Merge main into poc/certificate-posture

client/cmd/debug.go

@@ -23,7 +23,10 @@ import (
 	"github.com/netbirdio/netbird/version"
 )
 
-const errCloseConnection = "Failed to close connection: %v"
+const (
+	errCloseConnection = "Failed to close connection: %v"
+	noUpDownFlag       = "no-updown"
+)
 
 var (
 	logFileCount             uint32
@@ -257,13 +260,14 @@ func runForDuration(cmd *cobra.Command, args []string) error {
 	}
 
 	stateWasDown := stat.Status != string(internal.StatusConnected) && stat.Status != string(internal.StatusConnecting)
+	noUpDown, _ := cmd.Flags().GetBool(noUpDownFlag)
 
 	initialLogLevel, err := client.GetLogLevel(cmd.Context(), &proto.GetLogLevelRequest{})
 	if err != nil {
 		return fmt.Errorf("failed to get log level: %v", status.Convert(err).Message())
 	}
 
-	if stateWasDown {
+	if stateWasDown && !noUpDown {
 		if _, err := client.Up(cmd.Context(), &proto.UpRequest{}); err != nil {
 			cmd.PrintErrf("Failed to bring service up: %v\n", status.Convert(err).Message())
 		} else {
@@ -284,34 +288,20 @@ func runForDuration(cmd *cobra.Command, args []string) error {
 	}
 
 	needsRestoreUp := false
-	if _, err := client.Down(cmd.Context(), &proto.DownRequest{}); err != nil {
-		cmd.PrintErrf("Failed to bring service down: %v\n", status.Convert(err).Message())
+	if noUpDown {
+		enableSyncResponsePersistence(cmd, client)
 	} else {
-		needsRestoreUp = !stateWasDown
-		cmd.Println("netbird down")
+		needsRestoreUp = restartDaemon(cmd, client, stateWasDown)
 	}
 
-	time.Sleep(1 * time.Second)
-
-	// Enable sync response persistence before bringing the service up
-	if _, err := client.SetSyncResponsePersistence(cmd.Context(), &proto.SetSyncResponsePersistenceRequest{
-		Enabled: true,
-	}); err != nil {
-		cmd.PrintErrf("Failed to enable sync response persistence: %v\n", status.Convert(err).Message())
-	}
-
-	if _, err := client.Up(cmd.Context(), &proto.UpRequest{}); err != nil {
-		cmd.PrintErrf("Failed to bring service up: %v\n", status.Convert(err).Message())
-	} else {
-		needsRestoreUp = false
-		cmd.Println("netbird up")
-	}
-
-	time.Sleep(3 * time.Second)
-
 	cpuProfilingStarted := false
 	if _, err := client.StartCPUProfile(cmd.Context(), &proto.StartCPUProfileRequest{}); err != nil {
-		cmd.PrintErrf("Failed to start CPU profiling: %v\n", err)
+		if msg := status.Convert(err).Message(); strings.Contains(msg, "already in progress") {
+			cmd.PrintErrln("CPU profiling is already running (started with `netbird debug cpu start`). " +
+				"It is left running and is included in a bundle created after `netbird debug cpu stop`.")
+		} else {
+			cmd.PrintErrf("Failed to start CPU profiling: %v\n", msg)
+		}
 	} else {
 		cpuProfilingStarted = true
 		defer func() {
@@ -401,7 +391,7 @@ func runForDuration(cmd *cobra.Command, args []string) error {
 		}
 	}
 
-	if stateWasDown {
+	if stateWasDown && !noUpDown {
 		if _, err := client.Down(cmd.Context(), &proto.DownRequest{}); err != nil {
 			cmd.PrintErrf("Failed to restore service down state: %v\n", status.Convert(err).Message())
 		} else {
@@ -458,6 +448,47 @@ func setSyncResponsePersistence(cmd *cobra.Command, args []string) error {
 	return nil
 }
 
