controlclient, ipn: add endpoints to manage sleep/wakeangott/sleep-debug-apis

Updates #3363. Adds localapi + debug endpoints to set/unset sleep mode.

Signed-off-by: Andrea Gottardo <andrea@gottardo.me>

7 files changed, 85 insertions, 6 deletions

diff --git a/cmd/tailscale/cli/debug.go b/cmd/tailscale/cli/debug.go
index 51e825803..374e7fd98 100644
--- a/cmd/tailscale/cli/debug.go
+++ b/cmd/tailscale/cli/debug.go
@@ -132,7 +132,7 @@ var debugCmd = &ffcli.Command{
 		{
 			Name:      "derp-set-homeless",
 			Exec:      localAPIAction("derp-set-homeless"),
-			ShortHelp: "enable DERP homeless mode (breaks reachablility)",
+			ShortHelp: "enable DERP homeless mode (breaks reachability)",
 		},
 		{
 			Name:      "derp-unset-homeless",
@@ -140,6 +140,16 @@ var debugCmd = &ffcli.Command{
 			ShortHelp: "disable DERP homeless mode",
 		},
 		{
+			Name:      "sleep-set",
+			Exec:      localAPIAction("sleep-set"),
+			ShortHelp: "asks the backend to enter sleep mode",
+		},
+		{
+			Name:      "sleep-unset",
+			Exec:      localAPIAction("sleep-unset"),
+			ShortHelp: "asks the backend to leave sleep mode and resume all features",
+		},
+		{
 			Name:      "break-tcp-conns",
 			Exec:      localAPIAction("break-tcp-conns"),
 			ShortHelp: "break any open TCP connections from the daemon",
diff --git a/control/controlclient/auto.go b/control/controlclient/auto.go
index d0551cdab..f4d9a4f1a 100644
--- a/control/controlclient/auto.go
+++ b/control/controlclient/auto.go
@@ -139,6 +139,7 @@ type Auto struct {
 	loginGoal      *LoginGoal  // non-nil if some login activity is desired
 	inMapPoll      bool        // true once we get the first MapResponse in a stream; false when HTTP response ends
 	state          State       // TODO(bradfitz): delete this, make it computed by method from other state
+	isSleeping     bool        // whether we are ZZZing
 
 	authCtx    context.Context // context used for auth requests
 	mapCtx     context.Context // context used for netmap and update requests
@@ -200,6 +201,16 @@ func NewNoStart(opts Options) (_ *Auto, err error) {
 
 }
 
+func (c *Auto) SetSleepMode(enabled bool) {
+	c.logf("setSleepMode(%v)", enabled)
+	c.isSleeping = enabled
+	c.SetPaused(enabled)
+}
+
+func (c *Auto) IsSleeping() bool {
+	return c.isSleeping
+}
+
 // SetPaused controls whether HTTP activity should be paused.
 //
 // The client can be paused and unpaused repeatedly, unlike Start and Shutdown, which can only be used once.
diff --git a/control/controlclient/client.go b/control/controlclient/client.go
index ef5af68c6..2acd82ef8 100644
--- a/control/controlclient/client.go
+++ b/control/controlclient/client.go
@@ -54,6 +54,10 @@ type Client interface {
 	// TODO: It might be better to simply shutdown the controlclient and
 	// make a new one when it's time to unpause.
 	SetPaused(bool)
+	// SetSleepMode pauses the control client and prevents anybody else
+	// from unpausing it until SetSleepMode(false) is called again
+	SetSleepMode(bool)
+	IsSleeping() bool
 	// AuthCantContinue returns whether authentication is blocked. If it
 	// is, you either need to visit the auth URL (previously sent in a
 	// Status callback) or call the Login function appropriately.
diff --git a/ipn/ipnlocal/local.go b/ipn/ipnlocal/local.go
index b8aa769a1..3bc164124 100644
--- a/ipn/ipnlocal/local.go
+++ b/ipn/ipnlocal/local.go
@@ -53,6 +53,7 @@ import (
 	"tailscale.com/ipn/policy"
 	"tailscale.com/log/sockstatlog"
 	"tailscale.com/logpolicy"
+	"tailscale.com/logtail"
 	"tailscale.com/net/dns"
 	"tailscale.com/net/dnscache"
 	"tailscale.com/net/dnsfallback"
@@ -580,7 +581,12 @@ func (b *LocalBackend) pauseOrResumeControlClientLocked() {
 		return
 	}
 	networkUp := b.prevIfState.AnyInterfaceUp()
-	b.cc.SetPaused((b.state == ipn.Stopped && b.netMap != nil) || (!networkUp && !testenv.InTest()))
+	shouldPause := (b.state == ipn.Stopped && b.netMap != nil) || (!networkUp && !testenv.InTest())
+	if b.cc.IsSleeping() && shouldPause == false {
+		// Leave things untouched if a request to un-pause comes in while we should be sleeping.
+		return
+	}
+	b.cc.SetPaused(shouldPause)
 }
 
