package pxe

import (
	"context"
	"errors"
	"fmt"
	"io"
	"log"
	"net"
	"net/url"
	"os"
	"os/exec"
	"path/filepath"
	"runtime"
	"strings"
	"sync"
	"text/template"
	"time"

	"vetting/internal/model"
)

// SupervisorConfig controls how dnsmasq is launched and configured.
type SupervisorConfig struct {
	Enabled         bool
	Interface       string // e.g. "eth0"
	Subnet          string // LAN CIDR, e.g. "192.168.1.0/24"; scopes the proxy-DHCP response
	OrchestratorURL string // baked into iPXE scripts
	RuntimeDir      string // writable dir for dnsmasq.conf and leases
	TFTPRoot        string // holds ipxe.efi, undionly.kpxe
	LiveDir         string // holds vmlinuz, initrd.img (served via HTTP, not dnsmasq; "" disables validation)
	DNSMasqBin      string // path to dnsmasq binary (default: "dnsmasq")
}

// Supervisor owns a dnsmasq subprocess, rewrites its config when the
// host registry changes, and sends SIGHUP to reload. The MAC allowlist
// is the safety barrier: only registered MACs see a DHCP reply.
type Supervisor struct {
	cfg    SupervisorConfig
	mu     sync.Mutex
	cmd    *exec.Cmd
	cancel context.CancelFunc
}

func NewSupervisor(cfg SupervisorConfig) *Supervisor {
	if cfg.DNSMasqBin == "" {
		cfg.DNSMasqBin = "dnsmasq"
	}
	return &Supervisor{cfg: cfg}
}

// Validate checks the preconditions required for dnsmasq to actually
// serve PXE boots: the interface must exist, the iPXE payloads must
// be on disk, the DHCP range + orchestrator URL must parse. Returns
// nil when Enabled=false — tests and dev mode skip all of this.
//
// Without Validate(), dnsmasq starts cleanly on typo'd configs and
// the only symptom is a silent TFTP 404 when a real host PXE-boots.
func (s *Supervisor) Validate() error {
	if !s.cfg.Enabled {
		return nil
	}
	var errs []error

	if s.cfg.Interface == "" {
		errs = append(errs, fmt.Errorf("pxe.interface is required"))
	} else if _, err := net.InterfaceByName(s.cfg.Interface); err != nil {
		errs = append(errs, fmt.Errorf("pxe.interface %q not found on host — check `ip link` or fix pxe.interface in vetting.yaml", s.cfg.Interface))
	}

	if s.cfg.TFTPRoot == "" {
		errs = append(errs, fmt.Errorf("pxe.tftp_root is required"))
	} else {
		for _, name := range []string{"ipxe.efi", "undionly.kpxe"} {
			p := filepath.Join(s.cfg.TFTPRoot, name)
			if _, err := os.Stat(p); err != nil {
				errs = append(errs, fmt.Errorf("missing %s — run deploy/pxe-setup.sh to fetch iPXE binaries", p))
			}
		}
	}

	if s.cfg.LiveDir != "" {
		for _, name := range []string{"vmlinuz", "initrd.img"} {
			p := filepath.Join(s.cfg.LiveDir, name)
			if _, err := os.Stat(p); err != nil {
				errs = append(errs, fmt.Errorf("missing %s — build the live image (`make live-image`) and copy into pxe.live_dir, or use the release tarball", p))
			}
		}
	}

	if s.cfg.Subnet == "" {
		errs = append(errs, fmt.Errorf("pxe.subnet is required (e.g. \"192.168.1.0/24\") — the LAN CIDR dnsmasq proxy-DHCP scopes to"))
	} else if _, _, err := net.ParseCIDR(s.cfg.Subnet); err != nil {
		errs = append(errs, fmt.Errorf("pxe.subnet %q is not a valid CIDR: %v", s.cfg.Subnet, err))
	}

	if s.cfg.OrchestratorURL == "" {
		errs = append(errs, fmt.Errorf("pxe.orchestrator_url is required"))
	} else if u, err := url.Parse(s.cfg.OrchestratorURL); err != nil || (u.Scheme != "http" && u.Scheme != "https") || u.Host == "" {
		errs = append(errs, fmt.Errorf("pxe.orchestrator_url %q must be an http(s) URL with a host", s.cfg.OrchestratorURL))
	}

	return errors.Join(errs...)
}

// Start launches dnsmasq in the background. If cfg.Enabled is false
// Start is a no-op (useful for dev on Windows where dnsmasq isn't
// available).
func (s *Supervisor) Start(ctx context.Context, hosts []model.Host) error {
	if !s.cfg.Enabled {
		log.Printf("pxe: disabled in config — skipping dnsmasq")
		return nil
	}
	if runtime.GOOS == "windows" {
		return fmt.Errorf("dnsmasq supervision is not supported on Windows — run orchestrator on Linux")
	}
	if err := s.Validate(); err != nil {
		return fmt.Errorf("pxe preconditions failed: %w", err)
	}
	if err := os.MkdirAll(s.cfg.RuntimeDir, 0o755); err != nil {
		return fmt.Errorf("mkdir runtime: %w", err)
	}
	if err := s.writeConf(hosts); err != nil {
		return err
	}
	subCtx, cancel := context.WithCancel(ctx)
	s.mu.Lock()
	s.cancel = cancel
	s.mu.Unlock()

	confPath := filepath.Join(s.cfg.RuntimeDir, "dnsmasq.conf")
	cmd := exec.CommandContext(subCtx, s.cfg.DNSMasqBin,
		"--conf-file="+confPath,
		"--no-daemon",
		"--log-queries",
		"--log-dhcp",
	)
	cmd.Stdout = logWriter{prefix: "dnsmasq"}
	cmd.Stderr = logWriter{prefix: "dnsmasq"}
	if err := cmd.Start(); err != nil {
		cancel()
		return fmt.Errorf("start dnsmasq: %w", err)
	}
	s.mu.Lock()
	s.cmd = cmd
	s.mu.Unlock()
	go func() {
		if err := cmd.Wait(); err != nil && subCtx.Err() == nil {
			log.Printf("dnsmasq exited: %v", err)
		}
	}()
	return nil
}

