block less I guess

This reverts commit fa098c551a23d055f7ba040f14991b21dd72868f.

CHANGELOG.md

@@ -7,19 +7,6 @@ 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
@@ -788,8 +775,7 @@ created.)
 
 - Initial public release.
 
-[Unreleased]: https://github.com/slackhq/nebula/compare/v1.10.3...HEAD
-[1.10.3]: https://github.com/slackhq/nebula/releases/tag/v1.10.3
+[Unreleased]: https://github.com/slackhq/nebula/compare/v1.10.2...HEAD
 [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

cert/ca_pool.go

@@ -141,23 +141,10 @@ 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,
 	}
 
@@ -171,11 +158,6 @@ 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

@@ -5,7 +5,6 @@ import (
 	"testing"
 	"time"
 
-	"github.com/slackhq/nebula/cert/p256"
 	"github.com/stretchr/testify/assert"
 	"github.com/stretchr/testify/require"
 )
@@ -171,15 +170,6 @@ 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)
@@ -197,7 +187,7 @@ 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)
 
@@ -206,17 +196,7 @@ 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"})
-	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)
+	_, err = caPool.VerifyCertificate(time.Now(), c)
 	require.NoError(t, err)
 }
 
@@ -414,15 +394,6 @@ 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)
@@ -440,7 +411,7 @@ 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)
 
@@ -449,17 +420,7 @@ 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"})
-	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)
+	_, err = caPool.VerifyCertificate(time.Now(), c)
 	require.NoError(t, err)
 }
 

cert/cert.go

@@ -4,8 +4,6 @@ import (
 	"fmt"
 	"net/netip"
 	"time"
-
-	"github.com/slackhq/nebula/cert/p256"
 )
 
 type Version uint8
@@ -112,9 +110,6 @@ 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 {
@@ -157,31 +152,3 @@ 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/p256/p256.go

@@ -1,122 +0,0 @@
-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

@@ -1,28 +0,0 @@
-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

@@ -9,8 +9,6 @@ import (
 	"fmt"
 	"net/netip"
 	"time"
-
-	"github.com/slackhq/nebula/cert/p256"
 )
 
 // TBSCertificate represents a certificate intended to be signed.
@@ -128,13 +126,6 @@ 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

@@ -9,7 +9,6 @@ import (
 	"testing"
 	"time"
 
-	"github.com/slackhq/nebula/cert/p256"
 	"github.com/stretchr/testify/assert"
 	"github.com/stretchr/testify/require"
 )
@@ -90,48 +89,3 @@ 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-service/main.go

@@ -78,8 +78,16 @@ func main() {
 	}
 
 	if !*configTest {
-		ctrl.Start()
-		ctrl.ShutdownBlock()
+		wait, err := ctrl.Start()
+		if err != nil {
+			util.LogWithContextIfNeeded("Error while running", err, l)
+			os.Exit(1)
+		}
+
+		go ctrl.ShutdownBlock()
+		wait()
+
+		l.Info("Goodbye")
 	}
 
 	os.Exit(0)

cmd/nebula/main.go

@@ -72,9 +72,17 @@ func main() {
 	}
 
 	if !*configTest {
-		ctrl.Start()
+		wait, err := ctrl.Start()
+		if err != nil {
+			util.LogWithContextIfNeeded("Error while running", err, l)
+			os.Exit(1)
+		}
+
+		go ctrl.ShutdownBlock()
 		notifyReady(l)
-		ctrl.ShutdownBlock()
+		wait()
+
+		l.Info("Goodbye")
 	}
 
 	os.Exit(0)

control.go

@@ -2,9 +2,11 @@ package nebula
 
 import (
 	"context"
+	"errors"
 	"net/netip"
 	"os"
 	"os/signal"
+	"sync"
 	"syscall"
 
 	"github.com/sirupsen/logrus"
@@ -13,6 +15,16 @@ import (
 	"github.com/slackhq/nebula/overlay"
 )
 
