fips-core 0.3.72

Reusable FIPS mesh, endpoint, transport, and protocol library
Documentation
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
//! UDP Transport Implementation
//!
//! Provides UDP-based transport for FIPS peer communication.

#[cfg(target_os = "linux")]
use super::received_timestamp_ms;
use super::{
    DiscoveredPeer, PacketTx, ReceivedPacket, Transport, TransportAddr, TransportError,
    TransportId, TransportState, TransportType,
};
#[cfg(any(target_os = "linux", target_os = "macos"))]
pub(crate) mod connected_peer;
#[cfg(target_os = "macos")]
pub(crate) mod darwin_sockopts;
#[cfg(any(target_os = "linux", target_os = "macos"))]
pub(crate) mod peer_drain;
pub(crate) mod socket;
mod stats;
use super::resolve_socket_addr;
use crate::config::UdpConfig;
use crate::discovery::is_punch_packet;
use socket::{AsyncUdpSocket, UdpRawSocket};
use stats::UdpStats;
use std::collections::HashMap;
use std::net::SocketAddr;
use std::sync::{Arc, Mutex as StdMutex};
use std::time::{Duration, Instant};
use tokio::task::JoinHandle;
use tracing::{debug, info, trace, warn};

/// DNS cache TTL for hostname resolution (60 seconds).
const DNS_CACHE_TTL: Duration = Duration::from_secs(60);

/// Datagrams drained per UDP receive syscall / connected-peer poll cycle.
///
/// Keep one receive-batch width across the wildcard socket, connected peer
/// drain threads, and Linux recvmmsg wrapper. WireGuard-go and Tailscale use
/// 128 as their ideal userspace packet batch, and the current measured
/// bottleneck is pre-`PacketRx` dequeue backlog, so a wider receive batch
/// reduces syscall/channel-item churn without changing the priority/bulk lane
/// contract at the packet channel boundary.
#[cfg(any(target_os = "linux", target_os = "macos"))]
pub(crate) const UDP_RECV_BATCH_SIZE: usize = 128;

#[cfg(target_os = "linux")]
pub(crate) fn reset_recv_buffer(buffer: &mut Vec<u8>) {
    buffer.clear();
}

#[cfg(target_os = "macos")]
pub(crate) fn reset_recv_buffer(_buffer: &mut Vec<u8>) {}

#[cfg(target_os = "linux")]
fn linux_udp_rcvbuf_errors() -> Option<u64> {
    let contents = std::fs::read_to_string("/proc/net/snmp").ok()?;
    parse_proc_net_snmp_udp_rcvbuf_errors(&contents)
}

#[cfg(any(target_os = "linux", test))]
fn parse_proc_net_snmp_udp_rcvbuf_errors(contents: &str) -> Option<u64> {
    let mut lines = contents.lines();
    while let Some(header) = lines.next() {
        if !header.starts_with("Udp:") {
            continue;
        }
        let values = lines.next()?;
        if !values.starts_with("Udp:") {
            continue;
        }
        let header_fields: Vec<&str> = header.split_whitespace().collect();
        let value_fields: Vec<&str> = values.split_whitespace().collect();
        let idx = header_fields
            .iter()
            .position(|field| *field == "RcvbufErrors")?;
        return value_fields.get(idx)?.parse().ok();
    }
    None
}

fn socket_addr_families_compatible(local: SocketAddr, remote: SocketAddr) -> bool {
    matches!(
        (local, remote),
        (SocketAddr::V4(_), SocketAddr::V4(_)) | (SocketAddr::V6(_), SocketAddr::V6(_))
    )
}

