device: fix TOCTOU race during session state update (#77)
API introduced in a927a66e has two cases of state determination happening out of critical section for the state value: (1) expiredSession loads sessionExpiresNano, then releases all locks and calls noteSessionState(Expired). So a concurrent refresh that lands in that gap gets clobbered by a stale Expired -- and sticks until the next re-key. (2) Likewise in noteSessionHandshakeStopped, hasKeyMaterial check happens out of the session state lock and races with ZeroAndFlushAll. Both lead to a wrong state emitted via the device.sessionState.fn, but are otherwise benign. This moves the expiry timestamp under a lock to address the former, and provides a noteSessionStateLocked helper for the latter. Also changes API semantics to serialize events per-peer, to avoid sharing a single lock for all timestamps. Updates tailscale/corp#42874 Signed-off-by: Alex Valiushko <alexvaliushko@tailscale.com> Change-Id: Iee2cdf135375519e58a8e84362349d966a6a6964
This commit is contained in:
parent
35a60acb84
commit
15b912c1c0
3 changed files with 61 additions and 47 deletions
|
|
@ -18,16 +18,20 @@ import (
|
|||
)
|
||||
|
||||
type Peer struct {
|
||||
isRunning atomic.Bool
|
||||
keypairs Keypairs
|
||||
handshake Handshake
|
||||
device *Device
|
||||
stopping sync.WaitGroup // routines pending stop
|
||||
txBytes atomic.Uint64 // bytes send to peer (endpoint)
|
||||
rxBytes atomic.Uint64 // bytes received from peer
|
||||
lastHandshakeNano atomic.Int64 // nano seconds since epoch
|
||||
sessionExpiresNano atomic.Int64 // nano seconds since epoch
|
||||
sessionState PeerSessionState // guarded by device.sessionState.Mutex
|
||||
isRunning atomic.Bool
|
||||
keypairs Keypairs
|
||||
handshake Handshake
|
||||
device *Device
|
||||
stopping sync.WaitGroup // routines pending stop
|
||||
txBytes atomic.Uint64 // bytes send to peer (endpoint)
|
||||
rxBytes atomic.Uint64 // bytes received from peer
|
||||
lastHandshakeNano atomic.Int64 // nano seconds since epoch
|
||||
|
||||
sessionState struct {
|
||||
sync.Mutex
|
||||
current PeerSessionState
|
||||
sessionExpires time.Time
|
||||
}
|
||||
|
||||
queuedOutboundPackets atomic.Int32 // packets in staged+outbound queues, for input backpressure
|
||||
|
||||
|
|
@ -266,7 +270,6 @@ func (peer *Peer) Start() {
|
|||
|
||||
func (peer *Peer) ZeroAndFlushAll() {
|
||||
device := peer.device
|
||||
peer.sessionExpiresNano.Store(0)
|
||||
if peer.timers.sessionExpired != nil {
|
||||
peer.timers.sessionExpired.Del()
|
||||
}
|
||||
|
|
@ -292,7 +295,11 @@ func (peer *Peer) ZeroAndFlushAll() {
|
|||
handshake.mutex.Unlock()
|
||||
|
||||
peer.FlushStagedPackets()
|
||||
peer.noteSessionState(PeerSessionNone)
|
||||
|
||||
peer.sessionState.Lock()
|
||||
peer.sessionState.sessionExpires = time.Time{}
|
||||
peer.noteSessionStateLocked(PeerSessionNone)
|
||||
peer.sessionState.Unlock()
|
||||
}
|
||||
|
||||
func (peer *Peer) ExpireCurrentKeypairs() {
|
||||
|
|
@ -313,8 +320,10 @@ func (peer *Peer) ExpireCurrentKeypairs() {
|
|||
}
|
||||
keypairs.Unlock()
|
||||
|
||||
peer.sessionExpiresNano.Store(0)
|
||||
peer.noteSessionState(PeerSessionExpired)
|
||||
peer.sessionState.Lock()
|
||||
peer.sessionState.sessionExpires = time.Time{}
|
||||
peer.noteSessionStateLocked(PeerSessionExpired)
|
||||
peer.sessionState.Unlock()
|
||||
}
|
||||
|
||||
func (peer *Peer) Stop() {
|
||||
|
|
@ -338,42 +347,41 @@ func (peer *Peer) Stop() {
|
|||
}
|
||||
|
||||
func (peer *Peer) noteSessionState(state PeerSessionState) {
|
||||
device := peer.device
|
||||
device.sessionState.Lock()
|
||||
defer device.sessionState.Unlock()
|
||||
peer.sessionState.Lock()
|
||||
defer peer.sessionState.Unlock()
|
||||
peer.noteSessionStateLocked(state)
|
||||
}
|
||||
|
||||
if peer.sessionState == state {
|
||||
// noteSessionStateLocked records a session state transition and delivers the
|
||||
// callback. The caller must hold peer.sessionState.Mutex during the
|
||||
// state determination and transition.
|
||||
func (peer *Peer) noteSessionStateLocked(state PeerSessionState) {
|
||||
if peer.sessionState.current == state {
|
||||
return
|
||||
}
|
||||
peer.sessionState = state
|
||||
if f := device.sessionState.fn; f != nil {
|
||||
f(peer.handshake.remoteStatic, state)
|
||||
peer.sessionState.current = state
|
||||
if f := peer.device.peerStateFn.Load(); f != nil {
|
||||
(*f)(peer.handshake.remoteStatic, state)
|
||||
}
|
||||
}
|
||||
|
||||
func (peer *Peer) noteSessionHandshakeStarted() {
|
||||
device := peer.device
|
||||
device.sessionState.Lock()
|
||||
defer device.sessionState.Unlock()
|
||||
|
||||
switch peer.sessionState {
|
||||
case PeerSessionEstablished:
|
||||
return
|
||||
case PeerSessionHandshake:
|
||||
peer.sessionState.Lock()
|
||||
defer peer.sessionState.Unlock()
|
||||
if peer.sessionState.current == PeerSessionEstablished {
|
||||
return
|
||||
}
|
||||
peer.sessionState = PeerSessionHandshake
|
||||
if f := device.sessionState.fn; f != nil {
|
||||
f(peer.handshake.remoteStatic, PeerSessionHandshake)
|
||||
}
|
||||
peer.noteSessionStateLocked(PeerSessionHandshake)
|
||||
}
|
||||
|
||||
func (peer *Peer) noteSessionHandshakeStopped() {
|
||||
peer.sessionState.Lock()
|
||||
defer peer.sessionState.Unlock()
|
||||
state := PeerSessionNone
|
||||
if peer.hasKeyMaterial() {
|
||||
state = PeerSessionExpired
|
||||
}
|
||||
peer.noteSessionState(state)
|
||||
peer.noteSessionStateLocked(state)
|
||||
}
|
||||
|
||||
func (peer *Peer) hasKeyMaterial() bool {
|
||||
|
|
|
|||
Loading…
Add table
Add a link
Reference in a new issue