message saved if client crashes on retreival

This commit is contained in:
yves committed 2026-11-20 22:00:54 +01:00
1 parent 6d3ea3e2a6
commit 006bedc728
11 files changed
+591 -72

No files matched your search

+38 -7
View File
@@ -160,13 +160,42 @@ sautées sont bornées (`MAX_SKIP = 1000`). Le chiffrement des en-têtes est pr
### 3.5 Authentification des pulls
```
MailboxAuth { mailbox, timestamp, signature }
MailboxAuth { mailbox, timestamp, signature, acks[] }
signature = Ed25519_sign(mailbox_sk,
"meow/v2/pull" ‖ server_sign_pk ‖ mailbox ‖ timestamp ‖ request_id)
"meow/v2/pull" ‖ server_sign_pk ‖ mailbox ‖ timestamp
‖ SHA-256(acks, préfixés par leur longueur) ‖ request_id)
```
Le serveur rejette si `|now - timestamp| > 300 s`, si la signature est invalide ou si
`request_id` a déjà été vu (cache anti-rejeu avec TTL de 600 s).
`request_id` a déjà été vu (cache anti-rejeu avec TTL de 600 s). Les `acks` étant signés, seul
le propriétaire d'une mailbox peut en supprimer les messages.
### 3.6 Livraison au moins une fois
Un pull **ne supprime rien**. Chaque `StoredMessage` porte un `id` attribué par le serveur, et
reste stocké jusqu'à ce qu'un pull ultérieur l'acquitte dans `MailboxAuth.acks`.
```
client serveur
Pull{auths(acks=[])} ─────▶ fetch (sans supprimer)
◀───── PullResult{messages[id], more}
traitement, puis enregistrement de
l'état ET des acks en attente
Pull{auths(acks=[ids]), wait=0} ─────▶ ack(ids), puis fetch
◀───── PullResult{…}
```
* Côté client, les acks en attente (`PendingAck {server, mailbox, id}`) sont enregistrés
**dans la même sauvegarde** que l'état du ratchet. Un message redélivré dont l'id est encore
en attente n'est donc pas retraité : il est simplement acquitté de nouveau.
* `sync(wait)` relance un tour sans attente tant qu'il reçoit du nouveau ou que le serveur
signale `more` (limites de 256 messages et 16 Mio par pull). Un `sync` se termine donc avec
ses acks envoyés.
* Une mailbox qui disparaît alors qu'elle a encore des acks en attente (boîte de réponse
temporaire d'une invitation, contact supprimé) est « mise en retraite » : sa clé est conservée
le temps d'envoyer ces acks.
* Effacer les acks envoyés ne demande pas de sauvegarde, car un ack est idempotent : au pire,
il est renvoyé.
---
@@ -183,16 +212,18 @@ Les opérations (`oneof` de `Request`) :
| Requête | Effet |
|---|---|
| `Post { deposits[] }` | Stocke chaque payload dans `mailbox`, avec un TTL. Retourne les ids serveur. |
| `Pull { auths[], wait_seconds }` | Vide les mailbox authentifiées. Long-poll si elles sont vides (≤ 120 s). |
| `Pull { auths[], wait_seconds }` | Supprime d'abord les messages acquittés (`auths[].acks`), puis renvoie sans les supprimer les messages restants des mailbox authentifiées (`more` si limité). Long-poll si elles sont vides (≤ 120 s). Voir §3.6. |
| `InvitationPublish { payload, ttl, shortcode_len, password }` | Retourne `shortcode` et `expiry`. |
| `InvitationFetch { shortcode, password }` | 3 échecs → invitation détruite. Comparaison en temps constant. |
| `Ping` | Retourne l'heure du serveur (synchronisation des horodatages). |
Stockage (`ServerStore`, trait asynchrone) :
* `MemoryStore` : `HashMap` + `tokio::sync::Notify`, pour le dev et les tests.
* `RedisStore` : `LIST mb:{hex}` (avec `EXPIRE`), `PUBLISH mbch:{hex}`, `SET inv:{code} EX`,
`SET invpw:{code}`, `INCR invfa:{code}`, `SET replay:{id} NX EX`.
* `MemoryStore` : `HashMap` + canal `broadcast`, pour le dev et les tests.
* `RedisStore` (Redis ≥ 5) : `meow:v2:mb:{hex}` est un **stream** (`XADD MAXLEN`, `XRANGE` pour
lire, `XDEL` pour acquitter ; l'id du stream sert d'id de message) avec `EXPIRE`,
`PUBLISH meow:v2:new`, `SET inv:{code} EX`, `SET invpw:{code}`, `INCR invfa:{code}`,
`SET rq:{id} NX EX`.
---
Generated
+1
View File
@@ -1092,6 +1092,7 @@ dependencies = [
name = "meow-client"
version = "0.2.0"
dependencies = [
"async-trait",
"futures-util",
"meow-core",
"meow-crypto",
+4 -3
View File
@@ -14,7 +14,7 @@ Pour reprendre le travail : lire ce fichier, prendre le premier lot ⬜ dont les
### Ordre conseillé pour la suite
1. **3.8** livraison au moins une fois (correction de fond, petit lot).
1. ~~**3.8** livraison au moins une fois~~ ✅
2. **4.1 → 4.5** `meow-store` (prérequis du mobile).
3. **5.5, 5.6, 5.9** file d'envoi, accusés, flux d'événements (prérequis de l'UI).
4. **7.1 → 7.3** FFI et intégration dans bastet.
