document new config options

* connection-track ICMP and ICMPv6 traffic

* icmpv6 only has identifier on echo

.github/workflows/smoke-extra.yml

@@ -30,6 +30,9 @@ jobs:
     - name: add hashicorp source
       run: wget -O- https://apt.releases.hashicorp.com/gpg | gpg --dearmor | sudo tee /usr/share/keyrings/hashicorp-archive-keyring.gpg && echo "deb [signed-by=/usr/share/keyrings/hashicorp-archive-keyring.gpg] https://apt.releases.hashicorp.com $(lsb_release -cs) main" | sudo tee /etc/apt/sources.list.d/hashicorp.list
 
+    - name: workaround AMD-V issue  # https://github.com/cri-o/packaging/pull/306
+      run: sudo rmmod kvm_amd
+
     - name: install vagrant
       run: sudo apt-get update && sudo apt-get install -y vagrant virtualbox
 

.github/workflows/smoke/smoke.sh

@@ -37,17 +37,18 @@ docker run --name host4 --device /dev/net/tun:/dev/net/tun --cap-add NET_ADMIN -
 sleep 1
 
 # grab tcpdump pcaps for debugging
-docker exec lighthouse1 tcpdump -i nebula1 -q -w - -U 2>logs/lighthouse1.inside.log >logs/lighthouse1.inside.pcap &
+docker exec lighthouse1 tcpdump -i tun0 -q -w - -U 2>logs/lighthouse1.inside.log >logs/lighthouse1.inside.pcap &
 docker exec lighthouse1 tcpdump -i eth0 -q -w - -U 2>logs/lighthouse1.outside.log >logs/lighthouse1.outside.pcap &
-docker exec host2 tcpdump -i nebula1 -q -w - -U 2>logs/host2.inside.log >logs/host2.inside.pcap &
+docker exec host2 tcpdump -i tun0 -q -w - -U 2>logs/host2.inside.log >logs/host2.inside.pcap &
 docker exec host2 tcpdump -i eth0 -q -w - -U 2>logs/host2.outside.log >logs/host2.outside.pcap &
-docker exec host3 tcpdump -i nebula1 -q -w - -U 2>logs/host3.inside.log >logs/host3.inside.pcap &
+docker exec host3 tcpdump -i tun0 -q -w - -U 2>logs/host3.inside.log >logs/host3.inside.pcap &
 docker exec host3 tcpdump -i eth0 -q -w - -U 2>logs/host3.outside.log >logs/host3.outside.pcap &
-docker exec host4 tcpdump -i nebula1 -q -w - -U 2>logs/host4.inside.log >logs/host4.inside.pcap &
+docker exec host4 tcpdump -i tun0 -q -w - -U 2>logs/host4.inside.log >logs/host4.inside.pcap &
 docker exec host4 tcpdump -i eth0 -q -w - -U 2>logs/host4.outside.log >logs/host4.outside.pcap &
 
 docker exec host2 ncat -nklv 0.0.0.0 2000 &
 docker exec host3 ncat -nklv 0.0.0.0 2000 &
+docker exec host4 ncat -nkluv 0.0.0.0 4000 &
 docker exec host2 ncat -e '/usr/bin/echo host2' -nkluv 0.0.0.0 3000 &
 docker exec host3 ncat -e '/usr/bin/echo host3' -nkluv 0.0.0.0 3000 &
 
@@ -119,11 +120,11 @@ echo
 echo " *** Testing conntrack"
 echo
 set -x
-# host2 can ping host3 now that host3 pinged it first
+
-docker exec host2 ping -c1 192.168.100.3
+# host2 speaking to host4 on UDP 4000 should allow it to reply, when firewall rules would normally not permit this
-# host4 can ping host2 once conntrack established
+docker exec host2 sh -c "/usr/bin/echo host2 | ncat -nuv 192.168.100.4 4000"
-docker exec host2 ping -c1 192.168.100.4
+docker exec host2 ncat -e '/usr/bin/echo helloagainfromhost2' -nkluv 0.0.0.0 4000 &
-docker exec host4 ping -c1 192.168.100.2
+docker exec host4 sh -c "/usr/bin/echo host4 | ncat -nuv 192.168.100.2 4000"
 
 docker exec host4 sh -c 'kill 1'
 docker exec host3 sh -c 'kill 1'

CHANGELOG.md

@@ -7,6 +7,19 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 ## [Unreleased]
 
+## [1.10.3] - 2026-02-06
+
+### Security
+
+- Fix an issue where blocklist bypass is possible when using curve P256 since the signature can have 2 valid representations.
+  Both fingerprint representations will be tested against the blocklist.
+  Any newly issued P256 based certificates will have their signature clamped to the low-s form.
+  Nebula will assert the low-s signature form when validating certificates in a future version. [GHSA-69x3-g4r3-p962](https://github.com/slackhq/nebula/security/advisories/GHSA-69x3-g4r3-p962)
+
+### Changed
+
+- Improve error reporting if nebula fails to start due to a tun device naming issue. (#1588)
+
 ## [1.10.2] - 2026-01-21
 
 ### Fixed
@@ -775,7 +788,8 @@ created.)
 
 - Initial public release.
 
-[Unreleased]: https://github.com/slackhq/nebula/compare/v1.10.2...HEAD
+[Unreleased]: https://github.com/slackhq/nebula/compare/v1.10.3...HEAD
+[1.10.3]: https://github.com/slackhq/nebula/releases/tag/v1.10.3
 [1.10.2]: https://github.com/slackhq/nebula/releases/tag/v1.10.2
 [1.10.1]: https://github.com/slackhq/nebula/releases/tag/v1.10.1
 [1.10.0]: https://github.com/slackhq/nebula/releases/tag/v1.10.0

boring.go

@@ -1,5 +1,4 @@
 //go:build boringcrypto
-// +build boringcrypto
 
 package nebula
 

cert/ca_pool.go

@@ -141,10 +141,23 @@ func (ncp *CAPool) VerifyCertificate(now time.Time, c Certificate) (*CachedCerti
 		return nil, err
 	}
 
+	// Pre nebula v1.10.3 could generate signatures in either high or low s form and validation
+	// of signatures allowed for either. Nebula v1.10.3 and beyond clamps signature generation to low-s form
+	// but validation still allows for either. Since a change in the signature bytes affects the fingerprint, we
+	// need to test both forms until such a time comes that we enforce low-s form on signature validation.
+	fp2, err := CalculateAlternateFingerprint(c)
+	if err != nil {
+		return nil, fmt.Errorf("could not calculate alternate fingerprint to verify: %w", err)
+	}
+	if fp2 != "" && ncp.IsBlocklisted(fp2) {
+		return nil, ErrBlockListed
+	}
+
 	cc := CachedCertificate{
 		Certificate:       c,
 		InvertedGroups:    make(map[string]struct{}),
 		Fingerprint:       fp,
+		fingerprint2:      fp2,
 		signerFingerprint: signer.Fingerprint,
 	}
 
@@ -158,6 +171,11 @@ func (ncp *CAPool) VerifyCertificate(now time.Time, c Certificate) (*CachedCerti
 // VerifyCachedCertificate is the same as VerifyCertificate other than it operates on a pre-verified structure and
 // is a cheaper operation to perform as a result.
 func (ncp *CAPool) VerifyCachedCertificate(now time.Time, c *CachedCertificate) error {
+	// Check any available alternate fingerprint forms for this certificate, re P256 high-s/low-s
+	if c.fingerprint2 != "" && ncp.IsBlocklisted(c.fingerprint2) {
+		return ErrBlockListed
+	}
+
 	_, err := ncp.verify(c.Certificate, now, c.Fingerprint, c.signerFingerprint)
 	return err
 }

cert/ca_pool_test.go

@@ -5,6 +5,7 @@ import (
 	"testing"
 	"time"
 
+	"github.com/slackhq/nebula/cert/p256"
 	"github.com/stretchr/testify/assert"
 	"github.com/stretchr/testify/require"
 )
@@ -170,6 +171,15 @@ func TestCertificateV1_VerifyP256(t *testing.T) {
 	_, err = caPool.VerifyCertificate(time.Now(), c)
 	require.EqualError(t, err, "certificate is in the block list")
 
+	// Create a copy of the cert and swap to the alternate form for the signature
+	nc := c.Copy()
+	b, err := p256.Swap(c.Signature())
+	require.NoError(t, err)
+	require.NoError(t, nc.(*certificateV1).setSignature(b))
+
+	_, err = caPool.VerifyCertificate(time.Now(), nc)
+	require.EqualError(t, err, "certificate is in the block list")
+
 	caPool.ResetCertBlocklist()
 	_, err = caPool.VerifyCertificate(time.Now(), c)
 	require.NoError(t, err)
@@ -187,7 +197,7 @@ func TestCertificateV1_VerifyP256(t *testing.T) {
 	require.NoError(t, err)
 
 	caPool = NewCAPool()
-	b, err := caPool.AddCAFromPEM(caPem)
+	b, err = caPool.AddCAFromPEM(caPem)
 	require.NoError(t, err)
 	assert.Empty(t, b)
 
@@ -196,7 +206,17 @@ func TestCertificateV1_VerifyP256(t *testing.T) {
 	})
 
 	c, _, _, _ = NewTestCert(Version1, Curve_P256, ca, caKey, "test", time.Now(), time.Now().Add(5*time.Minute), nil, nil, []string{"test1"})
-	_, err = caPool.VerifyCertificate(time.Now(), c)
+	cc, err := caPool.VerifyCertificate(time.Now(), c)
+	require.NoError(t, err)
+
+	// Reset the blocklist and block the alternate form fingerprint
+	caPool.ResetCertBlocklist()
+	caPool.BlocklistFingerprint(cc.fingerprint2)
+	err = caPool.VerifyCachedCertificate(time.Now(), cc)
+	require.EqualError(t, err, "certificate is in the block list")
+
+	caPool.ResetCertBlocklist()
+	err = caPool.VerifyCachedCertificate(time.Now(), cc)
 	require.NoError(t, err)
 }
 
@@ -394,6 +414,15 @@ func TestCertificateV2_VerifyP256(t *testing.T) {
 	_, err = caPool.VerifyCertificate(time.Now(), c)
 	require.EqualError(t, err, "certificate is in the block list")
 
+	// Create a copy of the cert and swap to the alternate form for the signature
+	nc := c.Copy()
+	b, err := p256.Swap(c.Signature())
+	require.NoError(t, err)
+	require.NoError(t, nc.(*certificateV2).setSignature(b))
+
+	_, err = caPool.VerifyCertificate(time.Now(), nc)
+	require.EqualError(t, err, "certificate is in the block list")
+
 	caPool.ResetCertBlocklist()
 	_, err = caPool.VerifyCertificate(time.Now(), c)
 	require.NoError(t, err)
@@ -411,7 +440,7 @@ func TestCertificateV2_VerifyP256(t *testing.T) {
 	require.NoError(t, err)
 
 	caPool = NewCAPool()
-	b, err := caPool.AddCAFromPEM(caPem)
+	b, err = caPool.AddCAFromPEM(caPem)
 	require.NoError(t, err)
 	assert.Empty(t, b)
 
@@ -420,7 +449,17 @@ func TestCertificateV2_VerifyP256(t *testing.T) {
 	})
 
 	c, _, _, _ = NewTestCert(Version2, Curve_P256, ca, caKey, "test", time.Now(), time.Now().Add(5*time.Minute), nil, nil, []string{"test1"})
-	_, err = caPool.VerifyCertificate(time.Now(), c)
+	cc, err := caPool.VerifyCertificate(time.Now(), c)
+	require.NoError(t, err)
+
+	// Reset the blocklist and block the alternate form fingerprint
+	caPool.ResetCertBlocklist()
+	caPool.BlocklistFingerprint(cc.fingerprint2)
+	err = caPool.VerifyCachedCertificate(time.Now(), cc)
+	require.EqualError(t, err, "certificate is in the block list")
+
+	caPool.ResetCertBlocklist()
+	err = caPool.VerifyCachedCertificate(time.Now(), cc)
 	require.NoError(t, err)
 }
 

cert/cert.go

@@ -4,6 +4,8 @@ import (
 	"fmt"
 	"net/netip"
 	"time"
+
+	"github.com/slackhq/nebula/cert/p256"
 )
 
 type Version uint8
@@ -110,6 +112,9 @@ type CachedCertificate struct {
 	InvertedGroups    map[string]struct{}
 	Fingerprint       string
 	signerFingerprint string
+
+	// A place to store a 2nd fingerprint if the certificate could have one, such as with P256
+	fingerprint2 string
 }
 
 func (cc *CachedCertificate) String() string {
@@ -152,3 +157,31 @@ func Recombine(v Version, rawCertBytes, publicKey []byte, curve Curve) (Certific
 
 	return c, nil
 }
+
+// CalculateAlternateFingerprint calculates a 2nd fingerprint representation for P256 certificates
+// CAPool blocklist testing through `VerifyCertificate` and `VerifyCachedCertificate` automatically performs this step.
+func CalculateAlternateFingerprint(c Certificate) (string, error) {
+	if c.Curve() != Curve_P256 {
+		return "", nil
+	}
+
+	nc := c.Copy()
+	b, err := p256.Swap(nc.Signature())
+	if err != nil {
+		return "", err
+	}
+
+	switch v := nc.(type) {
+	case *certificateV1:
+		err = v.setSignature(b)
+	case *certificateV2:
+		err = v.setSignature(b)
+	default:
+		return "", ErrUnknownVersion
+	}
+
+	if err != nil {
+		return "", err
+	}
+	return nc.Fingerprint()
+}

cert/cert_v2.go

@@ -439,11 +439,7 @@ func (c *certificateV2) validate() error {
 				if !hasV6Networks {
 					return NewErrInvalidCertificateProperties("IPv6 unsafe networks require an IPv6 address assignment: %s", network)
 				}
-			} else if network.Addr().Is4() {
+			} // as long as we have any IP address, IPv4 UnsafeNetworks are allowed
-				if !hasV4Networks {
-					return NewErrInvalidCertificateProperties("IPv4 unsafe networks require an IPv4 address assignment: %s", network)
-				}
-			}
 		}
 	}
 

cert/p256/p256.go

@@ -0,0 +1,122 @@
+package p256
+
+import (
+	"crypto/elliptic"
+	"errors"
+	"math/big"
+
+	"filippo.io/bigmod"
+
+	"golang.org/x/crypto/cryptobyte"
+	"golang.org/x/crypto/cryptobyte/asn1"
+)
+
+var halfN = new(big.Int).Rsh(elliptic.P256().Params().N, 1)
+var nMod *bigmod.Modulus
+
+func init() {
+	n, err := bigmod.NewModulus(elliptic.P256().Params().N.Bytes())
+	if err != nil {
+		panic(err)
+	}
+	nMod = n
+}
+
+func IsNormalized(sig []byte) (bool, error) {
+	r, s, err := parseSignature(sig)
+	if err != nil {
+		return false, err
+	}
+	return checkLowS(r, s), nil
+}
+
+func checkLowS(_, s []byte) bool {
+	bigS := new(big.Int).SetBytes(s)
+	// Check if S <= (N/2), because we want to include the midpoint in the set of low-s
+	return bigS.Cmp(halfN) <= 0
+}
+
+func swap(r, s []byte) ([]byte, []byte, error) {
+	var err error
+	bigS, err := bigmod.NewNat().SetBytes(s, nMod)
+	if err != nil {
+		return nil, nil, err
+	}
+	sNormalized := nMod.Nat().Sub(bigS, nMod)
+
+	return r, sNormalized.Bytes(nMod), nil
+}
+
+func Normalize(sig []byte) ([]byte, error) {
+	r, s, err := parseSignature(sig)
+	if err != nil {
+		return nil, err
+	}
+
+	if checkLowS(r, s) {
+		return sig, nil
+	}
+
+	newR, newS, err := swap(r, s)
+	if err != nil {
+		return nil, err
+	}
+
+	return encodeSignature(newR, newS)
+}
+
+// Swap will change sig between its current form to the opposite high or low form.
+func Swap(sig []byte) ([]byte, error) {
+	r, s, err := parseSignature(sig)
+	if err != nil {
+		return nil, err
+	}
+
+	newR, newS, err := swap(r, s)
+	if err != nil {
+		return nil, err
+	}
+
+	return encodeSignature(newR, newS)
+}
+
+// parseSignature taken exactly from crypto/ecdsa/ecdsa.go
+func parseSignature(sig []byte) (r, s []byte, err error) {
+	var inner cryptobyte.String
+	input := cryptobyte.String(sig)
+	if !input.ReadASN1(&inner, asn1.SEQUENCE) ||
+		!input.Empty() ||
+		!inner.ReadASN1Integer(&r) ||
+		!inner.ReadASN1Integer(&s) ||
+		!inner.Empty() {
+		return nil, nil, errors.New("invalid ASN.1")
+	}
+	return r, s, nil
+}
+
+func encodeSignature(r, s []byte) ([]byte, error) {
+	var b cryptobyte.Builder
+	b.AddASN1(asn1.SEQUENCE, func(b *cryptobyte.Builder) {
+		addASN1IntBytes(b, r)
+		addASN1IntBytes(b, s)
+	})
+	return b.Bytes()
+}
+
+// addASN1IntBytes encodes in ASN.1 a positive integer represented as
+// a big-endian byte slice with zero or more leading zeroes.
+func addASN1IntBytes(b *cryptobyte.Builder, bytes []byte) {
+	for len(bytes) > 0 && bytes[0] == 0 {
+		bytes = bytes[1:]
+	}
+	if len(bytes) == 0 {
+		b.SetError(errors.New("invalid integer"))
+		return
+	}
+	b.AddASN1(asn1.INTEGER, func(c *cryptobyte.Builder) {
+		if bytes[0]&0x80 != 0 {
+			c.AddUint8(0)
+		}
+		c.AddBytes(bytes)
+	})
+}

cert/p256/p256_test.go

@@ -0,0 +1,28 @@
+package p256
+
+import (
+	"crypto/ecdsa"
+	"crypto/elliptic"
+	"crypto/rand"
+	"testing"
+
+	"github.com/stretchr/testify/require"
+)
+
+func TestFlipping(t *testing.T) {
+	priv, err1 := ecdsa.GenerateKey(elliptic.P256(), rand.Reader)
+	require.NoError(t, err1)
+
+	out, err := ecdsa.SignASN1(rand.Reader, priv, []byte("big chungus"))
+	require.NoError(t, err)
+
+	r, s, err := parseSignature(out)
+	require.NoError(t, err)
+
+	r, s1, err := swap(r, s)
+	require.NoError(t, err)
+	r, s2, err := swap(r, s1)
+	require.NoError(t, err)
+	require.Equal(t, s, s2)
+	require.NotEqual(t, s, s1)
+}

cert/sign.go

@@ -9,6 +9,8 @@ import (
 	"fmt"
 	"net/netip"
 	"time"
+
+	"github.com/slackhq/nebula/cert/p256"
 )
 
 // TBSCertificate represents a certificate intended to be signed.
@@ -126,6 +128,13 @@ func (t *TBSCertificate) SignWith(signer Certificate, curve Curve, sp SignerLamb
 		return nil, err
 	}
 
+	if curve == Curve_P256 {
+		sig, err = p256.Normalize(sig)
+		if err != nil {
+			return nil, err
+		}
+	}
+
 	err = c.setSignature(sig)
 	if err != nil {
 		return nil, err

cert/sign_test.go

@@ -9,6 +9,7 @@ import (
 	"testing"
 	"time"
 
+	"github.com/slackhq/nebula/cert/p256"
 	"github.com/stretchr/testify/assert"
 	"github.com/stretchr/testify/require"
 )
@@ -89,3 +90,48 @@ func TestCertificateV1_SignP256(t *testing.T) {
 	require.NoError(t, err)
 	assert.NotNil(t, uc)
 }
+
+func TestCertificate_SignP256_AlwaysNormalized(t *testing.T) {
+	before := time.Now().Add(time.Second * -60).Round(time.Second)
+	after := time.Now().Add(time.Second * 60).Round(time.Second)
+	pubKey := []byte("01234567890abcedfghij1234567890ab1234567890abcedfghij1234567890ab")
+
+	tbs := TBSCertificate{
+		Version: Version1,
+		Name:    "testing",
+		Networks: []netip.Prefix{
+			mustParsePrefixUnmapped("10.1.1.1/24"),
+			mustParsePrefixUnmapped("10.1.1.2/16"),
+		},
+		UnsafeNetworks: []netip.Prefix{
+			mustParsePrefixUnmapped("9.1.1.2/24"),
+			mustParsePrefixUnmapped("9.1.1.3/16"),
+		},
+		Groups:    []string{"test-group1", "test-group2", "test-group3"},
+		NotBefore: before,
+		NotAfter:  after,
+		PublicKey: pubKey,
+		IsCA:      true,
+		Curve:     Curve_P256,
+	}
+
+	priv, err := ecdsa.GenerateKey(elliptic.P256(), rand.Reader)
+	require.NoError(t, err)
+	pub := elliptic.Marshal(elliptic.P256(), priv.PublicKey.X, priv.PublicKey.Y)
+	rawPriv := priv.D.FillBytes(make([]byte, 32))
+
+	for i := 0; i < 1000; i++ {
+		if i&1 == 1 {
+			tbs.Version = Version1
+		} else {
+			tbs.Version = Version2
+		}
+		c, err := tbs.Sign(nil, Curve_P256, rawPriv)
+		require.NoError(t, err)
+		assert.NotNil(t, c)
+		assert.True(t, c.CheckSignature(pub))
+		normie, err := p256.IsNormalized(c.Signature())
+		require.NoError(t, err)
+		assert.True(t, normie)
+	}
+}

cmd/nebula-cert/test_freebsd.go

@@ -1,4 +0,0 @@
-package main
-
-const NoSuchFileError = "no such file or directory"
-const NoSuchDirError = "no such file or directory"

cmd/nebula-cert/test_openbsd.go

@@ -1,4 +0,0 @@
-package main
-
-const NoSuchFileError = "no such file or directory"
-const NoSuchDirError = "no such file or directory"

connection_manager_test.go

@@ -11,7 +11,6 @@ import (
 	"github.com/slackhq/nebula/cert"
 	"github.com/slackhq/nebula/config"
 	"github.com/slackhq/nebula/test"
-	"github.com/slackhq/nebula/test/device"
 	"github.com/slackhq/nebula/udp"
 	"github.com/stretchr/testify/assert"
 	"github.com/stretchr/testify/require"
@@ -53,7 +52,7 @@ func Test_NewConnectionManagerTest(t *testing.T) {
 	lh := newTestLighthouse()
 	ifce := &Interface{
 		hostMap:          hostMap,
-		inside:           &device.NoopTun{},
+		inside:           &test.NoopTun{},
 		outside:          &udp.NoopConn{},
 		firewall:         &Firewall{},
 		lightHouse:       lh,
@@ -136,7 +135,7 @@ func Test_NewConnectionManagerTest2(t *testing.T) {
 	lh := newTestLighthouse()
 	ifce := &Interface{
 		hostMap:          hostMap,
-		inside:           &device.NoopTun{},
+		inside:           &test.NoopTun{},
 		outside:          &udp.NoopConn{},
 		firewall:         &Firewall{},
 		lightHouse:       lh,
@@ -221,7 +220,7 @@ func Test_NewConnectionManager_DisconnectInactive(t *testing.T) {
 	lh := newTestLighthouse()
 	ifce := &Interface{
 		hostMap:          hostMap,
-		inside:           &device.NoopTun{},
+		inside:           &test.NoopTun{},
 		outside:          &udp.NoopConn{},
 		firewall:         &Firewall{},
 		lightHouse:       lh,
@@ -348,7 +347,7 @@ func Test_NewConnectionManagerTest_DisconnectInvalid(t *testing.T) {
 	lh := newTestLighthouse()
 	ifce := &Interface{
 		hostMap:          hostMap,
-		inside:           &device.NoopTun{},
+		inside:           &test.NoopTun{},
 		outside:          &udp.NoopConn{},
 		firewall:         &Firewall{},
 		lightHouse:       lh,

