Compare commits

..
Author SHA1 Message Date
Ryan Huber d3e21a8fc8 relay: Batch forwarded packets on the relay host 2026-10-17 18:53:36 +00:00
Wade Simmons cda3710d5a fix error return in handleOutsideMessagePacket (#1970)
smoke-extra / freebsd-amd64 (push) Failing after 17s
smoke-extra / linux-amd64-ipv6disable (push) Failing after 16s
smoke-extra / netbsd-amd64 (push) Failing after 16s
smoke-extra / openbsd-amd64 (push) Failing after 16s
smoke-extra / linux-386 (push) Failing after 17s
smoke / Run multi node smoke test (push) Failing after 1m39s
Build and test / Static checks (push) Successful in 45s
Build and test / Test linux (push) Failing after 1m18s
Build and test / Test linux-pkcs11 (push) Failing after 2m5s
Build and test / Test linux-boringcrypto (push) Failing after 3m1s
Build and test / Test linux-fips140 (push) Failing after 2m55s
Build and test / Cross-build linux-arm (push) Successful in 3m0s
Build and test / Cross-build linux-mips (push) Successful in 3m42s
Build and test / Cross-build linux-other (push) Successful in 3m5s
Build and test / Cross-build windows (push) Successful in 1m2s
Build and test / Cross-build freebsd (push) Successful in 1m32s
Build and test / Cross-build netbsd (push) Successful in 1m31s
Build and test / Cross-build openbsd (push) Successful in 1m32s
Build and test / Cross-build mobile (push) Successful in 3m21s
smoke-extra / Run windows smoke test (push) Canceled after 0s
smoke / Run self traffic smoke test on macOS (push) Canceled after 0s
Build and test / Test macos (push) Canceled after 0s
Build and test / Test windows (push) Canceled after 0s
Build and test / CI status (push) Canceled after 0s
Introduced by #1961, we should still return early for invalid packets
2026-10-17 11:09:25 -04:00
John Maguire c26871fe08 service: Close listeners and tear down the netstack on shutdown (#1954) 2026-10-17 11:08:29 -04:00
Caleb Jasik 7daacc6a48 overlay/darwin: test tun framing and zero-alloc Read/Write (#1906)
smoke-extra / freebsd-amd64 (push) Failing after 16s
smoke-extra / linux-amd64-ipv6disable (push) Failing after 16s
smoke-extra / netbsd-amd64 (push) Failing after 17s
smoke-extra / openbsd-amd64 (push) Failing after 17s
smoke-extra / linux-386 (push) Failing after 16s
smoke / Run multi node smoke test (push) Failing after 1m41s
Build and test / Static checks (push) Successful in 45s
Build and test / Test linux (push) Failing after 1m47s
Build and test / Test linux-pkcs11 (push) Failing after 2m6s
Build and test / Test linux-boringcrypto (push) Failing after 2m57s
Build and test / Test linux-fips140 (push) Failing after 2m53s
Build and test / Cross-build linux-arm (push) Successful in 3m7s
Build and test / Cross-build linux-mips (push) Successful in 3m55s
Build and test / Cross-build linux-other (push) Successful in 3m13s
Build and test / Cross-build windows (push) Successful in 1m4s
Build and test / Cross-build freebsd (push) Successful in 1m35s
Build and test / Cross-build netbsd (push) Successful in 1m35s
Build and test / Cross-build openbsd (push) Successful in 1m34s
Build and test / Cross-build mobile (push) Successful in 3m26s
smoke-extra / Run windows smoke test (push) Canceled after 0s
smoke / Run self traffic smoke test on macOS (push) Canceled after 0s
Build and test / Test macos (push) Canceled after 0s
Build and test / Test windows (push) Canceled after 0s
Build and test / CI status (push) Canceled after 0s
2026-10-16 18:12:58 -05:00
John Maguire 428dad94b3 Reject v2 cert names and groups at signing the parser rejects (#1967)
smoke-extra / freebsd-amd64 (push) Failing after 18s
smoke-extra / linux-amd64-ipv6disable (push) Failing after 18s
smoke-extra / netbsd-amd64 (push) Failing after 16s
smoke-extra / openbsd-amd64 (push) Failing after 17s
smoke-extra / linux-386 (push) Failing after 17s
smoke / Run multi node smoke test (push) Failing after 1m41s
Build and test / Static checks (push) Successful in 45s
Build and test / Test linux (push) Failing after 1m19s
Build and test / Test linux-pkcs11 (push) Failing after 2m5s
Build and test / Test linux-boringcrypto (push) Failing after 2m56s
Build and test / Test linux-fips140 (push) Failing after 2m51s
Build and test / Cross-build linux-arm (push) Successful in 3m0s
Build and test / Cross-build linux-mips (push) Successful in 3m43s
Build and test / Cross-build linux-other (push) Successful in 3m7s
Build and test / Cross-build windows (push) Successful in 1m2s
Build and test / Cross-build freebsd (push) Successful in 1m32s
Build and test / Cross-build netbsd (push) Successful in 1m33s
Build and test / Cross-build openbsd (push) Successful in 1m32s
Build and test / Cross-build mobile (push) Successful in 3m23s
smoke-extra / Run windows smoke test (push) Canceled after 0s
smoke / Run self traffic smoke test on macOS (push) Canceled after 0s
Build and test / Test macos (push) Canceled after 0s
Build and test / Test windows (push) Canceled after 0s
Build and test / CI status (push) Canceled after 0s
2026-10-16 18:05:51 -04:00
18 changed files with 831 additions and 114 deletions

No files matched your search

+2
View File
@@ -14,6 +14,8 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
### Fixed
- Signing a v2 certificate with an empty name, a name longer than 253 bytes, or an empty group now fails.
These certificates were signed successfully but could not be loaded.
