Find a refused packet's tunnel from its header, ask the lighthouse when probing, and address review

overlay/batch/tx_batch.go

@@ -4,16 +4,20 @@ import "net/netip"
 
 const SendBatchCap = 128
 
+// FailureHeadLen is how much of a refused packet a Failure keeps, nebula's header, which says whose packet it was
+const FailureHeadLen = 16
+
 // batchWriter is the minimal subset of udp.Conn needed by SendBatch to flush.
 type batchWriter interface {
 	WriteBatch(bufs [][]byte, addrs []netip.AddrPort, failed func(i int, err error)) (int, error)
 }
 
-// Failure is a packet the underlay would not send, with the tag it was committed under
+// Failure is a packet the underlay would not send
 type Failure struct {
-	Tag uint32
-	Dst netip.AddrPort
-	Err error
+	// Head is the start of the packet, zero padded if the packet was shorter
+	Head [FailureHeadLen]byte
+	Dst  netip.AddrPort
+	Err  error
 }
 
 // SendBatch accumulates encrypted UDP packets and flushes them via WriteBatch.
@@ -23,7 +27,6 @@ type SendBatch struct {
 	out   batchWriter
 	bufs  [][]byte
 	dsts  []netip.AddrPort
-	tags  []uint32
 	arena *Arena
 
 	failures []Failure
@@ -37,7 +40,6 @@ func NewSendBatch(out batchWriter, batchCap, arenaSize int) *SendBatch {
 		out:   out,
 		bufs:  make([][]byte, 0, batchCap),
 		dsts:  make([]netip.AddrPort, 0, batchCap),
-		tags:  make([]uint32, 0, batchCap),
 		arena: NewArena(arenaSize),
 	}
 	b.recordFailure = b.failed
@@ -45,7 +47,9 @@ func NewSendBatch(out batchWriter, batchCap, arenaSize int) *SendBatch {
 }
 
 func (b *SendBatch) failed(i int, err error) {
-	b.failures = append(b.failures, Failure{Tag: b.tags[i], Dst: b.dsts[i], Err: err})
+	f := Failure{Dst: b.dsts[i], Err: err}
+	copy(f.Head[:], b.bufs[i])
+	b.failures = append(b.failures, f)
 }
 
 func (b *SendBatch) Reserve(sz int) []byte {
@@ -57,12 +61,9 @@ func (b *SendBatch) Reserve(sz int) []byte {
 // bounding how long the first packet of a large read batch waits.
 func (b *SendBatch) Len() int { return len(b.bufs) }
 
-// Commit queues pkt for dst. tag comes back in Failures if the underlay refuses the packet, so the caller can tell
-// whose packet it was.
-func (b *SendBatch) Commit(pkt []byte, dst netip.AddrPort, tag uint32) {
+func (b *SendBatch) Commit(pkt []byte, dst netip.AddrPort) {
 	b.bufs = append(b.bufs, pkt)
 	b.dsts = append(b.dsts, dst)
-	b.tags = append(b.tags, tag)
 }
 
 // Flush writes every queued packet and reports how many actually went out. A short count means some destinations
@@ -79,7 +80,6 @@ func (b *SendBatch) Flush() (int, error) {
 	clear(b.bufs)
 	b.bufs = b.bufs[:0]
 	b.dsts = b.dsts[:0]
-	b.tags = b.tags[:0]
 	b.arena.Reset()
 	return written, err
 }

overlay/batch/tx_batch_test.go

@@ -36,7 +36,7 @@ func TestSendBatchReserveCommitFlush(t *testing.T) {
 			t.Fatalf("slot %d: cap=%d want 32", i, cap(slot))
 		}
 		pkt := append(slot[:0], byte(i), byte(i+1), byte(i+2))
-		b.Commit(pkt, ap, 0)
+		b.Commit(pkt, ap)
 	}
 	if _, err := b.Flush(); err != nil {
 		t.Fatalf("Flush: %v", err)
@@ -77,7 +77,7 @@ func TestSendBatchSlotsDoNotOverlap(t *testing.T) {
 	for i := range 3 {
 		s := b.Reserve(8)
 		pkt := append(s[:0], byte(0xA0+i), byte(0xB0+i))
-		b.Commit(pkt, ap, 0)
+		b.Commit(pkt, ap)
 	}
 	if _, err := b.Flush(); err != nil {
 		t.Fatalf("Flush: %v", err)
@@ -98,11 +98,11 @@ func TestSendBatchGrowPreservesCommitted(t *testing.T) {
 
 	s1 := b.Reserve(4)
 	pkt1 := append(s1[:0], 0x11, 0x22, 0x33, 0x44)
-	b.Commit(pkt1, ap, 0)
+	b.Commit(pkt1, ap)
 
 	s2 := b.Reserve(8) // exceeds remaining cap, triggers grow
 	pkt2 := append(s2[:0], 0xA, 0xB, 0xC, 0xD, 0xE)
-	b.Commit(pkt2, ap, 0)
+	b.Commit(pkt2, ap)
 