+type RunState int
+
+const (
+	Stopped  RunState = 0 // The control has yet to be started
+	Started  RunState = 1 // The control has been started
+	Stopping RunState = 2 // The control is stopping
+)
+
+var ErrAlreadyStarted = errors.New("nebula is already started")
+
 // Every interaction here needs to take extra care to copy memory and not return or use arguments "as is" when touching
 // core. This means copying IP objects, slices, de-referencing pointers and taking the actual value, etc
 
@@ -26,6 +38,9 @@ type controlHostLister interface {
 }
 
 type Control struct {
+	stateLock sync.Mutex
+	state     RunState
+
 	f                      *Interface
 	l                      *logrus.Logger
 	ctx                    context.Context
@@ -49,10 +64,21 @@ type ControlHostInfo struct {
 	CurrentRelaysThroughMe []netip.Addr     `json:"currentRelaysThroughMe"`
 }
 
-// Start actually runs nebula, this is a nonblocking call. To block use Control.ShutdownBlock()
-func (c *Control) Start() {
+// Start actually runs nebula, this is a nonblocking call.
+// The returned function can be used to wait for nebula to fully stop.
+func (c *Control) Start() (func(), error) {
+	c.stateLock.Lock()
+	if c.state != Stopped {
+		c.stateLock.Unlock()
+		return nil, ErrAlreadyStarted
+	}
+
 	// Activate the interface
-	c.f.activate()
+	err := c.f.activate()
+	if err != nil {
+		c.stateLock.Unlock()
+		return nil, err
+	}
 
 	// Call all the delayed funcs that waited patiently for the interface to be created.
 	if c.sshStart != nil {
@@ -72,15 +98,33 @@ func (c *Control) Start() {
 	}
 
 	// Start reading packets.
-	c.f.run()
+	c.state = Started
+	c.stateLock.Unlock()
+	return c.f.run()
+}
+
+func (c *Control) State() RunState {
+	c.stateLock.Lock()
+	defer c.stateLock.Unlock()
+	return c.state
 }
 
 func (c *Control) Context() context.Context {
 	return c.ctx
 }
 
-// Stop signals nebula to shutdown and close all tunnels, returns after the shutdown is complete
+// Stop is a non-blocking call that signals nebula to close all tunnels and shut down
 func (c *Control) Stop() {
+	c.stateLock.Lock()
+	if c.state != Started {
+		c.stateLock.Unlock()
+		// We are stopping or stopped already
+		return
+	}
+
+	c.state = Stopping
+	c.stateLock.Unlock()
+
 	// Stop the handshakeManager (and other services), to prevent new tunnels from
 	// being created while we're shutting them all down.
 	c.cancel()
@@ -89,7 +133,7 @@ func (c *Control) Stop() {
 	if err := c.f.Close(); err != nil {
 		c.l.WithError(err).Error("Close interface failed")
 	}
-	c.l.Info("Goodbye")
+	c.state = Stopped
 }
 
 // ShutdownBlock will listen for and block on term and interrupt signals, calling Control.Stop() once signalled

go.mod

@@ -4,7 +4,6 @@ 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

go.sum

@@ -1,8 +1,6 @@
 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=

interface.go

@@ -6,8 +6,8 @@ import (
 	"fmt"
 	"io"
 	"net/netip"
-	"os"
 	"runtime"
+	"sync"
 	"sync/atomic"
 	"time"
 
@@ -88,6 +88,7 @@ type Interface struct {
 
 	writers []udp.Conn
 	readers []io.ReadWriteCloser
+	wg      sync.WaitGroup
 
 	metricHandshakes    metrics.Histogram
 	messageMetrics      *MessageMetrics
@@ -210,7 +211,7 @@ func NewInterface(ctx context.Context, c *InterfaceConfig) (*Interface, error) {
 // activate creates the interface on the host. After the interface is created, any
 // other services that want to bind listeners to its IP may do so successfully. However,
 // the interface isn't going to process anything until run() is called.
-func (f *Interface) activate() {
+func (f *Interface) activate() error {
 	// actually turn on tun dev
 
 	addr, err := f.outside.LocalAddr()
@@ -238,33 +239,38 @@ func (f *Interface) activate() {
 		if i > 0 {
 			reader, err = f.inside.NewMultiQueueReader()
 			if err != nil {
-				f.l.Fatal(err)
+				return err
 			}
 		}
 		f.readers[i] = reader
 	}
 
