offline_protocol/protocol/receive.rs
1//! Message receive loop and file chunk handling.
2
3use super::mesh_relay::RelayAdmission as MeshRelayAdmission;
4use super::{
5 internal_prefixes, lock_shared_state, ChunkOutcome, InternalMessageResult, OfflineProtocol,
6 PendingMediaMetadataEntry, ProtocolState, RichPayloadV1, CRITICAL_RELAY_BATTERY_LEVEL,
7};
8use crate::constants::ACK_FOR_KEY;
9use crate::events::{Event, SecurityWarningCode};
10use crate::file_transfer::FileChunk;
11use crate::media_envelope::{decode_media_envelope, is_media_envelope, MediaChunkPlaintext};
12use crate::mls_observability::{DecryptionFailureKind, MlsErrorCategory, MlsOperationContext};
13use crate::SessionStateError;
14use offline_protocol_core::{ContentType, MediaMetadata, Message};
15use offline_protocol_mls::{EncryptedMessage, GroupId};
16use offline_protocol_router::relay::RelayPriority;
17use offline_protocol_transport::TransportType;
18use std::time::Instant;
19use tracing::{debug, error, info, warn};
20
21impl OfflineProtocol {
22 /// Applies a successful MLS decryption to an inbound message: swaps the
23 /// ciphertext for the plaintext, marks the message encrypted, drops the
24 /// outer `reply_context`, and restores rich fields from a sealed
25 /// `__RICH_V1__` body when present.
26 ///
27 /// The outer `reply_context` field is hop-visible cleartext that any
28 /// relay can inject or rewrite in transit — it sits outside the MLS AEAD
29 /// boundary. The strip happens unconditionally *before* the sealed-body
30 /// restore below, so the encrypted envelope is the only trusted carrier
31 /// for reply context on encrypted messages. Without this strip, a
32 /// `MessageReceived { encrypted: true }` event could surface an
33 /// attacker-controlled quote preview as if it were part of the
34 /// authenticated conversation.
35 ///
36 /// Sealed-body parsing is never capability-gated (mirroring envelope
37 /// parsing): whatever a peer chose to seal, we try to read. On a rich
38 /// message the sealed body is authoritative for `reply_context`,
39 /// `media_metadata`, and `forwarded_from` — the relay-writable outer
40 /// copies are overwritten wholesale, `None`s included — and for
41 /// `content_type` when the body carries one (absent from bodies sealed
42 /// by senders predating the field, where the outer hint stands). The
43 /// sealed `content_type` restore runs before the receive loop's
44 /// `FileChunk` routing, so a relay restamping an encrypted rich
45 /// message's outer hint can no longer misroute it; `FileChunk` itself
46 /// is refused from the sealed body (mirroring the send boundary), or a
47 /// hostile sender could steer an ordinary message into the
48 /// file-transfer manager, which drops it. A body that fails to parse
49 /// (including a hostile `reply_context.sender` rejected by `UserId`
50 /// validation) surfaces as raw text with a warning rather than
51 /// dropping an authenticated message.
52 pub(super) fn apply_decrypted_content(message: &mut Message, plaintext: String) {
53 message.reply_context = None;
54 if plaintext.starts_with(internal_prefixes::RICH_V1) {
55 match RichPayloadV1::parse_sealed(&plaintext, message.sender.as_str()) {
56 Some(rich) => {
57 message.content = rich.text;
58 message.reply_context = rich.reply_context;
59 message.media_metadata = rich.media_metadata;
60 message.forwarded_from = rich.forward_info;
61 // `parse_sealed` already refused a FileChunk claim; a
62 // remaining None keeps the outer value (bodies sealed by
63 // senders predating the field).
64 if let Some(content_type) = rich.content_type {
65 message.content_type = content_type;
66 }
67 }
68 None => {
69 message.content = plaintext;
70 }
71 }
72 } else {
73 message.content = plaintext;
74 }
75 message
76 .metadata
77 .insert("encrypted".to_string(), "true".to_string());
78 }
79
80 /// Receives the next available message.
81 pub fn receive_message(&mut self) -> Option<Message> {
82 let Ok(mut state) = lock_shared_state(&self.shared_state) else {
83 error!("Failed to lock shared state in receive_message");
84 return None;
85 };
86 let protocol_running = state.state == ProtocolState::Running;
87
88 if !state.received_messages.is_empty() {
89 return state.received_messages.pop_front();
90 }
91
92 drop(state);
93
94 // Drive confirmation maintenance from receive polling as an additional
95 // liveness source when the app does not call process() on a timer.
96 // Uses the same throttle as process() to avoid redundant storage I/O.
97 if protocol_running {
98 self.run_throttled_reconciliation("receive_message_poll");
99 }
100
101 loop {
102 match self.transport_manager.receive() {
103 Ok(Some((transport_used, mut message))) => {
104 // The peer whose link this frame physically arrived on.
105 // Distinct from `message.sender`, which is who wrote it:
106 // the two agree only at the first hop. `None` when the
107 // carrier does not identify links (Nostr) or the platform
108 // did not supply an id.
109 let arrival_peer = message.transport_peer_id().map(str::to_string);
110
111 // Traffic for someone else is forwarded, not processed, and
112 // that decision comes first.
113 //
114 // Everything below this point treats the frame as part of
115 // our own exchange: it merges the sender's clock into ours,
116 // settles our outbox from acknowledgements, and consumes the
117 // frame. None of that is true of a frame merely passing
118 // through us — its acknowledgements belong to the pair it
119 // travels between, and once a node carries the whole
120 // neighborhood's traffic, absorbing every clock it sees
121 // would drag ours to the network's maximum.
122 //
123 // Never forward a frame claiming our own origin: a genuine
124 // self-originated frame is never received inbound with a
125 // foreign recipient (the send path does not loop back), so
126 // this is a routing loop or a forgery aimed at
127 // `internet_control_op`'s self-origination gate — re-issuing
128 // it from our outbox would let a mesh peer drive
129 // relay-native control ops on our authenticated relay
130 // connection. (A sender==self frame addressed *to* us is the
131 // legitimate relay echo and falls through unchanged.)
132 if message.recipient.as_str() != self.local_id {
133 if message.sender.as_str() == self.local_id {
134 debug!(
135 message_id = %message.id,
136 recipient = %message.recipient,
137 "Dropping inbound frame forging our own origin"
138 );
139 continue;
140 }
141 self.try_relay_message(&message, arrival_peer.as_deref());
142 continue;
143 }
144
145 // Block filter (early): check before any side-effects so
146 // that blocked users cannot advance our Lamport clock,
147 // leak our presence via re-ACK, or trigger any processing.
148 let sender_blocked = self.is_user_blocked(message.sender.as_str());
149
150 // Merge Lamport clock for every non-blocked received message
151 // — including duplicates, ACKs, and internal protocol
152 // messages — so the local clock always advances past any
153 // observed peer value.
154 if !sender_blocked && message.lamport_clock.value() > 0 {
155 self.lamport_clock.merge(message.lamport_clock);
156 self.persist_lamport_clock();
157 }
158
159 if message.metadata.contains_key(ACK_FOR_KEY) {
160 if !sender_blocked {
161 self.handle_ack_message(&message);
162 }
163 continue;
164 }
165
166 if self.deduplicator.is_duplicate_mut(&message.id) {
167 // Re-ACK duplicate packets so the sender can stop
168 // retrying — but NOT for blocked users, to avoid
169 // leaking presence information.
170 //
171 // Every copy is answered, including the several that
172 // one delivery produces when it travels through nearby
173 // devices. Answering only the first was tried and
174 // rejected: a copy arriving moments later and a
175 // retransmission arriving because the answer was lost
176 // are not distinguishable here, and staying quiet for
177 // the second turns a delivered message into a failed
178 // one. The redundant answers are bounded — a delivery
179 // arrives by at most as many paths as neighbors carried
180 // it, answers never provoke further answers, and every
181 // frame that leaves is rate-capped like any other.
182 if !sender_blocked && message.requires_ack {
183 if let Err(err) = self.send_delivery_ack(&message, transport_used) {
184 error!(
185 message_id = %message.id,
186 error = %err,
187 "Failed to send delivery ACK for duplicate message"
188 );
189 }
190 }
191 continue;
192 }
193
194 // Block filter: silently drop messages from blocked users
195 // addressed to us. No ACK, no event, no side-effects.
196 // Checked before mark_seen so that if the user is later
197 // unblocked, retransmissions can still be delivered.