/// Threshold above which `send_async` triggers a sendmmsg flush
/// instead of just buffering. Matches the rx_loop's per-drain cap
/// (256) so the trailing-burst flush at the end of a drain cycle can
/// land in a single kernel syscall. The previous value (32) saw the
/// per-batch sendmmsg cost dominate at multi-Gbps single-stream: the
/// FIPS_PERF profile showed ~2.1 µs amortised per packet on the send
/// path (~37% of one core at 164 kpps) with threshold=32, almost all
/// UDP transport for FIPS.
///
/// Provides connectionless, unreliable packet delivery over UDP/IP.
/// A single socket serves all peers; links are virtual tuples of
/// (transport_id, remote_addr).
///
/// **No per-transport send buffering.** An earlier iteration of this
/// transport (commit 5929019) maintained a `pending_send` queue and
/// flushed it via `sendmmsg(2)` once a threshold was hit, in order
/// to amortise the per-syscall cost on the bulk-data hot path. That
/// path now flows through the encrypt worker pool — which does its
/// own `sendmmsg(2)` (target-grouped) directly on the raw fd — so
/// `send_async` is left handling only low-rate handshakes, MMP
/// reports, control messages, and rekeys (typical aggregate < 100
/// pps). The buffered version silently dropped packets in those
/// paths: idle tick / decrypt-fallback / control branches could leave
/// a heartbeat in the buffer until the next inbound batch arrived.
/// Result was MMP link-dead timeouts on idle peers + 60+ failing
/// integration tests (which construct `UdpTransport` outside the
/// rx_loop entirely). One sendmmsg-with-1 ≈ one sendto in kernel
/// time; the bulk path already gets real batching elsewhere.
pub struct UdpTransport {
    /// Unique transport identifier.
    transport_id: TransportId,
    /// Optional instance name (for named instances in config).
    name: Option<String>,
    /// Configuration.
    config: UdpConfig,
    /// Current state.
    state: TransportState,
    /// Bound socket (None until started).
    socket: Option<AsyncUdpSocket>,
    /// Channel for delivering received packets to Node.
    packet_tx: PacketTx,
    /// Receive loop task handle.
    recv_task: Option<JoinHandle<()>>,
    /// Local bound address (after start).
    local_addr: Option<SocketAddr>,
    /// Transport statistics.
    stats: Arc<UdpStats>,
    /// Linux namespace-level UDP receive-buffer error baseline.
    ///
    /// This is broader than the wildcard socket. It is reported separately
    /// from `SO_RXQ_OVFL` so benchmark artifacts can distinguish this socket
    /// dropping from unrelated UDP receive-buffer pressure in the namespace.
    #[cfg(target_os = "linux")]
    udp_rcvbuf_error_baseline: u64,
    /// DNS resolution cache for hostname addresses.
    dns_cache: StdMutex<HashMap<TransportAddr, (SocketAddr, Instant)>>,
}

impl UdpTransport {
    /// Create a new UDP transport.
    pub fn new(
        transport_id: TransportId,
        name: Option<String>,
        config: UdpConfig,
        packet_tx: PacketTx,
    ) -> Self {
        Self {
            transport_id,
            name,
            config,
            state: TransportState::Configured,
            socket: None,
            packet_tx,
            recv_task: None,
            local_addr: None,
            stats: Arc::new(UdpStats::new()),
            #[cfg(target_os = "linux")]
            udp_rcvbuf_error_baseline: linux_udp_rcvbuf_errors().unwrap_or(0),
            dns_cache: StdMutex::new(HashMap::new()),
        }
    }

    /// Get the instance name (if configured as a named instance).
    pub fn name(&self) -> Option<&str> {
        self.name.as_deref()
    }

    /// Get the local bound address (only valid after start).
    pub fn local_addr(&self) -> Option<SocketAddr> {
        self.local_addr
    }

    /// Configured recv buffer size — used when opening per-peer
    /// `ConnectedPeerSocket`s so they get the same buffer ceiling as
    /// the wildcard listen socket.
    pub fn recv_buf_size(&self) -> usize {
        self.config.recv_buf_size()
    }

    /// Configured send buffer size — companion to `recv_buf_size`.
    pub fn send_buf_size(&self) -> usize {
        self.config.send_buf_size()
    }

    /// Clone the `PacketTx` end of the packet channel for off-task
    /// receive paths (per-peer connected-socket drains, future shard
    /// recv loops). The clone is just a refcount bump.
    pub fn clone_packet_tx(&self) -> PacketTx {
        self.packet_tx.clone()
    }

    /// Get the transport statistics.
    pub fn stats(&self) -> &Arc<UdpStats> {
        &self.stats
    }

    /// Resolve a transport address (which may be a string like
    /// `"1.2.3.4:5678"` or a hostname) to a kernel `SocketAddr`,
    /// using the per-transport DNS cache. Public companion to
    /// `async_socket()` for off-task workers that want to skip the
    /// per-packet address parse / DNS lookup that `send_async` does
    /// inline. Returns `Err` if neither numeric parse nor DNS resolves
    /// the address.
    pub async fn resolve_for_off_task(
        &self,
        addr: &TransportAddr,
    ) -> Result<SocketAddr, TransportError> {
        self.resolve_cached(addr).await
    }

