edgeguard-native/internal/handlers/system.go

package handlers

import (
	"bufio"
	stdcontext "context"
	"errors"
	"log/slog"
	"net"
	"net/http"
	"os"
	"os/exec"
	"runtime"
	"strconv"
	"strings"
	"syscall"
	"time"

	"github.com/gin-gonic/gin"

	"github.com/jackc/pgx/v5/pgxpool"

	"git.netcell-it.de/projekte/edgeguard-native/internal/handlers/response"
	aptsvc "git.netcell-it.de/projekte/edgeguard-native/internal/services/apt"
	"git.netcell-it.de/projekte/edgeguard-native/internal/services/audit"
	"git.netcell-it.de/projekte/edgeguard-native/internal/services/setup"
	usersvc "git.netcell-it.de/projekte/edgeguard-native/internal/services/users"
)

// SystemHandler covers /system/health, /system/package-versions,
// /system/upgrade, /system/auto-update + /system/maintenance.
//
// Setup + HAProxyReloader sind optional — werden für /maintenance
// gebraucht. Wenn nil sind die Maintenance-Endpoints noch verfügbar
// aber returnen 503 (Setup nicht abgeschlossen).
type SystemHandler struct {
	Version         string
	Setup           *setup.Store
	HAProxyReloader func(stdcontext.Context) error
	Audit           *audit.Repo
	NodeID          string
	Pool            *pgxpool.Pool
	// ConfigPreviewers: generator name → render-to-string func. Wired
	// in main.go after generators are constructed. Read-only access —
	// no writes, no reloads.
	ConfigPreviewers map[string]func(stdcontext.Context) (string, error)
	// ExtraReloaders: additional service renderers triggered by
	// RenderConfigs. Keyed by service name (nftables, wireguard, etc.).
	ExtraReloaders map[string]func(stdcontext.Context) error
	// Users: optional — wired after DB pool opens. Used by AgentAuthCheck
	// so cluster peers can verify credentials against this node's DB.
	Users *usersvc.Repo
}

func NewSystemHandler(version string) *SystemHandler {
	return &SystemHandler{Version: version}
}

// WithMaintenance: setup + reloader injection. Aufrufer in main.go
// nach DB-Pool-Open.
func (h *SystemHandler) WithMaintenance(setupStore *setup.Store, reloader func(stdcontext.Context) error) *SystemHandler {
	h.Setup = setupStore
	h.HAProxyReloader = reloader
	return h
}

// WithAudit: Audit-Repo + NodeID damit Toggle-Aktionen (Maintenance,
// Auto-Update) ins audit_log fließen.
func (h *SystemHandler) WithAudit(a *audit.Repo, nodeID string) *SystemHandler {
	h.Audit = a
	h.NodeID = nodeID
	return h
}

// WithDB: Pool für DB-Size + andere PG-introspection-Endpoints.
func (h *SystemHandler) WithDB(pool *pgxpool.Pool) *SystemHandler {
	h.Pool = pool
	return h
}

// WithConfigPreviewers injectet die RenderToString-Funktionen der
// Config-Generatoren für den /system/config-preview Endpoint.
func (h *SystemHandler) WithConfigPreviewers(previewers map[string]func(stdcontext.Context) (string, error)) *SystemHandler {
	h.ConfigPreviewers = previewers
	return h
}

// WithAllReloaders injectet alle Service-Reloader für RenderConfigs.
// Reihenfolge: haproxy ist bereits in HAProxyReloader; extras sind die
// restlichen Dienste (nftables, wireguard, squid, unbound, chrony).
func (h *SystemHandler) WithAllReloaders(extras map[string]func(stdcontext.Context) error) *SystemHandler {
	h.ExtraReloaders = extras
	return h
}

// WithUsers injectet das Users-Repo für AgentAuthCheck.
func (h *SystemHandler) WithUsers(u *usersvc.Repo) *SystemHandler {
	h.Users = u
	return h
}

func (h *SystemHandler) Register(rg *gin.RouterGroup) {
	g := rg.Group("/system")
	g.GET("/health", h.Health)
	g.GET("/package-versions", h.PackageVersions)
	g.POST("/upgrade", h.Upgrade)
	g.GET("/interfaces", h.Interfaces)
	g.GET("/services", h.Services)
	g.GET("/resources", h.Resources)
	g.GET("/auto-update", h.AutoUpdate)
	g.POST("/auto-update", h.ToggleAutoUpdate)
	g.GET("/maintenance", h.Maintenance)
	g.POST("/maintenance", h.ToggleMaintenance)
	g.GET("/backup-retention", h.BackupRetention)
	g.POST("/backup-retention", h.SetBackupRetention)
	g.GET("/audit-retention", h.AuditRetention)
	g.POST("/audit-retention", h.SetAuditRetention)
	g.GET("/db-size", h.DBSize)
	g.POST("/haproxy-reload", h.HAProxyReload)
	g.POST("/render-configs", h.RenderConfigs)
	g.POST("/service-restart", h.ServiceRestart)
	g.GET("/upgrade-status", h.UpgradeStatus)
	g.GET("/ipv6", h.IPv6)
	g.POST("/ipv6", h.SetIPv6)
	g.GET("/config-preview", h.ConfigPreview)
	g.GET("/vip-status", h.VIPStatus)
}

// RegisterAgent mountet die read-only System-Endpoints auf der mTLS-
// Agent-Engine (Port :8443). Der Cluster-Aggregator auf der Main-API
// ruft diese Endpoints parallel auf allen Peers ab und kompiliert das
// Ergebnis für /cluster/system/load.
//
// Bewusst KEINE Mutations + KEIN /package-versions (würde apt-get update
// auf jedem Peer triggern), KEIN /upgrade.
func (h *SystemHandler) RegisterAgent(rg *gin.RouterGroup) {
	g := rg.Group("/agent/system")
	g.GET("/health", h.Health)
	g.GET("/resources", h.Resources)
	// Auth-Federation: Cluster-Nodes verifizieren Credentials gegen diesen
	// Node via mTLS. Nur über den Agent-Listener (:8443) erreichbar.
	rg.POST("/agent/auth/check", h.AgentAuthCheck)
}

