daemon: add ability to use syscall based reboot

internals/daemon/daemon.go

@@ -515,7 +515,7 @@ func (d *Daemon) HandleRestart(t restart.RestartType) {
 	case restart.RestartSystem:
 		// try to schedule a fallback slow reboot already here
 		// in case we get stuck shutting down
-		if err := reboot(rebootWaitTimeout); err != nil {
+		if err := rebootHandler(rebootWaitTimeout); err != nil {
 			logger.Noticef("%s", err)
 		}
 
@@ -700,11 +700,11 @@ func (d *Daemon) doReboot(sigCh chan<- os.Signal, waitTimeout time.Duration) err
 	}
 	// ask for shutdown and wait for it to happen.
 	// if we exit, pebble will be restarted by systemd
-	if err := reboot(rebootDelay); err != nil {
+	if err := rebootHandler(rebootDelay); err != nil {
 		return err
 	}
 	// wait for reboot to happen
-	logger.Noticef("Waiting for system reboot")
+	logger.Noticef("Waiting for system reboot...")
 	if sigCh != nil {
 		signal.Stop(sigCh)
 		if len(sigCh) > 0 {
@@ -717,21 +717,55 @@ func (d *Daemon) doReboot(sigCh chan<- os.Signal, waitTimeout time.Duration) err
 	return fmt.Errorf("expected reboot did not happen")
 }
 
-var shutdownMsg = "reboot scheduled to update the system"
+const rebootMsg = "reboot scheduled to update the system"
 
-func rebootImpl(rebootDelay time.Duration) error {
+var rebootHandler = commandReboot
+
+// SetSyscallReboot replaces the default command-based reboot
+// with a direct Linux kernel syscall based implementation.
+func SetSyscallReboot() {
+	rebootHandler = syscallReboot
+}
+
+// commandReboot assumes a userspace shutdown command exists.
+func commandReboot(rebootDelay time.Duration) error {
 	if rebootDelay < 0 {
 		rebootDelay = 0
 	}
 	mins := int64(rebootDelay / time.Minute)
-	cmd := exec.Command("shutdown", "-r", fmt.Sprintf("+%d", mins), shutdownMsg)
+	cmd := exec.Command("shutdown", "-r", fmt.Sprintf("+%d", mins), rebootMsg)
 	if out, err := cmd.CombinedOutput(); err != nil {
 		return osutil.OutputErr(out, err)
 	}
 	return nil
 }
 
-var reboot = rebootImpl
+var (
+	syncSyscall   = syscall.Sync
+	rebootSyscall = syscall.Reboot
+)
+
+// syscallReboot performs a delayed async reboot using direct Linux
+// kernel syscalls.
+//
+// Note: Reboot message not currently supported.
+func syscallReboot(rebootDelay time.Duration) error {
+	if rebootDelay < 0 {
+		rebootDelay = 0
+	}
+	// This has to be non-blocking, and scheduled for a future
+	// point in time to mimic shutdown.
+	time.AfterFunc(rebootDelay, func() {
+		// As per the requirements of the reboot syscall, we
+		// have to first call sync.
+		syncSyscall()
+		err := rebootSyscall(syscall.LINUX_REBOOT_CMD_RESTART)
+		if err != nil {
+			logger.Noticef("Failed on reboot syscall: %v", err)
+		}
+	})
+	return nil
+}
 
 func (d *Daemon) Dying() <-chan struct{} {
 	return d.tomb.Dying()

internals/daemon/daemon_test.go

@@ -30,12 +30,12 @@ import (
 	"time"
 
 	"github.com/gorilla/mux"
-
 	"gopkg.in/check.v1"
 
 	// XXX Delete import above and make this file like the other ones.
 	. "gopkg.in/check.v1"
 
+	"github.com/canonical/pebble/internals/logger"
 	"github.com/canonical/pebble/internals/osutil"
 	"github.com/canonical/pebble/internals/overlord/patch"
 	"github.com/canonical/pebble/internals/overlord/restart"
@@ -665,15 +665,15 @@ func (s *daemonSuite) TestRestartSystemWiring(c *check.C) {
 	oldRebootNoticeWait := rebootNoticeWait
 	oldRebootWaitTimeout := rebootWaitTimeout
 	defer func() {
-		reboot = rebootImpl
+		rebootHandler = commandReboot
 		rebootNoticeWait = oldRebootNoticeWait
 		rebootWaitTimeout = oldRebootWaitTimeout
 	}()
 	rebootWaitTimeout = 100 * time.Millisecond
 	rebootNoticeWait = 150 * time.Millisecond
 
 	var delays []time.Duration
-	reboot = func(d time.Duration) error {
+	rebootHandler = func(d time.Duration) error {
 		delays = append(delays, d)
 		return nil
 	}
@@ -740,7 +740,7 @@ func (s *daemonSuite) TestRebootHelper(c *check.C) {
 	}
 
