ping_core/conversation.rs
1//! Conversation state — wraps an OpenMLS `MlsGroup`.
2//!
3//! Each external conversation maps 1:1 to an MLS group whose leaves are devices. The DeviceGroup
4//! (one per user, devices only) is just a special-cased conversation with the same wrapper.
5//!
6//! Persistence: we snapshot the `MlsGroup` after every state-changing operation under
7//! `groups/{conversation_id}` and cache the result in-memory.
8
9use openmls::{
10 framing::{MlsMessageOut, ProcessedMessageContent},
11 group::{MlsGroup, MlsGroupCreateConfig, MlsGroupJoinConfig},
12 prelude::{
13 tls_codec::{Deserialize as TlsDeserialize, Serialize as TlsSerialize},
14 BasicCredential, Capabilities, Ciphersuite, CredentialWithKey, Extension, ExtensionType,
15 Extensions, MlsMessageBodyIn, MlsMessageIn, ProcessedMessage, ProtocolMessage,
16 ProtocolVersion, RequiredCapabilitiesExtension, UnknownExtension,
17 },
18};
19use openmls_basic_credential::SignatureKeyPair;
20use openmls_traits::OpenMlsProvider;
21use ping_mls_store::PersistentMlsProvider;
22use serde::{Deserialize, Serialize};
23use std::collections::BTreeMap;
24use std::sync::Arc;
25use ulid::Ulid;
26use zeroize::Zeroizing;
27
28use crate::{
29 clock::Hlc,
30 codec,
31 device::{DeviceId, GroupSnapshotEntry, GroupStateSnapshot, GROUP_SNAPSHOT_VERSION},
32 error::{Error, Result},
33 identity::UserId,
34 message::{IncomingMessage, MessageEnvelope, MessageKind},
35 storage::Storage,
36 sync::SyncCursor,
37};
38
39const DEFAULT_CIPHERSUITE: Ciphersuite = Ciphersuite::MLS_128_DHKEMX25519_AES128GCM_SHA256_Ed25519;
40
41/// Map an OpenMLS `WelcomeError` to a typed [`Error`]. The two variants that
42/// mean "the KeyPackage this Welcome was bound to (or its private init key) is
43/// no longer in our keystore" — i.e. the KeyPackage was already consumed
44/// (single-use reuse) or never persisted — become [`Error::KeyPackageNotFound`]
45/// so the host can top up KeyPackages and retry rather than treating it as an
46/// opaque MLS failure. Everything else stays a generic MLS error.
47fn map_welcome_error<S: std::fmt::Display>(e: openmls::group::WelcomeError<S>) -> Error {
48 use openmls::group::WelcomeError;
49 match e {
50 WelcomeError::NoMatchingKeyPackage | WelcomeError::PrivateInitKeyNotFound => {
51 Error::KeyPackageNotFound
52 }
53 other => Error::mls(other),
54 }
55}
56
57/// MLS GroupContext extension type carrying the human conversation `name`.
58///
59/// Stored in the group context so the name is part of shared MLS group state and
60/// therefore present on EVERY device that holds the group — including a device
61/// that joins via a Welcome. Without this the name was creator-local and a
62/// joiner/linked device saw `None`. A private-use `Unknown` extension type
63/// (not a GREASE `0x?A?A` value, not a registered type 0x0001–0x0005).
64///
65/// It is added WITHOUT a `RequiredCapabilities` extension, so openmls imposes no
66/// per-member capability check — existing KeyPackages keep working and no
67/// re-link is required. The value is UTF-8 bytes of the name.
68const GROUP_NAME_EXTENSION_TYPE: u16 = 0xFF00;
69
70/// Read the conversation `name` from a group's GroupContext extensions, if set.
71fn group_name_from_extensions(extensions: &Extensions) -> Option<String> {
72 extensions.iter().find_map(|ext| match ext {
73 Extension::Unknown(ext_type, data) if *ext_type == GROUP_NAME_EXTENSION_TYPE => {
74 String::from_utf8(data.0.clone())
75 .ok()
76 .filter(|s| !s.is_empty())
77 }
78 _ => None,
79 })
80}
81
82/// Build the GroupContext extensions carrying `name` (empty when `name` is
83/// `None`/blank), for `MlsGroupCreateConfig::with_group_context_extensions`.
84fn group_context_extensions_for_name(name: Option<&str>) -> Extensions {
85 match name {
86 Some(n) if !n.is_empty() => Extensions::single(Extension::Unknown(
87 GROUP_NAME_EXTENSION_TYPE,
88 UnknownExtension(n.as_bytes().to_vec()),
89 )),
90 _ => Extensions::empty(),
91 }
92}
93
94/// Leaf [`Capabilities`] advertising support for the group-name GroupContext
95/// extension ([`GROUP_NAME_EXTENSION_TYPE`]) ON TOP OF the MLS defaults.
96///
97/// Required so a later [`Conversation::set_name`] (rename / avatar-id change via
98/// `update_group_context_extensions`) passes openmls' GCE-proposal validation:
99/// that path demands every group-context extension be listed in a
100/// `RequiredCapabilities`, and that every member's leaf advertise those
101/// extension types. The CREATE path does not need this (genesis extensions skip
102/// the proposal validator), but a post-create UPDATE does. New KeyPackages and
103/// the creator's own leaf carry this; existing devices must RE-LINK once to pick
104/// it up (a pre-production-acceptable cost, aligned with the shared-identity
105/// re-link).
106pub(crate) fn ping_leaf_capabilities() -> Capabilities {
107 ping_leaf_capabilities_for(false)
108}
109
110/// Leaf capabilities, optionally also advertising the `LastResort` extension.
111///
112/// A KeyPackage marked last-resort (`mark_as_last_resort`) carries the
113/// `LastResort` extension on its leaf; openmls' KeyPackage validation rejects
114/// any leaf whose capabilities do not advertise every extension the leaf
115/// actually carries ("A key package extension is not supported in the leaf's
116/// capabilities"). So last-resort KeyPackages MUST list `LastResort` here, while
117/// ordinary KeyPackages must NOT (they don't carry it).
118pub(crate) fn ping_leaf_capabilities_for(last_resort: bool) -> Capabilities {
119 let extensions = if last_resort {
120 vec![
121 ExtensionType::Unknown(GROUP_NAME_EXTENSION_TYPE),
122 ExtensionType::LastResort,
123 ]
124 } else {
125 vec![ExtensionType::Unknown(GROUP_NAME_EXTENSION_TYPE)]
126 };
127 Capabilities::new(None, None, Some(&extensions), None, None)
128}
129
130/// GroupContext extensions for a NAME UPDATE commit (`set_name`), as opposed to
131/// genesis ([`group_context_extensions_for_name`]). The GCE-proposal validator
132/// requires every group-context extension to appear in a `RequiredCapabilities`,
133/// so we attach one listing [`GROUP_NAME_EXTENSION_TYPE`] alongside the name
134/// extension. Clearing the name (`None`/blank) keeps the `RequiredCapabilities`
135/// (harmless: required caps may list a type that isn't currently present).