// AgentAuthCheck verifies email+password against the local users table
// and setup.json. Called by cluster nodes over mTLS when local auth fails
// so users can log in on any cluster node with the primary's credentials.
func (h *SystemHandler) AgentAuthCheck(c *gin.Context) {
	var req struct {
		Email    string `json:"email"`
		Password string `json:"password"`
	}
	if err := c.ShouldBindJSON(&req); err != nil {
		response.BadRequest(c, err)
		return
	}
	email := strings.TrimSpace(strings.ToLower(req.Email))
	if email == "" || req.Password == "" {
		response.Unauthorized(c, errInvalidCreds)
		return
	}

	// Check local users table first.
	if h.Users != nil {
		u, hash, err := h.Users.FindByEmail(c.Request.Context(), email)
		if err == nil && u.Active && usersvc.VerifyPassword(hash, req.Password) {
			response.OK(c, gin.H{"actor": u.Email, "role": u.Role})
			return
		}
	}
	// Fallback: setup.json admin.
	if h.Setup != nil {
		st, _ := h.Setup.Load()
		if st != nil && strings.EqualFold(st.AdminEmail, email) && st.VerifyAdminPassword(req.Password) {
			response.OK(c, gin.H{"actor": st.AdminEmail, "role": "admin"})
			return
		}
	}
	response.Unauthorized(c, errInvalidCreds)
}

var errInvalidCreds = errors.New("invalid_credentials")

// servicesToCheck is the curated list shown on the dashboard
// service-health-grid. Order matters (UI renders in this sequence).
// Each entry is a (label, systemd-unit) pair — label is what the
// UI shows, unit is what `systemctl is-active` queries.
var servicesToCheck = []struct{ Label, Unit string }{
	{"edgeguard-api", "edgeguard-api"},
	{"edgeguard-scheduler", "edgeguard-scheduler"},
	{"haproxy", "haproxy"},
	{"nftables", "nftables"},
	{"keepalived", "keepalived"},
	{"unbound", "unbound"},
	{"chrony", "chrony"},
	{"squid", "squid"},
	{"postgresql", "postgresql"},
}

type serviceStatus struct {
	Label  string `json:"label"`
	Unit   string `json:"unit"`
	Active bool   `json:"active"`
	State  string `json:"state"`         // active|inactive|failed|activating|...
	Since  string `json:"since,omitempty"` // ActiveEnterTimestamp
}

// Services returns systemd-unit status for the curated stack.
func (h *SystemHandler) Services(c *gin.Context) {
	out := make([]serviceStatus, 0, len(servicesToCheck))
	for _, s := range servicesToCheck {
		st := serviceStatus{Label: s.Label, Unit: s.Unit}
		if s.Unit == "nftables" {
			// Distro-Unit nftables.service ist disabled — wir laden
			// die Rules direkt via 'nft -f' aus dem Renderer. Status
			// = ist unsere 'inet edgeguard'-Tabelle im Kernel?
			loaded, when := nftablesKernelState(c.Request.Context())
			st.Active = loaded
			if loaded {
				st.State = "kernel-loaded"
			} else {
				st.State = "no-table"
			}
			st.Since = when
			out = append(out, st)
			continue
		}
		raw, err := exec.CommandContext(c.Request.Context(),
			"systemctl", "show", "-p", "ActiveState,ActiveEnterTimestamp",
			s.Unit).Output()
		if err == nil {
			for _, line := range strings.Split(string(raw), "\n") {
				if k, v, ok := strings.Cut(line, "="); ok {
					switch k {
					case "ActiveState":
						st.State = v
						st.Active = v == "active"
					case "ActiveEnterTimestamp":
						st.Since = v
					}
				}
			}
		}
		out = append(out, st)
	}
	response.OK(c, gin.H{"services": out})
}

// nftablesKernelState reports whether our 'inet edgeguard' table is
// present in the kernel ruleset. Errors swallow to false. Returns
// the mtime of the source file as 'since' when loaded.
func nftablesKernelState(ctx stdcontext.Context) (bool, string) {
	out, err := exec.CommandContext(ctx, "sudo", "-n", "/usr/sbin/nft", "list", "tables").Output()
	if err != nil {
		return false, ""
	}
	if !strings.Contains(string(out), "inet edgeguard") {
		return false, ""
	}
	when := ""
	if fi, err := os.Stat("/etc/edgeguard/nftables.d/ruleset.nft"); err == nil {
		when = fi.ModTime().UTC().Format(time.RFC3339)
	}
	return true, when
}

type resources struct {
	LoadAvg1     float64 `json:"load_avg_1"`
	LoadAvg5     float64 `json:"load_avg_5"`
	LoadAvg15    float64 `json:"load_avg_15"`
	NumCPUs      int     `json:"num_cpus"`
	MemTotalKB   int64   `json:"mem_total_kb"`
	MemAvailKB   int64   `json:"mem_avail_kb"`
	MemUsedPct   float64 `json:"mem_used_pct"`
	DiskTotalGB  float64 `json:"disk_total_gb"`
	DiskFreeGB   float64 `json:"disk_free_gb"`
	DiskUsedPct  float64 `json:"disk_used_pct"`
	ConntrackCnt int64   `json:"conntrack_count"`
	ConntrackMax int64   `json:"conntrack_max"`
	UptimeSec    int64   `json:"uptime_sec"`
	BootTimeUnix int64   `json:"boot_time_unix"`
}

// Resources reads /proc + statfs for the box-level metrics card.
// All best-effort — missing files just leave the field at zero.
func (h *SystemHandler) Resources(c *gin.Context) {
	response.OK(c, computeLocalSystemResources())
}

