Files
continuwuity/src/service/federation/execute.rs
T
theS1LV3R 4f63edac33 feat: Support marking servers as unhealthy+stale during resolution
For when the server resolution process returns an error or Gone
2026-11-12 14:26:17 +00:00

444 lines
13 KiB
Rust

use std::{borrow::Cow, fmt::Debug, mem, time::Instant};
use conduwuit::{
Err, Error, Result, debug, debug_error, debug_info, debug_warn, err, metrics::KeyValue,
trace, utils::response::LimitReadExt,
};
use ipaddress::IPAddress;
use reqwest::{Client, Method, Request, Response, Url};
use resolvematrix::resolution::Resolution;
use ruma::{
ServerName,
api::{
EndpointError, IncomingResponseExt, OutgoingRequest, OutgoingRequestExt,
SupportedVersions,
auth_scheme::{AuthScheme, NoAuthentication},
error::Error as RumaError,
federation::authentication::{ServerSignatures, XMatrixSigningInput},
path_builder::PathBuilder,
},
};
use crate::SUPPORTED_VERSIONS;
impl super::Service {
/// Sends a signed request to a remote server over federation.
#[tracing::instrument(skip_all, name = "request", level = "debug")]
pub async fn execute<'i, T>(
&self,
dest: &ServerName,
request: T,
) -> Result<T::IncomingResponse>
where
T: OutgoingRequest<
Authentication = ServerSignatures,
PathBuilder: PathBuilder<Input<'i>: FederationPathBuilderInput>,
> + Debug
+ Send,
{
let client = &self.services.client.federation;
self.execute_signed(client, dest, request).await
}
/// Sends a signed request to a remote server over federation, with a
/// significantly higher timeout to account for slow endpoints.
#[tracing::instrument(skip_all, name = "request_slow", level = "debug")]
pub async fn execute_slow<'i, T>(
&self,
dest: &ServerName,
request: T,
) -> Result<T::IncomingResponse>
where
T: OutgoingRequest<
Authentication = ServerSignatures,
PathBuilder: PathBuilder<Input<'i>: FederationPathBuilderInput>,
> + Debug
+ Send,
{
let client = &self.services.client.federation_slow;
self.execute_signed(client, dest, request).await
}
/// Sends an unsigned (unauthenticated) request to a remote server over
/// federation.
pub async fn execute_unauthenticated<'i, T>(
&self,
dest: &ServerName,
request: T,
) -> Result<T::IncomingResponse>
where
T: OutgoingRequest<
Authentication = NoAuthentication,
PathBuilder: PathBuilder<Input<'i>: FederationPathBuilderInput>,
> + Debug
+ Send,
{
let client = &self.services.client.federation;
self.execute_on(client, dest, request, ()).await
}
/// Sends a signed request to a remote server over federation, given the
/// specific client.
pub async fn execute_signed<'i, T>(
&self,
client: &Client,
dest: &ServerName,
request: T,
) -> Result<T::IncomingResponse>
where
T: OutgoingRequest<
Authentication = ServerSignatures,
PathBuilder: PathBuilder<Input<'i>: FederationPathBuilderInput>,
> + Send,
{
let authentication = XMatrixSigningInput::new(
self.services.server.name.clone(),
dest.to_owned(),
self.services.server_keys.keypair(),
);
self.execute_on(client, dest, request, authentication).await
}
/// Prepares and sends a federation request with the given client.
/// If federation is disabled, an error is returned here. Likewise, if the
/// destination is forbidden.
/// Preparation includes resolving the server before sending the request.
#[tracing::instrument(
name = "fed",
level = "info",
skip(self, client, request, authentication)
)]
pub async fn execute_on<'i, T, PathBuilderInput>(
&self,
client: &Client,
dest: &ServerName,
request: T,
authentication: <T::Authentication as AuthScheme>::Input<'_>,
) -> Result<T::IncomingResponse>
where
T: OutgoingRequest<PathBuilder: PathBuilder<Input<'i> = PathBuilderInput>> + Send,
PathBuilderInput: FederationPathBuilderInput,
{
if !self.services.server.config.allow_federation {
return Err!(Config("allow_federation", "Federation is disabled."));
}
if self.services.moderation.is_remote_server_forbidden(dest) {
return Err!(Request(Forbidden(debug_warn!(
"Federation with {dest} is not allowed."