@@ -24,6 +24,7 @@ Pour reprendre le travail : lire ce fichier, prendre le premier lot ⬜ dont les
| Date | Lots | Notes |
|---|---|---|
| 2026-11-20 | 3.8 | Pull non destructif + acks signés dans `MailboxAuth`, `PendingAck` enregistrés avec l'état, redélivrances dédupliquées, pagination (`more`), mise en retraite des mailbox disparues. Redis : liste → stream. Profil client au format 2 (les `.mwid2` de format 1 sont refusés). 31 tests, e2e mémoire et Redis verts, et plus aucun message résiduel dans Redis à la fin de l'e2e. |
| 2026-11-20 | 0.x, 1.1–1.4, 2.1–2.4, 3.1–3.4, 4.0, 5.1–5.4, 6.1–6.5 | Squelette complet et fonctionnel. 29 tests Rust + `tests/e2e.sh` (≈ 60 appels CLI, 2 serveurs, mémoire et Redis, bascule quand un serveur tombe). Décision : protocole v2 sans compatibilité v1. |
---
@@ -68,7 +69,7 @@ Pour reprendre le travail : lire ce fichier, prendre le premier lot ⬜ dont les
| 3.5 | Limites : taille de corps, nombre de dépôts par requête, quota par mailbox, rate limiting par IP | Tests | 🟡 taille de corps et nombre de dépôts faits |
| 3.6 | Packaging : Dockerfile multi-stage, unité systemd, deb/rpm (`cargo-deb`, `cargo-generate-rpm`) | Image < 30 Mo | ⬜ |
| 3.7 | Statistiques agrégées `/v2/stats` (sans métadonnées par utilisateur) | | ⬜ |
| 3.8 | **Livraison au moins une fois** : aujourd'hui `Pull` *vide* la mailbox (comme en v1), donc un message est perdu si le client tombe entre la réponse et la sauvegarde. Il faut un pull qui renvoie des ids, puis un `Ack { ids }` au pull suivant pour supprimer, avec déduplication côté client par id de message | Test : coupure réseau simulée sans perte | ⬜ **prioritaire** |
| 3.8 | **Livraison au moins une fois** : aujourd'hui `Pull` *vide* la mailbox (comme en v1), donc un message est perdu si le client tombe entre la réponse et la sauvegarde. Il faut un pull qui renvoie des ids, puis un `Ack { ids }` au pull suivant pour supprimer, avec déduplication côté client par id de message | Test : coupure réseau simulée sans perte | ✅ (`no_message_lost_when_the_client_dies`) |
| 3.9 | Chiffrer l'offre publiée sur le serveur avec une clé placée dans le fragment de l'URL (`…/inv/<code>#<clé>`) : le serveur ne verrait plus le nom ni le message de l'initiateur | Test | ⬜ |
## Phase 4 — `meow-store` (persistance client)
@@ -92,7 +93,7 @@ Pour reprendre le travail : lire ce fichier, prendre le premier lot ⬜ dont les
| 5.4 | Réception : un pull par serveur regroupant toutes les mailbox, long-poll, traitement, événements | e2e | ✅ |
| 5.5 | File d'envoi persistante avec retries et backoff, envoi hors ligne | Tests | ⬜ |
| 5.6 | Accusés de réception et de lecture chiffrés (dans le ratchet) | e2e | ⬜ |
| 5.7 | Fichiers : découpage en morceaux, reprise, déduplication, limite de taille | e2e (`send-file`) | 🟡 envoi en un seul morceau |
| 5.7 | Fichiers : découpage en morceaux, reprise, déduplication, limite de taille. Le contenu reçu doit être écrit sur disque **avant** l'ack : aujourd'hui, la CLI écrit les fichiers après `sync`, alors que les acks sont déjà partis | e2e (`send-file`) | 🟡 envoi en un seul morceau |
| 5.8 | Trafic leurre : mailbox fictives dans les pulls, ordre aléatoire, padding des requêtes | Tests | ⬜ |
| 5.9 | Stream d'événements (`tokio::broadcast`) et tâche de synchronisation en arrière-plan | | ⬜ |
+1
View File
@@ -18,6 +18,7 @@ thiserror = "2"
tracing = "0.1"
[dev-dependencies]
async-trait = "0.1"
meow-server = { path = "../meow-server" }
tokio = { version = "1", features = ["rt-multi-thread", "macros", "net"] }
+154 -21
View File
@@ -28,7 +28,8 @@ use meow_proto::v2::{self, request::Body as Req, response::Body as Resp};
use meow_proto::Message;
pub use meow_core::contact::ContactState as State;
pub use profile::{Content, HistoryEntry};
use profile::RetiredMailbox;
pub use profile::{Content, HistoryEntry, PendingAck};
/// Maximum file size sent in a single message (chunking: roadmap 5.7).
pub const MAX_FILE_BYTES: usize = 4 * 1024 * 1024;
@@ -109,6 +110,17 @@ pub struct SendReport {
}
/// One identity, with its servers, contacts and history.
/// Acknowledged ids sent per mailbox and pull (server limit: 1024).
const MAX_ACKS_PER_MAILBOX: usize = 512;
#[derive(Default)]
struct PullRound {
/// Messages to process.
new: Vec<v2::StoredMessage>,
/// A server held messages back.
more: bool,
}
pub struct MeowClient {
path: PathBuf,
password: String,
@@ -268,8 +280,10 @@ impl MeowClient {
pub async fn remove_contact(&mut self, name: &str) -> Result<()> {
let i = self.contact_index(name)?;
let before = self.contact_mailboxes();
let c = self.profile.contacts.remove(i);
self.profile.history.retain(|h| h.contact != c.id);
self.retire_mailboxes(before);
self.save().await
}
@@ -521,45 +535,134 @@ impl MeowClient {
/// Pull every mailbox on every server (one request per server), process what
/// arrived and return the resulting events. With `wait > 0` the servers hold the
/// request until something arrives or the delay expires (long-poll).
///
/// Delivery is at-least-once: servers keep messages until they are acknowledged,
/// and a message is acknowledged (in the next pull) only once its effects are saved.