// computeLocalSystemResources extrahiert die /proc + statfs reads aus
// dem Handler damit der Cluster-Aggregator denselben Snapshot ohne
// gin.Context erstellen kann. Best-effort: fehlende Quellen lassen
// die jeweiligen Felder einfach auf 0.
func computeLocalSystemResources() resources {
	r := resources{NumCPUs: runtime.NumCPU()}
	if data, err := os.ReadFile("/proc/loadavg"); err == nil {
		f := strings.Fields(string(data))
		if len(f) >= 3 {
			r.LoadAvg1, _ = strconv.ParseFloat(f[0], 64)
			r.LoadAvg5, _ = strconv.ParseFloat(f[1], 64)
			r.LoadAvg15, _ = strconv.ParseFloat(f[2], 64)
		}
	}
	if data, err := os.ReadFile("/proc/meminfo"); err == nil {
		s := bufio.NewScanner(strings.NewReader(string(data)))
		for s.Scan() {
			line := s.Text()
			fields := strings.Fields(line)
			if len(fields) < 2 {
				continue
			}
			val, _ := strconv.ParseInt(fields[1], 10, 64)
			switch strings.TrimSuffix(fields[0], ":") {
			case "MemTotal":
				r.MemTotalKB = val
			case "MemAvailable":
				r.MemAvailKB = val
			}
		}
		if r.MemTotalKB > 0 {
			r.MemUsedPct = float64(r.MemTotalKB-r.MemAvailKB) * 100 / float64(r.MemTotalKB)
		}
	}
	var fs syscall.Statfs_t
	if err := syscall.Statfs("/", &fs); err == nil {
		total := float64(fs.Blocks) * float64(fs.Bsize)
		free := float64(fs.Bavail) * float64(fs.Bsize)
		r.DiskTotalGB = total / 1024 / 1024 / 1024
		r.DiskFreeGB = free / 1024 / 1024 / 1024
		if total > 0 {
			r.DiskUsedPct = (total - free) * 100 / total
		}
	}
	if data, err := os.ReadFile("/proc/sys/net/netfilter/nf_conntrack_count"); err == nil {
		r.ConntrackCnt, _ = strconv.ParseInt(strings.TrimSpace(string(data)), 10, 64)
	}
	if data, err := os.ReadFile("/proc/sys/net/netfilter/nf_conntrack_max"); err == nil {
		r.ConntrackMax, _ = strconv.ParseInt(strings.TrimSpace(string(data)), 10, 64)
	}
	if data, err := os.ReadFile("/proc/uptime"); err == nil {
		f := strings.Fields(string(data))
		if len(f) >= 1 {
			if up, err := strconv.ParseFloat(f[0], 64); err == nil {
				r.UptimeSec = int64(up)
				r.BootTimeUnix = time.Now().Unix() - r.UptimeSec
			}
		}
	}
	return r
}

// Maintenance liefert den aktuellen Whole-Box-Maintenance-Status.
// Wenn Setup nicht initialisiert: enabled=false.
func (h *SystemHandler) Maintenance(c *gin.Context) {
	if h.Setup == nil {
		response.OK(c, gin.H{"enabled": false, "message": ""})
		return
	}
	st, err := h.Setup.Load()
	if err != nil || st == nil {
		response.OK(c, gin.H{"enabled": false, "message": ""})
		return
	}
	response.OK(c, gin.H{
		"enabled": st.MaintenanceMode,
		"message": st.MaintenanceMessage,
	})
}

// ToggleMaintenance schaltet whole-box maintenance an/aus.
// Persistiert in setup-State + triggert HAProxy-Render damit das
// 503-Block aktiv wird (oder verschwindet).
func (h *SystemHandler) ToggleMaintenance(c *gin.Context) {
	if h.Setup == nil {
		response.Err(c, http.StatusServiceUnavailable,
			simpleErr("setup not initialised"))
		return
	}
	var req struct {
		Enabled bool   `json:"enabled"`
		Message string `json:"message"`
	}
	if err := c.ShouldBindJSON(&req); err != nil {
		response.BadRequest(c, err)
		return
	}
	if err := h.Setup.SetMaintenanceMode(req.Enabled, req.Message); err != nil {
		response.Internal(c, err)
		return
	}
	if h.HAProxyReloader != nil {
		ctx, cancel := stdcontext.WithTimeout(c.Request.Context(), 10*time.Second)
		defer cancel()
		if err := h.HAProxyReloader(ctx); err != nil {
			slog.Warn("system: haproxy reload after maintenance toggle failed", "error", err)
		}
	}
	if h.Audit != nil {
		action := "system.maintenance.off"
		if req.Enabled {
			action = "system.maintenance.on"
		}
		_ = h.Audit.Log(c.Request.Context(), actorOf(c), action,
			"", gin.H{"enabled": req.Enabled, "message": req.Message}, h.NodeID)
	}
	response.OK(c, gin.H{"enabled": req.Enabled, "message": req.Message})
}

type simpleErr string

func (e simpleErr) Error() string { return string(e) }

// BackupRetention liefert keep_n. 0 = Default (backup.DefaultKeepN).
func (h *SystemHandler) BackupRetention(c *gin.Context) {
	keep := 0
	if h.Setup != nil {
		if st, err := h.Setup.Load(); err == nil && st != nil {
			keep = st.BackupRetentionKeep
		}
	}
	response.OK(c, gin.H{"keep": keep, "default": 14})
}

