hamlab/internal/services/services.go

// Package services wraps systemctl and journalctl for the Services tab: list units,
// show one, read its unit file, read and tail its journal, and start/stop/etc.
//
// Reads run as the daemon's user (systemd exposes state to everyone; the journal needs
// the systemd-journal or adm group, or sudo). Actions go through passwordless sudo when
// it's available, otherwise plain systemctl so polkit gets its say.
package services

import (
	"bufio"
	"bytes"
	"context"
	"encoding/json"
	"errors"
	"fmt"
	"os"
	"os/exec"
	"regexp"
	"sort"
	"strconv"
	"strings"
	"sync"
	"syscall"
	"time"

	"hamlab/internal/priv"
)

type Unit struct {
	Name        string `json:"name"`
	Description string `json:"description"`
	Load        string `json:"load"`
	Active      string `json:"active"`
	Sub         string `json:"sub"`
	Enabled     string `json:"enabled"` // enabled | disabled | static | masked | … | "" (unknown)
	Failed      bool   `json:"failed"`
}

type Detail struct {
	Unit
	Props map[string]string `json:"props"`
}

type Entry struct {
	T        int64  `json:"t"` // unix ms
	Priority int    `json:"p"` // syslog priority 0-7
	Ident    string `json:"ident"`
	PID      string `json:"pid,omitempty"`
	Msg      string `json:"msg"`
}

var unitRe = regexp.MustCompile(`^[A-Za-z0-9][A-Za-z0-9@._:\\-]{0,255}\.service$`)

func Valid(unit string) bool { return unitRe.MatchString(unit) }

var Actions = map[string]bool{"start": true, "stop": true, "restart": true, "reload": true, "enable": true, "disable": true, "reset-failed": true}

// Properties shown on the detail page, in display order.
var showProps = []string{"Id", "Description", "LoadState", "ActiveState", "SubState", "UnitFileState", "FragmentPath",
	"MainPID", "ExecMainStartTimestamp", "ActiveEnterTimestamp", "InactiveEnterTimestamp", "NRestarts", "Result",
	"MemoryCurrent", "MemoryPeak", "CPUUsageNSec", "TasksCurrent", "ExecStart", "User", "Restart"}

type Manager struct {
	sudo        bool // actions via sudo -n
	journalSudo bool // journal needs sudo (user isn't in systemd-journal/adm)
	mu          sync.Mutex
	cache       []Unit
	cacheAt     time.Time
}

// ErrNoSystemd is what the Services tab says on OpenRC, runit and friends.
var ErrNoSystemd = errors.New("this host doesn't run systemd, which the Services tab needs")

// HasSystemd reports whether systemd is the running init (the test sd_booted() uses).
func HasSystemd() bool {
	_, err := os.Stat("/run/systemd/system")
	return err == nil
}

func New() *Manager {
	m := &Manager{}
	ctx, cancel := context.WithTimeout(context.Background(), 3*time.Second)
	defer cancel()
	m.sudo = priv.Available()
	// Prefer reading the journal as ourselves: a plain child can be killed cleanly when a
	// tail ends, whereas a root journalctl under sudo can only be reached with sudo kill.
	var errb bytes.Buffer
	probe := exec.CommandContext(ctx, "journalctl", "--system", "-n", "1", "-q", "-o", "cat")
	probe.Stderr = &errb
	if err := probe.Run(); err != nil || strings.Contains(strings.ToLower(errb.String()), "permission") {
		m.journalSudo = m.sudo
	}
	return m
}

// Sudo reports whether actions will run through passwordless sudo.
func (m *Manager) Sudo() bool { return m.sudo }

// JournalSudo reports whether reading the system journal needs sudo for this user.
func (m *Manager) JournalSudo() bool { return m.journalSudo }

func run(ctx context.Context, args ...string) (string, error) {
	ctx, cancel := context.WithTimeout(ctx, 20*time.Second)
	defer cancel()
	cmd := exec.CommandContext(ctx, args[0], args[1:]...)
	var out, errb bytes.Buffer
	cmd.Stdout, cmd.Stderr = &out, &errb
	if err := cmd.Run(); err != nil {
		msg := strings.TrimSpace(errb.String())
		if msg == "" {
			msg = err.Error()
		}
		return out.String(), errors.New(msg)
	}
	return out.String(), nil
}