- `create-tunnel -address <ip:port>` panicked on a nil RemoteList and, over the ssh debug server, took nebula
down with it. A hostinfo returned by StartHandshake had no RemoteList until the handshake loop's first
pass attached one, and `create-tunnel -address` set the remote before that. The list is now attached when
+10 -1
View File
@@ -389,7 +389,16 @@ func (c *certificateV2) fromTBSCertificate(t *TBSCertificate) error {
}
func (c *certificateV2) validate() error {
// Empty names are allowed
// Mirror the name and group rules in unmarshalDetails so signing fails instead of producing an unparseable certificate
if c.details.name == "" || len(c.details.name) > MaxNameLength {
return NewErrInvalidCertificateProperties("name must be between 1 and %d bytes", MaxNameLength)
}
for _, group := range c.details.groups {
if group == "" {
return NewErrInvalidCertificateProperties("groups may not be empty")
}
}
if len(c.publicKey) == 0 {
return ErrInvalidPublicKey
+45
View File
@@ -6,6 +6,7 @@ import (
"crypto/elliptic"
"crypto/rand"
"net/netip"
"strings"
"testing"
"time"
@@ -135,3 +136,47 @@ func TestCertificate_SignP256_AlwaysNormalized(t *testing.T) {
assert.True(t, normie)
}
}
func TestCertificateV2_SignRejectsUnparseable(t *testing.T) {
_, priv, err := ed25519.GenerateKey(rand.Reader)
require.NoError(t, err)
var invalid *ErrInvalidCertificateProperties
newTBS := func() TBSCertificate {
return TBSCertificate{
Version: Version2,
Name: "testing",
Networks: []netip.Prefix{mustParsePrefixUnmapped("10.1.1.1/24")},
Groups: []string{"test-group1"},
NotBefore: time.Now().Add(-time.Minute),
NotAfter: time.Now().Add(time.Minute),
PublicKey: []byte("1234567890abcedfghij1234567890ab"),
Curve: Curve_CURVE25519,
IsCA: true,
}
}
tbs := newTBS()
tbs.Name = strings.Repeat("a", MaxNameLength)
c, err := tbs.Sign(nil, Curve_CURVE25519, priv)
require.NoError(t, err)
b, err := c.Marshal()
require.NoError(t, err)
_, err = unmarshalCertificateV2(b, nil, Curve_CURVE25519)
require.NoError(t, err)
tbs = newTBS()
tbs.Name = strings.Repeat("a", MaxNameLength+1)
_, err = tbs.Sign(nil, Curve_CURVE25519, priv)
require.ErrorAs(t, err, &invalid)
tbs = newTBS()
tbs.Name = ""
_, err = tbs.Sign(nil, Curve_CURVE25519, priv)
require.ErrorAs(t, err, &invalid)
tbs = newTBS()
tbs.Groups = []string{"test-group1", ""}
_, err = tbs.Sign(nil, Curve_CURVE25519, priv)
require.ErrorAs(t, err, &invalid)
}
+17 -22
View File
@@ -65,7 +65,7 @@ type HandshakeManager struct {
lightHouse *LightHouse
outside udp.Conn
config HandshakeConfig
OutboundHandshakeTimer *LockingTimerWheel[*HandshakeHostInfo]
OutboundHandshakeTimer *LockingTimerWheel[netip.Addr]
messageMetrics *MessageMetrics
metricInitiated metrics.Counter
metricTimedOut metrics.Counter
@@ -81,7 +81,6 @@ type HandshakeHostInfo struct {
startTime time.Time // Time that we first started trying with this handshake
ready bool // Is the handshake ready
deleted bool // Has the handshake been removed from the pending maps, deleters hold the lock
initiatingVersionOverride cert.Version // Should we use a non-default cert version for this handshake?
counter int64 // How many attempts have we made so far
lastRemotes []netip.AddrPort // Remotes that we sent to during the previous attempt
@@ -126,7 +125,7 @@ func NewHandshakeManager(l *slog.Logger, mainHostMap *HostMap, lightHouse *Light
outside: outside,
config: config,
trigger: make(chan netip.Addr, config.triggerBuffer),
OutboundHandshakeTimer: NewLockingTimerWheel[*HandshakeHostInfo](config.tryInterval, hsTimeout(config.retries, config.tryInterval)),
OutboundHandshakeTimer: NewLockingTimerWheel[netip.Addr](config.tryInterval, hsTimeout(config.retries, config.tryInterval)),
messageMetrics: config.messageMetrics,
metricInitiated: metrics.GetOrRegisterCounter("handshake_manager.initiated", nil),
metricTimedOut: metrics.GetOrRegisterCounter("handshake_manager.timed_out", nil),
@@ -143,9 +142,7 @@ func (hm *HandshakeManager) Run(ctx context.Context) {
case <-ctx.Done():
return
case vpnIP := <-hm.trigger:
if hh := hm.queryVpnIp(vpnIP); hh != nil {
hm.handleOutbound(hh, true)
}
hm.handleOutbound(vpnIP, true)
case now := <-clockSource.C:
hm.NextOutboundHandshakeTimerTick(now)
}
@@ -200,25 +197,23 @@ func (hm *HandshakeManager) HandleIncoming(via ViaSender, packet []byte, h *head
func (hm *HandshakeManager) NextOutboundHandshakeTimerTick(now time.Time) {
hm.OutboundHandshakeTimer.Advance(now)
for {
hh, has := hm.OutboundHandshakeTimer.Purge()
vpnIp, has := hm.OutboundHandshakeTimer.Purge()
if !has {
break
}
hm.handleOutbound(hh, false)
hm.handleOutbound(vpnIp, false)
}
}
func (hm *HandshakeManager) handleOutbound(hh *HandshakeHostInfo, lighthouseTriggered bool) {
func (hm *HandshakeManager) handleOutbound(vpnIp netip.Addr, lighthouseTriggered bool) {
hh := hm.queryVpnIp(vpnIp)
if hh == nil {
return
}
hh.Lock()
defer hh.Unlock()
// Timer entries outlive their handshake. A new handshake for the same vpn addr runs its own timer chain.
if hh.deleted {
return
}
hostinfo := hh.hostinfo
vpnIp := hostinfo.vpnAddrs[0]
// If we are out of time, clean up
if hh.counter >= hm.config.retries {
fields := []any{
@@ -244,14 +239,10 @@ func (hm *HandshakeManager) handleOutbound(hh *HandshakeHostInfo, lighthouseTrig
// Increment the counter to increase our delay, linear backoff
hh.counter++
// A lighthouse triggered attempt is still in the timer wheel, adding it again would start a second retry chain
if !lighthouseTriggered {
defer hm.OutboundHandshakeTimer.Add(hh, hm.config.tryInterval*time.Duration(hh.counter))
}
// Check if we have a handshake packet to transmit yet
if !hh.ready {
if !hm.buildStage0Packet(hh) {
hm.OutboundHandshakeTimer.Add(vpnIp, hm.config.tryInterval*time.Duration(hh.counter))
return
}
}
@@ -332,6 +323,11 @@ func (hm *HandshakeManager) handleOutbound(hh *HandshakeHostInfo, lighthouseTrig
}
hm.f.relayManager.StartRelays(hm.f, vpnIp, hh, stage0)
// If a lighthouse triggered this attempt then we are still in the timer wheel and do not need to re-add
if !lighthouseTriggered {
hm.OutboundHandshakeTimer.Add(vpnIp, hm.config.tryInterval*time.Duration(hh.counter))
}
}
// GetOrHandshake will try to find a hostinfo with a fully formed tunnel or start a new handshake if one is not present
@@ -384,7 +380,7 @@ func (hm *HandshakeManager) StartHandshake(vpnAddr netip.Addr, cacheCb func(*Han
}
hm.vpnIps[vpnAddr] = hh
hm.metricInitiated.Inc(1)
hm.OutboundHandshakeTimer.Add(hh, hm.config.tryInterval)
hm.OutboundHandshakeTimer.Add(vpnAddr, hm.config.tryInterval)
if cacheCb != nil {
cacheCb(hh)
@@ -536,7 +532,6 @@ func (hm *HandshakeManager) DeleteHostInfo(hostinfo *HostInfo) {
func (hm *HandshakeManager) unlockedDeleteHostInfo(hostinfo *HostInfo) {
for _, addr := range hostinfo.vpnAddrs {
if cur, ok := hm.vpnIps[addr]; ok && cur.hostinfo == hostinfo {
cur.deleted = true
delete(hm.vpnIps, addr)
}
}
+1 -63
View File
@@ -65,69 +65,7 @@ func Test_NewHandshakeManagerVpnIp(t *testing.T) {
assert.NotContains(t, blah.vpnIps, ip)
}
// newTestHandshakeManager returns a manager whose handshakes never build a stage 0 packet (the cert has no key), so
// every attempt just bumps the counter and schedules the next one. tick advances its timer wheel by one tryInterval.