// SetBackupRetention persistiert die Operator-gewählte Retention.
// keep=0 → wieder Default, keep=1..365 → custom.
func (h *SystemHandler) SetBackupRetention(c *gin.Context) {
	if h.Setup == nil {
		response.Err(c, http.StatusServiceUnavailable, simpleErr("setup not initialised"))
		return
	}
	var req struct {
		Keep int `json:"keep"`
	}
	if err := c.ShouldBindJSON(&req); err != nil {
		response.BadRequest(c, err)
		return
	}
	if err := h.Setup.SetBackupRetention(req.Keep); err != nil {
		response.BadRequest(c, err)
		return
	}
	if h.Audit != nil {
		_ = h.Audit.Log(c.Request.Context(), actorOf(c), "system.backup_retention",
			"", gin.H{"keep": req.Keep}, h.NodeID)
	}
	response.OK(c, gin.H{"keep": req.Keep})
}

// AuditRetention liefert die konfigurierte Retention in Tagen.
// 0 = Default (90 — siehe scheduler/main.go).
func (h *SystemHandler) AuditRetention(c *gin.Context) {
	days := 0
	if h.Setup != nil {
		if st, err := h.Setup.Load(); err == nil && st != nil {
			days = st.AuditRetentionDays
		}
	}
	response.OK(c, gin.H{"days": days, "default": 90})
}

// SetAuditRetention setzt Audit-Retention in Tagen. 0..3650.
func (h *SystemHandler) SetAuditRetention(c *gin.Context) {
	if h.Setup == nil {
		response.Err(c, http.StatusServiceUnavailable, simpleErr("setup not initialised"))
		return
	}
	var req struct {
		Days int `json:"days"`
	}
	if err := c.ShouldBindJSON(&req); err != nil {
		response.BadRequest(c, err)
		return
	}
	if err := h.Setup.SetAuditRetention(req.Days); err != nil {
		response.BadRequest(c, err)
		return
	}
	if h.Audit != nil {
		_ = h.Audit.Log(c.Request.Context(), actorOf(c), "system.audit_retention",
			"", gin.H{"days": req.Days}, h.NodeID)
	}
	response.OK(c, gin.H{"days": req.Days})
}

// DBSize liefert pg_database_size + Top-N Tabellen-Größen für Capacity-
// Planning. Operator sieht so welche Tabellen Disk fressen (typisch
// audit_log + firewall_log → siehe Retention-Settings).
type dbSizeTable struct {
	Name    string `json:"name"`
	Bytes   int64  `json:"bytes"`
	HumanSz string `json:"human_size"`
}
type dbSizeResponse struct {
	TotalBytes int64         `json:"total_bytes"`
	HumanTotal string        `json:"human_total"`
	Tables     []dbSizeTable `json:"top_tables"`
}

func (h *SystemHandler) IPv6(c *gin.Context) {
	enabled := false
	if h.Setup != nil {
		if st, err := h.Setup.Load(); err == nil && st != nil {
			enabled = st.IPv6Enabled
		}
	}
	response.OK(c, gin.H{"enabled": enabled})
}

func (h *SystemHandler) SetIPv6(c *gin.Context) {
	if h.Setup == nil {
		response.Err(c, http.StatusServiceUnavailable, simpleErr("setup not initialised"))
		return
	}
	var req struct {
		Enabled bool `json:"enabled"`
	}
	if err := c.ShouldBindJSON(&req); err != nil {
		response.BadRequest(c, err)
		return
	}
	if err := h.Setup.SetIPv6Enabled(req.Enabled); err != nil {
		response.BadRequest(c, err)
		return
	}
	if h.HAProxyReloader != nil {
		_ = h.HAProxyReloader(c.Request.Context())
	}
	if h.Audit != nil {
		_ = h.Audit.Log(c.Request.Context(), actorOf(c), "system.ipv6",
			"", gin.H{"enabled": req.Enabled}, h.NodeID)
	}
	response.OK(c, gin.H{"enabled": req.Enabled})
}

func (h *SystemHandler) DBSize(c *gin.Context) {
	if h.Pool == nil {
		response.Err(c, http.StatusServiceUnavailable, simpleErr("db pool unavailable"))
		return
	}
	ctx, cancel := stdcontext.WithTimeout(c.Request.Context(), 3*time.Second)
	defer cancel()
	var total int64
	var totalH string
	if err := h.Pool.QueryRow(ctx, `
SELECT pg_database_size(current_database()),
       pg_size_pretty(pg_database_size(current_database()))`).Scan(&total, &totalH); err != nil {
		response.Internal(c, err)
		return
	}
	// Top 10 user-tables nach total_relation_size (inkl. Indizes + TOAST).
	rows, err := h.Pool.Query(ctx, `
SELECT c.relname,
       pg_total_relation_size(c.oid),
       pg_size_pretty(pg_total_relation_size(c.oid))
FROM pg_class c
JOIN pg_namespace n ON n.oid = c.relnamespace
WHERE c.relkind = 'r' AND n.nspname = 'public'
ORDER BY pg_total_relation_size(c.oid) DESC
LIMIT 10`)
	if err != nil {
		response.Internal(c, err)
		return
	}
	defer rows.Close()
	out := dbSizeResponse{TotalBytes: total, HumanTotal: totalH, Tables: []dbSizeTable{}}
	for rows.Next() {
		var t dbSizeTable
		if err := rows.Scan(&t.Name, &t.Bytes, &t.HumanSz); err != nil {
			response.Internal(c, err)
			return
		}
		out.Tables = append(out.Tables, t)
	}
	response.OK(c, out)
}

// HAProxyReload zwingt ein systemctl reload haproxy.service — nützlich
// wenn der Operator manuell in /etc/edgeguard/tls/ geschrieben hat
// (z. B. eigenes PEM per SSH kopiert) und HAProxy das neue Cert sehen
// soll, ohne eine UI-Mutation zu triggern die das automatisch täte.
func (h *SystemHandler) HAProxyReload(c *gin.Context) {
	out, err := exec.Command("sudo", "-n", "/usr/bin/systemctl", "reload", "haproxy.service").CombinedOutput()
	if err != nil {
		response.Err(c, http.StatusInternalServerError, simpleErr(strings.TrimSpace(string(out))+": "+err.Error()))
		return
	}
	if h.Audit != nil {
		_ = h.Audit.Log(c.Request.Context(), actorOf(c), "system.haproxy_reload",
			"", gin.H{}, h.NodeID)
	}
	response.OK(c, gin.H{"ok": true})
}

