add custom dialer

- Validate resolved IPs immediately before connecting.
- Guard Shoutrrr requests during initialization and delivery.
- Require admins for services without HTTP client support.
- Test DNS rebinding, redirects, internal hosts, and authorization.

SECURITY.md

@@ -2,6 +2,8 @@
 
 ## Reporting a Vulnerability
 
-If you find a vulnerability in the latest version, please [submit a private advisory](https://github.com/henrygd/beszel/security/advisories/new).
+**PLEASE ONLY USE SECURITY ADVISORIES FOR REAL HIGH SEVERITY VULNERABILITIES.**
 
-If it's low severity (use best judgement) you may open an issue instead of an advisory.
+If you find a vulnerability in the latest version, and it is not high severity, open an issue instead of an advisory.
+
+I am overwhelmed with advisories, often erroneous, which are clearly found and written by AI. I don't have the capacity to review all of them.

agent/client.go

@@ -25,7 +25,9 @@ import (
 )
 
 const (
-	wsDeadline = 70 * time.Second
+	// Keep the connection alive long enough for a slow collection cycle to
+	// finish before the hub considers the agent disconnected.
+	wsDeadline = 120 * time.Second
 )
 
 type caCertFileError struct {

agent/client_test.go

@@ -700,3 +700,11 @@ func TestGetToken(t *testing.T) {
 		assert.Equal(t, expectedToken, token, "Whitespace should be stripped from token file content")
 	})
 }
+
+func TestWebSocketDeadlineCoversSlowCollection(t *testing.T) {
+	const minimumDeadline = 120 * time.Second
+
+	if wsDeadline < minimumDeadline {
+		t.Fatalf("WebSocket deadline %s is shorter than the slow-collection window of %s", wsDeadline, minimumDeadline)
+	}
+}

go.mod

@@ -10,7 +10,7 @@ require (
 	github.com/gliderlabs/ssh v0.3.8
 	github.com/google/uuid v1.6.0
 	github.com/lxzan/gws v1.10.1
-	github.com/nicholas-fedor/shoutrrr v0.19.0
+	github.com/nicholas-fedor/shoutrrr v0.20.0
 	github.com/pocketbase/dbx v1.12.0
 	github.com/pocketbase/pocketbase v0.40.2
 	github.com/shirou/gopsutil/v4 v4.26.8

go.sum

@@ -54,8 +54,8 @@ github.com/golang-jwt/jwt/v5 v5.3.1/go.mod h1:fxCRLWMO43lRc8nhHWY6LGqRcf+1gQWArs
 github.com/golang/protobuf v1.3.1/go.mod h1:6lQm79b+lXiMfvg/cZm0SGofjICqVBUtrP5yJMmIC1U=
 github.com/google/go-cmp v0.7.0 h1:wk8382ETsv4JYUZwIsn6YpYiWiBsYLSJiTsyBybVuN8=
 github.com/google/go-cmp v0.7.0/go.mod h1:pXiqmnSA92OHEEa9HXL2W4E7lf9JzCmGVUdgjX3N/iU=
-github.com/google/pprof v0.0.0-20260902005441-ca85771921e4 h1:/6mPXfWmhv8eKck12I0YNIcIjwHtxP3YRIMKiEgTjWg=
-github.com/google/pprof v0.0.0-20260902005441-ca85771921e4/go.mod h1:jl5iWTm0/hd5PjEYEOuwAJ57L/CibdZfrqZ5XA5GrCk=
+github.com/google/pprof v0.0.0-20260906184651-6331bc6350fe h1:QAinXoAFJdGQYztXn3VpFey7KCwpedbZ/EkzbplQ0cY=
+github.com/google/pprof v0.0.0-20260906184651-6331bc6350fe/go.mod h1:jl5iWTm0/hd5PjEYEOuwAJ57L/CibdZfrqZ5XA5GrCk=
 github.com/google/uuid v1.6.0 h1:NIvaJDMOsjHA8n1jAhLSgzrAzy1Hgr+hNrb57e+94F0=
 github.com/google/uuid v1.6.0/go.mod h1:TIyPZe4MgqvfeYDBFedMoGGpEw/LqOeaOT+nhxU+yHo=
 github.com/gorilla/websocket v1.5.3 h1:saDtZ6Pbx/0u+bgYQ3q96pZgCzfhKXGPqt7kZ72aNNg=
@@ -83,8 +83,8 @@ github.com/mattn/go-isatty v0.0.24 h1:tGZZoVgT/KiqK1c8ocVLeDS8BSWMRd47J3Lbz7vsRe
 github.com/mattn/go-isatty v0.0.24/go.mod h1:nMCL3Zebbrt45jsMDgnfIwz6ydEQApk5oEI3HqDio6A=
 github.com/ncruces/go-strftime v1.0.0 h1:HMFp8mLCTPp341M/ZnA4qaf7ZlsbTc+miZjCLOFAw7w=
 github.com/ncruces/go-strftime v1.0.0/go.mod h1:Fwc5htZGVVkseilnfgOVb9mKy6w1naJmn9CehxcKcls=
-github.com/nicholas-fedor/shoutrrr v0.19.0 h1:Rl6bpK3DXuR2Trtx2JV8t+wjUwkHdRHrc8nBKoEpHr0=
-github.com/nicholas-fedor/shoutrrr v0.19.0/go.mod h1:Glfdi8AGTbnEn2k2+hW62n8oL0i9vqRVFtXaUIthNks=
+github.com/nicholas-fedor/shoutrrr v0.20.0 h1:hMAxIYlfAeZ1FcTDgU0kUOvVXUsOirWo8IWlnzGLkac=
+github.com/nicholas-fedor/shoutrrr v0.20.0/go.mod h1:hgde37yNWCXh8+N6WemyDRMNYLOFTf326GsBx8Z7CFA=
 github.com/onsi/ginkgo/v2 v2.32.1 h1:6tlvcDm/3sE8lGJbZ4+d4mO3RLy24/tQWOFzVSQNIfw=
 github.com/onsi/ginkgo/v2 v2.32.1/go.mod h1:+aXOY+vzZ5mu2iI2HpTZUPmM//oQfsNFX6gU9kNcA44=
 github.com/onsi/gomega v1.43.0 h1:VlG/1FxqNxhSO+lq/OHBNaaqwiBK/mO8JbVkX9Y+FeU=

internal/alerts/alerts.go

@@ -20,10 +20,10 @@ type hubLike interface {
 }
 
 type AlertManager struct {
-	hub                    hubLike
-	stopOnce               sync.Once
-	pendingAlerts          sync.Map
-	alertsCache            *AlertsCache
+	hub           hubLike
+	stopOnce      sync.Once
+	pendingAlerts sync.Map
+	alertsCache   *AlertsCache
 }
 
 type AlertMessageData struct {
@@ -231,8 +231,20 @@ func (am *AlertManager) SendAlert(data AlertMessageData) error {
 		am.hub.Logger().Error("Failed to unmarshal user settings", "err", err)
 	}
 	// send alerts via webhooks
+	send := sendPublicNotification
+	if len(userAlertSettings.Webhooks) > 0 {
+		// Read the owner's current role at delivery time, including for URLs
+		// saved before an admin was demoted. Never fall back on lookup failure.
+		owner, err := am.hub.FindRecordById("users", data.UserID)
+		if err != nil {
+			return fmt.Errorf("load notification owner: %w", err)
+		}
+		if owner.GetString("role") == "admin" {
+			send = shoutrrr.Send
+		}
+	}
 	for _, webhook := range userAlertSettings.Webhooks {
-		if err := am.SendShoutrrrAlert(webhook, data.Title, data.Message, data.Link, data.LinkText); err != nil {
+		if err := am.sendShoutrrrAlert(webhook, data.Title, data.Message, data.Link, data.LinkText, send); err != nil {
 			am.hub.Logger().Error("Failed to send shoutrrr alert", "err", err)
 		}
 	}
@@ -263,6 +275,10 @@ func (am *AlertManager) SendAlert(data AlertMessageData) error {
 
