Compare commits

..
Author SHA1 Message Date
JackDoan 0f59f895a4 cpupick: Keep single readers off tun.avoid_cpus without cpupick, warn and ignore bad lists 2026-10-15 12:21:59 -05:00
JackDoan df1c4d9364 cpupick: Make tun.avoid_cpus a pin selection rule instead of a process-wide mask 2026-10-15 12:13:21 -05:00
JackDoan 36579bae03 cpupick: Move AvoidCPUs from util into cpupick as Avoid 2026-10-15 12:00:28 -05:00
Ryan Huber 73d26dd87d tun: Allow excluding CPUs from process threads at startup
Set tun.avoid_cpus to a list of logical CPUs to remove from the affinity of
every thread of the process on Linux, before services start and reader CPUs
are chosen. Later threads inherit the restricted mask; enumeration repeats so
threads created meanwhile are covered. Inherited taskset and cgroup
restrictions are kept. Keeping Nebula off the underlay NIC's receive IRQ CPUs
and their SMT siblings is what this is for; the example configuration shows
how to find them.

Invalid IDs, a list that would leave a thread no CPU, and syscall failures
fail startup. Configuration tests only validate. Other platforms log and
ignore the setting. The exclusion is process-wide, including embedded use, and
not reloadable. It is independent of thread pinning.

New API: util.AvoidCPUs, next to the existing PinThreadToCPU and AllowedCPUs.

Tests run in a child process and check existing and new threads, preserved
inherited masks, validation without changes, and the unix.CPUSet bounds on
32- and 64-bit platforms. No isolated performance claim is made.
2026-10-15 14:40:45 +00:00
72 changed files with 607 additions and 1325 deletions

No files matched your search

+3 -3
View File
@@ -14,7 +14,7 @@ jobs:
- uses: actions/setup-go@v7
with:
go-version: '1.27'
go-version: '1.26'
check-latest: true
- name: Build
@@ -40,7 +40,7 @@ jobs:
- uses: actions/setup-go@v7
with:
go-version: '1.27'
go-version: '1.26'
check-latest: true
- name: Build
@@ -83,7 +83,7 @@ jobs:
- uses: actions/setup-go@v7
with:
go-version: '1.27'
go-version: '1.26'
check-latest: true
# GitHub holds ARNs, not credentials, and ARNs outlive a rotation
+3 -3
View File
@@ -34,7 +34,7 @@ jobs:
- uses: actions/setup-go@v7
with:
go-version: '1.27'
go-version: '1.26'
check-latest: true
- name: add hashicorp source
@@ -66,7 +66,7 @@ jobs:
- uses: actions/setup-go@v7
with:
go-version: '1.27'
go-version: '1.26'
check-latest: true
- name: add hashicorp source
@@ -92,7 +92,7 @@ jobs:
- uses: actions/setup-go@v7
with:
go-version: '1.27'
go-version: '1.26'
check-latest: true
# WSL2 + Ubuntu so the smoke can run a real linux peer with its own
+2 -2
View File
@@ -22,7 +22,7 @@ jobs:
- uses: actions/setup-go@v7
with:
go-version: '1.27'
go-version: '1.26'
check-latest: true
- name: Smoke Docker
@@ -51,7 +51,7 @@ jobs:
- uses: actions/setup-go@v7
with:
go-version: '1.27'
go-version: '1.26'
check-latest: true
- name: build
+4 -4
View File
@@ -22,7 +22,7 @@ jobs:
- uses: actions/setup-go@v7
with:
go-version: '1.27'
go-version: '1.26'
check-latest: true
- name: Install goimports
@@ -42,7 +42,7 @@ jobs:
- name: golangci-lint
uses: golangci/golangci-lint-action@v9
with:
version: v2.14
version: v2.12
test:
name: Test ${{ matrix.name }}
@@ -87,7 +87,7 @@ jobs:
- uses: actions/setup-go@v7
with:
go-version: '1.27'
go-version: '1.26'
check-latest: true
- name: Build
@@ -132,7 +132,7 @@ jobs:
- uses: actions/setup-go@v7
with:
go-version: '1.27'
go-version: '1.26'
check-latest: true
- name: Build ${{ matrix.name }}
+1 -2
View File
@@ -85,10 +85,9 @@ func main() {
fmt.Printf("failed to apply logging config: %s", err)
os.Exit(1)
}
c.RegisterReloadCallback(func(c *config.C, r *config.Reload) {
c.RegisterReloadCallback(func(c *config.C) {
if err := logging.ApplyConfig(l, c); err != nil {
l.Error("Failed to reconfigure logger on reload", "error", err)
r.Failed()
}
})
+1 -2
View File
@@ -34,10 +34,9 @@ func (p *program) Start(s service.Service) error {
if err := logging.ApplyConfig(l, c); err != nil {
return fmt.Errorf("failed to apply logging config: %s", err)
}
c.RegisterReloadCallback(func(c *config.C, r *config.Reload) {
c.RegisterReloadCallback(func(c *config.C) {
if err := logging.ApplyConfig(l, c); err != nil {
l.Error("Failed to reconfigure logger on reload", "error", err)
r.Failed()
}
})
+1 -2
View File
@@ -71,10 +71,9 @@ func main() {
fmt.Printf("failed to apply logging config: %s", err)
os.Exit(1)
}
c.RegisterReloadCallback(func(c *config.C, r *config.Reload) {
c.RegisterReloadCallback(func(c *config.C) {
if err := logging.ApplyConfig(l, c); err != nil {
l.Error("Failed to reconfigure logger on reload", "error", err)
r.Failed()
}
})
+31 -40
View File
@@ -26,12 +26,9 @@ type C struct {
files []string
Settings map[string]any
oldSettings map[string]any
callbacks []func(*C, *Reload)
callbacks []func(*C)
l *slog.Logger
reloadLock sync.Mutex
// Set by the first reload that needs a restart and never cleared. Guarded by reloadLock
needsRestart bool
}
func NewC(l *slog.Logger) *C {
@@ -75,9 +72,8 @@ func (c *C) LoadString(raw string) error {
// RegisterReloadCallback stores a function to be called when a config reload is triggered. The functions registered
// here should decide if they need to make a change to the current process before making the change. HasChanged can be
// used to help decide if a change is necessary.
// These functions should return quickly or spawn their own go routine if they will take a while.
// Use r to report anything that failed or needs a restart to apply.
func (c *C) RegisterReloadCallback(f func(c *C, r *Reload)) {
// These functions should return quickly or spawn their own go routine if they will take a while
func (c *C) RegisterReloadCallback(f func(*C)) {
c.callbacks = append(c.callbacks, f)
}
@@ -148,55 +144,50 @@ func (c *C) CatchHUP(ctx context.Context) {
return
case <-ch:
c.l.Info("Caught HUP, reloading config")
c.ReloadConfig().Log(c.l)
c.ReloadConfig()
}
}
}()
}
// ReloadConfig reloads the config from the path provided to Load and runs every registered callback. A config that
// fails to load is logged here, like any other failure, and the result carries the error for the summary.
func (c *C) ReloadConfig() ReloadResult {
func (c *C) ReloadConfig() {
c.reloadLock.Lock()
defer c.reloadLock.Unlock()
c.oldSettings = make(map[string]any)
maps.Copy(c.oldSettings, c.Settings)
if err := c.Load(c.path); err != nil {
c.l.Error("Error occurred while reloading config", "config_path", c.path, "error", err)
return ReloadResult{Err: err, RestartRequired: c.needsRestart}
err := c.Load(c.path)
if err != nil {
c.l.Error("Error occurred while reloading config",
"config_path", c.path,
"error", err,
)
return
}
return c.runCallbacks()
}
// ReloadConfigString is ReloadConfig for a config held in memory rather than on disk.
func (c *C) ReloadConfigString(raw string) ReloadResult {
c.reloadLock.Lock()
defer c.reloadLock.Unlock()
c.oldSettings = make(map[string]any)
maps.Copy(c.oldSettings, c.Settings)
if err := c.LoadString(raw); err != nil {
c.l.Error("Error occurred while reloading config", "error", err)
return ReloadResult{Err: err, RestartRequired: c.needsRestart}
}
return c.runCallbacks()
}
func (c *C) runCallbacks() ReloadResult {
r := &Reload{c: c}
for _, v := range c.callbacks {
v(c, r)
v(c)
}
}
func (c *C) ReloadConfigString(raw string) error {
c.reloadLock.Lock()
defer c.reloadLock.Unlock()
c.oldSettings = make(map[string]any)
maps.Copy(c.oldSettings, c.Settings)
err := c.LoadString(raw)
if err != nil {
return err
}
if r.restart {
c.needsRestart = true
for _, v := range c.callbacks {
v(c)
}
return ReloadResult{Failed: r.failed, RestartRequired: c.needsRestart, RestartKeys: r.keys}
return nil
}
// GetString will get the string for k or return the default d if not found or invalid
@@ -400,7 +391,7 @@ func (c *C) parse() error {
var nm map[string]any
err = yaml.Unmarshal(b, &nm)
if err != nil {
return fmt.Errorf("%s: %w", path, err)
return err
}
// We need to use WithAppendSlice so that firewall rules in separate
+2 -2
View File
@@ -19,7 +19,7 @@ func TestConfig_Load(t *testing.T) {
// invalid yaml
c := NewC(l)
os.WriteFile(filepath.Join(dir, "01.yaml"), []byte(" invalid yaml"), 0644)
require.EqualError(t, c.Load(dir), filepath.Join(dir, "01.yaml")+": yaml: unmarshal errors:\n line 1: cannot unmarshal !!str `invalid...` into map[string]interface {}")
require.EqualError(t, c.Load(dir), "yaml: unmarshal errors:\n line 1: cannot unmarshal !!str `invalid...` into map[string]interface {}")
// simple multi config merge
c = NewC(l)
@@ -129,7 +129,7 @@ func TestConfig_ReloadConfig(t *testing.T) {
os.WriteFile(filepath.Join(dir, "01.yaml"), []byte("outer:\n inner: ho"), 0644)
c.RegisterReloadCallback(func(c *C, _ *Reload) {
c.RegisterReloadCallback(func(c *C) {
done <- true
})
-98
View File
@@ -1,98 +0,0 @@
package config
import (
"context"
"fmt"
"log/slog"
"slices"
)
// ReloadResult describes how a reload went. Problems are logged where they happen, Log writes a one line summary.
type ReloadResult struct {
// Err is set when the new config could not be read or parsed, no callbacks ran
Err error
// Failed is true when a callback could not apply part of the new config
Failed bool
// RestartRequired is true once any reload has needed a restart, and stays true until the process restarts
RestartRequired bool
// RestartKeys are the keys this reload changed that only apply on a restart
RestartKeys []string
}
// Summary returns a one line summary of the reload and the level to log it at.
func (r ReloadResult) Summary() (slog.Level, string) {
switch {
case r.Err != nil && r.RestartRequired:
return slog.LevelError, "Config reload failed, the new config did not load and a restart is still required"
case r.Err != nil:
return slog.LevelError, "Config reload failed, the new config did not load"
case r.Failed && r.RestartRequired:
return slog.LevelError, "Config reload completed with errors and requires a restart to fully apply, review the logs"
case r.Failed:
return slog.LevelError, "Config reload completed with errors, review the logs"
case r.RestartRequired:
return slog.LevelWarn, "Config reload was completed but requires a restart to fully apply"
default:
return slog.LevelInfo, "Config reload complete"
}
}
// String is the summary with Err and RestartKeys included, for output other than the log.
func (r ReloadResult) String() string {
_, msg := r.Summary()
if r.Err != nil {
msg = fmt.Sprintf("%s: %v", msg, r.Err)
}
if len(r.RestartKeys) > 0 {
msg = fmt.Sprintf("%s, keys %v", msg, r.RestartKeys)
}
return msg
}
// Log writes Summary to l.
func (r ReloadResult) Log(l *slog.Logger) {
level, msg := r.Summary()
if len(r.RestartKeys) > 0 {
l.Log(context.Background(), level, msg, "keys", r.RestartKeys)
return
}
l.Log(context.Background(), level, msg)
}
// Reload is passed to each reload callback to report how applying the new config went.
type Reload struct {
c *C
failed bool
restart bool
// Changed keys that only apply on a restart, from every callback
keys []string
}
// Failed marks the reload as failed, the caller logs why. Safe to call on a nil Reload, which is what the initial load
// passes.
func (r *Reload) Failed() {
if r != nil {
r.failed = true
}
}
// RestartRequired marks the reload as needing a restart, the caller logs why. Safe to call on a nil Reload.
func (r *Reload) RestartRequired() {
if r != nil {
r.restart = true
}
}
// RestartRequiredIfChanged marks the reload as needing a restart if any of keys changed, and adds them to RestartKeys.
// Use it for keys that are only read at startup. Safe to call on a nil Reload.
func (r *Reload) RestartRequiredIfChanged(keys ...string) {
if r == nil {
return
}
for _, k := range keys {
if r.c.HasChanged(k) && !slices.Contains(r.keys, k) {
r.keys = append(r.keys, k)
r.restart = true
}
}
}
-108
View File
@@ -1,108 +0,0 @@
package config
import (
"errors"
"log/slog"
"testing"
"github.com/slackhq/nebula/test"
"github.com/stretchr/testify/assert"
"github.com/stretchr/testify/require"
)
func TestReloadConfigString(t *testing.T) {
c := NewC(test.NewLogger())
require.NoError(t, c.LoadString("a: 1"))
fail := false
ran := 0
c.RegisterReloadCallback(func(c *C, r *Reload) {
ran++
if fail {
r.Failed()
}
})
assert.Equal(t, ReloadResult{}, c.ReloadConfigString("a: 2"))
fail = true
assert.Equal(t, ReloadResult{Failed: true}, c.ReloadConfigString("a: 3"))
assert.Equal(t, 2, ran)
// A config that fails to load runs no callbacks and returns the error
res := c.ReloadConfigString("a: [")
require.Error(t, res.Err)
assert.False(t, res.Failed)
assert.Equal(t, 2, ran)
assert.Equal(t, 3, c.GetInt("a", 0))
}
func TestReloadRestartRequiredLatches(t *testing.T) {
c := NewC(test.NewLogger())
require.NoError(t, c.LoadString("a: 1\nb: 1"))
c.RegisterReloadCallback(func(c *C, r *Reload) { r.RestartRequiredIfChanged("a", "b") })
// Nothing that needs a restart changed
assert.Equal(t, ReloadResult{}, c.ReloadConfigString("a: 1\nb: 1\nc: 1"))
assert.Equal(t, ReloadResult{RestartRequired: true, RestartKeys: []string{"a"}}, c.ReloadConfigString("a: 2\nb: 1\nc: 1"))
// The restart sticks through a reload with no changes, reverting the key and a config that fails to load. Only the
// reload that changed the key lists it
assert.Equal(t, ReloadResult{RestartRequired: true}, c.ReloadConfigString("a: 2\nb: 1\nc: 1"))
assert.Equal(t, ReloadResult{RestartRequired: true, RestartKeys: []string{"a"}}, c.ReloadConfigString("a: 1\nb: 1\nc: 1"))
res := c.ReloadConfigString("a: [")
require.Error(t, res.Err)
assert.True(t, res.RestartRequired)
}
// A key reported by more than one callback is listed once
func TestReloadRestartKeysDeduped(t *testing.T) {
c := NewC(test.NewLogger())
require.NoError(t, c.LoadString("a: 1\nb: 1"))
c.RegisterReloadCallback(func(c *C, r *Reload) { r.RestartRequiredIfChanged("a") })
c.RegisterReloadCallback(func(c *C, r *Reload) { r.RestartRequiredIfChanged("a", "b") })
assert.Equal(t, []string{"a", "b"}, c.ReloadConfigString("a: 2\nb: 2").RestartKeys)
}
// A callback can require a restart without a key changing, a file the config points at may have
func TestReloadRestartRequiredWithoutAChange(t *testing.T) {
c := NewC(test.NewLogger())
require.NoError(t, c.LoadString("a: 1"))
c.RegisterReloadCallback(func(c *C, r *Reload) { r.RestartRequired() })
assert.Equal(t, ReloadResult{RestartRequired: true}, c.ReloadConfigString("a: 1"))
}
func TestReloadResultSummary(t *testing.T) {
cases := []struct {
res ReloadResult
level slog.Level
msg string
}{
{ReloadResult{}, slog.LevelInfo, "Config reload complete"},
{ReloadResult{RestartRequired: true}, slog.LevelWarn, "Config reload was completed but requires a restart to fully apply"},
{ReloadResult{Failed: true}, slog.LevelError, "Config reload completed with errors, review the logs"},
{ReloadResult{Failed: true, RestartRequired: true}, slog.LevelError, "Config reload completed with errors and requires a restart to fully apply, review the logs"},
{ReloadResult{Err: errors.New("bad yaml")}, slog.LevelError, "Config reload failed, the new config did not load"},
{ReloadResult{Err: errors.New("bad yaml"), RestartRequired: true}, slog.LevelError, "Config reload failed, the new config did not load and a restart is still required"},
}
for _, tc := range cases {
level, msg := tc.res.Summary()
assert.Equal(t, tc.level, level, tc.msg)
assert.Equal(t, tc.msg, msg)
}
res := ReloadResult{RestartRequired: true, RestartKeys: []string{"a", "b"}}
assert.Equal(t, "Config reload was completed but requires a restart to fully apply, keys [a b]", res.String())
// String includes the error
assert.Equal(t, "Config reload failed, the new config did not load: bad yaml", ReloadResult{Err: errors.New("bad yaml")}.String())
}
// The initial load passes a nil Reload
func TestReloadNilIsSafe(t *testing.T) {
var r *Reload
r.Failed()
r.RestartRequired()
r.RestartRequiredIfChanged("a")
}
+1 -2
View File
@@ -57,8 +57,7 @@ func newConnectionManagerFromConfig(l *slog.Logger, c *config.C, hm *HostMap, p
}
cm.reload(c, true)
c.RegisterReloadCallback(func(c *config.C, r *config.Reload) {
r.RestartRequiredIfChanged("timers.connection_alive_interval", "timers.pending_deletion_interval")
c.RegisterReloadCallback(func(c *config.C) {
cm.reload(c, false)
})
+3 -3
View File
@@ -151,9 +151,9 @@ func (c *fakeConn) WriteTo(_ []byte, _ netip.AddrPort) error { return nil }
func (c *fakeConn) WriteBatch(bufs [][]byte, _ []netip.AddrPort) (int, error) {
return len(bufs), nil
}
func (c *fakeConn) ReloadConfig(_ *config.C, _ *config.Reload) {}
func (c *fakeConn) SupportsMultipleReaders() bool { return true }
func (c *fakeConn) Close() error { c.closed = true; return nil }
func (c *fakeConn) ReloadConfig(_ *config.C) {}
func (c *fakeConn) SupportsMultipleReaders() bool { return true }
func (c *fakeConn) Close() error { c.closed = true; return nil }
type multiqueueDevice struct {
*fakeDevice
+75
View File
@@ -0,0 +1,75 @@
//go:build linux && !android && !e2e_testing
package cpupick
import (
"errors"
"fmt"
"os"
"strconv"
"unsafe"
"golang.org/x/sys/unix"
)
// Avoid removes cpus from the affinity of every thread of the process, for use at startup. New threads inherit
// their parent's mask, so it repeats until a pass finds nothing left to change; a thread that exits meanwhile is
// skipped. Every mask is checked before any is changed. With apply false it only validates, for configuration tests.