// restartAllowlist sind die Dienste die der Operator über die UI neu
// starten darf. edgeguard-api selbst ist bewusst ausgeschlossen (würde
// die eigene HTTP-Response killen). postgresql ebenfalls (Datenpfad).
var restartAllowlist = map[string]bool{
	"haproxy":   true,
	"squid":     true,
	"unbound":   true,
	"chrony":    true,
	"nftables":  true,
	"wireguard": true, // wireguard als Metadienst; einzelne wg-Ifaces über wg-quick@<name>
	"edgeguard-scheduler": true,
}

// ServiceRestart startet einen Dienst aus der Allowlist via
// `systemctl restart`. Gibt 400 zurück wenn der Dienst nicht auf der
// Allowlist steht, 500 wenn systemctl fehlschlägt.
func (h *SystemHandler) ServiceRestart(c *gin.Context) {
	var req struct {
		Service string `json:"service" binding:"required"`
	}
	if err := c.ShouldBindJSON(&req); err != nil {
		response.Err(c, http.StatusBadRequest, simpleErr("service required"))
		return
	}
	svc := strings.TrimSpace(req.Service)
	if !restartAllowlist[svc] {
		response.Err(c, http.StatusBadRequest, simpleErr("service not in allowlist: "+svc))
		return
	}
	unitName := svc + ".service"
	out, err := exec.Command("sudo", "-n", "/usr/bin/systemctl", "restart", unitName).CombinedOutput()
	if err != nil {
		response.Err(c, http.StatusInternalServerError, simpleErr(strings.TrimSpace(string(out))+": "+err.Error()))
		return
	}
	if h.Audit != nil {
		_ = h.Audit.Log(c.Request.Context(), actorOf(c), "system.service_restart",
			svc, gin.H{"service": svc}, h.NodeID)
	}
	response.OK(c, gin.H{"ok": true, "service": svc})
}

// RenderConfigs erzwingt ein Re-Render aller Service-Configs aus dem
// aktuellen DB-State. Läuft haproxy + alle ExtraReloaders (nftables,
// wireguard, squid, unbound, chrony) durch. Fehler werden gesammelt
// und als partial-Antwort zurückgegeben — erfolgreich gerenderte
// Dienste stehen in "rendered", fehlgeschlagene in "errors".
func (h *SystemHandler) RenderConfigs(c *gin.Context) {
	if h.HAProxyReloader == nil && len(h.ExtraReloaders) == 0 {
		response.Err(c, http.StatusServiceUnavailable, simpleErr("renderer not wired"))
		return
	}
	ctx, cancel := stdcontext.WithTimeout(c.Request.Context(), 30*time.Second)
	defer cancel()

	rendered := []string{}
	errs := map[string]string{}

	if h.HAProxyReloader != nil {
		if err := h.HAProxyReloader(ctx); err != nil {
			errs["haproxy"] = err.Error()
		} else {
			rendered = append(rendered, "haproxy")
		}
	}
	// Defined order so the audit log is deterministic.
	order := []string{"nftables", "wireguard", "squid", "unbound", "chrony"}
	for _, name := range order {
		fn, ok := h.ExtraReloaders[name]
		if !ok {
			continue
		}
		if err := fn(ctx); err != nil {
			errs[name] = err.Error()
		} else {
			rendered = append(rendered, name)
		}
	}
	if h.Audit != nil {
		_ = h.Audit.Log(c.Request.Context(), actorOf(c), "system.render_configs",
			"", gin.H{"rendered": rendered, "errors": errs}, h.NodeID)
	}
	response.OK(c, gin.H{"ok": len(errs) == 0, "rendered": rendered, "errors": errs})
}

// UpgradeStatus liefert den Status des letzten Self-Upgrade-Versuchs.
// Reads systemctl show + journalctl der edgeguard-upgrade.service
// transient unit. Hilft beim Debuggen wenn der Update-Banner nach
// einem Click nicht weg geht — typische Ursachen (apt-Resolver-fail,
// dpkg-broken) sind direkt im Log sichtbar.
type upgradeStatusResponse struct {
	State       string   `json:"state"`         // inactive / activating / failed / ...
	Result      string   `json:"result"`        // success / exit-code / ...
	ExecMainPID int      `json:"exec_main_pid"` // 0 wenn nie gelaufen
	ExitCode    int      `json:"exit_code"`     // exit-status des letzten Laufs
	StartedAt   string   `json:"started_at"`    // RFC3339 oder leer
	FinishedAt  string   `json:"finished_at"`   // RFC3339 oder leer
	Log         []string `json:"log"`           // letzte N Zeilen aus journalctl
}

func (h *SystemHandler) UpgradeStatus(c *gin.Context) {
	out := upgradeStatusResponse{Log: []string{}}

	// systemctl show liefert key=value pairs für die transient unit.
	// Wenn die Unit nie existiert hat → leeres Output / inactive.
	if data, err := exec.Command("systemctl", "show", "edgeguard-upgrade.service",
		"--no-page",
		"-p", "ActiveState",
		"-p", "Result",
		"-p", "ExecMainPID",
		"-p", "ExecMainStatus",
		"-p", "ExecMainStartTimestamp",
		"-p", "ExecMainExitTimestamp",
	).CombinedOutput(); err == nil {
		for _, line := range strings.Split(string(data), "\n") {
			kv := strings.SplitN(strings.TrimSpace(line), "=", 2)
			if len(kv) != 2 {
				continue
			}
			switch kv[0] {
			case "ActiveState":
				out.State = kv[1]
			case "Result":
				out.Result = kv[1]
			case "ExecMainPID":
				out.ExecMainPID, _ = strconv.Atoi(kv[1])
			case "ExecMainStatus":
				out.ExitCode, _ = strconv.Atoi(kv[1])
			case "ExecMainStartTimestamp":
				if t, err := time.Parse("Mon 2006-01-02 15:04:05 MST", kv[1]); err == nil {
					out.StartedAt = t.UTC().Format(time.RFC3339)
				}
			case "ExecMainExitTimestamp":
				if t, err := time.Parse("Mon 2006-01-02 15:04:05 MST", kv[1]); err == nil {
					out.FinishedAt = t.UTC().Format(time.RFC3339)
				}
			}
		}
	}