 // SendShoutrrrAlert sends an alert via a Shoutrrr URL
 func (am *AlertManager) SendShoutrrrAlert(notificationUrl, title, message, link, linkText string) error {
+	return am.sendShoutrrrAlert(notificationUrl, title, message, link, linkText, shoutrrr.Send)
+}
+
+func (am *AlertManager) sendShoutrrrAlert(notificationUrl, title, message, link, linkText string, send func(string, string) error) error {
 	// Parse the URL
 	parsedURL, err := url.Parse(notificationUrl)
 	if err != nil {
@@ -305,7 +321,7 @@ func (am *AlertManager) SendShoutrrrAlert(notificationUrl, title, message, link,
 	parsedURL.RawQuery = queryParams.Encode()
 	// log.Println("URL after modification:", parsedURL.String())
 
-	err = shoutrrr.Send(parsedURL.String(), message)
+	err = send(parsedURL.String(), message)
 
 	if err == nil {
 		am.hub.Logger().Info("Sent shoutrrr alert", "title", title)

internal/alerts/alerts_api.go

@@ -3,13 +3,11 @@ package alerts
 import (
 	"database/sql"
 	"errors"
-	"net"
 	"net/http"
-	"net/url"
 	"slices"
-	"strings"
 
 	"github.com/henrygd/beszel/internal/hub/utils"
+	"github.com/nicholas-fedor/shoutrrr"
 	"github.com/pocketbase/dbx"
 	"github.com/pocketbase/pocketbase/core"
 )
@@ -147,72 +145,16 @@ func (am *AlertManager) SendTestNotification(e *core.RequestEvent) error {
 	if err != nil || data.URL == "" {
 		return e.BadRequestError("URL is required", err)
 	}
-	// Only allow admins to send test notifications to internal URLs
+	send := shoutrrr.Send
 	if !e.Auth.IsSuperuser() && e.Auth.GetString("role") != "admin" {
-		internalURL, err := isInternalURL(data.URL)
-		if err != nil {
-			return e.BadRequestError(err.Error(), nil)
-		}
-		if internalURL {
-			return e.ForbiddenError("Only admins can send to internal destinations", nil)
-		}
+		send = sendPublicNotification
+	}
+	err = am.sendShoutrrrAlert(data.URL, "Test Alert", "This is a notification from Beszel.", am.hub.Settings().Meta.AppURL, "View Beszel", send)
+	if errors.Is(err, errInternalDestination) || errors.Is(err, errUnrestrictedService) {
+		return e.ForbiddenError(err.Error(), nil)
 	}
-	err = am.SendShoutrrrAlert(data.URL, "Test Alert", "This is a notification from Beszel.", am.hub.Settings().Meta.AppURL, "View Beszel")
 	if err != nil {
 		return e.JSON(200, map[string]string{"err": err.Error()})
 	}
 	return e.JSON(200, map[string]bool{"err": false})
 }
-
-// isInternalURL checks if the given shoutrrr URL points to an internal destination (localhost or private IP)
-func isInternalURL(rawURL string) (bool, error) {
-	parsedURL, err := url.Parse(rawURL)
-	if err != nil {
-		return false, err
-	}
-
-	host := parsedURL.Hostname()
-	if host == "" {
-		return false, nil
-	}
-
-	if strings.EqualFold(host, "localhost") {
-		return true, nil
-	}
-
-	if ip := net.ParseIP(host); ip != nil {
-		return isInternalIP(ip), nil
-	}
-
-	// Some Shoutrrr URLs use the host position for service identifiers rather than a
-	// network hostname (for example, discord://token@webhookid). Restrict DNS lookups
-	// to names that look like actual hostnames so valid service URLs keep working.
-	if !strings.Contains(host, ".") {
-		return false, nil
-	}
-
-	ips, err := net.LookupIP(host)
-	if err != nil {
-		return false, nil
-	}
-
-	if slices.ContainsFunc(ips, isInternalIP) {
-		return true, nil
-	}
-
-	return false, nil
-}
-
-var cgnatNetwork = &net.IPNet{
-	IP:   net.IPv4(100, 64, 0, 0),
-	Mask: net.CIDRMask(10, 32),
-}
-
-func isInternalIP(ip net.IP) bool {
-	return ip.IsPrivate() ||
-		ip.IsLoopback() ||
-		ip.IsUnspecified() ||
-		ip.IsLinkLocalUnicast() ||
-		ip.IsMulticast() ||
-		cgnatNetwork.Contains(ip)
-}

internal/alerts/alerts_api_test.go

@@ -7,10 +7,11 @@ import (
 	"encoding/json"
 	"io"
 	"net/http"
+	"net/http/httptest"
 	"strings"
+	"sync/atomic"
 	"testing"
 
-	"github.com/henrygd/beszel/internal/alerts"
 	beszelTests "github.com/henrygd/beszel/internal/tests"
 	pbTests "github.com/pocketbase/pocketbase/tests"
 
@@ -29,43 +30,6 @@ func jsonReader(v any) io.Reader {
 	return bytes.NewReader(data)
 }
 
-func TestIsInternalURL(t *testing.T) {
-	testCases := []struct {
-		name     string
-		url      string
-		internal bool
-	}{
-		{name: "loopback ipv4", url: "generic://127.0.0.1", internal: true},
-		{name: "private ipv4", url: "generic://10.0.0.1", internal: true},
-		{name: "localhost hostname", url: "generic://localhost", internal: true},
-		{name: "localhost with path", url: "generic+http://localhost/api/v1/postStuff", internal: true},
-		{name: "loopback with port and path", url: "generic+http://127.0.0.1:8080/api/v1/postStuff", internal: true},
-		{name: "public hostname", url: "generic+https://beszel.dev/api/v1/postStuff", internal: false},
-		{name: "cloud metadata ipv4", url: "generic://169.254.169.254", internal: true},
-		{name: "link-local ipv4", url: "generic://169.254.1.1", internal: true},
-		{name: "link-local ipv6", url: "generic://[fe80::1]", internal: true},
-		{name: "mapped link-local ipv4", url: "generic://[::ffff:169.254.169.254]", internal: true},
-		{name: "cgnat lower boundary", url: "generic://100.64.0.0", internal: true},
-		{name: "cgnat upper boundary", url: "generic://100.127.255.255", internal: true},
-		{name: "below cgnat", url: "generic://100.63.255.255", internal: false},
-		{name: "above cgnat", url: "generic://100.128.0.0", internal: false},
-		{name: "multicast ipv4", url: "generic://224.0.0.1", internal: true},
-		{name: "multicast ipv6", url: "generic://[ff02::1]", internal: true},
-		{name: "public ipv4", url: "generic://8.8.8.8", internal: false},
-		{name: "public ipv6", url: "generic://[2001:4860:4860::8888]", internal: false},
-		{name: "token style service url", url: "discord://abc123@123456789", internal: false},
-		{name: "single label service url", url: "slack://token@team/channel", internal: false},
-	}
-
-	for _, testCase := range testCases {
-		t.Run(testCase.name, func(t *testing.T) {
-			internal, err := alerts.IsInternalURL(testCase.url)
-			assert.NoError(t, err)
-			assert.Equal(t, testCase.internal, internal)
-		})
-	}
-}
-
 func TestUserAlertsApi(t *testing.T) {
 	hub, _ := beszelTests.NewTestHub(t.TempDir())
 	defer hub.Cleanup()
@@ -457,6 +421,17 @@ func TestSendTestNotification(t *testing.T) {
 	hub, user := beszelTests.GetHubWithUser(t)
 	defer hub.Cleanup()
 
+	var delivered atomic.Int32
+	server := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		delivered.Add(1)
+	}))
+	defer server.Close()
+	localURL := "generic+" + server.URL
+
+	readonlyUser, err := beszelTests.CreateUserWithRole(hub, "readonly@example.com", "password123", "readonly")
+	assert.NoError(t, err)
+	readonlyToken, err := readonlyUser.NewAuthToken()
+	assert.NoError(t, err)
 	userToken, err := user.NewAuthToken()
 
 	adminUser, err := beszelTests.CreateUserWithRole(hub, "admin@example.com", "password123", "admin")
@@ -481,11 +456,11 @@ func TestSendTestNotification(t *testing.T) {
 			ExpectedContent: []string{"requires valid"},
 			TestAppFactory:  testAppFactory,
 			Body: jsonReader(map[string]any{
-				"url": "generic://127.0.0.1",
+				"url": localURL,
 			}),
 		},
 		{
-			Name:           "POST /test-notification - with external auth should succeed",
+			Name:           "POST /test-notification - invalid service reports error",
 			Method:         http.MethodPost,
 			URL:            "/api/beszel/test-notification",
 			TestAppFactory: testAppFactory,
@@ -493,7 +468,7 @@ func TestSendTestNotification(t *testing.T) {
 				"Authorization": userToken,
 			},
 			Body: jsonReader(map[string]any{
-				"url": "generic://8.8.8.8",
+				"url": "unknown://example.com",
 			}),
 			ExpectedStatus:  200,
 			ExpectedContent: []string{"\"err\":"},
@@ -535,10 +510,10 @@ func TestSendTestNotification(t *testing.T) {
 				"Authorization": adminUserToken,
 			},
 			Body: jsonReader(map[string]any{
-				"url": "generic://127.0.0.1",
+				"url": localURL,
 			}),
 			ExpectedStatus:  200,
-			ExpectedContent: []string{"\"err\":"},
+			ExpectedContent: []string{"\"err\":false"},
 		},
 		{
 			Name:           "POST /test-notification - internal url with superuser auth should succeed",
@@ -549,14 +524,28 @@ func TestSendTestNotification(t *testing.T) {
 				"Authorization": superuserToken,
 			},
 			Body: jsonReader(map[string]any{
-				"url": "generic://127.0.0.1",
+				"url": localURL,
 			}),
 			ExpectedStatus:  200,
 			ExpectedContent: []string{"\"err\":"},
 		},
 	}
 
+	for _, url := range []string{localURL, "smtp://user:pass@127.0.0.1/?fromAddress=sender@example.com&toAddresses=recipient@example.com", "mqtt://127.0.0.1/topic"} {
+		scenarios = append(scenarios, beszelTests.ApiScenario{
+			Name:            "readonly cannot send to " + url,
+			Method:          http.MethodPost,
+			URL:             "/api/beszel/test-notification",
+			TestAppFactory:  testAppFactory,
+			Headers:         map[string]string{"Authorization": readonlyToken},
+			Body:            jsonReader(map[string]any{"url": url}),
+			ExpectedStatus:  403,
+			ExpectedContent: []string{"Only admins"},
+		})
+	}
+
 	for _, scenario := range scenarios {
 		scenario.Test(t)
 	}
+	assert.EqualValues(t, 2, delivered.Load(), "only admin and superuser requests should reach the server")
 }

internal/alerts/alerts_test_helpers.go

@@ -100,10 +100,6 @@ func (am *AlertManager) SetAlertTriggered(alert CachedAlertData, triggered bool)
 	return am.setAlertTriggered(alert, triggered)
 }
 
-func IsInternalURL(rawURL string) (bool, error) {
-	return isInternalURL(rawURL)
-}
-
 // BuildContainerLogExcerpt exposes buildContainerLogExcerpt for testing.
 func BuildContainerLogExcerpt(raw string) string {
 	return buildContainerLogExcerpt(raw)

internal/alerts/alerts_webhooks_test.go

@@ -0,0 +1,66 @@
+//go:build testing
+
+package alerts_test
+
+import (
+	"net/http"
+	"net/http/httptest"
+	"sync/atomic"
+	"testing"
+
+	"github.com/henrygd/beszel/internal/alerts"
+	beszelTests "github.com/henrygd/beszel/internal/tests"
+	"github.com/pocketbase/dbx"
+	"github.com/stretchr/testify/require"
+)
+
+func TestPersistedWebhooksUseCurrentOwnerRole(t *testing.T) {
+	hub, user := beszelTests.GetHubWithUser(t)
+	defer hub.Cleanup()
+	am := alerts.NewTestAlertManagerWithoutWorker(hub)
+
+	var delivered atomic.Int32
+	server := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		delivered.Add(1)
+	}))
+	defer server.Close()
+
+	settings, err := hub.FindFirstRecordByFilter("user_settings", "user={:user}", dbx.Params{"user": user.Id})
+	require.NoError(t, err)
+	settings.Set("settings", alerts.UserNotificationSettings{Webhooks: []string{"generic+" + server.URL}})
+	require.NoError(t, hub.Save(settings))
+	message := alerts.AlertMessageData{UserID: user.Id, Title: "Test", Message: "Persisted webhook"}
+
+	// Keep the same URL and manager while changing roles, so cached privileges
+	// or treating previously saved URLs as trusted would fail this test.
+	for _, tc := range []struct {
+		name string
+		role string
+		want int32
+	}{
+		{"regular user", "user", 0},
+		{"readonly user", "readonly", 0},
+		{"promoted admin", "admin", 1},
+		{"demoted admin", "user", 1},
+	} {
+		t.Run(tc.name, func(t *testing.T) {
+			user.Set("role", tc.role)
+			require.NoError(t, hub.Save(user))
+			// Webhook errors are logged; SendAlert continues to email delivery.
+			require.NoError(t, am.SendAlert(message))
+			require.Equal(t, tc.want, delivered.Load())
+		})
+	}
+
+	t.Run("missing owner fails closed", func(t *testing.T) {
+		// Model an orphaned settings record without deleting it through the
+		// normal user deletion cascade.
+		const missingOwner = "missingowner123"
+		settings.Set("user", missingOwner)
+		require.NoError(t, hub.SaveNoValidate(settings))
+		message.UserID = missingOwner
+		err := am.SendAlert(message)
+		require.ErrorContains(t, err, "load notification owner")
+		require.EqualValues(t, 1, delivered.Load())
+	})
+}