control_tester.go

@@ -1,5 +1,4 @@
 //go:build e2e_testing
-// +build e2e_testing
 
 package nebula
 

e2e/snat_test.go

@@ -0,0 +1,400 @@
+//go:build e2e_testing
+// +build e2e_testing
+
+package e2e
+
+import (
+	"encoding/binary"
+	"net/netip"
+	"testing"
+	"time"
+
+	"github.com/slackhq/nebula/cert"
+	"github.com/slackhq/nebula/cert_test"
+	"github.com/slackhq/nebula/e2e/router"
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+)
+
+// parseIPv4UDPPacket extracts source/dest IPs, ports, and payload from an IPv4 UDP packet.
+func parseIPv4UDPPacket(t testing.TB, pkt []byte) (srcIP, dstIP netip.Addr, srcPort, dstPort uint16, payload []byte) {
+	t.Helper()
+	require.True(t, len(pkt) >= 28, "packet too short for IPv4+UDP header")
+	require.Equal(t, byte(0x45), pkt[0]&0xF0|pkt[0]&0x0F, "not a simple IPv4 packet (IHL!=5)")
+
+	srcIP, _ = netip.AddrFromSlice(pkt[12:16])
+	dstIP, _ = netip.AddrFromSlice(pkt[16:20])
+
+	ihl := int(pkt[0]&0x0F) * 4
+	require.True(t, len(pkt) >= ihl+8, "packet too short for UDP header")
+	srcPort = binary.BigEndian.Uint16(pkt[ihl : ihl+2])
+	dstPort = binary.BigEndian.Uint16(pkt[ihl+2 : ihl+4])
+	udpLen := binary.BigEndian.Uint16(pkt[ihl+4 : ihl+6])
+	payload = pkt[ihl+8 : ihl+int(udpLen)]
+	return
+}
+
+func TestSNAT_IPv6OnlyPeer_IPv4UnsafeTraffic(t *testing.T) {
+	// Scenario: Two IPv6-only VPN nodes. The "router" node has unsafe networks
+	// (192.168.0.0/16) in its cert and a configured SNAT address. The "sender"
+	// node has an unsafe route for 192.168.0.0/16 via the router.
+	//
+	// When sender injects an IPv4 packet destined for the unsafe network, it
+	// gets tunneled to the router. The router's firewall detects this is IPv4
+	// from an IPv6-only peer and applies SNAT, rewriting the source IP to the
+	// SNAT address before delivering it to TUN.
+	//
+	// When a reply comes back from TUN addressed to the SNAT address, the
+	// router un-SNATs it (restoring the original destination) and tunnels it
+	// back to the sender.
+
+	ca, _, caKey, _ := cert_test.NewTestCaCert(cert.Version2, cert.Curve_CURVE25519, time.Now(), time.Now().Add(10*time.Minute), nil, nil, []string{})
+
+	unsafePrefix := "192.168.0.0/16"
+	snatAddr := netip.MustParseAddr("169.254.42.42")
+
+	// Router: IPv6-only with unsafe networks and a manual SNAT address.
+	// Override inbound firewall with local_cidr: "any" so both IPv4 (unsafe)
+	// and IPv6 (VPN) traffic is accepted.
+	routerControl, routerVpnIpNet, routerUdpAddr, _ := newSimpleServerWithUdpAndUnsafeNetworks(
+		cert.Version2, ca, caKey, "router", "ff::1/64",
+		netip.MustParseAddrPort("[beef::1]:4242"),
+		unsafePrefix,
+		m{
+			"firewall": m{
+				"inbound": []m{{
+					"proto":      "any",
+					"port":       "any",
+					"host":       "any",
+					"local_cidr": "any",
+				}},
+			},
+			"tun": m{
+				"snat_address_for_4over6": snatAddr.String(),
+			},
+		},
+	)
+
+	// Sender: IPv6-only with an unsafe route via the router
+	senderControl, _, _, _ := newSimpleServerWithUdp(
+		cert.Version2, ca, caKey, "sender", "ff::2/64",
+		netip.MustParseAddrPort("[beef::2]:4242"),
+		m{
+			"tun": m{
+				"unsafe_routes": []m{
+					{"route": unsafePrefix, "via": routerVpnIpNet[0].Addr().String()},
+				},
+			},
+		},
+	)
+
+	// Tell sender where the router lives
+	senderControl.InjectLightHouseAddr(routerVpnIpNet[0].Addr(), routerUdpAddr)
+
+	// Build the router and start both nodes
+	r := router.NewR(t, routerControl, senderControl)
+	defer r.RenderFlow()
+
+	routerControl.Start()
+	senderControl.Start()
+
+	// --- Outbound: sender -> IPv4 unsafe dest (via router with SNAT) ---
+
+	origSrcIP := netip.MustParseAddr("10.0.0.1")
+	unsafeDest := netip.MustParseAddr("192.168.1.1")
+	var origSrcPort uint16 = 12345
+	var dstPort uint16 = 80
+
+	t.Log("Sender injects an IPv4 packet to the unsafe network")
+	senderControl.InjectTunUDPPacket(unsafeDest, dstPort, origSrcIP, origSrcPort, []byte("snat me"))
+
+	t.Log("Route packets (handshake + data) until the router gets the packet on TUN")
+	snatPkt := r.RouteForAllUntilTxTun(routerControl)
+
+	t.Log("Verify the packet was SNATted")
+	gotSrcIP, gotDstIP, gotSrcPort, gotDstPort, gotPayload := parseIPv4UDPPacket(t, snatPkt)
+	assert.Equal(t, snatAddr, gotSrcIP, "source IP should be rewritten to the SNAT address")
+	assert.Equal(t, unsafeDest, gotDstIP, "destination IP should be unchanged")
+	assert.Equal(t, dstPort, gotDstPort, "destination port should be unchanged")
+	assert.Equal(t, []byte("snat me"), gotPayload, "payload should be unchanged")
+
+	// Capture the SNAT port (may differ from original if port was remapped)
+	snatPort := gotSrcPort
+	t.Logf("SNAT port: %d (original: %d)", snatPort, origSrcPort)
+
+	// --- Return: reply from unsafe dest -> un-SNATted back to sender ---
+
+	t.Log("Router injects a reply packet from the unsafe dest to the SNAT address")
+	routerControl.InjectTunUDPPacket(snatAddr, snatPort, unsafeDest, dstPort, []byte("reply from unsafe"))
+
+	t.Log("Route until sender gets the reply on TUN")
+	replyPkt := r.RouteForAllUntilTxTun(senderControl)
+
+	t.Log("Verify the reply was un-SNATted")
+	replySrcIP, replyDstIP, replySrcPort, replyDstPort, replyPayload := parseIPv4UDPPacket(t, replyPkt)
+	assert.Equal(t, unsafeDest, replySrcIP, "reply source should be the unsafe dest")
+	assert.Equal(t, origSrcIP, replyDstIP, "reply dest should be the original source IP (un-SNATted)")
+	assert.Equal(t, dstPort, replySrcPort, "reply source port should be the unsafe dest port")
+	assert.Equal(t, origSrcPort, replyDstPort, "reply dest port should be the original source port (un-SNATted)")
+	assert.Equal(t, []byte("reply from unsafe"), replyPayload, "payload should be unchanged")
+
+	r.RenderHostmaps("Final hostmaps", routerControl, senderControl)
+
+	// Also verify normal IPv6 VPN traffic still works between the nodes
+	t.Log("Verify normal IPv6 VPN tunnel still works")
+	assertTunnel(t, routerVpnIpNet[0].Addr(), senderControl.GetVpnAddrs()[0], routerControl, senderControl, r)
+
+	routerControl.Stop()
+	senderControl.Stop()
+}
+
+func TestSNAT_MultipleFlows(t *testing.T) {
+	// Test that multiple distinct IPv4 flows from the same IPv6-only peer
+	// are tracked independently through SNAT.
+
+	ca, _, caKey, _ := cert_test.NewTestCaCert(cert.Version2, cert.Curve_CURVE25519, time.Now(), time.Now().Add(10*time.Minute), nil, nil, []string{})
+
+	unsafePrefix := "192.168.0.0/16"
+	snatAddr := netip.MustParseAddr("169.254.42.42")
+
+	routerControl, routerVpnIpNet, routerUdpAddr, _ := newSimpleServerWithUdpAndUnsafeNetworks(
+		cert.Version2, ca, caKey, "router", "ff::1/64",
+		netip.MustParseAddrPort("[beef::1]:4242"),
+		unsafePrefix,
+		m{
+			"firewall": m{
+				"inbound": []m{{
+					"proto":      "any",
+					"port":       "any",
+					"host":       "any",
+					"local_cidr": "any",
+				}},
+			},
+			"tun": m{
+				"snat_address_for_4over6": snatAddr.String(),
+			},
+		},
+	)
+
+	senderControl, _, _, _ := newSimpleServerWithUdp(
+		cert.Version2, ca, caKey, "sender", "ff::2/64",
+		netip.MustParseAddrPort("[beef::2]:4242"),
+		m{
+			"tun": m{
+				"unsafe_routes": []m{
+					{"route": unsafePrefix, "via": routerVpnIpNet[0].Addr().String()},
+				},
+			},
+		},
+	)
+
+	senderControl.InjectLightHouseAddr(routerVpnIpNet[0].Addr(), routerUdpAddr)
+
+	r := router.NewR(t, routerControl, senderControl)
+	defer r.RenderFlow()
+	r.CancelFlowLogs()
+
+	routerControl.Start()
+	senderControl.Start()
+
+	unsafeDest := netip.MustParseAddr("192.168.1.1")
+
+	// Send first flow
+	senderControl.InjectTunUDPPacket(unsafeDest, 80, netip.MustParseAddr("10.0.0.1"), 1111, []byte("flow1"))
+	pkt1 := r.RouteForAllUntilTxTun(routerControl)
+	srcIP1, _, srcPort1, _, payload1 := parseIPv4UDPPacket(t, pkt1)
+	assert.Equal(t, snatAddr, srcIP1)
+	assert.Equal(t, []byte("flow1"), payload1)
+
+	// Send second flow (different source port)
+	senderControl.InjectTunUDPPacket(unsafeDest, 80, netip.MustParseAddr("10.0.0.1"), 2222, []byte("flow2"))
+	pkt2 := r.RouteForAllUntilTxTun(routerControl)
+	srcIP2, _, srcPort2, _, payload2 := parseIPv4UDPPacket(t, pkt2)
+	assert.Equal(t, snatAddr, srcIP2)
+	assert.Equal(t, []byte("flow2"), payload2)
+
+	// The two flows should have different SNAT ports (since they're different conntracks)
+	t.Logf("Flow 1 SNAT port: %d, Flow 2 SNAT port: %d", srcPort1, srcPort2)
+
+	// Reply to flow 2 first (out of order)
+	routerControl.InjectTunUDPPacket(snatAddr, srcPort2, unsafeDest, 80, []byte("reply2"))
+	reply2 := r.RouteForAllUntilTxTun(senderControl)
+	_, replyDst2, _, replyDstPort2, replyPayload2 := parseIPv4UDPPacket(t, reply2)
+	assert.Equal(t, netip.MustParseAddr("10.0.0.1"), replyDst2)
+	assert.Equal(t, uint16(2222), replyDstPort2, "reply to flow 2 should restore original port 2222")
+	assert.Equal(t, []byte("reply2"), replyPayload2)
+
+	// Reply to flow 1
+	routerControl.InjectTunUDPPacket(snatAddr, srcPort1, unsafeDest, 80, []byte("reply1"))
+	reply1 := r.RouteForAllUntilTxTun(senderControl)
+	_, replyDst1, _, replyDstPort1, replyPayload1 := parseIPv4UDPPacket(t, reply1)
+	assert.Equal(t, netip.MustParseAddr("10.0.0.1"), replyDst1)
+	assert.Equal(t, uint16(1111), replyDstPort1, "reply to flow 1 should restore original port 1111")
+	assert.Equal(t, []byte("reply1"), replyPayload1)
+
+	routerControl.Stop()
+	senderControl.Stop()
+}
+
+// --- Adversarial SNAT E2E Tests ---
+
+func TestSNAT_UnsolicitedReplyDropped(t *testing.T) {
+	// Without any outbound SNAT traffic, inject a packet from the router's TUN
+	// addressed to the SNAT address. The sender must never receive it because
+	// there's no conntrack entry to un-SNAT through.
+
+	ca, _, caKey, _ := cert_test.NewTestCaCert(cert.Version2, cert.Curve_CURVE25519, time.Now(), time.Now().Add(10*time.Minute), nil, nil, []string{})
+
+	unsafePrefix := "192.168.0.0/16"
+	snatAddr := netip.MustParseAddr("169.254.42.42")
+
+	routerControl, routerVpnIpNet, routerUdpAddr, _ := newSimpleServerWithUdpAndUnsafeNetworks(
+		cert.Version2, ca, caKey, "router", "ff::1/64",
+		netip.MustParseAddrPort("[beef::1]:4242"),
+		unsafePrefix,
+		m{
+			"firewall": m{
+				"inbound": []m{{
+					"proto":      "any",
+					"port":       "any",
+					"host":       "any",
+					"local_cidr": "any",
+				}},
+			},
+			"tun": m{
+				"snat_address_for_4over6": snatAddr.String(),
+			},
+		},
+	)
+
+	senderControl, _, _, _ := newSimpleServerWithUdp(
+		cert.Version2, ca, caKey, "sender", "ff::2/64",
+		netip.MustParseAddrPort("[beef::2]:4242"),
+		m{
+			"tun": m{
+				"unsafe_routes": []m{
+					{"route": unsafePrefix, "via": routerVpnIpNet[0].Addr().String()},
+				},
+			},
+		},
+	)
+
+	senderControl.InjectLightHouseAddr(routerVpnIpNet[0].Addr(), routerUdpAddr)
+
+	r := router.NewR(t, routerControl, senderControl)
+	defer r.RenderFlow()
+	r.CancelFlowLogs()
+
+	routerControl.Start()
+	senderControl.Start()
+
+	// First establish the tunnel with normal IPv6 traffic so handshake completes
+	assertTunnel(t, routerVpnIpNet[0].Addr(), senderControl.GetVpnAddrs()[0], routerControl, senderControl, r)
+
+	// Inject the unsolicited reply from router's TUN to the SNAT address.
+	// There is NO prior outbound SNAT flow, so no conntrack entry exists.
+	// The router should silently drop this because unSnat finds no matching conntrack.
+	routerControl.InjectTunUDPPacket(snatAddr, 55555, netip.MustParseAddr("192.168.1.1"), 80, []byte("unsolicited"))
+
+	// Send a canary IPv6 VPN packet after the bad one. Since the router processes
+	// TUN packets sequentially, the canary arriving proves the bad packet was processed first.
+	senderVpnAddr := senderControl.GetVpnAddrs()[0]
+	routerControl.InjectTunUDPPacket(senderVpnAddr, 90, routerVpnIpNet[0].Addr(), 80, []byte("canary"))
+	canaryPkt := r.RouteForAllUntilTxTun(senderControl)
+	assertUdpPacket(t, []byte("canary"), canaryPkt, routerVpnIpNet[0].Addr(), senderVpnAddr, 80, 90)
+
+	// The unsolicited packet should have been dropped — nothing else on sender's TUN
+	got := senderControl.GetFromTun(false)
+	assert.Nil(t, got, "sender should not receive unsolicited packet to SNAT address with no conntrack entry")
+
+	routerControl.Stop()
+	senderControl.Stop()
+}
+
+func TestSNAT_NonUnsafeDestDropped(t *testing.T) {
+	// An IPv6-only sender sends IPv4 traffic to a destination outside the router's
+	// unsafe networks (172.16.0.1 when unsafe is 192.168.0.0/16). The router should
+	// reject this because the local address is not routable. This verifies that
+	// willingToHandleLocalAddr enforces boundaries on what SNAT traffic can reach.
+
+	ca, _, caKey, _ := cert_test.NewTestCaCert(cert.Version2, cert.Curve_CURVE25519, time.Now(), time.Now().Add(10*time.Minute), nil, nil, []string{})
+
+	unsafePrefix := "192.168.0.0/16"
+	snatAddr := netip.MustParseAddr("169.254.42.42")
+
+	routerControl, routerVpnIpNet, routerUdpAddr, _ := newSimpleServerWithUdpAndUnsafeNetworks(
+		cert.Version2, ca, caKey, "router", "ff::1/64",
+		netip.MustParseAddrPort("[beef::1]:4242"),
+		unsafePrefix,
+		m{
+			"firewall": m{
+				"inbound": []m{{
+					"proto":      "any",
+					"port":       "any",
+					"host":       "any",
+					"local_cidr": "any",
+				}},
+			},
+			"tun": m{
+				"snat_address_for_4over6": snatAddr.String(),
+			},
+		},
+	)
+
+	// Sender has unsafe routes for BOTH 192.168.0.0/16 AND 172.16.0.0/12 via router.
+	// This means the sender will route 172.16.0.1 through the tunnel to the router.
+	// But the router should reject it because 172.16.0.0/12 is NOT in its unsafe networks.
+	senderControl, _, _, _ := newSimpleServerWithUdp(
+		cert.Version2, ca, caKey, "sender", "ff::2/64",
+		netip.MustParseAddrPort("[beef::2]:4242"),
+		m{
+			"tun": m{
+				"unsafe_routes": []m{
+					{"route": unsafePrefix, "via": routerVpnIpNet[0].Addr().String()},
+					{"route": "172.16.0.0/12", "via": routerVpnIpNet[0].Addr().String()},
+				},
+			},
+		},
+	)
+
+	senderControl.InjectLightHouseAddr(routerVpnIpNet[0].Addr(), routerUdpAddr)
+
+	r := router.NewR(t, routerControl, senderControl)
+	defer r.RenderFlow()
+	r.CancelFlowLogs()
+
+	routerControl.Start()
+	senderControl.Start()
+
+	// Establish the tunnel first
+	assertTunnel(t, routerVpnIpNet[0].Addr(), senderControl.GetVpnAddrs()[0], routerControl, senderControl, r)
+
+	// Send to 172.16.0.1 (NOT in router's unsafe networks 192.168.0.0/16).
+	// The router should reject this at willingToHandleLocalAddr.
+	senderControl.InjectTunUDPPacket(
+		netip.MustParseAddr("172.16.0.1"), 80,
+		netip.MustParseAddr("10.0.0.1"), 12345,
+		[]byte("wrong dest"),
+	)
+
+	// Send a canary to a valid unsafe destination to prove the bad packet was processed
+	senderControl.InjectTunUDPPacket(
+		netip.MustParseAddr("192.168.1.1"), 80,
+		netip.MustParseAddr("10.0.0.1"), 33333,
+		[]byte("canary"),
+	)
+
+	// Route until the canary arrives — the 172.16.0.1 packet should have been
+	// processed and dropped before the canary gets through
+	canaryPkt := r.RouteForAllUntilTxTun(routerControl)
+	_, canaryDst, _, _, canaryPayload := parseIPv4UDPPacket(t, canaryPkt)
+	assert.Equal(t, netip.MustParseAddr("192.168.1.1"), canaryDst, "canary should arrive at the valid unsafe dest")
+	assert.Equal(t, []byte("canary"), canaryPayload)
+
+	// No more packets — the 172.16.0.1 packet was dropped
+	got := routerControl.GetFromTun(false)
+	assert.Nil(t, got, "packet to non-unsafe destination 172.16.0.1 should be dropped by the router")
+
+	routerControl.Stop()
+	senderControl.Stop()
+}

examples/config.yml

@@ -283,6 +283,15 @@ tun:
   # If using massive routes updates, for example BGP, you may need to increase this value to avoid packet loss.
   # SO_RCVBUFFORCE is used to avoid having to raise the system wide max
   #use_system_route_table_buffer_size: 0
+  #
+  # When using the feature to carry IPv4 unsafe_routes through a IPv6-only overlay network (4over6),
+  # the system still needs to use IPv4 addresses to pass that traffic. 
+  # By default, they are randomly chosen out of 169.254.0.0/16 at startup, but if you need them to be static, you can set them here.
+  # Setting these options is not required for use of the 4over6 feature.
+  # this is the address that a client will use as a source IP for packets sent to the router
+  #unsafe_origin_address_for_4over6: 169.254.55.96
+  # this is the IP the router will use internally to hairpin-NAT 4over6 traffic before it hits the operating system.
+  #snat_address_for_4over6: 169.254.12.34
 
 # Configure logging level
 logging:
@@ -382,8 +391,8 @@ firewall:
   # Rules are comprised of a protocol, port, and one or more of host, group, or CIDR
   # Logical evaluation is roughly: port AND proto AND (ca_sha OR ca_name) AND (host OR group OR groups OR cidr) AND (local cidr)
   # - port: Takes `0` or `any` as any, a single number `80`, a range `200-901`, or `fragment` to match second and further fragments of fragmented packets (since there is no port available).
-  #   code: same as port but makes more sense when talking about ICMP, TODO: this is not currently implemented in a way that works, use `any`
   #   proto: `any`, `tcp`, `udp`, or `icmp`
+  #          a port specification is ignored if proto is `icmp`
   #   host: `any` or a literal hostname, ie `test-host`
   #   group: `any` or a literal group name, ie `default-group`
   #   groups: Same as group but accepts a list of values. Multiple values are AND'd together and a certificate would have to contain all groups to pass

firewall.go

@@ -2,6 +2,7 @@ package nebula
 
 import (
 	"crypto/sha256"
+	"encoding/binary"
 	"encoding/hex"
 	"errors"
 	"fmt"
@@ -22,10 +23,35 @@ import (
 	"github.com/slackhq/nebula/firewall"
 )
 
+var ErrCannotSNAT = errors.New("cannot SNAT this packet")
+var ErrSNATIdentityMismatch = errors.New("refusing to SNAT for mismatched host")
+var ErrSNATAddressCollision = errors.New("refusing to accept an incoming packet with my SNAT address")
+
+const ipv4SourcePosition = 12
+const ipv4DestinationPosition = 16
+const sourcePortOffset = 0
+const destinationPortOffset = 2
+
 type FirewallInterface interface {
 	AddRule(incoming bool, proto uint8, startPort int32, endPort int32, groups []string, host string, cidr, localCidr string, caName string, caSha string) error
 }
 
+type snatInfo struct {
+	//Src is the source IP+port to write into unsafe-route-bound packet
+	Src netip.AddrPort
+	//SrcVpnIp is the overlay IP associated with this flow. It's needed to associate reply traffic so we can get it back to the right host.
+	SrcVpnIp netip.Addr
+	//SnatPort is the port to rewrite into an overlay-bound packet
+	SnatPort uint16
+}
+
+func (s *snatInfo) Valid() bool {
+	if s == nil {
+		return false
+	}
+	return s.Src.IsValid()
+}
+
 type conn struct {
 	Expires time.Time // Time when this conntrack entry will expire
 
@@ -34,6 +60,9 @@ type conn struct {
 	// fields pack for free after the uint32 above
 	incoming     bool
 	rulesVersion uint16
+
+	//for SNAT support
+	snat *snatInfo
 }
 
 // TODO: need conntrack max tracked connections handling
@@ -66,6 +95,8 @@ type Firewall struct {
 	defaultLocalCIDRAny bool
 	incomingMetrics     firewallMetrics
 	outgoingMetrics     firewallMetrics
+	unsafeIPv4Origin    netip.Addr
+	snatAddr            netip.Addr
 
 	l *logrus.Logger
 }
@@ -81,14 +112,8 @@ type FirewallConntrack struct {
 
 	Conns      map[firewall.Packet]*conn
 	TimerWheel *TimerWheel[firewall.Packet]
-
-	// purgeCounter tracks lookups to trigger periodic purge instead of every lookup
-	purgeCounter uint32
 }
 
-// purgeInterval defines how many lookups between purge attempts
-const conntrackPurgeInterval = 1024
-
 // FirewallTable is the entry point for a rule, the evaluation order is:
 // Proto AND port AND (CA SHA or CA name) AND local CIDR AND (group OR groups OR name OR remote CIDR)
 type FirewallTable struct {
@@ -199,12 +224,12 @@ func NewFirewall(l *logrus.Logger, tcpTimeout, UDPTimeout, defaultTimeout time.D
 
 func NewFirewallFromConfig(l *logrus.Logger, cs *CertState, c *config.C) (*Firewall, error) {
 	certificate := cs.getCertificate(cert.Version2)
-	if certificate == nil {
+	if certificate == nil { //todo if config.initiating_version is set to 1, and unsafe_networks differ, things will suck
 		certificate = cs.getCertificate(cert.Version1)
 	}
 
 	if certificate == nil {
-		panic("No certificate available to reconfigure the firewall")
+		return nil, errors.New("no certificate available to reconfigure the firewall")
 	}
 
 	fw := NewFirewall(
@@ -213,7 +238,6 @@ func NewFirewallFromConfig(l *logrus.Logger, cs *CertState, c *config.C) (*Firew
 		c.GetDuration("firewall.conntrack.udp_timeout", time.Minute*3),
 		c.GetDuration("firewall.conntrack.default_timeout", time.Minute*10),
 		certificate,
-		//TODO: max_connections
 	)
 
 	fw.defaultLocalCIDRAny = c.GetBool("firewall.default_local_cidr_any", false)
@@ -255,20 +279,6 @@ func NewFirewallFromConfig(l *logrus.Logger, cs *CertState, c *config.C) (*Firew
 
 // AddRule properly creates the in memory rule structure for a firewall table.
 func (f *Firewall) AddRule(incoming bool, proto uint8, startPort int32, endPort int32, groups []string, host string, cidr, localCidr, caName string, caSha string) error {
-	// We need this rule string because we generate a hash. Removing this will break firewall reload.
-	ruleString := fmt.Sprintf(
-		"incoming: %v, proto: %v, startPort: %v, endPort: %v, groups: %v, host: %v, ip: %v, localIp: %v, caName: %v, caSha: %s",
-		incoming, proto, startPort, endPort, groups, host, cidr, localCidr, caName, caSha,
-	)
-	f.rules += ruleString + "\n"
-
-	direction := "incoming"
-	if !incoming {
-		direction = "outgoing"
-	}
-	f.l.WithField("firewallRule", m{"direction": direction, "proto": proto, "startPort": startPort, "endPort": endPort, "groups": groups, "host": host, "cidr": cidr, "localCidr": localCidr, "caName": caName, "caSha": caSha}).
-		Info("Firewall rule added")
-
 	var (
 		ft *FirewallTable
 		fp firewallPort
@@ -286,6 +296,12 @@ func (f *Firewall) AddRule(incoming bool, proto uint8, startPort int32, endPort
 	case firewall.ProtoUDP:
 		fp = ft.UDP
 	case firewall.ProtoICMP, firewall.ProtoICMPv6:
+		//ICMP traffic doesn't have ports, so we always coerce to "any", even if a value is provided
+		if startPort != firewall.PortAny {
+			f.l.WithField("startPort", startPort).Warn("ignoring port specification for ICMP firewall rule")
+		}
+		startPort = firewall.PortAny
+		endPort = firewall.PortAny
 		fp = ft.ICMP
 	case firewall.ProtoAny:
 		fp = ft.AnyProto
@@ -293,6 +309,20 @@ func (f *Firewall) AddRule(incoming bool, proto uint8, startPort int32, endPort
 		return fmt.Errorf("unknown protocol %v", proto)
 	}
 
+	// We need this rule string because we generate a hash. Removing this will break firewall reload.
+	ruleString := fmt.Sprintf(
+		"incoming: %v, proto: %v, startPort: %v, endPort: %v, groups: %v, host: %v, ip: %v, localIp: %v, caName: %v, caSha: %s",
+		incoming, proto, startPort, endPort, groups, host, cidr, localCidr, caName, caSha,
+	)
+	f.rules += ruleString + "\n"
+
+	direction := "incoming"
+	if !incoming {
+		direction = "outgoing"
+	}
+	f.l.WithField("firewallRule", m{"direction": direction, "proto": proto, "startPort": startPort, "endPort": endPort, "groups": groups, "host": host, "cidr": cidr, "localCidr": localCidr, "caName": caName, "caSha": caSha}).
+		Info("Firewall rule added")
+
 	return fp.addRule(f, startPort, endPort, groups, host, cidr, localCidr, caName, caSha)
 }
 
@@ -314,6 +344,18 @@ func (f *Firewall) GetRuleHashes() string {
 	return "SHA:" + f.GetRuleHash() + ",FNV:" + strconv.FormatUint(uint64(f.GetRuleHashFNV()), 10)
 }
 
+func (f *Firewall) SetSNATAddressFromInterface(i *Interface) {
+	//address-mutation-avoidance is done inside Interface, the firewall doesn't need to care
+	//todo should snatted conntracks get expired out? Probably not needed until if/when we allow reload
+	f.snatAddr = i.inside.SNATAddress().Addr()
+	f.unsafeIPv4Origin = i.inside.UnsafeIPv4OriginAddress().Addr()
+}
+
+func (f *Firewall) ShouldUnSNAT(fp *firewall.Packet) bool {
+	// f.snatAddr is only valid if we're a snat-capable router
+	return f.snatAddr.IsValid() && fp.RemoteAddr == f.snatAddr
+}
+
 func AddFirewallRulesFromConfig(l *logrus.Logger, inbound bool, c *config.C, fw FirewallInterface) error {
 	var table string
 	if inbound {
@@ -355,24 +397,31 @@ func AddFirewallRulesFromConfig(l *logrus.Logger, inbound bool, c *config.C, fw
 			sPort = r.Port
 		}
 
-		startPort, endPort, err := parsePort(sPort)
-		if err != nil {
-			return fmt.Errorf("%s rule #%v; %s %s", table, i, errPort, err)
-		}
-
 		var proto uint8
+		var startPort, endPort int32
 		switch r.Proto {
 		case "any":
 			proto = firewall.ProtoAny
+			startPort, endPort, err = parsePort(sPort)
 		case "tcp":
 			proto = firewall.ProtoTCP
+			startPort, endPort, err = parsePort(sPort)
 		case "udp":
 			proto = firewall.ProtoUDP
+			startPort, endPort, err = parsePort(sPort)
 		case "icmp":
 			proto = firewall.ProtoICMP
+			startPort = firewall.PortAny
+			endPort = firewall.PortAny
+			if sPort != "" {
+				l.WithField("port", sPort).Warn("ignoring port specification for ICMP firewall rule")
+			}
 		default:
 			return fmt.Errorf("%s rule #%v; proto was not understood; `%s`", table, i, r.Proto)
 		}
+		if err != nil {
+			return fmt.Errorf("%s rule #%v; %s %s", table, i, errPort, err)
+		}
 
 		if r.Cidr != "" && r.Cidr != "any" {
 			_, err = netip.ParsePrefix(r.Cidr)
@@ -407,50 +456,207 @@ var ErrInvalidRemoteIP = errors.New("remote address is not in remote certificate
 var ErrInvalidLocalIP = errors.New("local address is not in list of handled local addresses")
 var ErrNoMatchingRule = errors.New("no matching rule in firewall table")
 
+func (f *Firewall) unSnat(data []byte, fp *firewall.Packet) netip.Addr {
+	c := f.peek(*fp) //unfortunately this needs to lock. Surely there's a better way.
+	if c == nil {
+		return netip.Addr{}
+	}
+	if !c.snat.Valid() {
+		return netip.Addr{}
+	}
+	oldIP := netip.AddrPortFrom(f.snatAddr, fp.RemotePort)
+	rewritePacket(data, fp, oldIP, c.snat.Src, ipv4DestinationPosition, destinationPortOffset)
+	return c.snat.SrcVpnIp
+}
+
+func rewritePacket(data []byte, fp *firewall.Packet, oldIP netip.AddrPort, newIP netip.AddrPort, ipOffset int, portOffset int) {
+	//change address
+	copy(data[ipOffset:], newIP.Addr().AsSlice())
+	recalcIPv4Checksum(data, oldIP.Addr(), newIP.Addr())
+	ipHeaderLen := int(data[0]&0x0F) * 4
+
+	switch fp.Protocol {
+	case firewall.ProtoICMP:
+		binary.BigEndian.PutUint16(data[ipHeaderLen+4:ipHeaderLen+6], newIP.Port()) //we use the ID field as a "port" for ICMP
+		icmpCode := uint16(data[ipHeaderLen+1])                                     //todo not snatting on code yet (but Linux would)
+		recalcICMPv4Checksum(data, icmpCode, icmpCode, oldIP.Port(), newIP.Port())
+	case firewall.ProtoUDP:
+		dstport := ipHeaderLen + portOffset
+		binary.BigEndian.PutUint16(data[dstport:dstport+2], newIP.Port())
+		recalcUDPv4Checksum(data, oldIP, newIP)
+	case firewall.ProtoTCP:
+		dstport := ipHeaderLen + portOffset
+		binary.BigEndian.PutUint16(data[dstport:dstport+2], newIP.Port())
+		recalcTCPv4Checksum(data, oldIP, newIP)
+	}
+}
+
+func (f *Firewall) findUsableSNATPort(fp *firewall.Packet, c *conn) error {
+	const halfThePorts = 0x7fff
+	oldPort := fp.RemotePort
+	conntrack := f.Conntrack
+	conntrack.Lock()
+	defer conntrack.Unlock()
+	for numPortsChecked := 0; numPortsChecked < halfThePorts; numPortsChecked++ {
+		_, ok := conntrack.Conns[*fp]
+		if !ok {
+			//yay, we can use this port
+			//track the snatted flow with the same expiration as the unsnatted version
+			c.snat.SnatPort = fp.RemotePort
+			conntrack.Conns[*fp] = c
+			return nil
+		}
+		//increment and retry. There's probably better strategies out there
+		fp.RemotePort++
+		if fp.RemotePort < halfThePorts {
+			fp.RemotePort += halfThePorts // keep it ephemeral for now
+		}
+	}
+
+	//if we made it here, we failed
+	fp.RemotePort = oldPort
+	return ErrCannotSNAT
+}
+
+func (f *Firewall) applySnat(data []byte, fp *firewall.Packet, c *conn, hostinfo *HostInfo) error {
+	if !f.snatAddr.IsValid() {
+		return ErrCannotSNAT
+	}
+	if f.snatAddr == fp.LocalAddr { //a packet that came from UDP (incoming) should never ever have our snat address on it
+		return ErrSNATAddressCollision
+	}
+	if c.snat.Valid() {
+		//old flow: make sure it came from the right place
+		if !slices.Contains(hostinfo.vpnAddrs, c.snat.SrcVpnIp) {
+			return ErrSNATIdentityMismatch
+		}
+		fp.RemoteAddr = f.snatAddr
+		fp.RemotePort = c.snat.SnatPort
+	} else if hostinfo.vpnAddrs[0].Is6() {
+		//we got a new flow
+		c.snat = &snatInfo{
+			Src:      netip.AddrPortFrom(fp.RemoteAddr, fp.RemotePort),
+			SrcVpnIp: hostinfo.vpnAddrs[0],
+		}
+		fp.RemoteAddr = f.snatAddr
+		//find a new port to use, if needed
+		err := f.findUsableSNATPort(fp, c)
+		if err != nil {
+			c.snat = nil
+			return err
+		}
+	} else {
+		return ErrCannotSNAT
+	}
+
+	newIP := netip.AddrPortFrom(f.snatAddr, c.snat.SnatPort)
+	rewritePacket(data, fp, c.snat.Src, newIP, ipv4SourcePosition, sourcePortOffset)
+	return nil
+}
+
+func (f *Firewall) identifyRemoteNetworkType(h *HostInfo, fp firewall.Packet) NetworkType {
+	if h.networks == nil {
+		// Simple case: Certificate has one address and no unsafe networks
+		if h.vpnAddrs[0] == fp.RemoteAddr {
+			return NetworkTypeVPN
+		} //else, fallthrough
+	} else if nwType, ok := h.networks.Lookup(fp.RemoteAddr); ok {
+		return nwType //will return NetworkTypeVPN or NetworkTypeUnsafe
+	}
+
+	//RemoteAddr not in our networks table
+	if f.snatAddr.IsValid() && fp.IsIPv4() && h.HasOnlyV6Addresses() {
+		return NetworkTypeUnverifiedSNATPeer
+	} else {
+		return NetworkTypeInvalidPeer
+	}
+}
+
+func (f *Firewall) allowRemoteNetworkType(nwType NetworkType, fp firewall.Packet) error {
+	switch nwType {
+	case NetworkTypeVPN:
+		return nil
+	case NetworkTypeInvalidPeer:
+		return ErrInvalidRemoteIP
+	case NetworkTypeVPNPeer:
+		//one day we might need a specialSnatMode case in here to handle routers with v4 addresses when we don't also have a v4 address?
+		return ErrPeerRejected // reject for now, one day this may have different FW rules
+	case NetworkTypeUnsafe:
+		return nil // nothing special, one day this may have different FW rules
+	case NetworkTypeUnverifiedSNATPeer:
+		if f.unsafeIPv4Origin.IsValid() && fp.LocalAddr == f.unsafeIPv4Origin {
+			return nil //the client case
+		}
+		if f.snatAddr.IsValid() {
+			if fp.RemoteAddr == f.snatAddr {
+				return ErrInvalidRemoteIP //we should never get a packet with our SNAT addr as the destination, or "from" our SNAT addr
+			}
+			return nil
+		} else {
+			return ErrInvalidRemoteIP
+		}
+	default:
+		return ErrUnknownNetworkType //should never happen
+	}
+}
+
+func (f *Firewall) willingToHandleLocalAddr(incoming bool, fp firewall.Packet, remoteNwType NetworkType) error {
+	if f.routableNetworks.Contains(fp.LocalAddr) {
+		return nil //easy, this should handle NetworkTypeVPN in all cases, and NetworkTypeUnsafe on the router side
+	}
+	if incoming { //at least for now, reject all traffic other than what we've already decided is locally routable
+		return ErrInvalidLocalIP
+	}
+
+	//below this line, all traffic is outgoing. Outgoing traffic to NetworkTypeUnsafe is not required to be considered inbound-routable
+	if remoteNwType == NetworkTypeUnsafe {
+		return nil
+	}
+
+	return ErrInvalidLocalIP
+}
+
 // Drop returns an error if the packet should be dropped, explaining why. It
 // returns nil if the packet should not be dropped.
-func (f *Firewall) Drop(fp firewall.Packet, incoming bool, h *HostInfo, caPool *cert.CAPool, localCache firewall.ConntrackCache) error {
+func (f *Firewall) Drop(fp firewall.Packet, pkt []byte, incoming bool, h *HostInfo, caPool *cert.CAPool, localCache firewall.ConntrackCache) error {
+	table := f.OutRules
+	if incoming {
+		table = f.InRules
+	}
+
+	snatmode := fp.IsIPv4() && h.HasOnlyV6Addresses() && f.snatAddr.IsValid()
+	if snatmode {
+		//if this is an IPv4 packet from a V6 only host, and we're configured to snat that kind of traffic, it must be snatted,
+		//so it can never be in the localcache, which lacks SNAT data
+		//nil out the pointer to avoid ever using it
+		localCache = nil
+	}
+
 	// Check if we spoke to this tuple, if we did then allow this packet
-	if f.inConns(fp, h, caPool, localCache) {
+	if localCache != nil {
+		if _, ok := localCache[fp]; ok {
+			return nil //packet matched the cache, we're not snatting, we can return early
+		}
+	}
+	c := f.inConns(fp, h, caPool, localCache)
+	if c != nil {
+		if incoming && snatmode {
+			return f.applySnat(pkt, &fp, c, h)
+		}
 		return nil
 	}
 
 	// Make sure remote address matches nebula certificate, and determine how to treat it
-	if h.networks == nil {
+	remoteNetworkType := f.identifyRemoteNetworkType(h, fp)
-		// Simple case: Certificate has one address and no unsafe networks
+	if err := f.allowRemoteNetworkType(remoteNetworkType, fp); err != nil {
-		if h.vpnAddrs[0] != fp.RemoteAddr {
+		f.metrics(incoming).droppedRemoteAddr.Inc(1)
-			f.metrics(incoming).droppedRemoteAddr.Inc(1)
+		return err
-			return ErrInvalidRemoteIP
-		}
-	} else {
-		nwType, ok := h.networks.Lookup(fp.RemoteAddr)
-		if !ok {
-			f.metrics(incoming).droppedRemoteAddr.Inc(1)
-			return ErrInvalidRemoteIP
-		}
-		switch nwType {
-		case NetworkTypeVPN:
-			break // nothing special
-		case NetworkTypeVPNPeer:
-			f.metrics(incoming).droppedRemoteAddr.Inc(1)
-			return ErrPeerRejected // reject for now, one day this may have different FW rules
-		case NetworkTypeUnsafe:
-			break // nothing special, one day this may have different FW rules
-		default:
-			f.metrics(incoming).droppedRemoteAddr.Inc(1)
-			return ErrUnknownNetworkType //should never happen
-		}
 	}
 
 	// Make sure we are supposed to be handling this local ip address
-	if !f.routableNetworks.Contains(fp.LocalAddr) {
+	if err := f.willingToHandleLocalAddr(incoming, fp, remoteNetworkType); err != nil {
 		f.metrics(incoming).droppedLocalAddr.Inc(1)
-		return ErrInvalidLocalIP
+		return err
-	}
-
-	table := f.OutRules
-	if incoming {
-		table = f.InRules
 	}
 