	// Letzte 200 Zeilen Journal — reicht für apt-output + Stack-Traces.
	if data, err := exec.Command("journalctl",
		"-u", "edgeguard-upgrade.service",
		"--no-pager", "-n", "200", "-o", "cat",
	).CombinedOutput(); err == nil {
		lines := strings.Split(strings.TrimRight(string(data), "\n"), "\n")
		// Leere "no entries"-Antwort als leeres Log zurückgeben.
		if !(len(lines) == 1 && (lines[0] == "" || strings.HasPrefix(lines[0], "-- No entries"))) {
			out.Log = lines
		}
	}

	response.OK(c, out)
}

// AutoUpdate liefert den aktuellen Status (Conf-File existiert?).
func (h *SystemHandler) AutoUpdate(c *gin.Context) {
	response.OK(c, gin.H{"enabled": aptsvc.AutoUpdateEnabled()})
}

// ToggleAutoUpdate schaltet automatische Updates an/aus. Schreibt
// /etc/apt/apt.conf.d/52edgeguard-auto-updates (sudo tee) bzw.
// entfernt es (sudo rm). Idempotent.
func (h *SystemHandler) ToggleAutoUpdate(c *gin.Context) {
	var req struct {
		Enabled bool `json:"enabled"`
	}
	if err := c.ShouldBindJSON(&req); err != nil {
		response.BadRequest(c, err)
		return
	}
	if err := aptsvc.SetAutoUpdate(req.Enabled); err != nil {
		response.Internal(c, err)
		return
	}
	if h.Audit != nil {
		action := "system.auto_update.off"
		if req.Enabled {
			action = "system.auto_update.on"
		}
		_ = h.Audit.Log(c.Request.Context(), actorOf(c), action,
			"", gin.H{"enabled": req.Enabled}, h.NodeID)
	}
	response.OK(c, gin.H{"enabled": req.Enabled})
}

// ConfigPreview rendert eine Service-Config aus dem aktuellen DB-State
// und gibt sie als Plain-Text zurück — ohne sie auf Disk zu schreiben
// oder den Dienst zu reloaden. Query-Parameter: ?generator=haproxy|nftables|squid|unbound
func (h *SystemHandler) ConfigPreview(c *gin.Context) {
	gen := strings.TrimSpace(c.Query("generator"))
	if gen == "" {
		names := make([]string, 0, len(h.ConfigPreviewers))
		for k := range h.ConfigPreviewers {
			names = append(names, k)
		}
		response.OK(c, gin.H{"generators": names})
		return
	}
	if h.ConfigPreviewers == nil {
		response.Err(c, http.StatusServiceUnavailable, simpleErr("previewers not wired"))
		return
	}
	fn, ok := h.ConfigPreviewers[gen]
	if !ok {
		response.Err(c, http.StatusBadRequest, simpleErr("unknown generator: "+gen))
		return
	}
	ctx, cancel := stdcontext.WithTimeout(c.Request.Context(), 5*time.Second)
	defer cancel()
	content, err := fn(ctx)
	if err != nil {
		response.Internal(c, err)
		return
	}
	response.OK(c, gin.H{"generator": gen, "content": content})
}

func (h *SystemHandler) Health(c *gin.Context) {
	resp := gin.H{
		"status":  "ok",
		"version": h.Version,
	}
	if hn, err := os.Hostname(); err == nil {
		resp["hostname"] = hn
	}
	if data, err := os.ReadFile("/proc/version"); err == nil {
		// /proc/version: "Linux version 6.x.y (...)" — Kurzform: alles
		// bis zum ersten '(' kürzen.
		line := strings.TrimSpace(string(data))
		if idx := strings.Index(line, " ("); idx > 0 {
			line = strings.TrimSpace(line[:idx])
		}
		resp["kernel"] = line
	}
	if data, err := os.ReadFile("/etc/os-release"); err == nil {
		for _, line := range strings.Split(string(data), "\n") {
			if k, v, ok := strings.Cut(line, "="); ok && k == "PRETTY_NAME" {
				resp["os"] = strings.Trim(v, `"`)
				break
			}
		}
	}
	response.OK(c, resp)
}

// PackageVersions reports installed and available versions for the
// edgeguard-* APT packages. Called by the UI's update banner — it
// polls every 30s and lights up when available > installed.
//
// Implementierung delegiert an internal/services/apt: dort sitzt der
// 5-min-Throttle für apt-get update, der LC_ALL=C-Fix für deutsche
// Locales und das Background-Refresh-Timer. `?force=1` (UI-Button
// „Jetzt prüfen") bypassed den Throttle, damit nach einem `make publish`
// nicht 5 min auf das nächste Tick gewartet werden muss.
func (h *SystemHandler) PackageVersions(c *gin.Context) {
	force := c.Query("force") == "1" || c.Query("force") == "true"
	out := aptsvc.PackageVersions(c.Request.Context(), force)
	response.OK(c, out)
}

// Upgrade runs the apt upgrade detached via systemd-run so the API
// can reply BEFORE the package replaces it. Pattern from netcell-
// webpanel/management-agent/internal/handlers/update.go (see
// architecture.md §11). Without --collect on a transient service
// unit, the apt-get child dies when systemd-cleans up the scope as
// the API exits — leaves the box half-upgraded.
func (h *SystemHandler) Upgrade(c *gin.Context) {
	slog.Info("starting package upgrade (detached)")