+// enableSyncResponsePersistence asks the daemon to keep the latest sync
+// response so the bundle carries the network map. With a running daemon only
+// syncs received after the call are kept.
+func enableSyncResponsePersistence(cmd *cobra.Command, client proto.DaemonServiceClient) {
+	if _, err := client.SetSyncResponsePersistence(cmd.Context(), &proto.SetSyncResponsePersistenceRequest{
+		Enabled: true,
+	}); err != nil {
+		cmd.PrintErrf("Failed to enable sync response persistence: %v\n", status.Convert(err).Message())
+	}
+}
+
+// restartDaemon cycles the daemon down and up with sync response persistence
+// enabled so the bundle carries the network map. It reports whether the
+// daemon was left down although it was running before, so the caller can
+// bring it back up.
+func restartDaemon(cmd *cobra.Command, client proto.DaemonServiceClient, stateWasDown bool) bool {
+	needsRestoreUp := false
+	if _, err := client.Down(cmd.Context(), &proto.DownRequest{}); err != nil {
+		cmd.PrintErrf("Failed to bring service down: %v\n", status.Convert(err).Message())
+	} else {
+		needsRestoreUp = !stateWasDown
+		cmd.Println("netbird down")
+	}
+
+	time.Sleep(1 * time.Second)
+
+	// Enable sync response persistence before bringing the service up
+	enableSyncResponsePersistence(cmd, client)
+
+	if _, err := client.Up(cmd.Context(), &proto.UpRequest{}); err != nil {
+		cmd.PrintErrf("Failed to bring service up: %v\n", status.Convert(err).Message())
+	} else {
+		needsRestoreUp = false
+		cmd.Println("netbird up")
+	}
+
+	time.Sleep(3 * time.Second)
+
+	return needsRestoreUp
+}
+
 func waitForDurationOrCancel(ctx context.Context, duration time.Duration, cmd *cobra.Command) error {
 	ticker := time.NewTicker(1 * time.Second)
 	defer ticker.Stop()
@@ -546,4 +577,5 @@ func init() {
 	forCmd.Flags().StringVar(&uploadBundleURLFlag, "upload-bundle-url", types.DefaultBundleURL, "Service URL to get an URL to upload the debug bundle")
 	forCmd.Flags().BoolVar(&uploadBundleInsecureFlag, "upload-bundle-insecure", false, "Allow uploading to an http or untrusted-TLS upload server (self-hosted); requires root")
 	forCmd.Flags().Bool("capture", false, "Capture packets during the debug duration and include in bundle")
+	forCmd.Flags().Bool(noUpDownFlag, false, "Keep the daemon running instead of bringing it down and up before collecting. The bundle only includes the network map if a sync arrives during the run")
 }

client/cmd/debug_cpu.go

@@ -0,0 +1,83 @@
+package cmd
+
+import (
+	"fmt"
+
+	log "github.com/sirupsen/logrus"
+	"github.com/spf13/cobra"
+	"google.golang.org/grpc/status"
+
+	"github.com/netbirdio/netbird/client/proto"
+)
+
+var debugCPUCmd = &cobra.Command{
+	Use:   "cpu",
+	Short: "Profile the daemon's CPU usage",
+	Long: `Starts and stops CPU profiling in the running daemon without restarting it.
+The profile is included in the next debug bundle as cpu.prof.
+
+Profiling is not time limited: it keeps running, and keeps costing CPU, until
+"netbird debug cpu stop" is run.`,
+}
+
+var debugCPUStartCmd = &cobra.Command{
+	Use:     "start",
+	Short:   "Start CPU profiling in the daemon",
+	Example: "  netbird debug cpu start",
+	Args:    cobra.NoArgs,
+	RunE:    debugCPUStart,
+}
+
+var debugCPUStopCmd = &cobra.Command{
+	Use:   "stop",
+	Short: "Stop CPU profiling in the daemon",
+	Long: `Stops CPU profiling. The captured profile stays in the daemon until the next
+debug bundle is created, which includes it as cpu.prof.`,
+	Example: "  netbird debug cpu stop && netbird debug bundle",
+	Args:    cobra.NoArgs,
+	RunE:    debugCPUStop,
+}
+
+func debugCPUStart(cmd *cobra.Command, _ []string) error {
+	conn, err := getClient(cmd)
+	if err != nil {
+		return err
+	}
+	defer func() {
+		if err := conn.Close(); err != nil {
+			log.Errorf(errCloseConnection, err)
+		}
+	}()
+
+	if _, err := proto.NewDaemonServiceClient(conn).StartCPUProfile(cmd.Context(), &proto.StartCPUProfileRequest{}); err != nil {
+		return fmt.Errorf("start CPU profiling: %v", status.Convert(err).Message())
+	}
+
+	cmd.Println("CPU profiling started and runs until stopped. Run `netbird debug cpu stop` and then `netbird debug bundle` to collect it.")
+	return nil
+}
+
+func debugCPUStop(cmd *cobra.Command, _ []string) error {
+	conn, err := getClient(cmd)
+	if err != nil {
+		return err
+	}
+	defer func() {
+		if err := conn.Close(); err != nil {
+			log.Errorf(errCloseConnection, err)
+		}
+	}()
+
+	if _, err := proto.NewDaemonServiceClient(conn).StopCPUProfile(cmd.Context(), &proto.StopCPUProfileRequest{}); err != nil {
+		return fmt.Errorf("stop CPU profiling: %v", status.Convert(err).Message())
+	}
+
+	cmd.Println("CPU profiling stopped. Run `netbird debug bundle` to include cpu.prof.")
+	return nil
+}
+
+func init() {
+	debugCPUCmd.AddCommand(debugCPUStartCmd)
+	debugCPUCmd.AddCommand(debugCPUStopCmd)
+	debugCmd.AddCommand(debugCPUCmd)
+}

