Merge branch 'main' into fix/pkce-flow-session-extend

management.pb.go conflicted because both sides added to it: main gave
ResourceCompact an id and GroupCompact a resources list, this branch added
PKCEAuthorizationFlowRequest.SessionExtend. The .proto merged cleanly, so
the generated file was regenerated from it rather than resolved by hand.
This commit is contained in:
Zoltán Papp
2026-08-27 11:57:56 +02:00
353 changed files with 25426 additions and 3863 deletions
+87
View File
@@ -8,7 +8,9 @@ import (
"github.com/stretchr/testify/assert"
"github.com/stretchr/testify/require"
"github.com/netbirdio/netbird/client/iface/wgaddr"
nbdns "github.com/netbirdio/netbird/dns"
mgmProto "github.com/netbirdio/netbird/shared/management/proto"
)
func TestCreatePTRRecord_IPv4(t *testing.T) {
@@ -136,3 +138,88 @@ func TestAddReverseZone_IPv6(t *testing.T) {
assert.Len(t, reverseZone.Records, 1)
assert.Equal(t, int(dns.TypePTR), reverseZone.Records[0].Type)
}
// TestToDNSConfig_ZoneFlagsPreserved pins the per-zone NonAuthoritative flag
// through the legacy DNSConfig path. A non-authoritative zone is match-only:
// the local resolver falls through to the upstream for an in-zone name it does
// not define. The built-in peer zone is the authoritative one and must stay
// that way, so the flag has to travel per zone rather than be derived.
func TestToDNSConfig_ZoneFlagsPreserved(t *testing.T) {
config := toDNSConfig(&mgmProto.DNSConfig{
ServiceEnable: true,
CustomZones: []*mgmProto.CustomZone{
{
Domain: "netbird.cloud.",
Records: []*mgmProto.SimpleRecord{
{Name: "peer1.netbird.cloud.", Type: int64(dns.TypeA), Class: nbdns.DefaultClass, TTL: 300, RData: "100.64.0.1"},
},
},
{
Domain: "corp.internal.",
NonAuthoritative: true,
SearchDomainDisabled: true,
Records: []*mgmProto.SimpleRecord{
{Name: "db.corp.internal.", Type: int64(dns.TypeA), Class: nbdns.DefaultClass, TTL: 300, RData: "10.10.0.5"},
},
},
},
}, wgaddr.Address{
IP: netip.MustParseAddr("100.64.0.1"),
Network: netip.MustParsePrefix("100.64.0.0/16"),
})
zones := make(map[string]nbdns.CustomZone, len(config.CustomZones))
for _, zone := range config.CustomZones {
zones[zone.Domain] = zone
}
peerZone, ok := zones["netbird.cloud."]
require.True(t, ok, "peer zone must survive")
assert.False(t, peerZone.NonAuthoritative, "the built-in peer zone owns the account domain and stays authoritative")
accountZone, ok := zones["corp.internal."]
require.True(t, ok, "account zone must survive")
assert.True(t, accountZone.NonAuthoritative, "an account zone stays match-only, else undefined in-zone names get black-holed")
assert.True(t, accountZone.SearchDomainDisabled)
}
// TestToDNSConfig_SingleZoneForcedAuthoritative pins the compatibility clause
// in toDNSConfig: a config carrying exactly one zone is treated as
// authoritative no matter what the server said, because servers that predate
// the NonAuthoritative field send only the peer FQDN zone.
//
// The clause can only ever downgrade an explicit true to false, so a server
// that legitimately sends a single non-authoritative zone — an account whose
// only zone is a custom one, with no peer records to build the built-in zone
// from — gets that zone's whole apex black-holed on the client. Real accounts
// always carry the peer zone alongside, which is why this is latent. Narrowing
// it needs a way to tell "unset" from "false" on the wire, or the account
// domain passed down here; until then this test states the contract so a
// change to it is deliberate.
func TestToDNSConfig_SingleZoneForcedAuthoritative(t *testing.T) {
config := toDNSConfig(&mgmProto.DNSConfig{
ServiceEnable: true,
CustomZones: []*mgmProto.CustomZone{
{
Domain: "corp.internal.",
NonAuthoritative: true,
Records: []*mgmProto.SimpleRecord{
{Name: "db.corp.internal.", Type: int64(dns.TypeA), Class: nbdns.DefaultClass, TTL: 300, RData: "10.10.0.5"},
},
},
},
}, wgaddr.Address{
IP: netip.MustParseAddr("100.64.0.1"),
Network: netip.MustParsePrefix("100.64.0.0/16"),
})
require.NotEmpty(t, config.CustomZones)
assert.Equal(t, "corp.internal.", config.CustomZones[0].Domain)
assert.False(t, config.CustomZones[0].NonAuthoritative,
"a lone zone is forced authoritative for pre-NonAuthoritative servers")
// The reverse zone the config gains afterwards must not feed back into the
// decision: the compat gate counts the zones the server sent.
require.Len(t, config.CustomZones, 2, "a reverse zone is appended for the overlay prefix")
assert.Equal(t, "64.100.in-addr.arpa.", config.CustomZones[1].Domain)
}
+1 -1
View File
@@ -519,7 +519,7 @@ func startManagement(t *testing.T, dataDir, testFile string) (*grpc.Server, stri
updateManager := update_channel.NewPeersUpdateManager(metrics)
requestBuffer := server.NewAccountRequestBuffer(context.Background(), store)
networkMapController := controller.NewController(context.Background(), store, metrics, updateManager, requestBuffer, server.MockIntegratedValidator{}, settingsMockManager, "netbird.selfhosted", port_forwarding.NewControllerMock(), manager.NewEphemeralManager(store, peersManager), config)
networkMapController := controller.NewController(context.Background(), store, metrics, updateManager, requestBuffer, server.MockIntegratedValidator{}, settingsMockManager, "netbird.selfhosted", port_forwarding.NewControllerMock(), manager.NewEphemeralManager(store, peersManager), config, nil)
accountManager, err := server.BuildManager(context.Background(), config, store, networkMapController, jobManager, nil, "", eventStore, nil, false, ia, metrics, port_forwarding.NewControllerMock(), settingsMockManager, permissionsManager, false, cacheStore)
if err != nil {
return nil, "", err
+4
View File
@@ -109,6 +109,10 @@
// - Does NOT remove result.json (cleaned by ResultHandler after read)
// - Does NOT remove msi.log (kept for debugging)
//
// On Windows the updater copy is often still locked when the daemon it restarted
// runs cleanup, so removing it is retried briefly and otherwise left in place for
// the next update to overwrite rather than reported as a failure.
//
// # Dry-Run Mode
//
// Dry-run mode allows testing the update process without actually installing:
@@ -0,0 +1,67 @@
package installer
import (
"os"
"path/filepath"
"testing"
"time"
"golang.org/x/sys/windows"
)
// lockFile opens path without FILE_SHARE_DELETE, so os.Remove fails the way it does
// while the updater process still holds its own image.
func lockFile(t *testing.T, path string) windows.Handle {
t.Helper()
p, err := windows.UTF16PtrFromString(path)
if err != nil {
t.Fatalf("convert path: %v", err)
}
handle, err := windows.CreateFile(p, windows.GENERIC_READ, windows.FILE_SHARE_READ, nil, windows.OPEN_EXISTING, windows.FILE_ATTRIBUTE_NORMAL, 0)
if err != nil {
t.Fatalf("lock %s: %v", path, err)
}
return handle
}
// releaseAfter closes the handle once the delay has passed, standing in for the
// updater process finally exiting.