 	for _, t := range tests {
-		err := reboot(t.delay)
+		err := rebootHandler(t.delay)
 		c.Assert(err, check.IsNil)
 		c.Check(cmd.Calls(), check.DeepEquals, [][]string{
 			{"shutdown", "-r", t.delayArg, "reboot scheduled to update the system"},
@@ -1148,3 +1148,136 @@ services:
 	c.Assert(tasks, HasLen, 1)
 	c.Check(tasks[0].Kind(), Equals, "stop")
 }
+
+type rebootSuite struct{}
+
+var _ = check.Suite(&rebootSuite{})
+
+func mockSyncSyscall(f func()) (restore func()) {
+	old := syncSyscall
+	syncSyscall = f
+	return func() {
+		syncSyscall = old
+	}
+}
+
+func mockRebootSyscall(f func(cmd int) error) (restore func()) {
+	old := rebootSyscall
+	rebootSyscall = f
+	return func() {
+		rebootSyscall = old
+	}
+}
+
+func (s *rebootSuite) TestSyscallPosRebootDelay(c *C) {
+	wait := make(chan int)
+	defer mockSyncSyscall(func() {})()
+	defer mockRebootSyscall(func(cmd int) error {
+		if cmd == syscall.LINUX_REBOOT_CMD_RESTART {
+			wait <- 1
+		}
+		return nil
+	})()
+
+	period := 25 * time.Millisecond
+	syscallReboot(period)
+	start := time.Now()
+	select {
+	case <-wait:
+	case <-time.After(10 * time.Second):
+		c.Fatal("syscall did not take place and we timed out")
+	}
+	elapsed := time.Now().Sub(start)
+	c.Assert(elapsed >= period, Equals, true)
+}
+
+func (s *rebootSuite) TestSyscallNegRebootDelay(c *C) {
+	wait := make(chan int)
+	defer mockSyncSyscall(func() {})()
+	defer mockRebootSyscall(func(cmd int) error {
+		if cmd == syscall.LINUX_REBOOT_CMD_RESTART {
+			wait <- 1
+		}
+		return nil
+	})()
+
+	// Negative periods will be zeroed, so do not fear the huge negative.
+	// We do supply a rather big value here because this test is
+	// effectively a race, but given the huge timeout, it is not going
+	// to be a problem (c).
+	period := 10 * time.Second
+	syscallReboot(-period)
+	start := time.Now()
+	select {
+	case <-wait:
+	case <-time.After(10 * time.Second):
+		c.Fatal("syscall did not take place and we timed out")
+	}
+	elapsed := time.Now().Sub(start)
+	c.Assert(elapsed < period, Equals, true)
+}
+
+func (s *rebootSuite) TestSetSyscall(c *C) {
+	wait := make(chan int)
+	defer mockSyncSyscall(func() {})()
+	defer mockRebootSyscall(func(cmd int) error {
+		if cmd == syscall.LINUX_REBOOT_CMD_RESTART {
+			wait <- 1
+		}
+		return nil
+	})()
+
+	// We know the default is commandReboot otherwise the unit tests
+	// above will fail. We need to check the switch works.
+	SetSyscallReboot()
+	defer func() {
+		rebootHandler = commandReboot
+	}()
+
+	err := rebootHandler(0)
+	c.Assert(err, IsNil)
+	select {
+	case <-wait:
+	case <-time.After(10 * time.Second):
+		c.Fatal("syscall did not take place and we timed out")
+	}
+}
+
+type fakeLogger struct {
+	msg      string
+	noticeCh chan int
+}
+
+func (f *fakeLogger) Notice(msg string) {
+	f.msg = msg
+	f.noticeCh <- 1
+}
+
+func (f *fakeLogger) Debug(msg string) {}
+
+func (s *rebootSuite) TestSyscallRebootError(c *C) {
+	defer mockSyncSyscall(func() {})()
+	defer mockRebootSyscall(func(cmd int) error {
+		return fmt.Errorf("-EPERM")
+	})()
+
+	// We know the default is commandReboot otherwise the unit tests
+	// above will fail. We need to check the switch works.
+	SetSyscallReboot()
+	defer func() {
+		rebootHandler = commandReboot
+	}()
+	complete := make(chan int)
+	l := fakeLogger{noticeCh: complete}
+	old := logger.SetLogger(&l)
+	defer logger.SetLogger(old)
+
+	err := rebootHandler(0)
+	c.Assert(err, IsNil)
+	select {
+	case <-complete:
+	case <-time.After(10 * time.Second):
+		c.Fatal("syscall did not take place and we timed out")
+	}
+	c.Assert(l.msg, Matches, "*-EPERM")
+}