client/cmd/debug_cpu_test.go

@@ -0,0 +1,164 @@
+package cmd
+
+import (
+	"bytes"
+	"context"
+	"os/user"
+	"strings"
+	"testing"
+
+	"github.com/spf13/cobra"
+	"github.com/spf13/pflag"
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+
+	"github.com/netbirdio/netbird/client/internal"
+	"github.com/netbirdio/netbird/client/internal/profilemanager"
+)
+
+// startDebugTestDaemon starts an in-process daemon with an isolated profile
+// directory and returns the address the CLI should dial.
+func startDebugTestDaemon(t *testing.T) string {
+	t.Helper()
+
+	tempDir := t.TempDir()
+	origDefaultProfileDir := profilemanager.DefaultConfigPathDir
+	origActiveProfileStatePath := profilemanager.ActiveProfileStatePath
+	origConfigDirOverride := profilemanager.ConfigDirOverride
+	origDaemonAddr := daemonAddr
+	t.Cleanup(func() {
+		profilemanager.DefaultConfigPathDir = origDefaultProfileDir
+		profilemanager.ActiveProfileStatePath = origActiveProfileStatePath
+		profilemanager.ConfigDirOverride = origConfigDirOverride
+		daemonAddr = origDaemonAddr
+	})
+
+	profilemanager.DefaultConfigPathDir = tempDir
+	profilemanager.ActiveProfileStatePath = tempDir + "/active_profile.json"
+	profilemanager.ConfigDirOverride = tempDir
+
+	currUser, err := user.Current()
+	require.NoError(t, err)
+	sm := profilemanager.ServiceManager{}
+	created, err := sm.AddProfile("test1", currUser.Username)
+	require.NoError(t, err)
+	require.NoError(t, sm.SetActiveProfileState(&profilemanager.ActiveProfileState{
+		ID:       created.ID,
+		Username: currUser.Username,
+	}))
+
+	ctx, cancel := context.WithCancel(internal.CtxInitState(context.Background()))
+	srv, lis := startClientDaemon(t, ctx, "", tempDir+"/config.json")
+	t.Cleanup(func() {
+		cancel()
+		srv.Stop()
+	})
+
+	return "tcp://" + lis.Addr().String()
+}
+
+// runDebugCmd runs `netbird debug <args>` against the daemon at addr and
+// returns everything the command printed.
+func runDebugCmd(addr string, args ...string) (string, error) {
+	daemonAddr = addr
+	var out bytes.Buffer
+	rootCmd.SetOut(&out)
+	rootCmd.SetErr(&out)
+	rootCmd.SetArgs(append(append([]string{"debug"}, args...), "--daemon-addr", addr, "--log-file", ""))
+	err := rootCmd.Execute()
+	rootCmd.SetOut(nil)
+	rootCmd.SetErr(nil)
+	rootCmd.SetArgs(nil)
+	resetFlags(rootCmd)
+	return out.String(), err
+}
+
+// resetFlags puts every flag of the command and its subcommands back to its
+// default so a value parsed in one run does not leak into the next in-process
+// execution.
+func resetFlags(cmd *cobra.Command) {
+	reset := func(f *pflag.Flag) {
+		// Set appends to a slice flag and would parse the "[a,b]" default
+		// text as elements, so slices are replaced instead.
+		if sv, ok := f.Value.(pflag.SliceValue); ok {
+			var def []string
+			if trimmed := strings.Trim(f.DefValue, "[]"); trimmed != "" {
+				def = strings.Split(trimmed, ",")
+			}
+			_ = sv.Replace(def)
+		} else {
+			_ = f.Value.Set(f.DefValue)
+		}
+		f.Changed = false
+	}
+	cmd.Flags().VisitAll(reset)
+	cmd.PersistentFlags().VisitAll(reset)