136fn group_context_extensions_for_name_update(name: Option<&str>) -> Extensions {
137 let required = Extension::RequiredCapabilities(RequiredCapabilitiesExtension::new(
138 &[ExtensionType::Unknown(GROUP_NAME_EXTENSION_TYPE)],
139 &[],
140 &[],
141 ));
142 match name {
143 Some(n) if !n.is_empty() => Extensions::from_vec(vec![
144 Extension::Unknown(
145 GROUP_NAME_EXTENSION_TYPE,
146 UnknownExtension(n.as_bytes().to_vec()),
147 ),
148 required,
149 ])
150 // Two distinct extension types — `from_vec` only rejects duplicates.
151 .expect("name + required-capabilities are distinct extension types"),
152 _ => Extensions::single(required),
153 }
154}
155
156/// 16-byte conversation identifier (ULID encoded). Stable across epochs.
157#[derive(Debug, Clone, Copy, PartialEq, Eq, Hash, PartialOrd, Ord, Serialize, Deserialize)]
158pub struct ConversationId(#[serde(with = "serde_bytes_array16")] pub [u8; 16]);
159
160impl ConversationId {
161 pub fn new() -> Self {
162 Self(Ulid::new().to_bytes())
163 }
164 pub fn as_hex(&self) -> String {
165 hex::encode(self.0)
166 }
167}
168
169impl Default for ConversationId {
170 fn default() -> Self {
171 Self::new()
172 }
173}
174
175mod serde_bytes_array16 {
176 use serde::{Deserializer, Serializer};
177 pub fn serialize<S: Serializer>(b: &[u8; 16], s: S) -> Result<S::Ok, S::Error> {
178 serde_bytes::serialize(b.as_slice(), s)
179 }
180 pub fn deserialize<'de, D: Deserializer<'de>>(d: D) -> Result<[u8; 16], D::Error> {
181 let v: Vec<u8> = serde_bytes::deserialize(d)?;
182 v.try_into()
183 .map_err(|_| serde::de::Error::custom("expected 16 bytes"))
184 }
185}
186
187#[derive(Debug, Clone, Serialize, Deserialize)]
188pub struct ConversationMeta {
189 pub id: ConversationId,
190 pub name: Option<String>,
191 pub epoch: u64,
192 pub member_count: u32,
193 pub is_device_group: bool,
194 pub created_at_ms: u64,
195}
196
197/// One member leaf of a conversation's MLS group: the member's [`UserId`]
198/// (recovered from the leaf's `BasicCredential`) and its ratchet-tree leaf
199/// index. A user with multiple devices appears once **per device leaf** —
200/// callers that want a per-user roster should dedup by `user_id`.
201#[derive(Debug, Clone, Serialize, Deserialize)]
202pub struct MemberInfo {
203 pub user_id: UserId,
204 pub leaf_index: u32,
205}
206
207/// In-memory conversation handle. Holds the OpenMLS group plus our wire-level cursor.
208pub struct Conversation {
209 pub(crate) id: ConversationId,
210 pub(crate) meta: ConversationMeta,
211 pub(crate) group: MlsGroup,
212 pub(crate) crypto: Arc<PersistentMlsProvider>,
213 pub(crate) signing: Arc<SignatureKeyPair>,
214 pub(crate) own_device: DeviceId,
215 pub(crate) seq: u64,
216 pub(crate) hlc: Hlc,
217 pub(crate) cursor: SyncCursor,
218 pub(crate) storage: Arc<dyn Storage>,
219 /// Local device→leaf-index map for [CR-2] revocation.
220 ///
221 /// Populated when this device either (a) admits a peer via [`Self::add_members`] —
222 /// every entry in the `Vec<(DeviceId, KeyPackage)>` is recorded after the commit
223 /// merges — or (b) joins as the receiving device via [`Self::join`], at which point
224 /// we record our own leaf. Pruned when [`Self::remove_members`] is called.
225 ///
226 /// Not authoritative for *peers' devices we didn't admit*: those are visible in
227 /// `group.members()` but their device_ids are opaque to this client. `revoke_device`
228 /// is therefore best-effort across conversations we ourselves invited the device
229 /// into; see [`MessagingClient::revoke_device`] for the documented scope.
230 pub(crate) device_leaves: BTreeMap<DeviceId, u32>,
231}
232
233impl std::fmt::Debug for Conversation {
234 fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
235 f.debug_struct("Conversation")
236 .field("id", &self.id.as_hex())
237 .field("meta", &self.meta)
238 .finish()
239 }
240}
241
242impl Conversation {
243 pub fn id(&self) -> ConversationId {
244 self.id
245 }
246 pub fn meta(&self) -> &ConversationMeta {
247 &self.meta
248 }
249
250 /// Current member roster, recovered locally from the MLS group's leaf
251 /// credentials — no network and no out-of-band `ping.profile` message.
252 /// Each `BasicCredential` was built from the member's `UserId`
253 /// (`BasicCredential::new(own_user.0.clone())` in [`Self::create`] /
254 /// [`Self::join`]), so we round-trip it back here. Each entry is one
255 /// leaf; a multi-device user appears once per device leaf.
256 pub fn members(&self) -> Vec<MemberInfo> {
257 self.group
258 .members()
259 .filter_map(|m| {
260 let basic = BasicCredential::try_from(m.credential).ok()?;
261 Some(MemberInfo {
262 user_id: UserId(basic.identity().to_vec()),
263 leaf_index: m.index.u32(),
264 })
265 })
266 .collect()
267 }
268
269 pub fn epoch(&self) -> u64 {
270 self.group.epoch().as_u64()
271 }
272
273 /// True while the local device is still a member. Becomes false once we
274 /// process a Commit that removes our own leaf — OpenMLS transitions the group
275 /// to `MlsGroupState::Inactive`. The client uses this to tear the group down
276 /// on self-removal so a later re-invite Welcome is a clean first-join.
277 pub(crate) fn is_active(&self) -> bool {
278 self.group.is_active()
279 }
280
281 /// Delete this group's OpenMLS state from persistent storage. Called when the
282 /// local device has been removed from the group, so a subsequent re-invite
283 /// Welcome re-joins from CLEAN storage. OpenMLS keys group state by
284 /// `group_id`; a lingering removed group would (a) make the conversation look
285 /// "already joined" so the host drops the re-invite Welcome, and (b) collide
286 /// with the fresh `StagedWelcome::into_group` on re-join.
287 pub(crate) fn delete_group_state(&mut self) -> Result<()> {
288 self.group
289 .delete(self.crypto.storage())
290 .map_err(|e| Error::Storage(e.to_string()))
291 }
292
293 pub fn cursor(&self) -> &SyncCursor {
294 &self.cursor
295 }
296
297 /// Create a new conversation, with `self` as the only initial member.
298 // 8 args is a lot, but they're all needed for an internal constructor and a builder
299 // would be over-engineered for v0.1.