internal/alerts/notification_client.go

@@ -0,0 +1,150 @@
+package alerts
+
+import (
+	"context"
+	"errors"
+	"fmt"
+	"io"
+	"net"
+	"net/http"
+	"net/netip"
+	"sync/atomic"
+	"syscall"
+	"time"
+
+	"github.com/nicholas-fedor/shoutrrr/pkg/router"
+	"github.com/nicholas-fedor/shoutrrr/pkg/types"
+)
+
+var (
+	errInternalDestination   = errors.New("Only admins can send to internal destinations")
+	errUnrestrictedService   = errors.New("Only admins can use this notification service") // Restrict services w/o custom connection support
+	publicNotificationDialer = &net.Dialer{
+		Timeout: 10 * time.Second,
+		// Control checks each resolved address immediately before connecting.
+		Control: func(_, address string, _ syscall.RawConn) error { return checkNotificationAddress(address) },
+	}
+	publicNotificationClient = newPublicNotificationClient()
+)
+
+func newPublicNotificationClient() *http.Client {
+	return &http.Client{
+		Timeout: 15 * time.Second,
+		Transport: &http.Transport{
+			// Do not use proxies: they can resolve the target themselves and
+			// bypass the destination check on our socket.
+			DialContext:         publicNotificationDialer.DialContext,
+			TLSHandshakeTimeout: 10 * time.Second,
+			IdleConnTimeout:     90 * time.Second,
+		},
+	}
+}
+
+func checkNotificationAddress(address string) error {
+	addr, err := netip.ParseAddrPort(address)
+	if err != nil || addr.Addr().Zone() != "" {
+		return errInternalDestination
+	}
+	ip := net.IP(addr.Addr().AsSlice())
+	if !ip.IsGlobalUnicast() || isInternalIP(ip) {
+		return errInternalDestination
+	}
+	return nil
+}
+
+func sendPublicNotification(rawURL, message string) error {
+	client := &notificationClient{Client: publicNotificationClient}
+	service, err := newPublicNotificationService(rawURL, types.SenderOptions{HTTPClient: client, DialContext: client.dialContext})
+	if err == nil {
+		if closer, ok := service.(io.Closer); ok {
+			defer closer.Close()
+		}
+		err = service.Send(message, &types.Params{})
+	}
+	// Some services format errors without preserving their error chain.
+	if client.blocked.Load() {
+		return errInternalDestination
+	}
+	return err
+}
+
+type notificationClient struct {
+	*http.Client
+	blocked atomic.Bool
+}
+
+func (c *notificationClient) Do(req *http.Request) (*http.Response, error) {
+	response, err := c.Client.Do(req)
+	if errors.Is(err, errInternalDestination) {
+		c.blocked.Store(true)
+	}
+	return response, err
+}
+
+func (c *notificationClient) dialContext(ctx context.Context, network, address string) (net.Conn, error) {
+	conn, err := publicNotificationDialer.DialContext(ctx, network, address)
+	if errors.Is(err, errInternalDestination) {
+		c.blocked.Store(true)
+	}
+	return conn, err
+}
+
+func newPublicNotificationService(rawURL string, opts types.SenderOptions) (types.Service, error) {
+	r := &router.ServiceRouter{}
+	scheme, serviceURL, err := r.ExtractServiceName(rawURL)
+	if err != nil {
+		return nil, err
+	}
+	service, err := r.NewService(scheme)
+	if err != nil {
+		return nil, err
+	}
+	httpSetter, httpOK := service.(types.HTTPClientSetter)
+	dialSetter, dialOK := service.(types.DialContextSetter)
+	if (!httpOK || opts.HTTPClient == nil) && (!dialOK || opts.DialContext == nil) {
+		return nil, errUnrestrictedService
+	}
+	if serviceURL.Scheme != scheme {
+		custom, ok := service.(types.CustomURLService)
+		if !ok {
+			return nil, fmt.Errorf("%w: %s", router.ErrCustomURLsNotSupported, scheme)
+		}
+		serviceURL, err = custom.GetServiceURLFromCustom(serviceURL)
+		if err != nil {
+			return nil, err
+		}
+	}
+	// Shoutrrr v0.20.0 CreateSenderWithOptions injects only AFTER Initialize.
+	// Matrix can log in during Initialize, so inject before it as well.
+	if httpOK {
+		httpSetter.SetHTTPClient(opts.HTTPClient)
+	}
+	if dialOK {
+		dialSetter.SetDialContext(opts.DialContext)
+	}
+	if err := service.Initialize(serviceURL, nil); err != nil {
+		return nil, err
+	}
+	// Some initializers replace their HTTP client with a default client.
+	if httpOK {
+		httpSetter.SetHTTPClient(opts.HTTPClient)
+	}
+	if dialOK {
+		dialSetter.SetDialContext(opts.DialContext)
+	}
+	return service, nil
+}
+
+var cgnatNetwork = &net.IPNet{
+	IP:   net.IPv4(100, 64, 0, 0),
+	Mask: net.CIDRMask(10, 32),
+}
+
+func isInternalIP(ip net.IP) bool {
+	return ip.IsPrivate() ||
+		ip.IsLoopback() ||
+		ip.IsUnspecified() ||
+		ip.IsLinkLocalUnicast() ||
+		ip.IsMulticast() ||
+		cgnatNetwork.Contains(ip)
+}