+	// Commands pin their writers to the buffer of the run that first used
+	// them, so a later run would print into the old buffer.
+	cmd.SetOut(nil)
+	cmd.SetErr(nil)
+	for _, sub := range cmd.Commands() {
+		resetFlags(sub)
+	}
+}
+
+// TestResetFlagsSliceDefault guards against Set("[]") on slice flags, which
+// stores a literal "[]" element instead of the empty default.
+func TestResetFlagsSliceDefault(t *testing.T) {
+	cmd := &cobra.Command{Use: "x"}
+	var env, withDefault []string
+	cmd.Flags().StringSliceVar(&env, "env", nil, "")
+	cmd.Flags().StringSliceVar(&withDefault, "names", []string{"a", "b"}, "")
+	require.NoError(t, cmd.Flags().Parse([]string{"--env", "K=V", "--names", "c"}))
+
+	resetFlags(cmd)
+
+	assert.Empty(t, env, "slice flag with no default must reset to empty")
+	assert.Equal(t, []string{"a", "b"}, withDefault, "slice flag must reset to its default")
+}
+
+func TestDebugCPUStartStop(t *testing.T) {
+	addr := startDebugTestDaemon(t)
+
+	run := func(args ...string) error {
+		_, err := runDebugCmd(addr, append([]string{"cpu"}, args...)...)
+		return err
+	}
+
+	require.Error(t, run("stop"), "stop without a running profile must fail")
+	require.NoError(t, run("start"))
+	assert.Error(t, run("start"), "second start must be rejected while profiling")
+	require.NoError(t, run("stop"))
+	assert.Error(t, run("stop"), "second stop must be rejected")
+	assert.NoError(t, run("start"), "profiling can be started again after a stop")
+	assert.NoError(t, run("stop"))
+}
+
+// TestDebugForKeepsRunningCPUProfile covers `debug for` started while a
+// profile from `debug cpu start` is running: it must say so, leave the
+// profile alone, and still create the bundle.
+func TestDebugForKeepsRunningCPUProfile(t *testing.T) {
+	addr := startDebugTestDaemon(t)
+
+	_, err := runDebugCmd(addr, "cpu", "start")
+	require.NoError(t, err)
+
+	out, err := runDebugCmd(addr, "for", "1s", "-S=false", "--no-updown")
+	require.NoError(t, err, "output: %s", out)
+	assert.Contains(t, out, "CPU profiling is already running", "the conflict must be explained")
+	assert.NotContains(t, out, "rpc error", "the raw RPC error must not reach the user")
+	assert.Contains(t, out, "Local file:", "the bundle must still be created")
+
+	_, err = runDebugCmd(addr, "cpu", "stop")
+	assert.NoError(t, err, "the profile started by the user must still be running")
+}
+
+func TestDebugForNoUpDown(t *testing.T) {
+	addr := startDebugTestDaemon(t)
+
+	out, err := runDebugCmd(addr, "for", "1s", "-S=false", "--no-updown")
+	require.NoError(t, err, "output: %s", out)
+	assert.NotContains(t, out, "netbird down", "--no-updown must not bring the daemon down")
+	assert.NotContains(t, out, "netbird up", "--no-updown must not bring the daemon up")
+	assert.Contains(t, out, "Local file:", "the bundle must still be created")
+}

client/cmd/service.go

@@ -7,6 +7,7 @@ import (
 	"fmt"
 	"net/http"
 	"runtime"
+	"slices"
 	"strings"
 	"sync"
 
@@ -25,6 +26,30 @@ var serviceCmd = &cobra.Command{
 
 const defaultJSONSocket = "unix:///var/run/netbird-http.sock"
 