300 #[allow(clippy::too_many_arguments)]
301 pub(crate) fn create(
302 id: ConversationId,
303 name: Option<String>,
304 own_device: DeviceId,
305 own_user: &UserId,
306 crypto: Arc<PersistentMlsProvider>,
307 signing: Arc<SignatureKeyPair>,
308 storage: Arc<dyn Storage>,
309 now_ms: u64,
310 ) -> Result<Self> {
311 let credential = BasicCredential::new(own_user.0.clone());
312 let credential_with_key = CredentialWithKey {
313 credential: credential.into(),
314 signature_key: signing.public().into(),
315 };
316 // Carry the conversation name in the GroupContext so it travels in the
317 // Welcome to every joiner (see `GROUP_NAME_EXTENSION_TYPE`). No
318 // RequiredCapabilities is added, so this imposes no member capability
319 // check and stays compatible with existing KeyPackages.
320 let cfg = MlsGroupCreateConfig::builder()
321 .ciphersuite(DEFAULT_CIPHERSUITE)
322 .use_ratchet_tree_extension(true)
323 // Advertise the group-name extension capability on the creator's leaf
324 // so a later `set_name` (rename / avatar change) passes the GCE
325 // capability check. See `ping_leaf_capabilities`.
326 .capabilities(ping_leaf_capabilities())
327 .with_group_context_extensions(group_context_extensions_for_name(name.as_deref()))
328 .map_err(Error::mls)?
329 .build();
330 let group = MlsGroup::new_with_group_id(
331 crypto.as_ref(),
332 signing.as_ref(),
333 &cfg,
334 openmls::group::GroupId::from_slice(&id.0),
335 credential_with_key,
336 )
337 .map_err(Error::mls)?;
338
339 let meta = ConversationMeta {
340 id,
341 name,
342 epoch: 0,
343 member_count: 1,
344 is_device_group: false,
345 created_at_ms: now_ms,
346 };
347 // [CR-2] Group creator is always leaf 0; record so revoke_device can target it.
348 let mut device_leaves = BTreeMap::new();
349 device_leaves.insert(own_device.clone(), group.own_leaf_index().u32());
350 Ok(Self {
351 id,
352 meta,
353 group,
354 crypto,
355 signing,
356 own_device,
357 seq: 0,
358 hlc: Hlc::ZERO.tick(now_ms),
359 cursor: SyncCursor::default(),
360 storage,
361 device_leaves,
362 })
363 }
364
365 /// Join an existing conversation from a Welcome message.
366 ///
367 /// Maps the OpenMLS `WelcomeError` for a missing KeyPackage / init key to
368 /// the typed [`Error::KeyPackageNotFound`] (see `map_welcome_error`) so the
369 /// host can react (top up KeyPackages) instead of treating it as an opaque
370 /// MLS failure.
371 pub(crate) fn join(
372 welcome_bytes: &[u8],
373 own_device: DeviceId,
374 crypto: Arc<PersistentMlsProvider>,
375 signing: Arc<SignatureKeyPair>,
376 storage: Arc<dyn Storage>,
377 now_ms: u64,
378 ) -> Result<Self> {
379 let mls_in = MlsMessageIn::tls_deserialize_exact(welcome_bytes).map_err(Error::mls)?;
380 let welcome = match mls_in.extract() {
381 MlsMessageBodyIn::Welcome(w) => w,
382 _ => return Err(Error::Invalid("expected Welcome".into())),
383 };
384 let cfg = MlsGroupJoinConfig::builder()
385 .use_ratchet_tree_extension(true)
386 .build();
387 let staged =
388 openmls::group::StagedWelcome::new_from_welcome(crypto.as_ref(), &cfg, welcome, None)
389 .map_err(map_welcome_error)?;
390 let group = staged.into_group(crypto.as_ref()).map_err(Error::mls)?;
391
392 let id_bytes: [u8; 16] = group
393 .group_id()
394 .as_slice()
395 .try_into()
396 .map_err(|_| Error::Invalid("group id must be 16 bytes".into()))?;
397 let id = ConversationId(id_bytes);
398 // The creator stamped the name into the GroupContext, which is part of
399 // the shared group state the Welcome reconstructs — so a joining device
400 // recovers the real name here instead of starting with `None`.
401 let name = group_name_from_extensions(group.extensions());
402 let meta = ConversationMeta {
403 id,
404 name,
405 epoch: group.epoch().as_u64(),
406 member_count: group.members().count() as u32,
407 is_device_group: false,
408 created_at_ms: now_ms,
409 };
410
411 // Seed the cursor at the join epoch so subsequent fetches skip pre-join Commits
412 // (notably the Add commit that produced this Welcome — it lives in the conversation
413 // log at `epoch - 1`, which the joiner must not try to apply on top of its
414 // already-advanced group state).
415 let join_epoch = group.epoch().as_u64();
416 // [CR-2] Record our own (device_id → leaf_index) so the host can later revoke us
417 // via the standard `revoke_device` flow. `own_leaf_index()` is stable for the
418 // lifetime of this group membership.
419 let own_leaf = group.own_leaf_index().u32();
420 let mut device_leaves = BTreeMap::new();
421 device_leaves.insert(own_device.clone(), own_leaf);
422 Ok(Self {
423 id,
424 meta,
425 group,
426 crypto,
427 signing,
428 own_device,
429 seq: 0,
430 hlc: Hlc::ZERO.tick(now_ms),
431 cursor: SyncCursor {
432 epoch: join_epoch,
433 ..Default::default()
434 },
435 storage,
436 device_leaves,
437 })
438 }
439
440 /// The conversation name recovered from MLS GroupContext state (set at
441 /// creation, carried in the Welcome to every joiner). `None` if unnamed.
442 pub(crate) fn name_from_group_state(&self) -> Option<String> {
443 group_name_from_extensions(self.group.extensions())
444 }
445
446 /// [CR-4] Rehydrate a previously-persisted conversation on cold restart.
447 ///
448 /// Loads the OpenMLS group state via `MlsGroup::load` (which reads from the
449 /// provider's storage — populated by the SQLite-backed checkpoint on the
450 /// previous run). Pairs the loaded MLS state with the meta + cursor + device→leaf
451 /// map the host-side `Storage` trait kept for us. Returns `Ok(None)` if OpenMLS
452 /// finds no state for `id` — the host's `groups` namespace had a stale entry.
453 #[allow(clippy::too_many_arguments)]
454 pub(crate) fn load(
455 id: ConversationId,
456 meta: ConversationMeta,
457 cursor: SyncCursor,
458 device_leaves: BTreeMap<DeviceId, u32>,
459 own_device: DeviceId,
460 crypto: Arc<PersistentMlsProvider>,
461 signing: Arc<SignatureKeyPair>,
462 storage: Arc<dyn Storage>,
463 now_ms: u64,
464 ) -> Result<Option<Self>> {
465 use openmls::group::GroupId;
466 let group_id = GroupId::from_slice(&id.0);
467 let group = match MlsGroup::load(crypto.storage(), &group_id).map_err(Error::mls)? {
468 Some(g) => g,
469 None => return Ok(None),
470 };
471 // Restore the local outgoing-send counter from the persisted cursor. The cursor
472 // tracks the highest applied (epoch, sender, seq) for every device — including
473 // our own — so we can recover `self.seq` from `cursor.last_seq_per_device[own]`.