internal/alerts/notification_client_test.go

@@ -0,0 +1,217 @@
+package alerts
+
+import (
+	"context"
+	"encoding/binary"
+	"errors"
+	"io"
+	"net"
+	"net/http"
+	"net/http/httptest"
+	"strings"
+	"sync/atomic"
+	"testing"
+
+	"github.com/nicholas-fedor/shoutrrr/pkg/types"
+	"golang.org/x/net/dns/dnsmessage"
+)
+
+func TestCheckNotificationAddress(t *testing.T) {
+	for _, host := range []string{"127.0.0.1", "10.0.0.1", "172.16.0.1", "192.168.0.1", "169.254.169.254", "100.64.0.0", "100.127.255.255", "0.0.0.0", "224.0.0.1", "255.255.255.255", "::1", "::", "fc00::1", "fe80::1", "ff02::1", "::ffff:127.0.0.1", "::ffff:169.254.169.254", "fe80::1%lo", "localhost", "consul"} {
+		t.Run(host, func(t *testing.T) {
+			if err := checkNotificationAddress(net.JoinHostPort(host, "80")); !errors.Is(err, errInternalDestination) {
+				t.Fatalf("expected blocked address, got %v", err)
+			}
+		})
+	}
+	for _, host := range []string{"8.8.8.8", "100.63.255.255", "100.128.0.0", "2001:4860:4860::8888"} {
+		if err := checkNotificationAddress(net.JoinHostPort(host, "443")); err != nil {
+			t.Errorf("public address %s: %v", host, err)
+		}
+	}
+}
+
+func TestPublicNotificationBlocksInternalRequests(t *testing.T) {
+	var hits atomic.Int32
+	server := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		hits.Add(1)
+	}))
+	defer server.Close()
+	host := strings.TrimPrefix(server.URL, "http://")
+	for _, rawURL := range []string{
+		"generic+http://" + host,
+		"generic+https://" + host,
+		"generic+http://localhost:" + strings.Split(host, ":")[1],
+		"matrix://user:password@" + host + "/room?disabletls=yes",
+		"mattermost://" + host + "/token?disabletls=yes",
+	} {
+		t.Run(rawURL, func(t *testing.T) {
+			if err := sendPublicNotification(rawURL, "test"); !errors.Is(err, errInternalDestination) {
+				t.Fatalf("expected internal destination error, got %v", err)
+			}
+		})
+	}
+	if hits.Load() != 0 {
+		t.Fatal("internal server received a request")
+	}
+
+}
+
+type notificationRoundTripper func(*http.Request) (*http.Response, error)
+
+func (f notificationRoundTripper) RoundTrip(r *http.Request) (*http.Response, error) { return f(r) }
+
+func TestPublicNotificationRedirect(t *testing.T) {
+	client := newPublicNotificationClient()
+	defer client.CloseIdleConnections()
+	transport := client.Transport
+	client.Transport = notificationRoundTripper(func(r *http.Request) (*http.Response, error) {
+		if r.URL.Host == "public.example" {
+			return &http.Response{StatusCode: 307, Header: http.Header{"Location": {"http://127.0.0.1/"}}, Body: io.NopCloser(strings.NewReader("")), Request: r}, nil
+		}
+		return transport.RoundTrip(r)
+	})
+	_, err := client.Get("http://public.example/")
+	if !errors.Is(err, errInternalDestination) {
+		t.Fatalf("expected redirect to be blocked, got %v", err)
+	}
+}
+
+func TestPublicNotificationServiceClient(t *testing.T) {
+	for _, rawURL := range []string{"generic+http://public.example/path", "discord://token@123456789", "slack://hook:AAAAAAAAA-BBBBBBBBB-123456789123456789123456@webhook"} {
+		t.Run(rawURL, func(t *testing.T) {
+			var hits int
+			client := &http.Client{Transport: notificationRoundTripper(func(r *http.Request) (*http.Response, error) {
+				hits++
+				body := `{"ok":true}`
+				if strings.HasPrefix(rawURL, "slack:") {
+					body = "ok"
+				}
+				return &http.Response{StatusCode: 200, Header: make(http.Header), Body: io.NopCloser(strings.NewReader(body)), Request: r}, nil
+			})}
+			service, err := newPublicNotificationService(rawURL, types.SenderOptions{HTTPClient: client})
+			if err != nil {
+				t.Fatal(err)
+			}
+			if err := service.Send("test", &types.Params{}); err != nil {
+				t.Fatal(err)
+			}
+			if hits == 0 {
+				t.Fatal("injected client was not used")
+			}
+		})
+	}
+}
+
+func TestPublicNotificationDNS(t *testing.T) {
+	// Supply deterministic DNS responses over an in-memory TCP connection.
+	// The first lookup sees a public IP; subsequent lookups see loopback.
+	var rebound atomic.Bool
+	resolver := &net.Resolver{PreferGo: true, Dial: func(ctx context.Context, network, address string) (net.Conn, error) {
+		client, server := net.Pipe()
+		go func() {
+			defer server.Close()
+			var size [2]byte
+			if _, err := io.ReadFull(server, size[:]); err != nil {
+				return
+			}
+			buf := make([]byte, binary.BigEndian.Uint16(size[:]))
+			if _, err := io.ReadFull(server, buf); err != nil {
+				return
+			}
+			var msg dnsmessage.Message
+			if err := msg.Unpack(buf); err != nil {
+				return
+			}
+			msg.Header.Response = true
+			msg.Header.RecursionAvailable = true
+			q := msg.Questions[0]
+			if q.Type == dnsmessage.TypeA {
+				ip := [4]byte{8, 8, 8, 8}
+				if rebound.Load() {
+					ip = [4]byte{127, 0, 0, 1}
+				}
+				msg.Answers = []dnsmessage.Resource{{Header: dnsmessage.ResourceHeader{Name: q.Name, Type: q.Type, Class: dnsmessage.ClassINET}, Body: &dnsmessage.AResource{A: ip}}}
+			}
+			buf, err := msg.Pack()
+			if err != nil {
+				return
+			}
+			binary.BigEndian.PutUint16(size[:], uint16(len(buf)))
+			server.Write(append(size[:], buf...))
+		}()
+		return client, nil
+	}}
+	// These tests do not run in parallel; restore the process resolver afterward.
+	previous := net.DefaultResolver
+	net.DefaultResolver = resolver
+	t.Cleanup(func() { net.DefaultResolver = previous })
+	ips, err := resolver.LookupIP(context.Background(), "ip4", "rebind.example")
+	if err != nil || len(ips) != 1 || !ips[0].Equal(net.IPv4(8, 8, 8, 8)) {
+		t.Fatalf("initial DNS lookup: %v, %v", ips, err)
+	}
+	rebound.Store(true)
+	client := newPublicNotificationClient()
+	defer client.CloseIdleConnections()
+	for _, host := range []string{"rebind.example", "consul"} {
+		guarded := &notificationClient{Client: client}
+		conn, dialErr := guarded.dialContext(context.Background(), "tcp", net.JoinHostPort(host, "25"))
+		if conn != nil {
+			conn.Close()
+		}
+		if !errors.Is(dialErr, errInternalDestination) || !guarded.blocked.Load() {
+			t.Errorf("expected TCP dial-time rejection for %s, got %v", host, dialErr)
+		}
+		_, err := client.Get("http://" + host + "/")
+		if !errors.Is(err, errInternalDestination) {
+			t.Errorf("expected dial-time rejection for %s, got %v", host, err)
+		}
+	}
+}
+
+func TestPublicNotificationTCP(t *testing.T) {
+	for _, rawURL := range []string{
+		"smtp://user:pass@HOST:25/?fromAddress=sender@example.com&toAddresses=recipient@example.com",
+		"smtp://user:pass@HOST:465/?fromAddress=sender@example.com&toAddresses=recipient@example.com",
+		"mqtt://HOST:1883/topic",
+		"mqtts://HOST:8883/topic",
+	} {
+		t.Run(rawURL, func(t *testing.T) {
+			t.Parallel()
+			t.Run("internal destination", func(t *testing.T) {
+				err := sendPublicNotification(strings.ReplaceAll(rawURL, "HOST", "127.0.0.1"), "test")
+				if !errors.Is(err, errInternalDestination) {
+					t.Fatalf("expected blocked destination, got %v", err)
+				}
+			})
+			t.Run("public destination uses injected dialer", func(t *testing.T) {
+				var calls atomic.Int32
+				stopped := errors.New("test dial stopped")
+				service, err := newPublicNotificationService(strings.ReplaceAll(rawURL, "HOST", "8.8.8.8"), types.SenderOptions{
+					DialContext: func(ctx context.Context, network, address string) (net.Conn, error) {
+						calls.Add(1)
+						if network != "tcp" || !strings.HasPrefix(address, "8.8.8.8:") {
+							t.Errorf("unexpected dial: %s %s", network, address)
+						}
+						if err := checkNotificationAddress(address); err != nil {
+							t.Error(err)
+						}
+						return nil, stopped
+					},
+				})
+				if err != nil {
+					t.Fatal(err)
+				}
+				if closer, ok := service.(io.Closer); ok {
+					defer closer.Close()
+				}
+				if err := service.Send("test", &types.Params{}); err == nil {
+					t.Fatal("expected dial failure")
+				}
+				if calls.Load() == 0 {
+					t.Fatal("custom dialer was not used")
+				}
+			})
+		})
+	}
+}

internal/site/src/components/routes/system/smart-table.tsx

@@ -75,7 +75,12 @@ export const smartColumns: ColumnDef<SmartAttribute>[] = [
 		header: "Name",
 	},
 	{
-		accessorFn: (row) => row.rs || row.rv?.toString(),
+		accessorFn: (row) => {
+			if (row.n === "DataUnitsWritten" || row.n === "DataUnitsRead") {
+				return formatDataUnits(Number(row.rv ?? 0))
+			}
+			return row.rs || row.rv?.toString()
+		},
 		header: "Value",
 	},
 	{
@@ -103,6 +108,12 @@ function formatCapacity(bytes: number): string {
 	return `${toFixedFloat(value, value >= 10 ? 1 : 2)} ${unit}`
 }
 
+// Function to format NVMe data units
+// (1 unit = 1000 * 512 bytes) as a human-readable size
+function formatDataUnits(units: number): string {
+	return formatCapacity(units * 1000 * 512)
+}
+
 const SMART_DEVICE_FIELDS = "id,system,name,model,state,capacity,temp,type,hours,cycles,updated"
 
 export const createColumns = (

supplemental/scripts/install-agent.sh

@@ -43,7 +43,7 @@ is_glibc() {
 set_selinux_context() {
   # Check if SELinux is enabled and in enforcing or permissive mode
   if command -v getenforce >/dev/null 2>&1; then
-    SELINUX_MODE=$(getenforce)
+    SELINUX_MODE=$(getenforce) || { warn "Could not query SELinux mode."; return 0; }
     if [ "$SELINUX_MODE" != "Disabled" ]; then
       echo "SELinux is enabled (${SELINUX_MODE} mode). Setting appropriate context..."
 
@@ -51,7 +51,7 @@ set_selinux_context() {
       if command -v semanage >/dev/null 2>&1; then
         echo "Attempting to set persistent SELinux context..."
         if semanage fcontext -a -t bin_t "$BIN_PATH" >/dev/null 2>&1; then
-          restorecon -v "$BIN_PATH" >/dev/null 2>&1
+          restorecon -v "$BIN_PATH" >/dev/null 2>&1 || warn "Failed to restore persistent SELinux context; trying chcon."
         else
           echo "Warning: Failed to set persistent context, falling back to temporary context."
         fi
@@ -271,7 +271,7 @@ detect_mips_endianness() {
   for bin_to_check in $bins; do
     if [ -f "$bin_to_check" ]; then
       # The 6th byte in ELF header: 01 = little, 02 = big
-      endian=$(hexdump -n 1 -s 5 -e '1/1 "%02x"' "$bin_to_check" 2>/dev/null)
+      endian=$(hexdump -n 1 -s 5 -e '1/1 "%02x"' "$bin_to_check" 2>/dev/null) || continue
       if [ "$endian" = "01" ]; then
         echo "mipsle"
         return
@@ -286,6 +286,136 @@ detect_mips_endianness() {
   echo "mips"
 }
 