+// forbiddenServiceEnvVars are the environment variables the service is never
+// registered with, keyed in upper case since these are Windows names. Each one
+// decides where the daemon resolves something it then uses with the privileges
+// of the account it runs under — LocalSystem on Windows, root elsewhere: the
+// executables it runs (PATH, PATHEXT, COMSPEC, SystemRoot, windir) or the
+// directory it writes temporary files in (TEMP, TMP). The daemon needs none of
+// them, and the utilities it shells out to are resolved by absolute path.
+var forbiddenServiceEnvVars = map[string]struct{}{
+	"PATH":       {},
+	"PATHEXT":    {},
+	"SYSTEMROOT": {},
+	"WINDIR":     {},
+	"COMSPEC":    {},
+	"TEMP":       {},
+	"TMP":        {},
+}
+
+// forbiddenServiceEnvPrefixes are the dynamic-loader families, refused whole
+// rather than by name: LD_PRELOAD, DYLD_INSERT_LIBRARIES and their siblings all
+// reach the loader of the process, the set differs per platform and libc, and
+// new members arrive with new OS releases. Listing them one by one is a list
+// that is wrong the moment it is written.
+var forbiddenServiceEnvPrefixes = []string{"LD_", "DYLD_"}
+
 var (
 	serviceName      string
 	serviceEnvVars   []string
@@ -127,8 +152,33 @@ func parseServiceEnvVars(envVars []string) (map[string]string, error) {
 			return nil, fmt.Errorf("empty environment variable key in: %s", env)
 		}
 
+		if isForbiddenServiceEnvVar(key) {
+			return nil, fmt.Errorf("environment variable %s cannot be set on the service: it decides where the service resolves the executables, libraries or temporary files it uses", key)
+		}
+
 		envMap[key] = value
 	}
 
 	return envMap, nil
 }
+
+// isForbiddenServiceEnvVar reports whether name is one the service must not be
+// registered with.
+//
+// The names are matched case-insensitively only on Windows, where they are the
+// same variable however they are spelled. Elsewhere the environment is
+// case-sensitive, so Path and PATH are two different variables and only the
+// exact spelling is the one the loader reads.
+func isForbiddenServiceEnvVar(name string) bool {
+	if runtime.GOOS == "windows" {
+		name = strings.ToUpper(name)
+	}
+
+	if _, forbidden := forbiddenServiceEnvVars[name]; forbidden {
+		return true
+	}
+
+	return slices.ContainsFunc(forbiddenServiceEnvPrefixes, func(prefix string) bool {
+		return strings.HasPrefix(name, prefix)
+	})
+}

client/cmd/service_params.go

@@ -14,6 +14,7 @@ import (
 
 	"github.com/netbirdio/netbird/client/configs"
 	"github.com/netbirdio/netbird/client/internal/daemonaddr"
+	"github.com/netbirdio/netbird/client/internal/elevate"
 	"github.com/netbirdio/netbird/util"
 )
 
@@ -43,10 +44,33 @@ func serviceParamsPath() string {
 
 // loadServiceParams reads saved service parameters from disk.
 // Returns nil with no error if the file does not exist.
+//
+// The file is read by an elevated install and decides the arguments and the
+// environment of the service it then registers, so it is used only when its
+// ownership and permissions are the ones saveServiceParams leaves behind. That
+// restricted ACL is applied when the file is written, which is not necessarily
+// before it is first read, so this is checked rather than assumed. A file that
+// fails the check is treated as absent, and the install proceeds with its
+// defaults.
 func loadServiceParams() (*serviceParams, error) {
 	path := serviceParamsPath()
 