	// Skript landet NICHT in /tmp — edgeguard-api.service hat
	// PrivateTmp=true und sieht damit ein eigenes /tmp, das die
	// per `sudo systemd-run` gestartete Transient-Unit nicht sieht.
	// /var/lib/edgeguard ist edgeguard-owned + persistent + von
	// beiden Namespaces aus zugänglich.
	const scriptPath = "/var/lib/edgeguard/upgrade.sh"
	// Retry-Logik gegen Gitea-Packages.gz-Race: nach einem frischen
	// Publish kann der Packages-Index für ein paar Sekunden inkonsistent
	// sein (z. B. Meta uploaded, api/ui noch nicht in der regenerierten
	// Index-Datei) → apt-resolver-fail mit "no choices are installable".
	// Drei Versuche mit 15s/30s Backoff geben Gitea Zeit den Index
	// nachzuziehen. Befund 2026-05-17.
	const script = `#!/bin/bash
set -e
sleep 2
export DEBIAN_FRONTEND=noninteractive
echo "[upgrade] dpkg --configure -a"
dpkg --configure -a || true

retry_apt() {
    local attempt=0
    local max=3
    local wait_for=15
    while [ $attempt -lt $max ]; do
        attempt=$((attempt + 1))
        echo "[upgrade] attempt $attempt/$max: apt-get update + install"
        apt-get update -qq || true
        if apt-get install -y -qq -o Dpkg::Options::=--force-confold \
            edgeguard-api edgeguard-ui edgeguard; then
            return 0
        fi
        if [ $attempt -lt $max ]; then
            echo "[upgrade] failed (likely Packages-index race), waiting ${wait_for}s before retry"
            sleep $wait_for
            wait_for=$((wait_for * 2))
        fi
    done
    echo "[upgrade] all $max attempts failed"
    return 1
}

retry_apt
echo "[upgrade] complete"
rm -f /var/lib/edgeguard/upgrade.sh
`
	if err := os.WriteFile(scriptPath, []byte(script), 0o755); err != nil {
		response.Internal(c, err)
		return
	}

	const unitName = "edgeguard-upgrade.service"
	// API läuft als edgeguard-User; systemd-run + systemctl reset-failed
	// brauchen root. Sudoers-Whitelist in postinst lässt exakt diese
	// beiden Aufrufe durch. Ohne sudo schlug das früher mit
	// "Interactive authentication required" fehl und der Fallback
	// (setsid bash als edgeguard) konnte kein apt-get update — das
	// Modal blieb hängen und die Box nicht aktualisiert.
	_ = exec.Command("sudo", "-n", "/usr/bin/systemctl", "reset-failed", unitName).Run()
	cmd := exec.Command("sudo", "-n", "/usr/bin/systemd-run",
		"--unit="+unitName,
		"--description=EdgeGuard self-upgrade",
		"--collect",
		"bash", scriptPath)
	if err := cmd.Run(); err != nil {
		// systemd-run unavailable (dev env without sudo) — fall back
		// to setsid. In Prod sollte das nie greifen.
		slog.Warn("upgrade: sudo systemd-run failed, falling back to setsid", "error", err)
		fallback := exec.Command("setsid", "bash", scriptPath)
		fallback.SysProcAttr = &syscall.SysProcAttr{Setpgid: true}
		if err2 := fallback.Start(); err2 != nil {
			response.Internal(c, err2)
			return
		}
		_ = fallback.Process.Release()
	}

	if h.Audit != nil {
		_ = h.Audit.Log(c.Request.Context(), actorOf(c), "system.upgrade",
			"", gin.H{"unit": unitName}, h.NodeID)
	}
	c.JSON(http.StatusAccepted, response.Envelope{
		Data:    gin.H{"status": "upgrading", "unit": unitName},
		Error:   nil,
		Message: "upgrade started",
	})
}

// addrInfo + interfaceInfo mirror the relevant subset of `ip -j addr
// show` so the frontend keeps its existing parsing code.
type addrInfo struct {
	Family    string `json:"family"` // "inet" | "inet6"
	Local     string `json:"local"`
	PrefixLen int    `json:"prefixlen"`
}

type interfaceInfo struct {
	IfIndex   int        `json:"ifindex"`
	IfName    string     `json:"ifname"`
	Flags     []string   `json:"flags"`
	MTU       int        `json:"mtu"`
	LinkType  string     `json:"link_type,omitempty"`
	Address   string     `json:"address,omitempty"`
	AddrInfo  []addrInfo `json:"addr_info"`
	RxBytes   int64      `json:"rx_bytes"`
	TxBytes   int64      `json:"tx_bytes"`
	RxPackets int64      `json:"rx_packets"`
	TxPackets int64      `json:"tx_packets"`
	RxDrop    int64      `json:"rx_drop"`
	TxDrop    int64      `json:"tx_drop"`
}

type netDevStat struct {
	RxBytes, TxBytes     int64
	RxPackets, TxPackets int64
	RxDrop, TxDrop       int64
}

// readNetDevStats parses /proc/net/dev and returns per-interface counters.
// Missing or unreadable: returns empty map (caller gets zero-value stats).
func readNetDevStats() map[string]netDevStat {
	out := map[string]netDevStat{}
	data, err := os.ReadFile("/proc/net/dev")
	if err != nil {
		return out
	}
	scanner := bufio.NewScanner(strings.NewReader(string(data)))
	for scanner.Scan() {
		line := strings.TrimSpace(scanner.Text())
		idx := strings.Index(line, ":")
		if idx < 0 {
			continue
		}
		name := strings.TrimSpace(line[:idx])
		fields := strings.Fields(line[idx+1:])
		if len(fields) < 12 {
			continue
		}
		p := func(s string) int64 { n, _ := strconv.ParseInt(s, 10, 64); return n }
		out[name] = netDevStat{
			RxBytes:   p(fields[0]),
			RxPackets: p(fields[1]),
			RxDrop:    p(fields[3]),
			TxBytes:   p(fields[8]),
			TxPackets: p(fields[9]),
			TxDrop:    p(fields[11]),
		}
	}
	return out
}