func newTestHandshakeManager() (hm *HandshakeManager, tick func()) {
l := test.NewLogger()
mainHM := newHostMap(l)
mainHM.preferredRanges.Store(&[]netip.Prefix{})
hm = NewHandshakeManager(l, mainHM, newTestLighthouse(), &udp.NoopConn{}, defaultHandshakeConfig)
hm.f = &Interface{handshakeManager: hm, pki: &PKI{}, l: l}
hm.f.pki.cs.Store(&CertState{
initiatingVersion: cert.Version1,
privateKey: []byte{},
v1Cert: &dummyCert{version: cert.Version1},
})
now := time.Now()
hm.NextOutboundHandshakeTimerTick(now)
return hm, func() {
now = now.Add(DefaultHandshakeTryInterval)
hm.NextOutboundHandshakeTimerTick(now)
}
}
func Test_HandshakeManager_RestartDropsOldTimer(t *testing.T) {
ip := netip.MustParseAddr("172.1.1.2")
hm, tick := newTestHandshakeManager()
hm.StartHandshake(ip, nil)
old := hm.queryVpnIp(ip)
tick()
tick()
require.Equal(t, int64(1), old.counter)
// Restart the handshake while the old one still has a retry queued
hm.DeleteHostInfo(old.hostinfo)
hm.StartHandshake(ip, nil)
hh := hm.queryVpnIp(ip)
require.NotSame(t, old, hh)
// An entry for timeout d fires d/tryInterval+1 ticks after it is added, so linear backoff lands attempts on ticks
// 2, 4 and 7
var counters []int64
for range 7 {
tick()
counters = append(counters, hh.counter)
}
assert.Equal(t, []int64{0, 1, 1, 2, 2, 2, 3}, counters)
// The old handshake's queued retry ended its chain instead of retrying a deleted handshake
assert.Equal(t, int64(1), old.counter)
assert.Equal(t, 1, testCountTimerWheelEntries(hm.OutboundHandshakeTimer))
}
func Test_HandshakeManager_LighthouseTriggerKeepsOneTimer(t *testing.T) {
ip := netip.MustParseAddr("172.1.1.2")
hm, _ := newTestHandshakeManager()
hm.StartHandshake(ip, nil)
hm.handleOutbound(hm.queryVpnIp(ip), true)
assert.Equal(t, 1, testCountTimerWheelEntries(hm.OutboundHandshakeTimer))
}
func testCountTimerWheelEntries(tw *LockingTimerWheel[*HandshakeHostInfo]) (c int) {
func testCountTimerWheelEntries(tw *LockingTimerWheel[netip.Addr]) (c int) {
for _, i := range tw.t.wheel {
n := i.Head
for n != nil {
+3
View File
@@ -384,6 +384,8 @@ type rxContext struct {
hostmapCache map[uint32]*HostInfo
lhh *LightHouseHandler
ctCache *firewall.ConntrackCacheTicker
relayBatch *batch.SendBatch
relayRemote netip.AddrPort
}
func newRxContext(f *Interface, q int) *rxContext {
@@ -414,6 +416,7 @@ func (f *Interface) listenOut(i int) {
}
flusher := func() {
f.flushRelayBatch(rxc)
if err := f.batchers[i].Flush(); err != nil {
f.l.Error("Failed to flush tun coalescer", "error", err)
}
+35 -7
View File
@@ -25,7 +25,7 @@ var ErrOutOfWindow = errors.New("out of window packet")
// readOutsidePackets processes one received underlay packet.
// Message payloads are decrypted IN PLACE, so packet must stay untouched
// by the caller until the batcher for queue q has been flushed
// by the caller until the TUN and relay batches for queue q have been flushed.
func (f *Interface) readOutsidePackets(via ViaSender, packet []byte, rxc *rxContext) {
h := rxc.h
err := h.Parse(packet)
@@ -77,6 +77,12 @@ func (f *Interface) readOutsidePackets(via ViaSender, packet []byte, rxc *rxCont
f.messageMetrics.Rx(h.Type, h.Subtype, 1)
}
// Earlier forwards must leave before this packet can generate a response.
isMessageRelay := h.Type == header.Message && h.Subtype == header.MessageRelay
if !isMessageRelay {
f.flushRelayBatch(rxc)
}
// Unencrypted packets
switch h.Type {
case header.Handshake:
@@ -88,9 +94,6 @@ func (f *Interface) readOutsidePackets(via ViaSender, packet []byte, rxc *rxCont
return
}
// Relay packets are special
isMessageRelay := (h.Type == header.Message && h.Subtype == header.MessageRelay)
var hostinfo *HostInfo
if isMessageRelay {
hostinfo = f.hostMap.QueryRelayIndex(h.RemoteIndex)
@@ -102,6 +105,8 @@ func (f *Interface) readOutsidePackets(via ViaSender, packet []byte, rxc *rxCont
// recvError if necessary
if hostinfo == nil || hostinfo.ConnectionState == nil {
if !via.IsRelayed {
// An unknown relay index can also produce a response.
f.flushRelayBatch(rxc)
f.maybeSendRecvError(via.UdpAddr, h.RemoteIndex)
}
return
@@ -233,8 +238,10 @@ func (f *Interface) handleOutsideRelayPacket(hostinfo *HostInfo, via ViaSender,
// Forward this packet through the relay tunnel, rebuilding it in place.