/// After a round that brought messages, `sync` pulls again without waiting, which
/// sends the acknowledgements and fetches what the servers held back.
pub async fn sync(&mut self, wait: Duration) -> Result<Vec<Event>> {
// server (by signing key) → (server, mailbox keys)
let mut jobs: BTreeMap<SignPublic, (ServerInfo, Vec<SigningKeyPair>)> = BTreeMap::new();
for c in &self.profile.contacts {
for (kp, servers) in c.mailboxes() {
for s in servers {
jobs.entry(s.sign_key)
.or_insert_with(|| (s.clone(), Vec::new()))
.1
.push(kp.clone());
const MAX_ROUNDS: usize = 16;
let mut events = Vec::new();
let mut wait = wait;
for round in 0..MAX_ROUNDS {
let r = match self.pull_round(wait).await {
Ok(r) => r,
// A later round only flushes acks and backlog: keep what we already got.
Err(e) if round > 0 => {
tracing::warn!("follow-up pull failed: {e}");
break;
}
Err(e) => return Err(e),
};
let got_new = !r.new.is_empty();
if got_new {
events.extend(self.process(r.new).await?);
}
// Pull again (without waiting) only to acknowledge what just arrived and to
// fetch what servers held back. Redeliveries alone do not loop.
if !got_new && !r.more {
break;
}
wait = Duration::ZERO;
}
Ok(events)
}
/// One pull per server, carrying the pending acks. New (not yet processed) messages
/// are returned and recorded as pending acks; they are saved by [`Self::process`].
async fn pull_round(&mut self, wait: Duration) -> Result<PullRound> {
// server (by signing key) → (server, mailbox keys)
let pending = &self.profile.pending_acks;
self.profile.retired_mailboxes.retain(|r| {
pending
.iter()
.any(|a| a.mailbox.as_slice() == r.key.public().as_bytes().as_slice())
});
let mut jobs: BTreeMap<SignPublic, (ServerInfo, Vec<SigningKeyPair>)> = BTreeMap::new();
let retired = self
.profile
.retired_mailboxes
.iter()
.map(|r| (r.key.clone(), r.servers.clone()));
for (kp, servers) in self.contact_mailboxes().into_iter().chain(retired) {
for s in servers {
jobs.entry(s.sign_key)
.or_insert_with(|| (s.clone(), Vec::new()))
.1
.push(kp.clone());
}
}
// Drop acks nobody can send any more (contact or server removed).
self.profile.pending_acks.retain(|a| {
jobs.get(&a.server).is_some_and(|(_, kps)| {
kps.iter()
.any(|kp| kp.public().as_bytes().as_slice() == a.mailbox.as_slice())
})
});
let mut round = PullRound::default();
if jobs.is_empty() {
return Ok(Vec::new());
return Ok(round);
}
let wait_s = wait.as_secs().min(u32::MAX as u64) as u32;
let calls = jobs.values().map(|(server, kps)| {
let pending = &self.profile.pending_acks;
let mut sent_acks: Vec<Vec<PendingAck>> = Vec::with_capacity(jobs.len());
let mut requests = Vec::with_capacity(jobs.len());
for (server, kps) in jobs.values() {
let rid = meow_crypto::random_bytes::<16>().to_vec();
let ts = now();
let auths = kps.iter().map(|kp| rpc::mailbox_auth(kp, server, &rid, ts)).collect();
let req = v2::Request {
let mut sent = Vec::new();
let auths = kps
.iter()
.map(|kp| {
let mb = kp.public();
let acks: Vec<Vec<u8>> = pending
.iter()
.filter(|a| a.server == server.sign_key && a.mailbox.as_slice() == mb.as_bytes().as_slice())
.take(MAX_ACKS_PER_MAILBOX)
.inspect(|a| sent.push((*a).clone()))
.map(|a| a.id.clone())
.collect();
rpc::mailbox_auth(kp, server, &rid, acks, ts)
})
.collect();
sent_acks.push(sent);
requests.push(v2::Request {
request_id: rid,
timestamp: ts,
body: Some(Req::Pull(v2::Pull {
auths,
wait_seconds: wait_s,
})),
};
self.transport.call_request(server, req, wait)
});
});
}
let calls = jobs
.values()
.zip(requests)
.map(|((server, _), req)| self.transport.call_request(server, req, wait));
let results = join_all(calls).await;
let mut stored = Vec::new();
let mut failures = 0;
let mut last_err = None;
for ((server, _), r) in jobs.values().zip(results) {
for (((server, _), r), sent) in jobs.values().zip(results).zip(sent_acks) {
match r {
Ok(Resp::PullResult(p)) => stored.extend(p.messages),
Ok(Resp::PullResult(p)) => {
// The server applied our acks before answering. Forgetting them
// needs no save: at worst they are sent again (acks are idempotent).
self.profile.pending_acks.retain(|a| !sent.contains(a));
round.more |= p.more;
for m in p.messages {
let ack = PendingAck {
server: server.sign_key,
mailbox: m.mailbox.clone(),
id: m.id.clone(),
};
// Already processed (its ack was lost): only re-acknowledge.
if !self.profile.pending_acks.contains(&ack) {
self.profile.pending_acks.push(ack);
round.new.push(m);
}
}
}
Ok(_) => failures += 1,
Err(e) => {
tracing::warn!("pull on {} failed: {e}", server.url);
@@ -571,10 +674,39 @@ impl MeowClient {
if failures == jobs.len() {
return Err(last_err.unwrap_or_else(|| Error::Network("all servers failed".into())));
}
self.process(stored).await
Ok(round)
}
/// Every mailbox currently used by a contact, with its pull servers.
fn contact_mailboxes(&self) -> Vec<(SigningKeyPair, Vec<ServerInfo>)> {
self.profile
.contacts
.iter()
.flat_map(|c| c.mailboxes())
.map(|(kp, servers)| (kp.clone(), servers.to_vec()))
.collect()
}
/// Keep the keys of mailboxes that disappeared (`before` → now) while they still
/// have pending acks, so the next pull can acknowledge their last messages.
