flock/pkg/agent/bird.go

package agent

import (
	"errors"
	"fmt"
	"log/slog"
	"net"
	"os"
	"os/exec"
	"path/filepath"
	"sync"
	"time"

	flockv1alpha1 "code.fritzlab.net/fritzlab/flock/pkg/api/v1alpha1"
	"code.fritzlab.net/fritzlab/flock/pkg/routing/bird"
)

// BirdManager renders bird.conf and triggers birdc reload. Writes are
// debounced so a burst of NodeConfig / anycast changes coalesces.
type BirdManager struct {
	NodeName    string
	ConfigPath  string // /etc/flock/bird/bird.conf
	BirdcSocket string // /run/flock/bird6.ctl (BIRD2 single-socket default)
	BirdctlPath string // "birdc" — overridable for tests
	Logger      *slog.Logger

	mu          sync.Mutex
	last        string // last rendered output (de-dup)
	cooldown    *time.Timer
	cooldownEnd time.Time // window during which further reloads are coalesced
	pending     bool      // a render landed during cooldown; reload at window end
}

// reloadCooldown is the minimum spacing between two birdc reloads. The
// first change fires immediately (no leading-edge delay); follow-up
// changes within this window are coalesced into a single tail reload.
const reloadCooldown = 500 * time.Millisecond

// Render writes the config from a NodeConfig + anycast set. Idempotent —
// if the rendered content matches what we last wrote, no birdc reload.
func (b *BirdManager) Render(nc *flockv1alpha1.NodeConfig, anycast6, anycast4 []string, routerID string) error {
	if nc == nil {
		return fmt.Errorf("no NodeConfig")
	}
	in := bird.NodeBGP{
		NodeName: b.NodeName,
		RouterID: routerID,
		LocalASN: nc.Spec.BGP.ASN,
		CIDR6:    nc.Spec.CIDR6,
		CIDR4:    nc.Spec.CIDR4,
		Anycast6: anycast6,
		Anycast4: anycast4,
	}
	// Pick a local source address per family that's on the same subnet as
	// the BGP peer. crt001 rejects IPv6 advertisements whose next-hop is
	// link-local-only; an explicit `source address` makes BIRD use a
	// global next-hop self, which Cisco accepts.
	for _, p := range nc.Spec.BGP.Peers {
		fam := bird.FamilyOf(p.Address)
		if fam == "" {
			continue
		}
		in.Peers = append(in.Peers, bird.Peer{Family: fam, Address: p.Address, ASN: p.ASN})
		if local := localAddrSameSubnet(p.Address); local != "" {
			if fam == "v6" && in.LocalV6 == "" {
				in.LocalV6 = local
			}
			if fam == "v4" && in.LocalV4 == "" {
				in.LocalV4 = local
			}
		}
	}

	cfg, err := bird.Render(in)
	if err != nil {
		return err
	}
	b.mu.Lock()
	defer b.mu.Unlock()
	if cfg == b.last {
		return nil
	}
	if err := os.MkdirAll(filepath.Dir(b.ConfigPath), 0o755); err != nil {
		return fmt.Errorf("mkdir bird config dir: %w", err)
	}
	tmp := b.ConfigPath + ".tmp"
	if err := os.WriteFile(tmp, []byte(cfg), 0o644); err != nil {
		return fmt.Errorf("write bird.conf: %w", err)
	}
	if err := os.Rename(tmp, b.ConfigPath); err != nil {
		return fmt.Errorf("rename bird.conf: %w", err)
	}
	b.last = cfg
	b.scheduleReload()
	return nil
}