+# Expected failures must be handled explicitly; unexpected failures abort installation.
+set -eu
+
+fail() {
+  echo "Error: $*" >&2
+  exit 1
+}
+
+warn() {
+  echo "Warning: $*" >&2
+}
+
+require_value() {
+  [ "$#" -ge 2 ] && [ -n "$2" ] || fail "Option $1 requires a value."
+}
+
+validate_platform() {
+  case "$(uname -s)" in
+    Linux)
+      if is_alpine; then
+        command -v rc-service >/dev/null && command -v rc-update >/dev/null || fail "OpenRC is required."
+      elif is_openwrt; then
+        [ -f /etc/rc.common ] || fail "OpenWrt procd is required."
+      else
+        command -v systemctl >/dev/null && [ -d /run/systemd/system ] || fail "This Linux installer requires a running systemd, OpenRC (Alpine), or procd (OpenWrt)."
+      fi
+      ;;
+    FreeBSD)
+      command -v service >/dev/null && command -v sysrc >/dev/null || fail "FreeBSD service and sysrc commands are required."
+      ;;
+    Darwin) fail "For macOS, use the Homebrew installer: https://github.com/henrygd/beszel/blob/main/supplemental/scripts/install-agent-brew.sh" ;;
+    *) fail "Unsupported operating system: $(uname -s)" ;;
+  esac
+}
+
+agent_service() {
+  if is_alpine; then
+    rc-service beszel-agent "$1"
+  elif is_openwrt; then
+    /etc/init.d/beszel-agent "$1"
+  elif is_freebsd; then
+    service beszel-agent "$1"
+  else
+    systemctl "$1" beszel-agent.service
+  fi
+}
+
+# Match the files preserved by the service setup below. A binary or rc script
+# alone is not reusable configuration (FreeBSD stores its environment separately).
+agent_configuration_exists() {
+  if is_alpine || is_openwrt; then
+    [ -f /etc/init.d/beszel-agent ]
+  elif is_freebsd; then
+    [ -f "$AGENT_DIR/env" ]
+  else
+    [ -f /etc/systemd/system/beszel-agent.service ]
+  fi
+}
+
+# An orphaned binary can remain after a failed install. It does not imply
+# that the service manager knows about the agent yet.
+agent_service_registered() {
+  if is_alpine || is_openwrt; then
+    [ -f /etc/init.d/beszel-agent ]
+  elif is_freebsd; then
+    [ -f /usr/local/etc/rc.d/beszel-agent ]
+  else
+    service_load_state=$(systemctl show --property=LoadState --value beszel-agent.service) || return 2
+    case "$service_load_state" in
+      not-found) return 1 ;;
+      "") return 2 ;;
+      *) return 0 ;;
+    esac
+  fi
+}
+
+TEMP_DIR=""
+STAGED_BINARY=""
+INSTALL_STEP="validating installation options"
+UPGRADE_PENDING=false
+cleanup() {
+  cleanup_status=$?
+  trap - 0 HUP INT TERM
+  if [ "$cleanup_status" -ne 0 ]; then
+    warn "Installer failed while $INSTALL_STEP (exit $cleanup_status)."
+    if [ "$UPGRADE_PENDING" = true ]; then
+      warn "Restoring the previous binary and restarting its service if registered."
+      if [ -n "$STAGED_BINARY" ]; then
+        rm -f "$STAGED_BINARY" || warn "Could not remove staged binary."
+      fi
+      if STAGED_BINARY=$(mktemp "$BIN_PATH.XXXXXX") && cp -p "$BIN_PATH.bak" "$STAGED_BINARY" && mv -f "$STAGED_BINARY" "$BIN_PATH"; then
+        # The temporary inode does not inherit the installed binary's SELinux label.
+        set_selinux_context || warn "Could not restore SELinux context on the previous agent."
+        if agent_service_registered; then
+          agent_service restart || warn "Could not restart the previous agent; check the service configuration and logs."
+        else
+          service_check_status=$?
+          [ "$service_check_status" -eq 1 ] || warn "Could not determine whether the previous agent service is registered; check it manually."
+        fi
+      else
+        warn "Could not restore $BIN_PATH.bak. Restore it manually before restarting the service."
+      fi
+    fi
+  fi
+  if [ -n "$STAGED_BINARY" ]; then
+    rm -f "$STAGED_BINARY" || warn "Could not remove staged binary."
+  fi
+  if [ -n "$TEMP_DIR" ]; then
+    rm -rf "$TEMP_DIR" || warn "Could not remove temporary directory $TEMP_DIR."
+  fi
+  exit "$cleanup_status"
+}
+trap cleanup 0
+trap 'exit 129' HUP
+trap 'exit 130' INT
+trap 'exit 143' TERM
+
+# A missing crontab is normal. Keep the producer alive so the new job is written.
+read_root_crontab() {
+  crontab -u root -l 2>/dev/null || true
+}
+
+prompt_auto_update() {
+  printf "\nEnable automatic daily updates for beszel-agent? (y/n): "
+  if ! read -r AUTO_UPDATE; then
+    AUTO_UPDATE=n
+    echo "Skipping automatic updates (no input)."
+  fi
+}
+
 # Default values
 PORT=45876
 UNINSTALL=false
@@ -304,7 +434,7 @@ HUB_URL_PROVIDED=false
 VERSION="latest"
 
 # Check for help flag
-case "$1" in
+case "${1-}" in
 -h | --help)
   printf "Beszel Agent installation script\n\n"
   printf "Usage: ./install-agent.sh [options]\n\n"
@@ -324,6 +454,9 @@ case "$1" in
   ;;
 esac
 
+# Reject unsupported hosts before sudo or any system changes.
+validate_platform
+
 # Build sudo args by properly quoting everything
 build_sudo_args() {
   QUOTED_ARGS=""
@@ -354,26 +487,31 @@ fi
 while [ $# -gt 0 ]; do
   case "$1" in
   -k)
+    require_value "$@"
     shift
     KEY="$1"
     KEY_PROVIDED=true
     ;;
   -p)
+    require_value "$@"
     shift
     PORT="$1"
     PORT_PROVIDED=true
     ;;
   -t)
+    require_value "$@"
     shift
     TOKEN="$1"
     TOKEN_PROVIDED=true
     ;;
   -url)
+    require_value "$@"
     shift
     HUB_URL="$1"
     HUB_URL_PROVIDED=true
     ;;
   -v | --version)
+    require_value "$@"
     shift
     VERSION="$1"
     ;;
@@ -392,7 +530,7 @@ while [ $# -gt 0 ]; do
         GITHUB_PROXY_URL="https://gh.beszel.dev"
         GITHUB_URL="$GITHUB_PROXY_URL"
       fi
-    elif [ "$2" != "" ] && ! echo "$2" | grep -q '^-'; then
+    elif [ "${2-}" != "" ] && ! echo "$2" | grep -q '^-'; then
       # use custom proxy URL provided as next argument
       GITHUB_PROXY_URL="$2"
       GITHUB_URL="$(ensure_trailing_slash "$2")https://github.com"
@@ -415,7 +553,7 @@ while [ $# -gt 0 ]; do
       else
         echo "Invalid value for --auto-update flag: $AUTO_UPDATE_VALUE. Using default (prompt)."
       fi
-    elif [ "$2" = "true" ] || [ "$2" = "false" ]; then
+    elif [ "${2-}" = "true" ] || [ "${2-}" = "false" ]; then
       # Value provided as next argument
       AUTO_UPDATE_FLAG="$2"
       shift
@@ -443,20 +581,7 @@ else
   BIN_PATH="/opt/beszel-agent/beszel-agent"
 fi
 