 	// We now know which firewall table to check against
@@ -460,9 +666,14 @@ func (f *Firewall) Drop(fp firewall.Packet, incoming bool, h *HostInfo, caPool *
 	}
 
 	// We always want to conntrack since it is a faster operation
-	f.addConn(fp, incoming)
+	c = f.addConn(fp, incoming)
 
-	return nil
+	if incoming && remoteNetworkType == NetworkTypeUnverifiedSNATPeer {
+		return f.applySnat(pkt, &fp, c, h)
+	} else {
+		//outgoing snat is handled before this function is called
+		return nil
+	}
 }
 
 func (f *Firewall) metrics(incoming bool) firewallMetrics {
@@ -473,7 +684,7 @@ func (f *Firewall) metrics(incoming bool) firewallMetrics {
 	}
 }
 
-// Destroy cleans up any known cyclical references so the object can be free'd my GC. This should be called if a new
+// Destroy cleans up any known cyclical references so the object can be freed by GC. This should be called if a new
 // firewall object is created
 func (f *Firewall) Destroy() {
 	//TODO: clean references if/when needed
@@ -489,29 +700,28 @@ func (f *Firewall) EmitStats() {
 	metrics.GetOrRegisterGauge("firewall.rules.hash", nil).Update(int64(f.GetRuleHashFNV()))
 }
 
-func (f *Firewall) inConns(fp firewall.Packet, h *HostInfo, caPool *cert.CAPool, localCache firewall.ConntrackCache) bool {
+func (f *Firewall) peek(fp firewall.Packet) *conn {
-	if localCache != nil {
+	f.Conntrack.Lock()
-		if _, ok := localCache[fp]; ok {
+	c := f.Conntrack.Conns[fp]
-			return true
+	f.Conntrack.Unlock()
-		}
+	return c
-	}
+}
+
+func (f *Firewall) inConns(fp firewall.Packet, h *HostInfo, caPool *cert.CAPool, localCache firewall.ConntrackCache) *conn {
 	conntrack := f.Conntrack
 	conntrack.Lock()
 
-	// Periodic purge instead of every lookup (major CPU savings)
+	// Purge every time we test
-	conntrack.purgeCounter++
+	ep, has := conntrack.TimerWheel.Purge()
-	if conntrack.purgeCounter >= conntrackPurgeInterval {
+	if has {
-		conntrack.purgeCounter = 0
+		f.evict(ep)
-		ep, has := conntrack.TimerWheel.Purge()
-		if has {
-			f.evict(ep)
-		}
 	}
 
 	c, ok := conntrack.Conns[fp]
+
 	if !ok {
 		conntrack.Unlock()
-		return false
+		return nil
 	}
 
 	if c.rulesVersion != f.rulesVersion {
@@ -534,7 +744,7 @@ func (f *Firewall) inConns(fp firewall.Packet, h *HostInfo, caPool *cert.CAPool,
 			}
 			delete(conntrack.Conns, fp)
 			conntrack.Unlock()
-			return false
+			return nil
 		}
 
 		if f.l.Level >= logrus.DebugLevel {
@@ -564,12 +774,11 @@ func (f *Firewall) inConns(fp firewall.Packet, h *HostInfo, caPool *cert.CAPool,
 		localCache[fp] = struct{}{}
 	}
 
-	return true
+	return c
 }
 
-func (f *Firewall) addConn(fp firewall.Packet, incoming bool) {
+func (f *Firewall) packetTimeout(fp firewall.Packet) time.Duration {
 	var timeout time.Duration
-	c := &conn{}
 
 	switch fp.Protocol {
 	case firewall.ProtoTCP:
@@ -579,7 +788,13 @@ func (f *Firewall) addConn(fp firewall.Packet, incoming bool) {
 	default:
 		timeout = f.DefaultTimeout
 	}
+	return timeout
+}
 
+func (f *Firewall) addConn(fp firewall.Packet, incoming bool) *conn {
+	c := &conn{}
+
+	timeout := f.packetTimeout(fp)
 	conntrack := f.Conntrack
 	conntrack.Lock()
 	if _, ok := conntrack.Conns[fp]; !ok {
@@ -593,7 +808,9 @@ func (f *Firewall) addConn(fp firewall.Packet, incoming bool) {
 	c.rulesVersion = f.rulesVersion
 	c.Expires = time.Now().Add(timeout)
 	conntrack.Conns[fp] = c
+
 	conntrack.Unlock()
+	return c
 }
 
 // Evict checks if a conntrack entry has expired, if so it is removed, if not it is re-added to the wheel
@@ -669,6 +886,13 @@ func (fp firewallPort) match(p firewall.Packet, incoming bool, c *cert.CachedCer
 		return false
 	}
 
+	// this branch is here to catch traffic from FirewallTable.Any.match and FirewallTable.ICMP.match
+	if p.Protocol == firewall.ProtoICMP || p.Protocol == firewall.ProtoICMPv6 {
+		// port numbers are re-used for connection tracking of ICMP,
+		// but we don't want to actually filter on them.
+		return fp[firewall.PortAny].match(p, c, caPool)
+	}
+
 	var port int32
 
 	if p.Fragment {
@@ -813,10 +1037,8 @@ func (fr *FirewallRule) isAny(groups []string, host string, cidr string) bool {
 		return true
 	}
 
-	for _, group := range groups {
+	if slices.Contains(groups, "any") {
-		if group == "any" {
+		return true
-			return true
-		}
 	}
 
 	if host == "any" {
@@ -1027,54 +1249,56 @@ func (r *rule) sanity() error {
 		}
 	}
 
+	if r.Code != "" {
+		return fmt.Errorf("code specified as [%s]. Support for 'code' will be dropped in a future release, as it has never been functional", r.Code)
+	}
+
 	//todo alert on cidr-any
 
 	return nil
 }
 
-func parsePort(s string) (startPort, endPort int32, err error) {
+func parsePort(s string) (int32, int32, error) {
+	var err error
+	const notAPort int32 = -2
 	if s == "any" {
-		startPort = firewall.PortAny
+		return firewall.PortAny, firewall.PortAny, nil
-		endPort = firewall.PortAny
+	}
-
+	if s == "fragment" {
-	} else if s == "fragment" {
+		return firewall.PortFragment, firewall.PortFragment, nil
-		startPort = firewall.PortFragment
+	}
-		endPort = firewall.PortFragment
+	if !strings.Contains(s, `-`) {
-
-	} else if strings.Contains(s, `-`) {
-		sPorts := strings.SplitN(s, `-`, 2)
-		sPorts[0] = strings.Trim(sPorts[0], " ")
-		sPorts[1] = strings.Trim(sPorts[1], " ")
-
-		if len(sPorts) != 2 || sPorts[0] == "" || sPorts[1] == "" {
-			return 0, 0, fmt.Errorf("appears to be a range but could not be parsed; `%s`", s)
-		}
-
-		rStartPort, err := strconv.Atoi(sPorts[0])
-		if err != nil {
-			return 0, 0, fmt.Errorf("beginning range was not a number; `%s`", sPorts[0])
-		}
-
-		rEndPort, err := strconv.Atoi(sPorts[1])
-		if err != nil {
-			return 0, 0, fmt.Errorf("ending range was not a number; `%s`", sPorts[1])
-		}
-
-		startPort = int32(rStartPort)
-		endPort = int32(rEndPort)
-
-		if startPort == firewall.PortAny {
-			endPort = firewall.PortAny
-		}
-
-	} else {
 		rPort, err := strconv.Atoi(s)
 		if err != nil {
-			return 0, 0, fmt.Errorf("was not a number; `%s`", s)
+			return notAPort, notAPort, fmt.Errorf("was not a number; `%s`", s)
 		}
-		startPort = int32(rPort)
+		return int32(rPort), int32(rPort), nil
-		endPort = startPort
 	}
 
-	return
+	sPorts := strings.SplitN(s, `-`, 2)
+	for i := range sPorts {
+		sPorts[i] = strings.Trim(sPorts[i], " ")
+	}
+	if len(sPorts) != 2 || sPorts[0] == "" || sPorts[1] == "" {
+		return notAPort, notAPort, fmt.Errorf("appears to be a range but could not be parsed; `%s`", s)
+	}
+
+	rStartPort, err := strconv.Atoi(sPorts[0])
+	if err != nil {
+		return notAPort, notAPort, fmt.Errorf("beginning range was not a number; `%s`", sPorts[0])
+	}
+
+	rEndPort, err := strconv.Atoi(sPorts[1])
+	if err != nil {
+		return notAPort, notAPort, fmt.Errorf("ending range was not a number; `%s`", sPorts[1])
+	}
+
+	startPort := int32(rStartPort)
+	endPort := int32(rEndPort)
+
+	if startPort == firewall.PortAny {
+		endPort = firewall.PortAny
+	}
+
+	return startPort, endPort, nil
 }

firewall/packet.go

@@ -22,12 +22,19 @@ const (
 type Packet struct {
 	LocalAddr  netip.Addr
 	RemoteAddr netip.Addr
-	LocalPort  uint16
+	// LocalPort is the destination port for incoming traffic, or the source port for outgoing. Zero for ICMP.
+	LocalPort uint16
+	// RemotePort is the source port for incoming traffic, or the destination port for outgoing.
+	// For ICMP, it's the "identifier". This is only used for connection tracking, actual firewall rules will not filter on ICMP identifier
 	RemotePort uint16
 	Protocol   uint8
 	Fragment   bool
 }
 
+func (fp *Packet) IsIPv4() bool {
+	return fp.LocalAddr.Is4() && fp.RemoteAddr.Is4()
+}
+
 func (fp *Packet) Copy() *Packet {
 	return &Packet{
 		LocalAddr:  fp.LocalAddr,
@@ -46,6 +53,8 @@ func (fp Packet) MarshalJSON() ([]byte, error) {
 		proto = "tcp"
 	case ProtoICMP:
 		proto = "icmp"
+	case ProtoICMPv6:
+		proto = "icmpv6"
 	case ProtoUDP:
 		proto = "udp"
 	default:

firewall_test.go

@@ -87,9 +87,10 @@ func TestFirewall_AddRule(t *testing.T) {
 
 	fw = NewFirewall(l, time.Second, time.Minute, time.Hour, c)
 	require.NoError(t, fw.AddRule(true, firewall.ProtoICMP, 1, 1, []string{}, "h1", "", "", "", ""))
-	assert.Nil(t, fw.InRules.ICMP[1].Any.Any)
+	//no matter what port is given for icmp, it should end up as "any"
-	assert.Empty(t, fw.InRules.ICMP[1].Any.Groups)
+	assert.Nil(t, fw.InRules.ICMP[firewall.PortAny].Any.Any)
-	assert.Contains(t, fw.InRules.ICMP[1].Any.Hosts, "h1")
+	assert.Empty(t, fw.InRules.ICMP[firewall.PortAny].Any.Groups)
+	assert.Contains(t, fw.InRules.ICMP[firewall.PortAny].Any.Hosts, "h1")
 
 	fw = NewFirewall(l, time.Second, time.Minute, time.Hour, c)
 	require.NoError(t, fw.AddRule(false, firewall.ProtoAny, 1, 1, []string{}, "", ti.String(), "", "", ""))
@@ -212,44 +213,44 @@ func TestFirewall_Drop(t *testing.T) {
 	cp := cert.NewCAPool()
 
 	// Drop outbound
-	assert.Equal(t, ErrNoMatchingRule, fw.Drop(p, false, &h, cp, nil))
+	assert.Equal(t, ErrNoMatchingRule, fw.Drop(p, nil, false, &h, cp, nil))
 	// Allow inbound
 	resetConntrack(fw)
-	require.NoError(t, fw.Drop(p, true, &h, cp, nil))
+	require.NoError(t, fw.Drop(p, nil, true, &h, cp, nil))
 	// Allow outbound because conntrack
-	require.NoError(t, fw.Drop(p, false, &h, cp, nil))
+	require.NoError(t, fw.Drop(p, nil, false, &h, cp, nil))
 
 	// test remote mismatch
 	oldRemote := p.RemoteAddr
 	p.RemoteAddr = netip.MustParseAddr("1.2.3.10")
-	assert.Equal(t, fw.Drop(p, false, &h, cp, nil), ErrInvalidRemoteIP)
+	assert.Equal(t, fw.Drop(p, nil, false, &h, cp, nil), ErrInvalidRemoteIP)
 	p.RemoteAddr = oldRemote
 
 	// ensure signer doesn't get in the way of group checks
 	fw = NewFirewall(l, time.Second, time.Minute, time.Hour, &c)
 	require.NoError(t, fw.AddRule(true, firewall.ProtoAny, 0, 0, []string{"nope"}, "", "", "", "", "signer-shasum"))
 	require.NoError(t, fw.AddRule(true, firewall.ProtoAny, 0, 0, []string{"default-group"}, "", "", "", "", "signer-shasum-bad"))
-	assert.Equal(t, fw.Drop(p, true, &h, cp, nil), ErrNoMatchingRule)
+	assert.Equal(t, fw.Drop(p, nil, true, &h, cp, nil), ErrNoMatchingRule)
 
 	// test caSha doesn't drop on match
 	fw = NewFirewall(l, time.Second, time.Minute, time.Hour, &c)
 	require.NoError(t, fw.AddRule(true, firewall.ProtoAny, 0, 0, []string{"nope"}, "", "", "", "", "signer-shasum-bad"))
 	require.NoError(t, fw.AddRule(true, firewall.ProtoAny, 0, 0, []string{"default-group"}, "", "", "", "", "signer-shasum"))
-	require.NoError(t, fw.Drop(p, true, &h, cp, nil))
+	require.NoError(t, fw.Drop(p, nil, true, &h, cp, nil))
 
 	// ensure ca name doesn't get in the way of group checks
 	cp.CAs["signer-shasum"] = &cert.CachedCertificate{Certificate: &dummyCert{name: "ca-good"}}
 	fw = NewFirewall(l, time.Second, time.Minute, time.Hour, &c)
 	require.NoError(t, fw.AddRule(true, firewall.ProtoAny, 0, 0, []string{"nope"}, "", "", "", "ca-good", ""))
 	require.NoError(t, fw.AddRule(true, firewall.ProtoAny, 0, 0, []string{"default-group"}, "", "", "", "ca-good-bad", ""))
-	assert.Equal(t, fw.Drop(p, true, &h, cp, nil), ErrNoMatchingRule)
+	assert.Equal(t, fw.Drop(p, nil, true, &h, cp, nil), ErrNoMatchingRule)
 
 	// test caName doesn't drop on match
 	cp.CAs["signer-shasum"] = &cert.CachedCertificate{Certificate: &dummyCert{name: "ca-good"}}
 	fw = NewFirewall(l, time.Second, time.Minute, time.Hour, &c)
 	require.NoError(t, fw.AddRule(true, firewall.ProtoAny, 0, 0, []string{"nope"}, "", "", "", "ca-good-bad", ""))
 	require.NoError(t, fw.AddRule(true, firewall.ProtoAny, 0, 0, []string{"default-group"}, "", "", "", "ca-good", ""))
-	require.NoError(t, fw.Drop(p, true, &h, cp, nil))
+	require.NoError(t, fw.Drop(p, nil, true, &h, cp, nil))
 }
 
 func TestFirewall_DropV6(t *testing.T) {
@@ -291,44 +292,44 @@ func TestFirewall_DropV6(t *testing.T) {
 	cp := cert.NewCAPool()
 
 	// Drop outbound
-	assert.Equal(t, ErrNoMatchingRule, fw.Drop(p, false, &h, cp, nil))
+	assert.Equal(t, ErrNoMatchingRule, fw.Drop(p, nil, false, &h, cp, nil))
 	// Allow inbound
 	resetConntrack(fw)
-	require.NoError(t, fw.Drop(p, true, &h, cp, nil))
+	require.NoError(t, fw.Drop(p, nil, true, &h, cp, nil))
 	// Allow outbound because conntrack
-	require.NoError(t, fw.Drop(p, false, &h, cp, nil))
+	require.NoError(t, fw.Drop(p, nil, false, &h, cp, nil))
 
 	// test remote mismatch
 	oldRemote := p.RemoteAddr
 	p.RemoteAddr = netip.MustParseAddr("fd12::56")
-	assert.Equal(t, fw.Drop(p, false, &h, cp, nil), ErrInvalidRemoteIP)
+	assert.Equal(t, fw.Drop(p, nil, false, &h, cp, nil), ErrInvalidRemoteIP)
 	p.RemoteAddr = oldRemote
 
 	// ensure signer doesn't get in the way of group checks
 	fw = NewFirewall(l, time.Second, time.Minute, time.Hour, &c)
 	require.NoError(t, fw.AddRule(true, firewall.ProtoAny, 0, 0, []string{"nope"}, "", "", "", "", "signer-shasum"))
 	require.NoError(t, fw.AddRule(true, firewall.ProtoAny, 0, 0, []string{"default-group"}, "", "", "", "", "signer-shasum-bad"))
-	assert.Equal(t, fw.Drop(p, true, &h, cp, nil), ErrNoMatchingRule)
+	assert.Equal(t, fw.Drop(p, nil, true, &h, cp, nil), ErrNoMatchingRule)
 
 	// test caSha doesn't drop on match
 	fw = NewFirewall(l, time.Second, time.Minute, time.Hour, &c)
 	require.NoError(t, fw.AddRule(true, firewall.ProtoAny, 0, 0, []string{"nope"}, "", "", "", "", "signer-shasum-bad"))
 	require.NoError(t, fw.AddRule(true, firewall.ProtoAny, 0, 0, []string{"default-group"}, "", "", "", "", "signer-shasum"))
-	require.NoError(t, fw.Drop(p, true, &h, cp, nil))
+	require.NoError(t, fw.Drop(p, nil, true, &h, cp, nil))
 
 	// ensure ca name doesn't get in the way of group checks
 	cp.CAs["signer-shasum"] = &cert.CachedCertificate{Certificate: &dummyCert{name: "ca-good"}}
 	fw = NewFirewall(l, time.Second, time.Minute, time.Hour, &c)
 	require.NoError(t, fw.AddRule(true, firewall.ProtoAny, 0, 0, []string{"nope"}, "", "", "", "ca-good", ""))
 	require.NoError(t, fw.AddRule(true, firewall.ProtoAny, 0, 0, []string{"default-group"}, "", "", "", "ca-good-bad", ""))
-	assert.Equal(t, fw.Drop(p, true, &h, cp, nil), ErrNoMatchingRule)
+	assert.Equal(t, fw.Drop(p, nil, true, &h, cp, nil), ErrNoMatchingRule)
 
 	// test caName doesn't drop on match
 	cp.CAs["signer-shasum"] = &cert.CachedCertificate{Certificate: &dummyCert{name: "ca-good"}}
 	fw = NewFirewall(l, time.Second, time.Minute, time.Hour, &c)
 	require.NoError(t, fw.AddRule(true, firewall.ProtoAny, 0, 0, []string{"nope"}, "", "", "", "ca-good-bad", ""))
 	require.NoError(t, fw.AddRule(true, firewall.ProtoAny, 0, 0, []string{"default-group"}, "", "", "", "ca-good", ""))
-	require.NoError(t, fw.Drop(p, true, &h, cp, nil))
+	require.NoError(t, fw.Drop(p, nil, true, &h, cp, nil))
 }
 
 func BenchmarkFirewallTable_match(b *testing.B) {
@@ -536,10 +537,10 @@ func TestFirewall_Drop2(t *testing.T) {
 	cp := cert.NewCAPool()
 
 	// h1/c1 lacks the proper groups
-	require.ErrorIs(t, fw.Drop(p, true, &h1, cp, nil), ErrNoMatchingRule)
+	require.ErrorIs(t, fw.Drop(p, nil, true, &h1, cp, nil), ErrNoMatchingRule)
 	// c has the proper groups
 	resetConntrack(fw)
-	require.NoError(t, fw.Drop(p, true, &h, cp, nil))
+	require.NoError(t, fw.Drop(p, nil, true, &h, cp, nil))
 }
 
 func TestFirewall_Drop3(t *testing.T) {
@@ -617,18 +618,18 @@ func TestFirewall_Drop3(t *testing.T) {
 	cp := cert.NewCAPool()
 
 	// c1 should pass because host match
-	require.NoError(t, fw.Drop(p, true, &h1, cp, nil))
+	require.NoError(t, fw.Drop(p, nil, true, &h1, cp, nil))
 	// c2 should pass because ca sha match
 	resetConntrack(fw)
-	require.NoError(t, fw.Drop(p, true, &h2, cp, nil))
+	require.NoError(t, fw.Drop(p, nil, true, &h2, cp, nil))
 	// c3 should fail because no match
 	resetConntrack(fw)
-	assert.Equal(t, fw.Drop(p, true, &h3, cp, nil), ErrNoMatchingRule)
+	assert.Equal(t, fw.Drop(p, nil, true, &h3, cp, nil), ErrNoMatchingRule)
 
 	// Test a remote address match
 	fw = NewFirewall(l, time.Second, time.Minute, time.Hour, c.Certificate)
 	require.NoError(t, fw.AddRule(true, firewall.ProtoAny, 1, 1, []string{}, "", "1.2.3.4/24", "", "", ""))
-	require.NoError(t, fw.Drop(p, true, &h1, cp, nil))
+	require.NoError(t, fw.Drop(p, nil, true, &h1, cp, nil))
 }
 
 func TestFirewall_Drop3V6(t *testing.T) {
@@ -666,7 +667,7 @@ func TestFirewall_Drop3V6(t *testing.T) {
 	fw := NewFirewall(l, time.Second, time.Minute, time.Hour, c.Certificate)
 	cp := cert.NewCAPool()
 	require.NoError(t, fw.AddRule(true, firewall.ProtoAny, 1, 1, []string{}, "", "fd12::34/120", "", "", ""))
-	require.NoError(t, fw.Drop(p, true, &h, cp, nil))
+	require.NoError(t, fw.Drop(p, nil, true, &h, cp, nil))
 }
 
 func TestFirewall_DropConntrackReload(t *testing.T) {
@@ -708,12 +709,12 @@ func TestFirewall_DropConntrackReload(t *testing.T) {
 	cp := cert.NewCAPool()
 
 	// Drop outbound
-	assert.Equal(t, fw.Drop(p, false, &h, cp, nil), ErrNoMatchingRule)
+	assert.Equal(t, fw.Drop(p, nil, false, &h, cp, nil), ErrNoMatchingRule)
 	// Allow inbound
 	resetConntrack(fw)
-	require.NoError(t, fw.Drop(p, true, &h, cp, nil))
+	require.NoError(t, fw.Drop(p, nil, true, &h, cp, nil))
 	// Allow outbound because conntrack
-	require.NoError(t, fw.Drop(p, false, &h, cp, nil))
+	require.NoError(t, fw.Drop(p, nil, false, &h, cp, nil))
 
 	oldFw := fw
 	fw = NewFirewall(l, time.Second, time.Minute, time.Hour, c.Certificate)
@@ -722,7 +723,7 @@ func TestFirewall_DropConntrackReload(t *testing.T) {
 	fw.rulesVersion = oldFw.rulesVersion + 1
 
 	// Allow outbound because conntrack and new rules allow port 10
-	require.NoError(t, fw.Drop(p, false, &h, cp, nil))
+	require.NoError(t, fw.Drop(p, nil, false, &h, cp, nil))
 
 	oldFw = fw
 	fw = NewFirewall(l, time.Second, time.Minute, time.Hour, c.Certificate)
@@ -731,7 +732,151 @@ func TestFirewall_DropConntrackReload(t *testing.T) {
 	fw.rulesVersion = oldFw.rulesVersion + 1
 
 	// Drop outbound because conntrack doesn't match new ruleset
-	assert.Equal(t, fw.Drop(p, false, &h, cp, nil), ErrNoMatchingRule)
+	assert.Equal(t, fw.Drop(p, nil, false, &h, cp, nil), ErrNoMatchingRule)
+}
+
+func TestFirewall_ICMPPortBehavior(t *testing.T) {
+	l := test.NewLogger()
+	ob := &bytes.Buffer{}
+	l.SetOutput(ob)
+	myVpnNetworksTable := new(bart.Lite)
+	myVpnNetworksTable.Insert(netip.MustParsePrefix("1.1.1.1/8"))
+
+	network := netip.MustParsePrefix("1.2.3.4/24")
+
+	c := cert.CachedCertificate{
+		Certificate: &dummyCert{
+			name:     "host1",
+			networks: []netip.Prefix{network},
+			groups:   []string{"default-group"},
+			issuer:   "signer-shasum",
+		},
+		InvertedGroups: map[string]struct{}{"default-group": {}},
+	}
+	h := HostInfo{
+		ConnectionState: &ConnectionState{
+			peerCert: &c,
+		},
+		vpnAddrs: []netip.Addr{network.Addr()},
+	}
+	h.buildNetworks(myVpnNetworksTable, c.Certificate)
+
+	cp := cert.NewCAPool()
+
+	templ := firewall.Packet{
+		LocalAddr:  netip.MustParseAddr("1.2.3.4"),
+		RemoteAddr: netip.MustParseAddr("1.2.3.4"),
+		Protocol:   firewall.ProtoICMP,
+		Fragment:   false,
+	}
+
+	t.Run("ICMP allowed", func(t *testing.T) {
+		fw := NewFirewall(l, time.Second, time.Minute, time.Hour, c.Certificate)
+		require.NoError(t, fw.AddRule(true, firewall.ProtoICMP, 0, 0, []string{"any"}, "", "", "", "", ""))
+		t.Run("zero ports", func(t *testing.T) {
+			p := templ.Copy()
+			p.LocalPort = 0
+			p.RemotePort = 0
+			// Drop outbound
+			assert.Equal(t, fw.Drop(*p, nil, false, &h, cp, nil), ErrNoMatchingRule)
+			// Allow inbound
+			resetConntrack(fw)
+			require.NoError(t, fw.Drop(*p, nil, true, &h, cp, nil))
+			//now also allow outbound
+			require.NoError(t, fw.Drop(*p, nil, false, &h, cp, nil))
+		})
+
+		t.Run("nonzero ports", func(t *testing.T) {
+			p := templ.Copy()
+			p.LocalPort = 0xabcd
+			p.RemotePort = 0x1234
+			// Drop outbound
+			assert.Equal(t, fw.Drop(*p, nil, false, &h, cp, nil), ErrNoMatchingRule)
+			// Allow inbound
+			resetConntrack(fw)
+			require.NoError(t, fw.Drop(*p, nil, true, &h, cp, nil))
+			//now also allow outbound
+			require.NoError(t, fw.Drop(*p, nil, false, &h, cp, nil))
+		})
+	})
+
+	t.Run("Any proto, some ports allowed", func(t *testing.T) {
+		fw := NewFirewall(l, time.Second, time.Minute, time.Hour, c.Certificate)
+		require.NoError(t, fw.AddRule(true, firewall.ProtoAny, 80, 444, []string{"any"}, "", "", "", "", ""))
+		t.Run("zero ports, still blocked", func(t *testing.T) {
+			p := templ.Copy()
+			p.LocalPort = 0
+			p.RemotePort = 0
+			// Drop outbound
+			assert.Equal(t, fw.Drop(*p, nil, false, &h, cp, nil), ErrNoMatchingRule)
+			// Allow inbound
+			resetConntrack(fw)
+			assert.Equal(t, fw.Drop(*p, nil, true, &h, cp, nil), ErrNoMatchingRule)
+			//now also allow outbound
+			assert.Equal(t, fw.Drop(*p, nil, false, &h, cp, nil), ErrNoMatchingRule)
+		})
+
+		t.Run("nonzero ports, still blocked", func(t *testing.T) {
+			p := templ.Copy()
+			p.LocalPort = 0xabcd
+			p.RemotePort = 0x1234
+			// Drop outbound
+			assert.Equal(t, fw.Drop(*p, nil, false, &h, cp, nil), ErrNoMatchingRule)
+			// Allow inbound
+			resetConntrack(fw)
+			assert.Equal(t, fw.Drop(*p, nil, true, &h, cp, nil), ErrNoMatchingRule)
+			//now also allow outbound
+			assert.Equal(t, fw.Drop(*p, nil, false, &h, cp, nil), ErrNoMatchingRule)
+		})
+
+		t.Run("nonzero, matching ports, still blocked", func(t *testing.T) {
+			p := templ.Copy()
+			p.LocalPort = 80
+			p.RemotePort = 80
+			// Drop outbound
+			assert.Equal(t, fw.Drop(*p, nil, false, &h, cp, nil), ErrNoMatchingRule)
+			// Allow inbound
+			resetConntrack(fw)
+			assert.Equal(t, fw.Drop(*p, nil, true, &h, cp, nil), ErrNoMatchingRule)
+			//now also allow outbound
+			assert.Equal(t, fw.Drop(*p, nil, false, &h, cp, nil), ErrNoMatchingRule)
+		})
+	})
+	t.Run("Any proto, any port", func(t *testing.T) {
+		fw := NewFirewall(l, time.Second, time.Minute, time.Hour, c.Certificate)
+		require.NoError(t, fw.AddRule(true, firewall.ProtoAny, 0, 0, []string{"any"}, "", "", "", "", ""))
+		t.Run("zero ports, allowed", func(t *testing.T) {
+			resetConntrack(fw)
+			p := templ.Copy()
+			p.LocalPort = 0
+			p.RemotePort = 0
+			// Drop outbound
+			assert.Equal(t, fw.Drop(*p, nil, false, &h, cp, nil), ErrNoMatchingRule)
+			// Allow inbound
+			resetConntrack(fw)
+			require.NoError(t, fw.Drop(*p, nil, true, &h, cp, nil))
+			//now also allow outbound
+			require.NoError(t, fw.Drop(*p, nil, false, &h, cp, nil))
+		})
+
+		t.Run("nonzero ports, allowed", func(t *testing.T) {
+			resetConntrack(fw)
+			p := templ.Copy()
+			p.LocalPort = 0xabcd
+			p.RemotePort = 0x1234
+			// Drop outbound
+			assert.Equal(t, fw.Drop(*p, nil, false, &h, cp, nil), ErrNoMatchingRule)
+			// Allow inbound
+			resetConntrack(fw)
+			require.NoError(t, fw.Drop(*p, nil, true, &h, cp, nil))
+			//now also allow outbound
+			require.NoError(t, fw.Drop(*p, nil, false, &h, cp, nil))
+			//different ID is blocked
+			p.RemotePort++
+			require.Equal(t, fw.Drop(*p, nil, false, &h, cp, nil), ErrNoMatchingRule)
+		})
+	})
+
 }
 
 func TestFirewall_DropIPSpoofing(t *testing.T) {
@@ -777,7 +922,7 @@ func TestFirewall_DropIPSpoofing(t *testing.T) {
 		Protocol:   firewall.ProtoUDP,
 		Fragment:   false,
 	}
-	assert.Equal(t, fw.Drop(p, true, &h1, cp, nil), ErrInvalidRemoteIP)
+	assert.Equal(t, fw.Drop(p, nil, true, &h1, cp, nil), ErrInvalidRemoteIP)
 }
 
 func BenchmarkLookup(b *testing.B) {
@@ -897,7 +1042,7 @@ func TestNewFirewallFromConfig(t *testing.T) {
 	l := test.NewLogger()
 	// Test a bad rule definition
 	c := &dummyCert{}
-	cs, err := newCertState(cert.Version2, nil, c, false, cert.Curve_CURVE25519, nil)
+	cs, err := newCertState(l, cert.Version2, nil, c, false, cert.Curve_CURVE25519, nil)
 	require.NoError(t, err)
 
 	conf := config.NewC(l)
@@ -919,11 +1064,11 @@ func TestNewFirewallFromConfig(t *testing.T) {
 
 	// Test code/port error
 	conf = config.NewC(l)
-	conf.Settings["firewall"] = map[string]any{"outbound": []any{map[string]any{"code": "a", "host": "testh"}}}
+	conf.Settings["firewall"] = map[string]any{"outbound": []any{map[string]any{"code": "a", "host": "testh", "proto": "any"}}}
 	_, err = NewFirewallFromConfig(l, cs, conf)
 	require.EqualError(t, err, "firewall.outbound rule #0; code was not a number; `a`")
 
-	conf.Settings["firewall"] = map[string]any{"outbound": []any{map[string]any{"port": "a", "host": "testh"}}}
+	conf.Settings["firewall"] = map[string]any{"outbound": []any{map[string]any{"port": "a", "host": "testh", "proto": "any"}}}
 	_, err = NewFirewallFromConfig(l, cs, conf)
 	require.EqualError(t, err, "firewall.outbound rule #0; port was not a number; `a`")
 