// List merges `list-units --all` (runtime state) with `list-unit-files` (enablement).
// Cached for a few seconds because the UI polls it.
func (m *Manager) List(ctx context.Context) ([]Unit, error) {
	if !HasSystemd() {
		return nil, ErrNoSystemd
	}
	m.mu.Lock()
	if time.Since(m.cacheAt) < 200*time.Millisecond && m.cache != nil { // coalesce simultaneous clients only
		c := m.cache
		m.mu.Unlock()
		return c, nil
	}
	m.mu.Unlock()

	var out, files string
	var err error
	done := make(chan struct{})
	go func() {
		files, _ = run(ctx, "systemctl", "list-unit-files", "--type=service", "--output=json", "--no-pager")
		close(done)
	}()
	out, err = run(ctx, "systemctl", "list-units", "--type=service", "--all", "--output=json", "--no-pager", "--plain")
	<-done
	if err != nil {
		return nil, err
	}
	var units []listedUnit
	if err := json.Unmarshal([]byte(out), &units); err != nil {
		// systemd before 246 (RHEL 8, Ubuntu 20.04) ignores --output=json here and prints
		// the table; the columns have been the same since the beginning.
		if units = parseUnitTable(out); len(units) == 0 {
			return nil, fmt.Errorf("parse list-units: %w", err)
		}
	}
	var uf []listedFile
	if json.Unmarshal([]byte(files), &uf) != nil {
		uf = parseFileTable(files)
	}
	enabled := map[string]string{}
	for _, f := range uf {
		enabled[f.UnitFile] = f.State
	}

	seen := map[string]bool{}
	list := make([]Unit, 0, len(units)+len(uf))
	for _, u := range units {
		if strings.Contains(u.Unit, "@.") { // uninstantiated template
			continue
		}
		seen[u.Unit] = true
		list = append(list, Unit{Name: u.Unit, Description: u.Description, Load: u.Load, Active: u.Active, Sub: u.Sub,
			Enabled: enabled[u.Unit], Failed: u.Active == "failed"})
	}
	// Unit files that exist but were never loaded this boot (disabled and never started).
	for _, f := range uf {
		if seen[f.UnitFile] || strings.Contains(f.UnitFile, "@.") || f.State == "masked" || f.State == "alias" {
			continue
		}
		list = append(list, Unit{Name: f.UnitFile, Load: "not-loaded", Active: "inactive", Sub: "dead", Enabled: f.State})
	}
	sort.Slice(list, func(i, j int) bool { return list[i].Name < list[j].Name })

	m.mu.Lock()
	m.cache, m.cacheAt = list, time.Now()
	m.mu.Unlock()
	return list, nil
}

type listedUnit struct{ Unit, Load, Active, Sub, Description string }

type listedFile struct {
	UnitFile string `json:"unit_file"`
	State    string `json:"state"`
}

// parseUnitTable reads the plain `systemctl list-units` table: UNIT LOAD ACTIVE SUB DESCRIPTION…
func parseUnitTable(text string) []listedUnit {
	var units []listedUnit
	for _, line := range strings.Split(text, "\n") {
		f := strings.Fields(strings.TrimLeft(line, "●*× "))
		if len(f) < 4 || !strings.HasSuffix(f[0], ".service") {
			continue
		}
		units = append(units, listedUnit{Unit: f[0], Load: f[1], Active: f[2], Sub: f[3], Description: strings.Join(f[4:], " ")})
	}
	return units
}

// parseFileTable reads the plain `systemctl list-unit-files` table: UNIT FILE, STATE[, PRESET]
func parseFileTable(text string) []listedFile {
	var uf []listedFile
	for _, line := range strings.Split(text, "\n") {
		if f := strings.Fields(line); len(f) >= 2 && strings.HasSuffix(f[0], ".service") {
			uf = append(uf, listedFile{UnitFile: f[0], State: f[1]})
		}
	}
	return uf
}

func (m *Manager) invalidate() {
	m.mu.Lock()
	m.cacheAt = time.Time{}
	m.mu.Unlock()
}

func (m *Manager) Show(ctx context.Context, unit string) (Detail, error) {
	if !Valid(unit) {
		return Detail{}, errors.New("invalid unit name")
	}
	out, err := run(ctx, "systemctl", "show", unit, "-p", strings.Join(showProps, ","), "--no-pager")
	if err != nil {
		return Detail{}, err
	}
	props := map[string]string{}
	for _, line := range strings.Split(out, "\n") {
		if k, v, ok := strings.Cut(line, "="); ok && v != "" && v != "[not set]" && v != "infinity" {
			props[k] = v
		}
	}
	if props["LoadState"] == "not-found" {
		return Detail{}, errors.New("no such unit")
	}
	d := Detail{Props: props}
	d.Name, d.Description = unit, props["Description"]
	d.Load, d.Active, d.Sub, d.Enabled = props["LoadState"], props["ActiveState"], props["SubState"], props["UnitFileState"]
	d.Failed = d.Active == "failed"
	return d, nil
}