-# Stop existing service if it exists (for upgrades)
-if [ "$UNINSTALL" != true ] && [ -f "$BIN_PATH" ]; then
-  echo "Existing installation detected. Stopping service for upgrade..."
-  if is_alpine; then
-    rc-service beszel-agent stop 2>/dev/null || true
-  elif is_openwrt; then
-    /etc/init.d/beszel-agent stop 2>/dev/null || true
-  elif is_freebsd; then
-    service beszel-agent stop 2>/dev/null || true
-  else
-    systemctl stop beszel-agent.service 2>/dev/null || true
-  fi
-fi
-
+INSTALL_STEP="uninstalling the agent"
 # Uninstall process
 if [ "$UNINSTALL" = true ]; then
   # Clean up SELinux contexts before removing files
@@ -464,8 +589,8 @@ if [ "$UNINSTALL" = true ]; then
 
   if is_alpine; then
     echo "Stopping and disabling the agent service..."
-    rc-service beszel-agent stop
-    rc-update del beszel-agent default
+    rc-service beszel-agent stop || warn "Cleanup command failed: rc-service beszel-agent stop"
+    rc-update del beszel-agent default || warn "Cleanup command failed: rc-update del beszel-agent default"
 
     echo "Removing the OpenRC service files..."
     rm -f /etc/init.d/beszel-agent
@@ -481,8 +606,8 @@ if [ "$UNINSTALL" = true ]; then
     rm -f /var/log/beszel-agent.log /var/log/beszel-agent.err
   elif is_openwrt; then
     echo "Stopping and disabling the agent service..."
-    /etc/init.d/beszel-agent stop
-    /etc/init.d/beszel-agent disable
+    /etc/init.d/beszel-agent stop || warn "Cleanup command failed: /etc/init.d/beszel-agent stop"
+    /etc/init.d/beszel-agent disable || warn "Cleanup command failed: /etc/init.d/beszel-agent disable"
 
     echo "Removing the OpenWRT service files..."
     rm -f /etc/init.d/beszel-agent
@@ -498,7 +623,7 @@ if [ "$UNINSTALL" = true ]; then
 
   elif is_freebsd; then
     echo "Stopping and disabling the agent service..."
-    service beszel-agent stop
+    service beszel-agent stop || warn "Cleanup command failed: service beszel-agent stop"
     sysrc beszel_agent_enable="NO"
 
     echo "Removing the FreeBSD service files..."
@@ -525,16 +650,16 @@ if [ "$UNINSTALL" = true ]; then
 
   else
     echo "Stopping and disabling the agent service..."
-    systemctl stop beszel-agent.service
-    systemctl disable beszel-agent.service >/dev/null 2>&1
+    systemctl stop beszel-agent.service || warn "Cleanup command failed: systemctl stop beszel-agent.service"
+    systemctl disable beszel-agent.service >/dev/null 2>&1 || warn "Cleanup command failed: systemctl disable beszel-agent.service"
 
     echo "Removing the systemd service file..."
-    rm /etc/systemd/system/beszel-agent.service
+    rm -f /etc/systemd/system/beszel-agent.service
 
     # Remove the update timer and service if they exist
     echo "Removing the daily update service and timer..."
-    systemctl stop beszel-agent-update.timer 2>/dev/null
-    systemctl disable beszel-agent-update.timer >/dev/null 2>&1
+    systemctl stop beszel-agent-update.timer 2>/dev/null || warn "Cleanup command failed: systemctl stop beszel-agent-update.timer"
+    systemctl disable beszel-agent-update.timer >/dev/null 2>&1 || warn "Cleanup command failed: systemctl disable beszel-agent-update.timer"
     rm -f /etc/systemd/system/beszel-agent-update.service
     rm -f /etc/systemd/system/beszel-agent-update.timer
 
@@ -545,13 +670,15 @@ if [ "$UNINSTALL" = true ]; then
   rm -rf "$AGENT_DIR"
 
   echo "Removing the dedicated user for the agent service..."
-  killall beszel-agent 2>/dev/null
-  if is_alpine || is_openwrt; then
-    deluser beszel 2>/dev/null
-  elif is_freebsd; then
-    pw user del beszel 2>/dev/null
-  else
-    userdel beszel 2>/dev/null
+  killall beszel-agent 2>/dev/null || true # Usually already stopped by the service manager.
+  if id -u beszel >/dev/null 2>&1; then
+    if is_alpine || is_openwrt; then
+      deluser beszel || fail "Could not remove the beszel user."
+    elif is_freebsd; then
+      pw user del beszel || fail "Could not remove the beszel user."
+    else
+      userdel beszel || fail "Could not remove the beszel user."
+    fi
   fi
 
   echo "Beszel Agent has been uninstalled successfully!"
@@ -563,6 +690,7 @@ package_installed() {
   command -v "$1" >/dev/null 2>&1
 }
 
+INSTALL_STEP="installing required packages"
 # Check for package manager and install necessary packages if not installed
 if package_installed apk; then
   if ! package_installed tar || ! package_installed curl || ! package_installed sha256sum; then
@@ -596,13 +724,18 @@ else
   echo "Warning: Please ensure 'tar' and 'curl' and 'sha256sum (coreutils)' are installed."
 fi
 
-# If no SSH key is provided, ask for the SSH key interactively (skip if upgrading)
+for required_command in tar curl; do
+  command -v "$required_command" >/dev/null || fail "Required command is missing: $required_command"
+done
+
+# If no SSH key is provided, prompt unless service setup will reuse configuration.
 if [ -z "$KEY" ]; then
-  if [ -f "$BIN_PATH" ]; then
-    echo "Upgrading existing installation. Using existing service configuration."
+  if agent_configuration_exists; then
+    echo "Using existing service configuration."
   else
     printf "Enter your SSH key: "
-    read KEY
+    read -r KEY || fail "No SSH key received. Supply -k for noninteractive installation."
+    [ -n "$KEY" ] || fail "SSH key must not be empty."
   fi
 fi
 
@@ -623,6 +756,7 @@ else
   exit 1
 fi
 
+INSTALL_STEP="configuring the service user"
 # Create a dedicated user for the service if it doesn't exist
 AGENT_USER="beszel"
 echo "Configuring the dedicated user for the Beszel Agent service..."
@@ -696,6 +830,7 @@ else
   fi
 fi
 
+INSTALL_STEP="creating installation directories"
 # Create the directory for the Beszel Agent
 
 if [ ! -d "$AGENT_DIR" ]; then
@@ -709,6 +844,7 @@ if [ ! -d "$BIN_DIR" ]; then
   mkdir -p "$BIN_DIR"
 fi
 
+INSTALL_STEP="downloading and verifying the agent"
 # Download and install the Beszel Agent
 
 OS=$(uname -s | sed -e 'y/ABCDEFGHIJKLMNOPQRSTUVWXYZ/abcdefghijklmnopqrstuvwxyz/')
@@ -720,11 +856,12 @@ fi
 
 # Determine version to install
 if [ "$VERSION" = "latest" ]; then
-  INSTALL_VERSION=$(curl -s "https://get.beszel.dev/latest-version")
+  INSTALL_VERSION=$(curl -fsS --connect-timeout 10 --max-time 30 "https://get.beszel.dev/latest-version") || INSTALL_VERSION=""
   if [ -z "$INSTALL_VERSION" ]; then
     # Fallback to GitHub API
     API_RELEASE_URL="https://api.github.com/repos/henrygd/beszel/releases/latest"
-    INSTALL_VERSION=$(curl -s "$API_RELEASE_URL" | grep -o '"tag_name": "v[^"]*"' | cut -d'"' -f4 | tr -d 'v')
+    RELEASE_JSON=$(curl -fsS --connect-timeout 10 --max-time 30 "$API_RELEASE_URL") || fail "Could not fetch the latest release from GitHub."
+    INSTALL_VERSION=$(printf '%s\n' "$RELEASE_JSON" | grep -o '"tag_name": "v[^"]*"' | cut -d'"' -f4 | tr -d 'v')
   fi
   if [ -z "$INSTALL_VERSION" ]; then
     echo "Failed to get latest version"
@@ -741,7 +878,8 @@ echo "Downloading beszel-agent v${INSTALL_VERSION}..."
 # Download checksums file
 TEMP_DIR=$(mktemp -d)
 cd "$TEMP_DIR" || exit 1
-CHECKSUM=$(curl -fsSL "$GITHUB_URL/henrygd/beszel/releases/download/v${INSTALL_VERSION}/beszel_${INSTALL_VERSION}_checksums.txt" | grep "$FILE_NAME" | cut -d' ' -f1)
+curl -fsSL --connect-timeout 10 --max-time 60 "$GITHUB_URL/henrygd/beszel/releases/download/v${INSTALL_VERSION}/beszel_${INSTALL_VERSION}_checksums.txt" -o checksums.txt || fail "Could not download checksums. Try --mirror if GitHub is unreachable."
+CHECKSUM=$(awk -v name="$FILE_NAME" '$2 == name { print $1 }' checksums.txt)
 if [ -z "$CHECKSUM" ] || ! echo "$CHECKSUM" | grep -qE "^[a-fA-F0-9]{64}$"; then
   echo "Failed to get checksum or invalid checksum format"
   echo "Try again with --mirror (or --mirror <url>) if GitHub is not reachable."
@@ -763,10 +901,10 @@ if ! tar -tzf "$FILE_NAME" >/dev/null 2>&1; then
   exit 1
 fi
 