))));
}
self.ensure_remote_is_healthy(dest)?;
let actual = self
.services
.client
.matrix_resolver
.resolve_server(dest.as_str())
.await
.inspect_err(|_| {
self.hit_unhealthy(dest.to_owned());
self.mark_destination_stale(dest);
})?;
let request = Request::try_from(request.try_into_http_request::<Vec<u8>>(
actual.base_url().as_str(),
authentication,
PathBuilderInput::create(),
)?)?;
self.validate_url(request.url())?;
self.services.server.check_running()?;
self.perform::<T>(dest, &actual, request, client).await
}
/// Actually sends the federation request, handling the response before
/// returning.
///
/// Timing info and request logs are available when debug logs are enabled.
async fn perform<T>(
&self,
dest: &ServerName,
actual: &Resolution,
request: Request,
client: &Client,
) -> Result<T::IncomingResponse>
where
T: OutgoingRequest + Send,
{
let url = request.url().clone();
let method = request.method().clone();
debug_info!(
"Sending request: {} matrix-federation://{dest}{}",
method.as_str(),
url.path(),
);
let start = Instant::now();
let response = client.execute(request).await;
let elapsed = start.elapsed();
debug_info!(
"Request finished in {elapsed:?}: {} matrix-federation://{dest}{}",
method.as_str(),
url.path(),
);
let outcome = match &response {
| Ok(response) if response.status().is_success() => "success",
| Ok(response) if response.status().is_client_error() => "client_error",
| Ok(response) if response.status().is_server_error() => "server_error",
| Ok(_) => "other",
| Err(_) => "network_error",
};
self.services
.server
.metrics
.instruments
.federation_outgoing_request_duration
.record(elapsed.as_secs_f64(), &[KeyValue::new("outcome", outcome)]);
match response {
| Ok(response) =>
self.handle_response::<T>(dest, actual, &method, &url, response)
.await,
| Err(error) => {
// This awful wrapping hack is required because `reqwest::Error`
// does not implement `Clone`, but we need to convert it
// into a local error type to pass to should_mark_stale, but
// handle_error then itself expects the original
// reqwest error.
// handle_error *could* just take the wrapped error, but it'd
// have to unwrap it anyway.
let wrapped = Error::Reqwest(error.into());
if self.should_mark_stale(&wrapped) {
debug_info!("{dest} is unhealthy & stale due to a connect error");
self.mark_destination_stale(dest);
self.hit_unhealthy(dest.to_owned());
}
let Error::Reqwest(unwrapped) = wrapped else {
unreachable!("wrapped reqwest error must unwrap to a reqwest error");
};
Err(handle_error(actual, &method, &url, unwrapped.into())
.expect_err("always returns error"))
},
}
}
fn validate_url(&self, url: &Url) -> Result<()> {
if let Some(url_host) = url.host_str() {
if let Ok(ip) = IPAddress::parse(url_host) {
trace!("Checking request URL IP {ip:?}");
if !self.services.client.valid_cidr_range(&ip) {
return Err!(BadServerResponse("Not allowed to send requests to this IP"));
}
}
}
Ok(())
}
/// Handles a successful response from a remote server, reading and
/// deserializing the response body (if any).
async fn handle_response<T>(
&self,
dest: &ServerName,
actual: &Resolution,
method: &Method,
url: &Url,
response: Response,
) -> Result<T::IncomingResponse>
where
T: OutgoingRequest + Send,
{
const HUGE_ENDPOINTS: [&str; 2] =
["/_matrix/federation/v2/send_join/", "/_matrix/federation/v2/state/"];
let size_limit: u64 = if HUGE_ENDPOINTS.iter().any(|e| url.path().starts_with(e)) {
// Some federation endpoints can return huge response bodies, so
// we'll bump the limit for those endpoints specifically.
self.services
.server
.config
.max_request_size
.saturating_mul(10)
} else {
self.services.server.config.max_request_size
}
.try_into()
.expect("size_limit (usize) should fit within a u64");
let response =
into_http_response(dest, actual, method, url, response, size_limit).await?;
let (parts, body) = response.into_parts();
T::IncomingResponse::try_from_http_response(http::Response::from_parts(
parts,
body.as_ref(),
))
.inspect(|_| {
self.mark_healthy(dest);
})
.map_err(|e| {
// Bad 200 response is usually a sign of bad routing, so we'll mark
// as stale. We won't mark as offline though since we technically
// got a response.
self.mark_destination_stale(dest);
err!(BadServerResponse("Server returned bad 200 response: {e:?}"))
})
}
}
async fn into_http_response(
dest: &ServerName,
actual: &Resolution,
method: &Method,
url: &Url,
mut response: Response,
max_size: u64,
) -> Result<http::Response<Vec<u8>>> {
let status = response.status();
trace!(
%status, %method,
request_url = %url,
response_url = %response.url(),
"Received response from {}",
actual.base_url(),
);
if url.path() != "/.well-known/matrix/server" {
// For some reason, the spec starts off saying federation responses MUST
// include a `Content-Type` header of `application/json` on all
// responses. Then, the definition for the well-known server endpoint
// says that it's a SHOULD and that servers SHOULD just assume that the
// response is JSON regardless... And then there's media wayyyy down in
// the spec doc that just throws a `multipart/mixed` in there.