198 if sender_blocked {
199 debug!(
200 sender = %message.sender,
201 message_id = %message.id,
202 "Dropping message from blocked user"
203 );
204 continue;
205 }
206
207 self.mark_seen_persisted(message.id.clone());
208
209 // Handle internal MLS messages
210 let mut was_decrypted = false;
211 if let Some(result) =
212 self.process_internal_message_via(&message, Some(transport_used))
213 {
214 match result {
215 InternalMessageResult::Consumed => {
216 // Internal control messages are still delivery-sensitive for
217 // the sender (invites/accept/welcome). ACK before consume.
218 if message.requires_ack {
219 if let Err(err) =
220 self.send_delivery_ack(&message, transport_used)
221 {
222 error!(
223 message_id = %message.id,
224 error = %err,
225 "Failed to send delivery ACK for internal message"
226 );
227 }
228 }
229 // Internal message handled, don't surface to app
230 continue;
231 }
232 InternalMessageResult::SecurityRejected => {
233 // Security gate rejected this message (spoofed sender,
234 // bad signature, unsigned control traffic, or a signing
235 // key that does not derive to the claimed sender
236 // address). Do NOT send a delivery
237 // ACK — acknowledging would confirm to the attacker that
238 // the target peer is online and processing messages.
239 // Forget the id too, or an exact replay would hit the
240 // duplicate re-ACK path above and leak that presence
241 // anyway; reprocessing a replay costs no more than a
242 // fresh forged message.
243 self.unmark_seen_persisted(&message.id);
244 continue;
245 }
246 InternalMessageResult::Deferred => {
247 // The message was not delivered, but the sender
248 // can still recover it by resending: it could not
249 // be decrypted *yet* (session not ready, so queued
250 // for delayed decryption), or its envelope did not
251 // parse. See the conditions on
252 // `InternalMessageResult::Deferred`.
253 //
254 // Do NOT send a delivery ACK and do NOT keep the id
255 // dedup-marked: the message is not delivered, so the
256 // sender must be free to retry and that retry must
257 // re-enter processing rather than hit the duplicate
258 // re-ACK path above. For a queued copy, the session
259 // confirming drains it: the message is surfaced and
260 // the id re-marked (see `process_pending_decryption`),
261 // so the sender's next resend is then deduped +
262 // re-ACKed. For an unparseable envelope, recovery
263 // is that resend itself.
264 self.unmark_seen_persisted(&message.id);
265 continue;
266 }
267 InternalMessageResult::Undecryptable => {
268 // As `Deferred`, but answered: the sender's
269 // re-sealed resend must still re-enter
270 // processing, and whoever holds this copy
271 // must stop redelivering it.
272 self.unmark_seen_persisted(&message.id);
273 self.send_undecryptable_ack(&message, transport_used);
274 continue;
275 }
276 InternalMessageResult::Decrypted(plaintext) => {
277 was_decrypted = true;
278 Self::apply_decrypted_content(&mut message, plaintext);
279 }
280 }
281 }
282
283 // Policy gate for inbound plaintext (SEC: inbound text must
284 // honor require_encryption exactly like legacy media does).
285 // A message that reaches this point without decryption is
286 // unauthenticated cleartext with an attacker-controllable
287 // sender. FileChunk messages are exempt: encrypted media
288 // envelopes carry no content prefix (the ciphertext rides
289 // in binary_content), so they are indistinguishable from
290 // plaintext here — handle_incoming_file_chunk applies this
291 // same gate after telling the two apart. Rejection sends
292 // no delivery ACK, mirroring SecurityRejected: don't
293 // confirm to an injector that the target processes their
294 // messages. The id is forgotten so a replay re-enters
295 // this gate instead of the duplicate re-ACK path.
296 if !was_decrypted
297 && message.content_type != ContentType::FileChunk
298 && !self.accept_plaintext_content(message.sender.as_str())
299 {
300 warn!(
301 message_id = %message.id,
302 sender = %message.sender,
303 "Rejecting unencrypted inbound message (encryption policy)"
304 );
305 self.warn_plaintext_receive_rejected(
306 message.sender.as_str(),
307 "Inbound plaintext message rejected by encryption policy",
308 );
309 self.unmark_seen_persisted(&message.id);
310 continue;
311 }
312
313 // Route file-chunk messages to the transfer manager BEFORE
314 // the delivery ACK, and never surface them to the app as
315 // regular messages. An encrypted chunk that was not
316 // delivered but that the sender can still recover by
317 // resending — not decryptable yet (session not ready, so
318 // queued for delayed decryption), or undecryptable as it
319 // stands (epoch desync, crypto failure) — returns
320 // `Deferred`: it must NOT be ACKed and its id is unmarked,
321 // so the sender keeps retrying and the resend re-enters
322 // processing, matching the text `Deferred` path. Every other
323 // outcome (`Handled`: decrypted/assembled, or a terminal
324 // drop) is ACKed as before, since the sender cannot recover
325 // it by retrying.
326 if message.content_type == ContentType::FileChunk {
327 match self.handle_incoming_file_chunk_via(&message, Some(transport_used)) {
328 ChunkOutcome::Deferred => {
329 self.unmark_seen_persisted(&message.id);
330 }
331 // Plaintext chunk rejected by encryption policy:
332 // withhold the ACK and unmark the id, exactly like
333 // the plaintext-text rejection above and the text
334 // `SecurityRejected` path — don't confirm to an
335 // injector that we process their messages, and let a
336 // replay re-enter the gate rather than hit the
337 // duplicate re-ACK path.
338 ChunkOutcome::Rejected => {
339 self.unmark_seen_persisted(&message.id);
340 }
341 ChunkOutcome::Handled => {
342 if message.requires_ack {
343 if let Err(err) =
344 self.send_delivery_ack(&message, transport_used)
345 {
346 error!(
347 message_id = %message.id,
348 error = %err,
349 "Failed to send delivery ACK for media chunk"
350 );
351 }
352 }
353 }
354 }
355 continue;
356 }
357
358 if message.requires_ack {
359 if let Err(err) = self.send_delivery_ack(&message, transport_used) {
360 error!(
361 message_id = %message.id,
362 error = %err,
363 "Failed to send delivery ACK"
364 );
365 }
366 }
367
368 let forward_info = message
369 .forwarded_from
370 .as_ref()
371 .map(crate::events::ForwardInfoEvent::from);
372
373 let event = Event::MessageReceived {
374 message_id: message.id.as_str(),
375 sender: message.sender.as_str().to_string(),
376 recipient: message.recipient.as_str().to_string(),
377 content: message.content.clone(),
378 hop_count: message.hop_count.value(),
379 transport: transport_used.to_string(),
380 timestamp: message.timestamp.as_millis(),
381 lamport_clock: message.lamport_clock.value(),
382 reply_to_msg: message
383 .reply_to_msg
384 .as_ref()
385 .map(|id| id.as_str().to_string()),
386 reply_context: message
387 .reply_context
388 .as_ref()
389 .map(|rc| Box::new(crate::events::ReplyContextEvent::from(rc))),
390 content_type: message.content_type.to_string(),
391 media_metadata: message.media_metadata.clone(),
392 forward_info,
393 encrypted: was_decrypted,
394 app_id: message.app_id.as_str().to_string(),
395 };
396
397 let Ok(state) = lock_shared_state(&self.shared_state) else {
398 error!("Failed to lock shared state for message received event");
399 return None;
400 };
401 state.emit_event(event);
402 drop(state);
403
404 return Some(message);
405 }
406 Ok(None) => return None,
407 Err(err) => {
408 error!(error = %err, "Transport receive error");
409 return None;
410 }
411 }
412 }
413 }
414
415 /// Considers an inbound frame addressed to a third party for forwarding.
416 ///
417 /// Learns the route back toward the sender, then offers the frame to the
418 /// governor, which decides whether it travels any further. Nothing is
419 /// transmitted here: an accepted frame is queued and goes out from
420 /// [`Self::flush_mesh_relays`] once its delay elapses, which is what gives
421 /// a neighbor the chance to cover it first.
422 ///
423 /// `arrival_peer` is the neighbor that handed us this frame, when the
424 /// carrier identified the link.
425 fn try_relay_message(&mut self, message: &Message, arrival_peer: Option<&str>) {
426 // Cheapest answer first. This device sees the whole neighborhood's
427 // traffic, so in a crowded room most third-party arrivals are copies of
428 // a frame it has already dealt with — and everything below would be
429 // thrown away by the suppression check inside `admit`: a route-table
430 // write, a battery snapshot that locks and allocates across every
431 // transport, and an enumeration of every link. Answering a duplicate
432 // costs a hash lookup instead.
433 //
434 // Route learning is skipped along with the rest, which is deliberate:
435 // the copy teaches nothing the first one did not, and the table is not
436 // what forwarding decisions are made from. A duplicate we still hold a
437 // pending copy of is *not* absorbed here: it narrows that forward's
438 // fan-out, which `admit` records, so that case still takes the full
439 // path.
