nebula/connection_manager.go

316 lines
7.9 KiB
Go
Raw Normal View History

2019-11-19 18:00:20 +01:00
package nebula
import (
"context"
2019-11-19 18:00:20 +01:00
"sync"
"time"
2019-12-12 17:34:17 +01:00
"github.com/sirupsen/logrus"
"github.com/slackhq/nebula/header"
"github.com/slackhq/nebula/iputil"
2019-11-19 18:00:20 +01:00
)
// TODO: incount and outcount are intended as a shortcut to locking the mutexes for every single packet
// and something like every 10 packets we could lock, send 10, then unlock for a moment
type connectionManager struct {
hostMap *HostMap
in map[iputil.VpnIp]struct{}
2019-11-19 18:00:20 +01:00
inLock *sync.RWMutex
inCount int
out map[iputil.VpnIp]struct{}
2019-11-19 18:00:20 +01:00
outLock *sync.RWMutex
outCount int
TrafficTimer *SystemTimerWheel
intf *Interface
pendingDeletion map[iputil.VpnIp]int
2019-11-19 18:00:20 +01:00
pendingDeletionLock *sync.RWMutex
pendingDeletionTimer *SystemTimerWheel
checkInterval int
pendingDeletionInterval int
2021-03-26 15:46:30 +01:00
l *logrus.Logger
2019-11-19 18:00:20 +01:00
// I wanted to call one matLock
}
func newConnectionManager(ctx context.Context, l *logrus.Logger, intf *Interface, checkInterval, pendingDeletionInterval int) *connectionManager {
2019-11-19 18:00:20 +01:00
nc := &connectionManager{
hostMap: intf.hostMap,
in: make(map[iputil.VpnIp]struct{}),
2019-11-19 18:00:20 +01:00
inLock: &sync.RWMutex{},
inCount: 0,
out: make(map[iputil.VpnIp]struct{}),
2019-11-19 18:00:20 +01:00
outLock: &sync.RWMutex{},
outCount: 0,
TrafficTimer: NewSystemTimerWheel(time.Millisecond*500, time.Second*60),
intf: intf,
pendingDeletion: make(map[iputil.VpnIp]int),
2019-11-19 18:00:20 +01:00
pendingDeletionLock: &sync.RWMutex{},
pendingDeletionTimer: NewSystemTimerWheel(time.Millisecond*500, time.Second*60),
checkInterval: checkInterval,
pendingDeletionInterval: pendingDeletionInterval,
2021-03-26 15:46:30 +01:00
l: l,
2019-11-19 18:00:20 +01:00
}
nc.Start(ctx)
2019-11-19 18:00:20 +01:00
return nc
}
func (n *connectionManager) In(ip iputil.VpnIp) {
2019-11-19 18:00:20 +01:00
n.inLock.RLock()
// If this already exists, return
if _, ok := n.in[ip]; ok {
n.inLock.RUnlock()
return
}
n.inLock.RUnlock()
n.inLock.Lock()
n.in[ip] = struct{}{}
n.inLock.Unlock()
}
func (n *connectionManager) Out(ip iputil.VpnIp) {
2019-11-19 18:00:20 +01:00
n.outLock.RLock()
// If this already exists, return
if _, ok := n.out[ip]; ok {
n.outLock.RUnlock()
return
}
n.outLock.RUnlock()
n.outLock.Lock()
// double check since we dropped the lock temporarily
if _, ok := n.out[ip]; ok {
n.outLock.Unlock()
return
}
n.out[ip] = struct{}{}
n.AddTrafficWatch(ip, n.checkInterval)
n.outLock.Unlock()
}
func (n *connectionManager) CheckIn(vpnIp iputil.VpnIp) bool {
2019-11-19 18:00:20 +01:00
n.inLock.RLock()
if _, ok := n.in[vpnIp]; ok {
2019-11-19 18:00:20 +01:00
n.inLock.RUnlock()
return true
}