 	// pkt1 must still be intact even though backing reallocated.
 	if pkt1[0] != 0x11 || pkt1[3] != 0x44 {
@@ -140,11 +140,24 @@ func (w *failingBatchWriter) WriteBatch(bufs [][]byte, _ []netip.AddrPort, faile
 	return written, nil
 }
 
-func TestSendBatchReportsFailedTags(t *testing.T) {
+func TestSendBatchReportsFailures(t *testing.T) {
 	errNoRoute := errors.New("no route")
 	fw := &failingBatchWriter{refuse: map[byte]error{2: errNoRoute, 3: errNoRoute}}
+	b := NewSendBatch(fw, 4, 128)
 
-	b := NewSendBatch(fw, 4, 64)
+	// pkt is n bytes starting with the refuse key, each later byte its offset plus 0x10 so the head is recognizable
+	pkt := func(key byte, n int) []byte {
+		p := b.Reserve(n)[:n]
+		p[0] = key
+		for i := 1; i < n; i++ {
+			p[i] = byte(0x10 + i)
+		}
+		return p
+	}
+	head := func(p []byte) (h [FailureHeadLen]byte) {
+		copy(h[:], p)
+		return h
+	}
 
 	dsts := []netip.AddrPort{
 		netip.MustParseAddrPort("10.0.0.1:4242"),
@@ -152,10 +165,12 @@ func TestSendBatchReportsFailedTags(t *testing.T) {
 		netip.MustParseAddrPort("[fd00::3]:4242"),
 		netip.MustParseAddrPort("10.0.0.4:4242"),
 	}
-	for i, dst := range dsts {
-		pkt := append(b.Reserve(8)[:0], byte(i+1))
-		b.Commit(pkt, dst, uint32(100+i))
-	}
+	long := pkt(2, FailureHeadLen+8) // only its head is kept
+	short := pkt(3, 4)               // zero padded
+	b.Commit(pkt(1, 20), dsts[0])
+	b.Commit(long, dsts[1])
+	b.Commit(short, dsts[2])
+	b.Commit(pkt(1, 20), dsts[3])
 
 	written, err := b.Flush()
 	if err != nil {
@@ -164,22 +179,23 @@ func TestSendBatchReportsFailedTags(t *testing.T) {
 	if written != 2 {
 		t.Errorf("written = %d, want 2", written)
 	}
-	want := []Failure{{101, dsts[1], errNoRoute}, {102, dsts[2], errNoRoute}}
+	want := []Failure{{head(long), dsts[1], errNoRoute}, {head(short), dsts[2], errNoRoute}}
 	if got := b.Failures(); !slices.Equal(got, want) {
 		t.Errorf("failures = %v, want %v", got, want)
 	}
 
-	// Tags do not leak into the next flush
-	b.Commit(append(b.Reserve(8)[:0], 2), dsts[0], 7)
+	// Failures do not leak into the next flush
+	again := pkt(2, 20)
+	b.Commit(again, dsts[0])
 	if _, err := b.Flush(); err != nil {
 		t.Fatalf("Flush: %v", err)
 	}
-	if want := []Failure{{7, dsts[0], errNoRoute}}; !slices.Equal(b.Failures(), want) {
+	if want := []Failure{{head(again), dsts[0], errNoRoute}}; !slices.Equal(b.Failures(), want) {
 		t.Errorf("second flush failures = %v, want %v", b.Failures(), want)
 	}
 
 	// A clean flush forgets the last one's failures
-	b.Commit(append(b.Reserve(8)[:0], 1), dsts[0], 8)
+	b.Commit(pkt(1, 20), dsts[0])
 	if _, err := b.Flush(); err != nil {
 		t.Fatalf("Flush: %v", err)
 	}