// This affects the entire process, including other users of an embedded Nebula. Not wired up yet: tun.avoid_cpus
// only steers Default for now.
func Avoid(cpus []int, apply bool) error {
const capacity = int(unsafe.Sizeof(unix.CPUSet{})) * 8
for _, cpu := range cpus {
if cpu < 0 || cpu >= capacity {
return fmt.Errorf("CPU %d is out of range", cpu)
}
}
if len(cpus) == 0 {
return nil
}
for {
masks, err := avoidingMasks(cpus)
if err != nil || !apply || len(masks) == 0 {
return err
}
for tid, mask := range masks {
if err := unix.SchedSetaffinity(tid, &mask); err != nil && !errors.Is(err, unix.ESRCH) {
return fmt.Errorf("set thread %d affinity: %w", tid, err)
}
}
}
}
// avoidingMasks returns the mask without cpus of every thread whose mask has any of them.
func avoidingMasks(cpus []int) (map[int]unix.CPUSet, error) {
entries, err := os.ReadDir("/proc/self/task")
if err != nil {
return nil, err
}
masks := make(map[int]unix.CPUSet, len(entries))
for _, entry := range entries {
tid, err := strconv.Atoi(entry.Name())
if err != nil {
return nil, err
}
var mask unix.CPUSet
err = unix.SchedGetaffinity(tid, &mask)
if errors.Is(err, unix.ESRCH) {
continue
}
if err != nil {
return nil, fmt.Errorf("read thread %d affinity: %w", tid, err)
}
next := mask
for _, cpu := range cpus {
next.Clear(cpu)
}
if next.Count() == 0 {
return nil, fmt.Errorf("avoid_cpus leaves thread %d with no allowed CPU", tid)
}
if next != mask {
masks[tid] = next
}
}
return masks, nil
}
+98
View File
@@ -0,0 +1,98 @@
//go:build linux && !android && !e2e_testing
package cpupick
import (
"fmt"
"os"
"os/exec"
"runtime"
"strconv"
"sync"
"testing"
"unsafe"
"github.com/slackhq/nebula/util"
"github.com/stretchr/testify/require"
"golang.org/x/sys/unix"
)
// TestAvoid changes the affinity of the whole process, so it runs in a child, leaving the test runner untouched.
func TestAvoid(t *testing.T) {
if os.Getenv("NEBULA_CPUPICK_AVOID_TEST") != "1" {
exe, err := os.Executable()
require.NoError(t, err)
cmd := exec.Command(exe, "-test.run=^TestAvoid$", "-test.v")
cmd.Env = append(os.Environ(), "NEBULA_CPUPICK_AVOID_TEST=1")
out, err := cmd.CombinedOutput()
require.NoError(t, err, "%s", out)
return
}
allowed, err := util.AllowedCPUs()
require.NoError(t, err)
if len(allowed) < 2 {
t.Skip("requires at least two allowed CPUs")
}
cpu := allowed[0]
release := make(chan struct{})
var wg sync.WaitGroup
// lockThreads occupies 8 OS threads until release, so the runtime has to start more
lockThreads := func() {
var ready sync.WaitGroup
for range 8 {
ready.Add(1)
wg.Go(func() {
runtime.LockOSThread()
ready.Done()
<-release
})
}
ready.Wait()
}
lockThreads()
require.NoError(t, Avoid([]int{cpu}, false))
after, err := util.AllowedCPUs()
require.NoError(t, err)
require.Equal(t, allowed, after, "configuration check must not mutate affinity")
require.Error(t, Avoid(allowed, true), "must reject an empty mask before applying")
require.NoError(t, Avoid([]int{cpu, cpu}, true))
requireAvoided(t, cpu, allowed)
// Existing threads stay locked while new ones inherit the restriction
lockThreads()
requireAvoided(t, cpu, allowed)
close(release)
wg.Wait()
}
// requireAvoided checks no thread may run on cpu, and every thread still may on some CPU of allowed only.
func requireAvoided(t *testing.T, cpu int, allowed []int) {
t.Helper()
entries, err := os.ReadDir("/proc/self/task")
require.NoError(t, err)
for _, entry := range entries {
tid, err := strconv.Atoi(entry.Name())
require.NoError(t, err)
var mask unix.CPUSet
err = unix.SchedGetaffinity(tid, &mask)
if err == unix.ESRCH {
continue
}
require.NoError(t, err)
require.False(t, mask.IsSet(cpu), "thread %d", tid)
require.Positive(t, mask.Count())
for id := range len(mask) * 64 {
if mask.IsSet(id) {
require.Contains(t, allowed, id, "must not widen the inherited mask")
}
}
}
}
// TestAvoidRange checks the bounds of unix.CPUSet, whose words are 32 bits wide on 32-bit platforms.
func TestAvoidRange(t *testing.T) {
capacity := int(unsafe.Sizeof(unix.CPUSet{})) * 8
require.NoError(t, Avoid([]int{capacity - 1}, false))
require.EqualError(t, Avoid([]int{capacity}, false), fmt.Sprintf("CPU %d is out of range", capacity))
require.EqualError(t, Avoid([]int{-1}, false), "CPU -1 is out of range")
}
+8
View File
@@ -0,0 +1,8 @@
//go:build !linux || android || e2e_testing
package cpupick
import "errors"
// Avoid is unsupported outside Linux: there is no thread affinity API to restrict.
func Avoid(_ []int, _ bool) error { return errors.ErrUnsupported }
+17 -6
View File
@@ -16,10 +16,15 @@
// across distinct physical cores when the topology permits, CPU 0's physical
// core demoted to last resort, and the order rotated by a stable per-instance
// key so co-located instances spread instead of stacking.
//
// CPUs the operator asked to keep readers off (tun.avoid_cpus, typically the
// underlay NIC's receive IRQ CPUs) are removed from the allowed set before
// any of that, so no rule can bring one back.
package cpupick
import (
"log/slog"
"slices"
"github.com/slackhq/nebula/util"
)
@@ -57,12 +62,17 @@ func flatTopology(cpus []int) topology {
// Default computes the pin order for `routines` tun readers. key is any
// stable per-instance value; the bound UDP port is ideal — distinct across
// co-located instances, stable across restarts so benchmark runs stay
// comparable. Returns nil when there is nothing useful to say (no affinity
// support on this platform, lookup failure); callers keep their existing
// fallback spread.
func Default(routines int, key uint64, l *slog.Logger) []int {
// comparable. avoid lists CPUs never to return, matched literally. Returns
// nil when there is nothing useful to say (no affinity support on this
// platform, lookup failure, avoid covering every allowed CPU); callers keep
// their existing fallback spread.
func Default(routines int, key uint64, avoid []int, l *slog.Logger) []int {
allowed, err := util.AllowedCPUs()
if err != nil || len(allowed) == 0 {
if err != nil {
return nil
}
allowed = slices.DeleteFunc(allowed, func(c int) bool { return slices.Contains(avoid, c) })
if len(allowed) == 0 {
return nil
}
perf, signal := perfCPUs(allowed)
@@ -77,7 +87,8 @@ func Default(routines int, key uint64, l *slog.Logger) []int {
if l != nil {
l.Info("chose default pin CPUs for tun readers",
"cpus", cpus[:min(routines, len(cpus))],
"perfSignal", signal)
"perfSignal", signal,
"avoided", avoid)
}
return cpus
}
+35
View File
@@ -0,0 +1,35 @@
//go:build linux && !android && !e2e_testing
package cpupick
import (
"slices"
"testing"
"github.com/slackhq/nebula/util"
"github.com/stretchr/testify/assert"
"github.com/stretchr/testify/require"
)
func TestDefaultSkipsAvoided(t *testing.T) {
allowed, err := util.AllowedCPUs()
require.NoError(t, err)
if len(allowed) < 2 {
t.Skip("requires at least two allowed CPUs")
}
avoid := allowed[:len(allowed)/2]
for key := range uint64(16) {
got := Default(len(allowed), key, avoid, nil)
require.NotEmpty(t, got)
for _, cpu := range got {
assert.NotContains(t, avoid, cpu, "key %d", key)
assert.Contains(t, allowed, cpu, "key %d", key)
}
}
assert.Nil(t, Default(2, 0, allowed, nil), "avoiding everything leaves nothing to suggest")
// Without avoid every allowed CPU stays in play
got := Default(len(allowed), 0, nil, nil)
slices.Sort(got)
assert.Equal(t, allowed, got)
}
+1 -2
View File
@@ -69,10 +69,9 @@ func newDnsServerFromConfig(ctx context.Context, l *slog.Logger, pki *PKI, hostM
ds.mux = dns.NewServeMux()
ds.mux.HandleFunc(".", ds.handleDnsRequest)
c.RegisterReloadCallback(func(c *config.C, r *config.Reload) {
c.RegisterReloadCallback(func(c *config.C) {
if err := ds.reload(c, false); err != nil {
ds.l.Error("Failed to reload DNS responder from config", "error", err)
r.Failed()
}
})
+1 -1
View File
@@ -1513,7 +1513,7 @@ func TestLighthouseUpdateOnReload(t *testing.T) {
// Reload the config. The lighthouse.hosts change triggers TriggerUpdate,
// which wakes the update worker. It calls SendUpdate, initiating a
// handshake to the new lighthouse and caching the HostUpdateNotification.
require.NoError(t, myConfig.ReloadConfigString(string(newCfg)).Err)
require.NoError(t, myConfig.ReloadConfigString(string(newCfg)))
// Route until the lighthouse receives the HostUpdateNotification.
// This covers: handshake stage 1, stage 2, then the cached update.
+2 -2
View File
@@ -129,7 +129,7 @@ func TestCertUpgrade(t *testing.T) {
}
r.Logf("reload new v2-only config")
err = myC.ReloadConfigString(string(cb)).Err
err = myC.ReloadConfigString(string(cb))
assert.NoError(t, err)
r.Log("yay, spin until their sees it")
waitStart := time.Now()
@@ -222,7 +222,7 @@ func TestCertDowngrade(t *testing.T) {
}
r.Logf("reload new v1-only config")
err = myC.ReloadConfigString(string(cb)).Err
err = myC.ReloadConfigString(string(cb))
assert.NoError(t, err)
r.Log("yay, spin until their sees it")
waitStart := time.Now()
+16 -19
View File
@@ -43,7 +43,7 @@ static_host_map:
lighthouse:
# am_lighthouse is used to enable lighthouse functionality for a node. This should ONLY be true on nodes
# you have configured to be lighthouses in your network. Not reloadable.
# you have configured to be lighthouses in your network
am_lighthouse: false
# serve_dns optionally starts a dns listener that responds to various queries and can even be
# delegated to for resolution
@@ -126,9 +126,8 @@ lighthouse:
# Port Nebula will be listening on. The default here is 4242. For a lighthouse node, the port should be defined,
# however using port 0 will dynamically assign a port and is recommended for roaming nodes.
listen:
# To listen on only ipv4, use "0.0.0.0". Not reloadable.
# To listen on only ipv4, use "0.0.0.0"
host: "::"
# Not reloadable.
port: 4242
# Sets the max number of packets to pull from the kernel for each syscall (under systems that support recvmmsg)
# default is 64, does not support reload
@@ -136,8 +135,7 @@ listen:
# superpacket, so the receive scratch is batch * 64KiB per listening socket (~4MiB per routine at the
# default of 64). Lower this to trade peak per-syscall throughput for memory on constrained hosts.
#batch: 64
# Linux only. Reloadable, but removing a value leaves the socket at the old size until Nebula restarts.
# Configure socket buffers for the udp side (outside), leave unset to use the system defaults. Values will be doubled by the kernel
# Linux only. Configure socket buffers for the udp side (outside), leave unset to use the system defaults. Values will be doubled by the kernel
# Default is net.core.rmem_default and net.core.wmem_default (/proc/sys/net/core/rmem_default and /proc/sys/net/core/rmem_default)
# Maximum is limited by memory in the system, SO_RCVBUFFORCE and SO_SNDBUFFORCE is used to avoid having to raise the system wide
# max, net.core.rmem_max and net.core.wmem_max
@@ -181,7 +179,7 @@ listen:
# Currently, this defaults to 1 which means we have 1 tun queue reader and 1
# UDP queue reader. Setting this above one will set IFF_MULTI_QUEUE on the tun
# device and SO_REUSEPORT on the UDP socket to allow multiple queues.
# This option is only supported on Linux. Not reloadable.
# This option is only supported on Linux.
#routines: 1
punchy:
@@ -205,7 +203,6 @@ punchy:
# Cipher allows you to choose between the available ciphers for your network. Options are chachapoly or aes
# IMPORTANT: this value must be identical on ALL NODES/LIGHTHOUSES. We do not/will not support use of different ciphers simultaneously!
# Not reloadable.
#cipher: aes
# Preferred ranges is used to define a hint about the local network ranges, which speeds up discovering the fastest
@@ -237,7 +234,6 @@ punchy:
# and absolute paths outside of it will be rejected. Default is $TMP/nebula-debug.
# The directory is NOT automatically created.
# Overriding this to "" is the same as "/" and will allow overwriting any path on the host.
# Not reloadable.
#sandbox_dir: /var/tmp/nebula-debug
# EXPERIMENTAL: relay support for networks that can't establish direct connections.
@@ -257,25 +253,19 @@ relay:
# Configure the private interface. Note: addr is baked into the nebula certificate
tun:
# When tun is disabled, a lighthouse can be started without a local tun interface (and therefore without root)
# Not reloadable.
disabled: false
# Name of the device. If not set, a default will be chosen by the OS.
# For macOS: if set, must be in the form `utun[0-9]+`.
# For NetBSD: Required to be set, must be in the form `tun[0-9]+`
# Not reloadable.
dev: nebula1
# Toggles forwarding of local broadcast packets, the address of which depends on the ip/mask encoded in pki.cert
# Not reloadable.
drop_local_broadcast: false
# Toggles forwarding of multicast packets. Not reloadable.
# Toggles forwarding of multicast packets
drop_multicast: false
# Sets the transmit queue length, if you notice lots of transmit drops on the tun it may help to raise this number. Default is 5000
# Not reloadable.
tx_queue: 5000
# Default MTU for every packet, safe setting is (and the default) 1300 for internet based traffic
# Anything above 8937, here or on a route below, is capped to 8937 with a warning
# Reloadable on Linux only. Where the tun is handed to nebula by a host app (iOS, Android, macOS network extensions)
# the host app owns the MTU.
mtu: 1300
# the use_offloads setting controls if Nebula will attempt to enable GSO and GRO for the tun device. Linux only, not reloadable.
@@ -305,6 +295,15 @@ tun:
# - 2
# - 4
# Linux only. avoid_cpus lists CPU IDs the tun readers are never pinned to, exactly as listed (SMT siblings are not
# added for you). The default pin selection skips them, and a cpu_affinity entry naming one is rejected like any
# other invalid entry. Other threads are not restricted, and the setting does nothing with pin_threads: false. An
# invalid list, or one covering every allowed CPU, is ignored with a warning. Not reloadable. To keep Nebula off the
# underlay NIC's receive IRQ CPUs, find them in /proc/interrupts and /proc/irq/<IRQ>/effective_affinity_list, add
# their SMT siblings from /sys/devices/system/cpu/cpu<N>/topology/thread_siblings_list if you want those kept clear
# too, and revisit the list when IRQ placement changes.
#avoid_cpus: []
# Route based MTU overrides, you have known vpn ip paths that can support larger MTUs you can increase/decrease them here
routes:
#- mtu: 8800
@@ -368,7 +367,7 @@ tun:
#use_system_route_table: false
# Buffer size for reading routes updates. 0 means default system buffer size. (/proc/sys/net/core/rmem_default).
# 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. Not reloadable.
# SO_RCVBUFFORCE is used to avoid having to raise the system wide max
#use_system_route_table_buffer_size: 0
# Configure logging level
@@ -418,16 +417,14 @@ logging:
# Handshakes are sent to all known addresses at each interval with a linear backoff,
# Wait try_interval after the 1st attempt, 2 * try_interval after the 2nd, etc, until the handshake is older than timeout
# A 100ms interval with the default 10 retries will give a handshake 5.5 seconds to resolve before timing out
# Not reloadable.
#try_interval: 100ms
#retries: 20
# query_buffer is the size of the buffer channel for querying lighthouses, drops are recorded in the lighthouse.query.dropped metric.