-	if err := f.inside.Activate(); err != nil {
+	if err = f.inside.Activate(); err != nil {
 		f.inside.Close()
-		f.l.Fatal(err)
+		return err
 	}
+
+	return nil
 }
 
-func (f *Interface) run() {
+func (f *Interface) run() (func(), error) {
 	// Launch n queues to read packets from udp
 	for i := 0; i < f.routines; i++ {
+		f.wg.Add(1)
 		go f.listenOut(i)
 	}
 
 	// Launch n queues to read packets from tun dev
 	for i := 0; i < f.routines; i++ {
+		f.wg.Add(1)
 		go f.listenIn(f.readers[i], i)
 	}
+
+	return f.wg.Wait, nil
 }
 
 func (f *Interface) listenOut(i int) {
 	runtime.LockOSThread()
-
 	var li udp.Conn
 	if i > 0 {
 		li = f.writers[i]
@@ -279,14 +285,21 @@ func (f *Interface) listenOut(i int) {
 	fwPacket := &firewall.Packet{}
 	nb := make([]byte, 12, 12)
 
-	li.ListenOut(func(fromUdpAddr netip.AddrPort, payload []byte) {
+	err := 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))
 	})
+
+	if err != nil && !f.closed.Load() {
+		f.l.WithError(err).Error("Error while reading packet inbound packet, closing")
+		//TODO: Trigger Control to close
+	}
+
+	f.l.Debugf("underlay reader %v is done", i)
+	f.wg.Done()
 }
 
 func (f *Interface) listenIn(reader io.ReadWriteCloser, i int) {
 	runtime.LockOSThread()
-
 	packet := make([]byte, mtu)
 	out := make([]byte, mtu)
 	fwPacket := &firewall.Packet{}
@@ -297,17 +310,18 @@ func (f *Interface) listenIn(reader io.ReadWriteCloser, i int) {
 	for {
 		n, err := reader.Read(packet)
 		if err != nil {
-			if errors.Is(err, os.ErrClosed) && f.closed.Load() {
-				return
+			if !f.closed.Load() {
+				f.l.WithError(err).Error("Error while reading outbound packet, closing")
+				//TODO: Trigger Control to close
 			}
-
-			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)
+			break
 		}
 
 		f.consumeInsidePacket(packet[:n], fwPacket, nb, out, i, conntrackCache.Get(f.l))
 	}
+
+	f.l.Debugf("overlay reader %v is done", i)
+	f.wg.Done()
 }
 
 func (f *Interface) RegisterConfigChangeCallbacks(c *config.C) {
@@ -484,21 +498,23 @@ func (f *Interface) GetCertState() *CertState {
 func (f *Interface) Close() error {
 	f.closed.Store(true)
 
+	// Release the udp readers
 	for _, u := range f.writers {
 		err := u.Close()
 		if err != nil {
 			f.l.WithError(err).Error("Error while closing udp socket")
 		}
 	}
+
+	// Release the tun readers
 	for i, r := range f.readers {
 		if i == 0 {
-			continue // f.readers[0] is f.inside, which we want to save for last
+			continue // f.readers[0] is f.inside, which we want to save for last, since it closes other stuff too
 		}
 		if err := r.Close(); err != nil {
 			f.l.WithError(err).Error("Error while closing tun reader")
 		}
 	}
 