440 if self
441 .mesh_relay
442 .absorb_settled_duplicate(message, arrival_peer)
443 {
444 return;
445 }
446
447 // Carrying traffic for others costs this device's radio and battery,
448 // so it stays a local decision.
449 let relay_allowed = self.config.relay.allow_relay
450 && self.config.relay.relay_priority != RelayPriority::Never;
451
452 if !relay_allowed {
453 debug!(
454 message_id = %message.id,
455 "Not forwarding: relaying is disabled on this device"
456 );
457 return;
458 }
459
460 // Carrying other people's messages is the first thing to give up when
461 // the battery is going: a device that spends its last few percent
462 // relaying cannot send its own message when its owner needs to. The
463 // device keeps sending and receiving its own traffic either way.
464 //
465 // An unknown battery level is treated as willing — most platforms
466 // report one, and refusing to carry anything on a device that simply
467 // does not publish a level would quietly remove it from the network.
468 if !self.battery_allows_relaying() {
469 debug!(
470 message_id = %message.id,
471 "Not forwarding: battery is below the level for carrying traffic"
472 );
473 return;
474 }
475
476 let degree = self.transport_manager.mesh_neighbors().len();
477 match self.mesh_relay.admit(message, arrival_peer, degree) {
478 MeshRelayAdmission::Queued => {
479 debug!(
480 message_id = %message.id,
481 recipient = %message.recipient,
482 degree,
483 "Queued frame for forwarding"
484 );
485 }
486 MeshRelayAdmission::Rejected(reason) => {
487 debug!(
488 message_id = %message.id,
489 ?reason,
490 "Not forwarding"
491 );
492 }
493 }
494 }
495
496 /// Whether the battery is healthy enough to carry other people's traffic.
497 ///
498 /// Takes its own battery reading; the per-tick caller already holds one and
499 /// uses [`Self::battery_allows_relaying_with`] instead.
500 pub(super) fn battery_allows_relaying(&self) -> bool {
501 let (_statuses, available) = self.transport_manager.snapshot_status_and_available();
502 let (battery_level, is_charging) =
503 crate::telemetry::aggregator::device_battery_from_available(
504 self.transport_manager.current_transport(),
505 &available,
506 );
507 self.battery_allows_relaying_with(battery_level, is_charging)
508 }
509
510 /// [`Self::battery_allows_relaying`] against a reading the caller already
511 /// has.
512 ///
513 /// An unknown level means yes — see [`Self::try_relay_message`].
514 pub(super) fn battery_allows_relaying_with(
515 &self,
516 battery_level: Option<u8>,
517 is_charging: bool,
518 ) -> bool {
519 let Some(level) = battery_level else {
520 return true;
521 };
522 level >= self.relay_battery_floor(is_charging)
523 }
524
525 /// The battery level below which this device stops carrying other people's
526 /// traffic.
527 ///
528 /// [`RelayConfig::min_battery_for_relay`] is the floor in the ordinary
529 /// case, relaxed to the hard [`CRITICAL_RELAY_BATTERY_LEVEL`] for a device
530 /// that is either charging or configured [`RelayPriority::Always`] — the
531 /// two ways of saying "keep relaying anyway". The hard floor is never
532 /// crossed: it applies even to a configured minimum set below it, because
533 /// spending the last few percent on strangers' traffic leaves a device
534 /// unable to send its own.
535 pub(super) fn relay_battery_floor(&self, is_charging: bool) -> u8 {
536 let eager = matches!(self.config.relay.relay_priority, RelayPriority::Always);
537 if is_charging || eager {
538 CRITICAL_RELAY_BATTERY_LEVEL
539 } else {
540 self.config
541 .relay
542 .min_battery_for_relay
543 .max(CRITICAL_RELAY_BATTERY_LEVEL)
544 }
545 }
546
547 /// Transmits the forwards whose delay has elapsed.
548 ///
549 /// Called from the process tick. Each frame goes to a bounded set of
550 /// neighbors chosen by the governor, never back to the peer it came from or
551 /// to the peer that wrote it. A frame whose recipient is a neighbor of ours
552 /// is handed straight to them instead — the shortest path we can see.
553 pub(super) fn flush_mesh_relays(&mut self) {
554 self.flush_mesh_relays_at(Instant::now());
555 }
556
557 /// [`Self::flush_mesh_relays`] at a given instant, so a test can walk a
558 /// forward past the overdue cut-off without waiting it out.
559 pub(super) fn flush_mesh_relays_at(&mut self, now: Instant) {
560 let (due, abandoned) = self.mesh_relay.release_due(now);
561
562 // The drop point (`docs/spec/custody.md`): an ordinary forward here
563 // waited past the overdue cut-off without reaching a link, and its
564 // id is already released from the suppression cache; custody may
565 // take it on. A marked held forward goes back to the store, unjudged
566 // and uncounted.
567 for relay in abandoned {
568 self.judge_at_drop_point(relay);
569 }
570
571 if due.is_empty() {
572 return;
573 }
574
575 let neighbors = self.transport_manager.mesh_neighbors();
576
577 // `relay` is borrowed rather than destructured throughout: a frame that
578 // reaches no neighbor has to be handed back whole, and cloning the
579 // message to keep that option open would copy the payload — up to a
580 // whole media chunk — on every flush.
581 for relay in due {
582 let message = &relay.message;
583 let recipient = message.recipient.as_str().to_string();
584 let message_id = message.id.as_str();
585 let hop_count = message.hop_count.value();
586 let remaining_ttl = message.ttl.value();
587 let held = relay.custody_target.is_some();
588 let cause = if held {
589 "custody_redelivery"
590 } else {
591 "forward"
592 };
593
594 let targets = if let Some(target) = relay.custody_target.as_deref() {
595 // A held frame goes to the one neighbor it was queued toward,
596 // never through target selection: a fan-out here would defeat
597 // at most once per neighbor. Gone from range, it goes nowhere
598 // and back to the store below.
599 if neighbors.iter().any(|n| n.peer_id == target) {
600 vec![target.to_string()]
601 } else {
602 Vec::new()
603 }
604 } else {
605 let mut exclude: Vec<&str> = vec![message.sender.as_str()];
606 if let Some(peer) = relay.arrival_peer.as_deref() {
607 exclude.push(peer);
608 }
609 // Neighbors that handed us a copy while this waited hold the
610 // frame already (`MeshRelayGovernor::admit`).
611 exclude.extend(relay.heard_from.iter().map(String::as_str));
612 let onward = self.mesh_relay.select_targets(
613 neighbors
614 .iter()
615 .map(|n| (n.peer_id.as_str(), n.link_quality())),
616 &exclude,
617 &message_id,
618 );
619
620 // If the destination is one of our own neighbors, hand it over
621 // directly: no fan-out is worth more than arriving. Should that
622 // link fail between choosing it and writing to it, fall back to
623 // carrying it onward rather than dropping a frame we could still
624 // move.
625 if neighbors.iter().any(|n| n.peer_id == recipient) {
626 let mut ordered = vec![recipient.clone()];
627 ordered.extend(onward.into_iter().filter(|peer| peer != &recipient));
628 ordered
629 } else {
630 onward
631 }
632 };
633 let deliver_direct = targets.first() == Some(&recipient);
634
635 // Every neighbor we could hand it to has handed it to us. This is
636 // the one way a duplicate ends a forward: the frame is not lost,
637 // it is held all around us, and requeueing it would only spend
638 // the next ticks finding the same answer.
639 if targets.is_empty()
640 && !held
641 && !relay.heard_from.is_empty()
642 && neighbors
643 .iter()
644 .any(|n| relay.heard_from.iter().any(|peer| peer == &n.peer_id))
645 {
646 debug!(
647 message_id = %message.id,
648 "Every onward neighbor already holds this frame"
649 );
650 self.mesh_relay.drop_covered(relay);
651 continue;
652 }
653
654 if targets.is_empty() {
655 // Nowhere to hand it right now: we hold no link, or the only
656 // ones we hold are the peers this frame must not go back to.
657 // Falls through to the hand-back below rather than being
658 // dropped — a neighbor may well appear within the few seconds
659 // the frame is still worth carrying.
660 debug!(
661 message_id = %message.id,
662 cause,
663 "No onward neighbor for this frame"
664 );
665 }
666
667 let mut delivered_to = 0usize;
668 let mut reached_recipient = false;
669 for target in &targets {
670 // Each link this frame crosses is one transmission against the
671 // device's ceiling. Running out mid-fan-out stops the fan-out
672 // rather than the frame: the neighbors already reached carry it
673 // on, and the sender's retry covers the rest.
