Provisioning/internal/pxe/supervisor.go

package pxe

import (
	"context"
	"fmt"
	"log"
	"os"
	"os/exec"
	"path/filepath"
	"strings"
	"text/template"

	"provisioning/internal/model"
)

type SupervisorConfig struct {
	Enabled    bool
	Interface  string
	Subnet     string
	RuntimeDir string
	TFTPRoot   string
	DnsmasqBin string
	PublicURL  string
}

type Supervisor struct {
	cfg    SupervisorConfig
	cmd    *exec.Cmd
	cancel context.CancelFunc
}

func NewSupervisor(cfg SupervisorConfig) *Supervisor {
	return &Supervisor{cfg: cfg}
}

func (s *Supervisor) Start(ctx context.Context, hosts []model.Host) error {
	if !s.cfg.Enabled {
		log.Printf("pxe: dnsmasq disabled")
		return nil
	}
	if err := os.MkdirAll(s.cfg.RuntimeDir, 0o755); err != nil {
		return fmt.Errorf("pxe: create runtime dir: %w", err)
	}
	if err := s.writeConfig(hosts); err != nil {
		return err
	}
	return s.startProcess(ctx)
}

func (s *Supervisor) Reload(hosts []model.Host) error {
	if !s.cfg.Enabled {
		return nil
	}
	if err := s.writeConfig(hosts); err != nil {
		return err
	}
	if s.cmd != nil && s.cmd.Process != nil {
		return signalReload(s.cmd.Process)
	}
	return nil
}

func (s *Supervisor) Shutdown() error {
	if s.cancel != nil {
		s.cancel()
	}
	if s.cmd != nil && s.cmd.Process != nil {
		return s.cmd.Process.Kill()
	}
	return nil
}

func (s *Supervisor) writeConfig(hosts []model.Host) error {
	confPath := filepath.Join(s.cfg.RuntimeDir, "dnsmasq.conf")
	hostsPath := filepath.Join(s.cfg.RuntimeDir, "dhcp-hostsfile")

	var macs []string
	for _, h := range hosts {
		macs = append(macs, h.MAC)
	}
	if err := os.WriteFile(hostsPath, []byte(strings.Join(macs, "\n")+"\n"), 0o644); err != nil {
		return fmt.Errorf("pxe: write dhcp-hostsfile: %w", err)
	}

	conf := dnsmasqConf{
		Interface:  s.cfg.Interface,
		TFTPRoot:   s.cfg.TFTPRoot,
		HostsFile:  hostsPath,
		PublicURL:  s.cfg.PublicURL,
		RuntimeDir: s.cfg.RuntimeDir,
	}
	f, err := os.Create(confPath)
	if err != nil {
		return fmt.Errorf("pxe: create dnsmasq.conf: %w", err)
	}
	defer f.Close()
	if err := dnsmasqTmpl.Execute(f, conf); err != nil {
		return fmt.Errorf("pxe: render dnsmasq.conf: %w", err)
	}
	return nil
}

func (s *Supervisor) startProcess(ctx context.Context) error {
	confPath := filepath.Join(s.cfg.RuntimeDir, "dnsmasq.conf")
	procCtx, cancel := context.WithCancel(ctx)
	s.cancel = cancel
	s.cmd = exec.CommandContext(procCtx, s.cfg.DnsmasqBin, "--keep-in-foreground", "--conf-file="+confPath)
	s.cmd.Stdout = os.Stdout
	s.cmd.Stderr = os.Stderr
	if err := s.cmd.Start(); err != nil {
		cancel()
		return fmt.Errorf("pxe: start dnsmasq: %w", err)
	}
	go func() {
		if err := s.cmd.Wait(); err != nil {
			select {
			case <-procCtx.Done():
			default:
				log.Printf("pxe: dnsmasq exited: %v", err)
			}
		}
	}()
	log.Printf("pxe: dnsmasq started (pid %d)", s.cmd.Process.Pid)
	return nil
}

type dnsmasqConf struct {
	Interface  string
	TFTPRoot   string
	HostsFile  string
	PublicURL  string
	RuntimeDir string
}

var dnsmasqTmpl = template.Must(template.New("dnsmasq.conf").Parse(`
port=0
interface={{.Interface}}
bind-interfaces

dhcp-range=tag:known,192.168.1.0,proxy
dhcp-hostsfile={{.HostsFile}}
dhcp-ignore=tag:!known

enable-tftp
tftp-root={{.TFTPRoot}}

# Legacy BIOS
dhcp-match=set:bios,option:client-arch,0
dhcp-boot=tag:bios,undionly.kpxe

# UEFI
dhcp-match=set:efi64,option:client-arch,7
dhcp-boot=tag:efi64,ipxe.efi

# iPXE user-class: chain to HTTP script
dhcp-match=set:ipxe,option:user-class,iPXE
dhcp-boot=tag:ipxe,{{.PublicURL}}/ipxe/${mac:hexhyp}

log-dhcp
log-facility={{.RuntimeDir}}/dnsmasq.log
`))