    /// Clone the underlying async UDP socket (internally an
    /// `Arc<AsyncFd<UdpRawSocket>>`, so the "clone" is just a refcount
    /// bump). Returns `None` if the transport hasn't been started yet.
    ///
    /// Intended for off-task workers that need to issue raw
    /// `send_to` / `send_batch` calls — useful when the AEAD
    /// encrypt + UDP-send pipeline is parallelised across N worker
    /// threads that each own a shared handle to the same kernel
    /// socket. The kernel serialises concurrent `sendto` calls
    /// itself, so concurrent userland sends are safe.
    pub fn async_socket(&self) -> Option<AsyncUdpSocket> {
        self.socket.clone()
    }

    /// Resolve a transport address, using cached results for hostnames.
    ///
    /// Numeric IP addresses bypass the cache entirely. Hostnames are
    /// resolved via DNS and cached for `DNS_CACHE_TTL` to avoid
    /// per-packet resolution overhead.
    async fn resolve_cached(&self, addr: &TransportAddr) -> Result<SocketAddr, TransportError> {
        // Fast path: try numeric IP parse (no cache, no DNS)
        if let Some(s) = addr.as_str()
            && let Ok(sock_addr) = s.parse::<SocketAddr>()
        {
            return Ok(sock_addr);
        }

        // Check cache
        {
            let cache = self.dns_cache.lock().unwrap_or_else(|e| e.into_inner());
            if let Some((resolved, cached_at)) = cache.get(addr)
                && cached_at.elapsed() < DNS_CACHE_TTL
            {
                return Ok(*resolved);
            }
        }

        // Cache miss or expired — resolve via DNS
        let resolved = resolve_socket_addr(addr).await?;

        // Store in cache
        {
            let mut cache = self.dns_cache.lock().unwrap_or_else(|e| e.into_inner());
            cache.insert(addr.clone(), (resolved, Instant::now()));
        }

        Ok(resolved)
    }

    /// Query transport-local congestion indicators.
    pub fn congestion(&self) -> super::TransportCongestion {
        let socket_drops = self
            .stats
            .kernel_drops
            .load(std::sync::atomic::Ordering::Relaxed);
        #[cfg(target_os = "linux")]
        let namespace_drops = linux_udp_rcvbuf_errors()
            .unwrap_or(self.udp_rcvbuf_error_baseline)
            .saturating_sub(self.udp_rcvbuf_error_baseline);
        #[cfg(target_os = "linux")]
        let recv_drops = socket_drops.max(namespace_drops);
        #[cfg(not(target_os = "linux"))]
        let recv_drops = socket_drops;

        super::TransportCongestion {
            recv_drops: Some(recv_drops),
            socket_recv_drops: Some(socket_drops),
            #[cfg(target_os = "linux")]
            namespace_recv_drops: Some(namespace_drops),
            #[cfg(not(target_os = "linux"))]
            namespace_recv_drops: None,
        }
    }

    /// Start the transport asynchronously.
    ///
    /// Binds the UDP socket and spawns the receive loop.
    pub async fn start_async(&mut self) -> Result<(), TransportError> {
        if !self.state.can_start() {
            return Err(TransportError::AlreadyStarted);
        }

        self.state = TransportState::Starting;

        if self.config.outbound_only() && self.config.bind_addr.is_some() {
            warn!(
                configured_bind_addr = ?self.config.bind_addr,
                "udp.outbound_only = true; configured bind_addr is ignored, binding to 0.0.0.0:0"
            );
        }

        // Parse bind address
        let bind_addr: SocketAddr = self
            .config
            .bind_addr()
            .parse()
            .map_err(|e| TransportError::StartFailed(format!("invalid bind address: {}", e)))?;

        // Create, bind, and configure UDP socket
        let raw_socket = UdpRawSocket::open(
            bind_addr,
            self.config.recv_buf_size(),
            self.config.send_buf_size(),
        )?;

        let actual_recv = raw_socket.recv_buffer_size()?;
        let actual_send = raw_socket.send_buffer_size()?;
        self.local_addr = Some(raw_socket.local_addr());

        // Wrap in AsyncFd for tokio integration
        let async_socket = raw_socket.into_async()?;
        self.socket = Some(async_socket.clone());