674 if !self.mesh_relay.take_send_token() {
675 debug!(
676 message_id = %message.id,
677 "At the forwarding limit; stopping this fan-out here"
678 );
679 break;
680 }
681
682 match self.transport_manager.send_to_neighbor(target, message) {
683 Ok(transport) => {
684 delivered_to += 1;
685 // Handing it to the recipient themselves ends the
686 // journey; the remaining neighbors are only a fallback
687 // for that link failing.
688 if deliver_direct && target == &recipient {
689 reached_recipient = true;
690 debug!(
691 message_id = %message.id,
692 next_hop = %target,
693 transport = ?transport,
694 cause,
695 "Delivered to its recipient directly"
696 );
697 break;
698 }
699 debug!(
700 message_id = %message.id,
701 next_hop = %target,
702 transport = ?transport,
703 hop_count,
704 remaining_ttl,
705 cause,
706 "Forwarded frame to neighbor"
707 );
708 }
709 Err(err) => {
710 // A link that went away between selection and send.
711 // The other targets still carry the frame.
712 debug!(
713 message_id = %message.id,
714 next_hop = %target,
715 error = %err,
716 "Could not forward to neighbor"
717 );
718 }
719 }
720 }
721
722 // Nothing reached a neighbor. Hand it back instead of dropping it:
723 // its id is already recorded as handled here, so a drop would lose
724 // this copy *and* refuse both the copies arriving behind it and the
725 // sender's retransmissions, for the whole retention window.
726 //
727 // Why it got nowhere does not change that. The budget ran out (the
728 // frames released alongside this one spent the allowance between
729 // them), every link chosen for it went away between being picked
730 // and being written to, or there was no usable link to pick. In all
731 // three the frame has travelled nowhere and this queue is the only
732 // thing that still remembers it. It keeps its due time, so one that
733 // stays stuck is abandoned by the overdue cut-off rather than
734 // retried forever.
735 //
736 // A held frame refused room on the way back goes to the custody
737 // store instead: the store still holds it, nothing is lost, and
738 // its id was never recorded here.
739 if delivered_to == 0 {
740 if let Some(returned) = self.mesh_relay.requeue(relay) {
741 self.return_held_to_store(returned);
742 }
743 continue;
744 }
745
746 if held {
747 self.record_custody_transmission(&relay, reached_recipient);
748 }
749 self.mesh_relay.record_forwarded();
750 self.emit_event(Event::message_relayed(
751 message.id.as_str(),
752 message.sender.as_str().to_string(),
753 recipient,
754 hop_count,
755 remaining_ttl,
756 ));
757 }
758 }
759
760 /// [`Self::handle_incoming_file_chunk_via`] with no known arrival transport
761 /// (a deferred chunk falls back to the resend-driven ACK). Used only by
762 /// tests; production always routes through the `_via` form with the inbound
763 /// transport.
764 #[cfg(test)]
765 pub(super) fn handle_incoming_file_chunk(&mut self, message: &Message) -> ChunkOutcome {
766 self.handle_incoming_file_chunk_via(message, None)
767 }
768
769 /// Routes an inbound file-chunk message through the transfer manager,
770 /// recording `arrival_transport` on the pending entry if the chunk must be
771 /// deferred, so the drain can ACK it directly once the session confirms.
772 pub(super) fn handle_incoming_file_chunk_via(
773 &mut self,
774 message: &Message,
775 arrival_transport: Option<TransportType>,
776 ) -> ChunkOutcome {
777 let sender = message.sender.as_str().to_string();
778
779 // SEC-H1: encrypted chunks carry the chunk bytes, media metadata, and
780 // original content type inside the MLS ciphertext; legacy plaintext
781 // chunks carry them on the wire Message and are accepted only where
782 // policy allows. Rich extras (caption, reply, forward) only ever come
783 // from the sealed plaintext — never from the wire Message.
784 let (chunk, media_metadata, original_content_type, rich_extras, data_purpose) =
785 if let Some(ref binary) = message.binary_content {
786 if is_media_envelope(binary) {
787 let encrypted = match decode_media_envelope(binary) {
788 Ok(e) => e,
789 // A frame whose magic byte still reads as a media envelope
790 // but whose encoding does not decode — in-transit
791 // corruption past the magic, or injected garbage. Under
792 // crypto recovery this is the pre-decrypt sibling of a hard
793 // decrypt failure: the frame as it stands is dead, but the
794 // sender's resend carries a fresh encoding that would
795 // decode, so withhold the ACK (`Deferred`) instead of
796 // telling the sender "delivered" for a chunk we dropped.
797 // Do NOT enqueue: an unparseable frame can never become
798 // parseable, so a queued copy could never drain — the same
799 // reasoning as the spent-generation arm below.
800 Err(e) => {
801 warn!(
802 message_id = %message.id,
803 error = %e,
804 "Failed to decode encrypted media envelope"
805 );
806 if !self.config.encryption.crypto_recovery_enabled {
807 return ChunkOutcome::Handled;
808 }
809 // Advisory, not terminal — the chunk was never ACKed, so
810 // the transfer is *stalled* pending a resend rather than
811 // failed. The terminal media signal stays
812 // `FileReceiveFailed`.
813 if let Ok(state) = lock_shared_state(&self.shared_state) {
814 state.emit_event(Event::message_decryption_failed(
815 message.id.clone(),
816 sender.clone(),
817 crate::events::DecryptionFailureCode::InvalidPayload,
818 "encrypted media envelope failed to decode; its file transfer is stalled until the sender resends".to_string(),
819 ));
820 }
821 return ChunkOutcome::Deferred;
822 }
823 };
824 let plaintext = match self.decrypt_media_chunk(
825 &sender,
826 &encrypted,
827 message,
828 arrival_transport,
829 ) {
830 MediaChunkDecrypt::Plaintext(p) => p,
831 // Not delivered and recoverable by a resend (queued for
832 // delayed decryption, or an undecryptable frame the
833 // sender can re-send): defer the ACK so the sender
834 // retries and the resend re-enters processing.
835 MediaChunkDecrypt::Deferred => return ChunkOutcome::Deferred,
836 // Illegitimate frame (foreign session slot, or an MLS
837 // credential naming a different sender): answer with
838 // silence, exactly like the text `SecurityRejected`
839 // path. ACKing here would tell an injector that the
840 // target is online and processing their frames — the
841 // one thing the text path withholds.
842 MediaChunkDecrypt::SecurityRejected => return ChunkOutcome::Rejected,
843 // Terminal drop (permanent refusal, empty plaintext, MLS
844 // unavailable): ACK as before — the sender cannot
845 // recover it by retrying.
846 MediaChunkDecrypt::Dropped => return ChunkOutcome::Handled,
847 };
848 let inner = match MediaChunkPlaintext::decode(&plaintext) {
849 Ok(i) => i,
850 Err(e) => {
851 warn!(
852 message_id = %message.id,
853 error = %e,
854 "Failed to parse decrypted media chunk plaintext, dropping"
855 );
856 return ChunkOutcome::Handled;
857 }
858 };
859 let chunk = match FileChunk::from_bytes(&inner.chunk_bytes) {
860 Ok(c) => c,
861 Err(e) => {
862 warn!(
863 message_id = %message.id,
864 error = %e,
865 "Failed to deserialize decrypted file chunk, dropping"
866 );
867 return ChunkOutcome::Handled;
868 }
869 };
870 (
871 chunk,
872 inner.media_metadata,
873 inner.original_content_type,
874 inner.rich_extras,
875 inner.data_purpose,
876 )
877 } else {
878 if !self.accept_plaintext_content(&sender) {
879 warn!(
880 message_id = %message.id,
881 sender = %sender,
882 "Rejecting unencrypted media chunk (encryption policy)"
883 );
884 self.warn_plaintext_receive_rejected(
885 &sender,
886 "Inbound plaintext media chunk rejected by encryption policy",
887 );
888 return ChunkOutcome::Rejected;
889 }
890 let chunk = match FileChunk::from_bytes(binary) {
891 Ok(c) => c,
892 Err(e) => {
893 warn!(
894 message_id = %message.id,
895 error = %e,
896 "Failed to deserialize binary file chunk, dropping"
897 );
898 return ChunkOutcome::Handled;
899 }
900 };
901 let (meta, oct) = Self::wire_media_metadata(message);
902 (chunk, meta, oct, None, None)
903 }
904 } else {
905 if !self.accept_plaintext_content(&sender) {
906 warn!(
907 message_id = %message.id,
908 sender = %sender,
909 "Rejecting unencrypted media chunk (encryption policy)"
910 );
911 self.warn_plaintext_receive_rejected(
912 &sender,
913 "Inbound plaintext media chunk rejected by encryption policy",
914 );
915 return ChunkOutcome::Rejected;
916 }
917 let chunk = match FileChunk::from_json(&message.content) {
918 Ok(c) => c,
919 Err(e) => {
920 warn!(
921 message_id = %message.id,
922 error = %e,
923 "Failed to deserialize file chunk, dropping"
924 );
925 return ChunkOutcome::Handled;
926 }
927 };
928 let (meta, oct) = Self::wire_media_metadata(message);
929 (chunk, meta, oct, None, None)
930 };
931
932 let file_id = chunk.file_id.clone();
933 let file_name = chunk.file_name.clone();
934 let file_size = chunk.file_size;
935 let is_first_chunk = chunk.chunk_index == 0;
936
937 // A duplicate chunk 0 must not reclassify a transfer already under
938 // way. Duplicates are accepted by replacement, and the manager's
939 // consistency check covers only the fields on the chunk itself: the
940 // purpose rides the sealed envelope, outside them. Without this an
941 // authenticated peer sends chunk 0 twice, once marked and once not,
942 // and chooses afterwards which side of the invisibility rule the
943 // transfer lands on. Either direction is a defect the marking
944 // exists to prevent: a document snapshot surfaced to a person as a
945 // downloaded file, or an app's own download turned invisible and
946 // left with no terminal event ever.