# Not reloadable.
#query_buffer: 1024
# trigger_buffer is the size of the buffer channel for quickly sending handshakes
# after receiving the response for lighthouse queries. Not reloadable.
# after receiving the response for lighthouse queries
#trigger_buffer: 64
# Tunnel manager settings
+3 -4
View File
@@ -1,6 +1,6 @@
module github.com/slackhq/nebula
go 1.27.2
go 1.26.0
require (
dario.cat/mergo v1.0.2
@@ -33,7 +33,7 @@ require (
golang.zx2c4.com/wireguard/windows v1.0.1
google.golang.org/protobuf v1.36.12
gopkg.in/yaml.v3 v3.0.1
gvisor.dev/gvisor v0.0.0-20261014193016-04d31b385980
gvisor.dev/gvisor v0.0.0-20240423190808-9d7a357edefe
)
require (
@@ -45,6 +45,5 @@ require (
github.com/prometheus/common v0.70.1 // indirect
github.com/prometheus/procfs v0.21.1 // indirect
github.com/vishvananda/netns v0.0.5 // indirect
golang.org/x/exp v0.0.0-20250711185948-6ae5c78190dc // indirect
golang.org/x/time v0.15.0 // indirect
golang.org/x/time v0.5.0 // indirect
)
+4 -6
View File
@@ -157,8 +157,6 @@ golang.org/x/crypto v0.0.0-20200622213623-75b288015ac9/go.mod h1:LzIPMQfyMNhhGPh
golang.org/x/crypto v0.0.0-20210322153248-0c34fe9e7dc2/go.mod h1:T9bdIzuCu7OtxOm1hfPfRQxPLYneinmdGuTeoZ9dtd4=
golang.org/x/crypto v0.57.0 h1:3ZVCjf8Ggz7zneR/EHRVx68Ctf+2pmIMP2UFhh9cC6M=
golang.org/x/crypto v0.57.0/go.mod h1:Fdz0i5U6CoizGwLda9DttjSk6qlZo25zYNtR+ycvuZA=
golang.org/x/exp v0.0.0-20250711185948-6ae5c78190dc h1:TS73t7x3KarrNd5qAipmspBDS1rkMcgVG/fS1aRb4Rc=
golang.org/x/exp v0.0.0-20250711185948-6ae5c78190dc/go.mod h1:A+z0yzpGtvnG90cToK5n2tu8UJVP2XUATh+r+sfOOOc=
golang.org/x/lint v0.0.0-20200302205851-738671d3881b/go.mod h1:3xt1FjdF8hUf6vQPIChWIBhFzV8gjjsPE/fR3IyQdNY=
golang.org/x/mod v0.1.1-0.20191105210325-c90efee705ee/go.mod h1:QqPTAvyqsEbceGzBzNggFXnrqF1CaUcvgkdR5Ot7KZg=
golang.org/x/mod v0.2.0/go.mod h1:s0Qsj1ACt9ePp/hMypM3fl4fZqREWJwdYDEqhRiZZUA=
@@ -205,8 +203,8 @@ golang.org/x/term v0.46.0/go.mod h1:+K02xbkittuwc0Am4abfA3Fc+XRGXkvBXNO88NCXPoc=
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 v0.15.0 h1:bbrp8t3bGUeFOx08pvsMYRTCVSMk89u4tKbNOZbp88U=
golang.org/x/time v0.15.0/go.mod h1:Y4YMaQmXwGQZoFaVFk4YpCt4FLQMYKZe9oeV/f4MSno=
golang.org/x/time v0.5.0 h1:o7cqy6amK/52YcAKIPlM3a+Fpj35zvRj2TP+e1xFSfk=
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=
@@ -245,5 +243,5 @@ 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 v0.0.0-20261014193016-04d31b385980 h1:ribyrTYQ/QYeUIs8x8bVLYwGyepa7RnHLqJzegT5YYk=
gvisor.dev/gvisor v0.0.0-20261014193016-04d31b385980/go.mod h1:pK/dapF5Bf3KGW7/kd5CI5IC4lYh9FMFY1qETAIGLXk=
gvisor.dev/gvisor v0.0.0-20240423190808-9d7a357edefe h1:fre4i6mv4iBuz5lCMOzHD1rH1ljqHWSICFmZRbbgp3g=
gvisor.dev/gvisor v0.0.0-20240423190808-9d7a357edefe/go.mod h1:sxc3Uvk/vHcd3tj7/DHVBoR5wvWT/MmRq2pj7HRJnwU=
+1 -1
View File
@@ -317,7 +317,7 @@ func NewHostMapFromConfig(l *slog.Logger, c *config.C) *HostMap {
hm := newHostMap(l)
hm.reload(c, true)
c.RegisterReloadCallback(func(c *config.C, _ *config.Reload) {
c.RegisterReloadCallback(func(c *config.C) {
hm.reload(c, false)
})
+14 -8
View File
@@ -60,6 +60,9 @@ type InterfaceConfig struct {
// shorter lists than `routines` cycle. Empty list keeps the default
// pin-to-(i % NumCPU) behavior. Only consulted when PinThreads is true.
CpuAffinity []int
// AvoidCPUs (tun.avoid_cpus) lists CPUs the default pin spread skips when
// CpuAffinity is empty. Only consulted when PinThreads is true.
AvoidCPUs []int
// PinThreads controls whether each TUN reader OS thread is pinned to a
// single CPU (via tun.pin_threads, default true). Pinning keeps each
// goroutine's sendmmsg on one XPS-selected NIC TX ring so per-flow
@@ -95,6 +98,8 @@ type Interface struct {
// Empty falls back to the default pin-to-(allowed CPU) behavior.
// Only consulted when pinThreads is true.
cpuAffinity []int
// avoidCPUs are removed from that fallback spread, unless they would empty it.
avoidCPUs []int
// pinThreads controls whether listenIn pins each TUN reader OS thread to
// a CPU at all (tun.pin_threads, default true). When false, threads are
// left free to migrate as on stock nebula.
@@ -247,6 +252,7 @@ func NewInterface(ctx context.Context, c *InterfaceConfig) (*Interface, error) {
connectionManager: c.connectionManager,
conntrackCacheTimeout: c.ConntrackCacheTimeout,
cpuAffinity: c.CpuAffinity,
avoidCPUs: c.AvoidCPUs,
pinThreads: c.PinThreads,
udpTxErrLimiter: udpTxErrLimiter,
udpRelayTxErrLimiter: udpRelayTxErrLimiter,
@@ -437,12 +443,14 @@ func (f *Interface) pinThisThread(i int) {
var cpu int
if n := len(f.cpuAffinity); n > 0 {
// Explicit tun.cpu_affinity list wins; parseCpuAffinity already
// validated the entries against the allowed CPU set.
// validated the entries against the allowed CPU set and tun.avoid_cpus.
cpu = f.cpuAffinity[i%n]
} else if allowed, err := util.AllowedCPUs(); err == nil && len(allowed) > 0 {
// Default: spread queues across the CPUs we're actually allowed to
// run on. Under a cpuset/taskset mask these aren't 0..NumCPU-1, so
// i % NumCPU would pick unrunnable IDs and every pin would fail.
// parseAvoidCPUs already dropped an avoid list that covers them all.
allowed = slices.DeleteFunc(allowed, func(c int) bool { return slices.Contains(f.avoidCPUs, c) })
cpu = allowed[i%len(allowed)]
} else {
cpu = i % runtime.NumCPU()
@@ -509,11 +517,10 @@ func (f *Interface) flushSendBatch(sb *batch.SendBatch, q int) {
func (f *Interface) RegisterConfigChangeCallbacks(c *config.C) {
c.RegisterReloadCallback(f.reloadFirewall)
for _, reload := range []func(*config.C){
f.reloadSendRecvError, f.reloadAcceptRecvError, f.reloadDisconnectInvalid, f.reloadMisc,
} {
c.RegisterReloadCallback(func(c *config.C, _ *config.Reload) { reload(c) })
}
c.RegisterReloadCallback(f.reloadSendRecvError)
c.RegisterReloadCallback(f.reloadAcceptRecvError)
c.RegisterReloadCallback(f.reloadDisconnectInvalid)
c.RegisterReloadCallback(f.reloadMisc)
for _, udpConn := range f.writers {
c.RegisterReloadCallback(udpConn.ReloadConfig)
@@ -530,7 +537,7 @@ func (f *Interface) reloadDisconnectInvalid(c *config.C) {
}
}
func (f *Interface) reloadFirewall(c *config.C, r *config.Reload) {
func (f *Interface) reloadFirewall(c *config.C) {
cs := f.pki.getCertState()
curCert := cs.getCertificate(cert.Version2)
if curCert == nil {
@@ -549,7 +556,6 @@ func (f *Interface) reloadFirewall(c *config.C, r *config.Reload) {
fw, err := NewFirewallFromConfig(f.l, cs, c)
if err != nil {
f.l.Error("Error while creating firewall during reload", "error", err)
r.Failed()
return
}
+21 -23
View File
@@ -7,6 +7,7 @@ import (
"github.com/slackhq/nebula/cert"
"github.com/slackhq/nebula/config"
"github.com/slackhq/nebula/test"
"github.com/slackhq/nebula/util"
"github.com/stretchr/testify/assert"
"github.com/stretchr/testify/require"
)
@@ -65,10 +66,10 @@ func TestReloadFirewall_CertUnsafeNetworksChanged(t *testing.T) {
pki.cs.Store(&CertState{v2Cert: c2, initiatingVersion: cert.Version2})
// Reload with the same YAML so HasChanged("firewall") reports false.
require.NoError(t, cfg.ReloadConfigString(rawYAML).Err)
require.NoError(t, cfg.ReloadConfigString(rawYAML))
require.False(t, cfg.HasChanged("firewall"))
f.reloadFirewall(cfg, nil)
f.reloadFirewall(cfg)
assert.NotSame(t, fw, f.firewall, "firewall pointer should have been replaced")
assert.Equal(t, newUnsafe, f.firewall.unsafeNetworks)
@@ -113,30 +114,27 @@ func TestReloadFirewall_NoChange(t *testing.T) {
l: l,
}
require.NoError(t, cfg.ReloadConfigString(rawYAML).Err)
f.reloadFirewall(cfg, nil)
require.NoError(t, cfg.ReloadConfigString(rawYAML))
f.reloadFirewall(cfg)
assert.Same(t, fw, f.firewall, "firewall should not have been replaced")
}
// Rules that don't parse keep the old firewall, and the reload says it could not apply them
func TestReloadFirewall_BadRulesFailTheReload(t *testing.T) {
l := test.NewLogger()
pki := &PKI{}
pki.cs.Store(&CertState{v2Cert: &dummyCert{version: cert.Version2, networks: []netip.Prefix{netip.MustParsePrefix("10.0.0.1/24")}}, initiatingVersion: cert.Version2})
rules := func(proto string) string {
return "firewall:\n outbound:\n - port: any\n proto: any\n host: any\n inbound:\n - port: any\n proto: " + proto + "\n host: any\n"
}
cfg := config.NewC(l)
require.NoError(t, cfg.LoadString(rules("any")))
fw, err := NewFirewallFromConfig(l, pki.getCertState(), cfg)
// TestPinThisThreadFallbackSkipsAvoided covers the single reader, which never goes through cpupick: its fallback
// spread over the allowed set must still skip tun.avoid_cpus.
func TestPinThisThreadFallbackSkipsAvoided(t *testing.T) {
allowed, err := util.AllowedCPUs()
require.NoError(t, err)
f := &Interface{pki: pki, firewall: fw, l: l}
cfg.RegisterReloadCallback(f.reloadFirewall)
res := cfg.ReloadConfigString(rules("nope"))
require.NoError(t, res.Err)
assert.True(t, res.Failed)
assert.Same(t, fw, f.firewall)
if len(allowed) < 2 {
t.Skip("requires at least two allowed CPUs")
}
f := &Interface{l: test.NewLogger(), pinThreads: true, avoidCPUs: []int{allowed[0]}}
got := make(chan []int)
go func() {
// The pin leaves this goroutine locked to its thread, so the runtime discards the thread when it returns.
f.pinThisThread(0)
cpus, _ := util.AllowedCPUs()
got <- cpus
}()
assert.Equal(t, []int{allowed[1]}, <-got)
}
+1 -4
View File
@@ -143,16 +143,13 @@ func NewLightHouseFromConfig(ctx context.Context, l *slog.Logger, c *config.C, c
return nil, err
}
c.RegisterReloadCallback(func(c *config.C, r *config.Reload) {
r.RestartRequiredIfChanged("lighthouse.am_lighthouse", "listen.port", "handshakes.query_buffer", "stats.lighthouse_metrics")
c.RegisterReloadCallback(func(c *config.C) {
err := h.reload(c, false)
switch v := err.(type) {
case *util.ContextualError:
v.Log(l)
r.Failed()
case error:
l.Error("failed to reload lighthouse", "error", err)
r.Failed()
}
})
+13 -30
View File
@@ -87,15 +87,15 @@ func TestReloadLighthouseInterval(t *testing.T) {
lh.ifce = &mockEncWriter{}
// The first one routine is kicked off by main.go currently, lets make sure that one dies
require.NoError(t, c.ReloadConfigString("lighthouse:\n interval: 5").Err)
require.NoError(t, c.ReloadConfigString("lighthouse:\n interval: 5"))
assert.Equal(t, int64(5), lh.interval.Load())
// Subsequent calls are killed off by the LightHouse.Reload function
require.NoError(t, c.ReloadConfigString("lighthouse:\n interval: 10").Err)
require.NoError(t, c.ReloadConfigString("lighthouse:\n interval: 10"))
assert.Equal(t, int64(10), lh.interval.Load())
// If this completes then nothing is stealing our reload routine
require.NoError(t, c.ReloadConfigString("lighthouse:\n interval: 11").Err)
require.NoError(t, c.ReloadConfigString("lighthouse:\n interval: 11"))
assert.Equal(t, int64(11), lh.interval.Load())
}
@@ -334,7 +334,7 @@ func TestLighthouse_reloadStaticHostMap(t *testing.T) {
}
rc, err := yaml.Marshal(nc)
require.NoError(t, err)
require.NoError(t, c.ReloadConfigString(string(rc)).Err)
require.NoError(t, c.ReloadConfigString(string(rc)))
rl := lh.Query(staticHost)
require.NotNil(t, rl)
@@ -350,7 +350,7 @@ func TestLighthouse_reloadStaticHostMap(t *testing.T) {
}
rc, err = yaml.Marshal(nc)
require.NoError(t, err)
require.NoError(t, c.ReloadConfigString(string(rc)).Err)
require.NoError(t, c.ReloadConfigString(string(rc)))
rl = lh.Query(staticHost)
require.NotNil(t, rl)
@@ -358,7 +358,7 @@ func TestLighthouse_reloadStaticHostMap(t *testing.T) {
assert.Equal(t, []netip.AddrPort{netip.MustParseAddrPort("1.1.1.1:4242")}, rl.CopyAddrs([]netip.Prefix{}))
// Reload with the same config. An unchanged entry must not duplicate.
require.NoError(t, c.ReloadConfigString(string(rc)).Err)
require.NoError(t, c.ReloadConfigString(string(rc)))
rl = lh.Query(staticHost)
require.NotNil(t, rl)
@@ -373,7 +373,7 @@ func TestLighthouse_reloadStaticHostMap(t *testing.T) {
}
rc, err = yaml.Marshal(nc)
require.NoError(t, err)
require.NoError(t, c.ReloadConfigString(string(rc)).Err)
require.NoError(t, c.ReloadConfigString(string(rc)))
// Add a second host alongside the first. Both should be present, neither duplicated.
nc = map[string]any{
@@ -384,7 +384,7 @@ func TestLighthouse_reloadStaticHostMap(t *testing.T) {
}
rc, err = yaml.Marshal(nc)
require.NoError(t, err)
require.NoError(t, c.ReloadConfigString(string(rc)).Err)
require.NoError(t, c.ReloadConfigString(string(rc)))
rl = lh.Query(staticHost)
require.NotNil(t, rl)
@@ -406,7 +406,7 @@ func TestLighthouse_reloadStaticHostMap(t *testing.T) {
}
rc, err = yaml.Marshal(nc)
require.NoError(t, err)
require.NoError(t, c.ReloadConfigString(string(rc)).Err)
require.NoError(t, c.ReloadConfigString(string(rc)))
_, isStatic := lh.GetStaticHostList()[staticHost]
assert.False(t, isStatic)
@@ -966,13 +966,13 @@ func TestLighthouse_reloadRelaysTriggersUpdate(t *testing.T) {
require.NoError(t, err)
assert.Empty(t, lh.updateTrigger)
require.NoError(t, c.ReloadConfigString(base+"relay:\n relays: [10.128.0.6]\n").Err)
require.NoError(t, c.ReloadConfigString(base+"relay:\n relays: [10.128.0.6]\n"))
assert.Equal(t, []netip.Addr{netip.MustParseAddr("10.128.0.6")}, lh.GetRelaysForMe())
require.Len(t, lh.updateTrigger, 1)
<-lh.updateTrigger
// The config changed but what we advertise did not, a bad entry is skipped
require.NoError(t, c.ReloadConfigString(base+"relay:\n relays: [10.128.0.6, nope]\n").Err)
require.NoError(t, c.ReloadConfigString(base+"relay:\n relays: [10.128.0.6, nope]\n"))
assert.Equal(t, []netip.Addr{netip.MustParseAddr("10.128.0.6")}, lh.GetRelaysForMe())
assert.Empty(t, lh.updateTrigger)
}
@@ -988,29 +988,12 @@ func TestLighthouse_reloadAmRelayAlone(t *testing.T) {
assert.Equal(t, []netip.Addr{netip.MustParseAddr("10.128.0.5")}, lh.GetRelaysForMe())
// Relays aren't allowed to have relays, flipping am_relay alone must drop them and tell the lighthouses
require.NoError(t, c.ReloadConfigString(base+" am_relay: true\n").Err)
require.NoError(t, c.ReloadConfigString(base+" am_relay: true\n"))
assert.Empty(t, lh.GetRelaysForMe())
require.Len(t, lh.updateTrigger, 1)
<-lh.updateTrigger
require.NoError(t, c.ReloadConfigString(base+" am_relay: false\n").Err)
require.NoError(t, c.ReloadConfigString(base+" am_relay: false\n"))
assert.Equal(t, []netip.Addr{netip.MustParseAddr("10.128.0.5")}, lh.GetRelaysForMe())
require.Len(t, lh.updateTrigger, 1)
}
func TestLighthouse_reloadRestartRequired(t *testing.T) {
l := test.NewLogger()
c := config.NewC(l)
base := "lighthouse:\n am_lighthouse: true\n"
require.NoError(t, c.LoadString(base+"listen:\n port: 4242\n"))
_, err := NewLightHouseFromConfig(t.Context(), l, c, testCertState(netip.MustParsePrefix("10.128.0.1/24")), nil, nil)
require.NoError(t, err)
res := c.ReloadConfigString(base + "listen:\n port: 4242\n")
require.NoError(t, res.Err)
assert.False(t, res.RestartRequired)
res = c.ReloadConfigString(base + "listen:\n port: 4243\n")
require.NoError(t, res.Err)
assert.True(t, res.RestartRequired)
}
+53 -17
View File
@@ -197,7 +197,7 @@ func Main(c *config.C, configTest bool, buildVersion string, l *slog.Logger, dev
if err != nil {
return nil, util.NewContextualError("Failed to open udp listener", m{"queue": i}, err)
}
udpServer.ReloadConfig(c, nil)
udpServer.ReloadConfig(c)
udpConns[i] = udpServer
// If port is dynamic, discover it before the next pass through the for loop
@@ -243,11 +243,19 @@ func Main(c *config.C, configTest bool, buildVersion string, l *slog.Logger, dev
}
pinThreads := c.GetBool("tun.pin_threads", true)
cpuAffinity := parseCpuAffinity(c, l, routines)
avoidCPUs := parseAvoidCPUs(c, l)
if len(avoidCPUs) > 0 && !pinThreads {
l.Warn("tun.avoid_cpus has no effect with tun.pin_threads disabled")
}
// Keeping every thread of the process off avoidCPUs, not just the pinned readers, is parked for now:
//if err := cpupick.Avoid(avoidCPUs, !configTest); err != nil && !errors.Is(err, errors.ErrUnsupported) {
// return nil, fmt.Errorf("tun.avoid_cpus: %w", err)
//}
cpuAffinity := parseCpuAffinity(c, l, routines, avoidCPUs)
if pinThreads && routines > 1 && len(cpuAffinity) == 0 && !configTest {
// The operator didn't choose pin CPUs, so pick a default set that
// prefers performance cores and doesn't stack co-located instances
// onto allowed[0].