        // Spawn receive loop
        let transport_id = self.transport_id;
        let packet_tx = self.packet_tx.clone();
        let mtu = self.config.mtu();
        let stats = self.stats.clone();

        let recv_task = tokio::spawn(async move {
            udp_receive_loop(async_socket, transport_id, packet_tx, mtu, stats).await;
        });

        self.recv_task = Some(recv_task);
        self.state = TransportState::Up;

        if let Some(ref name) = self.name {
            info!(
                name = %name,
                local_addr = %self.local_addr.map_or_else(|| "<unbound>".to_string(), |addr| addr.to_string()),
                recv_buf = actual_recv,
                send_buf = actual_send,
                "UDP transport started"
            );
        } else {
            info!(
                local_addr = %self.local_addr.map_or_else(|| "<unbound>".to_string(), |addr| addr.to_string()),
                recv_buf = actual_recv,
                send_buf = actual_send,
                "UDP transport started"
            );
        }

        Ok(())
    }

    /// Start the transport using an already-bound UDP socket.
    ///
    /// This preserves an existing NAT mapping established by another
    /// subsystem, such as STUN or UDP hole punching.
    pub async fn adopt_socket_async(
        &mut self,
        socket: std::net::UdpSocket,
    ) -> Result<(), TransportError> {
        if !self.state.can_start() {
            return Err(TransportError::AlreadyStarted);
        }

        self.state = TransportState::Starting;

        let raw_socket = UdpRawSocket::adopt(
            socket,
            self.config.recv_buf_size(),
            self.config.send_buf_size(),
        )?;

        let actual_recv = raw_socket.recv_buffer_size()?;
        let actual_send = raw_socket.send_buffer_size()?;
        self.local_addr = Some(raw_socket.local_addr());

        let async_socket = raw_socket.into_async()?;
        self.socket = Some(async_socket.clone());

        let transport_id = self.transport_id;
        let packet_tx = self.packet_tx.clone();
        let mtu = self.config.mtu();
        let stats = self.stats.clone();

        let recv_task = tokio::spawn(async move {
            udp_receive_loop(async_socket, transport_id, packet_tx, mtu, stats).await;
        });

        self.recv_task = Some(recv_task);
        self.state = TransportState::Up;

        if let Some(ref name) = self.name {
            info!(
                name = %name,
                local_addr = %self.local_addr.map_or_else(|| "<unbound>".to_string(), |addr| addr.to_string()),
                recv_buf = actual_recv,
                send_buf = actual_send,
                "UDP transport adopted existing socket"
            );
        } else {
            info!(
                local_addr = %self.local_addr.map_or_else(|| "<unbound>".to_string(), |addr| addr.to_string()),
                recv_buf = actual_recv,
                send_buf = actual_send,
                "UDP transport adopted existing socket"
            );
        }

        Ok(())
    }

    /// Stop the transport asynchronously.
    pub async fn stop_async(&mut self) -> Result<(), TransportError> {
        if !self.state.is_operational() {
            return Err(TransportError::NotStarted);
        }

        // Abort receive task
        if let Some(task) = self.recv_task.take() {
            task.abort();
            let _ = task.await; // Ignore JoinError from abort
        }

        // Drop socket
        self.socket.take();
        self.local_addr = None;

        self.state = TransportState::Down;

        info!(
            transport_id = %self.transport_id,
            "UDP transport stopped"
        );

        Ok(())
    }

    /// Send a packet asynchronously.
    ///
    /// One syscall per call (`sendto(2)` on macOS / BSD, `sendmsg(2)`
    /// on Linux via the AsyncUdpSocket wrapper). No batching at this
    /// layer — see the module docs for why the previous buffered
    /// implementation was removed.
    pub async fn send_async(
        &self,
        addr: &TransportAddr,
        data: &[u8],
    ) -> Result<usize, TransportError> {
        if !self.state.is_operational() {
            return Err(TransportError::NotStarted);
        }

        if data.len() > self.config.mtu() as usize {
            self.stats.record_mtu_exceeded();
            return Err(TransportError::MtuExceeded {
                packet_size: data.len(),
                mtu: self.config.mtu(),
            });
        }