fn retire_mailboxes(&mut self, before: Vec<(SigningKeyPair, Vec<ServerInfo>)>) {
let current: Vec<SignPublic> = self.contact_mailboxes().iter().map(|(kp, _)| kp.public()).collect();
for (key, servers) in before {
let pk = key.public();
let has_acks = self
.profile
.pending_acks
.iter()
.any(|a| a.mailbox.as_slice() == pk.as_bytes().as_slice());
let known = self.profile.retired_mailboxes.iter().any(|r| r.key.public() == pk);
if has_acks && !known && !current.contains(&pk) {
self.profile.retired_mailboxes.push(RetiredMailbox { key, servers });
}
}
}
async fn process(&mut self, stored: Vec<v2::StoredMessage>) -> Result<Vec<Event>> {
let mailboxes_before = self.contact_mailboxes();
// Handshake messages first (step 2 before step 3), then ratchet messages by arrival.
let mut envs: Vec<(u32, u64, Vec<u8>, v2::PeerEnvelope)> = stored
.into_iter()
@@ -686,6 +818,7 @@ impl MeowClient {
_ => tracing::debug!("ignoring unsupported envelope"),
}
}
self.retire_mailboxes(mailboxes_before);
// Persist before anything leaves the device.
self.save().await?;
for (servers, deposit) in outgoing {
+24 -1
View File
@@ -9,11 +9,12 @@ use std::path::Path;
use meow_core::contact::{Contact, ContactId};
use meow_core::server::ServerInfo;
use meow_crypto::password::{self, KdfParams};
use meow_crypto::{SignPublic, SigningKeyPair};
use serde::{Deserialize, Serialize};
use crate::{Error, Result};
const FORMAT_VERSION: u32 = 1;
const FORMAT_VERSION: u32 = 2;
/// One entry of a conversation history.
#[derive(Clone, Debug, Serialize, Deserialize, PartialEq, Eq)]
@@ -33,6 +34,24 @@ pub enum Content {
File { name: String, size: u64 },
}
/// A message received from a server and processed (its effects are in this profile),
/// whose deletion the server has not confirmed yet. Sent with the next pull to that
/// server; also used to ignore a redelivery of the same message.
#[derive(Clone, Debug, Serialize, Deserialize, PartialEq, Eq)]
pub struct PendingAck {
pub server: SignPublic,
pub mailbox: Vec<u8>,
pub id: Vec<u8>,
}
/// A mailbox no longer in use (temporary invitation mailbox, removed contact) that
/// still has pending acks: its key is kept just long enough to send them.
#[derive(Clone, Serialize, Deserialize)]
pub struct RetiredMailbox {
pub key: SigningKeyPair,
pub servers: Vec<ServerInfo>,
}
#[derive(Clone, Serialize, Deserialize)]
pub struct Profile {
pub format: u32,
@@ -40,6 +59,8 @@ pub struct Profile {
pub servers: Vec<ServerInfo>,
pub contacts: Vec<Contact>,
pub history: Vec<HistoryEntry>,
pub pending_acks: Vec<PendingAck>,
pub retired_mailboxes: Vec<RetiredMailbox>,
}
impl Profile {
@@ -50,6 +71,8 @@ impl Profile {
servers: vec![],
contacts: vec![],
history: vec![],
pending_acks: vec![],
retired_mailboxes: vec![],
}
}
+157
View File
@@ -180,3 +180,160 @@ async fn file_and_server_invitations_then_messaging() {
let _ = std::fs::remove_dir_all(&dir);
}
// ───────────── at-least-once delivery (roadmap lot 3.8) ─────────────
/// Memory store whose acknowledgements can be "lost" on demand.
struct FlakyStore {
inner: MemoryStore,
drop_acks: std::sync::atomic::AtomicBool,
}
#[async_trait::async_trait]
impl meow_server::store::ServerStore for FlakyStore {
async fn post(&self, mb: [u8; 32], payload: Vec<u8>, now: u64) -> meow_server::store::StoreResult<()> {
self.inner.post(mb, payload, now).await
}
async fn fetch(
&self,
mbs: &[[u8; 32]],
limit: usize,
) -> meow_server::store::StoreResult<Vec<meow_proto::v2::StoredMessage>> {
self.inner.fetch(mbs, limit).await
}
async fn ack(&self, mb: [u8; 32], ids: &[Vec<u8>]) -> meow_server::store::StoreResult<()> {
if self.drop_acks.load(std::sync::atomic::Ordering::SeqCst) {
return Ok(());
}
self.inner.ack(mb, ids).await
}
async fn mark_request(&self, id: &[u8], ttl: Duration) -> meow_server::store::StoreResult<bool> {
self.inner.mark_request(id, ttl).await
}
async fn put_invitation(
&self,
payload: Vec<u8>,
password: &str,
ttl: Duration,
code_len: usize,
) -> meow_server::store::StoreResult<String> {
self.inner.put_invitation(payload, password, ttl, code_len).await
}
async fn get_invitation(&self, code: &str, password: &str) -> meow_server::store::StoreResult<Vec<u8>> {
self.inner.get_invitation(code, password).await
}
}
/// What a client that crashed right after receiving the response would have done:
/// pull without ever acknowledging.
async fn raw_pull(
server: &meow_core::server::ServerInfo,
mailbox: &meow_crypto::SigningKeyPair,
) -> Vec<meow_proto::v2::StoredMessage> {
use meow_proto::v2::{self, request::Body as Req, response::Body as Resp};
let rid = meow_crypto::random_bytes::<16>().to_vec();
let ts = std::time::SystemTime::now()
.duration_since(std::time::UNIX_EPOCH)
.unwrap()
.as_secs();
let req = v2::Request {
request_id: rid.clone(),
timestamp: ts,
body: Some(Req::Pull(v2::Pull {
auths: vec![meow_core::rpc::mailbox_auth(mailbox, server, &rid, vec![], ts)],
wait_seconds: 0,
})),
};
let t = meow_client::transport::Transport::new().unwrap();
match t.call_request(server, req, Duration::ZERO).await.unwrap() {
Resp::PullResult(p) => p.messages,
other => panic!("unexpected {other:?}"),
}
}
#[tokio::test(flavor = "multi_thread")]
async fn no_message_lost_when_the_client_dies() {
let listener = tokio::net::TcpListener::bind("127.0.0.1:0").await.unwrap();
let url = format!("http://{}", listener.local_addr().unwrap());
let keys = meow_core::server::ServerKeys::generate();
let (notify, _) = broadcast::channel(1024);
let store = Arc::new(FlakyStore {
inner: MemoryStore::new(notify.clone(), 1000),
drop_acks: false.into(),
});
let limits = Limits {
max_pull_messages: 2, // force the "more" path
..Limits::default()
};
let state = Arc::new(AppState {
card: encode_card(&keys, &url, "flaky", ""),
keys,
store: store.clone(),
notify,
limits,
});
tokio::spawn(meow_server::serve(listener, state));
let dir = tmpdir();
let servers = vec![url];
let mut bob = client(&dir, "bob", &servers).await;
let mut alice = client(&dir, "alice", &servers).await;
let offer = bob.create_invitation("alice", &[], "", 3600).await.unwrap();
let answer = alice.answer_invitation(&offer, None, &[]).await.unwrap();
let confirm = bob.accept_answer(&answer).await.unwrap();
alice.finalize_invitation(&confirm).await.unwrap();
bob.sync(Duration::ZERO).await.unwrap();
let server = alice.servers()[0].clone();
let mailbox = alice.contact("bob").unwrap().keys.as_ref().unwrap().mailbox.clone();
// 1. A client receives messages then dies before saving: nothing is lost.
for t in ["m1", "m2", "m3"] {
bob.send_text("alice", t).await.unwrap();
}
assert_eq!(raw_pull(&server, &mailbox).await.len(), 2); // limit 2, more = true
let ev = alice.sync(Duration::ZERO).await.unwrap();
assert_eq!(texts(&ev), vec!["m1", "m2", "m3"]); // several rounds ("more")
// Everything processed has been acknowledged and deleted.
assert!(raw_pull(&server, &mailbox).await.is_empty());
assert!(alice.sync(Duration::ZERO).await.unwrap().is_empty());
// 2. The acknowledgement is lost: the redelivered message is not processed twice,
// even after a restart (pending acks are persisted with the state).
store.drop_acks.store(true, std::sync::atomic::Ordering::SeqCst);
bob.send_text("alice", "m4").await.unwrap();
assert_eq!(texts(&alice.sync(Duration::ZERO).await.unwrap()), vec!["m4"]);
assert_eq!(raw_pull(&server, &mailbox).await.len(), 1);
let path = dir.join("alice.mwid2");
drop(alice);
let mut alice = MeowClient::open(&path, "pw", KdfParams::INSECURE_TEST).await.unwrap();
assert!(alice.sync(Duration::ZERO).await.unwrap().is_empty());
store.drop_acks.store(false, std::sync::atomic::Ordering::SeqCst);
assert!(alice.sync(Duration::ZERO).await.unwrap().is_empty());
assert!(raw_pull(&server, &mailbox).await.is_empty());
assert_eq!(alice.history("bob").unwrap().len(), 4);
assert_eq!(store.inner.stored_messages(), 0);
// 3. Mailboxes that disappear still get their last messages acknowledged:
// the temporary answer mailbox of a server invitation...
let mut carol = client(&dir, "carol", &servers).await;
let link = bob.publish_invitation("carol", &[], "", "", 3600).await.unwrap();
carol.accept_published_invitation(&link, "", None, &[]).await.unwrap();
bob.sync(Duration::ZERO).await.unwrap(); // step 2 → 3, answer mailbox retired
carol.sync(Duration::ZERO).await.unwrap(); // step 3 → 4
bob.sync(Duration::ZERO).await.unwrap(); // ack
carol.send_text("bob", "hi").await.unwrap();
assert_eq!(texts(&bob.sync(Duration::ZERO).await.unwrap()), vec!["hi"]);
assert_eq!(store.inner.stored_messages(), 0);
// ...and a contact removed while an ack is still pending.
store.drop_acks.store(true, std::sync::atomic::Ordering::SeqCst);
bob.send_text("alice", "bye").await.unwrap();
assert_eq!(texts(&alice.sync(Duration::ZERO).await.unwrap()), vec!["bye"]);
store.drop_acks.store(false, std::sync::atomic::Ordering::SeqCst);
alice.remove_contact("bob").await.unwrap();
assert_eq!(store.inner.stored_messages(), 1);
assert!(alice.sync(Duration::ZERO).await.unwrap().is_empty());
assert_eq!(store.inner.stored_messages(), 0);
let _ = std::fs::remove_dir_all(&dir);
}
+30 -8
View File
@@ -94,26 +94,44 @@ pub fn seal_response(key: &Key, resp: &v2::Response) -> Vec<u8> {
.encode_to_vec()
}
fn pull_message(server_sign: &SignPublic, mailbox: &SignPublic, ts: u64, request_id: &[u8]) -> Vec<u8> {
fn pull_message(
server_sign: &SignPublic,
mailbox: &SignPublic,
ts: u64,
acks: &[Vec<u8>],
request_id: &[u8],
) -> Vec<u8> {
let ack_refs: Vec<&[u8]> = acks.iter().map(Vec::as_slice).collect();
let acks_hash = meow_crypto::sha256(&ack_refs);
[
PULL_LABEL,
server_sign.as_bytes(),
mailbox.as_bytes(),
&ts.to_be_bytes(),
&acks_hash,
request_id,
]
.concat()
}
/// Client: prove ownership of a mailbox for one pull request.
pub fn mailbox_auth(mailbox: &SigningKeyPair, server: &ServerInfo, request_id: &[u8], now: u64) -> v2::MailboxAuth {
/// Client: prove ownership of a mailbox for one pull request, and acknowledge
/// (delete) the messages `acks` received by an earlier pull.
pub fn mailbox_auth(
mailbox: &SigningKeyPair,
server: &ServerInfo,
request_id: &[u8],
acks: Vec<Vec<u8>>,
now: u64,
) -> v2::MailboxAuth {
let pk = mailbox.public();
let signature = mailbox
.sign(&pull_message(&server.sign_key, &pk, now, &acks, request_id))
.to_vec();
v2::MailboxAuth {
mailbox: pk.to_vec(),
timestamp: now,
signature: mailbox
.sign(&pull_message(&server.sign_key, &pk, now, request_id))
.to_vec(),
signature,
acks,
}
}
@@ -130,7 +148,7 @@ pub fn verify_mailbox_auth(
return Err(Error::Expired);
}
mailbox.verify(
&pull_message(server_sign, &mailbox, auth.timestamp, request_id),
&pull_message(server_sign, &mailbox, auth.timestamp, &auth.acks, request_id),
&auth.signature,
)?;
Ok(mailbox)
@@ -178,9 +196,13 @@ mod tests {
let keys = ServerKeys::generate();
let info = ServerInfo::from_card(&keys.card("http://s", "s", "", 1)).unwrap();
let mb = SigningKeyPair::generate();
let auth = mailbox_auth(&mb, &info, b"rid", 1000);
let auth = mailbox_auth(&mb, &info, b"rid", vec![b"m1".to_vec()], 1000);
let sp = keys.sign.public();
assert_eq!(verify_mailbox_auth(&sp, b"rid", &auth, 1100).unwrap(), mb.public());
// Acks are covered by the signature.
let mut forged = auth.clone();
forged.acks.push(b"m2".to_vec());
assert!(verify_mailbox_auth(&sp, b"rid", &forged, 1100).is_err());
assert!(verify_mailbox_auth(&sp, b"other", &auth, 1100).is_err());
assert!(verify_mailbox_auth(&sp, b"rid", &auth, 1000 + AUTH_WINDOW_SECS + 1).is_err());
let other_server = ServerKeys::generate().sign.public();
+38 -4
View File
@@ -39,6 +39,12 @@ pub struct Limits {
pub max_deposits: usize,
pub max_payload_bytes: usize,
pub max_mailboxes_per_pull: usize,
/// Messages returned by one pull (the client pulls again when `more` is set).
pub max_pull_messages: usize,
/// Payload bytes returned by one pull (at least one message is always returned).
pub max_pull_bytes: usize,
/// Acknowledged ids per mailbox in one pull.
pub max_acks: usize,
pub max_wait: Duration,
pub max_invitation_ttl: Duration,
pub replay_ttl: Duration,
@@ -51,6 +57,9 @@ impl Default for Limits {
max_deposits: 128,
max_payload_bytes: 8 * 1024 * 1024,
max_mailboxes_per_pull: 512,
max_pull_messages: 256,
max_pull_bytes: 16 * 1024 * 1024,
max_acks: 1024,
max_wait: Duration::from_secs(120),
max_invitation_ttl: Duration::from_secs(7 * 86_400),
replay_ttl: Duration::from_secs(2 * rpc::AUTH_WINDOW_SECS),
@@ -196,13 +205,23 @@ async fn pull(st: &AppState, request_id: &[u8], p: v2::Pull, now: u64) -> HResul
let sign = st.keys.sign.public();
let mut mailboxes = Vec::with_capacity(p.auths.len());
for a in &p.auths {
if a.acks.len() > st.limits.max_acks || a.acks.iter().any(|id| id.len() > 64) {
return Err((Code::BadRequest, "bad acks"));
}
let mb = rpc::verify_mailbox_auth(&sign, request_id, a, now)
.map_err(|_| (Code::Unauthorized, "bad mailbox auth"))?;
mailboxes.push(mb.0);
}
// Subscribe before draining, so a deposit landing in between is not missed.
// Acknowledged messages were persisted by the client: delete them first.
for (mb, a) in mailboxes.iter().zip(&p.auths) {
if !a.acks.is_empty() {
st.store.ack(*mb, &a.acks).await.map_err(store_err)?;
}
}
// Subscribe before reading, so a deposit landing in between is not missed.
let mut rx = st.notify.subscribe();
let mut messages = st.store.drain(&mailboxes).await.map_err(store_err)?;
let limit = st.limits.max_pull_messages;
let mut messages = st.store.fetch(&mailboxes, limit + 1).await.map_err(store_err)?;
let wait = Duration::from_secs(p.wait_seconds as u64).min(st.limits.max_wait);
if messages.is_empty() && !wait.is_zero() {
let wanted: HashSet<Mailbox> = mailboxes.iter().copied().collect();
@@ -217,9 +236,24 @@ async fn pull(st: &AppState, request_id: &[u8], p: v2::Pull, now: u64) -> HResul
}
})
.await;
messages = st.store.drain(&mailboxes).await.map_err(store_err)?;
messages = st.store.fetch(&mailboxes, limit + 1).await.map_err(store_err)?;
}
Ok(Resp::PullResult(v2::PullResult { messages }))
// Honour the count and size budgets; always return at least one message.
let mut more = messages.len() > limit;
messages.truncate(limit);
let mut bytes = 0usize;
let keep = messages
.iter()
.position(|m| {
bytes += m.payload.len();
bytes > st.limits.max_pull_bytes
})
.map_or(messages.len(), |i| i.max(1));
if keep < messages.len() {
messages.truncate(keep);
more = true;
}
Ok(Resp::PullResult(v2::PullResult { messages, more }))
}
/// Serve on an already bound listener until the future is dropped.
+136 -26
View File
@@ -41,10 +41,13 @@ pub const MAX_INVITATION_ATTEMPTS: i64 = 3;
#[async_trait]
pub trait ServerStore: Send + Sync {
/// Append a payload to a mailbox and notify waiters.
