NexusGate/agent-go/health.go

package main

import (
	"math"
	"os"
	"path/filepath"
	"regexp"
	"runtime"
	"strconv"
	"strings"
	"sync"
	"syscall"
	"time"
)

// ---- systemInfo --------------------------------------------------------------

type systemReport struct {
	Hostname        string  `json:"hostname"`
	Platform        string  `json:"platform"`
	Release         string  `json:"release"`
	Arch            string  `json:"arch"`
	UptimeSeconds   int64   `json:"uptimeSeconds"`
	Load1           float64 `json:"load1"`
	CPUCount        int     `json:"cpuCount"`
	MemoryTotal     uint64  `json:"memoryTotal"`
	MemoryFree      uint64  `json:"memoryFree"`
	MemoryAvailable uint64  `json:"memoryAvailable"`
	DiskTotal       *uint64 `json:"diskTotal"`
	DiskAvailable   *uint64 `json:"diskAvailable"`
}

// nodeArch maps Go's GOARCH to Node's os.arch() vocabulary (amd64 → x64 etc).
func nodeArch() string {
	switch runtime.GOARCH {
	case "amd64":
		return "x64"
	case "386":
		return "ia32"
	case "arm64", "arm", "ppc64le", "s390x", "riscv64", "loong64", "mips64le":
		return runtime.GOARCH
	default:
		return runtime.GOARCH
	}
}

var memAvailableRe = regexp.MustCompile(`(?m)^MemAvailable:\s+(\d+) kB`)

func systemInfo() systemReport {
	// Node's os.totalmem()/freemem() use sysinfo(2) on Linux; do the same.
	var si syscall.Sysinfo_t
	var memTotal, memFree uint64
	if err := syscall.Sysinfo(&si); err == nil {
		memTotal = si.Totalram * uint64(si.Unit)
		memFree = si.Freeram * uint64(si.Unit)
	}
	memAvailable := memFree
	if data, err := os.ReadFile("/proc/meminfo"); err == nil {
		if m := memAvailableRe.FindSubmatch(data); m != nil {
			if kb, err := strconv.ParseUint(string(m[1]), 10, 64); err == nil {
				memAvailable = kb * 1024
			}
		}
	}
	var uptime int64
	if data, err := os.ReadFile("/proc/uptime"); err == nil {
		fields := strings.Fields(string(data))
		if len(fields) > 0 {
			if seconds, err := strconv.ParseFloat(fields[0], 64); err == nil {
				uptime = int64(math.Round(seconds))
			}
		}
	}
	var load1 float64
	if data, err := os.ReadFile("/proc/loadavg"); err == nil {
		fields := strings.Fields(string(data))
		if len(fields) > 0 {
			if v, err := strconv.ParseFloat(fields[0], 64); err == nil {
				load1 = math.Round(v*100) / 100 // Number(loadavg[0].toFixed(2))
			}
		}
	}
	release := ""
	if data, err := os.ReadFile("/proc/sys/kernel/osrelease"); err == nil {
		release = strings.TrimSpace(string(data))
	}
	hostname, _ := os.Hostname()
	report := systemReport{
		Hostname:        hostname,
		Platform:        runtime.GOOS,
		Release:         release,
		Arch:            nodeArch(),
		UptimeSeconds:   uptime,
		Load1:           load1,
		CPUCount:        runtime.NumCPU(),
		MemoryTotal:     memTotal,
		MemoryFree:      memFree,
		MemoryAvailable: memAvailable,
	}
	var fs syscall.Statfs_t
	if err := syscall.Statfs("/", &fs); err == nil {
		total := fs.Blocks * uint64(fs.Bsize)
		avail := fs.Bavail * uint64(fs.Bsize)
		report.DiskTotal = &total
		report.DiskAvailable = &avail
	}
	return report
}

