wgengine/magicsock: make debug-level stuff not logged by default

And add a CLI/localapi and c2n mechanism to enable it for a fixed
amount of time.

Updates #1548

Change-Id: I71674aaf959a9c6761ff33bbf4a417ffd42195a7
Signed-off-by: Brad Fitzpatrick <bradfitz@tailscale.com>

ipn/backend.go

@@ -168,6 +168,11 @@ type PartialFile struct {
 //     LocalBackend.userID, a string like "user-$USER_ID" (used in
 //     server mode).
 //   - on Linux/etc, it's always "_daemon" (ipn.GlobalDaemonStateKey)
+//
+// Additionally, the StateKey can be debug setting name:
+//
+//   - "_debug_magicsock_until" with value being a unix timestamp stringified
+//   - "_debug_<component>_until" with value being a unix timestamp stringified
 type StateKey string
 
 type Options struct {

ipn/ipnlocal/c2n.go

@@ -8,6 +8,8 @@ import (
 	"encoding/json"
 	"io"
 	"net/http"
+	"strconv"
+	"time"
 
 	"tailscale.com/tailcfg"
 	"tailscale.com/util/clientmetric"
@@ -32,6 +34,21 @@ func (b *LocalBackend) handleC2N(w http.ResponseWriter, r *http.Request) {
 	case "/debug/metrics":
 		w.Header().Set("Content-Type", "text/plain")
 		clientmetric.WritePrometheusExpositionFormat(w)
+	case "/debug/component-logging":
+		component := r.FormValue("component")
+		secs, _ := strconv.Atoi(r.FormValue("secs"))
+		if secs == 0 {
+			secs -= 1
+		}
+		until := time.Now().Add(time.Duration(secs) * time.Second)
+		err := b.SetComponentDebugLogging(component, until)
+		var res struct {
+			Error string `json:",omitempty"`
+		}
+		if err != nil {
+			res.Error = err.Error()
+		}
+		writeJSON(res)
 	case "/ssh/usernames":
 		var req tailcfg.C2NSSHUsernamesRequest
 		if r.Method == "POST" {

ipn/ipnlocal/local.go

@@ -24,6 +24,7 @@ import (
 	"time"
 
 	"go4.org/netipx"
+	"golang.org/x/exp/slices"
 	"tailscale.com/client/tailscale/apitype"
 	"tailscale.com/control/controlclient"
 	"tailscale.com/doctor"
@@ -55,6 +56,7 @@ import (
 	"tailscale.com/types/views"
 	"tailscale.com/util/deephash"
 	"tailscale.com/util/dnsname"
+	"tailscale.com/util/mak"
 	"tailscale.com/util/multierr"
 	"tailscale.com/util/osshare"
 	"tailscale.com/util/systemd"
@@ -186,6 +188,7 @@ type LocalBackend struct {
 	// *.partial file to its final name on completion.
 	directFileRoot          string
 	directFileDoFinalRename bool // false on macOS, true on several NAS platforms
+	componentLogUntil       map[string]componentLogState
 
 	// statusLock must be held before calling statusChanged.Wait() or
 	// statusChanged.Broadcast().
@@ -267,9 +270,91 @@ func NewLocalBackend(logf logger.Logf, logid string, store ipn.StateStore, diale
 		b.logf("[unexpected] failed to wire up peer API port for engine %T", e)
 	}
 
+	for _, component := range debuggableComponents {
+		key := componentStateKey(component)
+		if ut, err := ipn.ReadStoreInt(store, key); err == nil {
+			if until := time.Unix(ut, 0); until.After(time.Now()) {
+				// conditional to avoid log spam at start when off
+				b.SetComponentDebugLogging(component, until)
+			}
+		}
+	}
+
 	return b, nil
 }
 