func releaseAfter(t *testing.T, handle windows.Handle, delay time.Duration) {
t.Helper()
released := make(chan struct{})
t.Cleanup(func() { <-released })
go func() {
defer close(released)
time.Sleep(delay)
if err := windows.CloseHandle(handle); err != nil {
t.Errorf("close handle: %v", err)
}
}()
}
// TestCleanUpInstallerFilesLockedUpdater covers the post-update cleanup race: the
// daemon cleans up at startup while the updater that restarted it is still exiting,
// so the updater image is locked and Windows refuses the delete. Cleanup must wait
// the lock out instead of reporting a failure and leaving the binary behind.
func TestCleanUpInstallerFilesLockedUpdater(t *testing.T) {
tempDir := t.TempDir()
path := filepath.Join(tempDir, updaterBinary)
if err := os.WriteFile(path, []byte("x"), 0o600); err != nil {
t.Fatalf("write updater: %v", err)
}
releaseAfter(t, lockFile(t, path), 300*time.Millisecond)
u := NewWithDir(tempDir)
if err := u.CleanUpInstallerFiles(); err != nil {
t.Fatalf("cleanup must tolerate a still-locked updater: %v", err)
}
if _, err := os.Stat(path); !os.IsNotExist(err) {
t.Errorf("updater binary still present (stat err: %v)", err)
}
}
@@ -152,8 +152,8 @@ func (u *Installer) CleanUpInstallerFiles() error {
var merr *multierror.Error
if err := os.Remove(filepath.Join(u.tempDir, updaterBinary)); err != nil && !os.IsNotExist(err) {
merr = multierror.Append(merr, fmt.Errorf("failed to remove updater binary: %w", err))
if err := removeUpdaterBinary(filepath.Join(u.tempDir, updaterBinary)); err != nil {
merr = multierror.Append(merr, fmt.Errorf("remove updater binary: %w", err))
}
entries, err := os.ReadDir(u.tempDir)
@@ -167,10 +167,16 @@ func (u *Installer) CleanUpInstallerFiles() error {
}
name := entry.Name()
// The updater copy is handled above; on Windows its name also matches the
// extension sweep, which would report the same file twice.
if strings.EqualFold(name, updaterBinary) {
continue
}
for _, ext := range binaryExtensions {
if strings.HasSuffix(strings.ToLower(name), strings.ToLower(ext)) {
if err := os.Remove(filepath.Join(u.tempDir, name)); err != nil {
merr = multierror.Append(merr, fmt.Errorf("failed to remove %s: %w", name, err))
merr = multierror.Append(merr, fmt.Errorf("remove %s: %w", name, err))
}
break
}
@@ -0,0 +1,52 @@
//go:build windows || darwin
package installer
import (
"os"
"path/filepath"
"testing"
)
// TestCleanUpInstallerFiles checks that cleanup removes the updater copy and the
// downloaded installer while leaving the logs and the result file for the daemon.
func TestCleanUpInstallerFiles(t *testing.T) {
tempDir := t.TempDir()
installers := make([]string, 0, len(binaryExtensions))
for _, ext := range binaryExtensions {
installers = append(installers, "netbird_installer."+ext)
}
kept := []string{"installer.log", "result.json"}
for _, name := range append(append([]string{updaterBinary}, installers...), kept...) {
if err := os.WriteFile(filepath.Join(tempDir, name), []byte("x"), 0o600); err != nil {
t.Fatalf("write %s: %v", name, err)
}
}
u := NewWithDir(tempDir)
if err := u.CleanUpInstallerFiles(); err != nil {
t.Fatalf("CleanUpInstallerFiles: %v", err)
}
for _, name := range append([]string{updaterBinary}, installers...) {
if _, err := os.Stat(filepath.Join(tempDir, name)); !os.IsNotExist(err) {
t.Errorf("%s was not removed (stat err: %v)", name, err)
}
}
for _, name := range kept {
if _, err := os.Stat(filepath.Join(tempDir, name)); err != nil {
t.Errorf("%s should have been kept: %v", name, err)
}
}
}
func TestCleanUpInstallerFilesMissingTempDir(t *testing.T) {
u := NewWithDir(filepath.Join(t.TempDir(), "does-not-exist"))
if err := u.CleanUpInstallerFiles(); err != nil {
t.Errorf("a missing temp dir is not a cleanup failure, got: %v", err)
}
}
@@ -0,0 +1,12 @@
package installer
import "os"
// removeUpdaterBinary deletes the updater copy left in the temp dir. On darwin a
// running binary can be unlinked, so no retry is needed.