@@ -973,7 +1118,14 @@ func TestAddFirewallRulesFromConfig(t *testing.T) {
 	mf = &mockFirewall{}
 	conf.Settings["firewall"] = map[string]any{"outbound": []any{map[string]any{"port": "1", "proto": "icmp", "host": "a"}}}
 	require.NoError(t, AddFirewallRulesFromConfig(l, false, conf, mf))
-	assert.Equal(t, addRuleCall{incoming: false, proto: firewall.ProtoICMP, startPort: 1, endPort: 1, groups: nil, host: "a", ip: "", localIp: ""}, mf.lastCall)
+	assert.Equal(t, addRuleCall{incoming: false, proto: firewall.ProtoICMP, startPort: firewall.PortAny, endPort: firewall.PortAny, groups: nil, host: "a", ip: "", localIp: ""}, mf.lastCall)
+
+	// Test adding icmp rule no port
+	conf = config.NewC(l)
+	mf = &mockFirewall{}
+	conf.Settings["firewall"] = map[string]any{"outbound": []any{map[string]any{"proto": "icmp", "host": "a"}}}
+	require.NoError(t, AddFirewallRulesFromConfig(l, false, conf, mf))
+	assert.Equal(t, addRuleCall{incoming: false, proto: firewall.ProtoICMP, startPort: firewall.PortAny, endPort: firewall.PortAny, groups: nil, host: "a", ip: "", localIp: ""}, mf.lastCall)
 
 	// Test adding any rule
 	conf = config.NewC(l)
@@ -1184,7 +1336,7 @@ func (c *testcase) Test(t *testing.T, fw *Firewall) {
 	t.Helper()
 	cp := cert.NewCAPool()
 	resetConntrack(fw)
-	err := fw.Drop(c.p, true, c.h, cp, nil)
+	err := fw.Drop(c.p, nil, true, c.h, cp, nil)
 	if c.err == nil {
 		require.NoError(t, err, "failed to not drop remote address %s", c.p.RemoteAddr)
 	} else {
@@ -1192,7 +1344,7 @@ func (c *testcase) Test(t *testing.T, fw *Firewall) {
 	}
 }
 
-func buildTestCase(setup testsetup, err error, theirPrefixes ...netip.Prefix) testcase {
+func buildHostinfo(setup testsetup, theirPrefixes ...netip.Prefix) *HostInfo {
 	c1 := dummyCert{
 		name:     "host1",
 		networks: theirPrefixes,
@@ -1212,6 +1364,11 @@ func buildTestCase(setup testsetup, err error, theirPrefixes ...netip.Prefix) te
 		h.vpnAddrs[i] = theirPrefixes[i].Addr()
 	}
 	h.buildNetworks(setup.myVpnNetworksTable, &c1)
+	return &h
+}
+
+func buildTestCase(setup testsetup, err error, theirPrefixes ...netip.Prefix) testcase {
+	h := buildHostinfo(setup, theirPrefixes...)
 	p := firewall.Packet{
 		LocalAddr:  setup.c.Networks()[0].Addr(), //todo?
 		RemoteAddr: theirPrefixes[0].Addr(),
@@ -1221,9 +1378,9 @@ func buildTestCase(setup testsetup, err error, theirPrefixes ...netip.Prefix) te
 		Fragment:   false,
 	}
 	return testcase{
-		h:   &h,
+		h:   h,
 		p:   p,
-		c:   &c1,
+		c:   h.ConnectionState.peerCert.Certificate,
 		err: err,
 	}
 }
@@ -1245,6 +1402,19 @@ func newSetup(t *testing.T, l *logrus.Logger, myPrefixes ...netip.Prefix) testse
 	return newSetupFromCert(t, l, c)
 }
 
+func newSnatSetup(t *testing.T, l *logrus.Logger, myPrefix netip.Prefix, snatAddr netip.Addr) testsetup {
+	c := dummyCert{
+		name:     "me",
+		networks: []netip.Prefix{myPrefix},
+		groups:   []string{"default-group"},
+		issuer:   "signer-shasum",
+	}
+
+	out := newSetupFromCert(t, l, c)
+	out.fw.snatAddr = snatAddr
+	return out
+}
+
 func newSetupFromCert(t *testing.T, l *logrus.Logger, c dummyCert) testsetup {
 	myVpnNetworksTable := new(bart.Lite)
 	for _, prefix := range c.Networks() {
@@ -1380,3 +1550,59 @@ func resetConntrack(fw *Firewall) {
 	fw.Conntrack.Conns = map[firewall.Packet]*conn{}
 	fw.Conntrack.Unlock()
 }
+
+func TestFirewall_SNAT(t *testing.T) {
+	t.Parallel()
+	l := test.NewLogger()
+	ob := &bytes.Buffer{}
+	l.SetOutput(ob)
+	cp := cert.NewCAPool()
+	myPrefix := netip.MustParsePrefix("1.1.1.1/8")
+
+	MyCert := dummyCert{
+		name:     "me",
+		networks: []netip.Prefix{myPrefix},
+		groups:   []string{"default-group"},
+		issuer:   "signer-shasum",
+	}
+
+	theirPrefix := netip.MustParsePrefix("1.2.2.2/8")
+	snatAddr := netip.MustParseAddr("169.254.55.96")
+	t.Run("allow inbound all matching", func(t *testing.T) {
+		t.Parallel()
+		myCert := MyCert.Copy()
+		setup := newSnatSetup(t, l, myPrefix, snatAddr)
+		fw := NewFirewall(l, time.Second, time.Minute, time.Hour, myCert)
+		require.NoError(t, fw.AddRule(true, firewall.ProtoAny, 0, 0, []string{"any"}, "", "", "", "", ""))
+		resetConntrack(setup.fw)
+		h := buildHostinfo(setup, theirPrefix)
+		p := firewall.Packet{
+			LocalAddr:  setup.c.Networks()[0].Addr(), //todo?
+			RemoteAddr: h.vpnAddrs[0],
+			LocalPort:  10,
+			RemotePort: 90,
+			Protocol:   firewall.ProtoUDP,
+			Fragment:   false,
+		}
+		require.NoError(t, setup.fw.Drop(p, nil, true, h, cp, nil))
+	})
+	//t.Run("allow inbound unsafe route", func(t *testing.T) {
+	//	t.Parallel()
+	//	unsafePrefix := netip.MustParsePrefix("192.168.0.0/24")
+	//	c := dummyCert{
+	//		name:           "me",
+	//		networks:       []netip.Prefix{myPrefix},
+	//		unsafeNetworks: []netip.Prefix{unsafePrefix},
+	//		groups:         []string{"default-group"},
+	//		issuer:         "signer-shasum",
+	//	}
+	//	unsafeSetup := newSetupFromCert(t, l, c)
+	//	tc := buildTestCase(unsafeSetup, nil, twoPrefixes...)
+	//	tc.p.LocalAddr = netip.MustParseAddr("192.168.0.3")
+	//	tc.err = ErrNoMatchingRule
+	//	tc.Test(t, unsafeSetup.fw) //should hit firewall and bounce off
+	//	require.NoError(t, unsafeSetup.fw.AddRule(true, firewall.ProtoAny, 0, 0, []string{"any"}, "", "", unsafePrefix.String(), "", ""))
+	//	tc.err = nil
+	//	tc.Test(t, unsafeSetup.fw) //should pass
+	//})
+}

go.mod

@@ -4,6 +4,7 @@ go 1.25
 
 require (
 	dario.cat/mergo v1.0.2
+	filippo.io/bigmod v0.1.0
 	github.com/anmitsu/go-shlex v0.0.0-20200514113438-38f4b401e2be
 	github.com/armon/go-radix v1.0.0
 	github.com/cyberdelia/go-metrics-graphite v0.0.0-20161219230853-39f87cc3b432
@@ -30,11 +31,11 @@ require (
 	golang.org/x/sys v0.40.0
 	golang.org/x/term v0.39.0
 	golang.zx2c4.com/wintun v0.0.0-20230126152724-0fa3db229ce2
-	golang.zx2c4.com/wireguard v0.0.0-20250521234502-f333402bd9cb
+	golang.zx2c4.com/wireguard v0.0.0-20230325221338-052af4a8072b
 	golang.zx2c4.com/wireguard/windows v0.5.3
 	google.golang.org/protobuf v1.36.11
 	gopkg.in/yaml.v3 v3.0.1
-	gvisor.dev/gvisor v0.0.0-20250503011706-39ed1f5ac29c
+	gvisor.dev/gvisor v0.0.0-20240423190808-9d7a357edefe
 )
 
 require (
@@ -50,6 +51,6 @@ require (
 	github.com/vishvananda/netns v0.0.5 // indirect
 	go.yaml.in/yaml/v2 v2.4.2 // indirect
 	golang.org/x/mod v0.31.0 // indirect
-	golang.org/x/time v0.7.0 // indirect
+	golang.org/x/time v0.5.0 // indirect
 	golang.org/x/tools v0.40.0 // indirect
 )

go.sum

@@ -1,6 +1,8 @@
 cloud.google.com/go v0.34.0/go.mod h1:aQUYkXzVsufM+DwF1aE+0xfcU+56JwCaLick0ClmMTw=
 dario.cat/mergo v1.0.2 h1:85+piFYR1tMbRrLcDwR18y4UKJ3aH1Tbzi24VRW1TK8=
 dario.cat/mergo v1.0.2/go.mod h1:E/hbnu0NxMFBjpMIE34DRGLWqDy0g5FuKDhCb31ngxA=
+filippo.io/bigmod v0.1.0 h1:UNzDk7y9ADKST+axd9skUpBQeW7fG2KrTZyOE4uGQy8=
+filippo.io/bigmod v0.1.0/go.mod h1:OjOXDNlClLblvXdwgFFOQFJEocLhhtai8vGLy0JCZlI=
 github.com/alecthomas/template v0.0.0-20160405071501-a0175ee3bccc/go.mod h1:LOuyumcjzFXgccqObfd/Ljyb9UuFJ6TxHnclSeseNhc=
 github.com/alecthomas/template v0.0.0-20190718012654-fb15b899a751/go.mod h1:LOuyumcjzFXgccqObfd/Ljyb9UuFJ6TxHnclSeseNhc=
 github.com/alecthomas/units v0.0.0-20151022065526-2efee857e7cf/go.mod h1:ybxpYRFXyAe+OPACYpWeL0wqObRcbAqCMya13uyzqw0=
@@ -216,8 +218,8 @@ golang.org/x/term v0.39.0/go.mod h1:yxzUCTP/U+FzoxfdKmLaA0RV1WgE0VY7hXBwKtY/4ww=
 golang.org/x/text v0.3.0/go.mod h1:NqM8EUOU14njkJ3fqMW+pc6Ldnwhi/IjpwHt7yyuwOQ=
 golang.org/x/text v0.3.2/go.mod h1:bEr9sfX3Q8Zfm5fL9x+3itogRgK3+ptLWKqgva+5dAk=
 golang.org/x/text v0.3.3/go.mod h1:5Zoc/QRtKVWzQhOtBMvqHzDpF6irO9z98xDceosuGiQ=
-golang.org/x/time v0.7.0 h1:ntUhktv3OPE6TgYxXWv9vKvUSJyIFJlyohwbkEwPrKQ=
+golang.org/x/time v0.5.0 h1:o7cqy6amK/52YcAKIPlM3a+Fpj35zvRj2TP+e1xFSfk=
-golang.org/x/time v0.7.0/go.mod h1:3BpzKBy/shNhVucY/MWOyx10tF3SFh9QdLuxbVysPQM=
+golang.org/x/time v0.5.0/go.mod h1:3BpzKBy/shNhVucY/MWOyx10tF3SFh9QdLuxbVysPQM=
 golang.org/x/tools v0.0.0-20180917221912-90fa682c2a6e/go.mod h1:n7NCudcB/nEzxVGmLbDWY5pfWTLqBcC2KZ6jyYvM4mQ=
 golang.org/x/tools v0.0.0-20191119224855-298f0cb1881e/go.mod h1:b+2E5dAYhXwXZwtnZ6UAqBI28+e2cm9otk0dWdXHAEo=
 golang.org/x/tools v0.0.0-20200130002326-2f3ba24bd6e7/go.mod h1:TB2adYChydJhpapKDTa4BR/hXlZSLoq2Wpct/0txZ28=
@@ -231,8 +233,8 @@ golang.org/x/xerrors v0.0.0-20191204190536-9bdfabe68543/go.mod h1:I/5z698sn9Ka8T
 golang.org/x/xerrors v0.0.0-20200804184101-5ec99f83aff1/go.mod h1:I/5z698sn9Ka8TeJc9MKroUUfqBBauWjQqLJ2OPfmY0=
 golang.zx2c4.com/wintun v0.0.0-20230126152724-0fa3db229ce2 h1:B82qJJgjvYKsXS9jeunTOisW56dUokqW/FOteYJJ/yg=
 golang.zx2c4.com/wintun v0.0.0-20230126152724-0fa3db229ce2/go.mod h1:deeaetjYA+DHMHg+sMSMI58GrEteJUUzzw7en6TJQcI=
-golang.zx2c4.com/wireguard v0.0.0-20250521234502-f333402bd9cb h1:whnFRlWMcXI9d+ZbWg+4sHnLp52d5yiIPUxMBSt4X9A=
+golang.zx2c4.com/wireguard v0.0.0-20230325221338-052af4a8072b h1:J1CaxgLerRR5lgx3wnr6L04cJFbWoceSK9JWBdglINo=
-golang.zx2c4.com/wireguard v0.0.0-20250521234502-f333402bd9cb/go.mod h1:rpwXGsirqLqN2L0JDJQlwOboGHmptD5ZD6T2VmcqhTw=
+golang.zx2c4.com/wireguard v0.0.0-20230325221338-052af4a8072b/go.mod h1:tqur9LnfstdR9ep2LaJT4lFUl0EjlHtge+gAjmsHUG4=
 golang.zx2c4.com/wireguard/windows v0.5.3 h1:On6j2Rpn3OEMXqBq00QEDC7bWSZrPIHKIus8eIuExIE=
 golang.zx2c4.com/wireguard/windows v0.5.3/go.mod h1:9TEe8TJmtwyQebdFwAkEWOPr3prrtqm+REGFifP60hI=
 google.golang.org/appengine v1.4.0/go.mod h1:xpcJRLb0r/rnEns0DIKYYv+WjYCduHsrkT7/EB5XEv4=
@@ -258,5 +260,5 @@ gopkg.in/yaml.v2 v2.3.0/go.mod h1:hI93XBmqTisBFMUTm0b8Fm+jr3Dg1NNxqwp+5A1VGuI=
 gopkg.in/yaml.v3 v3.0.0-20200313102051-9f266ea9e77c/go.mod h1:K4uyk7z7BCEPqu6E+C64Yfv1cQ7kz7rIZviUmN+EgEM=
 gopkg.in/yaml.v3 v3.0.1 h1:fxVm/GzAzEWqLHuvctI91KS9hhNmmWOoWu0XTYJS7CA=
 gopkg.in/yaml.v3 v3.0.1/go.mod h1:K4uyk7z7BCEPqu6E+C64Yfv1cQ7kz7rIZviUmN+EgEM=
-gvisor.dev/gvisor v0.0.0-20250503011706-39ed1f5ac29c h1:m/r7OM+Y2Ty1sgBQ7Qb27VgIMBW8ZZhT4gLnUyDIhzI=
+gvisor.dev/gvisor v0.0.0-20240423190808-9d7a357edefe h1:fre4i6mv4iBuz5lCMOzHD1rH1ljqHWSICFmZRbbgp3g=
-gvisor.dev/gvisor v0.0.0-20250503011706-39ed1f5ac29c/go.mod h1:3r5CMtNQMKIvBlrmM9xWUNamjKBYPOWyXOjmg5Kts3g=
+gvisor.dev/gvisor v0.0.0-20240423190808-9d7a357edefe/go.mod h1:sxc3Uvk/vHcd3tj7/DHVBoR5wvWT/MmRq2pj7HRJnwU=

handshake_manager.go

@@ -590,7 +590,7 @@ func (hm *HandshakeManager) allocateIndex(hh *HandshakeHostInfo) error {
 	hm.Lock()
 	defer hm.Unlock()
 
-	for i := 0; i < 32; i++ {
+	for range 32 {
 		index, err := generateIndex(hm.l)
 		if err != nil {
 			return err

hostmap.go

@@ -224,6 +224,9 @@ const (
 	NetworkTypeVPNPeer
 	// NetworkTypeUnsafe is a network from Certificate.UnsafeNetworks()
 	NetworkTypeUnsafe
+	// NetworkTypeUnverifiedSNATPeer is used to indicate traffic we're willing to route, but never deliver to a NetworkTypeVPN
+	NetworkTypeUnverifiedSNATPeer
+	NetworkTypeInvalidPeer
 )
 
 type HostInfo struct {
@@ -277,6 +280,15 @@ type HostInfo struct {
 	lastUsed time.Time
 }
 
+func (i *HostInfo) HasOnlyV6Addresses() bool {
+	for _, vpnIp := range i.vpnAddrs {
+		if !vpnIp.Is6() {
+			return false
+		}
+	}
+	return true
+}
+
 type ViaSender struct {
 	UdpAddr   netip.AddrPort
 	relayHI   *HostInfo // relayHI is the host info object of the relay

hostmap_tester.go

@@ -1,5 +1,4 @@
 //go:build e2e_testing
-// +build e2e_testing
 
 package nebula
 

inside.go

@@ -2,8 +2,6 @@ package nebula
 
 import (
 	"net/netip"
-	"sync"
-	"time"
 
 	"github.com/sirupsen/logrus"
 	"github.com/slackhq/nebula/firewall"
@@ -13,258 +11,6 @@ import (
 	"github.com/slackhq/nebula/routing"
 )
 
-// preEncryptionPacket holds packet data before batch encryption
-type preEncryptionPacket struct {
-	hostinfo *HostInfo
-	ci       *ConnectionState
-	packet   []byte
-	out      []byte
-}
-
-// Pool for preEncryptionBatch slices to reduce allocations
-var preEncryptionBatchPool = sync.Pool{
-	New: func() any {
-		// Pre-allocate with reasonable capacity
-		batch := make([]preEncryptionPacket, 0, 128)
-		return &batch
-	},
-}
-
-// consumeInsidePackets processes multiple packets in a batch for improved performance
-// packets: slice of packet buffers to process
-// sizes: slice of packet sizes
-// count: number of packets to process
-// outs: slice of output buffers (one per packet) with virtio headroom
-// q: queue index
-// localCache: firewall conntrack cache
-// batchPackets: pre-allocated slice for accumulating encrypted packets
-// batchAddrs: pre-allocated slice for accumulating destination addresses
-func (f *Interface) consumeInsidePackets(packets [][]byte, sizes []int, count int, outs [][]byte, nb []byte, q int, localCache firewall.ConntrackCache, batchPackets *[][]byte, batchAddrs *[]netip.AddrPort) {
-	// Reusable per-packet state
-	fwPacket := &firewall.Packet{}
-
-	// Reset batch accumulation slices (reuse capacity)
-	*batchPackets = (*batchPackets)[:0]
-	*batchAddrs = (*batchAddrs)[:0]
-
-	// Collect packets for batched encryption
-	preEncryptionBatch := make([]preEncryptionPacket, 0, count)
-
-	// Process each packet in the batch
-	for i := 0; i < count; i++ {
-		packet := packets[i][:sizes[i]]
-		out := outs[i]
-
-		// Inline the consumeInsidePacket logic for better performance
-		err := newPacket(packet, false, fwPacket)
-		if err != nil {
-			if f.l.Level >= logrus.DebugLevel {
-				f.l.WithField("packet", packet).Debugf("Error while validating outbound packet: %s", err)
-			}
-			continue
-		}
-
-		// Ignore local broadcast packets
-		if f.dropLocalBroadcast {
-			if f.myBroadcastAddrsTable.Contains(fwPacket.RemoteAddr) {
-				continue
-			}
-		}
-
-		if f.myVpnAddrsTable.Contains(fwPacket.RemoteAddr) {
-			// Immediately forward packets from self to self.
-			if immediatelyForwardToSelf {
-				_, err := f.readers[q].Write(packet)
-				if err != nil {
-					f.l.WithError(err).Error("Failed to forward to tun")
-				}
-			}
-			continue
-		}
-
-		// Ignore multicast packets
-		if f.dropMulticast && fwPacket.RemoteAddr.IsMulticast() {
-			continue
-		}
-
-		hostinfo, ready := f.getOrHandshakeConsiderRouting(fwPacket, func(hh *HandshakeHostInfo) {
-			hh.cachePacket(f.l, header.Message, 0, packet, f.sendMessageNow, f.cachedPacketMetrics)
-		})
-
-		if hostinfo == nil {
-			f.rejectInside(packet, out, q)
-			if f.l.Level >= logrus.DebugLevel {
-				f.l.WithField("vpnAddr", fwPacket.RemoteAddr).
-					WithField("fwPacket", fwPacket).
-					Debugln("dropping outbound packet, vpnAddr not in our vpn networks or in unsafe networks")
-			}
-			continue
-		}
-
-		if !ready {
-			continue
-		}
-
-		dropReason := f.firewall.Drop(*fwPacket, false, hostinfo, f.pki.GetCAPool(), localCache)
-		if dropReason != nil {
-			f.rejectInside(packet, out, q)
-			if f.l.Level >= logrus.DebugLevel {
-				hostinfo.logger(f.l).
-					WithField("fwPacket", fwPacket).
-					WithField("reason", dropReason).
-					Debugln("dropping outbound packet")
-			}
-			continue
-		}
-
-		// Prepare packet for batch encryption
-		ci := hostinfo.ConnectionState
-		if ci.eKey == nil {
-			continue
-		}
-
-		// Check if this needs relay - if so, send immediately and skip batching
-		useRelay := !hostinfo.remote.IsValid()
-		if useRelay {
-			// Handle relay sends individually (less common path)
-			f.sendNoMetrics(header.Message, 0, ci, hostinfo, netip.AddrPort{}, packet, nb, out, q)
-			continue
-		}
-
-		// Collect for batched encryption
-		preEncryptionBatch = append(preEncryptionBatch, preEncryptionPacket{
-			hostinfo: hostinfo,
-			ci:       ci,
-			packet:   packet,
-			out:      out,
-		})
-	}
-
-	// BATCH ENCRYPTION: Process all collected packets
-	if len(preEncryptionBatch) > 0 {
-		f.encryptBatch(preEncryptionBatch, nb, batchPackets, batchAddrs)
-	}
-
-	// Send all accumulated packets in one batch
-	if len(*batchPackets) > 0 {
-		batchSize := len(*batchPackets)
-		n, err := f.writers[q].WriteMulti(*batchPackets, *batchAddrs)
-		if err != nil {
-			f.l.WithError(err).WithField("sent", n).WithField("total", batchSize).Error("Failed to send batch")
-		}
-	}
-}
-
-// encryptBatch processes multiple packets, grouping by ConnectionState to reduce lock acquisitions
-func (f *Interface) encryptBatch(batch []preEncryptionPacket, nb []byte, batchPackets *[][]byte, batchAddrs *[]netip.AddrPort) {
-	lockStart := time.Now()
-	lockAcquisitions := int64(0)
-
-	if noiseutil.EncryptLockNeeded {
-		// Group packets by ConnectionState to minimize lock acquisitions
-		// Process packets in order but batch lock acquisitions per CI
-		var currentCI *ConnectionState
-		var lockHeld bool
-
-		for i := range batch {
-			ci := batch[i].ci
-			hostinfo := batch[i].hostinfo
-
-			// Validate packet data to prevent nil pointer dereference
-			if ci == nil || hostinfo == nil {
-				continue
-			}
-
-			// Switch locks if we're moving to a different ConnectionState
-			if ci != currentCI {
-				if lockHeld {
-					currentCI.writeLock.Unlock()
-				}
-				ci.writeLock.Lock()
-				lockAcquisitions++
-				currentCI = ci
-				lockHeld = true
-			}
-
-			c := ci.messageCounter.Add(1)
-			out := header.Encode(batch[i].out, header.Version, header.Message, 0, hostinfo.remoteIndexId, c)
-			f.connectionManager.Out(hostinfo)
-
-			// Query lighthouse if needed
-			if hostinfo.lastRebindCount != f.rebindCount {
-				f.lightHouse.QueryServer(hostinfo.vpnAddrs[0])
-				hostinfo.lastRebindCount = f.rebindCount
-				if f.l.Level >= logrus.DebugLevel {
-					f.l.WithField("vpnAddrs", hostinfo.vpnAddrs).Debug("Lighthouse update triggered for punch due to rebind counter")
-				}
-			}
-
-			var err error
-			out, err = ci.eKey.EncryptDanger(out, out, batch[i].packet, c, nb)
-			if err != nil {
-				hostinfo.logger(f.l).WithError(err).
-					WithField("counter", c).
-					Error("Failed to encrypt outgoing packet")
-				continue
-			}
-
-			// Add to output batches
-			*batchPackets = append(*batchPackets, out)
-			*batchAddrs = append(*batchAddrs, hostinfo.remote)
-		}
-
-		// Release final lock
-		if lockHeld {
-			currentCI.writeLock.Unlock()
-		}
-	} else {
-		// No locks needed - process directly
-		for i := range batch {
-			ci := batch[i].ci
-			hostinfo := batch[i].hostinfo
-
-			// Validate packet data to prevent nil pointer dereference
-			if ci == nil || hostinfo == nil {
-				continue
-			}
-
-			c := ci.messageCounter.Add(1)
-			out := header.Encode(batch[i].out, header.Version, header.Message, 0, hostinfo.remoteIndexId, c)
-			f.connectionManager.Out(hostinfo)
-
-			// Query lighthouse if needed
-			if hostinfo.lastRebindCount != f.rebindCount {
-				f.lightHouse.QueryServer(hostinfo.vpnAddrs[0])
-				hostinfo.lastRebindCount = f.rebindCount
-				if f.l.Level >= logrus.DebugLevel {
-					f.l.WithField("vpnAddrs", hostinfo.vpnAddrs).Debug("Lighthouse update triggered for punch due to rebind counter")
-				}
-			}
-
-			var err error
-			out, err = ci.eKey.EncryptDanger(out, out, batch[i].packet, c, nb)
-			if err != nil {
-				hostinfo.logger(f.l).WithError(err).
-					WithField("counter", c).
-					Error("Failed to encrypt outgoing packet")
-				continue
-			}
-
-			// Add to output batches
-			*batchPackets = append(*batchPackets, out)
-			*batchAddrs = append(*batchAddrs, hostinfo.remote)
-		}
-	}
-
-	// Record metrics
-	encryptionTime := time.Since(lockStart)
-	if noiseutil.EncryptLockNeeded {
-		f.batchMetrics.lockAcquisitions.Inc(lockAcquisitions)
-	}
-	f.batchMetrics.encryptionTime.Update(encryptionTime.Nanoseconds())
-	f.batchMetrics.batchSize.Update(int64(len(batch)))
-}
-
 func (f *Interface) consumeInsidePacket(packet []byte, fwPacket *firewall.Packet, nb, out []byte, q int, localCache firewall.ConntrackCache) {
 	err := newPacket(packet, false, fwPacket)
 	if err != nil {
@@ -302,9 +48,7 @@ func (f *Interface) consumeInsidePacket(packet []byte, fwPacket *firewall.Packet
 		return
 	}
 
-	hostinfo, ready := f.getOrHandshakeConsiderRouting(fwPacket, func(hh *HandshakeHostInfo) {
+	hostinfo, ready := f.getHostinfo(packet, fwPacket)
-		hh.cachePacket(f.l, header.Message, 0, packet, f.sendMessageNow, f.cachedPacketMetrics)
-	})
 
 	if hostinfo == nil {
 		f.rejectInside(packet, out, q)
@@ -320,10 +64,9 @@ func (f *Interface) consumeInsidePacket(packet []byte, fwPacket *firewall.Packet
 		return
 	}
 
-	dropReason := f.firewall.Drop(*fwPacket, false, hostinfo, f.pki.GetCAPool(), localCache)
+	dropReason := f.firewall.Drop(*fwPacket, packet, false, hostinfo, f.pki.GetCAPool(), localCache)
 	if dropReason == nil {
 		f.sendNoMetrics(header.Message, 0, hostinfo.ConnectionState, hostinfo, netip.AddrPort{}, packet, nb, out, q)
-
 	} else {
 		f.rejectInside(packet, out, q)
 		if f.l.Level >= logrus.DebugLevel {
@@ -335,6 +78,26 @@ func (f *Interface) consumeInsidePacket(packet []byte, fwPacket *firewall.Packet
 	}
 }
 
+func (f *Interface) getHostinfo(packet []byte, fwPacket *firewall.Packet) (*HostInfo, bool) {
+	if f.firewall.ShouldUnSNAT(fwPacket) {
+		//unsnat packet re-writing also happens here, would be nice to not,
+		//but we need to do the unsnat lookup to find the hostinfo so we can run the firewall checks
+		destVpnAddr := f.firewall.unSnat(packet, fwPacket)
+		if destVpnAddr.IsValid() {
+			//because this was a snatted packet, we know it has an on-overlay destination, so no routing should be required.
+			return f.getOrHandshakeNoRouting(destVpnAddr, func(hh *HandshakeHostInfo) {
+				hh.cachePacket(f.l, header.Message, 0, packet, f.sendMessageNow, f.cachedPacketMetrics)
+			})
+		} else {
+			return nil, false
+		}
+	} else { //if we didn't need to unsnat
+		return f.getOrHandshakeConsiderRouting(fwPacket, func(hh *HandshakeHostInfo) {
+			hh.cachePacket(f.l, header.Message, 0, packet, f.sendMessageNow, f.cachedPacketMetrics)
+		})
+	}
+}
+
 func (f *Interface) rejectInside(packet []byte, out []byte, q int) {
 	if !f.firewall.InSendReject {
 		return
@@ -472,7 +235,7 @@ func (f *Interface) sendMessageNow(t header.MessageType, st header.MessageSubTyp
 	}
 
 	// check if packet is in outbound fw rules
-	dropReason := f.firewall.Drop(*fp, false, hostinfo, f.pki.GetCAPool(), nil)
+	dropReason := f.firewall.Drop(*fp, p, false, hostinfo, f.pki.GetCAPool(), nil)
 	if dropReason != nil {
 		if f.l.Level >= logrus.DebugLevel {
 			f.l.WithField("fwPacket", fp).

inside_bsd.go

@@ -1,5 +1,4 @@
 //go:build darwin || dragonfly || freebsd || netbsd || openbsd
-// +build darwin dragonfly freebsd netbsd openbsd
 
 package nebula
 

inside_generic.go

@@ -1,5 +1,4 @@
 //go:build !darwin && !dragonfly && !freebsd && !netbsd && !openbsd
-// +build !darwin,!dragonfly,!freebsd,!netbsd,!openbsd
 
 package nebula
 

interface.go

@@ -4,6 +4,7 @@ import (
 	"context"
 	"errors"
 	"fmt"
+	"io"
 	"net/netip"
 	"os"
 	"runtime"
@@ -21,7 +22,6 @@ import (
 )
 
 const mtu = 9001
-const virtioNetHdrLen = overlay.VirtioNetHdrLen
 
 type InterfaceConfig struct {
 	HostMap            *HostMap
@@ -50,13 +50,6 @@ type InterfaceConfig struct {
 	l                     *logrus.Logger
 }
 
-type batchMetrics struct {
-	udpReadSize      metrics.Histogram
-	encryptionTime   metrics.Histogram // Time spent in encryption (including lock waits)
-	batchSize        metrics.Histogram // Dynamic batch sizes being used
-	lockAcquisitions metrics.Counter   // Number of lock acquisitions (should be minimal)
-}
-
 type Interface struct {
 	hostMap               *HostMap
 	outside               udp.Conn
@@ -94,12 +87,11 @@ type Interface struct {
 	conntrackCacheTimeout time.Duration
 
 	writers []udp.Conn
-	readers []overlay.BatchReadWriter
+	readers []io.ReadWriteCloser
 
 	metricHandshakes    metrics.Histogram
 	messageMetrics      *MessageMetrics
 	cachedPacketMetrics *cachedPacketMetrics
-	batchMetrics        *batchMetrics
 
 	l *logrus.Logger
 }
@@ -186,7 +178,7 @@ func NewInterface(ctx context.Context, c *InterfaceConfig) (*Interface, error) {
 		routines:              c.routines,
 		version:               c.version,
 		writers:               make([]udp.Conn, c.routines),
-		readers:               make([]overlay.BatchReadWriter, c.routines),
+		readers:               make([]io.ReadWriteCloser, c.routines),
 		myVpnNetworks:         cs.myVpnNetworks,
 		myVpnNetworksTable:    cs.myVpnNetworksTable,
 		myVpnAddrs:            cs.myVpnAddrs,
@@ -202,12 +194,6 @@ func NewInterface(ctx context.Context, c *InterfaceConfig) (*Interface, error) {
 			sent:    metrics.GetOrRegisterCounter("hostinfo.cached_packets.sent", nil),
 			dropped: metrics.GetOrRegisterCounter("hostinfo.cached_packets.dropped", nil),
 		},
-		batchMetrics: &batchMetrics{
-			udpReadSize:      metrics.GetOrRegisterHistogram("batch.udp_read_size", nil, metrics.NewUniformSample(1024)),
-			encryptionTime:   metrics.GetOrRegisterHistogram("batch.encryption_time_ns", nil, metrics.NewUniformSample(1024)),
-			batchSize:        metrics.GetOrRegisterHistogram("batch.size", nil, metrics.NewUniformSample(1024)),
-			lockAcquisitions: metrics.GetOrRegisterCounter("batch.lock_acquisitions", nil),
-		},
 
 		l: c.l,
 	}
@@ -247,7 +233,7 @@ func (f *Interface) activate() {
 	metrics.GetOrRegisterGauge("routines", nil).Update(int64(f.routines))
 