// Reload rewrites the conf with the latest host registry and sends
// SIGHUP. It will restart the subprocess if SIGHUP is unsupported
// (e.g. when running behind an OS that doesn't support it).
func (s *Supervisor) Reload(hosts []model.Host) error {
	if !s.cfg.Enabled {
		return nil
	}
	if err := s.writeConf(hosts); err != nil {
		return err
	}
	s.mu.Lock()
	cmd := s.cmd
	s.mu.Unlock()
	if cmd == nil || cmd.Process == nil {
		return nil
	}
	if err := sighup(cmd.Process); err != nil {
		return fmt.Errorf("sighup dnsmasq: %w", err)
	}
	return nil
}

// Shutdown stops dnsmasq within the timeout.
func (s *Supervisor) Shutdown(timeout time.Duration) error {
	if !s.cfg.Enabled {
		return nil
	}
	s.mu.Lock()
	cancel := s.cancel
	cmd := s.cmd
	s.mu.Unlock()
	if cancel != nil {
		cancel()
	}
	if cmd != nil && cmd.Process != nil {
		done := make(chan struct{})
		go func() {
			_, _ = cmd.Process.Wait()
			close(done)
		}()
		select {
		case <-done:
		case <-time.After(timeout):
			_ = cmd.Process.Kill()
		}
	}
	return nil
}

func (s *Supervisor) writeConf(hosts []model.Host) error {
	tmpl, err := template.New("dnsmasq").Parse(dnsmasqTemplate)
	if err != nil {
		return err
	}
	conf := filepath.Join(s.cfg.RuntimeDir, "dnsmasq.conf")
	tmp := conf + ".new"
	f, err := os.Create(tmp)
	if err != nil {
		return fmt.Errorf("create conf: %w", err)
	}
	data := struct {
		Cfg   SupervisorConfig
		Hosts []model.Host
	}{s.cfg, hosts}
	if err := tmpl.Execute(f, data); err != nil {
		_ = f.Close()
		return fmt.Errorf("render conf: %w", err)
	}
	if err := f.Sync(); err != nil {
		_ = f.Close()
		return err
	}
	if err := f.Close(); err != nil {
		return err
	}
	if err := os.Rename(tmp, conf); err != nil {
		return fmt.Errorf("rename conf: %w", err)
	}
	return nil
}

// Exposed for the UI handlers to show operators what config is live.
func (s *Supervisor) ConfPath() string {
	return filepath.Join(s.cfg.RuntimeDir, "dnsmasq.conf")
}

type logWriter struct{ prefix string }

func (w logWriter) Write(p []byte) (int, error) {
	for _, line := range strings.Split(strings.TrimRight(string(p), "\n"), "\n") {
		if line == "" {
			continue
		}
		log.Printf("[%s] %s", w.prefix, line)
	}
	return len(p), nil
}

// Allow package consumers to swap io.Writer for logs in tests.
var _ io.Writer = logWriter{}

const dnsmasqTemplate = `# Generated by Vetting — do not hand-edit.
interface={{ .Cfg.Interface }}
bind-interfaces
port=0
domain-needed
bogus-priv
no-resolv

# Proxy DHCP: coexist with the LAN's real DHCP server. We never hand
# out an IP — we only answer the PXE options (option 66/67 and the
# PXE BINL on port 4011) when a registered MAC boots from the network.
dhcp-range={{ .Cfg.Subnet }},proxy

# MAC allowlist: dnsmasq only answers DHCP for MACs with a dhcp-host= below.
dhcp-ignore=tag:!known
{{- range .Hosts }}
dhcp-host={{ .MAC }},set:known
{{- end }}

# Keep runtime state inside RuntimeDir so the systemd sandbox
# (ReadWritePaths=/var/lib/vetting ...) doesn't block writes to the
# distro defaults (/var/lib/misc/dnsmasq.leases, /run/dnsmasq.pid).
dhcp-leasefile={{ .Cfg.RuntimeDir }}/dhcp.leases
pid-file={{ .Cfg.RuntimeDir }}/dnsmasq.pid

# TFTP for first-boot BIOS/UEFI clients; already-iPXE clients skip it.
enable-tftp
tftp-root={{ .Cfg.TFTPRoot }}

# Already-iPXE clients: chainload the per-MAC HTTP script directly.
dhcp-match=set:ipxe,175
dhcp-boot=tag:ipxe,{{ .Cfg.OrchestratorURL }}/ipxe/${mac}

# First-boot PXE ROM -> iPXE. In proxy-DHCP mode, chainloading uses
# pxe-service= (not dhcp-boot=) because the real LAN DHCP has already
# assigned the IP; we only supplement the boot menu.
pxe-service=tag:!ipxe,x86PC,"iPXE (BIOS)",undionly.kpxe
pxe-service=tag:!ipxe,X86-64_EFI,"iPXE (UEFI)",ipxe.efi

log-facility=-
`