func (m *Manager) Cat(ctx context.Context, unit string) (string, error) {
	if !Valid(unit) {
		return "", errors.New("invalid unit name")
	}
	return run(ctx, "systemctl", "cat", unit, "--no-pager")
}

func parseEntry(line []byte) (Entry, bool) {
	var raw map[string]any
	if json.Unmarshal(line, &raw) != nil {
		return Entry{}, false
	}
	e := Entry{Priority: 6}
	switch v := raw["MESSAGE"].(type) {
	case string:
		e.Msg = v
	case []any: // non-UTF-8 payload comes as a byte array
		b := make([]byte, 0, len(v))
		for _, x := range v {
			if f, ok := x.(float64); ok {
				b = append(b, byte(f))
			}
		}
		e.Msg = string(b)
	default:
		e.Msg = ""
	}
	if p, ok := raw["PRIORITY"].(string); ok {
		e.Priority, _ = strconv.Atoi(p)
	}
	if ts, ok := raw["__REALTIME_TIMESTAMP"].(string); ok {
		us, _ := strconv.ParseInt(ts, 10, 64)
		e.T = us / 1000
	}
	if id, ok := raw["SYSLOG_IDENTIFIER"].(string); ok {
		e.Ident = id
	} else if c, ok := raw["_COMM"].(string); ok {
		e.Ident = c
	}
	if pid, ok := raw["_PID"].(string); ok {
		e.PID = pid
	}
	return e, true
}

func (m *Manager) journalArgs(unit string, n int) []string {
	args := []string{"journalctl", "-u", unit, "-n", strconv.Itoa(n), "-o", "json", "--no-pager", "--all"}
	if m.journalSudo {
		args = priv.Wrap(args...)
	}
	return args
}

// Journal returns the last n entries.
func (m *Manager) Journal(ctx context.Context, unit string, n int) ([]Entry, error) {
	if !Valid(unit) {
		return nil, errors.New("invalid unit name")
	}
	out, err := run(ctx, m.journalArgs(unit, n)...)
	if err != nil {
		return nil, err
	}
	entries := []Entry{}
	for _, line := range bytes.Split([]byte(out), []byte("\n")) {
		if e, ok := parseEntry(line); ok {
			entries = append(entries, e)
		}
	}
	return entries, nil
}

// Tail streams entries (last n, then follow) to fn until ctx is cancelled.
func (m *Manager) Tail(ctx context.Context, unit string, n int, fn func(Entry)) error {
	if !Valid(unit) {
		return errors.New("invalid unit name")
	}
	args := append(m.journalArgs(unit, n), "-f")
	cmd := exec.CommandContext(ctx, args[0], args[1:]...)
	// Own process group so the kill reaches journalctl itself, not just a sudo wrapper.
	cmd.SysProcAttr = &syscall.SysProcAttr{Setpgid: true}
	cmd.Cancel = func() error {
		pg := "-" + strconv.Itoa(cmd.Process.Pid)
		if m.journalSudo { // children are root's; only sudo can signal them
			k := priv.Wrap("kill", "-9", "--", pg)
			return exec.Command(k[0], k[1:]...).Run()
		}
		return syscall.Kill(-cmd.Process.Pid, syscall.SIGKILL)
	}
	stdout, err := cmd.StdoutPipe()
	if err != nil {
		return err
	}
	if err := cmd.Start(); err != nil {
		return err
	}
	sc := bufio.NewScanner(stdout)
	sc.Buffer(make([]byte, 0, 64<<10), 4<<20)
	for sc.Scan() {
		if e, ok := parseEntry(sc.Bytes()); ok {
			fn(e)
		}
	}
	_ = cmd.Wait()
	return nil
}

// Action runs start/stop/restart/reload/enable/disable/reset-failed.
func (m *Manager) Action(ctx context.Context, action, unit string) error {
	if !Valid(unit) {
		return errors.New("invalid unit name")
	}
	if !Actions[action] {
		return errors.New("unknown action")
	}
	args := []string{"systemctl", action, unit, "--no-pager"}
	if m.sudo {
		args = priv.Wrap(args...)
	}
	_, err := run(ctx, args...)
	m.invalidate()
	return err
}