474 // Without this, the next `send_application()` re-uses an already-consumed seq
475 // and receivers silently dedupe (cursor.is_new returns false on their side).
476 let seq = cursor
477 .last_seq_per_device
478 .get(&own_device)
479 .copied()
480 .unwrap_or(0);
481 Ok(Some(Self {
482 id,
483 meta,
484 group,
485 crypto,
486 signing,
487 own_device,
488 seq,
489 hlc: Hlc::ZERO.tick(now_ms),
490 cursor,
491 storage,
492 device_leaves,
493 }))
494 }
495
496 /// Encrypt an application message and produce a wire envelope ready for transport.
497 ///
498 /// Uses the [CR-6] plaintext content_hash path: the envelope's `content_hash` is
499 /// `SHA-256(plaintext)`, not the MLS ciphertext. This is what makes rebase clean
500 /// and gives cross-binding hash parity.
501 pub fn send_application(&mut self, plaintext: &[u8], now_ms: u64) -> Result<MessageEnvelope> {
502 let out = self
503 .group
504 .create_message(self.crypto.as_ref(), self.signing.as_ref(), plaintext)
505 .map_err(Error::mls)?;
506
507 self.seq += 1;
508 self.hlc = self.hlc.tick(now_ms);
509 let bytes = out.tls_serialize_detached().map_err(Error::mls)?;
510 let env = MessageEnvelope::new_application(
511 self.id,
512 self.epoch(),
513 self.own_device.clone(),
514 self.seq,
515 self.hlc,
516 bytes,
517 plaintext,
518 );
519 // Advance the local cursor past our own send so a subsequent catch-up sync doesn't
520 // pull this envelope back to us (we've already applied it locally — re-processing
521 // would either fail or duplicate-deliver).
522 self.cursor.advance(
523 env.epoch,
524 self.own_device.clone(),
525 self.seq,
526 self.hlc,
527 now_ms,
528 );
529 Ok(env)
530 }
531
532 /// Add members by KeyPackage. Produces the Commit envelope to broadcast plus the Welcome
533 /// envelope(s) to deliver out-of-band to the newly-added devices.
534 ///
535 /// [CR-2] takes a `Vec<(DeviceId, KeyPackage)>` instead of a bare `Vec<KeyPackage>`. The
536 /// `DeviceId` for each entry is the *caller's* assertion of which device owns that
537 /// KeyPackage — hosts typically get it from the directory service alongside the
538 /// KeyPackage itself. The mapping is persisted per-conversation so [`MessagingClient::revoke_device`]
539 /// can later locate the leaf to remove without a fresh directory lookup. The SDK does
540 /// not cryptographically verify the device claim; that's a host policy concern
541 /// (typically: the directory authenticates the key_package_id → device_id mapping).
542 pub fn add_members(
543 &mut self,
544 entries: Vec<(DeviceId, Vec<u8>)>,
545 now_ms: u64,
546 ) -> Result<AddOutcome> {
547 // All-in-one (stage + immediate merge) for callers that commit and
548 // persist synchronously with NO networked rollback window — e.g. the
549 // device-group / device-linking paths. The networked group path in
550 // `client.rs` instead uses `stage_add_members` + `confirm_staged` /
551 // `abort_staged`, so a Commit the server REJECTS can be rolled back
552 // rather than leaving the local epoch ahead of the server (the desync
553 // that permanently bricks a group: every later Commit then 409s and
554 // peers can't decrypt our epoch).
555 let staged = self.stage_add_members(entries, now_ms)?;
556 self.confirm_staged(&staged, now_ms)?;
557 let StagedCommit {
558 commit, welcome, ..
559 } = staged;
560 let welcome =
561 welcome.ok_or_else(|| Error::Invalid("add_members produced no Welcome".into()))?;
562 Ok(AddOutcome { commit, welcome })
563 }
564
565 /// Stage an add-members Commit WITHOUT merging it — the group keeps a
566 /// *pending* commit and the local epoch is UNCHANGED. Returns the Commit +
567 /// Welcome envelopes to send. The caller MUST follow with exactly one of
568 /// [`Self::confirm_staged`] (server accepted the Commit → merge locally) or
569 /// [`Self::abort_staged`] (server rejected it → discard, epoch never moves).
570 ///
571 /// This is what makes the local epoch advance *only after* the server
572 /// accepts the Commit (send-then-merge), so a rejected/conflicting Commit
573 /// can't desync the group. Safe because the JS worker serializes top-level
574 /// requests — nothing else touches this group during the send round-trip
575 /// between stage and confirm/abort (worker.ts "SERIALIZE … dispatch").
576 pub(crate) fn stage_add_members(
577 &mut self,
578 entries: Vec<(DeviceId, Vec<u8>)>,
579 now_ms: u64,
580 ) -> Result<StagedCommit> {
581 let mut kps = Vec::with_capacity(entries.len());
582 // Track signature_key → device_id so we can resolve leaf indices post-commit.
583 let mut sig_to_device: Vec<(Vec<u8>, DeviceId)> = Vec::with_capacity(entries.len());
584 for (device_id, raw) in &entries {
585 let mls_in = MlsMessageIn::tls_deserialize_exact(raw).map_err(Error::mls)?;
586 let kp_in = match mls_in.extract() {
587 MlsMessageBodyIn::KeyPackage(kp) => kp,
588 _ => return Err(Error::Invalid("expected KeyPackage".into())),
589 };
590 // KeyPackages on the wire are unvalidated (`KeyPackageIn`); validate against the
591 // crypto provider before handing them to OpenMLS.
592 let kp = kp_in
593 .validate(self.crypto.crypto(), ProtocolVersion::default())
594 .map_err(Error::mls)?;
595 let sig_key = kp.leaf_node().signature_key().as_slice().to_vec();
596 sig_to_device.push((sig_key, device_id.clone()));
597 kps.push(kp);
598 }
599
600 // The Commit's wire `epoch` is the *source* epoch (where it was crafted). The
601 // Welcome's `epoch` is the *post-commit* epoch. An MLS Commit advances the epoch by
602 // exactly 1, so we can name the post-commit epoch as `pre + 1` WITHOUT merging.
603 let pre_commit_epoch = self.epoch();
604 let post_commit_epoch = pre_commit_epoch + 1;
605
606 let (commit_out, welcome_out, _gi) = self
607 .group
608 .add_members(self.crypto.as_ref(), self.signing.as_ref(), &kps)
609 .map_err(Error::mls)?;
610 // NB: NO merge here — the pending commit is merged by `confirm_staged`
611 // only once the server has accepted the Commit send.
612
613 let next_seq = self.seq + 1;
614 let next_hlc = self.hlc.tick(now_ms);
615
616 let commit_bytes = mls_message_out_bytes(commit_out)?;
617 let commit_env = MessageEnvelope::new(
618 self.id,
619 pre_commit_epoch,
620 MessageKind::Commit,
621 self.own_device.clone(),
622 next_seq,
623 next_hlc,
624 commit_bytes,
625 );
626
627 let welcome_bytes = mls_message_out_bytes(welcome_out)?;
628 let welcome_env = MessageEnvelope::new(
629 self.id,
630 post_commit_epoch,
631 MessageKind::Welcome,
632 self.own_device.clone(),
633 next_seq,
634 next_hlc,
635 welcome_bytes,
636 );
637
638 Ok(StagedCommit {
639 commit: commit_env,
640 welcome: Some(welcome_env),
641 next_seq,
642 next_hlc,
643 leaf_update: StagedLeafUpdate::Add(sig_to_device),
644 })
645 }
646
647 pub fn remove_members(
648 &mut self,
649 leaf_indexes: Vec<u32>,
650 now_ms: u64,
651 ) -> Result<MessageEnvelope> {
652 // All-in-one (stage + immediate merge). See `add_members` for why the
653 // networked path uses stage/confirm/abort instead.
654 let staged = self.stage_remove_members(leaf_indexes, now_ms)?;
655 self.confirm_staged(&staged, now_ms)?;
656 let StagedCommit { commit, .. } = staged;
657 Ok(commit)
658 }
659
660 /// Stage a remove-members Commit WITHOUT merging it — see
661 /// [`Self::stage_add_members`]. No Welcome (removals don't admit anyone).
662 pub(crate) fn stage_remove_members(
663 &mut self,
664 leaf_indexes: Vec<u32>,
665 now_ms: u64,
666 ) -> Result<StagedCommit> {
667 use openmls::prelude::LeafNodeIndex;
668 let leaves: Vec<LeafNodeIndex> = leaf_indexes
669 .iter()
670 .copied()
671 .map(LeafNodeIndex::new)
672 .collect();
673
674 let pre_commit_epoch = self.epoch();
675
676 let (commit_out, _welcome_opt, _gi) = self
677 .group
678 .remove_members(self.crypto.as_ref(), self.signing.as_ref(), &leaves)
679 .map_err(Error::mls)?;
680 // NB: NO merge here — see `stage_add_members`.
681
682 let next_seq = self.seq + 1;
683 let next_hlc = self.hlc.tick(now_ms);
684 let bytes = mls_message_out_bytes(commit_out)?;
685 let commit_env = MessageEnvelope::new(
686 self.id,
687 pre_commit_epoch,
688 MessageKind::Commit,
689 self.own_device.clone(),
690 next_seq,
691 next_hlc,
692 bytes,
693 );
694
695 let removed: std::collections::HashSet<u32> = leaf_indexes.iter().copied().collect();
696 Ok(StagedCommit {
697 commit: commit_env,
698 welcome: None,
699 next_seq,
700 next_hlc,
701 leaf_update: StagedLeafUpdate::Remove(removed),
702 })
703 }
704
705 /// Create a self-Remove PROPOSAL — "leave the group".
706 ///
707 /// MLS does not let a member commit their OWN removal, so leaving is a
708 /// two-step dance: this member broadcasts a Remove proposal for its own
709 /// leaf, and a DIFFERENT member commits it (see
710 /// [`Self::stage_commit_pending_proposals`]). A proposal does NOT advance
711 /// the epoch; we bump only our own seq/hlc and the caller then deletes the
712 /// conversation locally. Returns the proposal envelope
713 /// (`MessageKind::Proposal`) to broadcast on the conversation channel.
714 pub(crate) fn leave_group(&mut self, now_ms: u64) -> Result<MessageEnvelope> {
715 let proposal_out = self
716 .group
717 .leave_group(self.crypto.as_ref(), self.signing.as_ref())
718 .map_err(Error::mls)?;
719 let next_seq = self.seq + 1;
720 let next_hlc = self.hlc.tick(now_ms);
721 let bytes = mls_message_out_bytes(proposal_out)?;
722 let env = MessageEnvelope::new(
723 self.id,
724 self.epoch(),
725 MessageKind::Proposal,
726 self.own_device.clone(),
727 next_seq,
728 next_hlc,
729 bytes,
730 );
731 self.seq = next_seq;
732 self.hlc = next_hlc;
733 Ok(env)
734 }
735
736 /// True when OpenMLS has buffered pending proposals for this group (e.g. a
737 /// peer's leave proposal awaiting a Commit).
738 pub(crate) fn has_pending_proposals(&self) -> bool {
739 self.group.pending_proposals().next().is_some()
740 }
741
742 /// Stage a Commit covering all buffered pending proposals (the canonical
743 /// case: a peer's "leave" Remove proposal). Returns `None` when there is
744 /// nothing pending. Send-then-merge like add/remove so a server-rejected
745 /// Commit rolls back without desyncing the epoch. The removed leaves are
746 /// read off the pending Remove proposals so the device→leaf map is pruned
747 /// on confirm.
748 pub(crate) fn stage_commit_pending_proposals(
749 &mut self,
750 now_ms: u64,
751 ) -> Result<Option<StagedCommit>> {
752 use openmls::messages::proposals::Proposal;
753 let removed: std::collections::HashSet<u32> = self
754 .group
755 .pending_proposals()
756 .filter_map(|qp| match qp.proposal() {
757 Proposal::Remove(r) => Some(r.removed().u32()),
758 _ => None,
759 })
760 .collect();
761 if self.group.pending_proposals().next().is_none() {
762 return Ok(None);
763 }
764
765 let pre_commit_epoch = self.epoch();
766 let (commit_out, _welcome, _gi) = self
767 .group
768 .commit_to_pending_proposals(self.crypto.as_ref(), self.signing.as_ref())
769 .map_err(Error::mls)?;
770 // NB: NO merge here — confirm_staged merges once the server accepts.
771
772 let next_seq = self.seq + 1;
773 let next_hlc = self.hlc.tick(now_ms);
774 let bytes = mls_message_out_bytes(commit_out)?;
775 let commit_env = MessageEnvelope::new(
776 self.id,
777 pre_commit_epoch,
778 MessageKind::Commit,
779 self.own_device.clone(),
780 next_seq,
781 next_hlc,
782 bytes,
783 );
784
785 Ok(Some(StagedCommit {
786 commit: commit_env,
787 welcome: None,
788 next_seq,
789 next_hlc,
790 leaf_update: StagedLeafUpdate::Remove(removed),
791 }))
792 }
793
794 /// Change the conversation `name` carried in the GroupContext (RFC 9420
795 /// GroupContextExtensions commit). Unlike a hydration broadcast, this rides
796 /// MLS group STATE, so every member — and every future joiner via the
797 /// ratchet-tree/GroupInfo — converges on the new name. Hosts use this to make
798 /// a rename or an embedded avatar-media-id change BULLETPROOF (the name field
799 /// carries the `ping:meta:v1:` blob).
800 ///
801 /// Produces a Commit to broadcast; NO Welcome (membership is unchanged). All
802 /// members must advertise the group-name extension capability
803 /// ([`ping_leaf_capabilities`]) — i.e. have re-linked since that shipped —
804 /// else openmls rejects the commit (`RequiredExtensionNotSupportedByAllMembers`).
805 ///
806 /// All-in-one (stage + immediate merge). The networked group path uses
807 /// [`Self::stage_set_name`] + [`Self::confirm_staged`]/[`Self::abort_staged`]
808 /// so a server-rejected commit can roll back without desyncing the epoch.
809 pub fn set_name(&mut self, name: Option<String>, now_ms: u64) -> Result<MessageEnvelope> {
810 let staged = self.stage_set_name(name, now_ms)?;
811 self.confirm_staged(&staged, now_ms)?;
812 let StagedCommit { commit, .. } = staged;
813 Ok(commit)
814 }
815
816 /// Stage a name-update Commit WITHOUT merging it — see
817 /// [`Self::stage_add_members`]. No Welcome (no membership change). The local
818 /// epoch advances only on [`Self::confirm_staged`].
819 pub(crate) fn stage_set_name(
820 &mut self,
821 name: Option<String>,
822 now_ms: u64,
823 ) -> Result<StagedCommit> {
824 let pre_commit_epoch = self.epoch();
825 let extensions = group_context_extensions_for_name_update(name.as_deref());
826
827 let (commit_out, _welcome_opt, _gi) = self
828 .group
829 .update_group_context_extensions(
830 self.crypto.as_ref(),
831 extensions,
832 self.signing.as_ref(),
833 )
834 .map_err(Error::mls)?;
835 // NB: NO merge here — confirm_staged merges once the server accepts.
836
837 let next_seq = self.seq + 1;
838 let next_hlc = self.hlc.tick(now_ms);
839 let bytes = mls_message_out_bytes(commit_out)?;
840 let commit_env = MessageEnvelope::new(
841 self.id,
842 pre_commit_epoch,
843 MessageKind::Commit,
844 self.own_device.clone(),
845 next_seq,
846 next_hlc,
847 bytes,
848 );
849
850 Ok(StagedCommit {
851 commit: commit_env,
852 welcome: None,
853 next_seq,
854 next_hlc,
855 leaf_update: StagedLeafUpdate::None,
856 })
857 }
858
859 /// Merge a previously [staged](Self::stage_add_members) Commit into the local
860 /// group — call ONLY after the server has accepted the Commit send. Advances
861 /// the epoch, updates the roster + device→leaf map, bumps seq/hlc, and moves
862 /// the sync cursor past our own Commit so catch-up doesn't re-apply it.
863 pub(crate) fn confirm_staged(&mut self, staged: &StagedCommit, now_ms: u64) -> Result<()> {
864 self.group
865 .merge_pending_commit(self.crypto.as_ref())
866 .map_err(Error::mls)?;
867 self.meta.epoch = self.epoch();
868 self.meta.member_count = self.group.members().count() as u32;
869 // A GroupContextExtensions commit (e.g. `set_name`) changes the name
870 // carried in group state — refresh the cached meta name. Harmless for
871 // add/remove commits (the name is unchanged).
872 self.meta.name = self.name_from_group_state();
873
874 match &staged.leaf_update {
875 StagedLeafUpdate::Add(sig_to_device) => {
876 // [CR-2] Resolve leaf indexes for the devices we just added (match by the
877 // per-device MLS signature_key, unique per device).
878 for member in self.group.members() {
879 if let Some((_, device_id)) = sig_to_device
880 .iter()
881 .find(|(sig, _)| sig.as_slice() == member.signature_key.as_slice())
882 {
883 self.device_leaves
884 .insert(device_id.clone(), member.index.u32());
885 }
886 }
887 }
888 StagedLeafUpdate::Remove(removed) => {
889 // [CR-2] Prune the device→leaf map for removed leaves. Other entries' leaf
890 // indexes are stable (OpenMLS reuses blank slots, doesn't reshuffle).
891 self.device_leaves.retain(|_, idx| !removed.contains(idx));
892 }
893 StagedLeafUpdate::None => {
894 // No membership change (name-update commit) — leaf map unchanged.
895 }
896 }
897
898 self.seq = staged.next_seq;
899 self.hlc = staged.next_hlc;
900 self.cursor.advance(
901 self.meta.epoch,
902 self.own_device.clone(),
903 self.seq,
904 self.hlc,
905 now_ms,
906 );
907 Ok(())
908 }
909
910 /// Discard a previously [staged](Self::stage_add_members) Commit — call when
911 /// the server REJECTED the Commit send. Clears the pending commit so the
912 /// local epoch stays exactly where it was (no desync) and the conversation
913 /// is operational again. Idempotent / safe if there is no pending commit.
914 pub(crate) fn abort_staged(&mut self) -> Result<()> {
915 self.group
916 .clear_pending_commit(self.crypto.storage())
917 .map_err(Error::mls)?;
918 Ok(())
919 }
920
921 /// Process an inbound envelope. Returns Some(IncomingMessage) for application traffic.
922 pub fn process(
923 &mut self,
924 env: &MessageEnvelope,
925 now_ms: u64,
926 ) -> Result<Option<IncomingMessage>> {
927 if !self.cursor.is_new(env.epoch, &env.sender_device, env.seq) {
928 return Ok(None); // dedupe: already applied
929 }
930 let mls_in = MlsMessageIn::tls_deserialize_exact(&env.payload).map_err(Error::mls)?;
931
932 // OpenMLS' `process_message` expects an `impl Into<ProtocolMessage>`. `MlsMessageIn`
933 // itself doesn't implement that; we have to extract the body and convert the inner
934 // private/public message. Welcomes are handled at the client level, not here.
935 let protocol_msg: ProtocolMessage = match mls_in.extract() {
936 MlsMessageBodyIn::PrivateMessage(m) => m.into(),
937 MlsMessageBodyIn::PublicMessage(m) => m.into(),
938 MlsMessageBodyIn::Welcome(_) => {
939 return Err(Error::Invalid(
940 "Welcome must be handled at client level, not in-group".into(),
941 ));
942 }
943 _ => return Err(Error::Invalid("unsupported MLS message body".into())),
944 };
945
946 let processed: ProcessedMessage = self
947 .group
948 .process_message(self.crypto.as_ref(), protocol_msg)
949 .map_err(Error::mls)?;
950
951 // Recover the sender's account-level `UserId` from their authenticated
952 // leaf credential BEFORE `into_content()` consumes `processed`. Same
953 // round-trip as `members()`: the leaf was built as
954 // `BasicCredential::new(user.0)`, so `identity()` is the `UserId` bytes.
955 // This lets the host attribute messages to the right account across
956 // every linked device without any device→account side channel.
957 let sender_user_id = BasicCredential::try_from(processed.credential().clone())
958 .map(|c| UserId(c.identity().to_vec()))
959 .unwrap_or_else(|_| UserId(Vec::new()));
960
961 let out = match processed.into_content() {
962 ProcessedMessageContent::ApplicationMessage(app) => {
963 let pt = app.into_bytes();
964 // CR-6: for v=2 application envelopes the wire-contract validator can't
965 // check `content_hash` (the hash is over plaintext, which it didn't have).
