DebTik/internal/vrrp/manager.go

package vrrp

import (
	"encoding/json"
	"fmt"
	"net"
	"os"
	"os/exec"
	"path/filepath"
	"sort"
	"strings"
	"sync"
	"time"

	"github.com/vishvananda/netlink"
)

const (
	VRRPStorePath        = "/etc/debtik/vrrp.json"
	KeepalivedConfPath   = "/etc/keepalived/keepalived.conf"
	VRRPStatusDir        = "/run/debtik/vrrp"
	VRRPNotifyScriptPath = "/usr/local/bin/vrrp-notify.sh"
)

type Manager struct {
	mu        sync.RWMutex
	instances map[string]VRRPInstance
	storePath string
	confPath  string
}

func NewManager() *Manager {
	m := &Manager{
		instances: make(map[string]VRRPInstance),
		storePath: VRRPStorePath,
		confPath:  KeepalivedConfPath,
	}

	_ = os.MkdirAll(filepath.Dir(m.storePath), 0755)
	_ = os.MkdirAll(filepath.Dir(m.confPath), 0755)
	_ = os.MkdirAll(VRRPStatusDir, 0755)

	m.ensureNotifyScript()
	m.load()

	// Apply config on startup if instances exist
	_ = m.applyConfigLocked()

	return m
}

func (m *Manager) ensureNotifyScript() {
	scriptContent := `#!/bin/bash
STATE="$1"
NAME="$2"
mkdir -p /run/debtik/vrrp
echo "$STATE $(date -Iseconds)" > "/run/debtik/vrrp/${NAME}.state"
`
	_ = os.WriteFile(VRRPNotifyScriptPath, []byte(scriptContent), 0755)
}

func (m *Manager) load() {
	m.mu.Lock()
	defer m.mu.Unlock()

	data, err := os.ReadFile(m.storePath)
	if err != nil {
		return
	}

	var store VRRPConfigStore
	if err := json.Unmarshal(data, &store); err != nil {
		return
	}

	for _, inst := range store.Instances {
		m.instances[inst.ID] = inst
	}
}

func (m *Manager) saveLocked() error {
	store := VRRPConfigStore{
		Instances: make([]VRRPInstance, 0, len(m.instances)),
	}
	for _, inst := range m.instances {
		store.Instances = append(store.Instances, inst)
	}

	data, err := json.MarshalIndent(store, "", "  ")
	if err != nil {
		return err
	}
	return os.WriteFile(m.storePath, data, 0644)
}

func (m *Manager) List() []VRRPInstance {
	m.mu.RLock()
	defer m.mu.RUnlock()

	isDaemonActive := m.isKeepalivedRunning()

	list := make([]VRRPInstance, 0, len(m.instances))
	for _, inst := range m.instances {
		copyInst := inst
		copyInst.VirtualMAC = fmt.Sprintf("00:00:5E:00:01:%02X", copyInst.VRID)

		if copyInst.Disabled {
			copyInst.State = "DISABLED"
			copyInst.Running = false
		} else {
			copyInst.State, copyInst.Running, copyInst.LastTransition = m.detectInstanceState(copyInst, isDaemonActive)
		}

		list = append(list, copyInst)
	}

	sort.Slice(list, func(i, j int) bool {
		if list[i].Interface != list[j].Interface {
			return list[i].Interface < list[j].Interface
		}
		return list[i].VRID < list[j].VRID
	})

	return list
}

func (m *Manager) detectInstanceState(inst VRRPInstance, daemonActive bool) (string, bool, time.Time) {
	link, err := netlink.LinkByName(inst.Interface)
	if err != nil || link.Attrs().OperState == netlink.OperDown {
		return "FAULT", false, time.Time{}
	}

	// 1. Direct Kernel Check: Does this interface currently hold any of the virtual IPs?
	if len(inst.VirtualIPs) > 0 {
		addrs, err := netlink.AddrList(link, netlink.FAMILY_V4)
		if err == nil {
			for _, a := range addrs {
				for _, vip := range inst.VirtualIPs {
					cleanVIP := strings.Split(vip, "/")[0]
					if a.IP.String() == cleanVIP {
						return "MASTER", true, time.Time{}
					}
				}
			}
		}
	}

