Implement replay-safe browser synchronization

This commit is contained in:
ms2k 2026-07-18 00:16:21 +09:00
commit 459ffa04c6
7 changed files with 2734 additions and 25 deletions

View file

@ -0,0 +1,10 @@
CREATE TABLE live_event_tickets (
ticket_digest BYTEA PRIMARY KEY CHECK (octet_length(ticket_digest) = 32),
principal JSONB NOT NULL,
origin TEXT NOT NULL CHECK (origin <> ''),
created_at TIMESTAMPTZ NOT NULL,
expires_at TIMESTAMPTZ NOT NULL CHECK (expires_at > created_at)
);
CREATE INDEX live_event_tickets_expiry_idx
ON live_event_tickets (expires_at);

View file

@ -1,4 +1,4 @@
use std::time::Duration;
use std::{collections::HashSet, time::Duration};
use actix_web::{
HttpRequest, HttpResponse, ResponseError,
@ -51,6 +51,8 @@ pub struct PublicApiConfig {
pub event_retention: Duration,
pub event_poll_interval: Duration,
pub event_heartbeat_interval: Duration,
pub event_ticket_ttl: Duration,
pub event_allowed_origins: Vec<String>,
}
#[derive(Clone)]
@ -68,6 +70,8 @@ pub struct PublicApiState {
event_retention: Duration,
event_poll_interval: Duration,
event_heartbeat_interval: Duration,
event_ticket_ttl: Duration,
event_allowed_origins: HashSet<String>,
}
pub struct PublicApiServices {
@ -93,9 +97,28 @@ impl PublicApiState {
if config.event_retention.is_zero()
|| config.event_poll_interval.is_zero()
|| config.event_heartbeat_interval.is_zero()
|| config.event_ticket_ttl.is_zero()
{
return Err(DurableError::Config(
"event retention and intervals must be positive".into(),
"event retention, intervals, and ticket expiry must be positive".into(),
));
}
if config.event_ticket_ttl > Duration::from_secs(60) {
return Err(DurableError::Config(
"event ticket expiry must not exceed 60 seconds".into(),
));
}
let event_allowed_origins = config
.event_allowed_origins
.into_iter()
.collect::<HashSet<_>>();
if event_allowed_origins.is_empty()
|| event_allowed_origins
.iter()
.any(|origin| origin.is_empty() || origin.trim() != origin)
{
return Err(DurableError::Config(
"at least one exact event origin is required".into(),
));
}
Ok(Self {
@ -112,6 +135,8 @@ impl PublicApiState {
event_retention: config.event_retention,
event_poll_interval: config.event_poll_interval,
event_heartbeat_interval: config.event_heartbeat_interval,
event_ticket_ttl: config.event_ticket_ttl,
event_allowed_origins,
})
}
@ -193,6 +218,7 @@ pub fn configure_public_api(config: &mut web::ServiceConfig) {
"/deletions/{operation_id}",
web::get().to(get_deletion_operation),
)
.route("/event-tickets", web::post().to(events::issue_ticket))
.route("/events", web::get().to(events::connect)),
);
}