// scheduleReload uses leading-edge + cooldown semantics: the first call
// reloads immediately; subsequent calls within reloadCooldown coalesce
// into a single deferred reload at the cooldown's end. Caller holds b.mu.
func (b *BirdManager) scheduleReload() {
	now := time.Now()
	if now.After(b.cooldownEnd) {
		// Outside any active cooldown — fire now (leading edge).
		b.cooldownEnd = now.Add(reloadCooldown)
		b.pending = false
		go b.reload()
		return
	}
	// Inside cooldown — coalesce. If no tail timer is set, schedule one
	// at the cooldown end; if already set, just leave it.
	if b.pending {
		return
	}
	b.pending = true
	delay := b.cooldownEnd.Sub(now)
	b.cooldown = time.AfterFunc(delay, func() {
		b.mu.Lock()
		b.pending = false
		b.cooldownEnd = time.Now().Add(reloadCooldown)
		b.mu.Unlock()
		b.reload()
	})
}

func (b *BirdManager) reload() {
	birdctl := b.BirdctlPath
	if birdctl == "" {
		birdctl = "birdc"
	}
	socket := b.BirdcSocket
	if socket == "" {
		socket = "/run/flock/bird.ctl"
	}
	cmd := exec.Command(birdctl, "-s", socket, "configure")
	out, err := cmd.CombinedOutput()
	if err != nil {
		// First-run case: bird may not be ready yet — retry on next change.
		if errors.Is(err, exec.ErrNotFound) || os.IsNotExist(err) {
			b.Logger.Warn("birdc not available", "err", err)
			return
		}
		b.Logger.Warn("birdc reload failed", "err", err, "out", string(out))
		return
	}
	b.Logger.Info("birdc configure ok", "out", string(out))
}

// SummaryRoutes installs blackhole kernel routes for each NodeConfig CIDR.
// BIRD's protocol kernel imports them so they get advertised. Idempotent.
func (b *BirdManager) SummaryRoutes(nc *flockv1alpha1.NodeConfig) error {
	if nc == nil {
		return nil
	}
	for _, c := range nc.Spec.CIDR6 {
		if err := installBlackhole(c); err != nil {
			b.Logger.Warn("blackhole route v6", "cidr", c, "err", err)
		}
	}
	for _, c := range nc.Spec.CIDR4 {
		if err := installBlackhole(c); err != nil {
			b.Logger.Warn("blackhole route v4", "cidr", c, "err", err)
		}
	}
	return nil
}

// localAddrSameSubnet finds an IP on a local interface that's in the same
// /64 (v6) or /24 (v4) as `peer`. Returns "" if none. Used to derive the
// `source address` for a BGP session.
func localAddrSameSubnet(peer string) string {
	pip := net.ParseIP(peer)
	if pip == nil {
		return ""
	}
	addrs, err := net.InterfaceAddrs()
	if err != nil {
		return ""
	}
	v4 := pip.To4() != nil
	for _, a := range addrs {
		ipn, ok := a.(*net.IPNet)
		if !ok {
			continue
		}
		ip := ipn.IP
		if ip.IsLoopback() || ip.IsLinkLocalUnicast() {
			continue
		}
		if (ip.To4() != nil) != v4 {
			continue
		}
		// Use the peer's mask (assume same subnet) for membership test.
		var mask net.IPMask
		if v4 {
			mask = net.CIDRMask(24, 32)
		} else {
			mask = net.CIDRMask(64, 128)
		}
		peerSubnet := &net.IPNet{IP: pip, Mask: mask}
		if peerSubnet.Contains(ip) {
			if v4 {
				return ip.To4().String()
			}
			return ip.To16().String()
		}
	}
	return ""
}

func installBlackhole(cidr string) error {
	// Use `ip` rather than netlink so this file stays portable for non-Linux
	// builds (the agent on macOS just no-ops). The agent only runs in
	// Kubernetes pods on Linux nodes, so the exec is fine.
	_, _, err := net.ParseCIDR(cidr)
	if err != nil {
		return err
	}
	cmd := exec.Command("ip", "route", "replace", "blackhole", cidr)
	out, err := cmd.CombinedOutput()
	if err != nil {
		return fmt.Errorf("ip route replace blackhole %s: %w (%s)", cidr, err, string(out))
	}
	return nil
}