	// 2. Check notify state file
	stateFile := filepath.Join(VRRPStatusDir, fmt.Sprintf("%s.state", sanitizeName(inst.Name)))
	if data, err := os.ReadFile(stateFile); err == nil {
		parts := strings.Fields(string(data))
		if len(parts) >= 1 {
			st := parts[0]
			var t time.Time
			if len(parts) >= 2 {
				t, _ = time.Parse(time.RFC3339, parts[1])
			}
			if st == "MASTER" {
				return "MASTER", true, t
			}
			if st == "FAULT" {
				return "FAULT", false, t
			}
			if st == "BACKUP" && daemonActive {
				return "BACKUP", true, t
			}
		}
	}

	if daemonActive {
		return "BACKUP", true, time.Time{}
	}

	return "FAULT", false, time.Time{}
}

func (m *Manager) isKeepalivedRunning() bool {
	out, err := exec.Command("systemctl", "is-active", "keepalived").Output()
	return err == nil && strings.TrimSpace(string(out)) == "active"
}

func (m *Manager) Get(id string) (VRRPInstance, bool) {
	m.mu.RLock()
	defer m.mu.RUnlock()

	inst, ok := m.instances[id]
	if !ok {
		return VRRPInstance{}, false
	}
	inst.VirtualMAC = fmt.Sprintf("00:00:5E:00:01:%02X", inst.VRID)
	if inst.Disabled {
		inst.State = "DISABLED"
		inst.Running = false
	} else {
		inst.State, inst.Running, inst.LastTransition = m.detectInstanceState(inst, m.isKeepalivedRunning())
	}
	return inst, true
}

func (m *Manager) Save(inst VRRPInstance) (VRRPInstance, error) {
	m.mu.Lock()
	defer m.mu.Unlock()

	if inst.Name == "" || inst.Interface == "" {
		return inst, fmt.Errorf("name and interface are required")
	}
	if inst.VRID < 1 || inst.VRID > 255 {
		return inst, fmt.Errorf("VRID must be between 1 and 255")
	}
	if inst.Priority < 1 || inst.Priority > 255 {
		inst.Priority = 100
	}
	if inst.Interval < 1 {
		inst.Interval = 1
	}
	if inst.Version != 2 && inst.Version != 3 {
		inst.Version = 3
	}
	if len(inst.VirtualIPs) == 0 {
		return inst, fmt.Errorf("at least one Virtual IP (VIP) is required")
	}

	// Validate VIP formatting
	for _, vip := range inst.VirtualIPs {
		ipStr := strings.Split(vip, "/")[0]
		if net.ParseIP(strings.TrimSpace(ipStr)) == nil {
			return inst, fmt.Errorf("invalid virtual IP address: %s", vip)
		}
	}

	if inst.ID == "" {
		inst.ID = fmt.Sprintf("vrrp-%s", sanitizeName(inst.Name))
	}

	m.instances[inst.ID] = inst
	if err := m.saveLocked(); err != nil {
		return inst, err
	}

	if err := m.applyConfigLocked(); err != nil {
		return inst, err
	}

	return inst, nil
}

func (m *Manager) Delete(id string) error {
	m.mu.Lock()
	defer m.mu.Unlock()

	inst, ok := m.instances[id]
	if !ok {
		return fmt.Errorf("VRRP instance not found: %s", id)
	}

	delete(m.instances, id)
	_ = os.Remove(filepath.Join(VRRPStatusDir, fmt.Sprintf("%s.state", sanitizeName(inst.Name))))

	if err := m.saveLocked(); err != nil {
		return err
	}

	return m.applyConfigLocked()
}

func (m *Manager) ToggleState(id string, disabled bool) error {
	m.mu.Lock()
	defer m.mu.Unlock()

	inst, ok := m.instances[id]
	if !ok {
		return fmt.Errorf("VRRP instance not found: %s", id)
	}

	inst.Disabled = disabled
	m.instances[id] = inst

	if err := m.saveLocked(); err != nil {
		return err
	}

	return m.applyConfigLocked()
}

func (m *Manager) GetSummary() VRRPSummary {
	list := m.List()
	s := VRRPSummary{
		TotalInstances: len(list),
	}
	for _, inst := range list {
		if inst.State == "MASTER" {
			s.MasterInstances++
		} else if inst.State == "BACKUP" {
			s.BackupInstances++
		} else if inst.State == "FAULT" {
			s.FaultInstances++
		}
	}
	return s
}

func (m *Manager) applyConfigLocked() error {
	m.ensureNotifyScript()

	activeCount := 0
	for _, inst := range m.instances {
		if !inst.Disabled {
			activeCount++
		}
	}

	if activeCount == 0 {
		_ = exec.Command("systemctl", "stop", "keepalived").Run()
		_ = os.Remove(m.confPath)
		return nil
	}