View file

@ -1,16 +1,19 @@
use actix_web::{HttpRequest, HttpResponse, web};
use actix_web::{HttpRequest, HttpResponse, http::header, web};
use futures_util::StreamExt as _;
use remora_contracts::{
CursorDto, DateTimeDto, DeletionChangedDto, DeletionStateDto, EventAttemptDto, EventFrameDto,
EventProgressDto, HeartbeatDto, JobChangedDto, NullableDto, PositiveIntegerDto,
EventProgressDto, EventTicketDto, HeartbeatDto, JobChangedDto, NullableDto, PositiveIntegerDto,
ResyncReasonDto, ResyncRequiredDto, SafeIntegerDto, SchemaVersionDto,
};
use remora_domain::{JobLifecycle, JobPhase, OpaqueId};
use remora_domain::{
BucketName, JobLifecycle, JobPhase, OpaqueId, StorageScope, TargetPrefix, TargetStoreId,
};
use remora_durable::{DurableError, KeysetCursor};
use remora_identity::{ManagementOperation, Principal};
use serde::Deserialize;
use serde::{Deserialize, Serialize};
use serde_json::Value;
use sqlx::{FromRow, PgPool};
use sha2::{Digest as _, Sha256};
use sqlx::{FromRow, PgPool, types::Json};
use time::{OffsetDateTime, format_description::well_known::Rfc3339};
use super::{
@ -24,6 +27,60 @@ const EVENT_BATCH: i64 = 100;
#[derive(Deserialize)]
pub(super) struct EventQuery {
cursor: Option<String>,
ticket: Option<String>,
}
#[derive(Deserialize, Serialize)]
struct TicketPrincipal {
id: String,
issuer: String,
subject: String,
administrator: bool,
scopes: Vec<TicketScope>,
}
#[derive(Deserialize, Serialize)]
struct TicketScope {
store: String,
bucket: String,
prefix: String,
}
pub(super) async fn issue_ticket(
state: web::Data<PublicApiState>,
request: HttpRequest,
) -> Result<HttpResponse, PublicApiError> {
let principal = state.principal(&request).await?;
let origin = allowed_origin(&state, &request)?;
let ticket = random_ticket();
let digest = ticket_digest(ticket.as_str());
let mut tx = state.pool.begin().await?;
sqlx::query("DELETE FROM live_event_tickets WHERE expires_at <= now()")
.execute(tx.as_mut())
.await?;
let expires_at: OffsetDateTime = sqlx::query_scalar(
"INSERT INTO live_event_tickets (
ticket_digest, principal, origin, created_at, expires_at
) VALUES (
$1, $2, $3, now(),
now() + $4 * interval '1 second'
)
RETURNING expires_at",
)
.bind(digest.as_slice())
.bind(Json(ticket_principal_projection(&principal)))
.bind(&origin)
.bind(state.event_ticket_ttl.as_secs_f64())
.fetch_one(tx.as_mut())
.await?;
tx.commit().await?;
Ok(HttpResponse::Created()
.insert_header((header::CACHE_CONTROL, "no-store"))
.insert_header((header::PRAGMA, "no-cache"))
.json(EventTicketDto {
ticket,
expires_at: event_time(expires_at)?,
}))
}
pub(super) async fn connect(
@ -32,7 +89,13 @@ pub(super) async fn connect(
body: web::Payload,
query: web::Query<EventQuery>,
) -> Result<HttpResponse, PublicApiError> {
let principal = state.principal(&request).await?;
let bearer = request.headers().contains_key(header::AUTHORIZATION);
let principal = match (bearer, query.ticket.as_deref()) {
(true, None) => state.principal(&request).await?,
(false, Some(ticket)) => redeem_ticket(&state, &request, ticket).await?,