// Encode overwrites the old outer header, and the new AEAD tag lands where the old one was
fwdBuf := packet[:0]
//todo it would potentially be nice to batch these
f.SendVia(targetHI, targetRelay, signedPayload, rxc.nb, fwdBuf, true, rxc.q)
out, err := f.prepareSendVia(targetHI, targetRelay, signedPayload, rxc.nb, fwdBuf, true)
if err == nil {
f.queueRelayPacket(out, targetHI.GetRemote(), rxc)
}
case TerminalType:
hostinfo.logger(f.l).Error("Unexpected Relay Type of Terminal")
return
@@ -259,6 +266,27 @@ func (f *Interface) handleOutsideRelayPacket(hostinfo *HostInfo, via ViaSender,
}
}
// queueRelayPacket borrows a receive buffer until the batch is flushed. Destination changes and the send batch
// limit end a run, preserving arrival order while allowing WriteBatch to use UDP segmentation offload.
func (f *Interface) queueRelayPacket(packet []byte, remote netip.AddrPort, rxc *rxContext) {
if rxc.relayBatch == nil {
rxc.relayBatch = batch.NewSendBatch(f.writers[rxc.q], batch.SendBatchCap, 0)
} else if rxc.relayRemote != remote {
f.flushRelayBatch(rxc)
}
rxc.relayRemote = remote
rxc.relayBatch.Commit(packet, remote)
if rxc.relayBatch.Len() >= batch.SendBatchCap {
f.flushRelayBatch(rxc)
}
}
func (f *Interface) flushRelayBatch(rxc *rxContext) {
if rxc.relayBatch != nil {
f.flushSendBatch(rxc.relayBatch, rxc.q)
}
}
// closeTunnel closes a tunnel locally, it does not send a closeTunnel packet to the remote
func (f *Interface) closeTunnel(hostInfo *HostInfo) {
// Hold the hostmap lock across both the delete and the lighthouse eviction,
@@ -478,8 +506,8 @@ func (f *Interface) handleOutsideMessagePacket(hostinfo *HostInfo, messageCounte
if err != nil {
if f.l.Enabled(context.Background(), slog.LevelDebug) {
hostinfo.logger(f.l).Debug("Error while validating inbound packet", "error", err, "packet", out)
return
}
return
}
dropReason := f.firewall.Drop(rxc.fwPacket.Packet, true, hostinfo, f.pki.GetCAPool(), rxc.ctCache.Get())
+291
View File
@@ -0,0 +1,291 @@
package nebula
import (
"bytes"
"net/netip"
"sync"
"testing"
"github.com/flynn/noise"
"github.com/gaissmai/bart"
"github.com/rcrowley/go-metrics"
"github.com/slackhq/nebula/header"
"github.com/slackhq/nebula/noiseutil"
"github.com/slackhq/nebula/overlay/batch"
"github.com/slackhq/nebula/test"
"github.com/slackhq/nebula/udp"
"github.com/stretchr/testify/require"
)
type relayBatchWrite struct {
packets [][]byte
remotes []netip.AddrPort
batched bool
}
type relayBatchConn struct {
udp.NoopConn
read func(udp.EncReader, func())
writes []relayBatchWrite
}
func (c *relayBatchConn) ListenOut(r udp.EncReader, flush func()) error {
c.read(r, flush)
return nil
}
func (c *relayBatchConn) WriteBatch(bufs [][]byte, addrs []netip.AddrPort) (int, error) {
w := relayBatchWrite{remotes: append([]netip.AddrPort(nil), addrs...), batched: true}
for _, b := range bufs {
w.packets = append(w.packets, bytes.Clone(b))
}
c.writes = append(c.writes, w)
return len(bufs), nil
}
func (c *relayBatchConn) WriteTo(b []byte, addr netip.AddrPort) error {
c.writes = append(c.writes, relayBatchWrite{packets: [][]byte{bytes.Clone(b)}, remotes: []netip.AddrPort{addr}})
return nil
}
func (c *relayBatchConn) packets() [][]byte {
var packets [][]byte
for _, w := range c.writes {
packets = append(packets, w.packets...)
}
return packets
}
type relayBatchPath struct {
f *Interface
conn *relayBatchConn
source, target *HostInfo
incoming, outgoing *Relay
sourceKey noiseutil.CipherState
cipher noise.CipherFunc
counter uint64
}
// Separate cipher instances let the reference SendVia path use the same keys and
// counters without rewinding an AEAD instance (including FIPS nonce checks).
func relayBatchKey(cipher noise.CipherFunc, id byte) noiseutil.CipherState {
suite := noise.NewCipherSuite(noise.DH25519, cipher, noise.HashSHA256)
return noiseutil.NewCipherState(noise.UnsafeNewCipherState(suite, [32]byte{id}, 0), cipher)
}
func newRelayBatchPath(t *testing.T, cipher noise.CipherFunc) *relayBatchPath {
t.Helper()
l := test.NewLogger()
conn := &relayBatchConn{}
f := &Interface{
ctx: t.Context(), l: l, hostMap: newHostMap(l), outside: conn,
writers: []udp.Conn{conn}, lightHouse: &LightHouse{l: l},
myVpnNetworksTable: &bart.Lite{}, metricTxDropped: metrics.NewCounter(),
batchers: []*batch.MultiCoalescer{batch.NewMultiCoalescer(&capturingTun{}, l)},
}
f.connectionManager = &connectionManager{
intf: f, relayUsed: map[uint32]struct{}{}, relayUsedLock: &sync.RWMutex{},
}
newHost := func(addr string, remote string, id byte) *HostInfo {
h := &HostInfo{
vpnAddrs: []netip.Addr{netip.MustParseAddr(addr)},
ConnectionState: &ConnectionState{
eKey: relayBatchKey(cipher, id), dKey: relayBatchKey(cipher, id), window: NewBits(ReplayWindow),
},
relayState: RelayState{relayForByAddr: map[netip.Addr]*Relay{}, relayForByIdx: map[uint32]*Relay{}},
}
u := netip.MustParseAddrPort(remote)
h.remote.Store(&u)
f.hostMap.Hosts[h.vpnAddrs[0]] = h
return h
}
source := newHost("192.0.2.1", "198.51.100.1:4242", 1)
target := newHost("192.0.2.2", "198.51.100.2:4242", 2)
incoming := &Relay{Type: ForwardingType, State: Established, LocalIndex: 301, RemoteIndex: 302, PeerAddr: target.vpnAddrs[0]}
outgoing := &Relay{Type: ForwardingType, State: Established, LocalIndex: 303, RemoteIndex: 304, PeerAddr: source.vpnAddrs[0]}
source.relayState.InsertRelay(incoming.PeerAddr, incoming.LocalIndex, incoming)
target.relayState.InsertRelay(outgoing.PeerAddr, outgoing.LocalIndex, outgoing)
f.hostMap.Relays[incoming.LocalIndex] = source
f.hostMap.Indexes[100] = source
return &relayBatchPath{f: f, conn: conn, source: source, target: target, incoming: incoming, outgoing: outgoing, sourceKey: relayBatchKey(cipher, 1), cipher: cipher}
}
func (p *relayBatchPath) wrap(t *testing.T, payload []byte) []byte {
t.Helper()
p.counter++
b := header.Encode(make([]byte, header.Len), header.Version, header.Message, header.MessageRelay, p.incoming.LocalIndex, p.counter)
b = append(b, payload...)
