mirror of
https://github.com/tailscale/tailscale.git
synced 2026-09-15 15:29:33 -04:00
Brad Fitzpatrick
4c4d1c35f8
LocalBackend.Shutdown waits for the ACME refresh loop and active SSH sessions. Both can be blocked acquiring LocalBackend.mu, so waiting while holding that mutex deadlocks shutdown. Detach the SSH server under the mutex, then stop both subsystems after releasing it. Prevent their work from restarting once shutdown begins, and serialize repeated Shutdown calls with sync.Once. Add regression tests that verify subsystem shutdown runs without LocalBackend.mu held. Updates tailscale/corp#45964 Change-Id: I37ead4f26fbfb5703a83882668d98a8862ba7d67 Signed-off-by: Brad Fitzpatrick <bradfitz@tailscale.com>
85 lines
2.0 KiB
Go
85 lines
2.0 KiB
Go
// Copyright (c) Tailscale Inc & contributors
|
|
// SPDX-License-Identifier: BSD-3-Clause
|
|
|
|
package ipnlocal
|
|
|
|
import (
|
|
"errors"
|
|
"net"
|
|
"sync/atomic"
|
|
"testing"
|
|
|
|
"tailscale.com/ipn"
|
|
)
|
|
|
|
type shutdownTestSSHServer struct {
|
|
b *LocalBackend
|
|
calls atomic.Int32
|
|
t *testing.T
|
|
}
|
|
|
|
func (s *shutdownTestSSHServer) HandleSSHConn(net.Conn) error { return nil }
|
|
func (s *shutdownTestSSHServer) NumActiveConns() int { return 0 }
|
|
func (s *shutdownTestSSHServer) OnPolicyChange() {}
|
|
|
|
func (s *shutdownTestSSHServer) Shutdown() {
|
|
s.calls.Add(1)
|
|
if !s.b.mu.TryLock() {
|
|
s.t.Error("LocalBackend.mu held while shutting down SSH server")
|
|
return
|
|
}
|
|
s.b.mu.Unlock()
|
|
s.b.NodeKey() // do something that requires the lock
|
|
}
|
|
|
|
func TestShutdownReleasesMutexBeforeWaitingForSubsystems(t *testing.T) {
|
|
b := newTestLocalBackend(t)
|
|
ssh := &shutdownTestSSHServer{b: b, t: t}
|
|
b.mu.Lock()
|
|
b.sshServer = ssh
|
|
b.mu.Unlock()
|
|
|
|
var certShutdownCalls atomic.Int32
|
|
restore := HookShutdownCertRefreshLoop.SetForTest(func(b *LocalBackend) {
|
|
certShutdownCalls.Add(1)
|
|
if !b.mu.TryLock() {
|
|
t.Error("LocalBackend.mu held while shutting down cert refresh loop")
|
|
return
|
|
}
|
|
b.mu.Unlock()
|
|
b.ServeConfig()
|
|
})
|
|
t.Cleanup(restore)
|
|
|
|
b.Shutdown()
|
|
b.Shutdown()
|
|
|
|
if got := certShutdownCalls.Load(); got != 1 {
|
|
t.Errorf("cert refresh shutdown called %d times; want 1", got)
|
|
}
|
|
if got := ssh.calls.Load(); got != 1 {
|
|
t.Errorf("SSH shutdown called %d times; want 1", got)
|
|
}
|
|
if _, err := b.sshServerOrInit(); !errors.Is(err, errShutdown) {
|
|
t.Errorf("sshServerOrInit after Shutdown = %v; want %v", err, errShutdown)
|
|
}
|
|
}
|
|
|
|
func TestCertRefreshLoopDoesNotRestartAfterShutdown(t *testing.T) {
|
|
b := newTestLocalBackend(t)
|
|
var updates atomic.Int32
|
|
restore := HookUpdateCertRefreshLoop.SetForTest(func(*LocalBackend, ipn.State, ipn.ServeConfigView) {
|
|
updates.Add(1)
|
|
})
|
|
t.Cleanup(restore)
|
|
|
|
b.Shutdown()
|
|
b.mu.Lock()
|
|
b.updateCertRefreshLoopLocked()
|
|
b.mu.Unlock()
|
|
|
|
if got := updates.Load(); got != 0 {
|
|
t.Errorf("cert refresh loop restarted %d times after Shutdown; want 0", got)
|
|
}
|
|
}
|