// prefers performance cores, skips tun.avoid_cpus and doesn't stack
// co-located instances onto allowed[0].
// key is used to seed the spreading of routines->cores.
// use PID if you want to ensure many different Nebulas in VMs or containers land on different cores
@@ -270,7 +278,7 @@ func Main(c *config.C, configTest bool, buildVersion string, l *slog.Logger, dev
l.Warn("tun.pin_threads_key is invalid, using PID")
}
cpuAffinity = cpupick.Default(routines, key, l)
cpuAffinity = cpupick.Default(routines, key, avoidCPUs, l)
}
ifConfig := &InterfaceConfig{
@@ -295,6 +303,7 @@ func Main(c *config.C, configTest bool, buildVersion string, l *slog.Logger, dev
punchy: punchy,
ConntrackCacheTimeout: conntrackCacheTimeout,
CpuAffinity: cpuAffinity,
AvoidCPUs: avoidCPUs,
PinThreads: pinThreads,
l: l,
}
@@ -311,16 +320,6 @@ func Main(c *config.C, configTest bool, buildVersion string, l *slog.Logger, dev
lightHouse.ifce = ifce
ifce.RegisterConfigChangeCallbacks(c)
// These are only read at startup
c.RegisterReloadCallback(func(c *config.C, r *config.Reload) {
r.RestartRequiredIfChanged(
"routines", "tun.routines", "listen.routines", "firewall.conntrack.routine_cache_timeout",
"listen.port", "listen.host", "stats.message_metrics",
"handshakes.try_interval", "handshakes.retries", "handshakes.trigger_buffer",
"tun.drop_local_broadcast", "tun.drop_multicast",
)
r.RestartRequiredIfChanged(osStartupOnlyKeys...)
})
ifce.reloadDisconnectInvalid(c)
ifce.reloadSendRecvError(c)
ifce.reloadAcceptRecvError(c)
@@ -373,8 +372,10 @@ func Main(c *config.C, configTest bool, buildVersion string, l *slog.Logger, dev
// mask (util.AllowedCPUs) rather than 0..NumCPU-1: under a cgroup cpuset or
// taskset the runnable IDs are frequently not that contiguous range, and
// pinning to an unrunnable ID always fails. If the allowed set can't be
// determined we fall back to a plain non-negative check.
func parseCpuAffinity(c *config.C, l *slog.Logger, routines int) []int {
// determined we fall back to a plain non-negative check. An entry listed in
// avoid (tun.avoid_cpus) contradicts the operator's own config and is treated
// the same as any other invalid entry.
func parseCpuAffinity(c *config.C, l *slog.Logger, routines int, avoid []int) []int {
raw := c.Get("tun.cpu_affinity")
if raw == nil {
return nil
@@ -417,6 +418,11 @@ func parseCpuAffinity(c *config.C, l *slog.Logger, routines int) []int {
"index", i, "cpu", cpu, "allowed", allowed)
return nil
}
if slices.Contains(avoid, cpu) {
l.Warn("tun.cpu_affinity entry is listed in tun.avoid_cpus; ignoring affinity",
"index", i, "cpu", cpu)
return nil
}
cpus = append(cpus, cpu)
}
if len(cpus) != routines {
@@ -426,6 +432,36 @@ func parseCpuAffinity(c *config.C, l *slog.Logger, routines int) []int {
return cpus
}
// parseAvoidCPUs reads tun.avoid_cpus, a list of CPU IDs the tun readers must not be pinned to. IDs outside the
// allowed set are accepted, since one config may serve differently shaped hosts. Like tun.cpu_affinity, an invalid
// list is a warning and is ignored entirely, as is one covering every allowed CPU, which would leave nowhere to pin.
func parseAvoidCPUs(c *config.C, l *slog.Logger) []int {
raw := c.Get("tun.avoid_cpus")
if raw == nil {
return nil
}
values, ok := raw.([]any)
if !ok {
l.Warn("tun.avoid_cpus must be a list of integers; ignoring", "value", raw)
return nil
}
cpus := make([]int, 0, len(values))
for i, value := range values {
cpu, ok := value.(int)
if !ok || cpu < 0 {
l.Warn("tun.avoid_cpus entry not a non-negative integer; ignoring avoid_cpus", "index", i, "value", value)
return nil
}
cpus = append(cpus, cpu)
}
if allowed, err := util.AllowedCPUs(); err == nil && len(allowed) > 0 &&
!slices.ContainsFunc(allowed, func(cpu int) bool { return !slices.Contains(cpus, cpu) }) {
l.Warn("tun.avoid_cpus covers every allowed CPU; ignoring avoid_cpus", "cpus", cpus, "allowed", allowed)
return nil
}
return cpus
}
func moduleVersion() string {
info, ok := debug.ReadBuildInfo()
if !ok {
-6
View File
@@ -1,6 +0,0 @@
//go:build linux && !android && !e2e_testing
package nebula
// osStartupOnlyKeys are startup only keys that only matter on linux: udp batch size, udp offloads and thread pinning
var osStartupOnlyKeys = []string{"listen.batch", "listen.udp_offloads", "tun.pin_threads", "tun.pin_threads_key", "tun.cpu_affinity"}
-6
View File
@@ -1,6 +0,0 @@
//go:build !linux || android || e2e_testing
package nebula
// osStartupOnlyKeys is empty, none of the OS specific startup keys matter here
var osStartupOnlyKeys []string
+36 -5
View File
@@ -32,7 +32,7 @@ func TestParseCpuAffinity(t *testing.T) {
}
// unset -> nil (listenIn falls back to spreading across the allowed set)
assert.Nil(t, parseCpuAffinity(newConfig(nil), l, 1))
assert.Nil(t, parseCpuAffinity(newConfig(nil), l, 1, nil))
// Pick a CPU we're actually allowed to run on so a valid list survives
// validation regardless of the host's affinity mask.
@@ -43,19 +43,50 @@ func TestParseCpuAffinity(t *testing.T) {
}
// valid list -> parsed through unchanged
assert.Equal(t, []int{validCPU, validCPU}, parseCpuAffinity(newConfig([]any{validCPU, validCPU}), l, 2))
assert.Equal(t, []int{validCPU, validCPU}, parseCpuAffinity(newConfig([]any{validCPU, validCPU}), l, 2, nil))
// a negative entry is out of range on every platform -> disables the override
assert.Nil(t, parseCpuAffinity(newConfig([]any{validCPU, -1}), l, 2))
assert.Nil(t, parseCpuAffinity(newConfig([]any{validCPU, -1}), l, 2, nil))
// a non-integer entry -> disables the override
assert.Nil(t, parseCpuAffinity(newConfig([]any{validCPU, "not-a-cpu"}), l, 2))
assert.Nil(t, parseCpuAffinity(newConfig([]any{validCPU, "not-a-cpu"}), l, 2, nil))
// a CPU id outside the allowed set -> disables the override. Only assertable
// where we can enumerate the allowed set (e.g. linux); 1<<20 is far beyond
// any representable CPU id so it can never be in the mask.
if len(allowed) > 0 {
assert.Nil(t, parseCpuAffinity(newConfig([]any{1 << 20}), l, 1))
assert.Nil(t, parseCpuAffinity(newConfig([]any{1 << 20}), l, 1, nil))
}
// an entry also listed in tun.avoid_cpus -> disables the override
assert.Nil(t, parseCpuAffinity(newConfig([]any{validCPU}), l, 1, []int{validCPU}))
}
func TestParseAvoidCPUs(t *testing.T) {
l := test.NewLogger()
for _, tc := range []struct {
value any
want []int
}{
{nil, nil}, {[]any{}, []int{}}, {[]any{1, 1 << 20}, []int{1, 1 << 20}},
{"auto", nil}, {"1,2", nil}, {[]any{-1}, nil}, {[]any{1.5}, nil}, {[]any{1, "2"}, nil},
} {
c := config.NewC(l)
if tc.value != nil {
c.Settings["tun"] = map[string]any{"avoid_cpus": tc.value}
}
assert.Equal(t, tc.want, parseAvoidCPUs(c, l), "%v", tc.value)
}
// avoiding every allowed CPU leaves nowhere to pin; only checkable where the allowed set is known
if allowed, _ := util.AllowedCPUs(); len(allowed) > 0 {
all := make([]any, len(allowed))
for i, cpu := range allowed {
all[i] = cpu
}
c := config.NewC(l)
c.Settings["tun"] = map[string]any{"avoid_cpus": all}
assert.Nil(t, parseAvoidCPUs(c, l))
}
}
-7
View File
@@ -41,16 +41,9 @@ func (e *NameError) Error() string {
type DeviceFactory func(c *config.C, l *slog.Logger, vpnNetworks []netip.Prefix, routines int) (Device, error)
func NewDeviceFromConfig(c *config.C, l *slog.Logger, vpnNetworks []netip.Prefix, routines int) (Device, error) {
c.RegisterReloadCallback(func(c *config.C, r *config.Reload) {
r.RestartRequiredIfChanged("tun.disabled")
})
switch {
case c.GetBool("tun.disabled", false):
tun := newDisabledTun(vpnNetworks, c.GetInt("tun.tx_queue", 5000), c.GetBool("stats.message_metrics", false), l)
c.RegisterReloadCallback(func(c *config.C, r *config.Reload) {
r.RestartRequiredIfChanged("tun.tx_queue", "stats.message_metrics")
})
return tun, nil
default:
+4 -5
View File
@@ -38,17 +38,16 @@ func newTunFromFd(c *config.C, l *slog.Logger, deviceFd int, vpnNetworks []netip
l: l,
}
err := t.reload(c, true, nil)
err := t.reload(c, true)
if err != nil {
_ = file.Close()
return nil, err
}
c.RegisterReloadCallback(func(c *config.C, r *config.Reload) {
err := t.reload(c, false, r)
c.RegisterReloadCallback(func(c *config.C) {
err := t.reload(c, false)
if err != nil {
util.LogWithContextIfNeeded("failed to reload tun device", err, t.l)
r.Failed()
}
})
@@ -68,7 +67,7 @@ func (t *tun) Activate() error {
return nil
}
func (t *tun) reload(c *config.C, initial bool, r *config.Reload) error {
func (t *tun) reload(c *config.C, initial bool) error {
change, routes, err := getAllRoutesFromConfig(t.l, c, t.vpnNetworks, initial)
if err != nil {
return err
-3
View File
@@ -10,9 +10,6 @@ import (
"github.com/slackhq/nebula/wfp"
)
// interfaceBypassKeys are only read when the interface bypass is installed
var interfaceBypassKeys = []string{"tun.windows_bypass_wdf"}
// installInterfaceBypass installs a WFP PERMIT filter scoped to the wintun interface LUID so inbound traffic on the
// nebula adapter bypasses Windows Defender Firewall.
func installInterfaceBypass(l *slog.Logger, luid uint64) closer {
-3
View File
@@ -4,9 +4,6 @@ package overlay
import "log/slog"
// interfaceBypassKeys is empty, windows-386 has no bypass
var interfaceBypassKeys []string
// installInterfaceBypass is a no-op on windows-386 because we don't currently build for it.
func installInterfaceBypass(_ *slog.Logger, _ uint64) closer {
return nil
+13 -23
View File
@@ -132,17 +132,15 @@ func newTun(c *config.C, l *slog.Logger, vpnNetworks []netip.Prefix, _ bool) (*t
l: l,
}
err = t.reload(c, true, nil)
err = t.reload(c, true)
if err != nil {
return nil, err
}
c.RegisterReloadCallback(func(c *config.C, r *config.Reload) {
r.RestartRequiredIfChanged("tun.dev", "tun.mtu")
err := t.reload(c, false, r)
c.RegisterReloadCallback(func(c *config.C) {
err := t.reload(c, false)
if err != nil {
util.LogWithContextIfNeeded("failed to reload tun device", err, t.l)
r.Failed()
}
})
@@ -176,15 +174,14 @@ func newTunFromFd(c *config.C, l *slog.Logger, deviceFd int, vpnNetworks []netip
l: l,
}
if err := t.reload(c, true, nil); err != nil {
if err := t.reload(c, true); err != nil {
_ = file.Close()
return nil, err
}
c.RegisterReloadCallback(func(c *config.C, r *config.Reload) {
if err := t.reload(c, false, r); err != nil {
c.RegisterReloadCallback(func(c *config.C) {
if err := t.reload(c, false); err != nil {
util.LogWithContextIfNeeded("failed to reload tun device", err, t.l)
r.Failed()
}
})
@@ -271,7 +268,7 @@ func (t *tun) Activate() error {
}
// Unsafe path routes
return t.addRoutes(false, nil)
return t.addRoutes(false)
}
// activate4 assigns our vpn address as a /32 and routes the vpn network to the tun explicitly.
