client,cmd/tailscale,ipn/{ipnlocal,localapi}: add debug CLI command to clear netmap caches (#19213)

This is a follow-up to #19117, adding a debug CLI command allowing the operator
to explicitly discard cached netmap data, as a safety and recovery measure.

Updates #12639

Change-Id: I5c3c47c0204754b9c8e526a4ff8f69d6974db6d0
Signed-off-by: M. J. Fromberger <fromberger@tailscale.com>

cmd/tailscale/cli/debug-cachenetmap.go

@@ -0,0 +1,31 @@
+// Copyright (c) Tailscale Inc & contributors
+// SPDX-License-Identifier: BSD-3-Clause
+
+//go:build !ios && !ts_omit_cachenetmap
+
+package cli
+
+import (
+	"context"
+	"errors"
+
+	"github.com/peterbourgon/ff/v3/ffcli"
+)
+
+func init() {
+	debugClearNetmapCacheCmd = func() *ffcli.Command {
+		return &ffcli.Command{
+			Name:       "clear-netmap-cache",
+			ShortUsage: "tailscale debug clear-netmap-cache",
+			ShortHelp:  "Remove and discard cached network maps (if any)",
+			Exec:       runDebugClearNetmapCache,
+		}
+	}
+}
+
+func runDebugClearNetmapCache(ctx context.Context, args []string) error {
+	if len(args) != 0 {
+		return errors.New("unexpected arguments")
+	}
+	return localClient.DebugAction(ctx, "clear-netmap-cache")
+}

cmd/tailscale/cli/debug.go

@@ -51,9 +51,10 @@ import (
 )
 
 var (
-	debugCaptureCmd   func() *ffcli.Command // or nil
-	debugPortmapCmd   func() *ffcli.Command // or nil
-	debugPeerRelayCmd func() *ffcli.Command // or nil
+	debugCaptureCmd          func() *ffcli.Command // or nil
+	debugPortmapCmd          func() *ffcli.Command // or nil
+	debugPeerRelayCmd        func() *ffcli.Command // or nil
+	debugClearNetmapCacheCmd func() *ffcli.Command // or nil
 )
 
 func debugCmd() *ffcli.Command {
@@ -394,6 +395,7 @@ func debugCmd() *ffcli.Command {
 				Exec:       runPrintStateDir,
 			},
 			ccall(debugPeerRelayCmd),
+			ccall(debugClearNetmapCacheCmd),
 		}...),
 	}
 }

ipn/ipnlocal/diskcache.go

@@ -4,6 +4,10 @@
 package ipnlocal
 
 import (
+	"context"
+	"errors"
+	"fmt"
+
 	"tailscale.com/feature/buildfeatures"
 	"tailscale.com/ipn/ipnlocal/netmapcache"
 	"tailscale.com/types/netmap"
@@ -64,21 +68,50 @@ func (b *LocalBackend) discardDiskCacheLocked() {
 	if b.diskCache.cache == nil {
 		return // nothing to do, we do not have a cache
 	}
-
 	// Reaching here, we have a cache directory that needs to be purged.
 	// Log errors but do not fail for them.
 	store := netmapcache.FileStore(b.diskCache.dir)
-	ctx := b.currentNode().Context()
+	if err := b.clearStoreLocked(b.currentNode().Context(), store); err != nil {
+		b.logf("clearing netmap cache: %v", err)
+	}
+	b.diskCache = diskCache{} // drop in-memory state
+}
+
+// clearStoreLocked discards all the keys in the specified store.
+func (b *LocalBackend) clearStoreLocked(ctx context.Context, store netmapcache.Store) error {
+	var errs []error
 	for key, err := range store.List(ctx, "") {
 		if err != nil {
-			b.logf("listing cache contents: %v", err)
+			errs = append(errs, fmt.Errorf("list cache contest: %w", err))
 			break
 		}
 		if err := store.Remove(ctx, key); err != nil {
-			b.logf("discarding cache key %q: %v", key, err)
+			errs = append(errs, fmt.Errorf("discard cache key %q: %w", key, err))
+		}
+	}
+	return errors.Join(errs...)
+}
+
+// ClearNetmapCache discards stored netmap caches (if any) for profiles for the
+// current user of b. It also drops any cache from the active backend session,
+// if there is one.
+func (b *LocalBackend) ClearNetmapCache(ctx context.Context) error {
+	if !buildfeatures.HasCacheNetMap {
+		return nil // disabled
+	}
+
+	b.mu.Lock()
+	defer b.mu.Unlock()
+
+	var errs []error
+	for _, p := range b.pm.Profiles() {
+		store := netmapcache.FileStore(b.profileDataPathLocked(p.ID(), "netmap-cache"))
+		err := b.clearStoreLocked(ctx, store)
+		if err != nil {
+			errs = append(errs, fmt.Errorf("clear netmap cache for profile %q: %w", p.ID(), err))
 		}
 	}
 
-	b.diskCache.cache = nil // drop reference
-	b.diskCache.dir = ""
+	b.diskCache = diskCache{} // drop in-memory state
+	return errors.Join(errs...)
 }

ipn/localapi/debug.go

@@ -241,6 +241,8 @@ func (h *Handler) serveDebug(w http.ResponseWriter, r *http.Request) {
 		if err == nil {
 			return
 		}
+	case "clear-netmap-cache":
+		h.b.ClearNetmapCache(r.Context())
 	case "":
 		err = fmt.Errorf("missing parameter 'action'")
 	default: