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