@@ -369,7 +366,7 @@ func (t *tun) activate6(network netip.Prefix) error {
return nil
}
func (t *tun) reload(c *config.C, initial bool, r *config.Reload) error {
func (t *tun) reload(c *config.C, initial bool) error {
change, routes, err := getAllRoutesFromConfig(t.l, c, t.vpnNetworks, initial)
if err != nil {
return err
@@ -390,18 +387,16 @@ func (t *tun) reload(c *config.C, initial bool, r *config.Reload) error {
if !initial {
// Remove first, if the system removes a wanted route hopefully it will be re-added next
err := t.removeRoutes(findRemovedRoutes(routes, *oldRoutes), r)
err := t.removeRoutes(findRemovedRoutes(routes, *oldRoutes))
if err != nil {
util.LogWithContextIfNeeded("Failed to remove routes", err, t.l)
r.Failed()
}
// Ensure any routes we actually want are installed
err = t.addRoutes(true, r)
err = t.addRoutes(true)
if err != nil {
// Catch any stray logs
util.LogWithContextIfNeeded("Failed to add routes", err, t.l)
r.Failed()
}
}
@@ -494,7 +489,7 @@ func getLinkAddr(name string) (*netroute.LinkAddr, error) {
return nil, nil
}
func (t *tun) addRoutes(logErrors bool, rl *config.Reload) error {
func (t *tun) addRoutes(logErrors bool) error {
// The route tree is still ours, the system routing table is not
if t.hostOwned {
return nil
@@ -529,7 +524,6 @@ func (t *tun) addRoutes(logErrors bool, rl *config.Reload) error {
retErr := util.NewContextualError("Failed to add route", map[string]any{"route": r}, err)
if logErrors {
retErr.Log(t.l)
rl.Failed()
} else {
return retErr
}
@@ -544,7 +538,7 @@ func (t *tun) addRoutes(logErrors bool, rl *config.Reload) error {
return nil
}
func (t *tun) removeRoutes(routes []Route, rl *config.Reload) error {
func (t *tun) removeRoutes(routes []Route) error {
linkAddr := t.linkAddr.Load()
if t.hostOwned || linkAddr == nil {
return nil
@@ -557,12 +551,8 @@ func (t *tun) removeRoutes(routes []Route, rl *config.Reload) error {
}
err := delRoute(r.Cidr, linkAddr)
if errors.Is(err, unix.ESRCH) {
// Already gone, which is what we wanted
t.l.Info("Route was already removed", "route", r)
} else if err != nil {
if err != nil {
t.l.Error("Failed to remove route", "error", err, "route", r)
rl.Failed()
} else {
t.l.Info("Removed route", "route", r)
}
+3 -5
View File
@@ -72,8 +72,8 @@ func TestTunRoutesBeforeActivate(t *testing.T) {
}}
tn.Routes.Store(&routes)
assert.NoError(t, tn.addRoutes(true, nil))
assert.NoError(t, tn.removeRoutes(routes, nil))
assert.NoError(t, tn.addRoutes(true))
assert.NoError(t, tn.removeRoutes(routes))
}
// A tun.routes entry with no via is never installed, so removing it must not delete a route for the same cidr that
@@ -96,9 +96,7 @@ func TestTunReloadLeavesRoutesItNeverInstalled(t *testing.T) {
t.Cleanup(func() { _ = tn.Close() })
require.NoError(t, tn.Activate())
res := c.ReloadConfigString("tun: {}\n")
require.NoError(t, res.Err)
assert.False(t, res.Failed, "removing a route we never installed failed the reload")
require.NoError(t, c.ReloadConfigString("tun: {}\n"))
msgs, err := fetchRoutes()
require.NoError(t, err)
assert.Equal(t, lo0.Index, routeInterfaceIn(msgs, netip.MustParsePrefix(cidr)))
+10 -19
View File
@@ -385,18 +385,16 @@ func newTun(c *config.C, l *slog.Logger, vpnNetworks []netip.Prefix, _ bool) (*t
},
}
err = t.reload(c, true, nil)
err = t.reload(c, true)
if err != nil {
return nil, err
}
closeOnErr = false
c.RegisterReloadCallback(func(c *config.C, r *config.Reload) {
r.RestartRequiredIfChanged("tun.dev", "tun.mtu")
err := t.reload(c, false, r)
c.RegisterReloadCallback(func(c *config.C) {
err := t.reload(c, false)
if err != nil {
util.LogWithContextIfNeeded("failed to reload tun device", err, t.l)
r.Failed()
}
})
@@ -495,7 +493,7 @@ func (t *tun) Activate() error {
}
}
return t.addRoutes(false, nil)
return t.addRoutes(false)
}
func (t *tun) setMTU() error {
@@ -511,7 +509,7 @@ func (t *tun) setMTU() error {
return err
}
func (t *tun) reload(c *config.C, initial bool, r *config.Reload) error {
func (t *tun) reload(c *config.C, initial bool) error {
change, routes, err := getAllRoutesFromConfig(t.l, c, t.vpnNetworks, initial)
if err != nil {
return err
@@ -532,18 +530,16 @@ func (t *tun) reload(c *config.C, initial bool, r *config.Reload) error {
if !initial {
// Remove first, if the system removes a wanted route hopefully it will be re-added next
err := t.removeRoutes(findRemovedRoutes(routes, *oldRoutes), r)
err := t.removeRoutes(findRemovedRoutes(routes, *oldRoutes))
if err != nil {
util.LogWithContextIfNeeded("Failed to remove routes", err, t.l)
r.Failed()
}
// Ensure any routes we actually want are installed
err = t.addRoutes(true, r)
err = t.addRoutes(true)
if err != nil {
// Catch any stray logs
util.LogWithContextIfNeeded("Failed to add routes", err, t.l)
r.Failed()
}
}
@@ -567,7 +563,7 @@ func (t *tun) Queues(int) ([]tio.Queue, error) {
return []tio.Queue{tio.NewSingleQueue(t, defaultBatchBufSize)}, nil
}
func (t *tun) addRoutes(logErrors bool, rl *config.Reload) error {
func (t *tun) addRoutes(logErrors bool) error {
routes := *t.Routes.Load()
for _, r := range routes {
if len(r.Via) == 0 || !r.Install {
@@ -580,7 +576,6 @@ func (t *tun) addRoutes(logErrors bool, rl *config.Reload) error {
retErr := util.NewContextualError("Failed to add route", map[string]any{"route": r}, err)
if logErrors {
retErr.Log(t.l)
rl.Failed()
} else {
return retErr
}
@@ -592,7 +587,7 @@ func (t *tun) addRoutes(logErrors bool, rl *config.Reload) error {
return nil
}
func (t *tun) removeRoutes(routes []Route, rl *config.Reload) error {
func (t *tun) removeRoutes(routes []Route) error {
for _, r := range routes {
if len(r.Via) == 0 || !r.Install {
// addRoutes never installs these
@@ -600,12 +595,8 @@ func (t *tun) removeRoutes(routes []Route, rl *config.Reload) error {
}
err := delRoute(r.Cidr, t.linkAddr)
if errors.Is(err, unix.ESRCH) {
// Already gone, which is what we wanted
t.l.Info("Route was already removed", "route", r)
} else if err != nil {
if err != nil {
t.l.Error("Failed to remove route", "error", err, "route", r)
rl.Failed()
} else {
t.l.Info("Removed route", "route", r)
}
+4 -5
View File
@@ -47,17 +47,16 @@ func newTunFromFd(c *config.C, l *slog.Logger, deviceFd int, vpnNetworks []netip
l: l,
}
err := t.reload(c, true, nil)
err := t.reload(c, true)
if err != nil {
_ = file.Close()
return nil, err
}
c.RegisterReloadCallback(func(c *config.C, r *config.Reload) {
err := t.reload(c, false, r)
c.RegisterReloadCallback(func(c *config.C) {
err := t.reload(c, false)
if err != nil {
util.LogWithContextIfNeeded("failed to reload tun device", err, t.l)
r.Failed()
}
})
@@ -68,7 +67,7 @@ func (t *tun) Activate() error {
return nil
}
func (t *tun) reload(c *config.C, initial bool, r *config.Reload) error {
func (t *tun) reload(c *config.C, initial bool) error {
change, routes, err := getAllRoutesFromConfig(t.l, c, t.vpnNetworks, initial)
if err != nil {
return err
+31 -111
View File
@@ -185,10 +185,6 @@ func newTun(c *config.C, l *slog.Logger, vpnNetworks []netip.Prefix, multiqueue
}
t.Device = name
// Only read when we create the device, an fd handed to us already has its name and offloads
c.RegisterReloadCallback(func(c *config.C, r *config.Reload) {
r.RestartRequiredIfChanged("tun.dev", "tun.use_offloads")
})
return t, nil
}
@@ -232,30 +228,28 @@ func newTunGeneric(c *config.C, l *slog.Logger, fd int, vnetHdr bool, offloadFla
l: l,
}
if err = t.reload(c, true, nil); err != nil {
if err = t.reload(c, true); err != nil {
_ = t.Close()
return nil, err
}
c.RegisterReloadCallback(func(c *config.C, r *config.Reload) {
r.RestartRequiredIfChanged("tun.tx_queue", "tun.use_system_route_table", "tun.use_system_route_table_buffer_size")
err := t.reload(c, false, r)
c.RegisterReloadCallback(func(c *config.C) {
err := t.reload(c, false)
if err != nil {
util.LogWithContextIfNeeded("failed to reload tun device", err, t.l)
r.Failed()
}
})
return t, nil
}
func (t *tun) reload(c *config.C, initial bool, r *config.Reload) error {
func (t *tun) reload(c *config.C, initial bool) error {
routeChange, routes, err := getAllRoutesFromConfig(t.l, c, t.vpnNetworks, initial)
if err != nil {
return err
}
if !initial && !routeChange && !c.HasChanged("tun.mtu") && t.deviceHasConfiguredMTU() {
if !initial && !routeChange && !c.HasChanged("tun.mtu") {
return nil
}
@@ -280,6 +274,8 @@ func (t *tun) reload(c *config.C, initial bool, r *config.Reload) error {
}
t.routesFromSystemLock.Unlock()
oldDefaultMTU := t.DefaultMTU
oldMaxMTU := t.MaxMTU
newDefaultMTU := getMTU(t.l, c)
newMaxMTU := newDefaultMTU
for i, r := range routes {
@@ -300,39 +296,30 @@ func (t *tun) reload(c *config.C, initial bool, r *config.Reload) error {
t.routeTree.Store(routeTree)
if !initial {
// Compare with the device rather than the last config, the mtu may have been refused or changed under us
if !t.deviceHasMaxMTU() {
oldMTU, _ := t.deviceMTU()
if err := t.setMTU(t.MaxMTU); err != nil {
t.l.Error("Failed to set tun mtu", "mtu", t.MaxMTU, "error", err)
r.Failed()
} else {
t.l.Info("Set max MTU", "mtu", t.MaxMTU, "oldMTU", oldMTU)
}
if oldMaxMTU != newMaxMTU {
t.setMTU()
t.l.Info("Set max MTU", "mtu", t.MaxMTU, "oldMTU", oldMaxMTU)
}
for _, network := range t.vpnNetworks {
if t.defaultRouteHasMTU(network) {
continue
}
err := t.setDefaultRoute(network)
if err != nil {
t.l.Error(err.Error())
r.Failed()
} else {
t.l.Info("Set default MTU", "mtu", t.DefaultMTU, "network", network)
if oldDefaultMTU != newDefaultMTU {
for i := range t.vpnNetworks {
err := t.setDefaultRoute(t.vpnNetworks[i])
if err != nil {
t.l.Warn(err.Error())
} else {
t.l.Info("Set default MTU", "mtu", t.DefaultMTU, "oldMTU", oldDefaultMTU)
}
}
}
// Remove first, if the system removes a wanted route hopefully it will be re-added next
t.removeRoutes(findRemovedRoutes(routes, *oldRoutes), r)
t.removeRoutes(findRemovedRoutes(routes, *oldRoutes))
// Ensure any routes we actually want are installed
err = t.addRoutes(true, r)
err = t.addRoutes(true)
if err != nil {
// This should never be called since addRoutes should log its own errors in a reload condition
util.LogWithContextIfNeeded("Failed to refresh routes", err, t.l)
r.Failed()
}
}
@@ -487,10 +474,7 @@ func (t *tun) Activate() error {
t.deviceIndex = link.Attrs().Index
// Setup our default MTU
if err := t.setMTU(t.MaxMTU); err != nil {
// This is currently a non fatal condition because the route table must have the MTU set appropriately as well
t.l.Error("Failed to set tun mtu", "error", err)
}
t.setMTU()
// Set the transmit queue length
ifrq := ifreqQLEN{Name: devName, Value: int32(t.TXQueueLen)}
@@ -522,7 +506,7 @@ func (t *tun) Activate() error {
}
// Set the routes
if err = t.addRoutes(false, nil); err != nil {
if err = t.addRoutes(false); err != nil {
return err
}
@@ -535,72 +519,13 @@ func (t *tun) Activate() error {
return nil
}
func (t *tun) setMTU(mtu int) error {
ifm := ifreqMTU{Name: t.deviceBytes(), MTU: int32(mtu)}
return ioctl(t.ioctlFd, unix.SIOCSIFMTU, uintptr(unsafe.Pointer(&ifm)))
}
// deviceMTU asks the kernel for the device's mtu.
func (t *tun) deviceMTU() (int, error) {
ifm := ifreqMTU{Name: t.deviceBytes()}
if err := ioctl(t.ioctlFd, unix.SIOCGIFMTU, uintptr(unsafe.Pointer(&ifm))); err != nil {
return 0, err
func (t *tun) setMTU() {
// Set the MTU on the device
ifm := ifreqMTU{Name: t.deviceBytes(), MTU: int32(t.MaxMTU)}
if err := ioctl(t.ioctlFd, unix.SIOCSIFMTU, uintptr(unsafe.Pointer(&ifm))); err != nil {
// This is currently a non fatal condition because the route table must have the MTU set appropriately as well
t.l.Error("Failed to set tun mtu", "error", err)
}
return int(ifm.MTU), nil
}
// deviceHasConfiguredMTU is true when the device and the route for each of our networks carry the mtu the config asks
// for.
func (t *tun) deviceHasConfiguredMTU() bool {
if !t.deviceHasMaxMTU() {
return false
}
for _, network := range t.vpnNetworks {
if !t.defaultRouteHasMTU(network) {
return false
}
}
return true
}
// defaultRouteHasMTU is true when the kernel's route for network on our device has the default mtu and advmss. Before
// Activate there's nothing to check, Activate installs it.
func (t *tun) defaultRouteHasMTU(network netip.Prefix) bool {
if t.deviceIndex == 0 {
return true
}
family := netlink.FAMILY_V4
if network.Addr().Is6() {
family = netlink.FAMILY_V6
}
filter := &netlink.Route{
LinkIndex: t.deviceIndex,
Dst: &net.IPNet{
IP: network.Masked().Addr().AsSlice(),
Mask: net.CIDRMask(network.Bits(), network.Addr().BitLen()),
},
Table: unix.RT_TABLE_MAIN,
}
routes, err := netlink.RouteListFiltered(family, filter, netlink.RT_FILTER_OIF|netlink.RT_FILTER_DST|netlink.RT_FILTER_TABLE)
if err != nil {
return false
}
for _, r := range routes {
if r.MTU == t.DefaultMTU && r.AdvMSS == t.advMSS(Route{}) {
return true
}
}
return false
}
// deviceHasMaxMTU is true when the device's mtu is the one the config asks for. Before Activate there's nothing to
// check, Activate sets it.
func (t *tun) deviceHasMaxMTU() bool {
if t.ioctlFd == 0 {
return true
}
mtu, err := t.deviceMTU()
return err == nil && mtu == t.MaxMTU
}
func (t *tun) setDefaultRoute(cidr netip.Prefix) error {
@@ -652,7 +577,7 @@ func (t *tun) setDefaultRoute(cidr netip.Prefix) error {
return nil
}
func (t *tun) addRoutes(logErrors bool, rl *config.Reload) error {
func (t *tun) addRoutes(logErrors bool) error {
// Path routes
routes := *t.Routes.Load()
for _, r := range routes {
@@ -682,7 +607,6 @@ func (t *tun) addRoutes(logErrors bool, rl *config.Reload) error {
retErr := util.NewContextualError("Failed to add route", map[string]any{"route": r}, err)
if logErrors {
retErr.Log(t.l)
rl.Failed()
} else {
return retErr
}
@@ -694,7 +618,7 @@ func (t *tun) addRoutes(logErrors bool, rl *config.Reload) error {
return nil
}
func (t *tun) removeRoutes(routes []Route, rl *config.Reload) {
func (t *tun) removeRoutes(routes []Route) {
for _, r := range routes {
if !r.Install {
continue
@@ -718,12 +642,8 @@ func (t *tun) removeRoutes(routes []Route, rl *config.Reload) {
}
err := netlink.RouteDel(&nr)
if errors.Is(err, unix.ESRCH) {
// Already gone, which is what we wanted
t.l.Info("Route was already removed", "route", r)
} else if err != nil {
if err != nil {
t.l.Error("Failed to remove route", "error", err, "route", r)
rl.Failed()
} else {
t.l.Info("Removed route", "route", r)
}
+10 -19
View File
@@ -103,17 +103,15 @@ func newTun(c *config.C, l *slog.Logger, vpnNetworks []netip.Prefix, _ bool) (*t
l: l,
}
err = t.reload(c, true, nil)
err = t.reload(c, true)
if err != nil {
return nil, err
}
c.RegisterReloadCallback(func(c *config.C, r *config.Reload) {
r.RestartRequiredIfChanged("tun.dev", "tun.mtu")
err := t.reload(c, false, r)
c.RegisterReloadCallback(func(c *config.C) {
err := t.reload(c, false)
if err != nil {
util.LogWithContextIfNeeded("failed to reload tun device", err, t.l)
r.Failed()
}
})
@@ -326,7 +324,7 @@ func (t *tun) Activate() error {
}
}
return t.addRoutes(false, nil)
return t.addRoutes(false)
}
func (t *tun) doIoctlByName(ctl uintptr, value uint32) error {
@@ -341,7 +339,7 @@ func (t *tun) doIoctlByName(ctl uintptr, value uint32) error {
return err
}
func (t *tun) reload(c *config.C, initial bool, r *config.Reload) error {
func (t *tun) reload(c *config.C, initial bool) error {
change, routes, err := getAllRoutesFromConfig(t.l, c, t.vpnNetworks, initial)
if err != nil {