 // linkChange is our network monitor callback, called whenever the network changes.
@@ -5327,6 +5333,17 @@ func peerCanProxyDNS(p tailcfg.NodeView) bool {
 	return false
 }
 
+func (b *LocalBackend) SetSleep(enabled bool) {
+	b.logf("SetSleep: enabled = %v", enabled)
+	b.cc.SetSleepMode(enabled)
+	b.MagicConn().SetHomeless(enabled)
+	if enabled {
+		logtail.Disable()
+	} else {
+		logtail.Enable()
+	}
+}
+
 func (b *LocalBackend) DebugRebind() error {
 	b.MagicConn().Rebind()
 	return nil
diff --git a/ipn/ipnlocal/state_test.go b/ipn/ipnlocal/state_test.go
index 403233a11..395fc750a 100644
--- a/ipn/ipnlocal/state_test.go
+++ b/ipn/ipnlocal/state_test.go
@@ -91,10 +91,11 @@ func (nt *notifyThrottler) drain(count int) []ipn.Notify {
 // in the controlclient.Client, so by controlling it, we can check that
 // the state machine works as expected.
 type mockControl struct {
-	tb     testing.TB
-	logf   logger.Logf
-	opts   controlclient.Options
-	paused atomic.Bool
+	tb         testing.TB
+	logf       logger.Logf
+	opts       controlclient.Options
+	paused     atomic.Bool
+	isSleeping atomic.Bool
 
 	mu          sync.Mutex
 	machineKey  key.MachinePrivate
@@ -236,6 +237,18 @@ func (cc *mockControl) SetPaused(paused bool) {
 	}
 }
 
+func (cc *mockControl) SetSleepMode(enabled bool) {
+	cc.mu.Lock()
+	defer cc.mu.Unlock()
+	cc.isSleeping.Store(enabled)
+}
+
+func (cc *mockControl) IsSleeping() bool {
+	cc.mu.Lock()
+	defer cc.mu.Unlock()
+	return cc.isSleeping.Load()
+}
+
 func (cc *mockControl) AuthCantContinue() bool {
 	cc.mu.Lock()
 	defer cc.mu.Unlock()
diff --git a/ipn/localapi/localapi.go b/ipn/localapi/localapi.go
index a1b7da46f..cd5ebb811 100644
--- a/ipn/localapi/localapi.go
+++ b/ipn/localapi/localapi.go
@@ -110,6 +110,7 @@ var handler = map[string]localAPIHandler{
 	"serve-config":                (*Handler).serveServeConfig,
 	"set-dns":                     (*Handler).serveSetDNS,
 	"set-expiry-sooner":           (*Handler).serveSetExpirySooner,
+	"set-sleep":                   (*Handler).serveSetSleep,
 	"tailfs/fileserver-address":   (*Handler).serveTailFSFileServerAddr,
 	"tailfs/shares":               (*Handler).serveShares,
 	"start":                       (*Handler).serveStart,
@@ -573,6 +574,10 @@ func (h *Handler) serveDebug(w http.ResponseWriter, r *http.Request) {
 		h.b.MagicConn().SetHomeless(true)
 	case "derp-unset-homeless":
 		h.b.MagicConn().SetHomeless(false)
+	case "sleep-set":
+		h.b.SetSleep(true)
+	case "sleep-unset":
+		h.b.SetSleep(false)
 	case "rebind":
 		err = h.b.DebugRebind()
 	case "restun":
@@ -1695,6 +1700,20 @@ func (h *Handler) serveSetExpirySooner(w http.ResponseWriter, r *http.Request) {
 	io.WriteString(w, "done\n")
 }
 
+func (h *Handler) serveSetSleep(w http.ResponseWriter, r *http.Request) {
+	if !h.PermitWrite {
+		http.Error(w, "access denied", http.StatusForbidden)
+		return
+	}
+	if r.Method != "POST" {
+		http.Error(w, "want POST", http.StatusBadRequest)
+		return
+	}
+	h.b.SetSleep(r.FormValue("sleep") == "true")
+	w.Header().Set("Content-Type", "application/json")
+	json.NewEncoder(w).Encode(struct{}{})
+}
+
 func (h *Handler) servePing(w http.ResponseWriter, r *http.Request) {
 	ctx := r.Context()
 	if r.Method != "POST" {
diff --git a/logtail/logtail.go b/logtail/logtail.go
index fcaf80e41..90be1b515 100644
--- a/logtail/logtail.go
+++ b/logtail/logtail.go
@@ -510,6 +510,11 @@ func (l *Logger) StartFlush() {
 // logtailDisabled is whether logtail uploads to logcatcher are disabled.
 var logtailDisabled atomic.Bool
 
+// Enable enables logtail uploads for the lifetime of the process.
+func Enable() {
+	logtailDisabled.Store(false)
+}
+
 // Disable disables logtail uploads for the lifetime of the process.
 func Disable() {
 	logtailDisabled.Store(true)