Line data Source code
1 : /// A stand-alone program that routes connections, e.g. from
2 : /// `aaa--bbb--1234.external.domain` to `aaa.bbb.internal.domain:1234`.
3 : ///
4 : /// This allows connecting to pods/services running in the same Kubernetes cluster from
5 : /// the outside. Similar to an ingress controller for HTTPS.
6 : use std::{net::SocketAddr, sync::Arc};
7 :
8 : use futures::future::Either;
9 : use itertools::Itertools;
10 : use proxy::config::TlsServerEndPoint;
11 : use proxy::context::RequestMonitoring;
12 : use proxy::proxy::run_until_cancelled;
13 : use tokio::net::TcpListener;
14 :
15 : use anyhow::{anyhow, bail, ensure, Context};
16 : use clap::{self, Arg};
17 : use futures::TryFutureExt;
18 : use proxy::console::messages::MetricsAuxInfo;
19 : use proxy::stream::{PqStream, Stream};
20 :
21 : use tokio::io::{AsyncRead, AsyncWrite};
22 : use tokio_util::sync::CancellationToken;
23 : use utils::{project_git_version, sentry_init::init_sentry};
24 :
25 : use tracing::{error, info, Instrument};
26 :
27 : project_git_version!(GIT_VERSION);
28 :
29 0 : fn cli() -> clap::Command {
30 0 : clap::Command::new("Neon proxy/router")
31 0 : .version(GIT_VERSION)
32 0 : .arg(
33 0 : Arg::new("listen")
34 0 : .short('l')
35 0 : .long("listen")
36 0 : .help("listen for incoming client connections on ip:port")
37 0 : .default_value("127.0.0.1:4432"),
38 0 : )
39 0 : .arg(
40 0 : Arg::new("tls-key")
41 0 : .short('k')
42 0 : .long("tls-key")
43 0 : .help("path to TLS key for client postgres connections")
44 0 : .required(true),
45 0 : )
46 0 : .arg(
47 0 : Arg::new("tls-cert")
48 0 : .short('c')
49 0 : .long("tls-cert")
50 0 : .help("path to TLS cert for client postgres connections")
51 0 : .required(true),
52 0 : )
53 0 : .arg(
54 0 : Arg::new("dest")
55 0 : .short('d')
56 0 : .long("destination")
57 0 : .help("append this domain zone to the SNI hostname to get the destination address")
58 0 : .required(true),
59 0 : )
60 0 : }
61 :
62 : #[tokio::main]
63 0 : async fn main() -> anyhow::Result<()> {
64 0 : let _logging_guard = proxy::logging::init().await?;
65 0 : let _panic_hook_guard = utils::logging::replace_panic_hook_with_tracing_panic_hook();
66 0 : let _sentry_guard = init_sentry(Some(GIT_VERSION.into()), &[]);
67 0 :
68 0 : let args = cli().get_matches();
69 0 : let destination: String = args.get_one::<String>("dest").unwrap().parse()?;
70 0 :
71 0 : // Configure TLS
72 0 : let (tls_config, tls_server_end_point): (Arc<rustls::ServerConfig>, TlsServerEndPoint) = match (
73 0 : args.get_one::<String>("tls-key"),
74 0 : args.get_one::<String>("tls-cert"),
75 0 : ) {
76 0 : (Some(key_path), Some(cert_path)) => {
77 0 : let key = {
78 0 : let key_bytes = std::fs::read(key_path).context("TLS key file")?;
79 0 : let mut keys = rustls_pemfile::pkcs8_private_keys(&mut &key_bytes[..])
80 0 : .context(format!("Failed to read TLS keys at '{key_path}'"))?;
81 0 :
82 0 : ensure!(keys.len() == 1, "keys.len() = {} (should be 1)", keys.len());
83 0 : keys.pop().map(rustls::PrivateKey).unwrap()
84 0 : };
85 0 :
86 0 : let cert_chain_bytes = std::fs::read(cert_path)
87 0 : .context(format!("Failed to read TLS cert file at '{cert_path}.'"))?;
88 0 :
89 0 : let cert_chain = {
90 0 : rustls_pemfile::certs(&mut &cert_chain_bytes[..])