// Interfaces enumerates the kernel-side network interfaces using
// Go's net.Interfaces() — no shell-out, no AF_NETLINK exception
// in the systemd hardening required (the original `ip -j addr`
// approach was blocked by RestrictAddressFamilies).
//
// Output shape mirrors `ip -j addr show` enough for the UI's
// Networks "system-discovered" card.
func (h *SystemHandler) Interfaces(c *gin.Context) {
	ifaces, err := net.Interfaces()
	if err != nil {
		slog.Warn("system/interfaces: net.Interfaces failed", "error", err)
		response.OK(c, gin.H{"interfaces": []interfaceInfo{}})
		return
	}
	devStats := readNetDevStats()
	out := make([]interfaceInfo, 0, len(ifaces))
	for _, ifc := range ifaces {
		st := devStats[ifc.Name]
		info := interfaceInfo{
			IfIndex:   ifc.Index,
			IfName:    ifc.Name,
			MTU:       ifc.MTU,
			Address:   ifc.HardwareAddr.String(),
			LinkType:  classifyLinkType(ifc),
			Flags:     flagsToList(ifc.Flags),
			AddrInfo:  []addrInfo{},
			RxBytes:   st.RxBytes,
			TxBytes:   st.TxBytes,
			RxPackets: st.RxPackets,
			TxPackets: st.TxPackets,
			RxDrop:    st.RxDrop,
			TxDrop:    st.TxDrop,
		}
		addrs, err := ifc.Addrs()
		if err != nil {
			out = append(out, info)
			continue
		}
		for _, a := range addrs {
			ipnet, ok := a.(*net.IPNet)
			if !ok {
				continue
			}
			family := "inet"
			if ipnet.IP.To4() == nil {
				family = "inet6"
			}
			ones, _ := ipnet.Mask.Size()
			info.AddrInfo = append(info.AddrInfo, addrInfo{
				Family:    family,
				Local:     ipnet.IP.String(),
				PrefixLen: ones,
			})
		}
		out = append(out, info)
	}
	response.OK(c, gin.H{"interfaces": out})
}

func classifyLinkType(ifc net.Interface) string {
	if ifc.Flags&net.FlagLoopback != 0 {
		return "loopback"
	}
	if len(ifc.HardwareAddr) > 0 {
		return "ether"
	}
	return ""
}

// VIPStatus returns the VRRP state and active VIPs for this node.
// Uses net.Interfaces() (no shell-out) to check which VIPs from
// ip_addresses WHERE is_vip=true are currently assigned locally.
// MASTER = at least one VIP is locally present; BACKUP = none present.
func (h *SystemHandler) VIPStatus(c *gin.Context) {
	type vipEntry struct {
		Address string `json:"address"`
		Prefix  int    `json:"prefix"`
		Device  string `json:"device"`
		Active  bool   `json:"active"`
	}
	type vipStatus struct {
		VRRPState        string     `json:"vrrp_state"`
		KeepalivedActive bool       `json:"keepalived_active"`
		VIPs             []vipEntry `json:"vips"`
	}

	ctx := c.Request.Context()

	// keepalived service active?
	kaOut, _ := exec.CommandContext(ctx, "systemctl", "is-active", "keepalived").Output()
	kaActive := strings.TrimSpace(string(kaOut)) == "active"

	// query VIPs from DB
	var dbVIPs []vipEntry
	if h.Pool != nil {
		rows, err := h.Pool.Query(ctx,
			`SELECT a.address, a.prefix, COALESCE(i.name,'') AS device
			   FROM ip_addresses a
			   LEFT JOIN network_interfaces i ON i.id = a.interface_id
			  WHERE a.is_vip = true AND a.active = true
			  ORDER BY a.address`)
		if err == nil {
			defer rows.Close()
			for rows.Next() {
				var e vipEntry
				if err2 := rows.Scan(&e.Address, &e.Prefix, &e.Device); err2 == nil {
					dbVIPs = append(dbVIPs, e)
				}
			}
		}
	}

	// build set of locally assigned IPs
	localIPs := make(map[string]bool)
	if ifaces, err := net.Interfaces(); err == nil {
		for _, ifc := range ifaces {
			if addrs, err2 := ifc.Addrs(); err2 == nil {
				for _, a := range addrs {
					if ipnet, ok := a.(*net.IPNet); ok {
						localIPs[ipnet.IP.String()] = true
					}
				}
			}
		}
	}

	anyActive := false
	for i := range dbVIPs {
		dbVIPs[i].Active = localIPs[dbVIPs[i].Address]
		if dbVIPs[i].Active {
			anyActive = true
		}
	}

	state := "UNKNOWN"
	if kaActive {
		if anyActive {
			state = "MASTER"
		} else {
			state = "BACKUP"
		}
	}

	response.OK(c, vipStatus{
		VRRPState:        state,
		KeepalivedActive: kaActive,
		VIPs:             dbVIPs,
	})
}

func flagsToList(f net.Flags) []string {
	var out []string
	if f&net.FlagUp != 0 {
		out = append(out, "UP")
	}
	if f&net.FlagBroadcast != 0 {
		out = append(out, "BROADCAST")
	}
	if f&net.FlagLoopback != 0 {
		out = append(out, "LOOPBACK")
	}
	if f&net.FlagPointToPoint != 0 {
		out = append(out, "POINTOPOINT")
	}
	if f&net.FlagMulticast != 0 {
		out = append(out, "MULTICAST")
	}
	if f&net.FlagRunning != 0 {
		out = append(out, "LOWER_UP")
	}
	return out
}