        let socket_addr = self.resolve_cached(addr).await?;
        let socket = self.socket.as_ref().ok_or(TransportError::NotStarted)?;
        let local_addr = self.local_addr.ok_or(TransportError::NotStarted)?;
        if !socket_addr_families_compatible(local_addr, socket_addr) {
            return Err(TransportError::InvalidAddress(format!(
                "remote address family {socket_addr} is incompatible with local UDP socket {local_addr}"
            )));
        }
        match socket.send_to(data, &socket_addr).await {
            Ok(bytes_sent) => {
                self.stats.record_send(bytes_sent);
                trace!(
                    transport_id = %self.transport_id,
                    remote_addr = %socket_addr,
                    bytes = bytes_sent,
                    "UDP packet sent"
                );
                Ok(bytes_sent)
            }
            Err(e) => {
                self.stats.record_send_error();
                Err(e)
            }
        }
    }
}

impl Transport for UdpTransport {
    fn transport_id(&self) -> TransportId {
        self.transport_id
    }

    fn transport_type(&self) -> &TransportType {
        &TransportType::UDP
    }

    fn state(&self) -> TransportState {
        self.state
    }

    fn mtu(&self) -> u16 {
        self.config.mtu()
    }

    fn start(&mut self) -> Result<(), TransportError> {
        // Synchronous start not supported - use start_async()
        Err(TransportError::NotSupported(
            "use start_async() for UDP transport".into(),
        ))
    }

    fn stop(&mut self) -> Result<(), TransportError> {
        // Synchronous stop not supported - use stop_async()
        Err(TransportError::NotSupported(
            "use stop_async() for UDP transport".into(),
        ))
    }

    fn send(&self, _addr: &TransportAddr, _data: &[u8]) -> Result<(), TransportError> {
        // Synchronous send not supported - use send_async()
        Err(TransportError::NotSupported(
            "use send_async() for UDP transport".into(),
        ))
    }

    fn discover(&self) -> Result<Vec<DiscoveredPeer>, TransportError> {
        // UDP discovery not yet implemented (would use multicast/DNS-SD)
        // Peer configuration is handled at the node level, not transport level
        Ok(Vec::new())
    }

    /// Whether the transport accepts inbound handshake initiations.
    /// `outbound_only` mode forces this to false; otherwise reflects the
    /// `accept_connections` config field (default: true). Note that the
    /// hard gate is at the Node level (see ISSUE-2026-0004 fix in
    /// `src/node/handlers/handshake.rs`); this method is what that gate
    /// consults for transports that lack runtime-state-based filtering.
    fn accept_connections(&self) -> bool {
        if self.config.outbound_only() {
            false
        } else {
            self.config.accept_connections()
        }
    }
}

impl Drop for UdpTransport {
    fn drop(&mut self) {
        let had_task = self.recv_task.is_some();
        let had_socket = self.socket.is_some();
        if had_task || had_socket {
            debug!(
                transport_id = %self.transport_id,
                state = ?self.state,
                had_recv_task = had_task,
                had_socket = had_socket,
                "UdpTransport dropped without stop_async(); cleaning up",
            );
        }
        if let Some(task) = self.recv_task.take() {
            task.abort();
        }
        self.socket.take();
        self.local_addr = None;
    }
}

/// UDP receive loop - runs as a spawned task.
///
/// On Linux, drains the kernel UDP queue in `UDP_RECV_BATCH_SIZE` bursts via
/// `recvmmsg` to amortise the per-syscall + per-task-wakeup overhead. macOS /
/// Windows fall through to single-packet `recv_from`. Either way every
/// datagram is forwarded to `packet_tx` in arrival order.
async fn udp_receive_loop(
    socket: AsyncUdpSocket,
    transport_id: TransportId,
    packet_tx: PacketTx,
    mtu: u16,
    stats: Arc<UdpStats>,
) {
    debug!(transport_id = %transport_id, "UDP receive loop starting");

    #[cfg(target_os = "linux")]
    {
        const BATCH: usize = UDP_RECV_BATCH_SIZE;
        let buf_size = mtu as usize + 100;
        // Backing pool: one Vec<u8> per recvmmsg slot. We **own** each
        // slot here — when a packet lands, we `mem::replace` the filled
        // Vec out (handing the buffer directly to rx_loop via mpsc) and
        // drop in a fresh capacity-only Vec to refill that slot on the
        // next call.
        //
        // Previous code did `let data = buf.to_vec();` per packet,
        // which was 1 alloc + 1 memcpy of the entire packet (~1.5 KB)
        // for every received UDP datagram. At 100 kpps that's
        // ~150 MB/sec of avoidable memory bandwidth on the RX hot path.
        // The new code does the same alloc count (one fresh Vec to
        // refill the slot) but zero per-packet memcpy and no per-refill
        // memset on Linux — the receive buffer becomes the packet buffer
        // in one move.
        let mut backing: Vec<Vec<u8>> = (0..BATCH)
            .map(|_| packet_tx.recv_buffer(buf_size))
            .collect();
        let mut addrs: [Option<std::net::SocketAddr>; BATCH] = std::array::from_fn(|_| None);