-	// Release the tun device
 	return f.inside.Close()
 }

main.go

@@ -292,15 +292,15 @@ func Main(c *config.C, configTest bool, buildVersion string, logger *logrus.Logg
 	}
 
 	return &Control{
-		ifce,
-		l,
-		ctx,
-		cancel,
-		sshStart,
-		statsStart,
-		dnsStart,
-		lightHouse.StartUpdateWorker,
-		connManager.Start,
+		f:                      ifce,
+		l:                      l,
+		ctx:                    ctx,
+		cancel:                 cancel,
+		sshStart:               sshStart,
+		statsStart:             statsStart,
+		dnsStart:               dnsStart,
+		lighthouseStart:        lightHouse.StartUpdateWorker,
+		connectionManagerStart: connManager.Start,
 	}, nil
 }
 

overlay/tun_linux.go

@@ -72,6 +72,11 @@ type ifreqQLEN struct {
 }
 
 func newTunFromFd(c *config.C, l *logrus.Logger, deviceFd int, vpnNetworks []netip.Prefix) (*tun, error) {
+	err := unix.SetNonblock(deviceFd, true)
+	if err != nil {
+		return nil, err
+	}
+
 	file := os.NewFile(uintptr(deviceFd), "/dev/net/tun")
 
 	t, err := newTunGeneric(c, l, file, vpnNetworks)
@@ -122,6 +127,11 @@ func newTun(c *config.C, l *logrus.Logger, vpnNetworks []netip.Prefix, multiqueu
 	}
 	name := strings.Trim(string(req.Name[:]), "\x00")
 
+	err = unix.SetNonblock(fd, true)
+	if err != nil {
+		return nil, err
+	}
+
 	file := os.NewFile(uintptr(fd), "/dev/net/tun")
 	t, err := newTunGeneric(c, l, file, vpnNetworks)
 	if err != nil {
@@ -145,7 +155,12 @@ func newTunGeneric(c *config.C, l *logrus.Logger, file *os.File, vpnNetworks []n
 		l:                         l,
 	}
 
-	err := t.reload(c, true)
+	err := unix.SetNonblock(t.fd, true)
+	if err != nil {
+		return nil, err
+	}
+
+	err = t.reload(c, true)
 	if err != nil {
 		return nil, err
 	}
@@ -251,6 +266,11 @@ func (t *tun) NewMultiQueueReader() (io.ReadWriteCloser, error) {
 		return nil, err
 	}
 
+	err = unix.SetNonblock(fd, true)
+	if err != nil {
+		return nil, err
+	}
+
 	file := os.NewFile(uintptr(fd), "/dev/net/tun")
 
 	return file, nil
@@ -261,29 +281,6 @@ func (t *tun) RoutesFor(ip netip.Addr) routing.Gateways {
 	return r
 }
 
-func (t *tun) Write(b []byte) (int, error) {
-	var nn int
-	maximum := len(b)
-
-	for {
-		n, err := unix.Write(t.fd, b[nn:maximum])
-		if n > 0 {
-			nn += n
-		}
-		if nn == len(b) {
-			return nn, err
-		}
-
-		if err != nil {
-			return nn, err
-		}
-
-		if n == 0 {
-			return nn, io.ErrUnexpectedEOF
-		}
-	}
-}
-
 func (t *tun) deviceBytes() (o [16]byte) {
 	for i, c := range t.Device {
 		o[i] = byte(c)
@@ -712,11 +709,17 @@ func (t *tun) Close() error {
 	}
 
 	if t.ReadWriteCloser != nil {
-		_ = t.ReadWriteCloser.Close()
+		err := t.ReadWriteCloser.Close()
+		if err != nil {
+			t.l.WithField("error", err).Error("Failed to close read/write connection")
+		}
 	}
 
 	if t.ioctlFd > 0 {
-		_ = os.NewFile(t.ioctlFd, "ioctlFd").Close()
+		err := os.NewFile(t.ioctlFd, "ioctlFd").Close()
+		if err != nil {
+			t.l.WithField("error", err).Error("Failed to close ioctl fd")
+		}
 		t.ioctlFd = 0
 	}
 

service/service.go

@@ -44,7 +44,10 @@ type Service struct {
 }
 
 func New(control *nebula.Control) (*Service, error) {
-	control.Start()
+	wait, err := control.Start()
+	if err != nil {
+		return nil, err
+	}
 
 	ctx := control.Context()
 	eg, ctx := errgroup.WithContext(ctx)
@@ -141,6 +144,12 @@ func New(control *nebula.Control) (*Service, error) {
 		}
 	})
 