+type componentLogState struct {
+	until time.Time
+	timer *time.Timer // if non-nil, the AfterFunc to disable it
+}
+
+var debuggableComponents = []string{
+	"magicsock",
+}
+
+func componentStateKey(component string) ipn.StateKey {
+	return ipn.StateKey("_debug_" + component + "_until")
+}
+
+// SetComponentDebugLogging sets component's debug logging enabled until the until time.
+// If until is in the past, the component's debug logging is disabled.
+//
+// The following components are recognized:
+//
+//   - magicsock
+func (b *LocalBackend) SetComponentDebugLogging(component string, until time.Time) error {
+	b.mu.Lock()
+	defer b.mu.Unlock()
+
+	var setEnabled func(bool)
+	switch component {
+	case "magicsock":
+		mc, err := b.magicConn()
+		if err != nil {
+			return err
+		}
+		setEnabled = mc.SetDebugLoggingEnabled
+	}
+	if setEnabled == nil || !slices.Contains(debuggableComponents, component) {
+		return fmt.Errorf("unknown component %q", component)
+	}
+	timeUnixOrZero := func(t time.Time) int64 {
+		if t.IsZero() {
+			return 0
+		}
+		return t.Unix()
+	}
+	ipn.PutStoreInt(b.store, componentStateKey(component), timeUnixOrZero(until))
+	now := time.Now()
+	on := now.Before(until)
+	setEnabled(on)
+	var onFor time.Duration
+	if on {
+		onFor = until.Sub(now)
+		b.logf("debugging logging for component %q enabled for %v (until %v)", component, onFor.Round(time.Second), until.UTC().Format(time.RFC3339))
+	} else {
+		b.logf("debugging logging for component %q disabled", component)
+	}
+	if oldSt, ok := b.componentLogUntil[component]; ok && oldSt.timer != nil {
+		oldSt.timer.Stop()
+	}
+	newSt := componentLogState{until: until}
+	if on {
+		newSt.timer = time.AfterFunc(onFor, func() {
+			// Turn off logging after the timer fires, as long as the state is
+			// unchanged when the timer actually fires.
+			b.mu.Lock()
+			defer b.mu.Unlock()
+			if ls := b.componentLogUntil[component]; ls.until == until {
+				setEnabled(false)
+				b.logf("debugging logging for component %q disabled (by timer)", component)
+			}
+		})
+	}
+	mak.Set(&b.componentLogUntil, component, newSt)
+	return nil
+}
+
 // Dialer returns the backend's dialer.
 func (b *LocalBackend) Dialer() *tsdial.Dialer {
 	return b.dialer

ipn/localapi/localapi.go

@@ -146,6 +146,8 @@ func (h *Handler) ServeHTTP(w http.ResponseWriter, r *http.Request) {
 		h.serveMetrics(w, r)
 	case "/localapi/v0/debug":
 		h.serveDebug(w, r)
+	case "/localapi/v0/component-debug-logging":
+		h.serveComponentDebugLogging(w, r)
 	case "/localapi/v0/set-expiry-sooner":
 		h.serveSetExpirySooner(w, r)
 	case "/localapi/v0/dial":
@@ -329,6 +331,24 @@ func (h *Handler) serveDebug(w http.ResponseWriter, r *http.Request) {
 	io.WriteString(w, "done\n")
 }
 
+func (h *Handler) serveComponentDebugLogging(w http.ResponseWriter, r *http.Request) {
+	if !h.PermitWrite {
+		http.Error(w, "debug access denied", http.StatusForbidden)
+		return
+	}
+	component := r.FormValue("component")
+	secs, _ := strconv.Atoi(r.FormValue("secs"))
+	err := h.b.SetComponentDebugLogging(component, time.Now().Add(time.Duration(secs)*time.Second))
+	var res struct {
+		Error string
+	}
+	if err != nil {
+		res.Error = err.Error()
+	}
+	w.Header().Set("Content-Type", "application/json")
+	json.NewEncoder(w).Encode(res)
+}
+
 // serveProfileFunc is the implementation of Handler.serveProfile, after auth,
 // for platforms where we want to link it in.
 var serveProfileFunc func(http.ResponseWriter, *http.Request)

ipn/store.go

@@ -6,6 +6,8 @@ package ipn
 
 import (
 	"errors"
+	"fmt"
+	"strconv"
 )
 
 // ErrStateNotExist is returned by StateStore.ReadState when the
@@ -35,7 +37,7 @@ const (
 	// StateKey "user-1234".
 	ServerModeStartKey = StateKey("server-mode-start-key")
 
-	// NLKeyStateKey is the key under which we store the nodes'
+	// NLKeyStateKey is the key under which we store the node's
 	// network-lock node key, in its key.NLPrivate.MarshalText representation.
 	NLKeyStateKey = StateKey("_nl-node-key")
 )
@@ -48,3 +50,17 @@ type StateStore interface {
 	// WriteState saves bs as the state associated with ID.
 	WriteState(id StateKey, bs []byte) error
 }
+
+// ReadStoreInt reads an integer from a StateStore.
+func ReadStoreInt(store StateStore, id StateKey) (int64, error) {
+	v, err := store.ReadState(id)
+	if err != nil {
+		return 0, err
+	}
+	return strconv.ParseInt(string(v), 10, 64)
+}
+
+// PutStoreInt puts an integer into a StateStore.
+func PutStoreInt(store StateStore, id StateKey, val int64) error {
+	return store.WriteState(id, fmt.Appendf(nil, "%d", val))
+}