966 // We can now: verify SHA-256(pt) == env.content_hash and reject mismatches.
967 // For v=1 envelopes the wire-contract validator already checked the
968 // ciphertext-based hash, so no extra work here.
969 if env.v >= 2 {
970 let computed = crate::message::hash_application_plaintext(&pt);
971 if computed != env.content_hash {
972 return Err(Error::Invalid(
973 "v=2 application content_hash mismatch".into(),
974 ));
975 }
976 }
977 Some(IncomingMessage {
978 conversation_id: self.id,
979 sender_device: env.sender_device.clone(),
980 sender_user_id,
981 epoch: env.epoch,
982 hlc: env.hlc,
983 plaintext: pt,
984 content_hash: env.content_hash,
985 })
986 }
987 ProcessedMessageContent::StagedCommitMessage(staged) => {
988 self.group
989 .merge_staged_commit(self.crypto.as_ref(), *staged)
990 .map_err(Error::mls)?;
991 self.meta.epoch = self.epoch();
992 self.meta.member_count = self.group.members().count() as u32;
993 // A remote `set_name` (GroupContextExtensions) commit changes the
994 // name in group state — refresh the cached meta so this device
995 // picks up the rename / avatar-id change WITHOUT a side broadcast.
996 self.meta.name = self.name_from_group_state();
997 None
998 }
999 ProcessedMessageContent::ProposalMessage(qp) => {
1000 // OpenMLS does NOT auto-buffer a processed proposal — the caller
1001 // must explicitly store it, or `pending_proposals()` stays empty
1002 // and `commit_to_pending_proposals` covers nothing. This is the
1003 // mechanism behind "leave": a peer's self-Remove proposal is
1004 // stored here so a remaining member can commit it (evicting the
1005 // leaver) via `commit_pending_proposals`.
1006 self.group
1007 .store_pending_proposal(self.crypto.storage(), *qp)
1008 .map_err(Error::mls)?;
1009 None
1010 }
1011 ProcessedMessageContent::ExternalJoinProposalMessage(_) => {
1012 // External-join proposals are not part of any current flow; drop.
1013 None
1014 }
1015 };
1016
1017 self.cursor.advance(
1018 env.epoch,
1019 env.sender_device.clone(),
1020 env.seq,
1021 env.hlc,
1022 now_ms,
1023 );
1024 Ok(out)
1025 }
1026
1027 /// Export a derived secret keyed to this group's current epoch ([CR-8]).
1028 ///
1029 /// Wraps `MlsGroup::export_secret` (the MLS exporter, RFC 9420 §8.5) and surfaces the
1030 /// bytes in a `Zeroizing<Vec<u8>>` so the local copy is wiped on drop. Used by the host
1031 /// to seed:
1032 /// * the ephemeral channel (`ping/ephemeral`, §5.4 of the architecture)
1033 /// * call media keys (`ping/calls/media/{call_id}`, §7.2)
1034 /// * call-ephemeral framer keys (`ping/calls/ephemeral/{call_id}`, §7.5)
1035 ///
1036 /// `label` should use the documented `ping/*` namespacing convention. There is no
1037 /// runtime enforcement — cross-binding parity is enforced by conformance fixtures
1038 /// pinning specific label strings.
1039 ///
1040 /// Output is the secret. Callers MUST treat the buffer as a secret: never log, never
1041 /// persist unencrypted. The wrapper zeroes our local copy on drop; the caller is
1042 /// responsible for zeroing any copy they make.
1043 pub fn export_secret(
1044 &self,
1045 label: &str,
1046 context: &[u8],
1047 length: usize,
1048 ) -> Result<Zeroizing<Vec<u8>>> {
1049 if length == 0 {
1050 return Err(Error::Invalid("export_secret length must be > 0".into()));
1051 }
1052 // Soft cap to prevent runaway allocations from a malformed caller. Real labels never
1053 // need more than ~64 bytes (AES-256 key + 96-bit nonce + slack); 1 KiB is generous.
1054 if length > 1024 {
1055 return Err(Error::Invalid(
1056 "export_secret length exceeds 1024-byte cap".into(),
1057 ));
1058 }
1059 let bytes = self
1060 .group
1061 .export_secret(self.crypto.as_ref(), label, context, length)
1062 .map_err(Error::mls)?;
1063 Ok(Zeroizing::new(bytes))
1064 }
1065
1066 /// [CR-7] Export a portable snapshot of this group's MLS state.
1067 ///
1068 /// Walks the provider's working set, picks every entry whose key references this
1069 /// group's id, and bundles them with format metadata. Returns CBOR-encoded bytes
1070 /// suitable for inclusion in:
1071 /// * `LinkingTicket.catchup_snapshot.conversation_metas[i].group_state_bytes`
1072 /// (via [CR-13] — host calls this and passes the bytes through);
1073 /// * `IdentityBackup.device_group_snapshot` (the Permissive-recovery path per
1074 /// `docs/architecture/recovery.md`).
1075 ///
1076 /// Returns `Err` if the encoded snapshot exceeds [`GROUP_SNAPSHOT_HARD_CAP`].
1077 /// Output is wrapped in `Zeroizing` because the bytes contain past epoch secrets;
1078 /// the caller's copy on the FFI side is the host's responsibility to wipe.
1079 pub fn export_state_snapshot(&self, now_ms: u64) -> Result<Zeroizing<Vec<u8>>> {
1080 let entries = self.crypto.group_scoped_entries(&self.id.0);
1081 let snap = GroupStateSnapshot {
1082 v: GROUP_SNAPSHOT_VERSION,
1083 group_id: self.id,
1084 openmls_storage_version: openmls_traits::storage::CURRENT_VERSION,
1085 snapshot_created_at_ms: now_ms,
1086 entries: entries
1087 .into_iter()
1088 .map(|(key, value)| GroupSnapshotEntry { key, value })
1089 .collect(),
1090 };
1091 Ok(Zeroizing::new(snap.encode()?))
1092 }
1093
1094 /// Look up the leaf index this device controls, if known ([CR-2]).
1095 ///
1096 /// Returns the locally-tracked leaf for `device_id`. Only populated for devices we
1097 /// added via [`Self::add_members`] or for our own leaf via [`Self::create`] /
1098 /// [`Self::join`]. Devices a peer admitted on our behalf are not in this map.
1099 pub fn leaf_index_of(&self, device_id: &DeviceId) -> Option<u32> {
1100 self.device_leaves.get(device_id).copied()
1101 }
1102
1103 /// Synchronously capture everything [`ConversationSnapshot::flush`]
1104 /// needs to persist this conversation, so a caller can DROP the
1105 /// `conversations` lock BEFORE awaiting the async writes.
1106 ///
1107 /// Holding a `parking_lot` guard across `.await` is a latent bug: on
1108 /// the single-threaded wasm worker, a second client call that lands
1109 /// while the first is suspended (a waiting writer + a new reader)
1110 /// makes `parking_lot` try to PARK, and its wasm stub `panic!`s with
1111 /// "Parking not supported on this platform" — poisoning the module.