n.inLock.RUnlock()
return false
}
func (n *connectionManager) ClearIP(ip iputil.VpnIp) {
2019-11-19 18:00:20 +01:00
n.inLock.Lock()
n.outLock.Lock()
delete(n.in, ip)
delete(n.out, ip)
n.inLock.Unlock()
n.outLock.Unlock()
}
func (n *connectionManager) ClearPendingDeletion(ip iputil.VpnIp) {
2019-11-19 18:00:20 +01:00
n.pendingDeletionLock.Lock()
delete(n.pendingDeletion, ip)
n.pendingDeletionLock.Unlock()
}
func (n *connectionManager) AddPendingDeletion(ip iputil.VpnIp) {
2019-11-19 18:00:20 +01:00
n.pendingDeletionLock.Lock()
if _, ok := n.pendingDeletion[ip]; ok {
n.pendingDeletion[ip] += 1
} else {
n.pendingDeletion[ip] = 0
}
n.pendingDeletionTimer.Add(ip, time.Second*time.Duration(n.pendingDeletionInterval))
n.pendingDeletionLock.Unlock()
}
func (n *connectionManager) checkPendingDeletion(ip iputil.VpnIp) bool {
2019-11-19 18:00:20 +01:00
n.pendingDeletionLock.RLock()
if _, ok := n.pendingDeletion[ip]; ok {
n.pendingDeletionLock.RUnlock()
return true
}
n.pendingDeletionLock.RUnlock()
return false
}
func (n *connectionManager) AddTrafficWatch(vpnIp iputil.VpnIp, seconds int) {
n.TrafficTimer.Add(vpnIp, time.Second*time.Duration(seconds))
2019-11-19 18:00:20 +01:00
}
func (n *connectionManager) Start(ctx context.Context) {
go n.Run(ctx)
2019-11-19 18:00:20 +01:00
}
func (n *connectionManager) Run(ctx context.Context) {
clockSource := time.NewTicker(500 * time.Millisecond)
defer clockSource.Stop()
p := []byte("")
nb := make([]byte, 12, 12)
out := make([]byte, mtu)
2019-11-19 18:00:20 +01:00
for {
select {
case <-ctx.Done():
return
case now := <-clockSource.C:
n.HandleMonitorTick(now, p, nb, out)
n.HandleDeletionTick(now)
}
2019-11-19 18:00:20 +01:00
}
}
func (n *connectionManager) HandleMonitorTick(now time.Time, p, nb, out []byte) {
2019-11-19 18:00:20 +01:00
n.TrafficTimer.advance(now)
for {
ep := n.TrafficTimer.Purge()
if ep == nil {
break
}
vpnIp := ep.(iputil.VpnIp)
2019-11-19 18:00:20 +01:00
// Check for traffic coming back in from this host.
traf := n.CheckIn(vpnIp)
2019-11-19 18:00:20 +01:00
hostinfo, err := n.hostMap.QueryVpnIp(vpnIp)
if err != nil {
n.l.Debugf("Not found in hostmap: %s", vpnIp)
if !n.intf.disconnectInvalid {
n.ClearIP(vpnIp)
n.ClearPendingDeletion(vpnIp)
continue
}
}
if n.handleInvalidCertificate(now, vpnIp, hostinfo) {
continue
}
// If we saw an incoming packets from this ip and peer's certificate is not
// expired, just ignore.
2019-11-19 18:00:20 +01:00
if traf {
2021-03-26 15:46:30 +01:00
if n.l.Level >= logrus.DebugLevel {
n.l.WithField("vpnIp", vpnIp).
2019-11-19 18:00:20 +01:00
WithField("tunnelCheck", m{"state": "alive", "method": "passive"}).
Debug("Tunnel status")
}
n.ClearIP(vpnIp)
n.ClearPendingDeletion(vpnIp)
2019-11-19 18:00:20 +01:00
continue
}
2021-03-26 15:46:30 +01:00
hostinfo.logger(n.l).
2019-11-19 18:00:20 +01:00
WithField("tunnelCheck", m{"state": "testing", "method": "active"}).