+	// Add the nebula wait function to the group
+	eg.Go(func() error {
+		wait()
+		return nil
+	})
+
 	return &s, nil
 }
 

udp/conn.go

@@ -16,7 +16,7 @@ type EncReader func(
 type Conn interface {
 	Rebind() error
 	LocalAddr() (netip.AddrPort, error)
-	ListenOut(r EncReader)
+	ListenOut(r EncReader) error
 	WriteTo(b []byte, addr netip.AddrPort) error
 	ReloadConfig(c *config.C)
 	SupportsMultipleReaders() bool
@@ -31,8 +31,8 @@ func (NoopConn) Rebind() error {
 func (NoopConn) LocalAddr() (netip.AddrPort, error) {
 	return netip.AddrPort{}, nil
 }
-func (NoopConn) ListenOut(_ EncReader) {
-	return
+func (NoopConn) ListenOut(_ EncReader) error {
+	return nil
 }
 func (NoopConn) SupportsMultipleReaders() bool {
 	return false

udp/udp_darwin.go

@@ -165,7 +165,7 @@ func NewUDPStatsEmitter(udpConns []Conn) func() {
 	return func() {}
 }
 
-func (u *StdConn) ListenOut(r EncReader) {
+func (u *StdConn) ListenOut(r EncReader) error {
 	buffer := make([]byte, MTU)
 
 	for {
@@ -173,8 +173,7 @@ func (u *StdConn) ListenOut(r EncReader) {
 		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
+				return err
 			}
 
 			u.l.WithError(err).Error("unexpected udp socket receive error")

udp/udp_generic.go

@@ -10,11 +10,9 @@ package udp
 
 import (
 	"context"
-	"errors"
 	"fmt"
 	"net"
 	"net/netip"
-	"time"
 
 	"github.com/sirupsen/logrus"
 	"github.com/slackhq/nebula/config"
@@ -73,25 +71,14 @@ type rawMessage struct {
 	Len uint32
 }
 
-func (u *GenericConn) ListenOut(r EncReader) {
+func (u *GenericConn) ListenOut(r EncReader) error {
 	buffer := make([]byte, MTU)
 
-	var lastRecvErr time.Time
-
 	for {
 		// Just read one packet at a time
 		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
-			}
-			// Dampen unexpected message warns to once per minute
-			if lastRecvErr.IsZero() || time.Since(lastRecvErr) > time.Minute {
-				lastRecvErr = time.Now()
-				u.l.WithError(err).Warn("unexpected udp socket receive error")
-			}
-			continue
+			return err
 		}
 
 		r(netip.AddrPortFrom(rua.Addr().Unmap(), rua.Port()), buffer[:n])

udp/udp_linux.go

@@ -9,6 +9,7 @@ import (
 	"net"
 	"net/netip"
 	"syscall"
+	"time"
 	"unsafe"
 
 	"github.com/rcrowley/go-metrics"
@@ -17,6 +18,8 @@ import (
 	"golang.org/x/sys/unix"
 )
 
+var readTimeout = unix.NsecToTimeval(int64(3 * time.Second))
+
 type StdConn struct {
 	sysFd int
 	isV4  bool
@@ -24,14 +27,6 @@ type StdConn struct {
 	batch int
 }
 