b, err := p.sourceKey.EncryptDanger(b, b, nil, p.counter, make([]byte, 12))
require.NoError(t, err)
return b
}
func (p *relayBatchPath) reference(payloads [][]byte) [][]byte {
conn := &relayBatchConn{}
f := &Interface{writers: []udp.Conn{conn}, connectionManager: p.f.connectionManager, l: p.f.l}
h := &HostInfo{ConnectionState: &ConnectionState{eKey: relayBatchKey(p.cipher, 2)}}
u := p.target.GetRemote()
h.remote.Store(&u)
for _, payload := range payloads {
f.SendVia(h, p.outgoing, payload, make([]byte, 12), make([]byte, 0, len(payload)+header.MaxOverhead), false, 0)
}
return conn.packets()
}
func TestRelayReceiveBatch(t *testing.T) {
for _, cipher := range []noise.CipherFunc{noiseutil.CipherAESGCM, noise.CipherChaChaPoly} {
t.Run(cipher.CipherName(), func(t *testing.T) {
p := newRelayBatchPath(t, cipher)
const count = 8
var payloads, wire [][]byte
for i := range count * 2 {
payload := bytes.Repeat([]byte{byte(i)}, 64)
payloads = append(payloads, payload)
wire = append(wire, p.wrap(t, payload))
}
want := p.reference(payloads)
// Model a listener reusing the same slots for the next receive batch.
buffers := make([][]byte, count)
for i := range buffers {
buffers[i] = make([]byte, len(wire[i]))
}
p.conn.read = func(r udp.EncReader, flush func()) {
for b := range 2 {
for i := range count {
copy(buffers[i], wire[b*count+i])
r(p.source.GetRemote(), buffers[i])
require.Len(t, p.conn.writes, b, "forwarding waits for the batch boundary")
}
flush()
require.Len(t, p.conn.writes, b+1, "flush must send before receive slots are reused")
for _, buffer := range buffers {
clear(buffer)
}
}
}
p.f.listenOut(0)
require.Equal(t, want, p.conn.packets())
for _, w := range p.conn.writes {
require.True(t, w.batched)
require.Len(t, w.packets, count)
for _, remote := range w.remotes {
require.Equal(t, p.target.GetRemote(), remote)
}
}
})
}
}
func TestRelayBatchDestinationChange(t *testing.T) {
p := newRelayBatchPath(t, noise.CipherChaChaPoly)
first, second := p.target.GetRemote(), netip.MustParseAddrPort("198.51.100.3:4242")
remotes := []netip.AddrPort{first, first, second, second, first}
var payloads, wire [][]byte
for i := range remotes {
payload := []byte{byte(i)}
payloads = append(payloads, payload)
wire = append(wire, p.wrap(t, payload))
}
want := p.reference(payloads)
p.conn.read = func(r udp.EncReader, flush func()) {
for i, remote := range remotes {
p.target.remote.Store(&remote)
r(p.source.GetRemote(), wire[i])
}
flush()
}
p.f.listenOut(0)
require.Equal(t, want, p.conn.packets())
require.Len(t, p.conn.writes, 3)
for i, wantRemotes := range [][]netip.AddrPort{{first, first}, {second, second}, {first}} {
require.Equal(t, wantRemotes, p.conn.writes[i].remotes)
}
}
func TestRelayBatchCapacity(t *testing.T) {
p := newRelayBatchPath(t, noise.CipherChaChaPoly)
p.conn.read = func(r udp.EncReader, flush func()) {
for i := range batch.SendBatchCap + 1 {
r(p.source.GetRemote(), p.wrap(t, []byte{byte(i)}))
}
require.Len(t, p.conn.writes, 1)
require.Len(t, p.conn.writes[0].packets, batch.SendBatchCap)
flush()
}
p.f.listenOut(0)
require.Len(t, p.conn.writes, 2)
require.Len(t, p.conn.writes[1].packets, 1)
}
func TestRelayBatchBeforeResponse(t *testing.T) {
for _, unknownRelay := range []bool{false, true} {
name := "test request"
if unknownRelay {
name = "unknown relay"
}
t.Run(name, func(t *testing.T) {
p := newRelayBatchPath(t, noise.CipherChaChaPoly)
first, last := p.wrap(t, []byte("first")), p.wrap(t, []byte("last"))
p.counter++
request := header.Encode(make([]byte, header.Len), header.Version, header.Test, header.TestRequest, 100, p.counter)
request, err := p.sourceKey.EncryptDanger(request, request, []byte("ping"), p.counter, make([]byte, 12))
require.NoError(t, err)
if unknownRelay {
request = header.Encode(make([]byte, header.Len), header.Version, header.Message, header.MessageRelay, 999, p.counter)
}
p.conn.read = func(r udp.EncReader, flush func()) {
r(p.source.GetRemote(), first)
require.Empty(t, p.conn.writes)
r(p.source.GetRemote(), request)
require.Len(t, p.conn.writes, 2)
r(p.source.GetRemote(), last)
flush()
}
p.f.listenOut(0)
require.Len(t, p.conn.writes, 3)
require.True(t, p.conn.writes[0].batched)
require.False(t, p.conn.writes[1].batched)
require.True(t, p.conn.writes[2].batched)
require.Equal(t, p.source.GetRemote(), p.conn.writes[1].remotes[0])
h := &header.H{}
require.NoError(t, h.Parse(p.conn.writes[1].packets[0]))
if unknownRelay {
require.Equal(t, header.RecvError, h.Type)
require.EqualValues(t, 999, h.RemoteIndex)
} else {
require.Equal(t, header.Test, h.Type)
require.Equal(t, header.TestReply, h.Subtype)
reply := p.conn.writes[1].packets[0]
out, err := p.sourceKey.DecryptDanger(nil, reply[:header.Len], reply[header.Len:], h.MessageCounter, make([]byte, 12))
require.NoError(t, err)
require.Equal(t, []byte("ping"), out)
}
})
}
}
func TestRelayBatchRemovedTarget(t *testing.T) {
for _, remove := range []string{"host", "relay", "disestablished"} {
t.Run(remove, func(t *testing.T) {
p := newRelayBatchPath(t, noise.CipherChaChaPoly)
first, removed := p.wrap(t, []byte("first")), p.wrap(t, []byte("removed"))
want := p.reference([][]byte{[]byte("first")})
p.conn.read = func(r udp.EncReader, flush func()) {
r(p.source.GetRemote(), first)
switch remove {
case "host":
delete(p.f.hostMap.Hosts, p.target.vpnAddrs[0])
case "relay":
delete(p.target.relayState.relayForByAddr, p.source.vpnAddrs[0])
case "disestablished":
p.target.relayState.UpdateRelayForByIpState(p.source.vpnAddrs[0], Disestablished)
}
r(p.source.GetRemote(), removed)
flush()
}
p.f.listenOut(0)
require.Len(t, p.conn.writes, 1)
require.Equal(t, want, p.conn.packets(), "only the packet accepted before removal is sent")
})
}
}
+2 -2
View File
@@ -10,8 +10,8 @@ type batchWriter interface {
}
// SendBatch accumulates encrypted UDP packets and flushes them via WriteBatch.