/// Append a payload to a mailbox (under a fresh id) and notify waiters.
async fn post(&self, mailbox: Mailbox, payload: Vec<u8>, now: u64) -> StoreResult<()>;
/// Atomically take every message of the given mailboxes.
async fn drain(&self, mailboxes: &[Mailbox]) -> StoreResult<Vec<v2::StoredMessage>>;
/// Read, oldest first and **without removing them**, at most `limit` messages of the
/// given mailboxes. Messages stay until acknowledged with [`ServerStore::ack`].
async fn fetch(&self, mailboxes: &[Mailbox], limit: usize) -> StoreResult<Vec<v2::StoredMessage>>;
/// Delete the given messages of a mailbox. Unknown ids are ignored (idempotent).
async fn ack(&self, mailbox: Mailbox, ids: &[Vec<u8>]) -> StoreResult<()>;
/// Record a request id; returns `false` if it was already seen (replay).
async fn mark_request(&self, request_id: &[u8], ttl: Duration) -> StoreResult<bool>;
/// Store an invitation under a fresh shortcode.
@@ -90,6 +93,7 @@ struct MemInvitation {
#[derive(Default)]
struct MemInner {
mailboxes: HashMap<Mailbox, VecDeque<v2::StoredMessage>>,
next_id: u64,
requests: HashMap<Vec<u8>, Instant>,
invitations: HashMap<String, MemInvitation>,
}
@@ -110,6 +114,11 @@ impl MemoryStore {
}
}
/// Number of messages currently stored (all mailboxes).
pub fn stored_messages(&self) -> usize {
self.lock().map_or(0, |g| g.mailboxes.values().map(VecDeque::len).sum())
}
fn lock(&self) -> StoreResult<std::sync::MutexGuard<'_, MemInner>> {
self.inner
.lock()
@@ -122,11 +131,14 @@ impl ServerStore for MemoryStore {
async fn post(&self, mailbox: Mailbox, payload: Vec<u8>, now: u64) -> StoreResult<()> {
{
let mut g = self.lock()?;
g.next_id += 1;
let id = g.next_id.to_be_bytes().to_vec();
let q = g.mailboxes.entry(mailbox).or_default();
q.push_back(v2::StoredMessage {
mailbox: mailbox.to_vec(),
payload,
received_at: now,
id,
});
while q.len() > self.max_per_mailbox {
q.pop_front();
@@ -136,15 +148,28 @@ impl ServerStore for MemoryStore {
Ok(())
}
async fn drain(&self, mailboxes: &[Mailbox]) -> StoreResult<Vec<v2::StoredMessage>> {
let mut g = self.lock()?;
async fn fetch(&self, mailboxes: &[Mailbox], limit: usize) -> StoreResult<Vec<v2::StoredMessage>> {
let g = self.lock()?;
Ok(mailboxes
.iter()
.filter_map(|m| g.mailboxes.remove(m))
.filter_map(|m| g.mailboxes.get(m))
.flatten()
.take(limit)
.cloned()
.collect())
}
async fn ack(&self, mailbox: Mailbox, ids: &[Vec<u8>]) -> StoreResult<()> {
let mut g = self.lock()?;
if let Some(q) = g.mailboxes.get_mut(&mailbox) {
q.retain(|m| !ids.contains(&m.id));
if q.is_empty() {
g.mailboxes.remove(&mailbox);
}
}
Ok(())
}
async fn mark_request(&self, request_id: &[u8], ttl: Duration) -> StoreResult<bool> {
let mut g = self.lock()?;
let now = Instant::now();
@@ -207,7 +232,8 @@ pub const NOTIFY_CHANNEL: &str = "meow:v2:new";
/// Redis-backed store.
///
/// Keys: `meow:v2:mb:{hex}` (list of encoded `StoredMessage`), `meow:v2:rq:{hex}`
/// Keys: `meow:v2:mb:{hex}` (stream; the stream id is the message id, field `m` holds the
/// encoded `StoredMessage`), `meow:v2:rq:{hex}`
/// (replay cache), `meow:v2:inv:{code}`, `meow:v2:invpw:{code}`, `meow:v2:invfa:{code}`.
pub struct RedisStore {
conn: redis::aio::MultiplexedConnection,
@@ -219,6 +245,22 @@ fn hexs(b: &[u8]) -> String {
b.iter().map(|x| format!("{x:02x}")).collect()
}
fn mailbox_key(mb: &Mailbox) -> String {
format!("meow:v2:mb:{}", hexs(mb))
}
fn is_stream_id(s: &str) -> bool {
match s.split_once('-') {
Some((ms, seq)) => {
!ms.is_empty()
&& !seq.is_empty()
&& (ms.len() + seq.len()) <= 40
&& (ms.bytes().chain(seq.bytes())).all(|b| b.is_ascii_digit())
}
None => false,
}
}
impl RedisStore {
/// Connect and start the notification forwarder.