-	data, err := os.ReadFile(path)
+	// Resolve links first so the checks apply to the file that is actually read.
+	// Since the check covers every directory above it as well, nobody who fails
+	// it can swap the file between here and the read below.
+	resolved, err := filepath.EvalSymlinks(path)
+	if err != nil {
+		if os.IsNotExist(err) {
+			return nil, nil //nolint:nilnil
+		}
+		return nil, fmt.Errorf("resolve service params %s: %w", path, err)
+	}
+
+	if err := elevate.CheckOnlyOwnerWritable(resolved); err != nil {
+		return nil, fmt.Errorf("refusing to read service params from %s: %w", resolved, err)
+	}
+
+	data, err := os.ReadFile(resolved)
 	if err != nil {
 		if os.IsNotExist(err) {
 			return nil, nil //nolint:nilnil
@@ -182,10 +206,16 @@ func applyServiceParams(cmd *cobra.Command, params *serviceParams) {
 // If --service-env was explicitly set to empty, all saved env vars are cleared.
 // If --service-env was not set, saved env vars are used entirely.
 func applyServiceEnvParams(cmd *cobra.Command, params *serviceParams) {
+	// A forbidden name explicitly passed on the command line is an error the
+	// operator is told about, but one restored from a file written by an older
+	// version is dropped: an install that refuses to run would leave the host
+	// without a daemon over a variable nobody is asking for any more.
+	saved := dropForbiddenServiceEnvVars(cmd, params.ServiceEnvVars)
+
 	if !cmd.Flags().Changed("service-env") {
-		if len(params.ServiceEnvVars) > 0 {
+		if len(saved) > 0 {
 			// No explicit env vars: rebuild serviceEnvVars from saved params.
-			serviceEnvVars = envMapToSlice(params.ServiceEnvVars)
+			serviceEnvVars = envMapToSlice(saved)
 		}
 		return
 	}
@@ -204,13 +234,13 @@ func applyServiceEnvParams(cmd *cobra.Command, params *serviceParams) {
 		return
 	}
 
-	if len(params.ServiceEnvVars) == 0 {
+	if len(saved) == 0 {
 		return
 	}
 
 	// Merge saved values underneath explicit ones.
-	merged := make(map[string]string, len(params.ServiceEnvVars)+len(explicit))
-	maps.Copy(merged, params.ServiceEnvVars)
+	merged := make(map[string]string, len(saved)+len(explicit))
+	maps.Copy(merged, saved)
 	maps.Copy(merged, explicit) // explicit wins on conflict
 	serviceEnvVars = envMapToSlice(merged)
 }
@@ -233,6 +263,20 @@ var resetParamsCmd = &cobra.Command{
 	},
 }
 
+// dropForbiddenServiceEnvVars returns the saved entries that may still be
+// registered on the service, reporting every one it leaves behind.
+func dropForbiddenServiceEnvVars(cmd *cobra.Command, saved map[string]string) map[string]string {
+	kept := make(map[string]string, len(saved))
+	for key, value := range saved {
+		if isForbiddenServiceEnvVar(key) {
+			cmd.PrintErrf("Warning: ignoring saved service environment variable %s: it decides where the service resolves the executables, libraries or temporary files it uses\n", key)
+			continue
+		}
+		kept[key] = value
+	}
+	return kept
+}
+
 // envMapToSlice converts a map of env vars to a KEY=VALUE slice.
 func envMapToSlice(m map[string]string) []string {
 	s := make([]string, 0, len(m))

client/cmd/service_params_test.go

@@ -9,6 +9,7 @@ import (
 	"go/token"
 	"os"
 	"path/filepath"
+	"runtime"
 	"strings"
 	"testing"
 
@@ -353,6 +354,59 @@ func TestApplyServiceEnvParams_NotChanged(t *testing.T) {
 	assert.Equal(t, map[string]string{"FROM_SAVED": "val"}, result)
 }
 