 	// Prepare n tun queues
-	var reader overlay.BatchReadWriter = f.inside
+	var reader io.ReadWriteCloser = f.inside
 	for i := 0; i < f.routines; i++ {
 		if i > 0 {
 			reader, err = f.inside.NewMultiQueueReader()
@@ -262,6 +248,7 @@ func (f *Interface) activate() {
 		f.inside.Close()
 		f.l.Fatal(err)
 	}
+	f.firewall.SetSNATAddressFromInterface(f)
 }
 
 func (f *Interface) run() {
@@ -288,68 +275,39 @@ func (f *Interface) listenOut(i int) {
 
 	ctCache := firewall.NewConntrackCacheTicker(f.conntrackCacheTimeout)
 	lhh := f.lightHouse.NewRequestHandler()
-
+	plaintext := make([]byte, udp.MTU)
-	// Pre-allocate output buffers for batch processing
-	batchSize := li.BatchSize()
-	outs := make([][]byte, batchSize)
-	for idx := range outs {
-		// Allocate full buffer with virtio header space
-		outs[idx] = make([]byte, virtioNetHdrLen, virtioNetHdrLen+udp.MTU)
-	}
-
 	h := &header.H{}
 	fwPacket := &firewall.Packet{}
-	nb := make([]byte, 12)
+	nb := make([]byte, 12, 12)
-
-	li.ListenOutBatch(func(addrs []netip.AddrPort, payloads [][]byte, count int) {
-		f.readOutsidePacketsBatch(addrs, payloads, count, outs[:count], nb, i, h, fwPacket, lhh, ctCache.Get(f.l))
 
+	li.ListenOut(func(fromUdpAddr netip.AddrPort, payload []byte) {
+		f.readOutsidePackets(ViaSender{UdpAddr: fromUdpAddr}, plaintext[:0], payload, h, fwPacket, lhh, nb, i, ctCache.Get(f.l))
 	})
 }
 
-func (f *Interface) listenIn(reader overlay.BatchReadWriter, i int) {
+func (f *Interface) listenIn(reader io.ReadWriteCloser, i int) {
 	runtime.LockOSThread()
 
-	batchSize := reader.BatchSize()
+	packet := make([]byte, mtu)
-
+	out := make([]byte, mtu)
-	// Allocate buffers for batch reading
+	fwPacket := &firewall.Packet{}
-	bufs := make([][]byte, batchSize)
+	nb := make([]byte, 12, 12)
-	for idx := range bufs {
-		bufs[idx] = make([]byte, mtu)
-	}
-	sizes := make([]int, batchSize)
-
-	// Allocate output buffers for batch processing (one per packet)
-	// Each has virtio header headroom to avoid copies on write
-	outs := make([][]byte, batchSize)
-	for idx := range outs {
-		outBuf := make([]byte, virtioNetHdrLen+mtu)
-		outs[idx] = outBuf[virtioNetHdrLen:] // Slice starting after headroom
-	}
-
-	// Pre-allocate batch accumulation buffers for sending
-	batchPackets := make([][]byte, 0, batchSize)
-	batchAddrs := make([]netip.AddrPort, 0, batchSize)
-
-	// Pre-allocate nonce buffer (reused for all encryptions)
-	nb := make([]byte, 12)
 
 	conntrackCache := firewall.NewConntrackCacheTicker(f.conntrackCacheTimeout)
 
 	for {
-		n, err := reader.BatchRead(bufs, sizes)
+		n, err := reader.Read(packet)
 		if err != nil {
 			if errors.Is(err, os.ErrClosed) && f.closed.Load() {
 				return
 			}
 
-			f.l.WithError(err).Error("Error while batch reading outbound packets")
+			f.l.WithError(err).Error("Error while reading outbound packet")
 			// This only seems to happen when something fatal happens to the fd, so exit.
 			os.Exit(2)
 		}
 
-		// Process all packets in the batch at once
+		f.consumeInsidePacket(packet[:n], fwPacket, nb, out, i, conntrackCache.Get(f.l))
-		f.consumeInsidePackets(bufs, sizes, n, outs, nb, i, conntrackCache.Get(f.l), &batchPackets, &batchAddrs)
 	}
 }
 
@@ -387,6 +345,7 @@ func (f *Interface) reloadFirewall(c *config.C) {
 		f.l.WithError(err).Error("Error while creating firewall during reload")
 		return
 	}
+	fw.SetSNATAddressFromInterface(f)
 
 	oldFw := f.firewall
 	conntrack := oldFw.Conntrack

lighthouse.go

@@ -713,21 +713,14 @@ func (lh *LightHouse) unlockedShouldAddV6(vpnAddr netip.Addr, to *V6AddrPort) bo
 
 func (lh *LightHouse) IsLighthouseAddr(vpnAddr netip.Addr) bool {
 	l := lh.GetLighthouses()
-	for i := range l {
+	return slices.Contains(l, vpnAddr)
-		if l[i] == vpnAddr {
-			return true
-		}
-	}
-	return false
 }
 
 func (lh *LightHouse) IsAnyLighthouseAddr(vpnAddrs []netip.Addr) bool {
 	l := lh.GetLighthouses()
 	for i := range vpnAddrs {
-		for j := range l {
+		if slices.Contains(l, vpnAddrs[i]) {
-			if l[j] == vpnAddrs[i] {
+			return true
-				return true
-			}
 		}
 	}
 	return false

main.go

@@ -105,11 +105,7 @@ func Main(c *config.C, configTest bool, buildVersion string, logger *logrus.Logg
 		// deprecated and undocumented
 		tunQueues := c.GetInt("tun.routines", 1)
 		udpQueues := c.GetInt("listen.routines", 1)
-		if tunQueues > udpQueues {
+		routines = max(tunQueues, udpQueues)
-			routines = tunQueues
-		} else {
-			routines = udpQueues
-		}
 		if routines != 1 {
 			l.WithField("routines", routines).Warn("Setting tun.routines and listen.routines is deprecated. Use `routines` instead")
 		}
@@ -135,7 +131,8 @@ func Main(c *config.C, configTest bool, buildVersion string, logger *logrus.Logg
 			deviceFactory = overlay.NewDeviceFromConfig
 		}
 
-		tun, err = deviceFactory(c, l, pki.getCertState().myVpnNetworks, routines)
+		cs := pki.getCertState()
+		tun, err = deviceFactory(c, l, cs.myVpnNetworks, cs.GetDefaultCertificate().UnsafeNetworks(), routines)
 		if err != nil {
 			return nil, util.ContextualizeIfNeeded("Failed to get a tun/tap device", err)
 		}
@@ -171,7 +168,7 @@ func Main(c *config.C, configTest bool, buildVersion string, logger *logrus.Logg
 
 		for i := 0; i < routines; i++ {
 			l.Infof("listening on %v", netip.AddrPortFrom(listenHost, uint16(port)))
-			udpServer, err := udp.NewListener(l, listenHost, port, routines > 1, c.GetInt("listen.batch", 128))
+			udpServer, err := udp.NewListener(l, listenHost, port, routines > 1, c.GetInt("listen.batch", 64))
 			if err != nil {
 				return nil, util.NewContextualError("Failed to open udp listener", m{"queue": i}, err)
 			}

notboring.go

@@ -1,5 +1,4 @@
 //go:build !boringcrypto
-// +build !boringcrypto
 
 package nebula
 

outside.go

@@ -102,7 +102,7 @@ func (f *Interface) readOutsidePackets(via ViaSender, out []byte, packet []byte,
 					relay:     relay,
 					IsRelayed: true,
 				}
-				f.readOutsidePackets(via, out[:virtioNetHdrLen], signedPayload, h, fwPacket, lhf, nb, q, localCache)
+				f.readOutsidePackets(via, out[:0], signedPayload, h, fwPacket, lhf, nb, q, localCache)
 				return
 			case ForwardingType:
 				// Find the target HostInfo relay object
@@ -145,7 +145,7 @@ func (f *Interface) readOutsidePackets(via ViaSender, out []byte, packet []byte,
 		}
 
 		//TODO: assert via is not relayed
-		lhf.HandleRequest(via.UdpAddr, hostinfo.vpnAddrs, d[virtioNetHdrLen:], f)
+		lhf.HandleRequest(via.UdpAddr, hostinfo.vpnAddrs, d, f)
 
 		// Fallthrough to the bottom to record incoming traffic
 
@@ -167,7 +167,7 @@ func (f *Interface) readOutsidePackets(via ViaSender, out []byte, packet []byte,
 			// This testRequest might be from TryPromoteBest, so we should roam
 			// to the new IP address before responding
 			f.handleHostRoaming(hostinfo, via)
-			f.send(header.Test, header.TestReply, ci, hostinfo, d[virtioNetHdrLen:], nb, out)
+			f.send(header.Test, header.TestReply, ci, hostinfo, d, nb, out)
 		}
 
 		// Fallthrough to the bottom to record incoming traffic
@@ -210,7 +210,7 @@ func (f *Interface) readOutsidePackets(via ViaSender, out []byte, packet []byte,
 			return
 		}
 
-		f.relayManager.HandleControlMsg(hostinfo, d[virtioNetHdrLen:], f)
+		f.relayManager.HandleControlMsg(hostinfo, d, f)
 
 	default:
 		f.messageMetrics.Rx(h.Type, h.Subtype, 1)
@@ -327,13 +327,29 @@ func parseV6(data []byte, incoming bool, fp *firewall.Packet) error {
 		proto := layers.IPProtocol(data[protoAt])
 
 		switch proto {
-		case layers.IPProtocolICMPv6, layers.IPProtocolESP, layers.IPProtocolNoNextHeader:
+		case layers.IPProtocolESP, layers.IPProtocolNoNextHeader:
 			fp.Protocol = uint8(proto)
 			fp.RemotePort = 0
 			fp.LocalPort = 0
 			fp.Fragment = false
 			return nil
 
+		case layers.IPProtocolICMPv6:
+			if dataLen < offset+6 {
+				return ErrIPv6PacketTooShort
+			}
+			fp.Protocol = uint8(proto)
+			fp.LocalPort = 0 //incoming vs outgoing doesn't matter for icmpv6
+			icmptype := data[offset+1]
+			switch icmptype {
+			case layers.ICMPv6TypeEchoRequest, layers.ICMPv6TypeEchoReply:
+				fp.RemotePort = binary.BigEndian.Uint16(data[offset+4 : offset+6]) //identifier
+			default:
+				fp.RemotePort = 0
+			}
+			fp.Fragment = false
+			return nil
+
 		case layers.IPProtocolTCP, layers.IPProtocolUDP:
 			if dataLen < offset+4 {
 				return ErrIPv6PacketTooShort
@@ -423,34 +439,38 @@ func parseV4(data []byte, incoming bool, fp *firewall.Packet) error {
 
 	// Accounting for a variable header length, do we have enough data for our src/dst tuples?
 	minLen := ihl
-	if !fp.Fragment && fp.Protocol != firewall.ProtoICMP {
+	if !fp.Fragment {
-		minLen += minFwPacketLen
+		if fp.Protocol == firewall.ProtoICMP {
+			minLen += minFwPacketLen + 2
+		} else {
+			minLen += minFwPacketLen
+		}
 	}
+
 	if len(data) < minLen {
 		return ErrIPv4InvalidHeaderLength
 	}
 
-	// Firewall packets are locally oriented
+	if incoming { // Firewall packets are locally oriented
-	if incoming {
 		fp.RemoteAddr, _ = netip.AddrFromSlice(data[12:16])
 		fp.LocalAddr, _ = netip.AddrFromSlice(data[16:20])
-		if fp.Fragment || fp.Protocol == firewall.ProtoICMP {
-			fp.RemotePort = 0
-			fp.LocalPort = 0
-		} else {
-			fp.RemotePort = binary.BigEndian.Uint16(data[ihl : ihl+2])
-			fp.LocalPort = binary.BigEndian.Uint16(data[ihl+2 : ihl+4])
-		}
 	} else {
 		fp.LocalAddr, _ = netip.AddrFromSlice(data[12:16])
 		fp.RemoteAddr, _ = netip.AddrFromSlice(data[16:20])
-		if fp.Fragment || fp.Protocol == firewall.ProtoICMP {
+	}
-			fp.RemotePort = 0
+
-			fp.LocalPort = 0
+	if fp.Fragment {
-		} else {
+		fp.RemotePort = 0
-			fp.LocalPort = binary.BigEndian.Uint16(data[ihl : ihl+2])
+		fp.LocalPort = 0
-			fp.RemotePort = binary.BigEndian.Uint16(data[ihl+2 : ihl+4])
+	} else if fp.Protocol == firewall.ProtoICMP { //note that orientation doesn't matter on ICMP
-		}
+		fp.RemotePort = binary.BigEndian.Uint16(data[ihl+4 : ihl+6]) //identifier
+		fp.LocalPort = 0                                             //code would be uint16(data[ihl+1])
+	} else if incoming {
+		fp.RemotePort = binary.BigEndian.Uint16(data[ihl : ihl+2])  //src port
+		fp.LocalPort = binary.BigEndian.Uint16(data[ihl+2 : ihl+4]) //dst port
+	} else {
+		fp.LocalPort = binary.BigEndian.Uint16(data[ihl : ihl+2])    //src port
+		fp.RemotePort = binary.BigEndian.Uint16(data[ihl+2 : ihl+4]) //dst port
 	}
 
 	return nil
@@ -481,11 +501,9 @@ func (f *Interface) decryptToTun(hostinfo *HostInfo, messageCounter uint64, out
 		return false
 	}
 
-	packetData := out[virtioNetHdrLen:]
+	err = newPacket(out, true, fwPacket)
-
-	err = newPacket(packetData, true, fwPacket)
 	if err != nil {
-		hostinfo.logger(f.l).WithError(err).WithField("packet", packetData).
+		hostinfo.logger(f.l).WithError(err).WithField("packet", out).
 			Warnf("Error while validating inbound packet")
 		return false
 	}
@@ -496,11 +514,11 @@ func (f *Interface) decryptToTun(hostinfo *HostInfo, messageCounter uint64, out
 		return false
 	}
 
-	dropReason := f.firewall.Drop(*fwPacket, true, hostinfo, f.pki.GetCAPool(), localCache)
+	dropReason := f.firewall.Drop(*fwPacket, out, true, hostinfo, f.pki.GetCAPool(), localCache)
 	if dropReason != nil {
 		// NOTE: We give `packet` as the `out` here since we already decrypted from it and we don't need it anymore
 		// This gives us a buffer to build the reject packet in
-		f.rejectOutside(packetData, hostinfo.ConnectionState, hostinfo, nb, packet, q)
+		f.rejectOutside(out, hostinfo.ConnectionState, hostinfo, nb, packet, q)
 		if f.l.Level >= logrus.DebugLevel {
 			hostinfo.logger(f.l).WithField("fwPacket", fwPacket).
 				WithField("reason", dropReason).
@@ -564,108 +582,3 @@ func (f *Interface) handleRecvError(addr netip.AddrPort, h *header.H) {
 	// We also delete it from pending hostmap to allow for fast reconnect.
 	f.handshakeManager.DeleteHostInfo(hostinfo)
 }
-
-// readOutsidePacketsBatch processes multiple packets received from UDP in a batch
-// and writes all successfully decrypted packets to TUN in a single operation
-func (f *Interface) readOutsidePacketsBatch(addrs []netip.AddrPort, payloads [][]byte, count int, outs [][]byte, nb []byte, q int, h *header.H, fwPacket *firewall.Packet, lhf *LightHouseHandler, localCache firewall.ConntrackCache) {
-	// Pre-allocate slice for accumulating successful decryptions
-	tunPackets := make([][]byte, 0, count)
-
-	for i := 0; i < count; i++ {
-		payload := payloads[i]
-		addr := addrs[i]
-		out := outs[i]
-
-		// Parse header
-		err := h.Parse(payload)
-		if err != nil {
-			if len(payload) > 1 {
-				f.l.WithField("packet", payload).Infof("Error while parsing inbound packet from %s: %s", addr, err)
-			}
-			continue
-		}
-
-		if addr.IsValid() {
-			if f.myVpnNetworksTable.Contains(addr.Addr()) {
-				if f.l.Level >= logrus.DebugLevel {
-					f.l.WithField("udpAddr", addr).Debug("Refusing to process double encrypted packet")
-				}
-				continue
-			}
-		}
-
-		var hostinfo *HostInfo
-		if h.Type == header.Message && h.Subtype == header.MessageRelay {
-			hostinfo = f.hostMap.QueryRelayIndex(h.RemoteIndex)
-		} else {
-			hostinfo = f.hostMap.QueryIndex(h.RemoteIndex)
-		}
-
-		var ci *ConnectionState
-		if hostinfo != nil {
-			ci = hostinfo.ConnectionState
-		}
-
-		switch h.Type {
-		case header.Message:
-			if !f.handleEncrypted(ci, ViaSender{UdpAddr: addr}, h) {
-				continue
-			}
-
-			switch h.Subtype {
-			case header.MessageNone:
-				// Decrypt packet
-				out, err = hostinfo.ConnectionState.dKey.DecryptDanger(out, payload[:header.Len], payload[header.Len:], h.MessageCounter, nb)
-				if err != nil {
-					hostinfo.logger(f.l).WithError(err).Error("Failed to decrypt packet")
-					continue
-				}
-
-				packetData := out[virtioNetHdrLen:]
-
-				err = newPacket(packetData, true, fwPacket)
-				if err != nil {
-					hostinfo.logger(f.l).WithError(err).WithField("packet", packetData).Warnf("Error while validating inbound packet")
-					continue
-				}
-
-				if !hostinfo.ConnectionState.window.Update(f.l, h.MessageCounter) {
-					hostinfo.logger(f.l).WithField("fwPacket", fwPacket).Debugln("dropping out of window packet")
-					continue
-				}
-
-				dropReason := f.firewall.Drop(*fwPacket, true, hostinfo, f.pki.GetCAPool(), localCache)
-				if dropReason != nil {
-					f.rejectOutside(packetData, hostinfo.ConnectionState, hostinfo, nb, payload, q)
-					if f.l.Level >= logrus.DebugLevel {
-						hostinfo.logger(f.l).WithField("fwPacket", fwPacket).WithField("reason", dropReason).Debugln("dropping inbound packet")
-					}
-					continue
-				}
-
-				f.connectionManager.In(hostinfo)
-				// Add to batch for TUN write
-				tunPackets = append(tunPackets, out)
-
-			case header.MessageRelay:
-				// Skip relay packets in batch mode for now (less common path)
-				f.readOutsidePackets(ViaSender{UdpAddr: addr}, out[:virtioNetHdrLen], payload, h, fwPacket, lhf, nb, q, localCache)
-
-			default:
-				hostinfo.logger(f.l).Debugf("unexpected message subtype %d", h.Subtype)
-			}
-
-		default:
-			// Handle non-Message types using single-packet path
-			f.readOutsidePackets(ViaSender{UdpAddr: addr}, out[:virtioNetHdrLen], payload, h, fwPacket, lhf, nb, q, localCache)
-		}
-	}
-
-	if len(tunPackets) > 0 {
-		n, err := f.readers[q].WriteBatch(tunPackets, virtioNetHdrLen)
-		if err != nil {
-			f.l.WithError(err).WithField("sent", n).WithField("total", len(tunPackets)).Error("Failed to batch write to tun")
-		}
-
-	}
-}

outside_test.go

@@ -155,6 +155,7 @@ func Test_newPacket_v6(t *testing.T) {
 	// next layer, missing length byte
 	err = newPacket(buffer.Bytes()[:49], true, p)
 	require.ErrorIs(t, err, ErrIPv6CouldNotFindPayload)
+	err = nil
 
 	// A good ICMP packet
 	ip = layers.IPv6{
@@ -165,20 +166,26 @@ func Test_newPacket_v6(t *testing.T) {
 		DstIP:      net.IPv6linklocalallnodes,
 	}
 
-	icmp := layers.ICMPv6{}
+	icmp := layers.ICMPv6{
-
+		TypeCode: layers.ICMPv6TypeEchoRequest,
-	buffer.Clear()
+		Checksum: 0x1234,
-	err = gopacket.SerializeLayers(buffer, opt, &ip, &icmp)
-	if err != nil {
-		panic(err)
 	}
 
-	err = newPacket(buffer.Bytes(), true, p)
+	buffer.Clear()
-	require.NoError(t, err)
+	require.NoError(t, gopacket.SerializeLayers(buffer, opt, &ip, &icmp))
+	require.Error(t, newPacket(buffer.Bytes(), true, p))
+
+	buffer.Clear()
+	echo := layers.ICMPv6Echo{
+		Identifier: 0xabcd,
+		SeqNumber:  1234,
+	}
+	require.NoError(t, gopacket.SerializeLayers(buffer, opt, &ip, &icmp, &echo))
+	require.NoError(t, newPacket(buffer.Bytes(), true, p))
 	assert.Equal(t, uint8(layers.IPProtocolICMPv6), p.Protocol)
 	assert.Equal(t, netip.MustParseAddr("ff02::2"), p.RemoteAddr)
 	assert.Equal(t, netip.MustParseAddr("ff02::1"), p.LocalAddr)
-	assert.Equal(t, uint16(0), p.RemotePort)
+	assert.Equal(t, uint16(0xabcd), p.RemotePort)
 	assert.Equal(t, uint16(0), p.LocalPort)
 	assert.False(t, p.Fragment)
 
@@ -574,7 +581,7 @@ func BenchmarkParseV6(b *testing.B) {
 	}
 
 	evilBytes := buffer.Bytes()
-	for i := 0; i < 200; i++ {
+	for range 200 {
 		evilBytes = append(evilBytes, hopHeader...)
 	}
 	evilBytes = append(evilBytes, lastHopHeader...)

overlay/device.go

@@ -7,26 +7,15 @@ import (
 	"github.com/slackhq/nebula/routing"
 )
 
-// BatchReadWriter extends io.ReadWriteCloser with batch I/O operations
-type BatchReadWriter interface {
-	io.ReadWriteCloser
-
-	// BatchRead reads multiple packets at once
-	BatchRead(bufs [][]byte, sizes []int) (int, error)
-
-	// WriteBatch writes multiple packets at once
-	WriteBatch(bufs [][]byte, offset int) (int, error)
-
-	// BatchSize returns the optimal batch size for this device
-	BatchSize() int
-}
-
 type Device interface {
-	BatchReadWriter
+	io.ReadWriteCloser
 	Activate() error
 	Networks() []netip.Prefix
+	UnsafeNetworks() []netip.Prefix
+	UnsafeIPv4OriginAddress() netip.Prefix
+	SNATAddress() netip.Prefix
 	Name() string
 	RoutesFor(netip.Addr) routing.Gateways
 	SupportsMultiqueue() bool
-	NewMultiQueueReader() (BatchReadWriter, error)
+	NewMultiQueueReader() (io.ReadWriteCloser, error)
 }

overlay/tun.go

@@ -1,7 +1,9 @@
 package overlay
 
 import (
+	"crypto/rand"
 	"fmt"
+	"io"
 	"net"
 	"net/netip"
 
@@ -11,7 +13,6 @@ import (
 )
 
 const DefaultMTU = 1300
-const VirtioNetHdrLen = 10 // Size of virtio_net_hdr structure
 
 type NameError struct {
 	Name       string
@@ -23,22 +24,22 @@ func (e *NameError) Error() string {
 }
 
 // TODO: We may be able to remove routines
-type DeviceFactory func(c *config.C, l *logrus.Logger, vpnNetworks []netip.Prefix, routines int) (Device, error)
+type DeviceFactory func(c *config.C, l *logrus.Logger, vpnNetworks []netip.Prefix, unsafeNetworks []netip.Prefix, routines int) (Device, error)
 
-func NewDeviceFromConfig(c *config.C, l *logrus.Logger, vpnNetworks []netip.Prefix, routines int) (Device, error) {
+func NewDeviceFromConfig(c *config.C, l *logrus.Logger, vpnNetworks []netip.Prefix, unsafeNetworks []netip.Prefix, routines int) (Device, error) {
 	switch {
 	case c.GetBool("tun.disabled", false):
 		tun := newDisabledTun(vpnNetworks, c.GetInt("tun.tx_queue", 500), c.GetBool("stats.message_metrics", false), l)
 		return tun, nil
 
 	default:
-		return newTun(c, l, vpnNetworks, routines > 1)
+		return newTun(c, l, vpnNetworks, unsafeNetworks, routines > 1)
 	}
 }
 
 func NewFdDeviceFromConfig(fd *int) DeviceFactory {
-	return func(c *config.C, l *logrus.Logger, vpnNetworks []netip.Prefix, routines int) (Device, error) {
+	return func(c *config.C, l *logrus.Logger, vpnNetworks []netip.Prefix, unsafeNetworks []netip.Prefix, routines int) (Device, error) {
-		return newTunFromFd(c, l, *fd, vpnNetworks)
+		return newTunFromFd(c, l, *fd, vpnNetworks, unsafeNetworks)
 	}
 }
 
@@ -130,3 +131,85 @@ func selectGateway(dest netip.Prefix, gateways []netip.Prefix) (netip.Prefix, er
 
 	return netip.Prefix{}, fmt.Errorf("no gateway found for %v in the list of vpn networks", dest)
 }
+
+// genLinkLocal generates a random IPv4 link-local address.
+// If randomizer is nil, it uses rand.Reader to find two random bytes
+func genLinkLocal(randomizer io.Reader) netip.Prefix {
+	if randomizer == nil {
+		randomizer = rand.Reader
+	}
+	octets := []byte{169, 254, 0, 0}
+	_, _ = randomizer.Read(octets[2:4])
+	return coerceLinkLocal(octets)
+}
+
+func coerceLinkLocal(octets []byte) netip.Prefix {
+	if octets[3] == 0 {
+		octets[3] = 1 //please no .0 addresses
+	} else if octets[2] == 255 && octets[3] == 255 {
+		octets[3] = 254 //please no broadcast addresses
+	}
+	out, _ := netip.AddrFromSlice(octets)
+	return netip.PrefixFrom(out, 32)
+}
+
+// prepareUnsafeOriginAddr provides the IPv4 address used on IPv6-only clients that need to access IPv4 unsafe routes
+func prepareUnsafeOriginAddr(d Device, l *logrus.Logger, c *config.C, routes []Route) netip.Prefix {
+	if !d.Networks()[0].Addr().Is6() {
+		return netip.Prefix{} //if we have an IPv4 assignment within the overlay, we don't need an unsafe origin address
+	}
+
+	needed := false
+	for _, route := range routes { //or if we have a route defined into an IPv4 range
+		if route.Cidr.Addr().Is4() {
+			needed = true //todo should this only apply to unsafe routes? almost certainly
+			break
+		}
+	}
+	if !needed {
+		return netip.Prefix{}
+	}
+
+	//todo better config name for sure
+	if a := c.GetString("tun.unsafe_origin_address_for_4over6", ""); a != "" {
+		out, err := netip.ParseAddr(a)
+		if err != nil {
+			l.WithField("value", a).WithError(err).Warn("failed to parse tun.unsafe_origin_address_for_4over6, will use a random value")
+		} else if !out.Is4() || !out.IsLinkLocalUnicast() {
+			l.WithField("value", out).Warn("tun.unsafe_origin_address_for_4over6 must be an IPv4 address")
+		} else if out.IsValid() {
+			return netip.PrefixFrom(out, 32)
+		}
+	}
+	return genLinkLocal(nil)
+}
+
+// prepareSnatAddr provides the address that an IPv6-only unsafe router should use to SNAT traffic before handing it to the operating system
+func prepareSnatAddr(d Device, l *logrus.Logger, c *config.C) netip.Prefix {
+	if !d.Networks()[0].Addr().Is6() {
+		return netip.Prefix{} //if we have an IPv4 assignment within the overlay, we don't need a snat address
+	}
+
+	needed := false
+	for _, un := range d.UnsafeNetworks() { //if we are an unsafe router for an IPv4 range
+		if un.Addr().Is4() {
+			needed = true
+			break
+		}
+	}
+	if !needed {
+		return netip.Prefix{}
+	}
+
+	if a := c.GetString("tun.snat_address_for_4over6", ""); a != "" {
+		out, err := netip.ParseAddr(a)
+		if err != nil {
+			l.WithField("value", a).WithError(err).Warn("failed to parse tun.snat_address_for_4over6, will use a random value")
+		} else if !out.Is4() || !out.IsLinkLocalUnicast() {
+			l.WithField("value", out).Warn("tun.snat_address_for_4over6 must be an IPv4 address")
+		} else if out.IsValid() {
+			return netip.PrefixFrom(out, 32)
+		}
+	}
+	return genLinkLocal(nil)
+}

overlay/tun_android.go

@@ -19,14 +19,16 @@ import (
 
 type tun struct {
 	io.ReadWriteCloser
-	fd          int
+	fd               int
-	vpnNetworks []netip.Prefix
+	vpnNetworks      []netip.Prefix
-	Routes      atomic.Pointer[[]Route]
+	unsafeNetworks   []netip.Prefix
-	routeTree   atomic.Pointer[bart.Table[routing.Gateways]]
+	unsafeIPv4Origin netip.Prefix
-	l           *logrus.Logger
+	Routes           atomic.Pointer[[]Route]
+	routeTree        atomic.Pointer[bart.Table[routing.Gateways]]
+	l                *logrus.Logger
 }
 
-func newTunFromFd(c *config.C, l *logrus.Logger, deviceFd int, vpnNetworks []netip.Prefix) (*tun, error) {
+func newTunFromFd(c *config.C, l *logrus.Logger, deviceFd int, vpnNetworks []netip.Prefix, unsafeNetworks []netip.Prefix) (*tun, error) {
 	// XXX Android returns an fd in non-blocking mode which is necessary for shutdown to work properly.
 	// Be sure not to call file.Fd() as it will set the fd to blocking mode.
 	file := os.NewFile(uintptr(deviceFd), "/dev/net/tun")
@@ -35,6 +37,7 @@ func newTunFromFd(c *config.C, l *logrus.Logger, deviceFd int, vpnNetworks []net
 		ReadWriteCloser: file,
 		fd:              deviceFd,
 		vpnNetworks:     vpnNetworks,
+		unsafeNetworks:  unsafeNetworks,
 		l:               l,
 	}
 
@@ -53,7 +56,7 @@ func newTunFromFd(c *config.C, l *logrus.Logger, deviceFd int, vpnNetworks []net
 	return t, nil
 }
 
-func newTun(_ *config.C, _ *logrus.Logger, _ []netip.Prefix, _ bool) (*tun, error) {
+func newTun(_ *config.C, _ *logrus.Logger, _ []netip.Prefix, _ []netip.Prefix, _ bool) (*tun, error) {
 	return nil, fmt.Errorf("newTun not supported in Android")
 }
 
@@ -76,6 +79,8 @@ func (t *tun) reload(c *config.C, initial bool) error {
 		return nil
 	}
 
+	t.unsafeIPv4Origin = prepareUnsafeOriginAddr(t, t.l, c, routes)
+
 	routeTree, err := makeRouteTree(t.l, routes, false)
 	if err != nil {
 		return err
@@ -91,6 +96,18 @@ func (t *tun) Networks() []netip.Prefix {
 	return t.vpnNetworks
 }
 
+func (t *tun) UnsafeNetworks() []netip.Prefix {
+	return t.unsafeNetworks
+}
+
+func (t *tun) UnsafeIPv4OriginAddress() netip.Prefix {
+	return t.unsafeIPv4Origin
+}
+
+func (t *tun) SNATAddress() netip.Prefix {
+	return netip.Prefix{}
+}
+
 func (t *tun) Name() string {
 	return "android"
 }
@@ -99,29 +116,6 @@ func (t *tun) SupportsMultiqueue() bool {
 	return false
 }
 
-func (t *tun) NewMultiQueueReader() (BatchReadWriter, error) {
+func (t *tun) NewMultiQueueReader() (io.ReadWriteCloser, error) {
 	return nil, fmt.Errorf("TODO: multiqueue not implemented for android")
 }
-
-func (t *tun) BatchRead(bufs [][]byte, sizes []int) (int, error) {
-	n, err := t.Read(bufs[0])
-	if err != nil {
-		return 0, err
-	}
-	sizes[0] = n
-	return 1, nil
-}
-
-func (t *tun) WriteBatch(bufs [][]byte, offset int) (int, error) {
-	for i, buf := range bufs {
-		_, err := t.Write(buf[offset:])
-		if err != nil {
-			return i, err
-		}
-	}
-	return len(bufs), nil
-}
-
-func (t *tun) BatchSize() int {
-	return 1
-}

overlay/tun_darwin.go

@@ -24,13 +24,15 @@ import (
 
 type tun struct {
 	io.ReadWriteCloser
-	Device      string
+	Device           string
-	vpnNetworks []netip.Prefix
+	vpnNetworks      []netip.Prefix
-	DefaultMTU  int
+	unsafeNetworks   []netip.Prefix
-	Routes      atomic.Pointer[[]Route]
+	unsafeIPv4Origin netip.Prefix
-	routeTree   atomic.Pointer[bart.Table[routing.Gateways]]
+	DefaultMTU       int
-	linkAddr    *netroute.LinkAddr
+	Routes           atomic.Pointer[[]Route]
-	l           *logrus.Logger
+	routeTree        atomic.Pointer[bart.Table[routing.Gateways]]
+	linkAddr         *netroute.LinkAddr
+	l                *logrus.Logger
 
 	// cache out buffer since we need to prepend 4 bytes for tun metadata
 	out []byte
@@ -79,7 +81,7 @@ type ifreqAlias6 struct {
 	Lifetime   addrLifetime
 }
 
-func newTun(c *config.C, l *logrus.Logger, vpnNetworks []netip.Prefix, _ bool) (*tun, error) {
+func newTun(c *config.C, l *logrus.Logger, vpnNetworks []netip.Prefix, unsafeNetworks []netip.Prefix, _ bool) (*tun, error) {
 	name := c.GetString("tun.dev", "")
 	ifIndex := -1
 	if name != "" && name != "utun" {
@@ -127,6 +129,7 @@ func newTun(c *config.C, l *logrus.Logger, vpnNetworks []netip.Prefix, _ bool) (
 		ReadWriteCloser: os.NewFile(uintptr(fd), ""),
 		Device:          name,
 		vpnNetworks:     vpnNetworks,
+		unsafeNetworks:  unsafeNetworks,
 		DefaultMTU:      c.GetInt("tun.mtu", DefaultMTU),
 		l:               l,
 	}
@@ -153,7 +156,7 @@ func (t *tun) deviceBytes() (o [16]byte) {
 	return
 }
 
-func newTunFromFd(_ *config.C, _ *logrus.Logger, _ int, _ []netip.Prefix) (*tun, error) {
+func newTunFromFd(_ *config.C, _ *logrus.Logger, _ int, _ []netip.Prefix, _ []netip.Prefix) (*tun, error) {
 	return nil, fmt.Errorf("newTunFromFd not supported in Darwin")
 }
 
@@ -213,6 +216,11 @@ func (t *tun) Activate() error {
 			}
 		}
 	}
+	if t.unsafeIPv4Origin.IsValid() && t.unsafeIPv4Origin.Addr().Is4() {
+		if err = t.activate4(t.unsafeIPv4Origin); err != nil {
+			return err
+		}
+	}
 
 	// Run the interface
 	ifrf.Flags = ifrf.Flags | unix.IFF_UP | unix.IFF_RUNNING
@@ -314,6 +322,10 @@ func (t *tun) reload(c *config.C, initial bool) error {
 		return nil
 	}
 
+	if initial {
+		t.unsafeIPv4Origin = prepareUnsafeOriginAddr(t, t.l, c, routes)
+	}
+
 	routeTree, err := makeRouteTree(t.l, routes, false)
 	if err != nil {
 		return err
@@ -545,6 +557,18 @@ func (t *tun) Networks() []netip.Prefix {
 	return t.vpnNetworks
 }
 
+func (t *tun) UnsafeNetworks() []netip.Prefix {
+	return t.unsafeNetworks
+}
+
+func (t *tun) UnsafeIPv4OriginAddress() netip.Prefix {
+	return t.unsafeIPv4Origin
+}
+
+func (t *tun) SNATAddress() netip.Prefix {
+	return netip.Prefix{}
+}
+
 func (t *tun) Name() string {
 	return t.Device
 }
@@ -553,32 +577,6 @@ func (t *tun) SupportsMultiqueue() bool {
 	return false
 }
 
-func (t *tun) NewMultiQueueReader() (BatchReadWriter, error) {
+func (t *tun) NewMultiQueueReader() (io.ReadWriteCloser, error) {
 	return nil, fmt.Errorf("TODO: multiqueue not implemented for darwin")
 }
-
-// BatchRead reads a single packet (batch size 1 for non-Linux platforms)
-func (t *tun) BatchRead(bufs [][]byte, sizes []int) (int, error) {
-	n, err := t.Read(bufs[0])
-	if err != nil {
-		return 0, err
-	}
-	sizes[0] = n
-	return 1, nil
-}
-
-// WriteBatch writes packets individually (no batching for non-Linux platforms)
-func (t *tun) WriteBatch(bufs [][]byte, offset int) (int, error) {
-	for i, buf := range bufs {
-		_, err := t.Write(buf[offset:])
-		if err != nil {
-			return i, err
-		}
-	}
-	return len(bufs), nil
-}
-
-// BatchSize returns 1 for non-Linux platforms (no batching)
-func (t *tun) BatchSize() int {
-	return 1
-}

overlay/tun_disabled.go

@@ -52,6 +52,17 @@ func (t *disabledTun) Networks() []netip.Prefix {
 	return t.vpnNetworks
 }
 
+func (*disabledTun) UnsafeNetworks() []netip.Prefix {
+	return nil
+}
+func (*disabledTun) SNATAddress() netip.Prefix {
+	return netip.Prefix{}
+}
+
+func (*disabledTun) UnsafeIPv4OriginAddress() netip.Prefix {
+	return netip.Prefix{}
+}
+
 func (*disabledTun) Name() string {
 	return "disabled"
 }
@@ -109,36 +120,10 @@ func (t *disabledTun) SupportsMultiqueue() bool {
 	return true
 }
 
-func (t *disabledTun) NewMultiQueueReader() (BatchReadWriter, error) {
+func (t *disabledTun) NewMultiQueueReader() (io.ReadWriteCloser, error) {
 	return t, nil
 }
 
-// BatchRead reads a single packet (batch size 1 for disabled tun)
-func (t *disabledTun) BatchRead(bufs [][]byte, sizes []int) (int, error) {
-	n, err := t.Read(bufs[0])
-	if err != nil {
-		return 0, err
-	}
-	sizes[0] = n
-	return 1, nil
-}
-
-// WriteBatch writes packets individually (no batching for disabled tun)
-func (t *disabledTun) WriteBatch(bufs [][]byte, offset int) (int, error) {
-	for i, buf := range bufs {
-		_, err := t.Write(buf[offset:])
-		if err != nil {
-			return i, err
-		}
-	}
-	return len(bufs), nil
-}
-
-// BatchSize returns 1 for disabled tun (no batching)
-func (t *disabledTun) BatchSize() int {
-	return 1
-}
-
 func (t *disabledTun) Close() error {
 	if t.read != nil {
 		close(t.read)

overlay/tun_freebsd.go

@@ -7,6 +7,7 @@ import (
 	"bytes"
 	"errors"
 	"fmt"
+	"io"
 	"io/fs"
 	"net/netip"
 	"sync/atomic"
@@ -85,14 +86,16 @@ type ifreqAlias6 struct {
 }
 
 type tun struct {
-	Device      string
+	Device           string
-	vpnNetworks []netip.Prefix
+	vpnNetworks      []netip.Prefix
-	MTU         int
+	unsafeNetworks   []netip.Prefix
-	Routes      atomic.Pointer[[]Route]
+	unsafeIPv4Origin netip.Prefix
-	routeTree   atomic.Pointer[bart.Table[routing.Gateways]]
+	MTU              int
-	linkAddr    *netroute.LinkAddr
+	Routes           atomic.Pointer[[]Route]
-	l           *logrus.Logger
+	routeTree        atomic.Pointer[bart.Table[routing.Gateways]]
-	devFd       int
+	linkAddr         *netroute.LinkAddr
+	l                *logrus.Logger
+	devFd            int
 }
 
 func (t *tun) Read(to []byte) (int, error) {
@@ -198,11 +201,11 @@ func (t *tun) Close() error {
 	return nil
 }
 
-func newTunFromFd(_ *config.C, _ *logrus.Logger, _ int, _ []netip.Prefix) (*tun, error) {
+func newTunFromFd(_ *config.C, _ *logrus.Logger, _ int, _ []netip.Prefix, _ []netip.Prefix) (*tun, error) {
 	return nil, fmt.Errorf("newTunFromFd not supported in FreeBSD")
 }
 
-func newTun(c *config.C, l *logrus.Logger, vpnNetworks []netip.Prefix, _ bool) (*tun, error) {
+func newTun(c *config.C, l *logrus.Logger, vpnNetworks []netip.Prefix, unsafeNetworks []netip.Prefix, _ bool) (*tun, error) {
 	// Try to open existing tun device
 	var fd int
 	var err error
@@ -269,11 +272,12 @@ func newTun(c *config.C, l *logrus.Logger, vpnNetworks []netip.Prefix, _ bool) (
 	}
 
 	t := &tun{
-		Device:      deviceName,
+		Device:         deviceName,
-		vpnNetworks: vpnNetworks,
+		vpnNetworks:    vpnNetworks,
-		MTU:         c.GetInt("tun.mtu", DefaultMTU),
+		unsafeNetworks: unsafeNetworks,
-		l:           l,
+		MTU:            c.GetInt("tun.mtu", DefaultMTU),
-		devFd:       fd,
+		l:              l,
+		devFd:          fd,
 	}
 
 	err = t.reload(c, true)
@@ -409,6 +413,10 @@ func (t *tun) reload(c *config.C, initial bool) error {
 		return nil
 	}
 
+	if initial {
+		t.unsafeIPv4Origin = prepareUnsafeOriginAddr(t, t.l, c, routes)
+	}
+
 	routeTree, err := makeRouteTree(t.l, routes, false)
 	if err != nil {
 		return err
@@ -445,6 +453,18 @@ func (t *tun) Networks() []netip.Prefix {
 	return t.vpnNetworks
 }
 
+func (t *tun) UnsafeNetworks() []netip.Prefix {
+	return t.unsafeNetworks
+}
+
+func (t *tun) UnsafeIPv4OriginAddress() netip.Prefix {
+	return t.unsafeIPv4Origin
+}
+
+func (t *tun) SNATAddress() netip.Prefix {
+	return netip.Prefix{}
+}
+
 func (t *tun) Name() string {
 	return t.Device
 }
@@ -453,36 +473,10 @@ func (t *tun) SupportsMultiqueue() bool {
 	return false
 }
 
-func (t *tun) NewMultiQueueReader() (BatchReadWriter, error) {
+func (t *tun) NewMultiQueueReader() (io.ReadWriteCloser, error) {
 	return nil, fmt.Errorf("TODO: multiqueue not implemented for freebsd")
 }
 
-// BatchRead reads a single packet (batch size 1 for FreeBSD)
-func (t *tun) BatchRead(bufs [][]byte, sizes []int) (int, error) {
-	n, err := t.Read(bufs[0])
-	if err != nil {
-		return 0, err
-	}
-	sizes[0] = n
-	return 1, nil
-}
-
-// WriteBatch writes packets individually (no batching for FreeBSD)
-func (t *tun) WriteBatch(bufs [][]byte, offset int) (int, error) {
-	for i, buf := range bufs {
-		_, err := t.Write(buf[offset:])
-		if err != nil {
-			return i, err
-		}
-	}
-	return len(bufs), nil
-}
-
-// BatchSize returns 1 for FreeBSD (no batching)
-func (t *tun) BatchSize() int {
-	return 1
-}
-
 func (t *tun) addRoutes(logErrors bool) error {
 	routes := *t.Routes.Load()
 	for _, r := range routes {

overlay/tun_ios.go

@@ -22,20 +22,23 @@ import (
 
 type tun struct {
 	io.ReadWriteCloser
-	vpnNetworks []netip.Prefix
+	vpnNetworks      []netip.Prefix
-	Routes      atomic.Pointer[[]Route]
+	unsafeNetworks   []netip.Prefix
-	routeTree   atomic.Pointer[bart.Table[routing.Gateways]]
+	unsafeIPv4Origin netip.Prefix
-	l           *logrus.Logger
+	Routes           atomic.Pointer[[]Route]
+	routeTree        atomic.Pointer[bart.Table[routing.Gateways]]
+	l                *logrus.Logger
 }
 
-func newTun(_ *config.C, _ *logrus.Logger, _ []netip.Prefix, _ bool) (*tun, error) {
+func newTun(_ *config.C, _ *logrus.Logger, _ []netip.Prefix, _ []netip.Prefix, _ bool) (*tun, error) {
 	return nil, fmt.Errorf("newTun not supported in iOS")
 }
 
-func newTunFromFd(c *config.C, l *logrus.Logger, deviceFd int, vpnNetworks []netip.Prefix) (*tun, error) {
+func newTunFromFd(c *config.C, l *logrus.Logger, deviceFd int, vpnNetworks []netip.Prefix, unsafeNetworks []netip.Prefix) (*tun, error) {
 	file := os.NewFile(uintptr(deviceFd), "/dev/tun")
 	t := &tun{
 		vpnNetworks:     vpnNetworks,
+		unsafeNetworks:  unsafeNetworks,
 		ReadWriteCloser: &tunReadCloser{f: file},
 		l:               l,
 	}
@@ -69,6 +72,8 @@ func (t *tun) reload(c *config.C, initial bool) error {
 		return nil
 	}
 
+	t.unsafeIPv4Origin = prepareUnsafeOriginAddr(t, t.l, c, routes)
+
 	routeTree, err := makeRouteTree(t.l, routes, false)
 	if err != nil {
 		return err
@@ -147,6 +152,18 @@ func (t *tun) Networks() []netip.Prefix {
 	return t.vpnNetworks
 }
 
+func (t *tun) UnsafeNetworks() []netip.Prefix {
+	return t.unsafeNetworks
+}
+
+func (t *tun) UnsafeIPv4OriginAddress() netip.Prefix {
+	return t.unsafeIPv4Origin
+}
+
+func (t *tun) SNATAddress() netip.Prefix {
+	return netip.Prefix{}
+}
+
 func (t *tun) Name() string {
 	return "iOS"
 }
@@ -155,29 +172,6 @@ func (t *tun) SupportsMultiqueue() bool {
 	return false
 }
 
-func (t *tun) NewMultiQueueReader() (BatchReadWriter, error) {
+func (t *tun) NewMultiQueueReader() (io.ReadWriteCloser, error) {
 	return nil, fmt.Errorf("TODO: multiqueue not implemented for ios")
 }
-
-func (t *tun) BatchRead(bufs [][]byte, sizes []int) (int, error) {
-	n, err := t.Read(bufs[0])
-	if err != nil {
-		return 0, err
-	}
-	sizes[0] = n
-	return 1, nil
-}
-
-func (t *tun) WriteBatch(bufs [][]byte, offset int) (int, error) {
-	for i, buf := range bufs {
-		_, err := t.Write(buf[offset:])
-		if err != nil {
-			return i, err
-		}
-	}
-	return len(bufs), nil
-}
-
-func (t *tun) BatchSize() int {
-	return 1
-}

overlay/tun_linux.go

@@ -9,6 +9,7 @@ import (
 	"net"
 	"net/netip"
 	"os"
+	"strings"
 	"sync"
 	"sync/atomic"
 	"time"
@@ -21,20 +22,19 @@ import (
 	"github.com/slackhq/nebula/util"
 	"github.com/vishvananda/netlink"
 	"golang.org/x/sys/unix"
-	wgtun "golang.zx2c4.com/wireguard/tun"
 )
 
 type tun struct {
 	io.ReadWriteCloser
-	wgDevice    wgtun.Device
+	fd             int
-	fd          int
+	Device         string
-	Device      string
+	vpnNetworks    []netip.Prefix
-	vpnNetworks []netip.Prefix
+	unsafeNetworks []netip.Prefix
-	MaxMTU      int
+	MaxMTU         int
-	DefaultMTU  int
+	DefaultMTU     int
-	TXQueueLen  int
+	TXQueueLen     int
-	deviceIndex int
+	deviceIndex    int
-	ioctlFd     uintptr
+	ioctlFd        uintptr
 
 	Routes                    atomic.Pointer[[]Route]
 	routeTree                 atomic.Pointer[bart.Table[routing.Gateways]]
@@ -47,6 +47,9 @@ type tun struct {
 	routesFromSystem     map[netip.Prefix]routing.Gateways
 	routesFromSystemLock sync.Mutex
 
+	snatAddr         netip.Prefix
+	unsafeIPv4Origin netip.Prefix
+
 	l *logrus.Logger
 }
 
@@ -54,6 +57,18 @@ func (t *tun) Networks() []netip.Prefix {
 	return t.vpnNetworks
 }
 
+func (t *tun) UnsafeNetworks() []netip.Prefix {
+	return t.unsafeNetworks
+}
+
+func (t *tun) UnsafeIPv4OriginAddress() netip.Prefix {
+	return t.unsafeIPv4Origin
+}
+
+func (t *tun) SNATAddress() netip.Prefix {
+	return t.snatAddr
+}
+
 type ifReq struct {
 	Name  [16]byte
 	Flags uint16
@@ -72,164 +87,74 @@ type ifreqQLEN struct {
 	pad   [8]byte
 }
 
-// wgDeviceWrapper wraps a wireguard Device to implement io.ReadWriteCloser
+func newTunFromFd(c *config.C, l *logrus.Logger, deviceFd int, vpnNetworks []netip.Prefix, unsafeNetworks []netip.Prefix) (*tun, error) {
-// This allows multiqueue readers to use the same wireguard Device batching as the main device
+	file := os.NewFile(uintptr(deviceFd), "/dev/net/tun")
-type wgDeviceWrapper struct {
-	dev wgtun.Device
-	buf []byte // Reusable buffer for single packet reads
-}
 
-func (w *wgDeviceWrapper) Read(b []byte) (int, error) {
+	t, err := newTunGeneric(c, l, file, vpnNetworks, unsafeNetworks)
-	// Use wireguard Device's batch API for single packet
-	bufs := [][]byte{b}
-	sizes := make([]int, 1)
-	n, err := w.dev.Read(bufs, sizes, 0)
 	if err != nil {
-		return 0, err
-	}
-	if n == 0 {
-		return 0, io.EOF
-	}
-	return sizes[0], nil
-}
-
-func (w *wgDeviceWrapper) Write(b []byte) (int, error) {
-	// Buffer b should have virtio header space (10 bytes) at the beginning
-	// The decrypted packet data starts at offset 10
-	// Pass the full buffer to WireGuard with offset=virtioNetHdrLen
-	bufs := [][]byte{b}
-	n, err := w.dev.Write(bufs, VirtioNetHdrLen)
-	if err != nil {
-		return 0, err
-	}
-	if n == 0 {
-		return 0, io.ErrShortWrite
-	}
-	return len(b), nil
-}
-
-func (w *wgDeviceWrapper) WriteBatch(bufs [][]byte, offset int) (int, error) {
-	// Pass all buffers to WireGuard's batch write
-	return w.dev.Write(bufs, offset)
-}
-
-func (w *wgDeviceWrapper) Close() error {
-	return w.dev.Close()
-}
-
-// BatchRead implements batching for multiqueue readers
-func (w *wgDeviceWrapper) BatchRead(bufs [][]byte, sizes []int) (int, error) {
-	// The zero here is offset.
-	return w.dev.Read(bufs, sizes, 0)
-}
-
-// BatchSize returns the optimal batch size
-func (w *wgDeviceWrapper) BatchSize() int {
-	return w.dev.BatchSize()
-}
-
-func newTunFromFd(c *config.C, l *logrus.Logger, deviceFd int, vpnNetworks []netip.Prefix) (*tun, error) {
-	wgDev, name, err := wgtun.CreateUnmonitoredTUNFromFD(deviceFd)
-	if err != nil {
-		return nil, fmt.Errorf("failed to create TUN from FD: %w", err)
-	}
-
-	file := wgDev.File()
-	t, err := newTunGeneric(c, l, file, vpnNetworks)
-	if err != nil {
-		_ = wgDev.Close()
 		return nil, err
 	}
 
-	t.wgDevice = wgDev
+	t.Device = "tun0"
-	t.Device = name
 
 	return t, nil
 }
 
-func newTun(c *config.C, l *logrus.Logger, vpnNetworks []netip.Prefix, multiqueue bool) (*tun, error) {
+func newTun(c *config.C, l *logrus.Logger, vpnNetworks []netip.Prefix, unsafeNetworks []netip.Prefix, multiqueue bool) (*tun, error) {
-	// Check if /dev/net/tun exists, create if needed (for docker containers)
-	if _, err := os.Stat("/dev/net/tun"); os.IsNotExist(err) {
-		if err := os.MkdirAll("/dev/net", 0755); err != nil {
-			return nil, fmt.Errorf("/dev/net/tun doesn't exist, failed to mkdir -p /dev/net: %w", err)
-		}
-		if err := unix.Mknod("/dev/net/tun", unix.S_IFCHR|0600, int(unix.Mkdev(10, 200))); err != nil {
-			return nil, fmt.Errorf("failed to create /dev/net/tun: %w", err)
-		}
-	}
-
-	devName := c.GetString("tun.dev", "")
-	mtu := c.GetInt("tun.mtu", DefaultMTU)
-
-	// Create TUN device manually to support multiqueue
 	fd, err := unix.Open("/dev/net/tun", os.O_RDWR, 0)
 	if err != nil {
-		return nil, err
+		// If /dev/net/tun doesn't exist, try to create it (will happen in docker)
+		if os.IsNotExist(err) {
+			err = os.MkdirAll("/dev/net", 0755)
+			if err != nil {
+				return nil, fmt.Errorf("/dev/net/tun doesn't exist, failed to mkdir -p /dev/net: %w", err)
+			}
+			err = unix.Mknod("/dev/net/tun", unix.S_IFCHR|0600, int(unix.Mkdev(10, 200)))
+			if err != nil {
+				return nil, fmt.Errorf("failed to create /dev/net/tun: %w", err)
+			}
+
+			fd, err = unix.Open("/dev/net/tun", os.O_RDWR, 0)
+			if err != nil {
+				return nil, fmt.Errorf("created /dev/net/tun, but still failed: %w", err)
+			}
+		} else {
+			return nil, err
+		}
 	}
 
 	var req ifReq
-	req.Flags = uint16(unix.IFF_TUN | unix.IFF_NO_PI | unix.IFF_VNET_HDR)
+	req.Flags = uint16(unix.IFF_TUN | unix.IFF_NO_PI)
 	if multiqueue {
 		req.Flags |= unix.IFF_MULTI_QUEUE
 	}
-	copy(req.Name[:], devName)
+	nameStr := c.GetString("tun.dev", "")
+	copy(req.Name[:], nameStr)
 	if err = ioctl(uintptr(fd), uintptr(unix.TUNSETIFF), uintptr(unsafe.Pointer(&req))); err != nil {
-		unix.Close(fd)
+		return nil, &NameError{
-		return nil, err
+			Name:       nameStr,
-	}
+			Underlying: err,
-
+		}
-	// Set nonblocking
-	if err = unix.SetNonblock(fd, true); err != nil {
-		unix.Close(fd)
-		return nil, err
-	}
-
-	// Enable TCP and UDP offload (TSO/GRO) for performance
-	// This allows the kernel to handle segmentation/coalescing
-	const (
-		tunTCPOffloads = unix.TUN_F_CSUM | unix.TUN_F_TSO4 | unix.TUN_F_TSO6
-		tunUDPOffloads = unix.TUN_F_USO4 | unix.TUN_F_USO6
-	)
-	offloads := tunTCPOffloads | tunUDPOffloads
-	if err = unix.IoctlSetInt(fd, unix.TUNSETOFFLOAD, offloads); err != nil {
-		// Log warning but don't fail - offload is optional
-		l.WithError(err).Warn("Failed to enable TUN offload (TSO/GRO), performance may be reduced")
 	}
+	name := strings.Trim(string(req.Name[:]), "\x00")
 
 	file := os.NewFile(uintptr(fd), "/dev/net/tun")
-
+	t, err := newTunGeneric(c, l, file, vpnNetworks, unsafeNetworks)
-	// Create wireguard device from file descriptor
-	wgDev, err := wgtun.CreateTUNFromFile(file, mtu)
 	if err != nil {
-		file.Close()
-		return nil, fmt.Errorf("failed to create TUN from file: %w", err)
-	}
-
-	name, err := wgDev.Name()
-	if err != nil {
-		_ = wgDev.Close()
-		return nil, fmt.Errorf("failed to get TUN device name: %w", err)
-	}
-
-	// file is now owned by wgDev, get a new reference
-	file = wgDev.File()
-	t, err := newTunGeneric(c, l, file, vpnNetworks)
-	if err != nil {
-		_ = wgDev.Close()
 		return nil, err
 	}
 
-	t.wgDevice = wgDev
 	t.Device = name
 
 	return t, nil
 }
 
-func newTunGeneric(c *config.C, l *logrus.Logger, file *os.File, vpnNetworks []netip.Prefix) (*tun, error) {
+func newTunGeneric(c *config.C, l *logrus.Logger, file *os.File, vpnNetworks []netip.Prefix, unsafeNetworks []netip.Prefix) (*tun, error) {
 	t := &tun{
 		ReadWriteCloser:           file,
 		fd:                        int(file.Fd()),
 		vpnNetworks:               vpnNetworks,
+		unsafeNetworks:            unsafeNetworks,
 		TXQueueLen:                c.GetInt("tun.tx_queue", 500),
 		useSystemRoutes:           c.GetBool("tun.use_system_route_table", false),
 		useSystemRoutesBufferSize: c.GetInt("tun.use_system_route_table_buffer_size", 0),
@@ -262,6 +187,11 @@ func (t *tun) reload(c *config.C, initial bool) error {
 		return nil
 	}
 
+	if initial {
+		t.unsafeIPv4Origin = prepareUnsafeOriginAddr(t, t.l, c, routes) //todo MUST be different from t.snatAddr!
+		t.snatAddr = prepareSnatAddr(t, t.l, c)
+	}
+
 	routeTree, err := makeRouteTree(t.l, routes, true)
 	if err != nil {
 		return err
@@ -330,44 +260,22 @@ func (t *tun) SupportsMultiqueue() bool {
 	return true
 }
 
-func (t *tun) NewMultiQueueReader() (BatchReadWriter, error) {
+func (t *tun) NewMultiQueueReader() (io.ReadWriteCloser, error) {
 	fd, err := unix.Open("/dev/net/tun", os.O_RDWR, 0)
 	if err != nil {
 		return nil, err
 	}
 
 	var req ifReq
-	// MUST match the flags used in newTun - includes IFF_VNET_HDR
+	req.Flags = uint16(unix.IFF_TUN | unix.IFF_NO_PI | unix.IFF_MULTI_QUEUE)
-	req.Flags = uint16(unix.IFF_TUN | unix.IFF_NO_PI | unix.IFF_VNET_HDR | unix.IFF_MULTI_QUEUE)
 	copy(req.Name[:], t.Device)
 	if err = ioctl(uintptr(fd), uintptr(unix.TUNSETIFF), uintptr(unsafe.Pointer(&req))); err != nil {
-		unix.Close(fd)
 		return nil, err
 	}
 
-	// Set nonblocking mode - CRITICAL for proper netpoller integration
-	if err = unix.SetNonblock(fd, true); err != nil {
-		unix.Close(fd)
-		return nil, err
-	}
-
-	// Get MTU from main device
-	mtu := t.MaxMTU
-	if mtu == 0 {
-		mtu = DefaultMTU
-	}
-
 	file := os.NewFile(uintptr(fd), "/dev/net/tun")
 
-	// Create wireguard Device from the file descriptor (just like the main device)
+	return file, nil
-	wgDev, err := wgtun.CreateTUNFromFile(file, mtu)
-	if err != nil {
-		file.Close()
-		return nil, fmt.Errorf("failed to create multiqueue TUN device: %w", err)
-	}
-
-	// Return a wrapper that uses the wireguard Device for all I/O
-	return &wgDeviceWrapper{dev: wgDev}, nil
 }
 
 func (t *tun) RoutesFor(ip netip.Addr) routing.Gateways {
@@ -375,68 +283,7 @@ func (t *tun) RoutesFor(ip netip.Addr) routing.Gateways {
 	return r
 }
 
-func (t *tun) Read(b []byte) (int, error) {
-	if t.wgDevice != nil {
-		// Use wireguard device which handles virtio headers internally
-		bufs := [][]byte{b}
-		sizes := make([]int, 1)
-		n, err := t.wgDevice.Read(bufs, sizes, 0)
-		if err != nil {
-			return 0, err
-		}
-		if n == 0 {
-			return 0, io.EOF
-		}
-		return sizes[0], nil
-	}
-
-	// Fallback: direct read from file (shouldn't happen in normal operation)
-	return t.ReadWriteCloser.Read(b)
-}
-
-// BatchRead reads multiple packets at once for improved performance
-// bufs: slice of buffers to read into
-// sizes: slice that will be filled with packet sizes
-// Returns number of packets read
-func (t *tun) BatchRead(bufs [][]byte, sizes []int) (int, error) {
-	if t.wgDevice != nil {
-		return t.wgDevice.Read(bufs, sizes, 0)
-	}
-
-	// Fallback: single packet read
-	n, err := t.ReadWriteCloser.Read(bufs[0])
-	if err != nil {
-		return 0, err
-	}
-	sizes[0] = n
-	return 1, nil
-}
-
-// BatchSize returns the optimal number of packets to read/write in a batch
-func (t *tun) BatchSize() int {
-	if t.wgDevice != nil {
-		return t.wgDevice.BatchSize()
-	}
-	return 1
-}
-
 func (t *tun) Write(b []byte) (int, error) {
-	if t.wgDevice != nil {
-		// Buffer b should have virtio header space (10 bytes) at the beginning
-		// The decrypted packet data starts at offset 10
-		// Pass the full buffer to WireGuard with offset=virtioNetHdrLen
-		bufs := [][]byte{b}
-		n, err := t.wgDevice.Write(bufs, VirtioNetHdrLen)
-		if err != nil {
-			return 0, err
-		}
-		if n == 0 {
-			return 0, io.ErrShortWrite
-		}
-		return len(b), nil
-	}
-
-	// Fallback: direct write (shouldn't happen in normal operation)
 	var nn int
 	maximum := len(b)
 
@@ -459,22 +306,6 @@ func (t *tun) Write(b []byte) (int, error) {
 	}
 }
 
-// WriteBatch writes multiple packets to the TUN device in a single syscall
-func (t *tun) WriteBatch(bufs [][]byte, offset int) (int, error) {
-	if t.wgDevice != nil {
-		return t.wgDevice.Write(bufs, offset)
-	}
-
-	// Fallback: write individually (shouldn't happen in normal operation)
-	for i, buf := range bufs {
-		_, err := t.Write(buf)
-		if err != nil {
-			return i, err
-		}
-	}
-	return len(bufs), nil
-}
-
 func (t *tun) deviceBytes() (o [16]byte) {
 	for i, c := range t.Device {
 		o[i] = byte(c)
@@ -504,6 +335,17 @@ func (t *tun) addIPs(link netlink.Link) error {
 		}
 	}
 