Debug("Tunnel status")
if hostinfo != nil && hostinfo.ConnectionState != nil {
// Send a test packet to trigger an authenticated tunnel test, this should suss out any lingering tunnel issues
n.intf.SendMessageToVpnIp(header.Test, header.TestRequest, vpnIp, p, nb, out)
2019-11-19 18:00:20 +01:00
} else {
hostinfo.logger(n.l).Debugf("Hostinfo sadness: %s", vpnIp)
2019-11-19 18:00:20 +01:00
}
n.AddPendingDeletion(vpnIp)
2019-11-19 18:00:20 +01:00
}
}
func (n *connectionManager) HandleDeletionTick(now time.Time) {
n.pendingDeletionTimer.advance(now)
for {
ep := n.pendingDeletionTimer.Purge()
if ep == nil {
break
}
vpnIp := ep.(iputil.VpnIp)
2019-11-19 18:00:20 +01:00
hostinfo, err := n.hostMap.QueryVpnIp(vpnIp)
if err != nil {
n.l.Debugf("Not found in hostmap: %s", vpnIp)
if !n.intf.disconnectInvalid {
n.ClearIP(vpnIp)
n.ClearPendingDeletion(vpnIp)
continue
}
}
if n.handleInvalidCertificate(now, vpnIp, hostinfo) {
continue
}
// If we saw an incoming packets from this ip and peer's certificate is not
// expired, just ignore.
traf := n.CheckIn(vpnIp)
2019-11-19 18:00:20 +01:00
if traf {
n.l.WithField("vpnIp", vpnIp).
2019-11-19 18:00:20 +01:00
WithField("tunnelCheck", m{"state": "alive", "method": "active"}).
Debug("Tunnel status")
n.ClearIP(vpnIp)
n.ClearPendingDeletion(vpnIp)
2019-11-19 18:00:20 +01:00
continue
}
// If it comes around on deletion wheel and hasn't resolved itself, delete
if n.checkPendingDeletion(vpnIp) {
2019-11-19 18:00:20 +01:00
cn := ""
if hostinfo.ConnectionState != nil && hostinfo.ConnectionState.peerCert != nil {
cn = hostinfo.ConnectionState.peerCert.Details.Name
}
2021-03-26 15:46:30 +01:00
hostinfo.logger(n.l).
2019-11-19 18:00:20 +01:00
WithField("tunnelCheck", m{"state": "dead", "method": "active"}).
WithField("certName", cn).
Info("Tunnel status")
n.ClearIP(vpnIp)
n.ClearPendingDeletion(vpnIp)
2019-11-19 18:00:20 +01:00
// TODO: This is only here to let tests work. Should do proper mocking
if n.intf.lightHouse != nil {
n.intf.lightHouse.DeleteVpnIp(vpnIp)
2019-11-19 18:00:20 +01:00
}
n.hostMap.DeleteHostInfo(hostinfo)
2019-11-19 18:00:20 +01:00
} else {
n.ClearIP(vpnIp)
n.ClearPendingDeletion(vpnIp)
2019-11-19 18:00:20 +01:00
}
}
}
// handleInvalidCertificates will destroy a tunnel if pki.disconnect_invalid is true and the certificate is no longer valid
func (n *connectionManager) handleInvalidCertificate(now time.Time, vpnIp iputil.VpnIp, hostinfo *HostInfo) bool {
if !n.intf.disconnectInvalid {
return false
}
remoteCert := hostinfo.GetCert()
if remoteCert == nil {
return false
}
valid, err := remoteCert.Verify(now, n.intf.caPool)
if valid {
return false
}
fingerprint, _ := remoteCert.Sha256Sum()
n.l.WithField("vpnIp", vpnIp).WithError(err).
WithField("certName", remoteCert.Details.Name).
WithField("fingerprint", fingerprint).
Info("Remote certificate is no longer valid, tearing down the tunnel")
// Inform the remote and close the tunnel locally
n.intf.sendCloseTunnel(hostinfo)
n.intf.closeTunnel(hostinfo, false)
n.ClearIP(vpnIp)
n.ClearPendingDeletion(vpnIp)
return true
}