+func TestParseServiceEnvVars_RejectsForbiddenNames(t *testing.T) {
+	for _, env := range []string{"PATH=C:\\somewhere", "LD_PRELOAD=/tmp/lib.so", "DYLD_FALLBACK_LIBRARY_PATH=/tmp"} {
+		_, err := parseServiceEnvVars([]string{"KEEP=me", env})
+		require.Errorf(t, err, "%s selects what the service resolves and must be refused", env)
+	}
+}
+
+func TestIsForbiddenServiceEnvVar(t *testing.T) {
+	// The loader families are matched by prefix, so a name nobody has heard of
+	// yet is refused too.
+	for _, name := range []string{
+		"PATH", "PATHEXT", "COMSPEC", "SYSTEMROOT", "WINDIR", "TEMP", "TMP",
+		"LD_PRELOAD", "LD_AUDIT", "DYLD_INSERT_LIBRARIES", "DYLD_FALLBACK_FRAMEWORK_PATH",
+	} {
+		assert.Truef(t, isForbiddenServiceEnvVar(name), "%s must be refused", name)
+	}
+
+	// The prefix must not swallow names that merely start with the same letters.
+	for _, name := range []string{"NB_LOG_LEVEL", "NB_WG_DEBUG", "HTTPS_PROXY", "LDAP_URL", "DYLDX"} {
+		assert.Falsef(t, isForbiddenServiceEnvVar(name), "%s has no reason to be refused", name)
+	}
+
+	// On Windows a variable is the same one however it is spelled; elsewhere
+	// Path and PATH are two variables and only the exact one is read.
+	if runtime.GOOS == "windows" {
+		assert.True(t, isForbiddenServiceEnvVar("Path"))
+		assert.True(t, isForbiddenServiceEnvVar("ld_preload"))
+	} else {
+		assert.False(t, isForbiddenServiceEnvVar("Path"))
+		assert.False(t, isForbiddenServiceEnvVar("ld_preload"))
+	}
+}
+
+func TestApplyServiceEnvParams_DropsForbiddenSavedNames(t *testing.T) {
+	origServiceEnvVars := serviceEnvVars
+	t.Cleanup(func() { serviceEnvVars = origServiceEnvVars })
+
+	serviceEnvVars = nil
+
+	cmd := &cobra.Command{}
+	cmd.Flags().StringSlice("service-env", nil, "")
+
+	saved := &serviceParams{
+		ServiceEnvVars: map[string]string{"PATH": "C:\\attacker", "NB_LOG_FORMAT": "json"},
+	}
+
+	applyServiceEnvParams(cmd, saved)
+
+	result, err := parseServiceEnvVars(serviceEnvVars)
+	require.NoError(t, err, "a saved PATH must be dropped rather than fail the install")
+	assert.Equal(t, map[string]string{"NB_LOG_FORMAT": "json"}, result)
+}
+
 func TestApplyServiceEnvParams_ExplicitEmptyClears(t *testing.T) {
 	origServiceEnvVars := serviceEnvVars
 	t.Cleanup(func() { serviceEnvVars = origServiceEnvVars })

client/cmd/service_params_trust_test.go

@@ -0,0 +1,57 @@
+//go:build !windows && !ios && !android
+
+package cmd
+
+import (
+	"os"
+	"path/filepath"
+	"testing"
+
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+
+	"github.com/netbirdio/netbird/client/configs"
+)
+
+// The Windows equivalent of this is the ACL check in
+// elevate.CheckOnlyOwnerWritable, covered by that package's own tests; here the
+// point is that loadServiceParams asks the question at all.
+func TestLoadServiceParams_RefusesWorldWritableFile(t *testing.T) {
+	tmpDir := t.TempDir()
+
+	original := configs.StateDir
+	t.Cleanup(func() { configs.StateDir = original })
+	configs.StateDir = tmpDir
+
+	path := filepath.Join(tmpDir, serviceParamsFile)
+	require.NoError(t, os.WriteFile(path, []byte(`{"log_level":"debug"}`), 0o666))
+	// WriteFile is subject to the umask, so set the bits that matter explicitly.
+	require.NoError(t, os.Chmod(path, 0o666))
+
+	params, err := loadServiceParams()
+	require.Error(t, err, "a service.json anyone can rewrite must not be trusted")
+	assert.Nil(t, params)
+
+	require.NoError(t, os.Chmod(path, 0o600))
+	params, err = loadServiceParams()
+	require.NoError(t, err)
+	require.NotNil(t, params)
+	assert.Equal(t, "debug", params.LogLevel)
+}
+
+func TestLoadServiceParams_RefusesWorldWritableDirectory(t *testing.T) {
+	tmpDir := t.TempDir()
+	stateDir := filepath.Join(tmpDir, "state")
+	require.NoError(t, os.Mkdir(stateDir, 0o777))
+	require.NoError(t, os.Chmod(stateDir, 0o777))
+
+	original := configs.StateDir
+	t.Cleanup(func() { configs.StateDir = original })
+	configs.StateDir = stateDir
+
+	require.NoError(t, os.WriteFile(filepath.Join(stateDir, serviceParamsFile), []byte(`{}`), 0o600))
+
+	params, err := loadServiceParams()
+	require.Error(t, err, "a service.json in a directory anyone can replace entries in must not be trusted")
+	assert.Nil(t, params)
+}