func removeUpdaterBinary(path string) error {
if err := os.Remove(path); err != nil && !os.IsNotExist(err) {
return err
}
return nil
}
@@ -0,0 +1,45 @@
package installer
import (
"errors"
"os"
"time"
log "github.com/sirupsen/logrus"
"golang.org/x/sys/windows"
)
const (
// The updater is the process that restarted the daemon, so when the daemon
// cleans up at startup the updater is often still exiting and Windows refuses
// to delete its locked image. These bound how long cleanup waits for it.
updaterRemoveAttempts = 5
updaterRemoveDelay = 200 * time.Millisecond
)
// removeUpdaterBinary deletes the updater copy left in the temp dir, retrying
// while the still-exiting updater process holds its image. A binary that stays
// locked for the whole window is left in place and reported at info level: the
// next update overwrites it, so it is not worth failing cleanup over.
func removeUpdaterBinary(path string) error {
for attempt := 0; attempt < updaterRemoveAttempts; attempt++ {
if attempt > 0 {
time.Sleep(updaterRemoveDelay)
}
err := os.Remove(path)
if err == nil || os.IsNotExist(err) {
return nil
}
if !isFileLocked(err) {
return err
}
}
log.Infof("updater binary %s is still locked, leaving it for the next update to overwrite", path)
return nil
}
func isFileLocked(err error) bool {
return errors.Is(err, windows.ERROR_ACCESS_DENIED) || errors.Is(err, windows.ERROR_SHARING_VIOLATION)
}
@@ -0,0 +1,59 @@
package installer
import (
"os"
"path/filepath"
"testing"
"time"
"golang.org/x/sys/windows"
)
func TestRemoveUpdaterBinaryRetriesWhileLocked(t *testing.T) {
path := filepath.Join(t.TempDir(), updaterBinary)
if err := os.WriteFile(path, []byte("x"), 0o600); err != nil {
t.Fatalf("write updater: %v", err)
}
releaseAfter(t, lockFile(t, path), updaterRemoveDelay+50*time.Millisecond)
if err := removeUpdaterBinary(path); err != nil {
t.Fatalf("removeUpdaterBinary: %v", err)
}
if _, err := os.Stat(path); !os.IsNotExist(err) {
t.Errorf("updater binary still present (stat err: %v)", err)
}
}
// TestRemoveUpdaterBinaryStaysLocked covers an updater that never releases its
// image within the retry window. Cleanup gives up quietly and leaves the file
// behind rather than reporting a failure.
func TestRemoveUpdaterBinaryStaysLocked(t *testing.T) {
path := filepath.Join(t.TempDir(), updaterBinary)
if err := os.WriteFile(path, []byte("x"), 0o600); err != nil {
t.Fatalf("write updater: %v", err)
}
handle := lockFile(t, path)
t.Cleanup(func() {
if err := windows.CloseHandle(handle); err != nil {
t.Errorf("close handle: %v", err)
}
})
if err := removeUpdaterBinary(path); err != nil {
t.Fatalf("a permanently locked updater is not a cleanup failure, got: %v", err)
}
if _, err := os.Stat(path); err != nil {
t.Errorf("locked updater binary should be left in place, stat: %v", err)
}
}
func TestRemoveUpdaterBinaryMissingFile(t *testing.T) {
path := filepath.Join(t.TempDir(), updaterBinary)
if err := removeUpdaterBinary(path); err != nil {
t.Errorf("a missing updater binary is not a failure, got: %v", err)
}
}