mirror of
https://github.com/netbirdio/netbird.git
synced 2026-10-13 08:59:08 +02:00
Recover from tun device read/write panics and restart the client
This commit is contained in:
@@ -1,10 +1,13 @@
|
|||||||
package device
|
package device
|
||||||
|
|
||||||
import (
|
import (
|
||||||
|
"fmt"
|
||||||
"net/netip"
|
"net/netip"
|
||||||
|
"runtime/debug"
|
||||||
"sync"
|
"sync"
|
||||||
"sync/atomic"
|
"sync/atomic"
|
||||||
|
|
||||||
|
log "github.com/sirupsen/logrus"
|
||||||
"golang.zx2c4.com/wireguard/tun"
|
"golang.zx2c4.com/wireguard/tun"
|
||||||
)
|
)
|
||||||
|
|
||||||
@@ -41,10 +44,13 @@ type PacketCapture interface {
|
|||||||
type FilteredDevice struct {
|
type FilteredDevice struct {
|
||||||
tun.Device
|
tun.Device
|
||||||
|
|
||||||
filter PacketFilter
|
filter PacketFilter
|
||||||
capture atomic.Pointer[PacketCapture]
|
capture atomic.Pointer[PacketCapture]
|
||||||
mutex sync.RWMutex
|
// panicHandler is invoked after a panic in the underlying device is
|
||||||
closeOnce sync.Once
|
// recovered in Read or Write.
|
||||||
|
panicHandler atomic.Pointer[func()]
|
||||||
|
mutex sync.RWMutex
|
||||||
|
closeOnce sync.Once
|
||||||
}
|
}
|
||||||
|
|
||||||
// newDeviceFilter constructor function
|
// newDeviceFilter constructor function
|
||||||
@@ -70,7 +76,7 @@ func (d *FilteredDevice) Close() error {
|
|||||||
|
|
||||||
// Read wraps read method with filtering feature
|
// Read wraps read method with filtering feature
|
||||||
func (d *FilteredDevice) Read(bufs [][]byte, sizes []int, offset int) (n int, err error) {
|
func (d *FilteredDevice) Read(bufs [][]byte, sizes []int, offset int) (n int, err error) {
|
||||||
if n, err = d.Device.Read(bufs, sizes, offset); err != nil {
|
if n, err = d.deviceRead(bufs, sizes, offset); err != nil {
|
||||||
return 0, err
|
return 0, err
|
||||||
}
|
}
|
||||||
|
|
||||||
@@ -112,7 +118,7 @@ func (d *FilteredDevice) Write(bufs [][]byte, offset int) (int, error) {
|
|||||||
d.mutex.RUnlock()
|
d.mutex.RUnlock()
|
||||||
|
|
||||||
if filter == nil {
|
if filter == nil {
|
||||||
return d.Device.Write(bufs, offset)
|
return d.deviceWrite(bufs, offset)
|
||||||
}
|
}
|
||||||
|
|
||||||
filteredBufs := make([][]byte, 0, len(bufs))
|
filteredBufs := make([][]byte, 0, len(bufs))
|
||||||
@@ -125,9 +131,44 @@ func (d *FilteredDevice) Write(bufs [][]byte, offset int) (int, error) {
|
|||||||
}
|
}
|
||||||
}
|
}
|
||||||
|
|
||||||
n, err := d.Device.Write(filteredBufs, offset)
|
n, err := d.deviceWrite(filteredBufs, offset)
|
||||||
n += dropped
|
if err != nil {
|
||||||
return n, err
|
return n, err
|
||||||
|
}
|
||||||
|
return n + dropped, nil
|
||||||
|
}
|
||||||
|
|
||||||
|
// deviceRead calls the underlying device Read, recovering from panics in the
|
||||||
|
// wintun read path and converting them into errors.
|
||||||
|
func (d *FilteredDevice) deviceRead(bufs [][]byte, sizes []int, offset int) (n int, err error) {
|
||||||
|
defer d.recoverFromPanic("read", &n, &err)
|
||||||
|
return d.Device.Read(bufs, sizes, offset)
|
||||||
|
}
|
||||||
|
|
||||||
|
// deviceWrite calls the underlying device Write, recovering from panics in the
|
||||||
|
// wintun write path and converting them into errors.
|
||||||
|
func (d *FilteredDevice) deviceWrite(bufs [][]byte, offset int) (n int, err error) {
|
||||||
|
defer d.recoverFromPanic("write", &n, &err)
|
||||||
|
return d.Device.Write(bufs, offset)
|
||||||
|
}
|
||||||
|
|
||||||
|
// recoverFromPanic converts a panic in the underlying device into a regular
|
||||||
|
// error and invokes the registered panic handler. The wintun read path is
|
||||||
|
// known to panic on zero-length packets that third-party filter drivers can
|
||||||
|
// place in the ring.
|
||||||
|
func (d *FilteredDevice) recoverFromPanic(op string, n *int, err *error) {
|
||||||
|
r := recover()
|
||||||
|
if r == nil {
|
||||||
|
return
|
||||||
|
}
|
||||||
|
|
||||||
|
log.Errorf("recovered panic in tun device %s: %v\n%s", op, r, debug.Stack())
|
||||||
|
*n = 0
|
||||||
|
*err = fmt.Errorf("tun device %s panic: %v", op, r)
|
||||||
|
|
||||||
|
if handler := d.panicHandler.Load(); handler != nil {
|
||||||
|
(*handler)()
|
||||||
|
}
|
||||||
}
|
}
|
||||||
|
|
||||||
// SetFilter sets packet filter to device
|
// SetFilter sets packet filter to device
|
||||||
@@ -137,6 +178,17 @@ func (d *FilteredDevice) SetFilter(filter PacketFilter) {
|
|||||||
d.mutex.Unlock()
|
d.mutex.Unlock()
|
||||||
}
|
}
|
||||||
|
|
||||||
|
// SetPanicHandler registers a handler invoked after a recovered panic in Read
|
||||||
|
// or Write. The device is unusable after such a panic; the handler should
|
||||||
|
// trigger recreation of the interface. Pass nil to remove.