1112 /// Splitting the synchronous capture (under the lock) from the async
1113 /// flush (lock released) removes that hazard everywhere the snapshot
1114 /// runs for a conversation that lives inside the shared map. The
1115 /// capture is a consistent point-in-time view (cursor + meta + leaves
1116 /// + the Arc'd provider/storage handles).
1117 pub(crate) fn snapshot_inputs(&self) -> Result<ConversationSnapshot> {
1118 // [CR-2] Stable BTreeMap-of-pairs encoding → canonical CBOR so
1119 // every platform decodes identical bytes.
1120 let leaves_vec: Vec<(DeviceId, u32)> = self
1121 .device_leaves
1122 .iter()
1123 .map(|(d, i)| (d.clone(), *i))
1124 .collect();
1125 Ok(ConversationSnapshot {
1126 id: self.id,
1127 crypto: self.crypto.clone(),
1128 storage: self.storage.clone(),
1129 cursor: self.cursor.encode()?,
1130 meta: codec::encode(&self.meta)?,
1131 device_leaves: codec::encode(&leaves_vec)?,
1132 })
1133 }
1134
1135 /// Persist this conversation's state. Convenience wrapper used by
1136 /// call sites that hold an OWNED `Conversation` (not borrowed from the
1137 /// shared map) — e.g. just-created/just-joined conversations before
1138 /// they're inserted, where no lock is held across the await. Map-
1139 /// resident callers MUST instead use `snapshot_inputs()` + drop the
1140 /// guard + `flush().await` (see client.rs) to avoid the wasm parking
1141 /// panic described on `snapshot_inputs`.
1142 pub(crate) async fn snapshot_to_storage(&self) -> Result<()> {
1143 self.snapshot_inputs()?.flush().await
1144 }
1145}
1146
1147/// Point-in-time, lock-free snapshot of a [`Conversation`]'s persistable
1148/// state. Produced synchronously by [`Conversation::snapshot_inputs`] (so
1149/// the `conversations` lock can be dropped) and flushed asynchronously by
1150/// [`Self::flush`].
1151pub(crate) struct ConversationSnapshot {
1152 id: ConversationId,
1153 crypto: Arc<PersistentMlsProvider>,
1154 storage: Arc<dyn Storage>,
1155 cursor: Vec<u8>,
1156 meta: Vec<u8>,
1157 device_leaves: Vec<u8>,
1158}
1159
1160impl ConversationSnapshot {
1161 /// Flush the captured state to storage. Safe to `.await` with NO
1162 /// `conversations` lock held — it only touches the Arc'd provider +
1163 /// storage handles, never the shared map.
1164 pub(crate) async fn flush(self) -> Result<()> {
1165 // CRASH-CONSISTENCY ORDERING. The host `Storage` is a non-transactional
1166 // key-value store (iOS SQLCipher row, web IndexedDB slot, memory), so we
1167 // cannot make these four writes atomic without changing the host API on
1168 // every platform. Instead we order them so a crash between any two
1169 // leaves a SELF-HEALING state, never a permanent gap:
1170 //
1171 // 1. MLS working set (`checkpoint_async`) — the authoritative crypto
1172 // state (epoch, ratchet keys). Written FIRST so the persisted MLS
1173 // state is always >= what the cursor claims we've processed.
1174 // 2. meta (name / member_count) — cosmetic, re-derivable from group state.
1175 // 3. device→leaf map ([CR-2]) — needed by revoke_device after restart.
1176 // 4. cursor — the "processed up to here" gate, written LAST.
1177 //
1178 // Why cursor LAST is the key invariant: the cursor decides which events
1179 // we re-fetch on restart. If it committed BEFORE the MLS checkpoint, a
1180 // crash in between would leave the cursor ahead of the persisted MLS
1181 // state — we'd skip events the group never actually applied, a permanent
1182 // gap (the "stranded" condition). Writing MLS first and the cursor last
1183 // guarantees the cursor is never ahead of durable state: a crash just
1184 // means we re-fetch a few already-applied events, which `process` dedups
1185 // via `SyncCursor::is_new`. meta/device_leaves lagging the cursor is the
1186 // only residual skew and both are re-derivable from MLS group state.
1187 //
1188 // [CR-4] checkpoint MUST happen on every state-changing op so a cold
1189 // restart (iOS NSE, web SW) finds the latest epoch. `checkpoint_async`
1190 // is required for the WASM `IndexedDb` backend (IDB is async-only);
1191 // native Memory / Sqlite await trivially (their I/O is sync internally).
1192 self.crypto
1193 .checkpoint_async()
1194 .await
1195 .map_err(|e| Error::Storage(format!("checkpoint: {e}")))?;
1196
1197 let hex = self.id.as_hex();
1198 self.storage
1199 .put("groups", &format!("{hex}/meta"), self.meta)
1200 .await?;
1201 self.storage
1202 .put("device_leaves", &hex, self.device_leaves)
1203 .await?;
1204 // Cursor written LAST — see the ordering rationale above.
1205 self.storage.put("cursors", &hex, self.cursor).await?;
1206 Ok(())
1207 }
1208}
1209
1210/// Both halves of an Add commit. The Commit goes on the conversation channel; the Welcome is
1211/// delivered to the new members via whatever out-of-band path the host uses (often the same
1212/// transport, addressed to the new device's mailbox).
1213#[derive(Debug, Clone)]
1214pub struct AddOutcome {
1215 pub commit: MessageEnvelope,
1216 pub welcome: MessageEnvelope,
1217}
1218
1219/// The local-state mutation a staged Commit will apply on
1220/// [`Conversation::confirm_staged`]. Captured at stage time so confirm can run
1221/// after the (async) Commit send without re-deriving anything.
1222pub(crate) enum StagedLeafUpdate {
1223 /// Add: signature_key → device_id for each added device, resolved to a leaf
1224 /// index against the merged tree in `confirm_staged`.
1225 Add(Vec<(Vec<u8>, DeviceId)>),
1226 /// Remove: the leaf indexes being dropped from the device→leaf map.
1227 Remove(std::collections::HashSet<u32>),
1228 /// No membership change (e.g. a GroupContextExtensions / name-update commit).
1229 /// The device→leaf map is untouched.
1230 None,
1231}
1232
1233/// A Commit produced but NOT yet merged (see [`Conversation::stage_add_members`]).
1234/// Held by the client across the Commit send; merged via
1235/// [`Conversation::confirm_staged`] on success or discarded via
1236/// [`Conversation::abort_staged`] on a server rejection. This is the unit of the
1237/// send-then-merge protocol that keeps the local epoch from ever running ahead of
1238/// the server.
1239pub(crate) struct StagedCommit {
1240 pub commit: MessageEnvelope,
1241 pub welcome: Option<MessageEnvelope>,
1242 next_seq: u64,
1243 next_hlc: Hlc,
1244 leaf_update: StagedLeafUpdate,
1245}
1246
1247fn mls_message_out_bytes(m: MlsMessageOut) -> Result<Vec<u8>> {
1248 m.tls_serialize_detached().map_err(Error::mls)
1249}