return err
@@ -362,18 +360,16 @@ func (t *tun) reload(c *config.C, initial bool, r *config.Reload) error {
if !initial {
// Remove first, if the system removes a wanted route hopefully it will be re-added next
err := t.removeRoutes(findRemovedRoutes(routes, *oldRoutes), r)
err := t.removeRoutes(findRemovedRoutes(routes, *oldRoutes))
if err != nil {
util.LogWithContextIfNeeded("Failed to remove routes", err, t.l)
r.Failed()
}
// Ensure any routes we actually want are installed
err = t.addRoutes(true, r)
err = t.addRoutes(true)
if err != nil {
// Catch any stray logs
util.LogWithContextIfNeeded("Failed to add routes", err, t.l)
r.Failed()
}
}
@@ -397,7 +393,7 @@ func (t *tun) Queues(int) ([]tio.Queue, error) {
return []tio.Queue{tio.NewSingleQueue(t, defaultBatchBufSize)}, nil
}
func (t *tun) addRoutes(logErrors bool, rl *config.Reload) error {
func (t *tun) addRoutes(logErrors bool) error {
routes := *t.Routes.Load()
for _, r := range routes {
@@ -411,7 +407,6 @@ func (t *tun) addRoutes(logErrors bool, rl *config.Reload) error {
retErr := util.NewContextualError("Failed to add route", map[string]any{"route": r}, err)
if logErrors {
retErr.Log(t.l)
rl.Failed()
} else {
return retErr
}
@@ -423,7 +418,7 @@ func (t *tun) addRoutes(logErrors bool, rl *config.Reload) error {
return nil
}
func (t *tun) removeRoutes(routes []Route, rl *config.Reload) error {
func (t *tun) removeRoutes(routes []Route) error {
for _, r := range routes {
if len(r.Via) == 0 || !r.Install {
// addRoutes never installs these
@@ -431,12 +426,8 @@ func (t *tun) removeRoutes(routes []Route, rl *config.Reload) error {
}
err := delRoute(r.Cidr, t.vpnNetworks)
if errors.Is(err, unix.ESRCH) {
// Already gone, which is what we wanted
t.l.Info("Route was already removed", "route", r)
} else if err != nil {
if err != nil {
t.l.Error("Failed to remove route", "error", err, "route", r)
rl.Failed()
} else {
t.l.Info("Removed route", "route", r)
}
+10 -19
View File
@@ -94,17 +94,15 @@ func newTun(c *config.C, l *slog.Logger, vpnNetworks []netip.Prefix, _ bool) (*t
l: l,
}
err = t.reload(c, true, nil)
err = t.reload(c, true)
if err != nil {
return nil, err
}
c.RegisterReloadCallback(func(c *config.C, r *config.Reload) {
r.RestartRequiredIfChanged("tun.dev", "tun.mtu")
err := t.reload(c, false, r)
c.RegisterReloadCallback(func(c *config.C) {
err := t.reload(c, false)
if err != nil {
util.LogWithContextIfNeeded("failed to reload tun device", err, t.l)
r.Failed()
}
})
@@ -305,7 +303,7 @@ func (t *tun) Activate() error {
}
}
return t.addRoutes(false, nil)
return t.addRoutes(false)
}
func (t *tun) doIoctlByName(ctl uintptr, value uint32) error {
@@ -320,7 +318,7 @@ func (t *tun) doIoctlByName(ctl uintptr, value uint32) error {
return err
}
func (t *tun) reload(c *config.C, initial bool, r *config.Reload) error {
func (t *tun) reload(c *config.C, initial bool) error {
change, routes, err := getAllRoutesFromConfig(t.l, c, t.vpnNetworks, initial)
if err != nil {
return err
@@ -341,18 +339,16 @@ func (t *tun) reload(c *config.C, initial bool, r *config.Reload) error {
if !initial {
// Remove first, if the system removes a wanted route hopefully it will be re-added next
err := t.removeRoutes(findRemovedRoutes(routes, *oldRoutes), r)
err := t.removeRoutes(findRemovedRoutes(routes, *oldRoutes))
if err != nil {
util.LogWithContextIfNeeded("Failed to remove routes", err, t.l)
r.Failed()
}
// Ensure any routes we actually want are installed
err = t.addRoutes(true, r)
err = t.addRoutes(true)
if err != nil {
// Catch any stray logs
util.LogWithContextIfNeeded("Failed to add routes", err, t.l)
r.Failed()
}
}
@@ -376,7 +372,7 @@ func (t *tun) Queues(int) ([]tio.Queue, error) {
return []tio.Queue{tio.NewSingleQueue(t, defaultBatchBufSize)}, nil
}
func (t *tun) addRoutes(logErrors bool, rl *config.Reload) error {
func (t *tun) addRoutes(logErrors bool) error {
routes := *t.Routes.Load()
for _, r := range routes {
@@ -390,7 +386,6 @@ func (t *tun) addRoutes(logErrors bool, rl *config.Reload) error {
retErr := util.NewContextualError("Failed to add route", map[string]any{"route": r}, err)
if logErrors {
retErr.Log(t.l)
rl.Failed()
} else {
return retErr
}
@@ -402,7 +397,7 @@ func (t *tun) addRoutes(logErrors bool, rl *config.Reload) error {
return nil
}
func (t *tun) removeRoutes(routes []Route, rl *config.Reload) error {
func (t *tun) removeRoutes(routes []Route) error {
for _, r := range routes {
if len(r.Via) == 0 || !r.Install {
// addRoutes never installs these
@@ -410,12 +405,8 @@ func (t *tun) removeRoutes(routes []Route, rl *config.Reload) error {
}
err := delRoute(r.Cidr, t.vpnNetworks)
if errors.Is(err, unix.ESRCH) {
// Already gone, which is what we wanted
t.l.Info("Route was already removed", "route", r)
} else if err != nil {
if err != nil {
t.l.Error("Failed to remove route", "error", err, "route", r)
rl.Failed()
} else {
t.l.Info("Removed route", "route", r)
}
+4 -137
View File
@@ -4,16 +4,13 @@ package overlay
import (
"net/netip"
"os"
"testing"
"github.com/DefinedNet/netlink"
"github.com/slackhq/nebula/config"
"github.com/slackhq/nebula/routing"
"github.com/slackhq/nebula/test"
"github.com/stretchr/testify/assert"
"github.com/stretchr/testify/require"
"golang.org/x/sys/unix"
)
// Only tun.mtu changes. The system calls fail without a real device, what matters is the routes nebula keeps
@@ -28,10 +25,10 @@ func TestReloadMTUOnlyKeepsRoutes(t *testing.T) {
ioctlFd: ^uintptr(0),
routesFromSystem: map[netip.Prefix]routing.Gateways{},
}
require.NoError(t, tn.reload(c, true, nil))
require.NoError(t, tn.reload(c, true))
require.NoError(t, c.ReloadConfigString("tun:\n mtu: 1400\n"+routes).Err)
_ = tn.reload(c, false, nil)
require.NoError(t, c.ReloadConfigString("tun:\n mtu: 1400\n"+routes))
_ = tn.reload(c, false)
got := *tn.Routes.Load()
require.Len(t, got, 1)
@@ -53,137 +50,7 @@ func TestReloadMaxMTUIsTheLargestRoute(t *testing.T) {
ioctlFd: ^uintptr(0),
routesFromSystem: map[netip.Prefix]routing.Gateways{},
}
require.NoError(t, tn.reload(c, true, nil))
require.NoError(t, tn.reload(c, true))
assert.Equal(t, 1500, tn.MaxMTU)
assert.Equal(t, 1300, tn.DefaultMTU)
}
// There is no device behind this tun, so adding a route always fails. Removing one is refused without root, as root the
// kernel answers that the route isn't there, which is what we wanted
func TestReloadRouteFailures(t *testing.T) {
route := "tun:\n mtu: 1300\n unsafe_routes:\n - route: 192.168.100.0/24\n via: 10.1.0.1\n"
cases := []struct {
name string
from, to string
wantFails bool
}{
{"added", "tun:\n mtu: 1300\n", route, true},
{"removed", route, "tun:\n mtu: 1300\n", os.Geteuid() != 0},
{"unchanged", route, route + "punchy:\n punch: true\n", false},
}
for _, tc := range cases {
t.Run(tc.name, func(t *testing.T) {
c := config.NewC(test.NewLogger())
require.NoError(t, c.LoadString(tc.from))
tn := &tun{
l: test.NewLogger(),
vpnNetworks: []netip.Prefix{netip.MustParsePrefix("10.1.0.2/24")},
ioctlFd: 0, // not activated, the mtu is not checked
routesFromSystem: map[netip.Prefix]routing.Gateways{},
}
require.NoError(t, tn.reload(c, true, nil))
c.RegisterReloadCallback(func(c *config.C, r *config.Reload) {
require.NoError(t, tn.reload(c, false, r))
})
res := c.ReloadConfigString(tc.to)
require.NoError(t, res.Err)
assert.Equal(t, tc.wantFails, res.Failed)
})
}
}
// Keys only read when the device is created need a restart to apply
func TestTunReloadRestartKeys(t *testing.T) {
c := config.NewC(test.NewLogger())
require.NoError(t, c.LoadString("tun:\n dev: nebrk0\n"))
tn, err := NewDeviceFromConfig(c, test.NewLogger(), []netip.Prefix{netip.MustParsePrefix("10.1.0.2/24")}, 1)
if err != nil {
t.Skipf("can't create a tun device here: %v", err)
}
t.Cleanup(func() { _ = tn.Close() })
res := c.ReloadConfigString("tun:\n disabled: true\n dev: nebrk1\n tx_queue: 100\n")
require.NoError(t, res.Err)
assert.ElementsMatch(t, []string{"tun.disabled", "tun.dev", "tun.tx_queue"}, res.RestartKeys)
}
// There is no device behind this tun, so it refuses every mtu. Repeating the reload tries again rather than reporting
// it done
func TestReloadRefusedMaxMTURetries(t *testing.T) {
c := config.NewC(test.NewLogger())
require.NoError(t, c.LoadString("tun:\n mtu: 1300\n"))
tn := &tun{
l: test.NewLogger(),
vpnNetworks: []netip.Prefix{netip.MustParsePrefix("10.1.0.2/24")},
ioctlFd: ^uintptr(0),
routesFromSystem: map[netip.Prefix]routing.Gateways{},
}
require.NoError(t, tn.reload(c, true, nil))
c.RegisterReloadCallback(func(c *config.C, r *config.Reload) {
require.NoError(t, tn.reload(c, false, r))
})
for range 2 {
res := c.ReloadConfigString("tun:\n mtu: 1400\n")
require.NoError(t, res.Err)
assert.True(t, res.Failed)
}
}
// A reload puts the configured mtu back on the device and our network's route when they were changed outside nebula
func TestTunReloadRestoresMTU(t *testing.T) {
c := config.NewC(test.NewLogger())
require.NoError(t, c.LoadString("tun:\n dev: nebmtu0\n mtu: 1300\n"))
tn, err := newTun(c, test.NewLogger(), []netip.Prefix{netip.MustParsePrefix("10.1.0.2/24")}, false)
if err != nil {
t.Skipf("can't create a tun device here: %v", err)
}
t.Cleanup(func() { _ = tn.Close() })
require.NoError(t, tn.Activate())
link, err := netlink.LinkByName(tn.Device)
require.NoError(t, err)
require.NoError(t, netlink.LinkSetMTU(link, 1280))
network := &netlink.Route{LinkIndex: tn.deviceIndex, Table: unix.RT_TABLE_MAIN}
routeMTU := func() int {
routes, err := netlink.RouteListFiltered(netlink.FAMILY_V4, network, netlink.RT_FILTER_OIF|netlink.RT_FILTER_TABLE)
require.NoError(t, err)
for _, r := range routes {
if r.Dst != nil && r.Dst.String() == "10.1.0.0/24" {
return r.MTU
}
}
t.Fatal("no route for our network")
return 0
}
require.Equal(t, 1300, routeMTU())
setRouteMTU := func(mtu int) {
routes, err := netlink.RouteListFiltered(netlink.FAMILY_V4, network, netlink.RT_FILTER_OIF|netlink.RT_FILTER_TABLE)
require.NoError(t, err)
for _, r := range routes {
if r.Dst != nil && r.Dst.String() == "10.1.0.0/24" {
r.MTU = mtu
require.NoError(t, netlink.RouteReplace(&r))
}
}
require.Equal(t, mtu, routeMTU())
}
reload := func() {
res := c.ReloadConfigString("tun:\n dev: nebmtu0\n mtu: 1300\n")
require.NoError(t, res.Err)
assert.False(t, res.Failed)
}
setRouteMTU(1280)
reload()
mtu, err := tn.deviceMTU()
require.NoError(t, err)
assert.Equal(t, 1300, mtu)
assert.Equal(t, 1300, routeMTU())
// Only the route this time
setRouteMTU(1280)
reload()
assert.Equal(t, 1300, routeMTU())
}
-24
View File
@@ -1,24 +0,0 @@
package overlay
import (
"net/netip"
"testing"
"github.com/slackhq/nebula/config"
"github.com/slackhq/nebula/test"
"github.com/stretchr/testify/assert"
"github.com/stretchr/testify/require"
)
// tun.disabled and the disabled tun's settings are only read when the device is made
func TestDisabledTunRestartKeys(t *testing.T) {
c := config.NewC(test.NewLogger())
require.NoError(t, c.LoadString("tun:\n disabled: true\n"))
d, err := NewDeviceFromConfig(c, test.NewLogger(), []netip.Prefix{netip.MustParsePrefix("10.1.0.2/24")}, 1)
require.NoError(t, err)
t.Cleanup(func() { _ = d.Close() })
res := c.ReloadConfigString("tun:\n disabled: false\n tx_queue: 100\n")
require.NoError(t, res.Err)
assert.ElementsMatch(t, []string{"tun.disabled", "tun.tx_queue"}, res.RestartKeys)
}
+10 -20
View File
@@ -84,7 +84,7 @@ func newTun(c *config.C, l *slog.Logger, vpnNetworks []netip.Prefix, _ bool) (*w
l: l,
}
err = t.reload(c, true, nil)
err = t.reload(c, true)
if err != nil {
return nil, err
}
@@ -108,20 +108,17 @@ func newTun(c *config.C, l *slog.Logger, vpnNetworks []netip.Prefix, _ bool) (*w
// Disable registration before Activate assigns addresses and triggers it.
disableDNSRegistration(l, *guid)
c.RegisterReloadCallback(func(c *config.C, r *config.Reload) {
r.RestartRequiredIfChanged("tun.dev", "tun.mtu", "tun.network_category")
r.RestartRequiredIfChanged(interfaceBypassKeys...)
err := t.reload(c, false, r)
c.RegisterReloadCallback(func(c *config.C) {
err := t.reload(c, false)
if err != nil {
util.LogWithContextIfNeeded("failed to reload tun device", err, t.l)
r.Failed()
}
})
return t, nil
}
func (t *winTun) reload(c *config.C, initial bool, r *config.Reload) error {
func (t *winTun) reload(c *config.C, initial bool) error {
change, routes, err := getAllRoutesFromConfig(t.l, c, t.vpnNetworks, initial)
if err != nil {
return err
@@ -142,18 +139,16 @@ func (t *winTun) reload(c *config.C, initial bool, r *config.Reload) error {
if !initial {
// Remove first, if the system removes a wanted route hopefully it will be re-added next
err := t.removeRoutes(findRemovedSystemRoutes(routes, *oldRoutes), r)
err := t.removeRoutes(findRemovedSystemRoutes(routes, *oldRoutes))
if err != nil {
util.LogWithContextIfNeeded("Failed to remove routes", err, t.l)
r.Failed()
}
// Ensure any routes we actually want are installed
err = t.addRoutes(true, r)
err = t.addRoutes(true)
if err != nil {
// Catch any stray logs
util.LogWithContextIfNeeded("Failed to add routes", err, t.l)
r.Failed()
}
}
@@ -168,7 +163,7 @@ func (t *winTun) Activate() error {
return fmt.Errorf("failed to set address: %w", err)
}
err = t.addRoutes(false, nil)
err = t.addRoutes(false)
if err != nil {
return err
}
@@ -187,7 +182,7 @@ func (t *winTun) Activate() error {
return nil
}
func (t *winTun) addRoutes(logErrors bool, rl *config.Reload) error {
func (t *winTun) addRoutes(logErrors bool) error {
luid := winipcfg.LUID(t.tun.LUID())
routes := *t.Routes.Load()
foundDefault4 := false
@@ -218,7 +213,6 @@ func (t *winTun) addRoutes(logErrors bool, rl *config.Reload) error {
retErr := util.NewContextualError("Failed to add route", map[string]any{"route": r}, err)
if logErrors {
retErr.Log(t.l)
rl.Failed()
continue
} else {
return retErr
@@ -280,7 +274,7 @@ func (t *winTun) setMTU(luid winipcfg.LUID, foundDefault4, carriesV6 bool) error
return nil
}
func (t *winTun) removeRoutes(routes []Route, rl *config.Reload) error {
func (t *winTun) removeRoutes(routes []Route) error {
luid := winipcfg.LUID(t.tun.LUID())
for _, r := range routes {
@@ -290,12 +284,8 @@ func (t *winTun) removeRoutes(routes []Route, rl *config.Reload) error {
// See comment on luid.AddRoute
err := luid.DeleteRoute(r.Cidr, unspecifiedNextHop(r.Cidr))
if errors.Is(err, windows.ERROR_NOT_FOUND) {
// Already gone, which is what we wanted
t.l.Info("Route was already removed", "route", r)
} else if err != nil {
if err != nil {
t.l.Error("Failed to remove route", "error", err, "route", r)
rl.Failed()
} else {
t.l.Info("Removed route", "route", r)
}
+5 -25
View File
@@ -53,14 +53,6 @@ func (rt *routeTest) metric(t *testing.T, cidr string) uint32 {
return row.Metric
}
// reload applies cfg and expects nothing in it to fail
func (rt *routeTest) reload(t *testing.T, cfg string) {
t.Helper()
res := rt.c.ReloadConfigString(cfg)
require.NoError(t, res.Err)
assert.False(t, res.Failed, "the reload reported a failure")
}
func unsafeRoutes(routes ...string) string {
var b strings.Builder
b.WriteString("tun:\n dev: nebroutetest\n unsafe_routes:\n")
@@ -85,7 +77,7 @@ func TestWinTun_ReloadKeepsInstalledRoutes(t *testing.T) {
rt := newRouteTest(t, unsafeRoutes(a))
require.NoError(t, rt.tun.Activate())
rt.reload(t, unsafeRoutes(a, b))
require.NoError(t, rt.c.ReloadConfigString(unsafeRoutes(a, b)))
assert.NotContains(t, rt.log.String(), "Failed to add route")