-func maybeIPV4(ip net.IP) (net.IP, bool) {
-	ip4 := ip.To4()
-	if ip4 != nil {
-		return ip4, true
-	}
-	return ip, false
-}
-
 func NewListener(l *logrus.Logger, ip netip.Addr, port int, multi bool, batch int) (Conn, error) {
 	af := unix.AF_INET6
 	if ip.Is4() {
@@ -55,6 +50,11 @@ func NewListener(l *logrus.Logger, ip netip.Addr, port int, multi bool, batch in
 		}
 	}
 
+	// Set a read timeout
+	if err = unix.SetsockoptTimeval(fd, unix.SOL_SOCKET, unix.SO_RCVTIMEO, &readTimeout); err != nil {
+		return nil, fmt.Errorf("unable to set SO_RCVTIMEO: %s", err)
+	}
+
 	var sa unix.Sockaddr
 	if ip.Is4() {
 		sa4 := &unix.SockaddrInet4{Port: port}
@@ -122,7 +122,7 @@ func (u *StdConn) LocalAddr() (netip.AddrPort, error) {
 	}
 }
 
-func (u *StdConn) ListenOut(r EncReader) {
+func (u *StdConn) ListenOut(r EncReader) error {
 	var ip netip.Addr
 
 	msgs, buffers, names := u.PrepareRawMessages(u.batch)
@@ -134,8 +134,7 @@ func (u *StdConn) ListenOut(r EncReader) {
 	for {
 		n, err := read(msgs)
 		if err != nil {
-			u.l.WithError(err).Debug("udp socket is closed, exiting read loop")
-			return
+			return err
 		}
 
 		for i := 0; i < n; i++ {
@@ -163,6 +162,9 @@ func (u *StdConn) ReadSingle(msgs []rawMessage) (int, error) {
 		)
 
 		if err != 0 {
+			if err == unix.EAGAIN || err == unix.EINTR || err == unix.EWOULDBLOCK {
+				continue
+			}
 			return 0, &net.OpError{Op: "recvmsg", Err: err}
 		}
 
@@ -182,8 +184,14 @@ func (u *StdConn) ReadMulti(msgs []rawMessage) (int, error) {
 			0,
 			0,
 		)
-
-		if err != 0 {
+		if err == unix.EAGAIN || err == unix.EINTR || err == unix.EWOULDBLOCK {
+			if int64(n) > 0 {
+				//ran out of time, but have some messages to return
+				return int(n), nil
+			} else {
+				continue
+			}
+		} else if err != 0 {
 			return 0, &net.OpError{Op: "recvmmsg", Err: err}
 		}
 

udp/udp_rio_windows.go

@@ -140,7 +140,7 @@ func (u *RIOConn) bind(l *logrus.Logger, sa windows.Sockaddr) error {
 	return nil
 }
 
-func (u *RIOConn) ListenOut(r EncReader) {
+func (u *RIOConn) ListenOut(r EncReader) error {
 	buffer := make([]byte, MTU)
 
 	var lastRecvErr time.Time
@@ -151,8 +151,7 @@ func (u *RIOConn) ListenOut(r EncReader) {
 
 		if err != nil {
 			if errors.Is(err, net.ErrClosed) {
-				u.l.WithError(err).Debug("udp socket is closed, exiting read loop")
-				return
+				return err
 			}
 			// Dampen unexpected message warns to once per minute
 			if lastRecvErr.IsZero() || time.Since(lastRecvErr) > time.Minute {

udp/udp_tester.go

@@ -6,6 +6,7 @@ package udp
 import (
 	"io"
 	"net/netip"
+	"os"
 	"sync/atomic"
 
 	"github.com/sirupsen/logrus"
@@ -106,11 +107,11 @@ func (u *TesterConn) WriteTo(b []byte, addr netip.AddrPort) error {
 	return nil
 }
 
-func (u *TesterConn) ListenOut(r EncReader) {
+func (u *TesterConn) ListenOut(r EncReader) error {
 	for {
 		p, ok := <-u.RxPackets
 		if !ok {
-			return
+			return os.ErrClosed
 		}
 		r(p.From, p.Data)
 	}