947 //
948 // Scoped to the sender the assembly belongs to. A chunk under
949 // another peer's file_id is refused by the manager's own sender
950 // check, and refusing it here instead would let any peer tombstone
951 // a transfer it merely named.
952 if is_first_chunk
953 && self
954 .pending_media_metadata
955 .get(&file_id)
956 .is_some_and(|entry| entry.sender == sender && entry.data_purpose != data_purpose)
957 {
958 warn!(
959 file_id = %file_id,
960 peer = %sender,
961 "Refusing a chunk 0 that reclassifies a transfer already under way"
962 );
963 self.file_transfer_manager.refuse_transfer(&file_id);
964 self.pending_media_metadata.remove(&file_id);
965 return ChunkOutcome::Handled;
966 }
967
968 // Refuse a blob nobody asked for at the door, not at the end.
969 //
970 // The check also runs on completion, which is where the hash is
971 // verified, but by then the whole transfer has been buffered,
972 // reassembled and checksummed. That is the storage and the battery
973 // this rule exists to protect: an authenticated peer could otherwise
974 // open as many transfers as the resource caps allow, hold them for
975 // the stale timeout, and pay one frame per transfer to do it.
976 //
977 // Chunk 0 carries the purpose, so this is where the question can
978 // first be asked. Answered with an ACK rather than silence: the
979 // sender is wrong rather than unlucky, and a retry would be refused
980 // exactly the same way.
981 if is_first_chunk
982 && !self.admits_data_media_transfer(&sender, data_purpose.as_ref(), file_size)
983 {
984 // Refused, not merely skipped. Chunks need not arrive in order,
985 // so any that landed before this one already opened an assembly,
986 // and any behind it would open another. The refusal has to
987 // outlive the chunk that prompted it, which is what the
988 // tombstone is for.
989 self.file_transfer_manager.refuse_transfer(&file_id);
990 self.pending_media_metadata.remove(&file_id);
991 return ChunkOutcome::Handled;
992 }
993
994 // Metadata is recorded only for chunks the manager accepts — a
995 // rejected chunk must leave no state behind (SEC-H2).
996 match self.file_transfer_manager.process_chunk(&sender, chunk) {
997 Ok(progress) => {
998 if is_first_chunk {
999 self.pending_media_metadata.insert(
1000 file_id.clone(),
1001 PendingMediaMetadataEntry {
1002 content_type: original_content_type.unwrap_or(ContentType::File),
1003 media_metadata,
1004 last_updated_at: Instant::now(),
1005 sender: sender.clone(),
1006 rich_extras,
1007 timestamp_ms: message.timestamp.as_millis(),
1008 app_id: message.app_id.as_str().to_string(),
1009 data_purpose,
1010 },
1011 );
1012 } else if let Some(entry) = self.pending_media_metadata.get_mut(&file_id) {
1013 entry.last_updated_at = Instant::now();
1014 }
1015 // What this transfer is, as far as this device can tell
1016 // right now. Read once and used twice below.
1017 let arriving = self
1018 .pending_media_metadata
1019 .get(&file_id)
1020 .and_then(|entry| entry.data_purpose.clone());
1021 // A chunk landing is the sign that an answer is coming, so
1022 // the question it answers is kept alive. The fetch record is
1023 // what admits the bytes at the end, and it expires on a
1024 // clock that a large blob over a slow radio outlives: without
1025 // this the record dies mid-carriage and the completed
1026 // transfer is then dropped as unsolicited, forever, because
1027 // every retry takes just as long.
1028 if let Some(purpose) = &arriving {
1029 self.note_data_media_progress(&sender, purpose);
1030 }
1031 // A data-purposed transfer is invisible to the application
1032 // for its whole life, not merely at the end. Reporting
1033 // progress on it would put a download nobody started in
1034 // front of a person, counting up to a file that never
1035 // appears.
1036 //
1037 // Positive knowledge, not absence of it: the purpose rides
1038 // chunk 0, so a chunk arriving before it leaves this device
1039 // unable to say what the transfer is. Reading "no entry yet"
1040 // as "not data-purposed" would emit exactly the phantom
1041 // progress this suppression exists to prevent, on any
1042 // transport that delivers chunk 1 first.
1043 //
1044 // The cost is that an ordinary transfer whose chunk 0 is
1045 // delayed loses the progress events until it lands. That is
1046 // cheap: progress is advisory and each event supersedes the
1047 // last, so the app sees the count resume rather than a gap.
1048 let reports_progress =
1049 arriving.is_none() && self.pending_media_metadata.contains_key(&file_id);
1050 if reports_progress {
1051 if let Ok(state) = lock_shared_state(&self.shared_state) {
1052 state.emit_event(Event::file_progress(
1053 file_id.clone(),
1054 progress.chunks_completed,
1055 progress.total_chunks,
1056 ));
1057 }
1058 }
1059 }
1060 Err(rejection) if rejection.is_resource_exhaustion() => {
1061 // A well-formed transfer was dropped by a receiver-side
1062 // resource limit. The chunk was already ACKed and will not
1063 // be retransmitted, so the transfer is unrecoverable —
1064 // surface that to the application instead of going silent.
1065 let dropped = self.pending_media_metadata.remove(&file_id);
1066 // Positive knowledge, the same rule the progress event
1067 // follows, reached by a different road. On chunk 0 the
1068 // purpose is in hand right here and the stored entry is not:
1069 // an entry is written only once the manager accepts a chunk,
1070 // which is exactly what did not happen. Off chunk 0 the
1071 // stored entry is the only knowledge there is.
1072 let identified = is_first_chunk || dropped.is_some();
1073 let purpose = if is_first_chunk { data_purpose } else { None }
1074 .or_else(|| dropped.and_then(|entry| entry.data_purpose));
1075 if let Some(purpose) = purpose {
1076 self.report_data_media_transfer_failure(&sender, &purpose, rejection.as_str());
1077 return ChunkOutcome::Handled;
1078 }
1079 if !identified {
1080 // A transfer whose chunk 0 has not landed cannot be named
1081 // as anything, so it must not be failed in front of a
1082 // person: it may be the document layer's. The app has
1083 // heard nothing about it either (progress is withheld by
1084 // the same rule), so there is no report to correct.
1085 debug!(
1086 file_id = %file_id,
1087 "Dropping a transfer whose purpose is not yet known; no failure reported"
1088 );
1089 return ChunkOutcome::Handled;
1090 }
1091 if let Ok(state) = lock_shared_state(&self.shared_state) {
1092 state.emit_event(Event::file_receive_failed(
1093 file_id,
1094 file_name,
1095 sender,
1096 rejection.as_str().to_string(),
1097 ));
1098 }
1099 return ChunkOutcome::Handled;
1100 }
1101 // Malformed or mismatched chunks are logged by the manager; the
1102 // assembly they targeted (if any) stays intact. Chunks of an
1103 // already-failed transfer land here too (`previously_failed`) —
1104 // its FileReceiveFailed event already fired exactly once.
1105 Err(_) => return ChunkOutcome::Handled,
1106 }
1107
1108 if self.file_transfer_manager.is_complete(&file_id) {
1109 let Some(file_data) = self.file_transfer_manager.finalize_file(&file_id) else {
1110 warn!(
1111 file_id = %file_id,
1112 "File transfer marked complete but reassembly or integrity checks failed"
1113 );
1114 let dropped = self.pending_media_metadata.remove(&file_id);
1115 // Completion implies chunk 0 was accepted, so this is
1116 // normally `true`. Checked anyway, and for the same reason as
1117 // the arm above: the alternative is naming a transfer this
1118 // device cannot identify as a file the app never saw begin.
1119 let identified = dropped.is_some();
1120 if let Some(purpose) = dropped.and_then(|entry| entry.data_purpose) {
1121 self.report_data_media_transfer_failure(
1122 &sender,
1123 &purpose,
1124 "integrity_check_failed",
1125 );
1126 return ChunkOutcome::Handled;
1127 }
1128 if !identified {
1129 debug!(
1130 file_id = %file_id,
1131 "Dropping a transfer whose purpose is not yet known; no failure reported"
1132 );
1133 return ChunkOutcome::Handled;
1134 }
1135 if let Ok(state) = lock_shared_state(&self.shared_state) {
1136 state.emit_event(Event::file_receive_failed(
1137 file_id,
1138 file_name,
1139 sender,
1140 "integrity_check_failed".to_string(),
1141 ));
1142 }
1143 return ChunkOutcome::Handled;
1144 };
1145 let metadata_entry = self.pending_media_metadata.remove(&file_id);
1146 // The one branch that makes a data-purposed transfer worth
1147 // having: the bytes go to the document layer and the person is
1148 // told nothing, because nothing happened that concerns them.
1149 if let Some(purpose) = metadata_entry
1150 .as_ref()
1151 .and_then(|entry| entry.data_purpose.clone())
1152 {
1153 self.route_data_media_payload(&sender, &purpose, file_data);
1154 return ChunkOutcome::Handled;
1155 }
1156 let (content_type, media_metadata, rich_extras, timestamp_ms, app_id) = metadata_entry
1157 .map(|entry| {
1158 (
1159 entry.content_type,
1160 entry.media_metadata,
1161 entry.rich_extras,
1162 Some(entry.timestamp_ms),
1163 Some(entry.app_id),
1164 )
1165 })
1166 .unwrap_or((ContentType::File, None, None, None, None));
1167 let rich_extras = rich_extras.unwrap_or_default();
1168
1169 if let Ok(state) = lock_shared_state(&self.shared_state) {
1170 state.emit_event(Event::file_received(
1171 file_id,
1172 file_name,
1173 file_size,
1174 sender,
1175 content_type,
1176 media_metadata,
1177 file_data,
1178 timestamp_ms,
1179 rich_extras.caption,
1180 rich_extras.reply_to_msg,
1181 rich_extras.reply_context.as_ref(),
1182 rich_extras.forward_info.as_ref(),
1183 app_id,
1184 ));
1185 }
1186 }
1187
1188 ChunkOutcome::Handled
1189 }
1190
1191 /// Whether a data-purposed transfer may be admitted at all.
1192 ///
1193 /// Only the attachment case is answerable by the fetch record: an
1194 /// attachment is bytes this device asked a named peer for, so the request
1195 /// either exists or it does not. A snapshot is unsolicited by design (it
1196 /// answers a version exchange rather than a fetch) and is bounded instead
1197 /// by the size check and the containment on the import path.
1198 ///
1199 /// The runtime kill switch is checked ahead of both, and what it saves is
1200 /// the buffering rather than the import. A device with the layer off was
1201 /// never going to write anything: `require_data_storage` fails closed
1202 /// with `DataDisabled` further down. But a transfer admitted at chunk 0
1203 /// is reassembled in full before it reaches that seam, so without this
1204 /// gate a device that opted out still holds a peer's whole blob in memory
1205 /// on its way to a refusal that was certain from the start.
1206 ///
1207 /// A conforming peer never sends one, because a build with the layer off
1208 /// never advertises the capability that permits it. This is the arm for
1209 /// the peer whose knowledge of us is stale or wrong.
1210 fn admits_data_media_transfer(
1211 &self,
1212 sender: &str,
1213 purpose: Option<&crate::media_envelope::DataPurpose>,
1214 declared_size: u64,
1215 ) -> bool {
1216 let Some(purpose) = purpose else {
1217 return true;
1218 };
1219 #[cfg(feature = "data")]
1220 {
1221 if !self.config.data.enabled {
1222 warn!(
1223 peer = %sender,
1224 "Refusing a document-layer transfer: the data layer is off"
1225 );
1226 return false;
1227 }
1228 match purpose {
1229 crate::media_envelope::DataPurpose::Attachment { hash } => {
1230 if !self.awaiting_attachment(sender, hash) {
1231 warn!(
1232 peer = %sender,
1233 "Refusing attachment bytes for a fetch this device never made"
1234 );
1235 return false;
1236 }
1237 }
1238 // A snapshot has no fetch record to judge it by, so the size
1239 // it declares is the only thing that can be judged here. It
1240 // is judged here rather than at the end because the ceiling
1241 // is certain: a document past it cannot be persisted by this
1242 // device even if every byte arrives, so admitting one spends
1243 // a whole reassembly to reach a refusal that was decided
1244 // before the first chunk. The completion check stays as the
1245 // backstop against a header that lied.
1246 crate::media_envelope::DataPurpose::Snapshot { doc } => {
1247 // The same refusal the import path makes, made before
1248 // the reassembly rather than after it, and for the same
1249 // reason the size check is here: the answer is already
1250 // certain. A document outside this space's interest is
1251 // refused whatever arrives, so admitting one spends a
1252 // whole transfer, up to the record ceiling, to reach a
1253 // decision that was made when the application narrowed
1254 // the space. Only a peer without entry 5 sends one,
1255 // which is exactly the peer whose traffic the declared
1256 // interest could not save.
1257 if !self.data_wants(sender, doc) {
1258 debug!(
1259 peer = %sender,
1260 doc,
1261 "Refusing a carried snapshot for a document outside this space's interest"
1262 );
1263 return false;
1264 }
1265 if declared_size > crate::protocol::data_sync::MAX_MEDIA_SNAPSHOT_BYTES as u64 {
1266 warn!(
1267 peer = %sender,
1268 declared = declared_size,
1269 "Refusing a carried snapshot larger than a record can hold"
1270 );
1271 return false;
1272 }
1273 }
1274 }
1275 true
1276 }
1277 #[cfg(not(feature = "data"))]
1278 {
1279 let _ = (sender, purpose, declared_size);
1280 // Nothing here can use these bytes, so nothing here should buffer
1281 // them either.
1282 false
1283 }
1284 }
1285
1286 /// Hand a completed data-purposed transfer to the document layer.
1287 ///
1288 /// Split behind a feature gate because the layer is optional and the
1289 /// bytes are not: a build compiled without it still has to recognise
1290 /// such a transfer, since recognising it is what keeps a document
1291 /// snapshot from being handed to a person as a downloaded file. It has
1292 /// nowhere to put the bytes, so it drops them. A conforming peer never
1293 /// sends one, because a build without the layer never advertises the
1294 /// capability that permits it.
1295 pub(super) fn route_data_media_payload(
1296 &mut self,
1297 sender: &str,
1298 purpose: &crate::media_envelope::DataPurpose,
1299 bytes: Vec<u8>,
1300 ) {
1301 #[cfg(feature = "data")]
1302 {
1303 self.accept_data_media_payload(sender, purpose, bytes);
1304 }
1305 #[cfg(not(feature = "data"))]
1306 {
1307 let _ = bytes;
1308 warn!(
1309 peer = %sender,
1310 ?purpose,
1311 "Dropping a document-layer transfer: this build has no data layer"
1312 );
1313 }
1314 }
1315
1316 /// Note that a data-purposed transfer is still arriving, so whatever
1317 /// this device is waiting on does not expire underneath it. See
1318 /// [`Self::route_data_media_payload`] for the gate.
1319 pub(super) fn note_data_media_progress(
1320 &mut self,
1321 sender: &str,
1322 purpose: &crate::media_envelope::DataPurpose,
1323 ) {
1324 #[cfg(feature = "data")]
1325 {
1326 self.refresh_attachment_fetch(sender, purpose);
1327 }
1328 #[cfg(not(feature = "data"))]
1329 {
1330 // A build with no data layer never admitted this transfer, so
1331 // there is nothing here that could be waiting on it.
1332 let _ = (sender, purpose);
1333 }
1334 }
1335
1336 /// Report a failed data-purposed transfer, where there is anything to
1337 /// report it to. See [`Self::route_data_media_payload`] for the gate.
1338 pub(super) fn report_data_media_transfer_failure(
1339 &mut self,
1340 sender: &str,
1341 purpose: &crate::media_envelope::DataPurpose,
1342 reason: &str,
1343 ) {
1344 #[cfg(feature = "data")]
1345 {
1346 self.report_data_media_failure(sender, purpose, reason);
1347 }
1348 #[cfg(not(feature = "data"))]
1349 {
1350 debug!(
1351 peer = %sender,
1352 ?purpose,
1353 reason,
1354 "A document-layer transfer failed; this build has no data layer"
1355 );
1356 }
1357 }
1358
1359 /// Extracts the chunk-0 metadata a legacy (unencrypted) chunk carries on
1360 /// the wire `Message`.