|
||||||
|
func (d *FilteredDevice) SetPanicHandler(handler func()) {
|
||||||
|
if handler == nil {
|
||||||
|
d.panicHandler.Store(nil)
|
||||||
|
return
|
||||||
|
}
|
||||||
|
d.panicHandler.Store(&handler)
|
||||||
|
}
|
||||||
|
|
||||||
// SetCapture sets or clears the packet capture sink. Pass nil to disable.
|
// SetCapture sets or clears the packet capture sink. Pass nil to disable.
|
||||||
// Uses atomic store so the hot path (Read/Write) is a single pointer load
|
// Uses atomic store so the hot path (Read/Write) is a single pointer load
|
||||||
// with no locking overhead when capture is off.
|
// with no locking overhead when capture is off.
|
||||||
|
|||||||
@@ -221,3 +221,60 @@ func TestDeviceWrapperRead(t *testing.T) {
|
|||||||
}
|
}
|
||||||
})
|
})
|
||||||
}
|
}
|
||||||
|
|
||||||
|
func TestDeviceWrapperReadPanic(t *testing.T) {
|
||||||
|
ctrl := gomock.NewController(t)
|
||||||
|
defer ctrl.Finish()
|
||||||
|
|
||||||
|
tun := mocks.NewMockDevice(ctrl)
|
||||||
|
tun.EXPECT().Read(gomock.Any(), gomock.Any(), gomock.Any()).
|
||||||
|
DoAndReturn(func(bufs [][]byte, sizes []int, offset int) (int, error) {
|
||||||
|
// Reproduce the wintun zero-length packet panic (index out of range).
|
||||||
|
packet := make([]byte, 0)
|
||||||
|
return int(packet[0]), nil
|
||||||
|
})
|
||||||
|
|
||||||
|
wrapped := newDeviceFilter(tun)
|
||||||
|
|
||||||
|
handlerCalled := false
|
||||||
|
wrapped.SetPanicHandler(func() { handlerCalled = true })
|
||||||
|
|
||||||
|
n, err := wrapped.Read([][]byte{{}}, []int{0}, 0)
|
||||||
|
if err == nil {
|
||||||
|
t.Errorf("expected error from recovered panic, got nil")
|
||||||
|
}
|
||||||
|
if n != 0 {
|
||||||
|
t.Errorf("expected n=0, got %d", n)
|
||||||
|
}
|
||||||
|
if !handlerCalled {
|
||||||
|
t.Errorf("expected panic handler to be called")
|
||||||
|
}
|
||||||
|
}
|
||||||
|
|
||||||
|
func TestDeviceWrapperWritePanic(t *testing.T) {
|
||||||
|
ctrl := gomock.NewController(t)
|
||||||
|
defer ctrl.Finish()
|
||||||
|
|
||||||
|
tun := mocks.NewMockDevice(ctrl)
|
||||||
|
tun.EXPECT().Write(gomock.Any(), gomock.Any()).
|
||||||
|
DoAndReturn(func(bufs [][]byte, offset int) (int, error) {
|
||||||
|
packet := make([]byte, 0)
|
||||||
|
return int(packet[0]), nil
|
||||||
|
})
|
||||||
|
|
||||||
|
wrapped := newDeviceFilter(tun)
|
||||||
|
|
||||||
|
handlerCalled := false
|
||||||
|
wrapped.SetPanicHandler(func() { handlerCalled = true })
|
||||||
|
|
||||||
|
n, err := wrapped.Write([][]byte{{0x45, 0x00}}, 0)
|
||||||
|
if err == nil {
|
||||||
|
t.Errorf("expected error from recovered panic, got nil")
|
||||||
|
}
|
||||||
|
if n != 0 {
|
||||||
|
t.Errorf("expected n=0, got %d", n)
|
||||||
|
}
|
||||||
|
if !handlerCalled {
|
||||||
|
t.Errorf("expected panic handler to be called")
|
||||||
|
}
|
||||||
|
}
|
||||||
|
|||||||
@@ -240,7 +240,7 @@ type Engine struct {
|
|||||||
syncStore syncstore.Store
|
syncStore syncstore.Store
|
||||||
syncStoreDir string
|
syncStoreDir string
|
||||||
|
|
||||||
flowManager nftypes.FlowManager
|
flowManager nftypes.FlowManager
|
||||||
|
|
||||||
// auto-update
|
// auto-update
|
||||||
updateManager *updater.Manager
|
updateManager *updater.Manager
|
||||||
@@ -531,6 +531,10 @@ func (e *Engine) Start(netbirdConfig *mgmProto.NetbirdConfig, mgmtURL *url.URL)
|
|||||||
return fmt.Errorf("create wg interface: %w", err)
|
return fmt.Errorf("create wg interface: %w", err)
|
||||||
}
|
}
|
||||||
|
|
||||||
|
if filteredDevice := e.wgInterface.GetDevice(); filteredDevice != nil {
|
||||||
|
filteredDevice.SetPanicHandler(e.triggerClientRestart)
|
||||||
|
}
|
||||||
|
|
||||||
if err := e.createFirewall(); err != nil {
|
if err := e.createFirewall(); err != nil {
|
||||||
e.close()
|
e.close()
|
||||||
return err
|
return err
|
||||||
|
|||||||
Reference in New Issue
Block a user