// One SendBatch is owned by each listenIn goroutine; no locking is needed.
// Slots are backed by an Arena (see its docs)
// Each reader goroutine owns its own SendBatch; no locking is needed.
// Slots can borrow receive buffers or be backed by an Arena (see its docs).
type SendBatch struct {
out batchWriter
bufs [][]byte
+121
View File
@@ -0,0 +1,121 @@
//go:build !ios && !e2e_testing
package overlay
import (
"bytes"
"os"
"syscall"
"testing"
"github.com/stretchr/testify/require"
)
func TestTun_WriteFraming(t *testing.T) {
r, w, err := os.Pipe()
require.NoError(t, err)
defer r.Close()
defer w.Close()
tn := &tun{f: w}
for _, tc := range []struct {
name string
pkt []byte
af byte
}{
// Literal families rather than syscall.AF_*, so a wrong constant in Write can't also fix the expectation.
{"ipv4", []byte{0x45, 1, 2, 3, 4, 5}, 2},
{"ipv6", []byte{0x60, 1, 2, 3, 4, 5}, 30},
} {
t.Run(tc.name, func(t *testing.T) {
n, err := tn.Write(tc.pkt)
require.NoError(t, err)
require.Equal(t, len(tc.pkt), n)
got := make([]byte, 64)
n, err = r.Read(got)
require.NoError(t, err)
require.Equal(t, append([]byte{0, 0, 0, tc.af}, tc.pkt...), got[:n])
})
}
_, err = tn.Write([]byte{0x10, 1, 2, 3})
require.Error(t, err)
_, err = tn.Write([]byte{})
require.ErrorIs(t, err, syscall.EIO)
// The rejected writes must not have put anything on the fd, so the next read sees only this packet.
pkt := []byte{0x45, 9, 9}
_, err = tn.Write(pkt)
require.NoError(t, err)
got := make([]byte, 64)
n, err := r.Read(got)
require.NoError(t, err)
require.Equal(t, append([]byte{0, 0, 0, 2}, pkt...), got[:n])
}
func TestTun_ReadStripsHeader(t *testing.T) {
r, w, err := os.Pipe()
require.NoError(t, err)
defer r.Close()
defer w.Close()
tn := &tun{f: r}
pkt := []byte{0x45, 1, 2, 3, 4, 5}
_, err = w.Write(append([]byte{0, 0, 0, syscall.AF_INET}, pkt...))
require.NoError(t, err)
got := make([]byte, 64)
n, err := tn.Read(got)
require.NoError(t, err)
require.Equal(t, pkt, got[:n])
}
func TestTun_ReadShort(t *testing.T) {
r, w, err := os.Pipe()
require.NoError(t, err)
defer r.Close()
defer w.Close()
tn := &tun{f: r}
// Fewer bytes than the protocol header is not a packet.
_, err = w.Write([]byte{0, 0})
require.NoError(t, err)
n, err := tn.Read(make([]byte, 64))
require.NoError(t, err)
require.Zero(t, n)
// The next packet reads cleanly only if the short read consumed exactly what was written.
pkt := []byte{0x45, 1, 2, 3}
_, err = w.Write(append([]byte{0, 0, 0, syscall.AF_INET}, pkt...))
require.NoError(t, err)
got := make([]byte, 64)
n, err = tn.Read(got)
require.NoError(t, err)
require.Equal(t, pkt, got[:n])
}
// A heap allocation per packet rarely shows in a short throughput run, only across GC pauses, so
// this pins Read and Write to zero allocations directly.
func TestTun_ReadWriteDoNotAllocate(t *testing.T) {
r, w, err := os.Pipe()
require.NoError(t, err)
defer r.Close()
defer w.Close()
tw := &tun{f: w}
tr := &tun{f: r}
pkts := [][]byte{bytes.Repeat([]byte{0x45}, 1300), bytes.Repeat([]byte{0x60}, 1300)}
buf := make([]byte, 1500)
i := 0
allocs := testing.AllocsPerRun(1000, func() {
i++
if _, err := tw.Write(pkts[i%len(pkts)]); err != nil {
t.Fatal(err)
}
if _, err := tr.Read(buf); err != nil {
t.Fatal(err)
}
})
require.Zero(t, allocs)
}
+4 -2
View File
@@ -36,11 +36,13 @@ func (l *tcpListener) Accept() (net.Conn, error) {
func (l *tcpListener) Close() error {
l.closeOnce.Do(func() {
// Close before leaving the map so a concurrent closeListeners that
// no longer sees this listener knows Accept is already released.
close(l.closed)
l.s.mu.Lock()
delete(l.s.mu.listeners, l.port)
l.s.mu.Unlock()
close(l.closed)
})
return nil
}
+70 -7
View File
@@ -32,13 +32,18 @@ import (
const nicID = 1
type Service struct {
eg *errgroup.Group
control *nebula.Control
ipstack *stack.Stack
// done is closed once the forwarding goroutines have exited and the
// netstack is torn down, at which point err holds the first fatal error.