1361 fn wire_media_metadata(message: &Message) -> (Option<MediaMetadata>, Option<ContentType>) {
1362 use crate::constants::ORIGINAL_CONTENT_TYPE_KEY;
1363 let original_ct = message
1364 .metadata
1365 .get(ORIGINAL_CONTENT_TYPE_KEY)
1366 .map(|s| ContentType::parse(s));
1367 (message.media_metadata.clone(), original_ct)
1368 }
1369
1370 /// Policy gate for inbound plaintext content — text messages and legacy
1371 /// (unencrypted) media chunks alike: rejected when this node requires
1372 /// encryption, and rejected once the sender is known to run MLS — an
1373 /// encryption-capable peer sending plaintext is a downgrade/forgery attempt
1374 /// (plaintext carries no sender authentication, so anyone could inject it
1375 /// under a contact's name).
1376 ///
1377 /// # Capability, not confirmation
1378 ///
1379 /// This asks whether the peer *can* encrypt, not whether a session is
1380 /// confirmed. The distinction is the whole gate: a sender only ever emits
1381 /// plaintext when its own `should_auto_encrypt()` is false, which precludes
1382 /// it having established a session with us — while the session is merely
1383 /// pending it *queues* (`prepare_outbound_content`) rather than downgrading.
1384 /// So no honest peer sends cleartext while we know it speaks MLS, and
1385 /// asking only about *confirmed* sessions left a hole: an attacker with
1386 /// app-container write access deletes the `session_states` record, restore
1387 /// re-bootstraps it as `Pending` (which it must), the peer silently drops
1388 /// out of `confirmed_sessions`, and this gate re-opens. See
1389 /// [`OfflineProtocol::encryption_capable_peers`] for why that signal now
1390 /// comes from the credential store instead.
1391 ///
1392 /// The capability check is also *cheaper* than the confirmation one it
1393 /// precedes: it is an in-memory set lookup, where `is_session_confirmed`
1394 /// falls through to a protocol-state read plus a sealed-record open for any
1395 /// sender not already in `confirmed_sessions` — a path driven by the
1396 /// attacker-controlled `message.sender`. Ordering the cheap check first
1397 /// makes a flood of forged sender ids cost less than it used to.
1398 ///
1399 /// The group path has gated on existence rather than confirmation since it
1400 /// was written (`has_mls_group_state`); this brings the 1:1 path into line.
1401 fn accept_plaintext_content(&mut self, sender: &str) -> bool {
1402 if self.config.encryption.require_encryption {
1403 return false;
1404 }
1405 // `should_auto_encrypt()` guards both checks: a node with encryption
1406 // disabled or MLS uninitialized has no standing to call anything a
1407 // downgrade, and removing this guard would reject legitimate plaintext
1408 // on every plaintext-only deployment.
1409 if !self.should_auto_encrypt() {
1410 return true;
1411 }
1412 if self.is_encryption_capable(sender) {
1413 return false;
1414 }
1415 // Fail closed: if the confirmation lookup errors (storage failure),
1416 // treat the session as confirmed and reject — accepting plaintext on
1417 // error would let a storage fault disable the downgrade gate.
1418 if self.is_session_confirmed(sender).unwrap_or(true) {
1419 return false;
1420 }
1421 true
1422 }
1423
1424 /// Decrypts an encrypted media chunk envelope. Four failure dispositions,
1425 /// mirroring the text path in [`Self::handle_encrypted_message`]:
1426 ///
1427 /// - **Session not ready**: the whole message is queued for delayed
1428 /// decryption and [`MediaChunkDecrypt::Deferred`] is returned.
1429 /// - **Recoverable** (epoch desync, or a crypto/transport failure while
1430 /// `crypto_recovery_enabled`): [`MediaChunkDecrypt::Deferred`] *without*
1431 /// queueing — the ciphertext is dead either way, so recovery is the
1432 /// sender's resend, driven by the withheld ACK.
1433 /// - **Security rejection** (the envelope names another pair's session
1434 /// slot, or the MLS credential authenticates a different sender than the
1435 /// wire envelope claims): [`MediaChunkDecrypt::SecurityRejected`], which
1436 /// the caller answers with silence. Deliberately *not* gated on
1437 /// `crypto_recovery_enabled` — this is about what the receiver reveals,
1438 /// not about recovery, and the text path's equivalent is unconditional.
1439 /// - **Terminal** (a permanent refusal, an empty plaintext, or any crypto
1440 /// failure with recovery switched off): decryption telemetry plus
1441 /// [`MediaChunkDecrypt::Dropped`], which the caller still ACKs.
1442 ///
1443 /// On `Deferred` or `SecurityRejected` the caller must skip the ACK and
1444 /// unmark the id — for the former so the sender keeps retrying, for the
1445 /// latter so a replay re-enters this gate rather than hitting the duplicate
1446 /// re-ACK path. A successful decrypt doubles as a session confirmation
1447 /// signal, exactly like text decrypts.
1448 fn decrypt_media_chunk(
1449 &mut self,
1450 sender: &str,
1451 encrypted: &EncryptedMessage,
1452 message: &Message,
1453 arrival_transport: Option<TransportType>,
1454 ) -> MediaChunkDecrypt {
1455 let group_id = encrypted.group_id.as_str().to_string();
1456
1457 // Media envelopes are only ever produced for the sender's 1:1 session,
1458 // whose MLS group id is deterministic. Enforce that binding before
1459 // decrypting: MLS authenticates group membership, not the wire sender
1460 // claim, so without this check any peer holding a valid session with
1461 // us could deliver its own ciphertext under an arbitrary
1462 // `message.sender` and have the file attributed to that identity.
1463 let Ok(expected_group) = GroupId::for_session(&self.local_id, sender) else {
1464 warn!(
1465 sender = %sender,
1466 "Cannot derive 1:1 session id for media chunk sender, dropping"
1467 );
1468 return MediaChunkDecrypt::Dropped;
1469 };
1470 if encrypted.group_id != expected_group {
1471 error!(
1472 sender = %sender,
1473 group_id = %group_id,
1474 expected = %expected_group,
1475 "SECURITY: encrypted media chunk MLS group does not match the claimed sender's session, rejecting"
1476 );
1477 self.emit_security_warning(
1478 sender,
1479 SecurityWarningCode::MediaSenderGroupMismatch,
1480 "Encrypted media chunk MLS group does not match the claimed sender",
1481 );
1482 // Not ACKed: the text path answers the identical condition
1483 // (`MlsError::SessionIdentityMismatch` →
1484 // `InternalMessageResult::SecurityRejected`) with silence, and an
1485 // ACK here would leak exactly what that silence protects — an
1486 // injector who gets an ACK for a media chunk but nothing for the
1487 // same text frame learns the target is live either way.
1488 return MediaChunkDecrypt::SecurityRejected;
1489 }
1490
1491 let Some(mls) = self.mls_manager.clone() else {
1492 warn!(
1493 sender = %sender,
1494 "Encrypted media chunk received but MLS is not initialized, dropping"
1495 );
1496 self.emit_mls_decryption_failed(
1497 sender,
1498 Some(&group_id),
1499 DecryptionFailureKind::NotInitialized,
1500 MlsOperationContext::Receive,
1501 );
1502 return MediaChunkDecrypt::Dropped;
1503 };
1504
1505 let decrypt_result = {
1506 let manager = match mls.read() {
1507 Ok(m) => m,
1508 Err(_) => {
1509 error!("MLS lock poisoned while decrypting media chunk");
1510 return MediaChunkDecrypt::Dropped;
1511 }
1512 };
1513 manager.decrypt(encrypted, sender)
1514 };
1515
1516 match decrypt_result {
1517 Ok(Some(plaintext)) => {
1518 // A group-aware decrypt is the same confirmation signal the
1519 // text path uses; skip when already cache-confirmed to avoid
1520 // per-chunk storage I/O.
1521 if !self.confirmed_sessions.contains(sender) {
1522 self.confirm_session_from_successful_decrypt(sender, &group_id);
1523 }
1524 MediaChunkDecrypt::Plaintext(plaintext)
1525 }
1526 Ok(None) => {
1527 warn!(sender = %sender, "Media chunk decryption returned empty, dropping");
1528 MediaChunkDecrypt::Dropped
1529 }
1530 // Intercepted BEFORE classification, mirroring the text path in
1531 // `handle_encrypted_message`. Both variants classify as
1532 // `SessionStateError::Unknown`, whose disposition must stay
1533 // terminal-drop-and-ACK for the refusals that genuinely belong
1534 // there (`CommitNotAuthorized`), so the split has to happen here
1535 // rather than in the classifier.