-if [ "$($CHECK_CMD "$FILE_NAME" | cut -d' ' -f1)" != "$CHECKSUM" ]; then
-  echo "Checksum verification failed: $($CHECK_CMD "$FILE_NAME" | cut -d' ' -f1) & $CHECKSUM"
-  rm -rf "$TEMP_DIR"
-  exit 1
+ACTUAL_CHECKSUM=$($CHECK_CMD "$FILE_NAME") || fail "Could not calculate archive checksum."
+ACTUAL_CHECKSUM=${ACTUAL_CHECKSUM%% *}
+if [ "$ACTUAL_CHECKSUM" != "$CHECKSUM" ]; then
+  fail "Checksum verification failed: $ACTUAL_CHECKSUM != $CHECKSUM"
 fi
 
 if ! tar -xzf "$FILE_NAME" beszel-agent; then
@@ -781,20 +919,34 @@ if [ ! -s "$TEMP_DIR/beszel-agent" ]; then
   exit 1
 fi
 
+INSTALL_STEP="replacing the agent binary"
+# Stage on the destination filesystem so replacement and rollback use atomic renames.
+STAGED_BINARY=$(mktemp "$BIN_PATH.XXXXXX") || fail "Could not create a staged binary."
+cp beszel-agent "$STAGED_BINARY" || fail "Could not stage the agent binary."
+chown "${AGENT_USER}:${AGENT_USER}" "$STAGED_BINARY" || fail "Could not set binary ownership."
+chmod 755 "$STAGED_BINARY" || fail "Could not set binary permissions."
+
 if [ -f "$BIN_PATH" ]; then
   echo "Backing up existing binary..."
-  cp "$BIN_PATH" "$BIN_PATH.bak"
+  cp -p "$BIN_PATH" "$BIN_PATH.bak" || fail "Could not back up the existing binary."
+  UPGRADE_PENDING=true
+  if agent_service_registered; then
+    agent_service stop || fail "Could not stop the existing agent."
+  else
+    service_check_status=$?
+    [ "$service_check_status" -eq 1 ] || fail "Could not determine whether the existing agent service is registered."
+  fi
 fi
 
-mv beszel-agent "$BIN_PATH"
-chown "${AGENT_USER}:${AGENT_USER}" "$BIN_PATH"
-chmod 755 "$BIN_PATH"
+mv -f "$STAGED_BINARY" "$BIN_PATH" || fail "Could not install the agent binary."
+STAGED_BINARY=""
 
 # Set SELinux context if needed
 set_selinux_context
 
 # Cleanup
 rm -rf "$TEMP_DIR"
+TEMP_DIR=""
 
 # Make sure /etc/machine-id exists and is non-empty for persistent fingerprint
 if [ ! -s /etc/machine-id ]; then
@@ -819,6 +971,7 @@ detect_nvidia_devices() {
   echo "$devices"
 }
 
+INSTALL_STEP="configuring and starting the service"
 # Modify service installation part, add Alpine check before systemd service creation
 if is_alpine; then
   if [ ! -f /etc/init.d/beszel-agent ]; then
@@ -868,7 +1021,7 @@ EOF
   chown "${AGENT_USER}:${AGENT_USER}" /var/log/beszel-agent.log /var/log/beszel-agent.err
 
   # Start the service
-  rc-service beszel-agent restart
+  rc-service beszel-agent restart || fail "Could not start the agent; check service logs."
 
   # Check if service started successfully
   sleep 2
@@ -884,8 +1037,7 @@ EOF
   elif [ "$AUTO_UPDATE_FLAG" = "false" ]; then
     AUTO_UPDATE="n"
   else
-    printf "\nEnable automatic daily updates for beszel-agent? (y/n): "
-    read AUTO_UPDATE
+    prompt_auto_update
   fi
   case "$AUTO_UPDATE" in
   [Yy]*)
@@ -893,7 +1045,7 @@ EOF
 
     # Create cron job to run beszel-agent update command daily at midnight
     if ! crontab -u root -l 2>/dev/null | grep -q "beszel-agent.*update"; then
-      (crontab -u root -l 2>/dev/null; echo "12 0 * * * $BIN_PATH update >/dev/null 2>&1") | crontab -u root -
+      (read_root_crontab; echo "12 0 * * * $BIN_PATH update >/dev/null 2>&1") | crontab -u root -
     fi
 
     printf "\nDaily updates have been enabled via cron job.\n"
@@ -963,7 +1115,7 @@ EOF
   fi
 
   # Start the service
-  /etc/init.d/beszel-agent restart
+  /etc/init.d/beszel-agent restart || fail "Could not start the agent; check service logs."
 
   # Auto-update service for OpenWRT using a crontab job
   if [ "$AUTO_UPDATE_FLAG" = "true" ]; then
@@ -973,15 +1125,14 @@ EOF
     AUTO_UPDATE="n"
     sleep 1 # give time for the service to start
   else
-    printf "\nEnable automatic daily updates for beszel-agent? (y/n): "
-    read AUTO_UPDATE
+    prompt_auto_update
   fi
   case "$AUTO_UPDATE" in
   [Yy]*)
     echo "Setting up daily automatic updates for beszel-agent..."
 
     if ! crontab -u root -l 2>/dev/null | grep -q "beszel-agent.*update"; then
-      (crontab -u root -l 2>/dev/null; echo "12 0 * * * /etc/init.d/beszel-agent update") | crontab -u root -
+      (read_root_crontab; echo "12 0 * * * /etc/init.d/beszel-agent update") | crontab -u root -
     fi
 
     /etc/init.d/cron restart
@@ -1072,7 +1223,7 @@ EOF
     esac
   fi
 
-  service beszel-agent restart
+  service beszel-agent restart || fail "Could not start the agent; check service logs."
   
   # Check if service started successfully
   sleep 2
@@ -1088,8 +1239,7 @@ EOF
   elif [ "$AUTO_UPDATE_FLAG" = "false" ]; then
     AUTO_UPDATE="n"
   else
-    printf "\nEnable automatic daily updates for beszel-agent? (y/n): "
-    read AUTO_UPDATE
+    prompt_auto_update
   fi
   case "$AUTO_UPDATE" in
   [Yy]*)
@@ -1170,7 +1320,7 @@ EOF
   printf "\nLoading and starting the agent service...\n"
   systemctl daemon-reload
   systemctl enable beszel-agent.service >/dev/null 2>&1
-  systemctl restart beszel-agent.service
+  systemctl restart beszel-agent.service || fail "Could not start the agent; check service logs."
 
 
 
@@ -1182,8 +1332,7 @@ EOF
     AUTO_UPDATE="n"
     sleep 1 # give time for the service to start
   else
-    printf "\nEnable automatic daily updates for beszel-agent? (y/n): "
-    read AUTO_UPDATE
+    prompt_auto_update
   fi
   case "$AUTO_UPDATE" in
   [Yy]*)
@@ -1229,6 +1378,7 @@ EOF
   fi
 fi
 
+UPGRADE_PENDING=false
 RUNNING_ADDRESS=$(configured_address)
 [ -n "$RUNNING_ADDRESS" ] || RUNNING_ADDRESS=$PORT