done chan struct{}
err error
mu struct {
sync.Mutex
closed bool
listeners map[uint16]*tcpListener
}
}
@@ -65,7 +70,6 @@ func New(control *nebula.Control) (_ *Service, reterr error) {
ctx := control.Context()
eg, ctx := errgroup.WithContext(ctx)
s := Service{
eg: eg,
control: control,
}
s.mu.listeners = map[uint16]*tcpListener{}
@@ -74,6 +78,12 @@ func New(control *nebula.Control) (_ *Service, reterr error) {
NetworkProtocols: []stack.NetworkProtocolFactory{ipv4.NewProtocol, ipv6.NewProtocol},
TransportProtocols: []stack.TransportProtocolFactory{tcp.NewProtocol, udp.NewProtocol, icmp.NewProtocol4, icmp.NewProtocol6},
})
// The teardown goroutine only owns the stack once New succeeds
defer func() {
if reterr != nil {
s.ipstack.Destroy()
}
}()
sackEnabledOpt := tcpip.TCPSACKEnabled(true) // TCP SACK is disabled by default
tcpipErr := s.ipstack.SetTransportProtocolOption(tcp.ProtocolNumber, &sackEnabledOpt)
if tcpipErr != nil {
@@ -123,6 +133,12 @@ func New(control *nebula.Control) (_ *Service, reterr error) {
// this will read exactly one packet
n, err := reader.Read(buf)
if err != nil {
// An error after shutdown began is noise from the device pipes
// closing, anything else is fatal and must stop nebula too
if ctx.Err() != nil {
return nil
}
s.control.Stop()
return err
}
packetBuf := stack.NewPacketBuffer(stack.PacketBufferOptions{
@@ -130,8 +146,8 @@ func New(control *nebula.Control) (_ *Service, reterr error) {
})
linkEP.InjectInbound(header.IPv4ProtocolNumber, packetBuf)
if err := ctx.Err(); err != nil {
return err
if ctx.Err() != nil {
return nil
}
}
})
@@ -139,13 +155,18 @@ func New(control *nebula.Control) (_ *Service, reterr error) {
for {
packet := linkEP.ReadContext(ctx)
if packet == nil {
if err := ctx.Err(); err != nil {
return err
if ctx.Err() != nil {
return nil
}
continue
}
bufView := packet.ToView()
if _, err := bufView.WriteTo(writer); err != nil {
// Same shutdown noise handling as the inbound reader
if ctx.Err() != nil {
return nil
}
s.control.Stop()
return err
}
bufView.Release()
@@ -158,6 +179,22 @@ func New(control *nebula.Control) (_ *Service, reterr error) {
return control.Wait()
})
// Tear down here rather than in Close so it also happens when nebula stops
// on its own. The forwarding goroutines must exit before the stack they feed
// is destroyed.
s.done = make(chan struct{})
go func() {
s.err = eg.Wait()
s.closeListeners()
// Destroy aborts the endpoints that exist when it starts, then waits on
// every endpoint, so a Dial registering in between would hold Wait until
// its connect timed out. Without the NIC new dials fail route lookup.
// RemoveNIC also closes linkEP.
_ = s.ipstack.RemoveNIC(nicID)
s.ipstack.Destroy()
close(s.done)
}()
return &s, nil
}
@@ -237,6 +274,9 @@ func (s *Service) Listen(network, address string) (net.Listener, error) {
s.mu.Lock()
defer s.mu.Unlock()
if s.mu.closed {
return nil, net.ErrClosed
}
if _, ok := s.mu.listeners[port]; ok {
return nil, fmt.Errorf("already listening on port %d", port)
}
@@ -245,15 +285,38 @@ func (s *Service) Listen(network, address string) (net.Listener, error) {
return l, nil
}
// Wait blocks until the service has fully stopped, either via Close or a fatal
// error, and returns the first fatal error if there was one.
func (s *Service) Wait() error {
return s.eg.Wait()
<-s.done
return s.err
}
// Close closes all listeners and stops nebula. Use Wait to block until
// shutdown has completed.
func (s *Service) Close() error {
s.closeListeners()
s.control.Stop()
return nil
}
// closeListeners unblocks pending Accept calls and rejects new listeners.
func (s *Service) closeListeners() {
// Every caller closes every listener rather than returning early once
// closed is set, so none returns while another is still closing them.
s.mu.Lock()
s.mu.closed = true
listeners := make([]*tcpListener, 0, len(s.mu.listeners))
for _, l := range s.mu.listeners {
listeners = append(listeners, l)
}
s.mu.Unlock()
for _, l := range listeners {
_ = l.Close()
}
}
func (s *Service) tcpHandler(r *tcp.ForwarderRequest) {
endpointID := r.ID()
+159
View File
@@ -327,3 +327,162 @@ func TestListenerAcceptAfterClose(t *testing.T) {
}
}
}
// Close must release blocked Accept calls and let Wait report a clean shutdown.
func TestServiceClose(t *testing.T) {
ca, _, caKey, _ := cert_test.NewTestCaCert(cert.Version2, cert.Curve_CURVE25519, time.Now(), time.Now().Add(10*time.Minute), nil, nil, []string{})
a := newSimpleService(ca, caKey, "a", netip.MustParseAddr("10.0.0.1"), m{
"static_host_map": m{},
"listen": m{
"host": "127.0.0.1",
"port": 0,
},
})
ln, err := a.Listen("tcp", ":1234")
if err != nil {
t.Fatal(err)
}
accepted := make(chan error, 1)
go func() {
_, err := ln.Accept()
accepted <- err
}()
if err := a.Close(); err != nil {
t.Fatal(err)
}
select {
case err := <-accepted:
if !errors.Is(err, net.ErrClosed) {
t.Fatalf("expected net.ErrClosed from Accept, got %v", err)
}
case <-time.After(5 * time.Second):
t.Fatal("Accept still blocked after Close")
}
waited := make(chan error, 1)
go func() { waited <- a.Wait() }()
select {
case err := <-waited:
if err != nil {
t.Fatalf("expected clean shutdown from Wait, got %v", err)
}
case <-time.After(5 * time.Second):
t.Fatal("Wait still blocked after Close")
}
if a.ipstack.HasNIC(nicID) {
t.Fatal("netstack not torn down after Wait")
}
if err := a.Close(); err != nil {
t.Fatalf("second close: %v", err)
}
if _, err := a.Listen("tcp", ":1235"); !errors.Is(err, net.ErrClosed) {
t.Fatalf("expected net.ErrClosed from Listen after Close, got %v", err)
}
}
// If nebula stops without Close being called, the service must still release
// Accept and tear down the netstack.
func TestServiceNebulaStopped(t *testing.T) {
ca, _, caKey, _ := cert_test.NewTestCaCert(cert.Version2, cert.Curve_CURVE25519, time.Now(), time.Now().Add(10*time.Minute), nil, nil, []string{})
a := newSimpleService(ca, caKey, "a", netip.MustParseAddr("10.0.0.1"), m{
"static_host_map": m{},
"listen": m{
"host": "127.0.0.1",
"port": 0,
},
})
ln, err := a.Listen("tcp", ":1234")
if err != nil {
t.Fatal(err)
}
accepted := make(chan error, 1)
go func() {
_, err := ln.Accept()
accepted <- err
}()
a.control.Stop()
waited := make(chan error, 1)
go func() { waited <- a.Wait() }()
select {
case err := <-waited:
if err != nil {
t.Fatalf("expected clean shutdown from Wait, got %v", err)
}
case <-time.After(5 * time.Second):
t.Fatal("Wait still blocked after nebula stopped")
}
select {
case err := <-accepted:
if !errors.Is(err, net.ErrClosed) {
t.Fatalf("expected net.ErrClosed from Accept, got %v", err)
}
case <-time.After(5 * time.Second):
t.Fatal("Accept still blocked after nebula stopped")
}
if a.ipstack.HasNIC(nicID) {
t.Fatal("netstack not torn down after Wait")
}
}
// Dials racing Close must not hold Wait until their connect times out.