        loop {
            let recv_result = {
                let _t = crate::perf_profile::Timer::start(crate::perf_profile::Stage::UdpRecv);
                socket.recv_batch(&mut backing, &mut addrs).await
            };
            match recv_result {
                Ok((count, kernel_drops)) => {
                    stats.set_kernel_drops(kernel_drops as u64);
                    let timestamp_ms = received_timestamp_ms();
                    let trace_enqueued_at = crate::perf_profile::stamp();
                    let mut packets = packet_tx.packet_batch(count);
                    for i in 0..count {
                        let len = backing[i].len();
                        let Some(remote_addr) = addrs[i].take() else {
                            reset_recv_buffer(&mut backing[i]);
                            continue;
                        };
                        stats.record_recv(len);

                        // Peek before swap: punch probes / acks are
                        // discarded without consuming a buffer move.
                        if is_punch_packet(&backing[i][..len]) {
                            trace!(
                                transport_id = %transport_id,
                                remote_addr = %remote_addr,
                                bytes = len,
                                "Dropping stray punch probe/ack on UDP transport"
                            );
                            reset_recv_buffer(&mut backing[i]);
                            continue;
                        }

                        // Move the filled buffer out of the slot and
                        // refill with a fresh one. `mem::replace`
                        // returns the OLD value and writes the new one
                        // — single pointer swap, no copy.
                        let data =
                            std::mem::replace(&mut backing[i], packet_tx.recv_buffer(buf_size));
                        let addr = TransportAddr::from_socket_addr(remote_addr);
                        let packet = ReceivedPacket::with_trace_timestamp(
                            transport_id,
                            addr,
                            packet_tx.packet_buffer(data),
                            timestamp_ms,
                            trace_enqueued_at,
                        );

                        trace!(
                            transport_id = %transport_id,
                            remote_addr = %remote_addr,
                            bytes = len,
                            "UDP packet received"
                        );

                        packets.push(packet);
                    }

                    if !packets.is_empty() && packet_tx.send_packet_batch(packets).is_err() {
                        debug!(
                            transport_id = %transport_id,
                            "Packet channel closed, stopping receive loop"
                        );
                        return;
                    }
                }
                Err(e) => {
                    stats.record_recv_error();
                    warn!(
                        transport_id = %transport_id,
                        error = %e,
                        "UDP receive error"
                    );
                }
            }
        }
    }

    #[cfg(not(target_os = "linux"))]
    {
        let mut buf = vec![0u8; mtu as usize + 100];

        loop {
            match socket.recv_from(&mut buf).await {
                Ok((len, remote_addr, kernel_drops)) => {
                    stats.record_recv(len);
                    stats.set_kernel_drops(kernel_drops as u64);

                    if is_punch_packet(&buf[..len]) {
                        trace!(
                            transport_id = %transport_id,
                            remote_addr = %remote_addr,
                            bytes = len,
                            "Dropping stray punch probe/ack on UDP transport"
                        );
                        continue;
                    }

                    let data = buf[..len].to_vec();
                    let addr = TransportAddr::from_socket_addr(remote_addr);
                    let packet = ReceivedPacket::new(transport_id, addr, data);

                    trace!(
                        transport_id = %transport_id,
                        remote_addr = %remote_addr,
                        bytes = len,
                        "UDP packet received"
                    );

                    if packet_tx.send(packet).is_err() {
                        debug!(
                            transport_id = %transport_id,
                            "Packet channel closed, stopping receive loop"
                        );
                        break;
                    }
                }
                Err(e) => {
                    stats.record_recv_error();
                    warn!(
                        transport_id = %transport_id,
                        error = %e,
                        "UDP receive error"
                    );
                }
            }
        }
    }
}

// ============================================================================
// Tests
// ============================================================================

#[cfg(test)]
mod tests;