91 0 : .context(format!(
92 0 : "Failed to read TLS certificate chain from bytes from file at '{cert_path}'."
93 0 : ))?
94 0 : .into_iter()
95 0 : .map(rustls::Certificate)
96 0 : .collect_vec()
97 0 : };
98 0 :
99 0 : // needed for channel bindings
100 0 : let first_cert = cert_chain.first().context("missing certificate")?;
101 0 : let tls_server_end_point = TlsServerEndPoint::new(first_cert)?;
102 0 :
103 0 : let tls_config = rustls::ServerConfig::builder()
104 0 : .with_safe_default_cipher_suites()
105 0 : .with_safe_default_kx_groups()
106 0 : .with_protocol_versions(&[&rustls::version::TLS13, &rustls::version::TLS12])?
107 0 : .with_no_client_auth()
108 0 : .with_single_cert(cert_chain, key)?
109 0 : .into();
110 0 :
111 0 : (tls_config, tls_server_end_point)
112 0 : }
113 0 : _ => bail!("tls-key and tls-cert must be specified"),
114 0 : };
115 0 :
116 0 : // Start listening for incoming client connections
117 0 : let proxy_address: SocketAddr = args.get_one::<String>("listen").unwrap().parse()?;
118 0 : info!("Starting sni router on {proxy_address}");
119 0 : let proxy_listener = TcpListener::bind(proxy_address).await?;
120 0 :
121 0 : let cancellation_token = CancellationToken::new();
122 0 :
123 0 : let main = tokio::spawn(task_main(
124 0 : Arc::new(destination),
125 0 : tls_config,
126 0 : tls_server_end_point,
127 0 : proxy_listener,
128 0 : cancellation_token.clone(),
129 0 : ));
130 0 : let signals_task = tokio::spawn(proxy::handle_signals(cancellation_token));
131 0 :
132 0 : // the signal task cant ever succeed.
133 0 : // the main task can error, or can succeed on cancellation.
134 0 : // we want to immediately exit on either of these cases
135 0 : let signal = match futures::future::select(signals_task, main).await {
136 0 : Either::Left((res, _)) => proxy::flatten_err(res)?,
137 0 : Either::Right((res, _)) => return proxy::flatten_err(res),
138 0 : };
139 0 :
140 0 : // maintenance tasks return `Infallible` success values, this is an impossible value
141 0 : // so this match statically ensures that there are no possibilities for that value
142 0 : match signal {}
143 0 : }
144 :
145 0 : async fn task_main(
146 0 : dest_suffix: Arc<String>,
147 0 : tls_config: Arc<rustls::ServerConfig>,
148 0 : tls_server_end_point: TlsServerEndPoint,
149 0 : listener: tokio::net::TcpListener,
150 0 : cancellation_token: CancellationToken,
151 0 : ) -> anyhow::Result<()> {
152 0 : // When set for the server socket, the keepalive setting
153 0 : // will be inherited by all accepted client sockets.
154 0 : socket2::SockRef::from(&listener).set_keepalive(true)?;
155 :
156 0 : let connections = tokio_util::task::task_tracker::TaskTracker::new();
157 :
158 0 : while let Some(accept_result) =
159 0 : run_until_cancelled(listener.accept(), &cancellation_token).await
160 : {
161 0 : let (socket, peer_addr) = accept_result?;
162 :
163 0 : let session_id = uuid::Uuid::new_v4();
164 0 : let tls_config = Arc::clone(&tls_config);
165 0 : let dest_suffix = Arc::clone(&dest_suffix);
166 0 :
167 0 : connections.spawn(
168 0 : async move {
169 0 : socket
170 0 : .set_nodelay(true)
171 0 : .context("failed to set socket option")?;
172 :
173 0 : info!(%peer_addr, "serving");
174 0 : let ctx = RequestMonitoring::new(session_id, peer_addr.ip(), "sni_router", "sni");
175 0 : handle_client(ctx, dest_suffix, tls_config, tls_server_end_point, socket).await
176 0 : }
177 0 : .unwrap_or_else(|e| {
178 0 : // Acknowledge that the task has finished with an error.
179 0 : error!("per-client task finished with an error: {e:#}");
180 0 : })
181 0 : .instrument(tracing::info_span!("handle_client", ?session_id)),
182 : );
183 : }
184 :
185 0 : connections.close();
186 0 : drop(listener);
187 0 :
188 0 : connections.wait().await;
189 :
190 0 : info!("all client connections have finished");
191 0 : Ok(())
192 0 : }
193 :
194 : const ERR_INSECURE_CONNECTION: &str = "connection is insecure (try using `sslmode=require`)";
195 :
196 0 : async fn ssl_handshake<S: AsyncRead + AsyncWrite + Unpin>(
197 0 : raw_stream: S,
198 0 : tls_config: Arc<rustls::ServerConfig>,
199 0 : tls_server_end_point: TlsServerEndPoint,
200 0 : ) -> anyhow::Result<Stream<S>> {
201 0 : let mut stream = PqStream::new(Stream::from_raw(raw_stream));
202 :
203 0 : let msg = stream.read_startup_packet().await?;
204 : use pq_proto::FeStartupPacket::*;
205 :
206 0 : match msg {
207 : SslRequest => {
208 0 : stream
209 0 : .write_message(&pq_proto::BeMessage::EncryptionResponse(true))
210 0 : .await?;
211 : // Upgrade raw stream into a secure TLS-backed stream.
212 : // NOTE: We've consumed `tls`; this fact will be used later.
213 :
214 0 : let (raw, read_buf) = stream.into_inner();
215 0 : // TODO: Normally, client doesn't send any data before
216 0 : // server says TLS handshake is ok and read_buf is empy.
217 0 : // However, you could imagine pipelining of postgres
218 0 : // SSLRequest + TLS ClientHello in one hunk similar to
219 0 : // pipelining in our node js driver. We should probably
220 0 : // support that by chaining read_buf with the stream.
221 0 : if !read_buf.is_empty() {
222 0 : bail!("data is sent before server replied with EncryptionResponse");
223 0 : }
224 0 :
225 0 : Ok(Stream::Tls {
226 0 : tls: Box::new(raw.upgrade(tls_config).await?),
227 0 : tls_server_end_point,
228 : })
229 : }
230 0 : unexpected => {
231 0 : info!(
232 0 : ?unexpected,
233 0 : "unexpected startup packet, rejecting connection"
234 0 : );
235 0 : stream
236 0 : .throw_error_str(ERR_INSECURE_CONNECTION, proxy::error::ErrorKind::User)
237 0 : .await?
238 : }
239 : }
240 0 : }
241 :
242 0 : async fn handle_client(
243 0 : mut ctx: RequestMonitoring,
244 0 : dest_suffix: Arc<String>,
245 0 : tls_config: Arc<rustls::ServerConfig>,
246 0 : tls_server_end_point: TlsServerEndPoint,
247 0 : stream: impl AsyncRead + AsyncWrite + Unpin,
248 0 : ) -> anyhow::Result<()> {
249 0 : let tls_stream = ssl_handshake(stream, tls_config, tls_server_end_point).await?;
250 :
251 : // Cut off first part of the SNI domain
252 : // We receive required destination details in the format of
253 : // `{k8s_service_name}--{k8s_namespace}--{port}.non-sni-domain`
254 0 : let sni = tls_stream.sni_hostname().ok_or(anyhow!("SNI missing"))?;
255 0 : let dest: Vec<&str> = sni
256 0 : .split_once('.')
257 0 : .context("invalid SNI")?
258 : .0
259 0 : .splitn(3, "--")
260 0 : .collect();
261 0 : let port = dest[2].parse::<u16>().context("invalid port")?;
262 0 : let destination = format!("{}.{}.{}:{}", dest[0], dest[1], dest_suffix, port);
263 :
264 0 : info!("destination: {}", destination);
265 :
266 0 : let client = tokio::net::TcpStream::connect(destination).await?;
267 :
268 0 : let metrics_aux: MetricsAuxInfo = Default::default();
269 0 :
270 0 : // doesn't yet matter as pg-sni-router doesn't report analytics logs
271 0 : ctx.set_success();
272 0 : ctx.log();
273 0 :
274 0 : proxy::proxy::passthrough::proxy_pass(tls_stream, client, metrics_aux).await
275 0 : }
|