rt.metric(t, a)
rt.metric(t, b)
@@ -100,7 +92,7 @@ func TestWinTun_ReloadMetricChangeApplies(t *testing.T) {
require.NoError(t, rt.tun.Activate())
require.Equal(t, uint32(100), rt.metric(t, a))
rt.reload(t, unsafeRoutes(a+"@200"))
require.NoError(t, rt.c.ReloadConfigString(unsafeRoutes(a+"@200")))
assert.NotContains(t, rt.log.String(), "Failed to add route")
assert.NotContains(t, rt.log.String(), "Removed route")
assert.Equal(t, uint32(200), rt.metric(t, a))
@@ -128,7 +120,7 @@ func TestWinTun_DuplicateCidrInstallsTheLastEntry(t *testing.T) {
require.NoError(t, rt.tun.Activate())
assert.Equal(t, uint32(200), rt.metric(t, a))
rt.reload(t, unsafeRoutes(a+"@100", a+"@200", b))
require.NoError(t, rt.c.ReloadConfigString(unsafeRoutes(a+"@100", a+"@200", b)))
assert.Equal(t, uint32(200), rt.metric(t, a))
assert.NotContains(t, rt.log.String(), "Updated route metric")
assert.NotContains(t, rt.log.String(), "Failed to add route")
@@ -142,7 +134,7 @@ func TestWinTun_DroppingAnUninstalledDuplicateKeepsTheRoute(t *testing.T) {
rt := newRouteTest(t, unsafeRoutes(a+"@100", a+"@200"))
require.NoError(t, rt.tun.Activate())
rt.reload(t, unsafeRoutes(a+"@200"))
require.NoError(t, rt.c.ReloadConfigString(unsafeRoutes(a+"@200")))
assert.NotContains(t, rt.log.String(), "Removed route")
assert.Equal(t, uint32(200), rt.metric(t, a))
})
@@ -163,24 +155,12 @@ func TestWinTun_ReloadMatchesAFreshStart(t *testing.T) {
rt := newRouteTest(t, unsafeRoutes(a+"@100"))
require.NoError(t, rt.tun.Activate())
rt.reload(t, routes)
require.NoError(t, rt.c.ReloadConfigString(routes))
_, err = rt.luid.Route(cidr, unspecifiedNextHop(cidr))
assert.Error(t, err, "the reload left it installed")
})
}
// Removing a route that is already gone is what we wanted, not a failure
func TestWinTun_ReloadRouteAlreadyGone(t *testing.T) {
forFamilies(t, func(t *testing.T, a, b string) {
rt := newRouteTest(t, unsafeRoutes(a, b))
require.NoError(t, rt.tun.Activate())
p := netip.MustParsePrefix(b)
require.NoError(t, rt.luid.DeleteRoute(p, unspecifiedNextHop(p)))
rt.reload(t, unsafeRoutes(a))
})
}
// forFamilies runs f with two unsafe_routes CIDRs of each address family, the tun's own network stays IPv4
func forFamilies(t *testing.T, f func(t *testing.T, a, b string)) {
t.Run("ipv4", func(t *testing.T) { f(t, "10.251.1.0/24", "10.251.2.0/24") })
+9 -30
View File
@@ -50,19 +50,15 @@ type CertState struct {
myVpnBroadcastAddrsTable *bart.Lite
}
// errCertNeedsRestart marks a new cert that is refused on reload because only a restart can apply it
var errCertNeedsRestart = errors.New("a restart is required to use the new cert")
func NewPKIFromConfig(l *slog.Logger, c *config.C) (*PKI, error) {
pki := &PKI{l: l}
err := pki.reload(c, true, nil)
err := pki.reload(c, true)
if err != nil {
return nil, err
}
c.RegisterReloadCallback(func(c *config.C, r *config.Reload) {
r.RestartRequiredIfChanged("cipher")
rErr := pki.reload(c, false, r)
c.RegisterReloadCallback(func(c *config.C) {
rErr := pki.reload(c, false)
if rErr != nil {
util.LogWithContextIfNeeded("Failed to reload PKI from config", rErr, l)
}
@@ -79,29 +75,13 @@ func (p *PKI) getCertState() *CertState {
return p.cs.Load()
}
// r is nil on the initial load
func (p *PKI) reload(c *config.C, initial bool, r *config.Reload) error {
before := p.cs.Load()
func (p *PKI) reload(c *config.C, initial bool) error {
err := p.reloadCerts(c, initial)
if err != nil {
if initial {
return err
}
err.Log(p.l)
if errors.Is(err, errCertNeedsRestart) {
r.RestartRequired()
} else {
r.Failed()
}
} else if !initial {
// A cert for a version we weren't running is accepted without comparing its networks to ours. The tun and the
// lighthouse keep the networks they started with
after := p.cs.Load()
if !slices.Equal(before.myVpnNetworks, after.myVpnNetworks) {
p.l.Warn("The new cert changes our networks, a restart is required to use them",
"old_networks", before.myVpnNetworks, "new_networks", after.myVpnNetworks)
r.RestartRequired()
}
}
err = p.reloadCAPool(c)
@@ -110,7 +90,6 @@ func (p *PKI) reload(c *config.C, initial bool, r *config.Reload) error {
return err
}
err.Log(p.l)
r.Failed()
}
return nil
@@ -153,7 +132,7 @@ func (p *PKI) reloadCerts(c *config.C, initial bool) *util.ContextualError {
return util.NewContextualError(
"Networks in new cert was different from old",
m{"new_networks": newState.v1Cert.Networks(), "old_networks": currentState.v1Cert.Networks(), "cert_version": cert.Version1},
errCertNeedsRestart,
nil,
)
}
@@ -161,7 +140,7 @@ func (p *PKI) reloadCerts(c *config.C, initial bool) *util.ContextualError {
return util.NewContextualError(
"Curve in new v1 cert was different from old",
m{"new_curve": newState.v1Cert.Curve(), "old_curve": currentState.v1Cert.Curve(), "cert_version": cert.Version1},
errCertNeedsRestart,
nil,
)
}
}
@@ -176,7 +155,7 @@ func (p *PKI) reloadCerts(c *config.C, initial bool) *util.ContextualError {
return util.NewContextualError(
"Networks in new cert was different from old",
m{"new_networks": newState.v2Cert.Networks(), "old_networks": currentState.v2Cert.Networks(), "cert_version": cert.Version2},
errCertNeedsRestart,
nil,
)
}
@@ -184,7 +163,7 @@ func (p *PKI) reloadCerts(c *config.C, initial bool) *util.ContextualError {
return util.NewContextualError(
"Curve in new cert was different from old",
m{"new_curve": newState.v2Cert.Curve(), "old_curve": currentState.v2Cert.Curve(), "cert_version": cert.Version2},
errCertNeedsRestart,
nil,
)
}
}
@@ -199,7 +178,7 @@ func (p *PKI) reloadCerts(c *config.C, initial bool) *util.ContextualError {
return util.NewContextualError(
"Removing a V2 cert is not permitted unless it has identical networks to the new V1 cert",
m{"new_v1_networks": newState.v1Cert.Networks(), "old_v2_networks": currentState.v2Cert.Networks()},
errCertNeedsRestart,
nil,
)
}
}
-86
View File
@@ -1,16 +1,10 @@
package nebula
import (
"fmt"
"net/netip"
"strings"
"testing"
"time"
"github.com/slackhq/nebula/cert"
"github.com/slackhq/nebula/cert_test"
"github.com/slackhq/nebula/config"
"github.com/slackhq/nebula/test"
"github.com/stretchr/testify/assert"
"github.com/stretchr/testify/require"
)
@@ -98,83 +92,3 @@ func TestNewCipherSuiteUnsupportedCurve(t *testing.T) {
assert.True(t, strings.HasPrefix(err.Error(), "unsupported curve:"), "got: %v", err)
assert.Nil(t, cs)
}
func TestPKIReloadRestartRequired(t *testing.T) {
ca, _, caKey, caPEM := cert_test.NewTestCaCert(cert.Version2, cert.Curve_CURVE25519, time.Time{}, time.Time{}, nil, nil, nil)
pkiConfig := func(cipher, network string) string {
_, _, keyPEM, certPEM := cert_test.NewTestCert(cert.Version2, cert.Curve_CURVE25519, ca, caKey, "host", time.Time{}, time.Time{}, []netip.Prefix{netip.MustParsePrefix(network)}, nil, nil)
return fmt.Sprintf("cipher: %s\npki:\n ca: %q\n cert: %q\n key: %q\n", cipher, caPEM, certPEM, keyPEM)
}
c := config.NewC(test.NewLogger())
require.NoError(t, c.LoadString(pkiConfig("aes", "10.1.0.2/24")))
p, err := NewPKIFromConfig(test.NewLogger(), c)
require.NoError(t, err)
// A cert that changes our networks is refused and needs a restart, it is not a failure
res := c.ReloadConfigString(pkiConfig("aes", "10.1.0.3/24"))
require.NoError(t, res.Err)
assert.True(t, res.RestartRequired)
assert.False(t, res.Failed)
assert.Equal(t, []netip.Prefix{netip.MustParsePrefix("10.1.0.2/24")}, p.getCertState().myVpnNetworks)
// The cipher only applies at startup
c2 := config.NewC(test.NewLogger())
require.NoError(t, c2.LoadString(pkiConfig("aes", "10.1.0.2/24")))
p, err = NewPKIFromConfig(test.NewLogger(), c2)
require.NoError(t, err)
res = c2.ReloadConfigString(pkiConfig("chachapoly", "10.1.0.2/24"))
require.NoError(t, res.Err)
assert.True(t, res.RestartRequired)
assert.Equal(t, "aes", p.getCertState().cipher)
// A cert that keeps our networks applies without a restart
c3 := config.NewC(test.NewLogger())
require.NoError(t, c3.LoadString(pkiConfig("aes", "10.1.0.2/24")))
_, err = NewPKIFromConfig(test.NewLogger(), c3)
require.NoError(t, err)
res = c3.ReloadConfigString(pkiConfig("aes", "10.1.0.2/24"))
require.NoError(t, res.Err)
assert.False(t, res.RestartRequired)
// A cert that can't be read is a failure, and a restart wouldn't help
res = c3.ReloadConfigString(strings.Replace(pkiConfig("aes", "10.1.0.2/24"), "cert: \"", "cert: \"garbage", 1))
require.NoError(t, res.Err)
assert.True(t, res.Failed)
assert.False(t, res.RestartRequired)
}
// A cert for a version we weren't running is accepted without comparing networks, so switching to one with different
// networks needs a restart. One with the same networks doesn't
func TestPKIReloadNewVersionWithOtherNetworks(t *testing.T) {
ca1, _, ca1Key, ca1PEM := cert_test.NewTestCaCert(cert.Version1, cert.Curve_CURVE25519, time.Time{}, time.Time{}, nil, nil, nil)
ca2, _, ca2Key, ca2PEM := cert_test.NewTestCaCert(cert.Version2, cert.Curve_CURVE25519, time.Time{}, time.Time{}, nil, nil, nil)
caPEM := string(ca1PEM) + string(ca2PEM)
pkiConfig := func(v cert.Version, network string) string {
ca, caKey := ca1, ca1Key
if v == cert.Version2 {
ca, caKey = ca2, ca2Key
}
_, _, keyPEM, certPEM := cert_test.NewTestCert(v, cert.Curve_CURVE25519, ca, caKey, "host", time.Time{}, time.Time{}, []netip.Prefix{netip.MustParsePrefix(network)}, nil, nil)
return fmt.Sprintf("pki:\n ca: %q\n cert: %q\n key: %q\n", caPEM, certPEM, keyPEM)
}
for _, tc := range []struct {
network string
restart bool
}{
{"10.9.0.7/24", true},
{"10.1.0.2/24", false},
} {
c := config.NewC(test.NewLogger())
require.NoError(t, c.LoadString(pkiConfig(cert.Version1, "10.1.0.2/24")))
p, err := NewPKIFromConfig(test.NewLogger(), c)
require.NoError(t, err)
res := c.ReloadConfigString(pkiConfig(cert.Version2, tc.network))
require.NoError(t, res.Err)
assert.False(t, res.Failed, tc.network)
assert.Equal(t, tc.restart, res.RestartRequired, tc.network)
assert.Equal(t, []netip.Prefix{netip.MustParsePrefix(tc.network)}, p.getCertState().myVpnNetworks, "accepted either way, as before")
}
}
+1 -2
View File
@@ -67,8 +67,7 @@ func NewPunchyFromConfig(l *slog.Logger, c *config.C, punchConn udp.Conn) *Punch
}
p.reload(c, true)
c.RegisterReloadCallback(func(c *config.C, r *config.Reload) {
r.RestartRequiredIfChanged("stats.lighthouse_metrics")
c.RegisterReloadCallback(func(c *config.C) {
p.reload(c, false)
})
+6 -16
View File
@@ -76,7 +76,7 @@ punchy:
punch: true
delay: 10m
respond: true
`).Err)
`))
p.reload(c, false)
assert.True(t, p.punch.Load())
assert.Equal(t, newDelay, time.Duration(p.delay.Load()))
@@ -170,7 +170,7 @@ func TestPunchy_LogFormat_ReloadPunch(t *testing.T) {
NewPunchyFromConfig(l, c, nil)
hook.entries = nil
require.NoError(t, c.ReloadConfigString(`punchy: {punch: true}`).Err)
require.NoError(t, c.ReloadConfigString(`punchy: {punch: true}`))
entry := findEntry(t, hook.entries, "punchy.punch changed")
assert.Equal(t, slog.LevelInfo, entry.Level)
@@ -184,7 +184,7 @@ func TestPunchy_LogFormat_ReloadRespond(t *testing.T) {
NewPunchyFromConfig(l, c, nil)
hook.entries = nil
require.NoError(t, c.ReloadConfigString(`punchy: {respond: true}`).Err)
require.NoError(t, c.ReloadConfigString(`punchy: {respond: true}`))
entry := findEntry(t, hook.entries, "punchy.respond changed")
assert.Equal(t, slog.LevelInfo, entry.Level)
@@ -198,7 +198,7 @@ func TestPunchy_LogFormat_ReloadDelay(t *testing.T) {
NewPunchyFromConfig(l, c, nil)
hook.entries = nil
require.NoError(t, c.ReloadConfigString(`punchy: {delay: 10s}`).Err)
require.NoError(t, c.ReloadConfigString(`punchy: {delay: 10s}`))
entry := findEntry(t, hook.entries, "punchy.delay changed")
assert.Equal(t, slog.LevelInfo, entry.Level)
@@ -212,7 +212,7 @@ func TestPunchy_LogFormat_ReloadTargetAllRemotes(t *testing.T) {
NewPunchyFromConfig(l, c, nil)
hook.entries = nil
require.NoError(t, c.ReloadConfigString(`punchy: {target_all_remotes: true}`).Err)
require.NoError(t, c.ReloadConfigString(`punchy: {target_all_remotes: true}`))
entry := findEntry(t, hook.entries, "punchy.target_all_remotes changed")
assert.Equal(t, slog.LevelInfo, entry.Level)
@@ -226,19 +226,9 @@ func TestPunchy_LogFormat_ReloadRespondDelay(t *testing.T) {
NewPunchyFromConfig(l, c, nil)
hook.entries = nil
require.NoError(t, c.ReloadConfigString(`punchy: {respond_delay: 15s}`).Err)
require.NoError(t, c.ReloadConfigString(`punchy: {respond_delay: 15s}`))
entry := findEntry(t, hook.entries, "punchy.respond_delay changed")
assert.Equal(t, slog.LevelInfo, entry.Level)
assert.Equal(t, map[string]any{"respond_delay": 15 * time.Second}, entry.Attrs)
}
func TestPunchyReloadRestartRequired(t *testing.T) {
c := config.NewC(test.NewLogger())
require.NoError(t, c.LoadString("stats:\n lighthouse_metrics: false\n"))
NewPunchyFromConfig(test.NewLogger(), c, nil)
res := c.ReloadConfigString("stats:\n lighthouse_metrics: true\n")
require.NoError(t, res.Err)
assert.True(t, res.RestartRequired)
}
+1 -2
View File
@@ -28,11 +28,10 @@ func NewRelayManager(ctx context.Context, l *slog.Logger, hostmap *HostMap, c *c
hostmap: hostmap,
}
rm.reload(c, true)
c.RegisterReloadCallback(func(c *config.C, r *config.Reload) {
c.RegisterReloadCallback(func(c *config.C) {
err := rm.reload(c, false)
if err != nil {
rm.l.Error("Failed to reload relay_manager", "error", err)
r.Failed()
}
})
return rm
+10 -25
View File
@@ -1,8 +1,8 @@
package service
import (
"io"
"net"
"sync"
)
type tcpListener struct {
@@ -10,38 +10,23 @@ type tcpListener struct {
s *Service
addr *net.TCPAddr
accept chan net.Conn
// closed is closed instead of accept so a handler blocked handing off a
// connection can't send on a closed channel.
closed chan struct{}
closeOnce sync.Once
}
func (l *tcpListener) Accept() (net.Conn, error) {
// select picks randomly among ready cases, so without this a handler still
// waiting to hand off could win over closed after Close has returned.
select {
case <-l.closed:
return nil, net.ErrClosed
default:
}
select {
case conn := <-l.accept:
return conn, nil
case <-l.closed:
return nil, net.ErrClosed
conn, ok := <-l.accept
if !ok {
return nil, io.EOF
}
return conn, nil
}
func (l *tcpListener) Close() error {
l.closeOnce.Do(func() {
l.s.mu.Lock()
delete(l.s.mu.listeners, l.port)
l.s.mu.Unlock()
l.s.mu.Lock()
defer l.s.mu.Unlock()
delete(l.s.mu.listeners, uint16(l.addr.Port))
close(l.accept)
close(l.closed)
})
return nil
}
+3 -9
View File
@@ -231,7 +231,6 @@ func (s *Service) Listen(network, address string) (net.Listener, error) {
s: s,
addr: addr,
accept: make(chan net.Conn),
closed: make(chan struct{}),
}
s.mu.Lock()
@@ -258,8 +257,9 @@ func (s *Service) tcpHandler(r *tcp.ForwarderRequest) {
endpointID := r.ID()
s.mu.Lock()
defer s.mu.Unlock()
l, ok := s.mu.listeners[endpointID.LocalPort]
s.mu.Unlock()
if !ok {
r.Complete(true)
return
@@ -275,12 +275,6 @@ func (s *Service) tcpHandler(r *tcp.ForwarderRequest) {
r.Complete(false)
ep.SocketOptions().SetKeepAlive(true)
// The handoff blocks until Accept is called, so it must not hold s.mu or a
// listener nobody accepts from stalls every other port.
conn := gonet.NewTCPConn(&wq, ep)
select {
case l.accept <- conn:
case <-l.closed:
_ = conn.Close()
}
l.accept <- conn
}
-139
View File
@@ -188,142 +188,3 @@ func TestService(t *testing.T) {
t.Fatal(err)
}
}
// A listener that is never accepted from must not stall connections to other
// ports, and closing it must not block.