// ---- installedCertificates ---------------------------------------------------

type certCache struct {
	fingerprint string
	checkedAt   time.Time
	domains     []string
}

var (
	certCacheMu sync.Mutex
	certCacheV  certCache
)

func opensslOK(timeout time.Duration, args ...string) (string, bool) {
	stdout, _, code := runCapture(timeout, "openssl", args...)
	return stdout, code == 0
}

// installedCertificates lists TLS domains under NG_TLS_DIR whose cert/key
// pairs verify — agent.js installedCertificates(), including its 5-minute
// fingerprint cache.
func installedCertificates() []string {
	root := cfg.TLSDir
	entries, err := os.ReadDir(root)
	if err != nil {
		return []string{}
	}
	domains := []string{}
	for _, entry := range entries {
		domains = append(domains, entry.Name())
	}
	if len(domains) > 50 {
		domains = domains[:50]
	}
	// Fingerprint over (domain, cert mtime/size, key mtime/size) tuples.
	fpParts := []interface{}{}
	for _, domain := range domains {
		certStat, certErr := os.Stat(filepath.Join(root, domain, "fullchain.pem"))
		keyStat, keyErr := os.Stat(filepath.Join(root, domain, "privkey.pem"))
		if certErr == nil && keyErr == nil {
			fpParts = append(fpParts, []interface{}{domain,
				float64(certStat.ModTime().UnixMilli()), certStat.Size(),
				float64(keyStat.ModTime().UnixMilli()), keyStat.Size()})
		} else {
			fpParts = append(fpParts, []interface{}{domain})
		}
	}
	fingerprint, _ := marshalJSON(fpParts)
	certCacheMu.Lock()
	defer certCacheMu.Unlock()
	if certCacheV.fingerprint == fingerprint && time.Since(certCacheV.checkedAt) < 300*time.Second {
		return append([]string{}, certCacheV.domains...)
	}
	valid := []string{}
	for _, domain := range domains {
		if !domainRe.MatchString(domain) {
			continue
		}
		cert := filepath.Join(root, domain, "fullchain.pem")
		key := filepath.Join(root, domain, "privkey.pem")
		if !fileExists(cert) || !fileExists(key) {
			continue
		}
		if _, ok := opensslOK(3*time.Second, "x509", "-in", cert, "-noout", "-checkend", "86400"); !ok {
			continue
		}
		hostOut, ok := opensslOK(3*time.Second, "x509", "-in", cert, "-noout", "-checkhost", domain)
		if !ok || !strings.Contains(hostOut, "does match") {
			continue
		}
		if _, ok := opensslOK(3*time.Second, "verify", "-verify_hostname", domain, "-untrusted", cert, cert); !ok {
			continue
		}
		certKey, ok1 := opensslOK(3*time.Second, "x509", "-in", cert, "-pubkey", "-noout")
		privateKey, ok2 := opensslOK(3*time.Second, "pkey", "-in", key, "-pubout")
		if !ok1 || !ok2 || strings.TrimSpace(certKey) != strings.TrimSpace(privateKey) {
			continue
		}
		valid = append(valid, domain)
	}
	certCacheV = certCache{fingerprint: fingerprint, checkedAt: time.Now(), domains: valid}
	return append([]string{}, valid...)
}

// ---- engineHealth / heartbeat -------------------------------------------------

type engineReport struct {
	Status           string   `json:"status"`
	Detail           string   `json:"detail"`
	SingBoxInstalled bool     `json:"singBoxInstalled"`
	Certificates     []string `json:"certificates"`
	SingBoxStatus    string   `json:"singBoxStatus"`
	SingBoxVersion   string   `json:"singBoxVersion,omitempty"`
}

// ---- sing-box version probe -------------------------------------------------
// The core version is probed by running `sing-box version` exactly once and
// cached against the binary's mtime+size. It is re-probed only when the file
// changes (an engine install/update replaces the binary) — the first probe
// happens on the first heartbeat after agent start. Steady-state heartbeats
// pay one stat, never a subprocess.

var singBoxVersionMu sync.Mutex
var singBoxVersionCache struct {
	path    string
	mtimeNs int64
	size    int64
	version string
	probed  bool
}