+	if t.unsafeIPv4Origin.IsValid() {
+		newAddrs = append(newAddrs, &netlink.Addr{
+			IPNet: &net.IPNet{
+				IP:   t.unsafeIPv4Origin.Addr().AsSlice(),
+				Mask: net.CIDRMask(t.unsafeIPv4Origin.Bits(), t.unsafeIPv4Origin.Addr().BitLen()),
+			},
+			Label: t.unsafeIPv4Origin.Addr().Zone(),
+		})
+		t.l.WithField("address", t.unsafeIPv4Origin).Info("Adding origin address for IPv4 unsafe_routes")
+	}
+
 	//add all new addresses
 	for i := range newAddrs {
 		//AddrReplace still adds new IPs, but if their properties change it will change them as well
@@ -591,7 +433,13 @@ func (t *tun) Activate() error {
 	//set route MTU
 	for i := range t.vpnNetworks {
 		if err = t.setDefaultRoute(t.vpnNetworks[i]); err != nil {
-			return fmt.Errorf("failed to set default route MTU: %w", err)
+			return fmt.Errorf("failed to set default route MTU for %s: %w", t.vpnNetworks[i], err)
+		}
+	}
+
+	if t.unsafeIPv4Origin.IsValid() {
+		if err = t.setDefaultRoute(t.unsafeIPv4Origin); err != nil {
+			return fmt.Errorf("failed to set default route MTU for %s: %w", t.unsafeIPv4Origin, err)
 		}
 	}
 
@@ -618,6 +466,23 @@ func (t *tun) setMTU() {
 	}
 }
 
+func (t *tun) setSnatRoute() error {
+	dr := &net.IPNet{
+		IP:   t.snatAddr.Masked().Addr().AsSlice(),
+		Mask: net.CIDRMask(t.snatAddr.Bits(), t.snatAddr.Addr().BitLen()),
+	}
+
+	nr := netlink.Route{
+		LinkIndex: t.deviceIndex,
+		Dst:       dr,
+		Scope:     unix.RT_SCOPE_LINK,
+		//Protocol: unix.RTPROT_KERNEL,
+		Table: unix.RT_TABLE_MAIN,
+		Type:  unix.RTN_UNICAST,
+	}
+	return netlink.RouteReplace(&nr)
+}
+
 func (t *tun) setDefaultRoute(cidr netip.Prefix) error {
 	dr := &net.IPNet{
 		IP:   cidr.Masked().Addr().AsSlice(),
@@ -694,6 +559,13 @@ func (t *tun) addRoutes(logErrors bool) error {
 		}
 	}
 
+	if t.snatAddr.IsValid() {
+		//at least for Linux, we need to set a return route for the SNATted traffic in order to satisfy the reverse-path filter,
+		//and to help the kernel deliver our reply traffic to the tun device.
+		//however, it is important that we do not actually /assign/ the SNAT address,
+		//since link-local addresses will not be routed between interfaces without significant trickery.
+		return t.setSnatRoute()
+	}
 	return nil
 }
 
@@ -902,10 +774,6 @@ func (t *tun) Close() error {
 		close(t.routeChan)
 	}
 
-	if t.wgDevice != nil {
-		_ = t.wgDevice.Close()
-	}
-
 	if t.ReadWriteCloser != nil {
 		_ = t.ReadWriteCloser.Close()
 	}

overlay/tun_netbsd.go

@@ -6,6 +6,7 @@ package overlay
 import (
 	"errors"
 	"fmt"
+	"io"
 	"net/netip"
 	"os"
 	"regexp"
@@ -57,23 +58,25 @@ type addrLifetime struct {
 }
 
 type tun struct {
-	Device      string
+	Device           string
-	vpnNetworks []netip.Prefix
+	vpnNetworks      []netip.Prefix
-	MTU         int
+	unsafeNetworks   []netip.Prefix
-	Routes      atomic.Pointer[[]Route]
+	unsafeIPv4Origin netip.Prefix
-	routeTree   atomic.Pointer[bart.Table[routing.Gateways]]
+	MTU              int
-	l           *logrus.Logger
+	Routes           atomic.Pointer[[]Route]
-	f           *os.File
+	routeTree        atomic.Pointer[bart.Table[routing.Gateways]]
-	fd          int
+	l                *logrus.Logger
+	f                *os.File
+	fd               int
 }
 
 var deviceNameRE = regexp.MustCompile(`^tun[0-9]+$`)
 
-func newTunFromFd(_ *config.C, _ *logrus.Logger, _ int, _ []netip.Prefix) (*tun, error) {
+func newTunFromFd(_ *config.C, _ *logrus.Logger, _ int, _ []netip.Prefix, _ []netip.Prefix) (*tun, error) {
 	return nil, fmt.Errorf("newTunFromFd not supported in NetBSD")
 }
 
-func newTun(c *config.C, l *logrus.Logger, vpnNetworks []netip.Prefix, _ bool) (*tun, error) {
+func newTun(c *config.C, l *logrus.Logger, vpnNetworks []netip.Prefix, unsafeNetworks []netip.Prefix, _ bool) (*tun, error) {
 	// Try to open tun device
 	var err error
 	deviceName := c.GetString("tun.dev", "")
@@ -349,6 +352,10 @@ func (t *tun) reload(c *config.C, initial bool) error {
 		return nil
 	}
 
+	if initial {
+		t.unsafeIPv4Origin = prepareUnsafeOriginAddr(t, t.l, c, routes)
+	}
+
 	routeTree, err := makeRouteTree(t.l, routes, false)
 	if err != nil {
 		return err
@@ -385,6 +392,18 @@ func (t *tun) Networks() []netip.Prefix {
 	return t.vpnNetworks
 }
 
+func (t *tun) UnsafeNetworks() []netip.Prefix {
+	return t.unsafeNetworks
+}
+
+func (t *tun) UnsafeIPv4OriginAddress() netip.Prefix {
+	return t.unsafeIPv4Origin
+}
+
+func (t *tun) SNATAddress() netip.Prefix {
+	return netip.Prefix{}
+}
+
 func (t *tun) Name() string {
 	return t.Device
 }
@@ -393,33 +412,10 @@ func (t *tun) SupportsMultiqueue() bool {
 	return false
 }
 
-func (t *tun) NewMultiQueueReader() (BatchReadWriter, error) {
+func (t *tun) NewMultiQueueReader() (io.ReadWriteCloser, error) {
 	return nil, fmt.Errorf("TODO: multiqueue not implemented for netbsd")
 }
 
-func (t *tun) BatchRead(bufs [][]byte, sizes []int) (int, error) {
-	n, err := t.Read(bufs[0])
-	if err != nil {
-		return 0, err
-	}
-	sizes[0] = n
-	return 1, nil
-}
-
-func (t *tun) WriteBatch(bufs [][]byte, offset int) (int, error) {
-	for i, buf := range bufs {
-		_, err := t.Write(buf[offset:])
-		if err != nil {
-			return i, err
-		}
-	}
-	return len(bufs), nil
-}
-
-func (t *tun) BatchSize() int {
-	return 1
-}
-
 func (t *tun) addRoutes(logErrors bool) error {
 	routes := *t.Routes.Load()
 

overlay/tun_openbsd.go

@@ -6,6 +6,7 @@ package overlay
 import (
 	"errors"
 	"fmt"
+	"io"
 	"net/netip"
 	"os"
 	"regexp"
@@ -48,25 +49,27 @@ type ifreq struct {
 }
 
 type tun struct {
-	Device      string
+	Device           string
-	vpnNetworks []netip.Prefix
+	vpnNetworks      []netip.Prefix
-	MTU         int
+	unsafeNetworks   []netip.Prefix
-	Routes      atomic.Pointer[[]Route]
+	unsafeIPv4Origin netip.Prefix
-	routeTree   atomic.Pointer[bart.Table[routing.Gateways]]
+	MTU              int
-	l           *logrus.Logger
+	Routes           atomic.Pointer[[]Route]
-	f           *os.File
+	routeTree        atomic.Pointer[bart.Table[routing.Gateways]]
-	fd          int
+	l                *logrus.Logger
+	f                *os.File
+	fd               int
 	// cache out buffer since we need to prepend 4 bytes for tun metadata
 	out []byte
 }
 
 var deviceNameRE = regexp.MustCompile(`^tun[0-9]+$`)
 
-func newTunFromFd(_ *config.C, _ *logrus.Logger, _ int, _ []netip.Prefix) (*tun, error) {
+func newTunFromFd(_ *config.C, _ *logrus.Logger, _ int, _ []netip.Prefix, _ []netip.Prefix) (*tun, error) {
 	return nil, fmt.Errorf("newTunFromFd not supported in openbsd")
 }
 
-func newTun(c *config.C, l *logrus.Logger, vpnNetworks []netip.Prefix, _ bool) (*tun, error) {
+func newTun(c *config.C, l *logrus.Logger, vpnNetworks []netip.Prefix, unsafeNetworks []netip.Prefix, _ bool) (*tun, error) {
 	// Try to open tun device
 	var err error
 	deviceName := c.GetString("tun.dev", "")
@@ -88,12 +91,13 @@ func newTun(c *config.C, l *logrus.Logger, vpnNetworks []netip.Prefix, _ bool) (
 	}
 
 	t := &tun{
-		f:           os.NewFile(uintptr(fd), ""),
+		f:              os.NewFile(uintptr(fd), ""),
-		fd:          fd,
+		fd:             fd,
-		Device:      deviceName,
+		Device:         deviceName,
-		vpnNetworks: vpnNetworks,
+		vpnNetworks:    vpnNetworks,
-		MTU:         c.GetInt("tun.mtu", DefaultMTU),
+		unsafeNetworks: unsafeNetworks,
-		l:           l,
+		MTU:            c.GetInt("tun.mtu", DefaultMTU),
+		l:              l,
 	}
 
 	err = t.reload(c, true)
@@ -269,6 +273,10 @@ func (t *tun) reload(c *config.C, initial bool) error {
 		return nil
 	}
 
+	if initial {
+		t.unsafeIPv4Origin = prepareUnsafeOriginAddr(t, t.l, c, routes)
+	}
+
 	routeTree, err := makeRouteTree(t.l, routes, false)
 	if err != nil {
 		return err
@@ -305,6 +313,18 @@ func (t *tun) Networks() []netip.Prefix {
 	return t.vpnNetworks
 }
 
+func (t *tun) UnsafeNetworks() []netip.Prefix {
+	return t.unsafeNetworks
+}
+
+func (t *tun) UnsafeIPv4OriginAddress() netip.Prefix {
+	return t.unsafeIPv4Origin
+}
+
+func (t *tun) SNATAddress() netip.Prefix {
+	return netip.Prefix{}
+}
+
 func (t *tun) Name() string {
 	return t.Device
 }
@@ -313,33 +333,10 @@ func (t *tun) SupportsMultiqueue() bool {
 	return false
 }
 
-func (t *tun) NewMultiQueueReader() (BatchReadWriter, error) {
+func (t *tun) NewMultiQueueReader() (io.ReadWriteCloser, error) {
 	return nil, fmt.Errorf("TODO: multiqueue not implemented for openbsd")
 }
 
-func (t *tun) BatchRead(bufs [][]byte, sizes []int) (int, error) {
-	n, err := t.Read(bufs[0])
-	if err != nil {
-		return 0, err
-	}
-	sizes[0] = n
-	return 1, nil
-}
-
-func (t *tun) WriteBatch(bufs [][]byte, offset int) (int, error) {
-	for i, buf := range bufs {
-		_, err := t.Write(buf[offset:])
-		if err != nil {
-			return i, err
-		}
-	}
-	return len(bufs), nil
-}
-
-func (t *tun) BatchSize() int {
-	return 1
-}
-
 func (t *tun) addRoutes(logErrors bool) error {
 	routes := *t.Routes.Load()
 

overlay/tun_snat_test.go

@@ -0,0 +1,179 @@
+package overlay
+
+import (
+	"io"
+	"net/netip"
+	"testing"
+
+	"github.com/sirupsen/logrus"
+	"github.com/slackhq/nebula/config"
+	"github.com/slackhq/nebula/routing"
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+)
+
+// mockDevice is a minimal Device implementation for testing prepareUnsafeOriginAddr.
+type mockDevice struct {
+	networks       []netip.Prefix
+	unsafeNetworks []netip.Prefix
+	snatAddr       netip.Prefix
+	unsafeSnatAddr netip.Prefix
+}
+
+func (d *mockDevice) Read([]byte) (int, error)                         { return 0, nil }
+func (d *mockDevice) Write([]byte) (int, error)                        { return 0, nil }
+func (d *mockDevice) Close() error                                     { return nil }
+func (d *mockDevice) Activate() error                                  { return nil }
+func (d *mockDevice) Networks() []netip.Prefix                         { return d.networks }
+func (d *mockDevice) UnsafeNetworks() []netip.Prefix                   { return d.unsafeNetworks }
+func (d *mockDevice) SNATAddress() netip.Prefix                        { return d.snatAddr }
+func (d *mockDevice) UnsafeIPv4OriginAddress() netip.Prefix            { return d.unsafeSnatAddr }
+func (d *mockDevice) Name() string                                     { return "mock" }
+func (d *mockDevice) RoutesFor(netip.Addr) routing.Gateways            { return routing.Gateways{} }
+func (d *mockDevice) SupportsMultiqueue() bool                         { return false }
+func (d *mockDevice) NewMultiQueueReader() (io.ReadWriteCloser, error) { return nil, nil }
+
+func TestPrepareSnatAddr_V4Primary_NoSnat(t *testing.T) {
+	l := logrus.New()
+	l.SetLevel(logrus.PanicLevel)
+	c := config.NewC(l)
+
+	// If the device has an IPv4 primary address, no SNAT needed
+	d := &mockDevice{
+		networks: []netip.Prefix{netip.MustParsePrefix("10.0.0.1/24")},
+	}
+	result := prepareUnsafeOriginAddr(d, l, c, nil)
+	assert.Equal(t, netip.Prefix{}, result, "should not assign SNAT addr when device has IPv4 primary")
+}
+
+func TestPrepareSnatAddr_V6Primary_NoUnsafeOrRoutes(t *testing.T) {
+	l := logrus.New()
+	l.SetLevel(logrus.PanicLevel)
+	c := config.NewC(l)
+
+	// IPv6 primary but no unsafe networks or IPv4 routes
+	d := &mockDevice{
+		networks: []netip.Prefix{netip.MustParsePrefix("fd00::1/128")},
+	}
+	result := prepareUnsafeOriginAddr(d, l, c, nil)
+	assert.Equal(t, netip.Prefix{}, result, "should not assign SNAT addr without IPv4 unsafe networks or routes")
+}
+
+func TestPrepareSnatAddr_V6Primary_WithV4Unsafe(t *testing.T) {
+	l := logrus.New()
+	l.SetLevel(logrus.PanicLevel)
+	c := config.NewC(l)
+
+	// IPv6 primary with IPv4 unsafe network -> should get SNAT addr
+	d := &mockDevice{
+		networks:       []netip.Prefix{netip.MustParsePrefix("fd00::1/128")},
+		unsafeNetworks: []netip.Prefix{netip.MustParsePrefix("192.168.0.0/16")},
+	}
+	result := prepareSnatAddr(d, l, c)
+	require.True(t, result.IsValid(), "should assign SNAT addr")
+	assert.True(t, result.Addr().Is4(), "SNAT addr should be IPv4")
+	assert.True(t, result.Addr().IsLinkLocalUnicast(), "SNAT addr should be link-local")
+	assert.Equal(t, 32, result.Bits(), "SNAT addr should be /32")
+
+	result = prepareUnsafeOriginAddr(d, l, c, nil)
+	require.False(t, result.IsValid(), "no routes = no origin addr needed")
+}
+
+func TestPrepareUnsafeOriginAddr_V6Primary_WithV4Route(t *testing.T) {
+	l := logrus.New()
+	l.SetLevel(logrus.PanicLevel)
+	c := config.NewC(l)
+
+	// IPv6 primary with IPv4 route -> should get SNAT addr
+	d := &mockDevice{
+		networks: []netip.Prefix{netip.MustParsePrefix("fd00::1/128")},
+	}
+	routes := []Route{
+		{Cidr: netip.MustParsePrefix("10.0.0.0/8")},
+	}
+	result := prepareUnsafeOriginAddr(d, l, c, routes)
+	require.True(t, result.IsValid(), "should assign SNAT addr when IPv4 route exists")
+	assert.True(t, result.Addr().Is4())
+	assert.True(t, result.Addr().IsLinkLocalUnicast())
+
+	result = prepareSnatAddr(d, l, c)
+	require.False(t, result.IsValid(), "no UnsafeNetworks = no snat addr needed")
+}
+
+func TestPrepareSnatAddr_V6Primary_V6UnsafeOnly(t *testing.T) {
+	l := logrus.New()
+	l.SetLevel(logrus.PanicLevel)
+	c := config.NewC(l)
+
+	// IPv6 primary with only IPv6 unsafe network -> no SNAT needed
+	d := &mockDevice{
+		networks:       []netip.Prefix{netip.MustParsePrefix("fd00::1/128")},
+		unsafeNetworks: []netip.Prefix{netip.MustParsePrefix("fd01::/64")},
+	}
+	result := prepareUnsafeOriginAddr(d, l, c, nil)
+	assert.Equal(t, netip.Prefix{}, result, "should not assign SNAT addr for IPv6-only unsafe networks")
+}
+
+func TestPrepareSnatAddr_ManualAddress(t *testing.T) {
+	l := logrus.New()
+	l.SetLevel(logrus.PanicLevel)
+	c := config.NewC(l)
+	c.Settings["tun"] = map[string]any{
+		"snat_address_for_4over6": "169.254.42.42",
+	}
+
+	d := &mockDevice{
+		networks:       []netip.Prefix{netip.MustParsePrefix("fd00::1/128")},
+		unsafeNetworks: []netip.Prefix{netip.MustParsePrefix("192.168.0.0/16")},
+	}
+	result := prepareSnatAddr(d, l, c)
+	require.True(t, result.IsValid())
+	assert.Equal(t, netip.MustParseAddr("169.254.42.42"), result.Addr())
+	assert.Equal(t, 32, result.Bits())
+}
+
+func TestPrepareSnatAddr_InvalidManualAddress_Fallback(t *testing.T) {
+	l := logrus.New()
+	l.SetLevel(logrus.PanicLevel)
+	c := config.NewC(l)
+	c.Settings["tun"] = map[string]any{
+		"snat_address_for_4over6": "not-an-ip",
+	}
+
+	d := &mockDevice{
+		networks:       []netip.Prefix{netip.MustParsePrefix("fd00::1/128")},
+		unsafeNetworks: []netip.Prefix{netip.MustParsePrefix("192.168.0.0/16")},
+	}
+	result := prepareSnatAddr(d, l, c)
+	// Should fall back to auto-assignment
+	require.True(t, result.IsValid(), "should fall back to auto-assigned address")
+	assert.True(t, result.Addr().Is4())
+	assert.True(t, result.Addr().IsLinkLocalUnicast())
+}
+
+func TestPrepareSnatAddr_AutoGenerated_Range(t *testing.T) {
+	l := logrus.New()
+	l.SetLevel(logrus.PanicLevel)
+	c := config.NewC(l)
+
+	d := &mockDevice{
+		networks:       []netip.Prefix{netip.MustParsePrefix("fd00::1/128")},
+		unsafeNetworks: []netip.Prefix{netip.MustParsePrefix("192.168.0.0/16")},
+	}
+
+	// Generate several addresses and verify they're all in the expected range
+	for i := 0; i < 100; i++ {
+		result := prepareSnatAddr(d, l, c)
+		require.True(t, result.IsValid())
+		addr := result.Addr()
+		octets := addr.As4()
+		assert.Equal(t, byte(169), octets[0], "first octet should be 169")
+		assert.Equal(t, byte(254), octets[1], "second octet should be 254")
+		// Should not have .0 in the last octet
+		assert.NotEqual(t, byte(0), octets[3], "last octet should not be 0")
+		// Should not be 169.254.255.255 (broadcast)
+		if octets[2] == 255 {
+			assert.NotEqual(t, byte(255), octets[3], "should not be broadcast address")
+		}
+	}
+}

overlay/tun_test.go

@@ -0,0 +1,37 @@
+package overlay
+
+import (
+	"bytes"
+	"net/netip"
+	"testing"
+
+	"github.com/stretchr/testify/assert"
+)
+
+func TestLinkLocal(t *testing.T) {
+	r := bytes.NewReader([]byte{42, 99})
+	result := genLinkLocal(r)
+	assert.Equal(t, netip.MustParsePrefix("169.254.42.99/32"), result, "genLinkLocal with a deterministic randomizer")
+
+	result = genLinkLocal(nil)
+	assert.True(t, result.IsValid(), "genLinkLocal with nil randomizer should be valid")
+	assert.True(t, result.Addr().IsLinkLocalUnicast(), "genLinkLocal with nil randomizer should be link-local")
+
+	result = coerceLinkLocal([]byte{169, 254, 100, 50})
+	assert.Equal(t, netip.MustParsePrefix("169.254.100.50/32"), result, "coerceLinkLocal should pass through normal values")
+
+	result = coerceLinkLocal([]byte{169, 254, 0, 0})
+	assert.Equal(t, netip.MustParsePrefix("169.254.0.1/32"), result, "coerceLinkLocal should bump .0 last octet to .1")
+
+	result = coerceLinkLocal([]byte{169, 254, 255, 255})
+	assert.Equal(t, netip.MustParsePrefix("169.254.255.254/32"), result, "coerceLinkLocal should bump broadcast 255.255 to 255.254")
+
+	result = coerceLinkLocal([]byte{169, 254, 0, 1})
+	assert.Equal(t, netip.MustParsePrefix("169.254.0.1/32"), result, "coerceLinkLocal should leave .1 last octet unchanged")
+
+	result = coerceLinkLocal([]byte{169, 254, 255, 254})
+	assert.Equal(t, netip.MustParsePrefix("169.254.255.254/32"), result, "coerceLinkLocal should leave 255.254 unchanged")
+
+	result = coerceLinkLocal([]byte{169, 254, 255, 100})
+	assert.Equal(t, netip.MustParsePrefix("169.254.255.100/32"), result, "coerceLinkLocal should leave 255.100 unchanged")
+}

overlay/tun_tester.go

@@ -17,18 +17,21 @@ import (
 )
 
 type TestTun struct {
-	Device      string
+	Device           string
-	vpnNetworks []netip.Prefix
+	vpnNetworks      []netip.Prefix
-	Routes      []Route
+	unsafeNetworks   []netip.Prefix
-	routeTree   *bart.Table[routing.Gateways]
+	snatAddr         netip.Prefix
-	l           *logrus.Logger
+	unsafeIPv4Origin netip.Prefix
+	Routes           []Route
+	routeTree        *bart.Table[routing.Gateways]
+	l                *logrus.Logger
 
 	closed    atomic.Bool
 	rxPackets chan []byte // Packets to receive into nebula
 	TxPackets chan []byte // Packets transmitted outside by nebula
 }
 
-func newTun(c *config.C, l *logrus.Logger, vpnNetworks []netip.Prefix, _ bool) (*TestTun, error) {
+func newTun(c *config.C, l *logrus.Logger, vpnNetworks []netip.Prefix, unsafeNetworks []netip.Prefix, _ bool) (*TestTun, error) {
 	_, routes, err := getAllRoutesFromConfig(c, vpnNetworks, true)
 	if err != nil {
 		return nil, err
@@ -38,18 +41,22 @@ func newTun(c *config.C, l *logrus.Logger, vpnNetworks []netip.Prefix, _ bool) (
 		return nil, err
 	}
 
-	return &TestTun{
+	tt := &TestTun{
-		Device:      c.GetString("tun.dev", ""),
+		Device:         c.GetString("tun.dev", ""),
-		vpnNetworks: vpnNetworks,
+		vpnNetworks:    vpnNetworks,
-		Routes:      routes,
+		unsafeNetworks: unsafeNetworks,
-		routeTree:   routeTree,
+		Routes:         routes,
-		l:           l,
+		routeTree:      routeTree,
-		rxPackets:   make(chan []byte, 10),
+		l:              l,
-		TxPackets:   make(chan []byte, 10),
+		rxPackets:      make(chan []byte, 10),
-	}, nil
+		TxPackets:      make(chan []byte, 10),
+	}
+	tt.unsafeIPv4Origin = prepareUnsafeOriginAddr(tt, l, c, routes)
+	tt.snatAddr = prepareSnatAddr(tt, tt.l, c)
+	return tt, nil
 }
 
-func newTunFromFd(_ *config.C, _ *logrus.Logger, _ int, _ []netip.Prefix) (*TestTun, error) {
+func newTunFromFd(_ *config.C, _ *logrus.Logger, _ int, _ []netip.Prefix, _ []netip.Prefix) (*TestTun, error) {
 	return nil, fmt.Errorf("newTunFromFd not supported")
 }
 
@@ -109,12 +116,8 @@ func (t *TestTun) Write(b []byte) (n int, err error) {
 		return 0, io.ErrClosedPipe
 	}
 
-	// Skip virtio header (consistent with production Linux tun)
+	packet := make([]byte, len(b), len(b))
-	// The buffer b has VirtioNetHdrLen bytes of header followed by the actual packet
+	copy(packet, b)
-	data := b[VirtioNetHdrLen:]
-
-	packet := make([]byte, len(data))
-	copy(packet, data)
 	t.TxPackets <- packet
 	return len(b), nil
 }
@@ -140,34 +143,18 @@ func (t *TestTun) SupportsMultiqueue() bool {
 	return false
 }
 
-func (t *TestTun) NewMultiQueueReader() (BatchReadWriter, error) {
+func (t *TestTun) NewMultiQueueReader() (io.ReadWriteCloser, error) {
 	return nil, fmt.Errorf("TODO: multiqueue not implemented")
 }
 
-func (t *TestTun) BatchRead(bufs [][]byte, sizes []int) (int, error) {
+func (t *TestTun) UnsafeNetworks() []netip.Prefix {
-	n, err := t.Read(bufs[0])
+	return t.unsafeNetworks
-	if err != nil {
-		return 0, err
-	}
-	sizes[0] = n
-	return 1, nil
 }
 
-func (t *TestTun) WriteBatch(bufs [][]byte, offset int) (int, error) {
+func (t *TestTun) UnsafeIPv4OriginAddress() netip.Prefix {
-	if t.closed.Load() {
+	return t.unsafeIPv4Origin
-		return 0, io.ErrClosedPipe
-	}
-
-	for _, buf := range bufs {
-		// Strip the header at offset and send directly to channel
-		data := buf[offset:]
-		packet := make([]byte, len(data))
-		copy(packet, data)
-		t.TxPackets <- packet
-	}
-	return len(bufs), nil
 }
 
-func (t *TestTun) BatchSize() int {
+func (t *TestTun) SNATAddress() netip.Prefix {
-	return 1
+	return t.snatAddr
 }

overlay/tun_windows.go

@@ -6,6 +6,7 @@ package overlay
 import (
 	"crypto"
 	"fmt"
+	"io"
 	"net/netip"
 	"os"
 	"path/filepath"
@@ -27,21 +28,23 @@ import (
 const tunGUIDLabel = "Fixed Nebula Windows GUID v1"
 
 type winTun struct {
-	Device      string
+	Device           string
-	vpnNetworks []netip.Prefix
+	vpnNetworks      []netip.Prefix
-	MTU         int
+	unsafeNetworks   []netip.Prefix
-	Routes      atomic.Pointer[[]Route]
+	unsafeIPv4Origin netip.Prefix
-	routeTree   atomic.Pointer[bart.Table[routing.Gateways]]
+	MTU              int
-	l           *logrus.Logger
+	Routes           atomic.Pointer[[]Route]
+	routeTree        atomic.Pointer[bart.Table[routing.Gateways]]
+	l                *logrus.Logger
 
 	tun *wintun.NativeTun
 }
 
-func newTunFromFd(_ *config.C, _ *logrus.Logger, _ int, _ []netip.Prefix) (Device, error) {
+func newTunFromFd(_ *config.C, _ *logrus.Logger, _ int, _ []netip.Prefix, _ []netip.Prefix) (Device, error) {
 	return nil, fmt.Errorf("newTunFromFd not supported in Windows")
 }
 
-func newTun(c *config.C, l *logrus.Logger, vpnNetworks []netip.Prefix, _ bool) (*winTun, error) {
+func newTun(c *config.C, l *logrus.Logger, vpnNetworks []netip.Prefix, unsafeNetworks []netip.Prefix, _ bool) (*winTun, error) {
 	err := checkWinTunExists()
 	if err != nil {
 		return nil, fmt.Errorf("can not load the wintun driver: %w", err)
@@ -54,10 +57,11 @@ func newTun(c *config.C, l *logrus.Logger, vpnNetworks []netip.Prefix, _ bool) (
 	}
 
 	t := &winTun{
-		Device:      deviceName,
+		Device:         deviceName,
-		vpnNetworks: vpnNetworks,
+		vpnNetworks:    vpnNetworks,
-		MTU:         c.GetInt("tun.mtu", DefaultMTU),
+		unsafeNetworks: unsafeNetworks,
-		l:           l,
+		MTU:            c.GetInt("tun.mtu", DefaultMTU),
+		l:              l,
 	}
 
 	err = t.reload(c, true)
@@ -101,6 +105,10 @@ func (t *winTun) reload(c *config.C, initial bool) error {
 		return nil
 	}
 
+	if initial {
+		t.unsafeIPv4Origin = prepareUnsafeOriginAddr(t, t.l, c, routes)
+	}
+
 	routeTree, err := makeRouteTree(t.l, routes, false)
 	if err != nil {
 		return err
@@ -131,7 +139,12 @@ func (t *winTun) reload(c *config.C, initial bool) error {
 func (t *winTun) Activate() error {
 	luid := winipcfg.LUID(t.tun.LUID())
 
-	err := luid.SetIPAddresses(t.vpnNetworks)
+	prefixes := t.vpnNetworks
+	if t.unsafeIPv4Origin.IsValid() {
+		prefixes = append(prefixes, t.unsafeIPv4Origin)
+	}
+
+	err := luid.SetIPAddresses(prefixes)
 	if err != nil {
 		return fmt.Errorf("failed to set address: %w", err)
 	}
@@ -224,6 +237,18 @@ func (t *winTun) Networks() []netip.Prefix {
 	return t.vpnNetworks
 }
 
+func (t *winTun) UnsafeNetworks() []netip.Prefix {
+	return t.unsafeNetworks
+}
+
+func (t *winTun) UnsafeIPv4OriginAddress() netip.Prefix {
+	return t.unsafeIPv4Origin
+}
+
+func (t *winTun) SNATAddress() netip.Prefix {
+	return netip.Prefix{}
+}
+
 func (t *winTun) Name() string {
 	return t.Device
 }
@@ -240,36 +265,10 @@ func (t *winTun) SupportsMultiqueue() bool {
 	return false
 }
 
-func (t *winTun) NewMultiQueueReader() (BatchReadWriter, error) {
+func (t *winTun) NewMultiQueueReader() (io.ReadWriteCloser, error) {
 	return nil, fmt.Errorf("TODO: multiqueue not implemented for windows")
 }
 
-// BatchRead reads a single packet (batch size 1 for Windows)
-func (t *winTun) BatchRead(bufs [][]byte, sizes []int) (int, error) {
-	n, err := t.Read(bufs[0])
-	if err != nil {
-		return 0, err
-	}
-	sizes[0] = n
-	return 1, nil
-}
-
-// WriteBatch writes packets individually (no batching for Windows)
-func (t *winTun) WriteBatch(bufs [][]byte, offset int) (int, error) {
-	for i, buf := range bufs {
-		_, err := t.Write(buf[offset:])
-		if err != nil {
-			return i, err
-		}
-	}
-	return len(bufs), nil
-}
-
-// BatchSize returns 1 for Windows (no batching)
-func (t *winTun) BatchSize() int {
-	return 1
-}
-
 func (t *winTun) Close() error {
 	// It seems that the Windows networking stack doesn't like it when we destroy interfaces that have active routes,
 	// so to be certain, just remove everything before destroying.

overlay/user.go

@@ -9,7 +9,7 @@ import (
 	"github.com/slackhq/nebula/routing"
 )
 
-func NewUserDeviceFromConfig(c *config.C, l *logrus.Logger, vpnNetworks []netip.Prefix, routines int) (Device, error) {
+func NewUserDeviceFromConfig(c *config.C, l *logrus.Logger, vpnNetworks []netip.Prefix, unsafeNetworks []netip.Prefix, routines int) (Device, error) {
 	return NewUserDevice(vpnNetworks)
 }
 
@@ -36,6 +36,17 @@ type UserDevice struct {
 	inboundWriter *io.PipeWriter
 }
 
+func (d *UserDevice) UnsafeNetworks() []netip.Prefix {
+	return nil
+}
+
+func (d *UserDevice) SNATAddress() netip.Prefix {
+	return netip.Prefix{}
+}
+func (d *UserDevice) UnsafeIPv4OriginAddress() netip.Prefix {
+	return netip.Prefix{}
+}
+
 func (d *UserDevice) Activate() error {
 	return nil
 }
@@ -50,36 +61,10 @@ func (d *UserDevice) SupportsMultiqueue() bool {
 	return true
 }
 