//
// Media has a special case regarding status codes. Only 200 OK
// responses should have the multipart assertion, as errors are JSON.
let ct = if url.path().starts_with("/_matrix/federation/v1/media/")
&& response.status().is_success()
{
"multipart/mixed"
} else {
"application/json"
};
enforce_content_type(&response, ct)?;
}
let mut http_response_builder = http::Response::builder()
.status(status)
.version(response.version());
mem::swap(
response.headers_mut(),
http_response_builder
.headers_mut()
.expect("http::response::Builder is usable"),
);
trace!("Waiting for response body...");
let http_response = http_response_builder
.body(response.limit_read(max_size).await.unwrap_or_default())
.expect("reqwest body is valid http body");
debug!("Got {status:?} for {method} {url}");
if !status.is_success() {
let (parts, body) = http_response.into_parts();
return Err(Error::Federation(
dest.to_owned(),
RumaError::from_http_response(http::Response::from_parts(parts, body.as_ref())),
));
}
Ok(http_response)
}
fn handle_error(
actual: &Resolution,
method: &Method,
url: &Url,
mut e: reqwest::Error,
) -> Result {
if e.is_timeout() || e.is_connect() {
e = e.without_url();
debug_warn!("{e:?}");
} else if e.is_redirect() {
debug_error!(
%method,
%url,
final_url = e.url().map(tracing::field::display),
"Redirect loop {}: {}",
actual.host,
e,
);
} else {
debug_error!("{e:?}");
}
Err(e.into())
}
/// Enforces that the `Content-Type` response header is present, and that the
/// first part matches the expected header. This does not enforce that multipart
/// responses have a boundary as only the text left of a semicolon, if present,
/// is compared.
///
/// Returns `M_UNKNOWN` with HTTP 415 (Unsupported Media Type) if the response
/// header is missing or mismatched.
fn enforce_content_type(response: &Response, expected: &str) -> Result<()> {
// N.B. Content type's value is case-insensitive per RFC 9110. The Matrix
// spec explicitly prescribes `application/json` and `multipart/mixed` etc,
// but it's assumed they're intended to be case-insensitive.
debug_assert_eq!(
expected,
expected.to_lowercase(),
"Content type checks should use the lowercase content type"
); // ^ Saves us double-folding
match response
.headers()
.get(&http::header::CONTENT_TYPE)
.and_then(|h| h.to_str().ok())
.map(|h| h.split_once(';').unwrap_or((h, "")).0.to_lowercase())
{
| Some(ct) if ct == expected => Ok(()),
| Some(ct) => Err!(Request(
Unknown(
"Remote server returned the wrong Content-Type: expected {expected}, got {ct:?}"
),
UNSUPPORTED_MEDIA_TYPE
)),
| None => Err!(Request(
Unknown("Remote server did not return a Content-Type header"),
UNSUPPORTED_MEDIA_TYPE
)),
}
}
/// A trait for the input types of acceptable path builders for outgoing
/// federation requests.
///
/// Ruma uses Rust's type system to encode the versioning scheme of endpoints in
/// the Matrix spec. Every endpoint has a `PathBuilder` associated type, which
/// has an `Input` associated type. Endpoints with multiple versions have
/// `VersionHistory` as their `PathBuilder`, which has `SupportedVersions`
/// as its `Input` type. Endpoints with no version have `SinglePath` as their
/// `PathBuilder`, which has `()` as its `Input` type. Both `SupportedVersions`
/// and `()` can be created out of thin air using static data (or no data at
/// all). This property is what the `FederationPathBuilderInput` trait
/// represents.
///
/// This trait allows the federation sender service's functions to accept
/// requests for either versioned or unversioned endpoints, by requiring that
/// the `Input` of the `PathBuilder` of the endpoint implements
/// `FederationPathBuilderInput`.
pub trait FederationPathBuilderInput {
fn create() -> Self;
}
impl FederationPathBuilderInput for () {
fn create() -> Self {}
}
impl FederationPathBuilderInput for Cow<'_, SupportedVersions> {
fn create() -> Self { Cow::Borrowed(&SUPPORTED_VERSIONS) }
}