1536 Err(ref e) if is_media_security_rejection(e) => {
1537 error!(
1538 sender = %sender,
1539 error = %e,
1540 "SECURITY: encrypted media chunk failed its sender/session identity check, rejecting"
1541 );
1542 MediaChunkDecrypt::SecurityRejected
1543 }
1544 Err(e) => {
1545 let classification = SessionStateError::from(&e);
1546 match classification {
1547 SessionStateError::SessionNotReady | SessionStateError::GroupNotFound => {
1548 info!(
1549 sender = %sender,
1550 error_code = classification.code(),
1551 "Encrypted media chunk received before session ready, queuing"
1552 );
1553 self.emit_mls_session_missing(
1554 Some(sender),
1555 Some(&group_id),
1556 MlsOperationContext::SessionLookup,
1557 MlsErrorCategory::SessionStateMissing,
1558 );
1559 self.enqueue_pending_decryption_via(sender, message, arrival_transport);
1560 MediaChunkDecrypt::Deferred
1561 }
1562 SessionStateError::SessionDesync
1563 if self.config.encryption.crypto_recovery_enabled =>
1564 {
1565 // Epoch desync (see the text path in `handle_encrypted_message`):
1566 // heal the channel with a rate-limited re-key and return
1567 // Deferred so the chunk is not ACKed and its id is
1568 // unmarked. We do NOT enqueue — the chunk is sealed to the
1569 // dead epoch and can never decrypt after the re-key.
1570 //
1571 // Unlike a DM, an in-flight media chunk cannot be
1572 // re-sealed for recovery: Tier 2 reseal is a deliberate
1573 // no-op for media (the media outbox is not persisted and
1574 // chunks are re-encoded, not replayed). Withholding the
1575 // ACK here is what drives recovery: the sender's media
1576 // outbox keeps retrying and, once its retry/ACK-timeout
1577 // budget lapses, surfaces `MediaResendRequired`, and the
1578 // app re-supplies the bytes via `send_media_with` — which
1579 // re-encodes fresh chunks against the now-healed session.
1580 // So media recovers via descriptor-based resend, not reseal.
1581 info!(
1582 sender = %sender,
1583 error_code = classification.code(),
1584 "Encrypted media chunk failed to decrypt due to epoch desync, re-keying"
1585 );
1586 self.schedule_session_rekey(sender);
1587 MediaChunkDecrypt::Deferred
1588 }
1589 SessionStateError::TransportFailure | SessionStateError::CryptoFailure
1590 if self.config.encryption.crypto_recovery_enabled =>
1591 {
1592 // A genuine crypto/transport failure on a chunk (AEAD
1593 // failure, spent ratchet generation, malformed frame).
1594 // Mirrors the text path in `handle_encrypted_message`:
1595 // withhold the ACK and unmark the id rather than lying
1596 // "delivered" for a chunk we dropped. Do NOT enqueue —
1597 // the generation is spent, so a queued copy could never
1598 // drain — and do NOT re-key, which stays desync-only.
1599 //
1600 // Recovery differs from a DM: chunks have no Tier 2
1601 // re-seal (they are re-encoded, not replayed), so the
1602 // un-ACKed chunk drives the media outbox's retry/ACK
1603 // budget to lapse and surface `MediaResendRequired`,
1604 // and the app re-supplies the bytes via `send_media_with`
1605 // — re-encoded against a live generation. Same
1606 // descriptor-based recovery as the media desync arm.
1607 warn!(
1608 sender = %sender,
1609 error = %e,
1610 error_code = classification.code(),
1611 "Failed to decrypt media chunk; withholding ACK so the transfer can be resent"
1612 );
1613 let kind = DecryptionFailureKind::from_mls_error(&e);
1614 self.emit_mls_decryption_failed(
1615 sender,
1616 Some(&group_id),
1617 kind,
1618 MlsOperationContext::Receive,
1619 );
1620 // Advisory, not terminal — matching a pending-queue
1621 // eviction: the chunk was never ACKed, so the transfer
1622 // is *stalled* pending a resend, not failed. The
1623 // terminal media signal is still `FileReceiveFailed`.
1624 if let Ok(state) = lock_shared_state(&self.shared_state) {
1625 state.emit_event(Event::message_decryption_failed(
1626 message.id.clone(),
1627 sender.to_string(),
1628 Self::decryption_failure_code_from_kind(kind),
1629 format!(
1630 "encrypted media chunk failed to decrypt ({}); its file transfer is stalled until the sender resends",
1631 classification.code()
1632 ),
1633 ));
1634 }
1635 MediaChunkDecrypt::Deferred
1636 }
1637 _ => {
1638 warn!(
1639 sender = %sender,
1640 error = %e,
1641 error_code = classification.code(),
1642 "Failed to decrypt media chunk, dropping"
1643 );
1644 let kind = DecryptionFailureKind::from_mls_error(&e);
1645 self.emit_mls_decryption_failed(
1646 sender,
1647 Some(&group_id),
1648 kind,
1649 MlsOperationContext::Receive,
1650 );
1651 // Permanently undecryptable (or crypto recovery is
1652 // switched off): the chunk is ACKed and stays
1653 // dedup-marked, so this loss is terminal for the
1654 // transfer — surface it to the app, not just as MLS
1655 // telemetry.
1656 if let Ok(state) = lock_shared_state(&self.shared_state) {
1657 state.emit_event(Event::message_decryption_failed(
1658 message.id.clone(),
1659 sender.to_string(),
1660 Self::decryption_failure_code_from_kind(kind),
1661 format!(
1662 "encrypted media chunk failed to decrypt ({}); its file transfer cannot complete",
1663 classification.code()
1664 ),
1665 ));
1666 }
1667 MediaChunkDecrypt::Dropped
1668 }
1669 }
1670 }
1671 }
1672 }
1673}
1674
1675/// Four-way outcome of [`OfflineProtocol::decrypt_media_chunk`]: the plaintext,
1676/// a deferral (not delivered, but recoverable by the sender's resend — either
1677/// queued because the session is not ready, or dropped as undecryptable with
1678/// the ACK withheld), a security rejection (an illegitimate frame the caller
1679/// must answer with silence), or a terminal drop the caller ACKs.
1680enum MediaChunkDecrypt {
1681 Plaintext(Vec<u8>),
1682 Deferred,
1683 /// The frame is not a legitimate message from the claimed sender: its
1684 /// envelope named another pair's session slot, or the MLS credential
1685 /// authenticated a different member than the wire envelope claims. Mirrors
1686 /// the text path's [`InternalMessageResult::SecurityRejected`] — the caller
1687 /// must NOT ACK (an ACK confirms to an injector that the target is online
1688 /// and processing their frames, which is exactly what the text path refuses
1689 /// to reveal) and must unmark the id.
1690 ///
1691 /// [`InternalMessageResult::SecurityRejected`]: super::InternalMessageResult::SecurityRejected
1692 SecurityRejected,
1693 Dropped,
1694}
1695
1696/// Whether an MLS decrypt error is a *security* rejection rather than a session
1697/// state problem — the two classes for which the receiver must stay silent
1698/// instead of ACKing.
1699///
1700/// Used as the match guard that intercepts these variants *before*
1701/// [`SessionStateError`] classification, exactly as the text path does inline in
1702/// `handle_encrypted_message`. The intercept cannot be replaced by a
1703/// classification arm: both variants map to [`SessionStateError::Unknown`], and
1704/// `Unknown` must keep its terminal drop-and-ACK disposition for the classes
1705/// that genuinely belong there (notably `CommitNotAuthorized`, a permanent
1706/// refusal whose sender gains nothing from retrying).
1707///
1708/// Kept as a named predicate rather than an inline pattern so the classification
1709/// itself is testable: `SenderIdentityMismatch` is not reachable through any
1710/// SDK-built ciphertext (see
1711/// `test_media_security_rejection_classifies_both_identity_mismatches`).
1712pub(super) fn is_media_security_rejection(error: &offline_protocol_mls::MlsError) -> bool {
1713 matches!(
1714 error,
1715 offline_protocol_mls::MlsError::SenderIdentityMismatch { .. }
1716 | offline_protocol_mls::MlsError::SessionIdentityMismatch { .. }
1717 // The leaf-binding refusals join them for the same reason: an
1718 // `__MLS_ENC__` envelope naming a `group:` id reaches group decrypt
1719 // through this path, both classify as `Unknown`, and `Unknown` must
1720 // keep drop-and-ACK for `CommitNotAuthorized`.
1721 | offline_protocol_mls::MlsError::LeafAddressMismatch { .. }
1722 | offline_protocol_mls::MlsError::UnsupportedSender { .. }
1723 )
1724}