-func (d *UserDevice) NewMultiQueueReader() (BatchReadWriter, error) {
+func (d *UserDevice) NewMultiQueueReader() (io.ReadWriteCloser, error) {
 	return d, nil
 }
 
-// BatchRead reads a single packet (batch size 1 for UserDevice)
-func (d *UserDevice) BatchRead(bufs [][]byte, sizes []int) (int, error) {
-	n, err := d.Read(bufs[0])
-	if err != nil {
-		return 0, err
-	}
-	sizes[0] = n
-	return 1, nil
-}
-
-// WriteBatch writes packets individually (no batching for UserDevice)
-func (d *UserDevice) WriteBatch(bufs [][]byte, offset int) (int, error) {
-	for i, buf := range bufs {
-		_, err := d.Write(buf[offset:])
-		if err != nil {
-			return i, err
-		}
-	}
-	return len(bufs), nil
-}
-
-// BatchSize returns 1 for UserDevice (no batching)
-func (d *UserDevice) BatchSize() int {
-	return 1
-}
-
 func (d *UserDevice) Pipe() (*io.PipeReader, *io.PipeWriter) {
 	return d.inboundReader, d.outboundWriter
 }

pki.go

@@ -91,7 +91,7 @@ func (p *PKI) reload(c *config.C, initial bool) error {
 }
 
 func (p *PKI) reloadCerts(c *config.C, initial bool) *util.ContextualError {
-	newState, err := newCertStateFromConfig(c)
+	newState, err := newCertStateFromConfig(c, p.l)
 	if err != nil {
 		return util.NewContextualError("Could not load client cert", nil, err)
 	}
@@ -102,7 +102,7 @@ func (p *PKI) reloadCerts(c *config.C, initial bool) *util.ContextualError {
 			if currentState.v1Cert == nil {
 				//adding certs is fine, actually. Networks-in-common confirmed in newCertState().
 			} else {
-				// did IP in cert change? if so, don't set
+				// did IP in cert change? if so, don't set. If we ever allow this, need to set p.firewallReloadNeeded
 				if !slices.Equal(currentState.v1Cert.Networks(), newState.v1Cert.Networks()) {
 					return util.NewContextualError(
 						"Networks in new cert was different from old",
@@ -158,6 +158,14 @@ func (p *PKI) reloadCerts(c *config.C, initial bool) *util.ContextualError {
 			}
 		}
 
+		newUN := newState.GetDefaultCertificate().UnsafeNetworks()
+		oldUN := currentState.GetDefaultCertificate().UnsafeNetworks()
+		if !slices.Equal(newUN, oldUN) {
+			//todo I don't love this, because other clients will see the new assignments and act on them, but we will not be able to.
+			//I think we need to wire this into the firewall reload.
+			p.l.WithFields(m{"previous": oldUN, "new": newUN}).Warning("UnsafeNetworks assignments differ. A restart is required in order for this to take effect.")
+		}
+
 		// Cipher cant be hot swapped so just leave it at what it was before
 		newState.cipher = currentState.cipher
 
@@ -260,7 +268,7 @@ func (cs *CertState) MarshalJSON() ([]byte, error) {
 	return json.Marshal(msg)
 }
 
-func newCertStateFromConfig(c *config.C) (*CertState, error) {
+func newCertStateFromConfig(c *config.C, l *logrus.Logger) (*CertState, error) {
 	var err error
 
 	privPathOrPEM := c.GetString("pki.key", "")
@@ -344,10 +352,33 @@ func newCertStateFromConfig(c *config.C) (*CertState, error) {
 		return nil, fmt.Errorf("unknown pki.initiating_version: %v", rawInitiatingVersion)
 	}
 
-	return newCertState(initiatingVersion, v1, v2, isPkcs11, curve, rawKey)
+	return newCertState(l, initiatingVersion, v1, v2, isPkcs11, curve, rawKey)
 }
 
-func newCertState(dv cert.Version, v1, v2 cert.Certificate, pkcs11backed bool, privateKeyCurve cert.Curve, privateKey []byte) (*CertState, error) {
+func compareUnsafeNetworksAcrossCertVersions(v1, v2 cert.Certificate) error {
+	if v1 == nil || v2 == nil {
+		return nil //can't be a problem if we don't have one of the kinds of cert
+	}
+
+	v4UnsafeNets := 0
+	for _, n := range v2.UnsafeNetworks() {
+		if n.Addr().Is6() {
+			continue // V1 certs can't have IPv6 unsafe networks
+		} else {
+			v4UnsafeNets++
+		}
+		if !slices.Contains(v1.UnsafeNetworks(), n) {
+			return errors.New("UnsafeNetworks mismatch")
+		}
+	}
+	if len(v1.UnsafeNetworks()) != v4UnsafeNets {
+		return errors.New("UnsafeNetworks mismatch")
+	}
+
+	return nil
+}
+
+func newCertState(l *logrus.Logger, dv cert.Version, v1, v2 cert.Certificate, pkcs11backed bool, privateKeyCurve cert.Curve, privateKey []byte) (*CertState, error) {
 	cs := CertState{
 		privateKey:               privateKey,
 		pkcs11Backed:             pkcs11backed,
@@ -370,6 +401,12 @@ func newCertState(dv cert.Version, v1, v2 cert.Certificate, pkcs11backed bool, p
 		}
 
 		cs.initiatingVersion = dv
+
+		warn := compareUnsafeNetworksAcrossCertVersions(v1, v2)
+		if warn != nil {
+			l.WithFields(m{"UnsafeNetworksV1": v1.UnsafeNetworks(), "UnsafeNetworksV2": v2.UnsafeNetworks()}).
+				Warning("the IPv4 UnsafeNetworks assigned in the V1 certificate do not match the ones in V2")
+		}
 	}
 
 	if v1 != nil {

relay_manager.go

@@ -55,7 +55,7 @@ func (rm *relayManager) setAmRelay(v bool) {
 func AddRelay(l *logrus.Logger, relayHostInfo *HostInfo, hm *HostMap, vpnIp netip.Addr, remoteIdx *uint32, relayType int, state int) (uint32, error) {
 	hm.Lock()
 	defer hm.Unlock()
-	for i := 0; i < 32; i++ {
+	for range 32 {
 		index, err := generateIndex(l)
 		if err != nil {
 			return 0, err

remote_list.go

@@ -404,12 +404,7 @@ func (r *RemoteList) Rebuild(preferredRanges []netip.Prefix) {
 
 // unlockedIsBad assumes you have the write lock and checks if the remote matches any entry in the blocked address list
 func (r *RemoteList) unlockedIsBad(remote netip.AddrPort) bool {
-	for _, v := range r.badRemotes {
+	return slices.Contains(r.badRemotes, remote)
-		if v == remote {
-			return true
-		}
-	}
-	return false
 }
 
 // unlockedSetLearnedV4 assumes you have the write lock and sets the current learned address for this owner and marks the

snat.go

@@ -0,0 +1,91 @@
+package nebula
+
+import (
+	"encoding/binary"
+	"net/netip"
+)
+
+func recalcIPv4Checksum(data []byte, oldSrcIP netip.Addr, newSrcIP netip.Addr) {
+	oldChecksum := binary.BigEndian.Uint16(data[10:12])
+	//because of how checksums work, we can re-use this function
+	checksum := calcNewTransportChecksum(oldChecksum, oldSrcIP, 0, newSrcIP, 0)
+	binary.BigEndian.PutUint16(data[10:12], checksum)
+}
+
+func calcNewTransportChecksum(oldChecksum uint16, oldSrcIP netip.Addr, oldSrcPort uint16, newSrcIP netip.Addr, newSrcPort uint16) uint16 {
+	oldIP := binary.BigEndian.Uint32(oldSrcIP.AsSlice())
+	newIP := binary.BigEndian.Uint32(newSrcIP.AsSlice())
+
+	// Start with inverted checksum
+	checksum := uint32(^oldChecksum)
+
+	// Subtract old IP (as two 16-bit words)
+	checksum += uint32(^uint16(oldIP >> 16))
+	checksum += uint32(^uint16(oldIP & 0xFFFF))
+
+	// Subtract old port
+	checksum += uint32(^oldSrcPort)
+
+	// Add new IP (as two 16-bit words)
+	checksum += uint32(newIP >> 16)
+	checksum += uint32(newIP & 0xFFFF)
+
+	// Add new port
+	checksum += uint32(newSrcPort)
+
+	// Fold carries
+	for checksum > 0xFFFF {
+		checksum = (checksum & 0xFFFF) + (checksum >> 16)
+	}
+
+	// Return ones' complement
+	return ^uint16(checksum)
+}
+
+func recalcV4TransportChecksum(offsetInsideHeader int, data []byte, oldSrcIP netip.AddrPort, newSrcIP netip.AddrPort) {
+	ipHeaderOffset := int(data[0]&0x0F) * 4
+	offset := ipHeaderOffset + offsetInsideHeader
+	oldcsum := binary.BigEndian.Uint16(data[offset : offset+2])
+	checksum := calcNewTransportChecksum(oldcsum, oldSrcIP.Addr(), oldSrcIP.Port(), newSrcIP.Addr(), newSrcIP.Port())
+	binary.BigEndian.PutUint16(data[offset:offset+2], checksum)
+}
+
+func recalcUDPv4Checksum(data []byte, oldSrcIP netip.AddrPort, newSrcIP netip.AddrPort) {
+	const offsetInsideHeader = 6
+	recalcV4TransportChecksum(offsetInsideHeader, data, oldSrcIP, newSrcIP)
+}
+
+func recalcTCPv4Checksum(data []byte, oldSrcIP netip.AddrPort, newSrcIP netip.AddrPort) {
+	const offsetInsideHeader = 16
+	recalcV4TransportChecksum(offsetInsideHeader, data, oldSrcIP, newSrcIP)
+}
+
+func calcNewICMPChecksum(oldChecksum uint16, oldCode uint16, newCode uint16, oldID uint16, newID uint16) uint16 {
+	// Start with inverted checksum
+	checksum := uint32(^oldChecksum)
+
+	// Subtract old stuff
+	checksum += uint32(^oldCode)
+	checksum += uint32(^oldID)
+
+	// Add new stuff
+	checksum += uint32(newCode)
+	checksum += uint32(newID)
+
+	// Fold carries
+	for checksum > 0xFFFF {
+		checksum = (checksum & 0xFFFF) + (checksum >> 16)
+	}
+
+	// Return ones' complement
+	return ^uint16(checksum)
+}
+
+func recalcICMPv4Checksum(data []byte, oldCode uint16, newCode uint16, oldID uint16, newID uint16) {
+	const offsetInsideHeader = 2
+	ipHeaderOffset := int(data[0]&0x0F) * 4
+	offset := ipHeaderOffset + offsetInsideHeader
+	oldChecksum := binary.BigEndian.Uint16(data[offset : offset+2])
+	checksum := calcNewICMPChecksum(oldChecksum, oldCode, newCode, oldID, newID)
+	binary.BigEndian.PutUint16(data[offset:offset+2], checksum)
+}

ssh.go

@@ -6,6 +6,7 @@ import (
 	"errors"
 	"flag"
 	"fmt"
+	"maps"
 	"net"
 	"net/netip"
 	"os"
@@ -831,9 +832,7 @@ func sshPrintRelays(ifce *Interface, fs any, a []string, w sshd.StringWriter) er
 
 	relays := map[uint32]*HostInfo{}
 	ifce.hostMap.Lock()
-	for k, v := range ifce.hostMap.Relays {
+	maps.Copy(relays, ifce.hostMap.Relays)
-		relays[k] = v
-	}
 	ifce.hostMap.Unlock()
 
 	type RelayFor struct {

stats.go

@@ -6,7 +6,6 @@ import (
 	"log"
 	"net"
 	"net/http"
-	_ "net/http/pprof"
 	"runtime"
 	"strconv"
 	"time"

test/tun.go

@@ -1,16 +1,27 @@
-package device
+package test
 
 import (
 	"errors"
 	"io"
 	"net/netip"
 
-	"github.com/slackhq/nebula/overlay"
 	"github.com/slackhq/nebula/routing"
 )
 
 type NoopTun struct{}
 
+func (NoopTun) UnsafeNetworks() []netip.Prefix {
+	return nil
+}
+
+func (NoopTun) SNATAddress() netip.Prefix {
+	return netip.Prefix{}
+}
+
+func (NoopTun) UnsafeIPv4OriginAddress() netip.Prefix {
+	return netip.Prefix{}
+}
+
 func (NoopTun) RoutesFor(addr netip.Addr) routing.Gateways {
 	return routing.Gateways{}
 }
@@ -39,25 +50,10 @@ func (NoopTun) SupportsMultiqueue() bool {
 	return false
 }
 
-func (NoopTun) NewMultiQueueReader() (overlay.BatchReadWriter, error) {
+func (NoopTun) NewMultiQueueReader() (io.ReadWriteCloser, error) {
 	return nil, errors.New("unsupported")
 }
 
 func (NoopTun) Close() error {
 	return nil
 }
-
-// BatchRead implements BatchReadWriter interface
-func (NoopTun) BatchRead(bufs [][]byte, sizes []int) (int, error) {
-	return 0, io.EOF
-}
-
-// WriteBatch implements BatchReadWriter interface
-func (NoopTun) WriteBatch(bufs [][]byte, offset int) (int, error) {
-	return len(bufs), nil
-}
-
-// BatchSize implements BatchReadWriter interface
-func (NoopTun) BatchSize() int {
-	return 1
-}

timeout_test.go

@@ -134,7 +134,7 @@ func TestTimerWheel_Purge(t *testing.T) {
 	assert.True(t, tw.lastTick.After(lastTick))
 
 	// Make sure we get all 4 packets back
-	for i := 0; i < 4; i++ {
+	for i := range 4 {
 		p, has := tw.Purge()
 		assert.True(t, has)
 		assert.Equal(t, fps[i], p)
@@ -149,7 +149,7 @@ func TestTimerWheel_Purge(t *testing.T) {
 	// Make sure we cached the free'd items
 	assert.Equal(t, 4, tw.itemsCached)
 	ci := tw.itemCache
-	for i := 0; i < 4; i++ {
+	for range 4 {
 		assert.NotNil(t, ci)
 		ci = ci.Next
 	}

udp/conn.go

@@ -13,21 +13,13 @@ type EncReader func(
 	payload []byte,
 )
 
-type EncBatchReader func(
-	addrs []netip.AddrPort,
-	payloads [][]byte,
-	count int,
-)
-
 type Conn interface {
 	Rebind() error
 	LocalAddr() (netip.AddrPort, error)
-	ListenOutBatch(r EncBatchReader)
+	ListenOut(r EncReader)
 	WriteTo(b []byte, addr netip.AddrPort) error
-	WriteMulti(packets [][]byte, addrs []netip.AddrPort) (int, error)
 	ReloadConfig(c *config.C)
 	SupportsMultipleReaders() bool
-	BatchSize() int
 	Close() error
 }
 
@@ -39,25 +31,17 @@ func (NoopConn) Rebind() error {
 func (NoopConn) LocalAddr() (netip.AddrPort, error) {
 	return netip.AddrPort{}, nil
 }
-
+func (NoopConn) ListenOut(_ EncReader) {
-func (NoopConn) ListenOut(_ EncReader) {}
+	return
-
+}
 func (NoopConn) SupportsMultipleReaders() bool {
 	return false
 }
-
-func (NoopConn) ListenOutBatch(_ EncBatchReader) {}
-
 func (NoopConn) WriteTo(_ []byte, _ netip.AddrPort) error {
 	return nil
 }
-func (NoopConn) WriteMulti(_ [][]byte, _ []netip.AddrPort) (int, error) {
+func (NoopConn) ReloadConfig(_ *config.C) {
-	return 0, nil
+	return
-}
-func (NoopConn) ReloadConfig(_ *config.C) {}
-
-func (NoopConn) BatchSize() int {
-	return 1
 }
 func (NoopConn) Close() error {
 	return nil

udp/udp_darwin.go

@@ -140,17 +140,6 @@ func (u *StdConn) WriteTo(b []byte, ap netip.AddrPort) error {
 	}
 }
 
-// WriteMulti sends multiple packets - fallback implementation without sendmmsg
-func (u *StdConn) WriteMulti(packets [][]byte, addrs []netip.AddrPort) (int, error) {
-	for i := range packets {
-		err := u.WriteTo(packets[i], addrs[i])
-		if err != nil {
-			return i, err
-		}
-	}
-	return len(packets), nil
-}
-
 func (u *StdConn) LocalAddr() (netip.AddrPort, error) {
 	a := u.UDPConn.LocalAddr()
 
@@ -199,34 +188,6 @@ func (u *StdConn) SupportsMultipleReaders() bool {
 	return false
 }
 
-// ListenOutBatch - fallback to single-packet reads for Darwin
-func (u *StdConn) ListenOutBatch(r EncBatchReader) {
-	buffer := make([]byte, MTU)
-	addrs := make([]netip.AddrPort, 1)
-	payloads := make([][]byte, 1)
-
-	for {
-		// Just read one packet at a time and call batch callback with count=1
-		n, rua, err := u.ReadFromUDPAddrPort(buffer)
-		if err != nil {
-			if errors.Is(err, net.ErrClosed) {
-				u.l.WithError(err).Debug("udp socket is closed, exiting read loop")
-				return
-			}
-
-			u.l.WithError(err).Error("unexpected udp socket receive error")
-		}
-
-		addrs[0] = netip.AddrPortFrom(rua.Addr().Unmap(), rua.Port())
-		payloads[0] = buffer[:n]
-		r(addrs, payloads, 1)
-	}
-}
-
-func (u *StdConn) BatchSize() int {
-	return 1
-}
-
 func (u *StdConn) Rebind() error {
 	var err error
 	if u.isV4 {

udp/udp_generic.go

@@ -101,38 +101,3 @@ func (u *GenericConn) ListenOut(r EncReader) {
 func (u *GenericConn) SupportsMultipleReaders() bool {
 	return false
 }
-
-// ListenOutBatch - fallback to single-packet reads for generic platforms
-func (u *GenericConn) ListenOutBatch(r EncBatchReader) {
-	buffer := make([]byte, MTU)
-	addrs := make([]netip.AddrPort, 1)
-	payloads := make([][]byte, 1)
-
-	for {
-		// Just read one packet at a time and call batch callback with count=1
-		n, rua, err := u.ReadFromUDPAddrPort(buffer)
-		if err != nil {
-			u.l.WithError(err).Debug("udp socket is closed, exiting read loop")
-			return
-		}
-
-		addrs[0] = netip.AddrPortFrom(rua.Addr().Unmap(), rua.Port())
-		payloads[0] = buffer[:n]
-		r(addrs, payloads, 1)
-	}
-}
-
-// WriteMulti sends multiple packets - fallback implementation
-func (u *GenericConn) WriteMulti(packets [][]byte, addrs []netip.AddrPort) (int, error) {
-	for i := range packets {
-		err := u.WriteTo(packets[i], addrs[i])
-		if err != nil {
-			return i, err
-		}
-	}
-	return len(packets), nil
-}
-
-func (u *GenericConn) BatchSize() int {
-	return 1
-}

udp/udp_linux.go

@@ -22,11 +22,6 @@ type StdConn struct {
 	isV4  bool
 	l     *logrus.Logger
 	batch int
-
-	// Pre-allocated buffers for batch writes (sized for IPv6, works for both)
-	writeMsgs   []rawMessage
-	writeIovecs []iovec
-	writeNames  [][]byte
 }
 
 func maybeIPV4(ip net.IP) (net.IP, bool) {
@@ -74,26 +69,7 @@ func NewListener(l *logrus.Logger, ip netip.Addr, port int, multi bool, batch in
 		return nil, fmt.Errorf("unable to bind to socket: %s", err)
 	}
 
-	c := &StdConn{sysFd: fd, isV4: ip.Is4(), l: l, batch: batch}
+	return &StdConn{sysFd: fd, isV4: ip.Is4(), l: l, batch: batch}, err
-
-	// Pre-allocate write message structures for batching (sized for IPv6, works for both)
-	c.writeMsgs = make([]rawMessage, batch)
-	c.writeIovecs = make([]iovec, batch)
-	c.writeNames = make([][]byte, batch)
-
-	for i := range c.writeMsgs {
-		// Allocate for IPv6 size (larger than IPv4, works for both)
-		c.writeNames[i] = make([]byte, unix.SizeofSockaddrInet6)
-
-		// Point to the iovec in the slice
-		c.writeMsgs[i].Hdr.Iov = &c.writeIovecs[i]
-		c.writeMsgs[i].Hdr.Iovlen = 1
-
-		c.writeMsgs[i].Hdr.Name = &c.writeNames[i][0]
-		// Namelen will be set appropriately in writeMulti4/writeMulti6
-	}
-
-	return c, err
 }
 
 func (u *StdConn) SupportsMultipleReaders() bool {
@@ -146,7 +122,7 @@ func (u *StdConn) LocalAddr() (netip.AddrPort, error) {
 	}
 }
 
-func (u *StdConn) ListenOutBatch(r EncBatchReader) {
+func (u *StdConn) ListenOut(r EncReader) {
 	var ip netip.Addr
 
 	msgs, buffers, names := u.PrepareRawMessages(u.batch)
@@ -155,12 +131,6 @@ func (u *StdConn) ListenOutBatch(r EncBatchReader) {
 		read = u.ReadSingle
 	}
 
-	udpBatchHist := metrics.GetOrRegisterHistogram("batch.udp_read_size", nil, metrics.NewUniformSample(1024))
-
-	// Pre-allocate slices for batch callback
-	addrs := make([]netip.AddrPort, u.batch)
-	payloads := make([][]byte, u.batch)
-
 	for {
 		n, err := read(msgs)
 		if err != nil {
@@ -168,21 +138,15 @@ func (u *StdConn) ListenOutBatch(r EncBatchReader) {
 			return
 		}
 
-		udpBatchHist.Update(int64(n))
-
-		// Prepare batch data
 		for i := 0; i < n; i++ {
+			// Its ok to skip the ok check here, the slicing is the only error that can occur and it will panic
 			if u.isV4 {
 				ip, _ = netip.AddrFromSlice(names[i][4:8])
 			} else {
 				ip, _ = netip.AddrFromSlice(names[i][8:24])
 			}
-			addrs[i] = netip.AddrPortFrom(ip.Unmap(), binary.BigEndian.Uint16(names[i][2:4]))
+			r(netip.AddrPortFrom(ip.Unmap(), binary.BigEndian.Uint16(names[i][2:4])), buffers[i][:msgs[i].Len])
-			payloads[i] = buffers[i][:msgs[i].Len]
 		}
-
-		// Call batch callback with all packets
-		r(addrs, payloads, n)
 	}
 }
 
@@ -234,19 +198,6 @@ func (u *StdConn) WriteTo(b []byte, ip netip.AddrPort) error {
 	return u.writeTo6(b, ip)
 }
 
-func (u *StdConn) WriteMulti(packets [][]byte, addrs []netip.AddrPort) (int, error) {
-	if len(packets) != len(addrs) {
-		return 0, fmt.Errorf("packets and addrs length mismatch")
-	}
-	if len(packets) == 0 {
-		return 0, nil
-	}
-	if u.isV4 {
-		return u.writeMulti4(packets, addrs)
-	}
-	return u.writeMulti6(packets, addrs)
-}
-
 func (u *StdConn) writeTo6(b []byte, ip netip.AddrPort) error {
 	var rsa unix.RawSockaddrInet6
 	rsa.Family = unix.AF_INET6
@@ -301,123 +252,6 @@ func (u *StdConn) writeTo4(b []byte, ip netip.AddrPort) error {
 	}
 }
 
-func (u *StdConn) writeMulti4(packets [][]byte, addrs []netip.AddrPort) (int, error) {
-	sent := 0
-	for sent < len(packets) {
-		// Determine batch size based on remaining packets and buffer capacity
-		batchSize := len(packets) - sent
-		if batchSize > len(u.writeMsgs) {
-			batchSize = len(u.writeMsgs)
-		}
-
-		// Use pre-allocated buffers
-		msgs := u.writeMsgs[:batchSize]
-		iovecs := u.writeIovecs[:batchSize]
-		names := u.writeNames[:batchSize]
-
-		// Setup message structures for this batch
-		for i := 0; i < batchSize; i++ {
-			pktIdx := sent + i
-			if !addrs[pktIdx].Addr().Is4() {
-				return sent + i, ErrInvalidIPv6RemoteForSocket
-			}
-
-			// Setup the packet buffer
-			iovecs[i].Base = &packets[pktIdx][0]
-			iovecs[i].Len = uint(len(packets[pktIdx]))
-
-			// Setup the destination address
-			rsa := (*unix.RawSockaddrInet4)(unsafe.Pointer(&names[i][0]))
-			rsa.Family = unix.AF_INET
-			rsa.Addr = addrs[pktIdx].Addr().As4()
-			binary.BigEndian.PutUint16((*[2]byte)(unsafe.Pointer(&rsa.Port))[:], addrs[pktIdx].Port())
-
-			// Set the appropriate address length for IPv4
-			msgs[i].Hdr.Namelen = unix.SizeofSockaddrInet4
-		}
-
-		// Send this batch
-		nsent, _, err := unix.Syscall6(
-			unix.SYS_SENDMMSG,
-			uintptr(u.sysFd),
-			uintptr(unsafe.Pointer(&msgs[0])),
-			uintptr(batchSize),
-			0,
-			0,
-			0,
-		)
-
-		if err != 0 {
-			return sent + int(nsent), &net.OpError{Op: "sendmmsg", Err: err}
-		}
-
-		sent += int(nsent)
-		if int(nsent) < batchSize {
-			// Couldn't send all packets in batch, return what we sent
-			return sent, nil
-		}
-	}
-
-	return sent, nil
-}
-
-func (u *StdConn) writeMulti6(packets [][]byte, addrs []netip.AddrPort) (int, error) {
-	sent := 0
-	for sent < len(packets) {
-		// Determine batch size based on remaining packets and buffer capacity
-		batchSize := len(packets) - sent
-		if batchSize > len(u.writeMsgs) {
-			batchSize = len(u.writeMsgs)
-		}
-
-		// Use pre-allocated buffers
-		msgs := u.writeMsgs[:batchSize]
-		iovecs := u.writeIovecs[:batchSize]
-		names := u.writeNames[:batchSize]
-
-		// Setup message structures for this batch
-		for i := 0; i < batchSize; i++ {
-			pktIdx := sent + i
-
-			// Setup the packet buffer
-			iovecs[i].Base = &packets[pktIdx][0]
-			iovecs[i].Len = uint(len(packets[pktIdx]))
-
-			// Setup the destination address
-			rsa := (*unix.RawSockaddrInet6)(unsafe.Pointer(&names[i][0]))
-			rsa.Family = unix.AF_INET6
-			rsa.Addr = addrs[pktIdx].Addr().As16()
-			binary.BigEndian.PutUint16((*[2]byte)(unsafe.Pointer(&rsa.Port))[:], addrs[pktIdx].Port())
-
-			// Set the appropriate address length for IPv6
-			msgs[i].Hdr.Namelen = unix.SizeofSockaddrInet6
-		}
-
-		// Send this batch
-		nsent, _, err := unix.Syscall6(
-			unix.SYS_SENDMMSG,
-			uintptr(u.sysFd),
-			uintptr(unsafe.Pointer(&msgs[0])),
-			uintptr(batchSize),
-			0,
-			0,
-			0,
-		)
-
-		if err != 0 {
-			return sent + int(nsent), &net.OpError{Op: "sendmmsg", Err: err}
-		}
-
-		sent += int(nsent)
-		if int(nsent) < batchSize {
-			// Couldn't send all packets in batch, return what we sent
-			return sent, nil
-		}
-	}
-
-	return sent, nil
-}
-
 func (u *StdConn) ReloadConfig(c *config.C) {
 	b := c.GetInt("listen.read_buffer", 0)
 	if b > 0 {
@@ -475,10 +309,6 @@ func (u *StdConn) getMemInfo(meminfo *[unix.SK_MEMINFO_VARS]uint32) error {
 	return nil
 }
 
-func (u *StdConn) BatchSize() int {
-	return u.batch
-}
-
 func (u *StdConn) Close() error {
 	return syscall.Close(u.sysFd)
 }

udp/udp_linux_32.go

@@ -12,7 +12,7 @@ import (
 
 type iovec struct {
 	Base *byte
-	Len  uint
+	Len  uint32
 }
 
 type msghdr struct {
@@ -40,7 +40,7 @@ func (u *StdConn) PrepareRawMessages(n int) ([]rawMessage, [][]byte, [][]byte) {
 		names[i] = make([]byte, unix.SizeofSockaddrInet6)
 
 		vs := []iovec{
-			{Base: &buffers[i][0], Len: uint(len(buffers[i]))},
+			{Base: &buffers[i][0], Len: uint32(len(buffers[i]))},
 		}
 
 		msgs[i].Hdr.Iov = &vs[0]

udp/udp_linux_64.go

@@ -12,7 +12,7 @@ import (
 
 type iovec struct {
 	Base *byte
-	Len  uint
+	Len  uint64
 }
 
 type msghdr struct {
@@ -43,7 +43,7 @@ func (u *StdConn) PrepareRawMessages(n int) ([]rawMessage, [][]byte, [][]byte) {
 		names[i] = make([]byte, unix.SizeofSockaddrInet6)
 
 		vs := []iovec{
-			{Base: &buffers[i][0], Len: uint(len(buffers[i]))},
+			{Base: &buffers[i][0], Len: uint64(len(buffers[i]))},
 		}
 
 		msgs[i].Hdr.Iov = &vs[0]

udp/udp_rio_windows.go

@@ -338,50 +338,6 @@ func (u *RIOConn) Rebind() error {
 
 func (u *RIOConn) ReloadConfig(*config.C) {}
 
-// BatchSize returns 1 since RIO reads packets one at a time
-func (u *RIOConn) BatchSize() int {
-	return 1
-}
-
-// ListenOutBatch - fallback to single-packet reads for RIO
-func (u *RIOConn) ListenOutBatch(r EncBatchReader) {
-	buffer := make([]byte, MTU)
-	addrs := make([]netip.AddrPort, 1)
-	payloads := make([][]byte, 1)
-
-	var lastRecvErr time.Time
-
-	for {
-		n, rua, err := u.receive(buffer)
-		if err != nil {
-			if errors.Is(err, net.ErrClosed) {
-				u.l.WithError(err).Debug("udp socket is closed, exiting read loop")
-				return
-			}
-			if lastRecvErr.IsZero() || time.Since(lastRecvErr) > time.Minute {
-				lastRecvErr = time.Now()
-				u.l.WithError(err).Warn("unexpected udp socket receive error")
-			}
-			continue
-		}
-
-		addrs[0] = netip.AddrPortFrom(netip.AddrFrom16(rua.Addr).Unmap(), (rua.Port>>8)|((rua.Port&0xff)<<8))
-		payloads[0] = buffer[:n]
-		r(addrs, payloads, 1)
-	}
-}
-
-// WriteMulti sends multiple packets - fallback implementation
-func (u *RIOConn) WriteMulti(packets [][]byte, addrs []netip.AddrPort) (int, error) {
-	for i := range packets {
-		err := u.WriteTo(packets[i], addrs[i])
-		if err != nil {
-			return i, err
-		}
-	}
-	return len(packets), nil
-}
-
 func (u *RIOConn) Close() error {
 	if !u.isOpen.CompareAndSwap(true, false) {
 		return nil

udp/udp_tester.go

@@ -116,31 +116,6 @@ func (u *TesterConn) ListenOut(r EncReader) {
 	}
 }
 
-func (u *TesterConn) ListenOutBatch(r EncBatchReader) {
-	addrs := make([]netip.AddrPort, 1)
-	payloads := make([][]byte, 1)
-
-	for {
-		p, ok := <-u.RxPackets
-		if !ok {
-			return
-		}
-		addrs[0] = p.From
-		payloads[0] = p.Data
-		r(addrs, payloads, 1)
-	}
-}
-
-func (u *TesterConn) WriteMulti(packets [][]byte, addrs []netip.AddrPort) (int, error) {
-	for i := range packets {
-		err := u.WriteTo(packets[i], addrs[i])
-		if err != nil {
-			return i, err
-		}
-	}
-	return len(packets), nil
-}
-
 func (u *TesterConn) ReloadConfig(*config.C) {}
 
 func NewUDPStatsEmitter(_ []Conn) func() {
@@ -156,10 +131,6 @@ func (u *TesterConn) SupportsMultipleReaders() bool {
 	return false
 }
 
-func (u *TesterConn) BatchSize() int {
-	return 1
-}
-
 func (u *TesterConn) Rebind() error {
 	return nil
 }