// parseSingBoxVersion extracts the version token from the first line of
// `sing-box version` output ("sing-box version 1.13.21"). Unstamped local
// builds print "unknown"; those report no version.
func parseSingBoxVersion(out string) string {
	line := out
	if i := strings.IndexByte(line, '\n'); i >= 0 {
		line = line[:i]
	}
	fields := strings.Fields(line)
	if len(fields) >= 3 && fields[0] == "sing-box" && fields[1] == "version" && fields[2] != "unknown" {
		return fields[2]
	}
	return ""
}

func singBoxVersion() string {
	info, err := os.Stat(cfg.SingBoxBin)
	if err != nil {
		return ""
	}
	singBoxVersionMu.Lock()
	defer singBoxVersionMu.Unlock()
	c := &singBoxVersionCache
	if c.probed && c.path == cfg.SingBoxBin && c.mtimeNs == info.ModTime().UnixNano() && c.size == info.Size() {
		return c.version
	}
	version := ""
	if out, _, code := runCapture(5*time.Second, cfg.SingBoxBin, "version"); code == 0 {
		version = parseSingBoxVersion(out)
	}
	c.path, c.mtimeNs, c.size, c.version, c.probed = cfg.SingBoxBin, info.ModTime().UnixNano(), info.Size(), version, true
	return version
}

// engineHealth mirrors agent.js engineHealth(). Callers must hold stateMu
// and pass the resources they read under the same lock.
func engineHealthLocked(resources []map[string]interface{}) engineReport {
	installed := fileExists(cfg.SingBoxBin) && fileExists(cfg.SingBoxStatsBin)
	service := serviceHealth("nexusgate-sing-box")
	result := engineReport{
		SingBoxInstalled: installed,
		Certificates:     installedCertificates(),
		SingBoxStatus:    service.Status,
	}
	if installed {
		result.Status = service.Status
		result.Detail = service.Detail
		result.SingBoxVersion = singBoxVersion()
	} else {
		result.Status = "error"
		result.Detail = "sing-box 及统计组件未安装；请在入口机器运行 ng-agent engine install"
	}
	if result.Status == "ready" && len(resources) > 0 {
		missing, err := missingHysteriaListeners(resources)
		if err != nil {
			result.Status = "error"
			result.Detail = safeError(err.Error())
		} else if len(missing) > 0 {
			parts := make([]string, len(missing))
			for i, port := range missing {
				parts[i] = strconv.Itoa(port)
			}
			result.Status = "error"
			result.Detail = "Hysteria 2 UDP 未监听：" + strings.Join(parts, ", ")
		}
	}
	return result
}

func heartbeat() error {
	stateMu.Lock()
	// A readResources error skips the heartbeat entirely (JS throws).
	resources, err := readResources()
	if err != nil {
		stateMu.Unlock()
		return err
	}
	engine := engineHealthLocked(resources)
	stateMu.Unlock()
	// The panel reconciles against this list: any expected resource the node
	// no longer holds (e.g. local state was wiped) is re-dispatched
	// automatically. An empty list is meaningful, so keep it non-nil.
	resourceIDs := []string{}
	for _, res := range resources {
		if s, ok := res["id"].(string); ok && s != "" {
			resourceIDs = append(resourceIDs, s)
		}
	}
	body := map[string]interface{}{
		"version":   version,
		"system":    systemInfo(),
		"engine":    engine,
		"resources": resourceIDs,
	}
	if _, err := requestJSON("/api/agent/heartbeat", body); err != nil {
		return err
	}
	stamp, _ := marshalJSON(map[string]interface{}{"at": isoNow()})
	if err := os.WriteFile("/etc/nexusgate/last-heartbeat.json", []byte(stamp+"\n"), 0o600); err != nil {
		// JS logs this but does not fail the heartbeat.
		logf("Heartbeat reached controller but readiness file failed %s", err.Error())
	}
	return nil
}