expr.MetaKeyIIFNAME, Register: 1},
+			&expr.Cmp{Op: expr.CmpOpEq, Register: 1, Data: ifname("lo")},
+			&expr.Payload{DestRegister: 1, Base: expr.PayloadBaseNetworkHeader, Offset: 12, Len: 4}, // saddr
+			&expr.Cmp{Op: expr.CmpOpEq, Register: 1, Data: loopback},
+			&expr.Payload{DestRegister: 1, Base: expr.PayloadBaseNetworkHeader, Offset: 16, Len: 4}, // daddr
+			&expr.Cmp{Op: expr.CmpOpEq, Register: 1, Data: loopback},
+			&expr.Meta{Key: expr.MetaKeyL4PROTO, Register: 1},
+			&expr.Cmp{Op: expr.CmpOpEq, Register: 1, Data: []byte{unix.IPPROTO_UDP}},
+			&expr.Payload{DestRegister: 1, Base: expr.PayloadBaseTransportHeader, Offset: 2, Len: 2},
+			&expr.Cmp{Op: expr.CmpOpEq, Register: 1, Data: proxyPortBytes}, // dport=proxyPort
+			&expr.Counter{},
+			&expr.Notrack{},
+		},
+	})
+
+	if err := m.rConn.Flush(); err != nil {
+		return fmt.Errorf("flush notrack rules: %w", err)
+	}
+
+	log.Debugf("set up ebpf proxy notrack rules for ports %d,%d", proxyPort, wgPort)
+	return nil
+}
+
+func (m *Manager) initNoTrackChains(table *nftables.Table) error {
+	m.notrackOutputChain = m.rConn.AddChain(&nftables.Chain{
+		Name:     chainNameRawOutput,
+		Table:    table,
+		Type:     nftables.ChainTypeFilter,
+		Hooknum:  nftables.ChainHookOutput,
+		Priority: nftables.ChainPriorityRaw,
+	})
+
+	m.notrackPreroutingChain = m.rConn.AddChain(&nftables.Chain{
+		Name:     chainNameRawPrerouting,
+		Table:    table,
+		Type:     nftables.ChainTypeFilter,
+		Hooknum:  nftables.ChainHookPrerouting,
+		Priority: nftables.ChainPriorityRaw,
+	})
+
+	if err := m.rConn.Flush(); err != nil {
+		return fmt.Errorf("flush chain creation: %w", err)
+	}
+
+	return nil
+}
+
+func (m *Manager) refreshNoTrackChains() error {
+	chains, err := m.rConn.ListChainsOfTableFamily(nftables.TableFamilyIPv4)
+	if err != nil {
+		return fmt.Errorf("list chains: %w", err)
+	}
+
+	tableName := getTableName()
+	for _, c := range chains {
+		if c.Table.Name != tableName {
+			continue
+		}
+		switch c.Name {
+		case chainNameRawOutput:
+			m.notrackOutputChain = c
+		case chainNameRawPrerouting:
+			m.notrackPreroutingChain = c
+		}
+	}
+
+	return nil
+}
+
 func (m *Manager) createWorkTable() (*nftables.Table, error) {
 	tables, err := m.rConn.ListTablesOfFamily(nftables.TableFamilyIPv4)
 	if err != nil {

client/firewall/uspfilter/filter.go

@@ -570,6 +570,14 @@ func (m *Manager) SetLegacyManagement(isLegacy bool) error {
 // Flush doesn't need to be implemented for this manager
 func (m *Manager) Flush() error { return nil }
 
+// SetupEBPFProxyNoTrack creates notrack rules for eBPF proxy loopback traffic.
+func (m *Manager) SetupEBPFProxyNoTrack(proxyPort, wgPort uint16) error {
+	if m.nativeFirewall == nil {
+		return nil
+	}
+	return m.nativeFirewall.SetupEBPFProxyNoTrack(proxyPort, wgPort)
+}
+
 // UpdateSet updates the rule destinations associated with the given set
 // by merging the existing prefixes with the new ones, then deduplicating.
 func (m *Manager) UpdateSet(set firewall.Set, prefixes []netip.Prefix) error {