	var sb strings.Builder
	sb.WriteString("# ==================================================================\n")
	sb.WriteString("# Enterprise VRRP (Virtual Router Redundancy Protocol) Configuration\n")
	sb.WriteString("# Managed by DebTik Network Appliance Controller. DO NOT EDIT DIRECTLY.\n")
	sb.WriteString("# ==================================================================\n\n")

	sb.WriteString("global_defs {\n")
	sb.WriteString("    router_id debtik\n")
	sb.WriteString("    vrrp_version 3\n")
	sb.WriteString("    script_user root\n")
	sb.WriteString("    enable_script_security\n")
	sb.WriteString("}\n\n")

	for _, inst := range m.instances {
		if inst.Disabled {
			continue
		}

		instTag := sanitizeName(inst.Name)
		sb.WriteString(fmt.Sprintf("# --- VRRP Instance: %s on %s (VRID: %d) ---\n", inst.Name, inst.Interface, inst.VRID))
		sb.WriteString(fmt.Sprintf("vrrp_instance %s {\n", instTag))
		sb.WriteString("    state BACKUP\n")
		sb.WriteString(fmt.Sprintf("    interface %s\n", inst.Interface))
		sb.WriteString(fmt.Sprintf("    virtual_router_id %d\n", inst.VRID))
		sb.WriteString(fmt.Sprintf("    priority %d\n", inst.Priority))
		sb.WriteString(fmt.Sprintf("    advert_int %d\n", inst.Interval))

		if inst.Preempt {
			sb.WriteString("    preempt\n")
		} else {
			sb.WriteString("    nopreempt\n")
		}

		if inst.Version == 2 {
			sb.WriteString("    version 2\n")
		}

		if inst.AuthType == "pass" && inst.AuthPassword != "" {
			sb.WriteString("    authentication {\n")
			sb.WriteString("        auth_type PASS\n")
			sb.WriteString(fmt.Sprintf("        auth_pass %s\n", inst.AuthPassword))
			sb.WriteString("    }\n")
		}

		sb.WriteString("    virtual_ipaddress {\n")
		for _, vip := range inst.VirtualIPs {
			vipTrim := strings.TrimSpace(vip)
			if vipTrim != "" {
				if strings.Contains(vipTrim, "/") {
					sb.WriteString(fmt.Sprintf("        %s dev %s\n", vipTrim, inst.Interface))
				} else {
					sb.WriteString(fmt.Sprintf("        %s/32 dev %s\n", vipTrim, inst.Interface))
				}
			}
		}
		sb.WriteString("    }\n")

		sb.WriteString(fmt.Sprintf("    notify_master \"%s MASTER %s\"\n", VRRPNotifyScriptPath, instTag))
		sb.WriteString(fmt.Sprintf("    notify_backup \"%s BACKUP %s\"\n", VRRPNotifyScriptPath, instTag))
		sb.WriteString(fmt.Sprintf("    notify_fault  \"%s FAULT %s\"\n", VRRPNotifyScriptPath, instTag))
		sb.WriteString(fmt.Sprintf("    notify_stop   \"%s STOP %s\"\n", VRRPNotifyScriptPath, instTag))
		sb.WriteString("}\n\n")
	}

	if err := os.WriteFile(m.confPath, []byte(sb.String()), 0644); err != nil {
		return fmt.Errorf("failed to write keepalived.conf: %w", err)
	}

	cmd := exec.Command("systemctl", "restart", "keepalived")
	if out, err := cmd.CombinedOutput(); err != nil {
		return fmt.Errorf("failed to restart keepalived: %s", string(out))
	}

	return nil
}

// RenameInterface updates VRRP instance interface bindings when an interface is renamed.
func (m *Manager) RenameInterface(oldName, newName string) {
	m.mu.Lock()
	defer m.mu.Unlock()

	changed := false
	for id, inst := range m.instances {
		if inst.Interface == oldName {
			inst.Interface = newName
			m.instances[id] = inst
			changed = true
		}
	}
	if changed {
		_ = m.saveLocked()
		_ = m.applyConfigLocked()
	}
}

func sanitizeName(n string) string {
	var res strings.Builder
	for _, r := range n {
		if (r >= 'a' && r <= 'z') || (r >= 'A' && r <= 'Z') || (r >= '0' && r <= '9') || r == '_' {
			res.WriteRune(r)
		} else {
			res.WriteRune('_')
		}
	}
	s := res.String()
	if s == "" {
		return "vrrp1"
	}
	return s
}