func TestServiceDialDuringClose(t *testing.T) {
ca, _, caKey, _ := cert_test.NewTestCaCert(cert.Version2, cert.Curve_CURVE25519, time.Now(), time.Now().Add(10*time.Minute), nil, nil, []string{})
a := newSimpleService(ca, caKey, "a", netip.MustParseAddr("10.0.0.1"), m{
"static_host_map": m{},
"listen": m{
"host": "127.0.0.1",
"port": 0,
},
})
stop := make(chan struct{})
var eg errgroup.Group
for range 8 {
eg.Go(func() error {
for {
select {
case <-stop:
return nil
default:
}
// 10.0.0.2 has no tunnel, so each dial sits in SYN-SENT until aborted
if c, err := a.Dial("tcp", "10.0.0.2:80"); err == nil {
_ = c.Close()
}
}
})
}
time.Sleep(50 * time.Millisecond)
if err := a.Close(); err != nil {
t.Fatal(err)
}
waited := make(chan error, 1)
go func() { waited <- a.Wait() }()
select {
case err := <-waited:
if err != nil {
t.Fatalf("expected clean shutdown from Wait, got %v", err)
}
case <-time.After(5 * time.Second):
t.Fatal("Wait blocked behind a dial that raced Close")
}
close(stop)
dialersDone := make(chan error, 1)
go func() { dialersDone <- eg.Wait() }()
select {
case <-dialersDone:
case <-time.After(5 * time.Second):
t.Fatal("Dial still blocked after Wait")
}
}
+2 -5
View File
@@ -147,10 +147,7 @@ func (tw *TimerWheel[T]) Purge() (T, bool) {
tw.expired.Tail = nil
}
// Clear out the items references, a cached item must not keep its value alive
v := ti.Item
var zero T
ti.Item = zero
// Clear out the items references
ti.Next = nil
// Maybe cache it for later
@@ -160,7 +157,7 @@ func (tw *TimerWheel[T]) Purge() (T, bool) {
tw.itemsCached++
}
return v, true
return ti.Item, true
}
// findWheel find the next position in the wheel for the provided timeout given the current tick
+1 -3
View File
@@ -7,7 +7,6 @@ import (
"github.com/slackhq/nebula/firewall"
"github.com/stretchr/testify/assert"
"github.com/stretchr/testify/require"
)
func TestNewTimerWheel(t *testing.T) {
@@ -151,8 +150,7 @@ func TestTimerWheel_Purge(t *testing.T) {
assert.Equal(t, 4, tw.itemsCached)
ci := tw.itemCache
for range 4 {
require.NotNil(t, ci)
assert.Zero(t, ci.Item)
assert.NotNil(t, ci)
ci = ci.Next
}
assert.Nil(t, ci)
+1 -1
View File
@@ -41,7 +41,7 @@ type Conn interface {
//
// Returns the number of packets successfully written. A destination the kernel rejects costs only
// its own packet, so a short count means some peers were undeliverable, not that the batch failed.
// Not safe for concurrent use on the same Conn.
// Safe for concurrent use on the same Conn.
WriteBatch(bufs [][]byte, addrs []netip.AddrPort) (int, error)
// ReloadConfig applies the listen settings that can change, r is nil at startup
ReloadConfig(c *config.C, r *config.Reload)
+6 -1
View File
@@ -12,6 +12,7 @@ import (
"net/netip"
"strconv"
"strings"
"sync"
"unsafe"
"golang.org/x/sys/unix"
@@ -20,7 +21,7 @@ import (
// batchWriter owns the sendmmsg(2)/UDP-GSO transmit path for a StdConn: the
// scratch WriteBatch packs mmsghdr entries into, plus the GSO capability
// state probed at socket creation. Each queue has its own StdConn and
// batchWriter, so no locking is needed.
// batchWriter; mu serializes batches from the TUN reader and relay forwarding.
//
// Terminology, smallest to largest:
//
@@ -39,6 +40,7 @@ import (
// chunk the entries packed for one sendmmsg call, at most MaxWriteBatch.
// batch the caller's whole bufs/addrs pair, processed as one or more chunks.
type batchWriter struct {
mu sync.Mutex
fd int
isV4 bool
@@ -192,6 +194,9 @@ func parseRelease(r string) (major, minor int) {
// Returns the number of packets sent. An error means the call itself failed.
// A short count means some destinations were undeliverable.
func (w *batchWriter) WriteBatch(bufs [][]byte, addrs []netip.AddrPort) (int, error) {
w.mu.Lock()
defer w.mu.Unlock()
if len(bufs) != len(addrs) {
return 0, fmt.Errorf("WriteBatch: len(bufs)=%d != len(addrs)=%d", len(bufs), len(addrs))
}
@@ -0,0 +1,61 @@
//go:build linux && !android && !e2e_testing
package udp
import (
"net/netip"
"runtime"
"sync"
"testing"
"github.com/stretchr/testify/require"
)
// The TUN reader and relay forwarding share a socket. Exercise their concurrent
// batches against the actual GSO planner and scratch, without relying on a NIC.
func TestWriteBatchConcurrent(t *testing.T) {
const batches = 64
w := newRewindTestWriter()
sent := make(chan [][]byte, 2*batches)
w.sendFn = func(start, n int) (int, error) {
runtime.Gosched()
sent <- capturePrepared(w, start, n)
return n, nil
}
conn := &StdConn{bw: w}
remote := netip.MustParseAddrPort("192.0.2.1:4242")
var wg sync.WaitGroup
start := make(chan struct{})
for producer := range 2 {
wg.Go(func() {
<-start
for i := range batches {
var bufs [][]byte
var addrs []netip.AddrPort
for j := range 4 {
bufs = append(bufs, []byte{byte(producer), byte(i), byte(j), 0})
addrs = append(addrs, remote)
}
n, err := conn.WriteBatch(bufs, addrs)
if err != nil || n != len(bufs) {
t.Errorf("WriteBatch = %d, %v; want %d, nil", n, err, len(bufs))
return
}
}
})
}
close(start)
wg.Wait()
close(sent)
var counts [2]int
for packets := range sent {
require.Len(t, packets, 4)
producer := packets[0][0]
require.Less(t, int(producer), len(counts))
for j, packet := range packets {
require.Equal(t, []byte{producer, byte(counts[producer]), byte(j), 0}, packet)
}
counts[producer]++
}
require.Equal(t, [2]int{batches, batches}, counts)
}