(false, None) => return Err(PublicApiError::unauthenticated()),
(true, Some(_)) => return Err(PublicApiError::invalid_request()),
};
let now = OffsetDateTime::now_utc();
let retention =
time::Duration::try_from(state.event_retention).map_err(|_| PublicApiError::internal())?;
@ -54,6 +117,94 @@ pub(super) async fn connect(
Ok(response)
}
fn allowed_origin(state: &PublicApiState, request: &HttpRequest) -> Result<String, PublicApiError> {
let mut origins = request.headers().get_all(header::ORIGIN);
let origin = origins
.next()
.and_then(|value| value.to_str().ok())
.ok_or_else(PublicApiError::forbidden)?;
if origins.next().is_some() || !state.event_allowed_origins.contains(origin) {
return Err(PublicApiError::forbidden());
}
Ok(origin.to_owned())
}
async fn redeem_ticket(
state: &PublicApiState,
request: &HttpRequest,
ticket: &str,
) -> Result<Principal, PublicApiError> {
let origin = allowed_origin(state, request)?;
let digest = ticket_digest(ticket);
let projection: Option<Json<TicketPrincipal>> = sqlx::query_scalar(
"DELETE FROM live_event_tickets
WHERE ticket_digest = $1 AND origin = $2 AND expires_at > now()
RETURNING principal",
)
.bind(digest.as_slice())
.bind(origin)
.fetch_optional(&state.pool)
.await?;
projection
.ok_or_else(PublicApiError::unauthenticated)
.and_then(|projection| ticket_principal(projection.0))
}
fn ticket_principal_projection(principal: &Principal) -> TicketPrincipal {
TicketPrincipal {
id: principal.id.as_str().to_owned(),
issuer: principal.issuer.clone(),
subject: principal.subject.clone(),
administrator: principal.administrator,
scopes: principal
.scopes
.iter()
.map(|scope| TicketScope {
store: scope.store.as_str().to_owned(),
bucket: scope.bucket.as_str().to_owned(),
prefix: scope.prefix.canonical(),
})
.collect(),
}
}
fn ticket_principal(projection: TicketPrincipal) -> Result<Principal, PublicApiError> {
let scopes = projection
.scopes
.into_iter()
.map(|scope| {
Ok(StorageScope {
store: TargetStoreId::parse(scope.store).map_err(|_| PublicApiError::internal())?,
bucket: BucketName::parse(scope.bucket).map_err(|_| PublicApiError::internal())?,
prefix: TargetPrefix::parse(&scope.prefix, std::iter::empty())
.map_err(|_| PublicApiError::internal())?,
})
})
.collect::<Result<Vec<_>, PublicApiError>>()?;
Ok(Principal {
id: OpaqueId::parse(projection.id).map_err(|_| PublicApiError::internal())?,
issuer: projection.issuer,
subject: projection.subject,
administrator: projection.administrator,
scopes,
})
}
fn random_ticket() -> OpaqueId {
let mut bytes = [0_u8; 32];
getrandom::fill(&mut bytes).expect("operating system randomness is required");
let mut value = String::from("t");
for byte in bytes {
use std::fmt::Write as _;
write!(value, "{byte:02x}").expect("writing to String cannot fail");
}
OpaqueId::parse(value).expect("random event ticket is valid")
}
fn ticket_digest(ticket: &str) -> [u8; 32] {
Sha256::digest(ticket.as_bytes()).into()
}
pub(super) async fn prune_history(
pool: &PgPool,
before: OffsetDateTime,