func TestServiceUnacceptedListener(t *testing.T) {
lighthousePort := ephemeralUDPPort(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{},
"lighthouse": m{
"am_lighthouse": true,
},
"listen": m{
"host": "0.0.0.0",
"port": lighthousePort,
},
})
t.Cleanup(func() { _ = a.Close() })
b := newSimpleService(ca, caKey, "b", netip.MustParseAddr("10.0.0.2"), m{
"static_host_map": m{
"10.0.0.1": []string{fmt.Sprintf("127.0.0.1:%d", lighthousePort)},
},
"lighthouse": m{
"hosts": []string{"10.0.0.1"},
"interval": 1,
},
})
t.Cleanup(func() { _ = b.Close() })
idle, err := a.Listen("tcp", ":1234")
if err != nil {
t.Fatal(err)
}
active, err := a.Listen("tcp", ":1235")
if err != nil {
t.Fatal(err)
}
ctx, cancel := context.WithTimeout(context.Background(), 10*time.Second)
defer cancel()
idleConn, err := b.DialContext(ctx, "tcp", "10.0.0.1:1234")
if err != nil {
t.Fatal(err)
}
defer idleConn.Close()
accepted := make(chan error, 1)
go func() {
conn, err := active.Accept()
if err == nil {
err = conn.Close()
}
accepted <- err
}()
ctx, cancel = context.WithTimeout(context.Background(), 5*time.Second)
defer cancel()
activeConn, err := b.DialContext(ctx, "tcp", "10.0.0.1:1235")
if err != nil {
t.Fatalf("dial to accepted port stalled behind idle listener: %v", err)
}
defer activeConn.Close()
select {
case err := <-accepted:
if err != nil {
t.Fatal(err)
}
case <-time.After(5 * time.Second):
t.Fatal("accept stalled behind idle listener")
}
closed := make(chan error, 1)
go func() { closed <- idle.Close() }()
select {
case err := <-closed:
if err != nil {
t.Fatal(err)
}
case <-time.After(5 * time.Second):
t.Fatal("closing idle listener blocked")
}
if _, err := idle.Accept(); !errors.Is(err, net.ErrClosed) {
t.Fatalf("expected net.ErrClosed from closed listener, got %v", err)
}
if err := idle.Close(); err != nil {
t.Fatalf("second close: %v", err)
}
}
// Accept must report net.ErrClosed after Close even while a handler is still
// trying to hand off a connection.
func TestListenerAcceptAfterClose(t *testing.T) {
s := &Service{}
s.mu.listeners = map[uint16]*tcpListener{}
l := &tcpListener{
port: 1234,
s: s,
accept: make(chan net.Conn),
closed: make(chan struct{}),
}
s.mu.listeners[l.port] = l
if err := l.Close(); err != nil {
t.Fatal(err)
}
// Keep a send pending so every Accept has a connection ready alongside closed.
stop := make(chan struct{})
done := make(chan struct{})
go func() {
defer close(done)
for {
server, client := net.Pipe()
_ = client.Close()
select {
case l.accept <- server:
case <-stop:
_ = server.Close()
return
}
}
}()
defer func() {
close(stop)
<-done
}()
for range 100 {
time.Sleep(time.Millisecond)
conn, err := l.Accept()
if conn != nil {
_ = conn.Close()
}
if !errors.Is(err, net.ErrClosed) {
t.Fatalf("Accept after Close: got err %v, want net.ErrClosed", err)
}
}
}
+6 -15
View File
@@ -58,9 +58,7 @@ type sshDeviceInfoFlags struct {
}
func wireSSHReload(l *slog.Logger, ssh *sshd.SSHServer, c *config.C) {
c.RegisterReloadCallback(func(c *config.C, r *config.Reload) {
// sshd.sandbox_dir is read once, when the commands are attached
r.RestartRequiredIfChanged("sshd.sandbox_dir")
c.RegisterReloadCallback(func(c *config.C) {
if c.GetBool("sshd.enabled", false) {
// Restarting the server drops every open session, the one that asked for this reload included. Only restart
// when it isn't running or its config changed. The host key file can change under the same config, new
@@ -73,7 +71,6 @@ func wireSSHReload(l *slog.Logger, ssh *sshd.SSHServer, c *config.C) {
if err != nil {
l.Error("Failed to reconfigure the sshd", "error", err)
ssh.Stop()
r.Failed()
}
return
}
@@ -81,7 +78,6 @@ func wireSSHReload(l *slog.Logger, ssh *sshd.SSHServer, c *config.C) {
if err != nil {
l.Error("Failed to reconfigure the sshd", "error", err)
ssh.Stop()
r.Failed()
}
if sshRun != nil {
go sshRun()
@@ -280,7 +276,7 @@ func attachCommands(l *slog.Logger, c *config.C, ssh *sshd.SSHServer, f *Interfa
Name: "reload",
ShortDescription: "Reloads configuration from disk, same as sending HUP to the process",
Callback: func(fs any, a []string, w sshd.StringWriter) error {
return sshReload(l, c, w)
return sshReload(c, w)
},
})
@@ -1079,13 +1075,8 @@ func sshDeviceInfo(ifce *Interface, fs any, w sshd.StringWriter) error {
}
}
// sshReload reloads the config and writes the result to the client. It reloads even if the client has gone away
func sshReload(l *slog.Logger, c *config.C, w sshd.StringWriter) error {
writeErr := w.WriteLine("Reloading config")
res := c.ReloadConfig()
res.Log(l)
if err := w.WriteLine(res.String()); writeErr == nil {
writeErr = err
}
return writeErr
func sshReload(c *config.C, w sshd.StringWriter) error {
err := w.WriteLine("Reloading config")
c.ReloadConfig()
return err
}
-83
View File
@@ -1,83 +0,0 @@
package nebula
import (
"errors"
"io"
"os"
"path/filepath"
"testing"
"github.com/slackhq/nebula/config"
"github.com/slackhq/nebula/test"
"github.com/stretchr/testify/assert"
"github.com/stretchr/testify/require"
)
type lineWriter struct {
lines []string
err error
}
func (w *lineWriter) WriteLine(s string) error {
w.lines = append(w.lines, s)
return w.err
}
func (w *lineWriter) Write(s string) error { return w.WriteLine(s) }
func (w *lineWriter) WriteBytes(b []byte) error { return w.WriteLine(string(b)) }
func (w *lineWriter) GetWriter() io.Writer { return io.Discard }
func TestSSHReloadReportsTheResult(t *testing.T) {
dir := t.TempDir()
write := func(body string) {
require.NoError(t, os.WriteFile(filepath.Join(dir, "config.yml"), []byte(body), 0o600))
}
write("a: 1\nb: 1")
l := test.NewLogger()
c := config.NewC(l)
require.NoError(t, c.Load(dir))
fail := false
c.RegisterReloadCallback(func(c *config.C, r *config.Reload) {
r.RestartRequiredIfChanged("b")
if fail {
r.Failed()
}
})
reload := func() []string {
w := &lineWriter{}
require.NoError(t, sshReload(l, c, w))
return w.lines
}
write("a: 2\nb: 1")
assert.Equal(t, []string{"Reloading config", "Config reload complete"}, reload())
fail = true
write("a: 3\nb: 1")
assert.Equal(t, []string{"Reloading config", "Config reload completed with errors, review the logs"}, reload())
fail = false
write("a: 3\nb: 2")
assert.Equal(t, []string{"Reloading config", "Config reload was completed but requires a restart to fully apply, keys [b]"}, reload())
write("a: [")
lines := reload()
require.Len(t, lines, 2)
assert.Contains(t, lines[1], "Config reload failed, the new config did not load")
assert.Contains(t, lines[1], dir, "names the config that did not load")
}
// A client that went away still gets the reload it asked for
func TestSSHReloadReloadsWhenTheClientIsGone(t *testing.T) {
dir := t.TempDir()
require.NoError(t, os.WriteFile(filepath.Join(dir, "config.yml"), []byte("a: 1"), 0o600))
l := test.NewLogger()
c := config.NewC(l)
require.NoError(t, c.Load(dir))
reloaded := false
c.RegisterReloadCallback(func(*config.C, *config.Reload) { reloaded = true })
gone := errors.New("client is gone")
require.ErrorIs(t, sshReload(l, c, &lineWriter{err: gone}), gone)
assert.True(t, reloaded)
}
+6 -6
View File
@@ -130,7 +130,7 @@ func TestSSHDReload_UnchangedKeepsSessions(t *testing.T) {
h := newSSHDReloadHarness(t)
client, _ := h.connect(t)
require.NoError(t, h.c.ReloadConfigString(h.config()+"punchy:\n punch: true\n").Err)
require.NoError(t, h.c.ReloadConfigString(h.config()+"punchy:\n punch: true\n"))
assert.False(t, dropped(client, 500*time.Millisecond), "an unchanged sshd was restarted")
}
@@ -139,7 +139,7 @@ func TestSSHDReload_ChangedConfigRestarts(t *testing.T) {
client, _ := h.connect(t)
h.extra = " trusted_cas: []\n"
require.NoError(t, h.c.ReloadConfigString(h.config()).Err)
require.NoError(t, h.c.ReloadConfigString(h.config()))
assert.True(t, dropped(client, 5*time.Second), "a changed sshd config was not applied")
require.Eventually(t, h.server.Running, 5*time.Second, 10*time.Millisecond)
h.connect(t)
@@ -152,7 +152,7 @@ func TestSSHDReload_ChangedHostKeyFile(t *testing.T) {
client, _ := h.connect(t)
want := h.writeHostKey(t, newEd25519Key(t))
require.NoError(t, h.c.ReloadConfigString(h.config()).Err)
require.NoError(t, h.c.ReloadConfigString(h.config()))
_, got := h.connect(t)
assert.Equal(t, want.Marshal(), got.Marshal(), "a new host key was not applied")
assert.False(t, dropped(client, 500*time.Millisecond), "a new host key restarted sshd")
@@ -166,7 +166,7 @@ func TestSSHDReload_HostKeyAlgorithmChange(t *testing.T) {
rsaKey, err := rsa.GenerateKey(rand.Reader, 2048)
require.NoError(t, err)
want := h.writeHostKey(t, rsaKey)
require.NoError(t, h.c.ReloadConfigString(h.config()).Err)
require.NoError(t, h.c.ReloadConfigString(h.config()))
_, got := h.connect(t)
assert.Equal(t, want.Marshal(), got.Marshal())
@@ -180,7 +180,7 @@ func TestSSHDReload_NotRunningRestarts(t *testing.T) {
h.server.Stop()
require.Eventually(t, func() bool { return !h.server.Running() }, 5*time.Second, 10*time.Millisecond)
require.NoError(t, h.c.ReloadConfigString(h.config()).Err)
require.NoError(t, h.c.ReloadConfigString(h.config()))
require.Eventually(t, h.server.Running, 5*time.Second, 10*time.Millisecond)
h.connect(t)
}
@@ -191,6 +191,6 @@ func TestSSHDReload_BadHostKeyFileStops(t *testing.T) {
h.connect(t)
require.NoError(t, os.WriteFile(h.keyFile, []byte("not a key"), 0o600))
assert.True(t, h.c.ReloadConfigString(h.config()).Failed)
require.NoError(t, h.c.ReloadConfigString(h.config()))
require.Eventually(t, func() bool { return !h.server.Running() }, 5*time.Second, 10*time.Millisecond)
}
+1 -2
View File
@@ -99,10 +99,9 @@ func newStatsServerFromConfig(ctx context.Context, l *slog.Logger, c *config.C,
emitStats: emitStats,
}
c.RegisterReloadCallback(func(c *config.C, r *config.Reload) {
c.RegisterReloadCallback(func(c *config.C) {
if err := s.reload(c, false); err != nil {
s.l.Error("Failed to reload stats from config", "error", err)
r.Failed()
}
})
+2 -3
View File
@@ -43,8 +43,7 @@ type Conn interface {
// 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.
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)
ReloadConfig(c *config.C)
SupportsMultipleReaders() bool
Close() error
}
@@ -69,7 +68,7 @@ func (NoopConn) WriteTo(_ []byte, _ netip.AddrPort) error {
func (NoopConn) WriteBatch(bufs [][]byte, _ []netip.AddrPort) (int, error) {
return len(bufs), nil
}
func (NoopConn) ReloadConfig(_ *config.C, _ *config.Reload) {
func (NoopConn) ReloadConfig(_ *config.C) {
return
}
func (NoopConn) Close() error {
-6
View File
@@ -24,12 +24,6 @@ type NetworkChangeMonitor struct {
// NewNetworkChangeMonitor builds a monitor for local network changes. The returned monitor is always usable: Start
// is safe to call unconditionally, it no-ops when disabled or on a platform that does not need it.
func NewNetworkChangeMonitor(ctx context.Context, l *slog.Logger, c *config.C) *NetworkChangeMonitor {
if watchesNetworkChanges {
c.RegisterReloadCallback(func(c *config.C, r *config.Reload) {
r.RestartRequiredIfChanged("listen.rebind_on_network_change")
})
}
return &NetworkChangeMonitor{
l: l,
ctx: ctx,
-3
View File
@@ -13,9 +13,6 @@ import (
"golang.org/x/sys/unix"
)
// watchesNetworkChanges is true where Start does anything
const watchesNetworkChanges = true
const (
// netChangeSettleWindow is how long we keep swallowing routing messages after the first interesting one. A
// single network change is never a single message, it is a burst: the link drops, addresses go away, new ones
-3
View File
@@ -7,9 +7,6 @@ import (
"log/slog"
)
// watchesNetworkChanges is true where Start does anything
const watchesNetworkChanges = false
// watchNetworkChanges is a no-op outside of darwin.
//
// Darwin is the platform that scopes a udp socket to the interface it came up on, so it is the platform whose socket
-16
View File
@@ -37,19 +37,3 @@ func TestNetworkChangeMonitorNilRebindIsANoOp(t *testing.T) {
m := newMonitor(t, context.Background(), "listen:\n rebind_on_network_change: true\n")
m.Start(nil)
}
// The monitor only reads listen.rebind_on_network_change when it is made
func TestNetworkChangeMonitorRestartKey(t *testing.T) {
l := test.NewLogger()
c := config.NewC(l)
require.NoError(t, c.LoadString("listen:\n rebind_on_network_change: true\n"))
NewNetworkChangeMonitor(context.Background(), l, c)
res := c.ReloadConfigString("listen:\n rebind_on_network_change: false\n")
require.NoError(t, res.Err)
if watchesNetworkChanges {
assert.Equal(t, []string{"listen.rebind_on_network_change"}, res.RestartKeys)
} else {
assert.Empty(t, res.RestartKeys, "the setting does nothing here")
}
}
+2 -4
View File
@@ -26,9 +26,7 @@ type bypassConn struct {
session *wfp.Session
}
func (b *bypassConn) ReloadConfig(c *config.C, r *config.Reload) {
// Only read on the first call, at startup
r.RestartRequiredIfChanged("listen.windows_bypass_wdf")
func (b *bypassConn) ReloadConfig(c *config.C) {
b.installOnce.Do(func() {
if !c.GetBool("listen.windows_bypass_wdf", true) {
return
@@ -47,7 +45,7 @@ func (b *bypassConn) ReloadConfig(c *config.C, r *config.Reload) {
"port", addr.Port())
b.session = s
})
b.Conn.ReloadConfig(c, r)
b.Conn.ReloadConfig(c)
}
func (b *bypassConn) Close() error {
+1 -1
View File
@@ -171,7 +171,7 @@ func (u *StdConn) LocalAddr() (netip.AddrPort, error) {
}
}
func (u *StdConn) ReloadConfig(c *config.C, r *config.Reload) {
func (u *StdConn) ReloadConfig(c *config.C) {
// TODO
}
+1 -1
View File
@@ -73,7 +73,7 @@ func (u *GenericConn) LocalAddr() (netip.AddrPort, error) {
}
}
func (u *GenericConn) ReloadConfig(c *config.C, r *config.Reload) {
func (u *GenericConn) ReloadConfig(c *config.C) {
}
+1 -4
View File
@@ -392,7 +392,7 @@ func writeSockaddr(buf []byte, addr netip.AddrPort, isV4 bool) (int, error) {
return unix.SizeofSockaddrInet6, nil
}
func (u *StdConn) ReloadConfig(c *config.C, r *config.Reload) {
func (u *StdConn) ReloadConfig(c *config.C) {
b := c.GetInt("listen.read_buffer", 0)
if b > 0 {
if err := u.SetRecvBuffer(b); err == nil {
@@ -403,7 +403,6 @@ func (u *StdConn) ReloadConfig(c *config.C, r *config.Reload) {
}
} else {
u.l.Error("Failed to set listen.read_buffer", "error", err)
r.Failed()
}
}
@@ -417,7 +416,6 @@ func (u *StdConn) ReloadConfig(c *config.C, r *config.Reload) {
}
} else {
u.l.Error("Failed to set listen.write_buffer", "error", err)
r.Failed()
}
}
@@ -432,7 +430,6 @@ func (u *StdConn) ReloadConfig(c *config.C, r *config.Reload) {
}
} else {
u.l.Error("Failed to set listen.so_mark", "error", err)
r.Failed()
}
}
}
-31
View File
@@ -1,31 +0,0 @@
//go:build linux && !android && !e2e_testing
package udp
import (
"net/netip"
"os"
"testing"
"github.com/slackhq/nebula/config"
"github.com/stretchr/testify/assert"
"github.com/stretchr/testify/require"
)
// A setting the kernel refuses fails the reload
func TestReloadConfig_RefusedSettingFailsTheReload(t *testing.T) {
if os.Geteuid() == 0 {
t.Skip("root may set listen.so_mark, run as an unprivileged user")
}
conn, err := NewListener(testLogger(), Settings{Listen: netip.MustParseAddrPort("127.0.0.1:0"), Batch: 1})
require.NoError(t, err)
t.Cleanup(func() { _ = conn.Close() })
c := config.NewC(testLogger())
require.NoError(t, c.LoadString("listen:\n port: 0\n"))
c.RegisterReloadCallback(conn.ReloadConfig)
res := c.ReloadConfigString("listen:\n port: 0\n so_mark: 42\n")
require.NoError(t, res.Err)
assert.True(t, res.Failed)
}
+1 -1
View File
@@ -351,7 +351,7 @@ func (u *RIOConn) Rebind() error {
return nil
}
func (u *RIOConn) ReloadConfig(*config.C, *config.Reload) {}
func (u *RIOConn) ReloadConfig(*config.C) {}
func (u *RIOConn) Close() error {
if !u.isOpen.CompareAndSwap(true, false) {
+1 -1
View File
@@ -197,7 +197,7 @@ func (u *TesterConn) ListenOut(r EncReader, flush func()) error {
}
}
func (u *TesterConn) ReloadConfig(*config.C, *config.Reload) {}
func (u *TesterConn) ReloadConfig(*config.C) {}
func NewUDPStatsEmitter(_ []Conn) func() {
// No UDP stats for non-linux