pub async fn connect(
@@ -252,20 +294,24 @@ impl RedisStore {
#[async_trait]
impl ServerStore for RedisStore {
async fn post(&self, mailbox: Mailbox, payload: Vec<u8>, now: u64) -> StoreResult<()> {
let key = format!("meow:v2:mb:{}", hexs(&mailbox));
let key = mailbox_key(&mailbox);
let item = v2::StoredMessage {
mailbox: mailbox.to_vec(),
payload,
received_at: now,
id: Vec::new(),
}
.encode_to_vec();
let mut c = self.conn.clone();
let max = self.max_per_mailbox as isize;
let _: () = redis::pipe()
.atomic()
.rpush(&key, item)
.ignore()
.ltrim(&key, -max, -1)
.cmd("XADD")
.arg(&key)
.arg("MAXLEN")
.arg(self.max_per_mailbox)
.arg("*")
.arg("m")
.arg(item)
.ignore()
.expire(&key, self.mailbox_ttl.as_secs() as i64)
.ignore()
@@ -276,27 +322,61 @@ impl ServerStore for RedisStore {
Ok(())
}
async fn drain(&self, mailboxes: &[Mailbox]) -> StoreResult<Vec<v2::StoredMessage>> {
async fn fetch(&self, mailboxes: &[Mailbox], limit: usize) -> StoreResult<Vec<v2::StoredMessage>> {
let mut out = Vec::new();
let mut c = self.conn.clone();
for mb in mailboxes {
let key = format!("meow:v2:mb:{}", hexs(mb));
let (items, _): (Vec<Vec<u8>>, i64) = redis::pipe()
.atomic()
.lrange(&key, 0, -1)
.del(&key)
let left = limit.saturating_sub(out.len());
if left == 0 {
break;
}
let key = mailbox_key(mb);
let entries: Vec<(String, Vec<Vec<u8>>)> = redis::cmd("XRANGE")
.arg(&key)
.arg("-")
.arg("+")
.arg("COUNT")
.arg(left)
.query_async(&mut c)
.await?;
for it in items {
match v2::StoredMessage::decode(it.as_slice()) {
Ok(m) => out.push(m),
Err(e) => tracing::warn!("dropping corrupted stored message: {e}"),
for (id, fields) in entries {
// fields = ["m", item]
let decoded = fields.get(1).map(|it| v2::StoredMessage::decode(it.as_slice()));
match decoded {
Some(Ok(mut m)) => {
m.id = id.into_bytes();
out.push(m);
}
_ => {
tracing::warn!("dropping corrupted stored message");
let _: i64 = redis::cmd("XDEL").arg(&key).arg(&id).query_async(&mut c).await?;
}
}
}
}
Ok(out)
}
async fn ack(&self, mailbox: Mailbox, ids: &[Vec<u8>]) -> StoreResult<()> {
// Stream ids are ASCII `ms-seq`; anything else cannot exist and is skipped
// (XDEL would reject the whole command on a malformed id).
let ids: Vec<&str> = ids
.iter()
.filter_map(|i| std::str::from_utf8(i).ok())
.filter(|i| is_stream_id(i))
.collect();
if ids.is_empty() {
return Ok(());
}
let mut c = self.conn.clone();
let _: i64 = redis::cmd("XDEL")
.arg(mailbox_key(&mailbox))
.arg(ids)
.query_async(&mut c)
.await?;
Ok(())
}
async fn mark_request(&self, request_id: &[u8], ttl: Duration) -> StoreResult<bool> {
let mut c = self.conn.clone();
let r: Option<String> = redis::cmd("SET")
@@ -380,6 +460,15 @@ impl ServerStore for RedisStore {
mod tests {
use super::*;
#[test]
fn stream_ids() {
assert!(is_stream_id("1700000000000-0"));
assert!(!is_stream_id("1700000000000"));
assert!(!is_stream_id("-0"));
assert!(!is_stream_id("12-a"));
assert!(!is_stream_id("1-2-3"));
}
#[test]
fn shortcodes() {
let c = random_shortcode(8);
@@ -392,13 +481,34 @@ mod tests {
let mb = meow_crypto::random_bytes::<32>();
store.post(mb, b"one".to_vec(), 1).await.unwrap();
store.post(mb, b"two".to_vec(), 2).await.unwrap();
store.post(mb, b"three".to_vec(), 3).await.unwrap();
assert_eq!(rx.recv().await.unwrap(), mb);
let got = store.drain(&[mb]).await.unwrap();
let payloads = |v: &[v2::StoredMessage]| v.iter().map(|m| m.payload.clone()).collect::<Vec<_>>();
let got = store.fetch(&[mb], 10).await.unwrap();
assert_eq!(
got.iter().map(|m| m.payload.clone()).collect::<Vec<_>>(),
vec![b"one".to_vec(), b"two".to_vec()]
payloads(&got),
vec![b"one".to_vec(), b"two".to_vec(), b"three".to_vec()]
);
assert!(store.drain(&[mb]).await.unwrap().is_empty());
// Not acknowledged: delivered again. The limit is honoured.
let again = store.fetch(&[mb], 2).await.unwrap();
assert_eq!(payloads(&again), vec![b"one".to_vec(), b"two".to_vec()]);
assert_eq!(again[0].id, got[0].id);
assert_ne!(got[0].id, got[1].id);
// Ack two of them (plus garbage, which is ignored): only "three" remains.
store
.ack(mb, &[got[0].id.clone(), got[1].id.clone(), b"bogus".to_vec()])
.await
.unwrap();
assert_eq!(
payloads(&store.fetch(&[mb], 10).await.unwrap()),
vec![b"three".to_vec()]
);
// Acks are scoped to their mailbox.
let other = meow_crypto::random_bytes::<32>();
store.ack(other, &[got[2].id.clone()]).await.unwrap();
assert_eq!(store.fetch(&[mb], 10).await.unwrap().len(), 1);
store.ack(mb, &[got[2].id.clone()]).await.unwrap();
assert!(store.fetch(&[mb], 10).await.unwrap().is_empty());
let rid = meow_crypto::random_bytes::<16>();
assert!(store.mark_request(&rid, Duration::from_secs(60)).await.unwrap());
+8 -2
View File
@@ -201,7 +201,10 @@ message PostAck { uint32 accepted = 1; }
message MailboxAuth {
bytes mailbox = 1;
uint64 timestamp = 2;
bytes signature = 3; // see ARCHITECTURE.md §3.5
bytes signature = 3; // see ARCHITECTURE.md §3.5 (covers `acks`)
// Ids of messages of this mailbox, received by an earlier pull and now safely
// persisted by the client: the server deletes them before answering.
repeated bytes acks = 4;
}
message Pull { repeated MailboxAuth auths = 1; uint32 wait_seconds = 2; }
@@ -209,8 +212,11 @@ message StoredMessage {
bytes mailbox = 1;
bytes payload = 2;
uint64 received_at = 3;
bytes id = 4; // server-assigned, opaque, unique within the mailbox
}
message PullResult { repeated StoredMessage messages = 1; }
// Messages stay on the server until acknowledged (at-least-once delivery).
// `more`: the server held back messages (count/size limit), pull again.
message PullResult { repeated StoredMessage messages = 1; bool more = 2; }
message InvitationPublish {
bytes payload = 1; // encoded InvitationOffer