View file

@ -29,9 +29,12 @@ use time::OffsetDateTime;
use tokio_tungstenite::{
MaybeTlsStream, WebSocketStream, connect_async,
tungstenite::{
Message,
Error as WebSocketError, Message,
client::IntoClientRequest,
http::{HeaderValue, header::AUTHORIZATION},
http::{
HeaderValue,
header::{AUTHORIZATION, ORIGIN},
},
},
};
@ -40,6 +43,7 @@ mod support;
use support::{mint_provider_token, oidc_provider, required_env, revoke_remora_grants};
type Socket = WebSocketStream<MaybeTlsStream<tokio::net::TcpStream>>;
const BROWSER_ORIGIN: &str = "https://remora.test";
struct Harness {
pool: PgPool,
@ -158,6 +162,8 @@ impl Harness {
event_retention: Duration::from_millis(500),
event_poll_interval: Duration::from_millis(250),
event_heartbeat_interval: Duration::from_millis(100),
event_ticket_ttl: Duration::from_secs(30),
event_allowed_origins: vec!["https://remora.test".into()],
},
)
.unwrap();
@ -227,7 +233,50 @@ impl Harness {
response.json().await.unwrap()
}
async fn issue_ticket(&self, token: &str) -> String {
let response = self
.client
.post(format!("{}/event-tickets", self.http_base))
.bearer_auth(token)
.header("Origin", BROWSER_ORIGIN)
.send()
.await
.unwrap();
assert_eq!(response.status(), StatusCode::CREATED);
assert_eq!(response.headers().get("Cache-Control").unwrap(), "no-store");
response.json::<Value>().await.unwrap()["ticket"]
.as_str()
.unwrap()
.to_owned()
}
async fn connect(&self, token: &str, cursor: Option<&str>) -> Socket {
let ticket = self.issue_ticket(token).await;
self.connect_ticket(&ticket, cursor, BROWSER_ORIGIN)
.await
.unwrap()
}
async fn connect_ticket(
&self,
ticket: &str,
cursor: Option<&str>,
origin: &str,
) -> Result<Socket, WebSocketError> {
let url = cursor.map_or_else(
|| format!("{}?ticket={ticket}", self.ws_base),
|cursor| format!("{}?ticket={ticket}&cursor={cursor}", self.ws_base),
);
let mut request = url.into_client_request().unwrap();
request
.headers_mut()
.insert(ORIGIN, HeaderValue::from_str(origin).unwrap());
let (socket, response) = connect_async(request).await?;
assert_eq!(response.status(), 101);
Ok(socket)
}
async fn connect_bearer(&self, token: &str, cursor: Option<&str>) -> Socket {
let url = cursor.map_or_else(
|| self.ws_base.clone(),
|cursor| format!("{}?cursor={cursor}", self.ws_base),
@ -255,6 +304,60 @@ async fn live_event_replay_authorization_progress_and_resync_converge() {
return;
};
let ticket = harness.issue_ticket(&harness.token).await;
let forbidden_ticket = harness
.client
.post(format!("{}/event-tickets", harness.http_base))
.bearer_auth(&harness.token)
.header("Origin", "https://attacker.invalid")
.send()
.await
.unwrap();
assert_eq!(forbidden_ticket.status(), StatusCode::FORBIDDEN);
assert!(
harness
.connect_ticket(&ticket, None, "https://attacker.invalid")
.await
.is_err()
);
let valid = harness
.connect_ticket(&ticket, None, BROWSER_ORIGIN)
.await
.unwrap();
drop(valid);
assert!(
harness
.connect_ticket(&ticket, None, BROWSER_ORIGIN)
.await
.is_err()
);
let concurrent_ticket = harness.issue_ticket(&harness.token).await;
let (left, right) = tokio::join!(
harness.connect_ticket(&concurrent_ticket, None, BROWSER_ORIGIN),
harness.connect_ticket(&concurrent_ticket, None, BROWSER_ORIGIN)
);
assert_eq!(usize::from(left.is_ok()) + usize::from(right.is_ok()), 1);
let expired_ticket = harness.issue_ticket(&harness.token).await;
sqlx::query(
"UPDATE live_event_tickets
SET expires_at = created_at + interval '1 microsecond'
WHERE expires_at > now()",
)
.execute(&harness.pool)
.await
.unwrap();
assert!(
harness
.connect_ticket(&expired_ticket, None, BROWSER_ORIGIN)
.await
.is_err()
);
let mut bearer = harness.connect_bearer(&harness.token, None).await;
assert_resync(next_frame(&mut bearer).await, ResyncReasonDto::CursorAbsent);
let mut absent = harness.connect(&harness.token, None).await;
assert_resync(next_frame(&mut absent).await, ResyncReasonDto::CursorAbsent);

View file

@ -144,6 +144,8 @@ impl Harness {
event_retention: Duration::from_secs(60),
event_poll_interval: Duration::from_millis(50),
event_heartbeat_interval: Duration::from_secs(1),
event_ticket_ttl: Duration::from_secs(30),
event_allowed_origins: vec!["https://remora.test".into()],
},
)
.unwrap();

View file

@ -15,6 +15,13 @@ pub enum EventFrameDto {
Heartbeat(HeartbeatDto),
}
#[derive(Clone, Debug, Deserialize, Eq, PartialEq, Serialize)]
#[serde(deny_unknown_fields)]
pub struct EventTicketDto {
pub ticket: OpaqueId,
pub expires_at: DateTimeDto,
}
#[derive(Clone, Debug, Deserialize, Eq, PartialEq, Serialize)]
#[serde(deny_unknown_fields)]
pub struct JobChangedDto {

View file

@ -25,7 +25,7 @@ pub use dto::{
};
pub use events::{
DeletionChangedDto, DeletionStateDto, EventAttemptDto, EventFrameDto, EventProgressDto,
HeartbeatDto, JobChangedDto, ResyncReasonDto, ResyncRequiredDto,
EventTicketDto, HeartbeatDto, JobChangedDto, ResyncReasonDto, ResyncRequiredDto,
};
pub use internal::{
ActivationResponseDto, AttemptActivateRequestDto, AttemptCheckpointRequestDto,